RFC 4695
RTP Payload Format for MIDI
RFC 4695: RTP Payload Format for MIDI 🎹✨
Part 1 of 38 🦋
Okay bestie, this one is from the Network Working Group 🌸 and it's written by J. Lazzaro and J. Wawrzynek, both from UC Berkeley. It came out in November 2006 and it's a Standards Track doc, which means it's an actual official protocol spec, not just a vibe. It's all about how to send MIDI (the music language that keyboards and synths speak) over RTP, the protocol that carries real-time media across the internet. Iconic. 💜
Status of This Memo ✨
So this document is an Internet standards track protocol for the Internet community 💅 and it's basically inviting everyone to chime in with discussion and suggestions for improvements 🫶. If you want to know where this protocol stands in the official standardization process, check the current edition of the "Internet Official Protocol Standards" (STD 1) 📚. And the distribution of this memo is unlimited, so share it with everyone 🌈.
Copyright Notice 🎀
Copyright (C) The IETF Trust (2006). Legally boring but still very much a must-have 💖.
Abstract 🦋
Okay, the big picture: this memo describes an RTP payload format for the MIDI (Musical Instrument Digital Interface) command language 🎶. The format encodes all commands that may legally appear on a MIDI 1.0 DIN cable. Every single one. No cap 🔥.
It's built for two different vibes. First, interactive apps like network musical performance, where you're jamming live with someone across the internet 🎹💜. Second, content-delivery apps like file streaming 🎧. Both are valid, and the format works for both, which is so iconic.
It can run over unicast and multicast UDP and TCP 🌐, and it comes with tools for graceful recovery from packet loss 🛟. Lowkey the best part, because packet loss is the worst. Also, stream behavior, including the MIDI rendering method, can be customized during session setup ⚙️. So you get to tweak the stream to your liking before the music even starts 💅.
The format also doubles as a mode for the mpeg4-generic format 👑. That lets it support the MPEG 4 Audio Object Types for General MIDI, Downloadable Sounds Level 2, and Structured Audio 🌸. It's giving multitasking queen energy.
Table of Contents 📚🧚
1. Introduction ....................................................4
1.1. Terminology ................................................5
1.2. Bitfield Conventions .......................................6
2. Packet Format ...................................................6
2.1. RTP Header .................................................7
2.2. MIDI Payload ..............................................11
3. MIDI Command Section ...........................................12
3.1. Timestamps ...............................................14
3.2. Command Coding ...........................................16
4. The Recovery Journal System ....................................22
5. Recovery Journal Format ........................................24
6. Session Description Protocol ...................................28
6.1. Session Descriptions for Native Streams ...................29
6.2. Session Descriptions for mpeg4-generic Streams ............30
6.3. Parameters ................................................33
7. Extensibility ..................................................34
8. Congestion Control .............................................35
9. Security Considerations ........................................35
10. Acknowledgements ..............................................36
11. IANA Considerations ...........................................37
11.1. rtp-midi Media Type Registration .........................37
11.1.1. Repository Request for "audio/rtp-midi" ...........40
11.2. mpeg4-generic Media Type Registration ....................41
11.2.1. Repository Request for Mode rtp-midi for
mpeg4-generic .....................................44
11.3. asc Media Type Registration ..............................46
A. The Recovery Journal Channel Chapters ..........................48
A.1. Recovery Journal Definitions ..............................48
A.2. Chapter P: MIDI Program Change ............................52
A.3. Chapter C: MIDI Control Change ............................53
A.3.1. Log Inclusion Rules ................................54
A.3.2. Controller Log Format ..............................55
A.3.3. Log List Coding Rules ..............................57
A.3.4. The Parameter System ...............................60
A.4. Chapter M: MIDI Parameter System ..........................62
A.4.1. Log Inclusion Rules ................................64
A.4.2. Log Coding Rules ...................................65
A.4.2.1. The Value Tool .............................67
A.4.2.2. The Count Tool .............................70
A.5. Chapter W: MIDI Pitch Wheel ...............................71
A.6. Chapter N: MIDI NoteOff and NoteOn ........................71
A.6.1. Header Structure ...................................73
A.6.2. Note Structures ....................................74
A.7. Chapter E: MIDI Note Command Extras .......................75
A.7.1. Note Log Format ....................................76
A.7.2. Log Inclusion Rules ................................76
A.8. Chapter T: MIDI Channel Aftertouch ........................77
A.9. Chapter A: MIDI Poly Aftertouch ...........................78
B. The Recovery Journal System Chapters ...........................79
B.1. System Chapter D: Simple System Commands ..................79
B.1.1. Undefined System Commands ..........................80
B.2. System Chapter V: Active Sense Command ....................83
B.3. System Chapter Q: Sequencer State Commands ................83
B.3.1. Non-compliant Sequencers ...........................85
B.4. System Chapter F: MIDI Time Code Tape Position ............86
B.4.1. Partial Frames .....................................88
B.5. System Chapter X: System Exclusive ........................89
B.5.1. Chapter Format .....................................90
B.5.2. Log Inclusion Semantics ............................92
B.5.3. TCOUNT and COUNT Fields ............................95
C. Session Configuration Tools ....................................95
C.1. Configuration Tools: Stream Subsetting ....................97
C.2. Configuration Tools: The Journalling System ..............101
C.2.1. The j_sec Parameter ...............................102
C.2.2. The j_update Parameter ............................103
C.2.2.1. The anchor Sending Policy .................104
C.2.2.2. The closed-loop Sending Policy ............104
C.2.2.3. The open-loop Sending Policy ..............108
C.2.3. Recovery Journal Chapter Inclusion Parameters .....110
C.3. Configuration Tools: Timestamp Semantics .................115
C.3.1. The comex Algorithm ...............................115
C.3.2. The async Algorithm ...............................116
C.3.3. The buffer Algorithm ..............................117
C.4. Configuration Tools: Packet Timing Tools .................118
C.4.1. Packet Duration Tools .............................119
C.4.2. The guardtime Parameter ...........................120
C.5. Configuration Tools: Stream Description ..................121
C.6. Configuration Tools: MIDI Rendering ......................128
C.6.1. The multimode Parameter ...........................129
C.6.2. Renderer Specification ............................129
C.6.3. Renderer Initialization ...........................131
C.6.4. MIDI Channel Mapping ..............................133
C.6.4.1. The smf_info Parameter ....................134
C.6.4.2. The smf_inline, smf_url, and smf_cid
Parameters ................................136
C.6.4.3. The chanmask Parameter ....................136
C.6.5. The audio/asc Media Type ..........................137
C.7. Interoperability .........................................139
C.7.1. MIDI Content Streaming Applications ...............139
C.7.2. MIDI Network Musical Performance Applications .....142
D. Parameter Syntax Definitions ..................................150
E. A MIDI Overview for Networking Specialists ....................156
E.1. Commands Types ...........................................159
E.2. Running Status ...........................................159
E.3. Command Timing ...........................................160
E.4. AudioSpecificConfig Templates for MMA Renderers ..........160
References .......................................................165
Normative References .............................................165
Informative References ...........................................1661. Introduction ✨
okay so the Internet Engineering Task Force (IETF) has cooked up a bunch of focused tools for multimedia networking 🎀 ([RFC3550] [RFC4566] [RFC3261] [RFC2326]). You can mix and match them in different ways to power all kinds of real-time apps over Internet Protocol (IP) networks 🦋
for example, a telephony app might use the Session Initiation Protocol (SIP, [RFC3261]) to set up a phone call 📞 call setup would include negotiating to agree on a common audio codec [RFC3264]. and those negotiations would use the Session Description Protocol (SDP, [RFC4566]) to describe the candidate codecs 💅
once the call is set up, the audio data flows between the two parties using the Real Time Protocol (RTP, [RFC3550]) under whatever applicable profile (for example, the Audio/Visual Profile (AVP, [RFC3551])) 🌸 and the tools from this telephony example (SIP, SDP, RTP) could be combined in a totally different way to support a content streaming app, maybe alongside other tools like the Real Time Streaming Protocol (RTSP, [RFC2326]) 🧚
now the MIDI (Musical Instrument Digital Interface) command language [MIDI] is literally everywhere in music apps that are kind of analogous to the examples above 🎹 on stage and in the recording studio, MIDI is used for interactive remote control of musical instruments, which is giving telephony vibes ✨ and on web pages, Standard MIDI Files (SMFs, [MIDI]) rendered using the General MIDI standard [MIDI] are a low-bandwidth substitute for audio streaming 🍓
this memo is motivated by a simple premise 💜 if MIDI performances could be sent as RTP streams that are managed by IETF session tools, the MIDI and IETF application domains might hybridize. iykyk, it's a whole glow-up 🌈
for example, interoperable MIDI networking could foster network music performance apps, where a group of musicians in different physical locations interact over a network to perform as if they were in the same room [NMP]. ngl that's iconic 😭 and as a second example, the streaming community may begin to use MIDI for low-bitrate audio coding, perhaps alongside normative sound synthesis methods [MPEGSA] 🔥
so to let MIDI apps use RTP, this memo defines an RTP payload format and its media type 🫶 Sections 2-5 and Appendices A-B define the RTP payload format. Section 6 and Appendices C-D define the media types that identify the payload format, the parameters needed for configuration, and how those parameters are used in SDP 👑
Appendix C also includes interoperability guidelines for the example applications described above: network musical performance using SIP (Appendix C.7.2) and content-streaming using RTSP (Appendix C.7.1). ✨
another potential app area for RTP MIDI is MIDI networking for professional audio equipment and electronic musical instruments 🎧 we do not offer interoperability guidelines for this application in this memo. however, RTP MIDI was designed with stage and studio apps in mind, and we expect that efforts to define a stage and studio framework will rely on RTP MIDI for MIDI transport services 🎀
some apps may require MIDI media delivery at a certain service quality level (latency, jitter, packet loss, etc) 🙈 RTP itself does not provide service guarantees. however, apps may use lower-layer network protocols to configure the quality of the transport services that RTP uses. those protocols may act to reserve network resources for RTP flows [RFC2205], or they may simply direct RTP traffic onto a dedicated "media network" in a local installation 🛤️ and note that RTP and the MIDI payload format do provide tools that apps may use to hit the best possible real-time performance at a given service level 💖
this memo normatively defines the syntax and semantics of the MIDI payload format. however, this memo does not define algorithms for sending and receiving packets 🙃 an ancillary document [RFC4696] gives informative guidance on algorithms, and supplemental info can be found in related conference publications [NMP] [GRAME] 📚
throughout this memo, the phrase "native stream" refers to a stream that uses the rtp-midi media type. the phrase "mpeg4-generic stream" refers to a stream that uses the mpeg4-generic media type (in mode rtp-midi) to operate in an MPEG 4 environment [RFC3640] 🧩 Section 6 describes this distinction in detail 💅
1.1. Terminology 🔤
in this document, the key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" are to be interpreted as described in BCP 14, RFC 2119 [RFC2119] 🫶
1.2. Bitfield Conventions 🧃
in this document, the packet bitfields that share a common name often have identical semantics. since most of these bitfields appear in Appendices A-B, we define the common bitfield names in Appendix A.1 ✨
however, a few of these common names also show up in the main text of this document, so for convenience, here are those definitions:
- R flag bit. R flag bits are reserved for future use. Senders MUST set R bits to 0. Receivers MUST ignore R bit values. 🚫
- LENGTH field. All fields named LENGTH (as distinct from LEN) code the number of octets in the structure that contains them, including the header they reside in and all hierarchical levels below them. If a structure contains a LENGTH field, a receiver MUST use the LENGTH field value to advance past the structure during parsing, rather than use knowledge about the internal format of the structure. 📏💜
2. Packet Format 📦
in this section, we introduce the format of RTP MIDI packets 🎀 the description includes some background on RTP, for the benefit of MIDI implementors who are new to IETF tools. implementors should consult [RFC3550] for an authoritative description of RTP 📖
this memo assumes the reader is familiar with MIDI syntax and semantics 🎹 Appendix E provides a MIDI overview, at a level of detail sufficient to understand most of this memo. implementors should consult [MIDI] for an authoritative description of MIDI 🌈
the MIDI payload format maps a MIDI command stream (16 voice channels + systems) onto an RTP stream 🦋 an RTP media stream is a sequence of logical packets that share a common format. each packet has two parts: the RTP header and the MIDI payload. Figure 1 shows this format (vertical space separates the header and payload) 🧁
we call RTP packets "logical" packets to highlight the fact that RTP itself is not a network-layer protocol 🫧 instead, RTP packets get mapped onto network protocols (such as unicast UDP, multicast UDP, or TCP) by an application [ALF]. the interleaved mode of the Real Time Streaming Protocol (RTSP, [RFC2326]) is one example of an RTP mapping to TCP transport, and so is [RFC4571] 🌸
2.1. RTP Header 💖
[RFC3550] has the full scoop on every RTP header field 📚, so we're just zooming in on the few fields that matter for MIDI apps ✨. Everything is coded in network byte order (big-endian), so keep that in mind 🦋.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| V |P|X| CC |M| PT | Sequence number |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Timestamp |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| SSRC |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| MIDI command section ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Journal section ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+Figure 1 -- Packet format
The 1-bit M field acts differently depending on the media type of the stream 🌸. For native streams, the M bit MUST be set to 1 if the MIDI command section has a non-zero LEN field, and MUST be set to 0 otherwise. For mpeg4-generic streams, the M bit MUST be set to 1 for all packets (to conform to [RFC3640]).
In an RTP MIDI stream, the 16-bit sequence number field starts at a randomly chosen value and goes up by one (modulo 2^16) for each packet sent in the stream 🔢. A related quantity, the 32-bit extended packet sequence number, can be computed by tracking rollovers of the 16-bit sequence number. Note that different receivers of the same stream may compute different extended packet sequence numbers, depending on when the receiver joined the session 👀.
The 32-bit timestamp field sets the base timestamp value for the packet ⏰. The payload codes MIDI command timing relative to this value. The timestamp units are set by the clock rate parameter. For example, if the clock rate has a value of 44100 Hz, two packets whose base timestamp values differ by 2 seconds have RTP timestamp fields that differ by 88200.
Note that the clock rate parameter is not encoded within each RTP MIDI packet 🙈. A receiver of an RTP MIDI stream becomes aware of the clock rate as part of the session setup process. For example, if a session management tool uses the Session Description Protocol (SDP, [RFC4566]) to describe a media session, the clock rate parameter is set using the rtpmap attribute. We show examples of session setup in Section 6.
For RTP MIDI streams destined to be rendered into audio 🎧, the clock rate SHOULD be an audio sample rate of 32 KHz or higher. This recommendation is due to the sensitivity of human musical perception to small timing errors in musical note sequences, and due to the timbral changes that occur when two near-simultaneous MIDI NoteOns are rendered with a different timing than that desired by the content author due to clock rate quantization. RTP MIDI streams that are not destined for audio rendering (such as MIDI streams that control stage lighting) MAY use a lower clock rate but SHOULD use a clock rate high enough to avoid timing artifacts in the application 💡.
For RTP MIDI streams destined to be rendered into audio, the clock rate SHOULD be chosen from rates in common use in professional audio applications or in consumer audio distribution 🎶. At the time of this writing, these rates include 32 KHz, 44.1 KHz, 48 KHz, 64 KHz, 88.2 KHz, 96 KHz, 176.4 KHz, and 192 KHz. If the RTP MIDI session is a part of a synchronized media session that includes another (non-MIDI) RTP audio stream with a clock rate of 32 KHz or higher, the RTP MIDI stream SHOULD use a clock rate that matches the clock rate of the other audio stream. However, if the RTP MIDI stream is destined to be rendered into audio, the RTP MIDI stream SHOULD NOT use a clock rate lower than 32 KHz, even if this second stream has a clock rate less than 32 KHz 🚫.
Timestamps of consecutive packets don't necessarily increment at a fixed rate, because RTP MIDI packets aren't necessarily sent at a fixed rate 🏃♀️. The degree of packet transmission regularity reflects the underlying application dynamics. Interactive applications may vary the packet sending rate to track the gestural rate of a human performer, whereas content-streaming applications may send packets at a fixed rate.
So the timestamps for two sequential RTP packets may be identical, or the second packet may have a timestamp arbitrarily larger than the first packet (modulo 2^32). Section 3 places additional restrictions on the RTP timestamps for two sequential RTP packets, as does the guardtime parameter (Appendix C.4.2) 🔒.
We use the term "media time" to denote the temporal duration of the media coded by an RTP packet ⌛. The media time coded by a packet is computed by subtracting the last command timestamp in the MIDI command section from the RTP timestamp (modulo 2^32). If the MIDI list of the MIDI command section of a packet is empty, the media time coded by the packet is 0 ms. Appendix C.4.1 discusses media time issues in detail.
We now define RTP session semantics, in the context of sessions specified using the session description protocol [RFC4566] 🗂️. A session description media line ("m=") specifies an RTP session. An RTP session has an independent space of 2^32 synchronization sources. Synchronization source identifiers are coded in the SSRC header field of RTP session packets. The payload types that may appear in the PT header field of RTP session packets are listed at the end of the media line.
Several RTP MIDI streams may appear in an RTP session 🎀. Each stream is distinguished by a unique SSRC value and has a unique sequence number and RTP timestamp space. Multiple streams in the RTP session may be sent by a single party. Multiple parties may send streams in the RTP session. An RTP MIDI stream encodes data for a single MIDI command name space (16 voice channels + Systems).
Streams in an RTP session may use different payload types, or they may use the same payload type. However, each party may send, at most, one RTP MIDI stream for each payload type mapped to an RTP MIDI payload format in an RTP session. Recall that dynamic binding of payload type numbers in [RFC4566] lets a party map many payload type numbers to the RTP MIDI payload format; thus a party may send many RTP MIDI streams in a single RTP session 🧩. Pairs of streams (unicast or multicast) that communicate between two parties in an RTP session and that share a payload type have the same association as a MIDI cable pair that cross-connects two devices in a MIDI 1.0 DIN network 🔌.
The RTP session architecture described above is efficient in its use of network ports 🏆, as one RTP session (using a port pair per party) supports the transport of many MIDI name spaces (16 MIDI channels + systems). We define tools for grouping and labelling MIDI name spaces across streams and sessions in Appendix C.5 of this memo.
The RTP header timestamps for each stream in an RTP session have separately and randomly chosen initialization values 🎲. Receivers use the timing fields encoded in the RTP control protocol (RTCP, [RFC3550]) sender reports to synchronize the streams sent by a party. The SSRC values for each stream in an RTP session are also separately and randomly chosen, as described in [RFC3550]. Receivers use the CNAME field encoded in RTCP sender reports to verify that streams were sent by the same party, and to detect SSRC collisions, as described in [RFC3550] 🕵️♀️.
In some applications, a receiver renders MIDI commands into audio (or into control actions, such as the rewind of a tape deck or the dimming of stage lights) 🎛️. In other applications, a receiver presents a MIDI stream to software programs via an Application Programmer Interface (API). Appendix C.6 defines session configuration tools to specify what receivers should do with a MIDI command stream.
If a multimedia session uses different RTP MIDI streams to send different classes of media, the streams MUST be sent over different RTP sessions 🚦. For example, if a multimedia session uses one MIDI stream for audio and a second MIDI stream to control a lighting system, the audio and lighting streams MUST be sent over different RTP sessions, each with its own media line 🎶💡.
Session description tools defined in Appendix C.5 let a sending party split a single MIDI name space (16 voice channels + systems) over several RTP MIDI streams. Split transport of a MIDI command stream is a delicate task 😭, because correct command stream reconstruction by a receiver depends on exact timing synchronization across the streams.
To support split name spaces, we define the following requirements 📋:
- A party MUST NOT send several RTP MIDI streams that share a MIDI name space in the same RTP session. Instead, each stream MUST be sent from a different RTP session.
- If several RTP MIDI streams sent by a party share a MIDI name space, all streams MUST use the same SSRC value and MUST use the same randomly chosen RTP timestamp initialization value.
These rules let a receiver identify streams that share a MIDI name space (by matching SSRC values) and also let a receiver accurately reconstruct the source MIDI command stream (by using RTP timestamps to interleave commands from the two streams) 🧠. Care MUST be taken by senders to ensure that SSRC changes due to collisions are reflected in both streams. Receivers MUST regularly examine the RTCP CNAME fields associated with the linked streams, to ensure that the assumed link is legitimate and not the result of an SSRC collision by another sender 🛡️.
Except for the special cases described above, a party may send many RTP MIDI streams in the same session 💜. However, it is sometimes advantageous for two RTP MIDI streams to be sent over different RTP sessions. For example, two streams may need different values for RTP session-level attributes (such as the sendonly and recvonly attributes). As a second example, two RTP sessions may be needed to send two unicast streams in a multimedia session that originate on different computers (with different IP numbers). Two RTP sessions are needed in this case because transport addresses are specified on the RTP-session or multimedia-session level, not on a payload type level.
On a final note, in some uses of MIDI, parties send bidirectional traffic to conduct transactions (such as file exchange) 📁. These commands were designed to work over MIDI 1.0 DIN cable networks may be configured in a multicast topology, which use pure "party-line" signalling. Thus, if a multimedia session ensures a multicast connection between all parties, bidirectional MIDI commands will work without additional support from the RTP MIDI payload format 🥳.
2.2. MIDI Payload 🎹
okay so the payload (Figure 1) MUST begin with the MIDI command section 💖 This is the part that codes a (possibly empty) list of timestamped MIDI commands, and honestly it's the whole reason the payload format exists ✨
The payload MAY also contain a journal section 🧚 The journal section is the resiliency queen 👑 It codes the recent history of the stream so things can recover if packets get lost. A flag in the MIDI command section tells you whether a journal section is in the payload 🫶
Section 3 defines the MIDI command section. Sections 4-5 and Appendices A-B define the recovery journal, which is the default format for the journal section 📚 We're going to talk about how these payload sections behave in a stream in an RTP session, so stay with me 🌸
The journalling method for a stream gets set at the start of a session and MUST NOT be changed after that 🔒 A stream may be set to use the recovery journal, to use an alternative journal format (none are defined in this memo), or not to use a journal at all 🦋
The default journalling method is inferred from the stream's transport type 🌈 Streams that use unreliable transport (such as UDP) default to using the recovery journal, and streams that use reliable transport (such as TCP) default to not using a journal. Appendix C.2.1 defines session configuration tools for overriding these defaults. For all types of transport, a sender MUST transmit an RTP packet stream with consecutive sequence numbers (modulo 2^16) 💅
If a stream uses the recovery journal, every payload in the stream MUST include a journal section 🍓 If a stream does not use journalling, a journal section MUST NOT appear in a stream payload. If a stream uses an alternative journal format, the specification for that journal format defines the inclusion policy 🙈
If a stream is sent over UDP, the Maximum Transmission Unit (MTU) of the underlying network limits how big the payload section can practically be (an Ethernet MTU is 1500 octets, for example). This matters for applications where predictable and minimal packet transmission latency is critical 😭 A sender SHOULD NOT create RTP MIDI UDP packets whose size exceeds the MTU of the underlying network. Instead, the sender SHOULD take steps to keep the maximum packet size under the MTU limit 🔥
There are a bunch of ways to do that 💜 The default closed-loop recovery journal sending policy (defined in Appendix C.2.2.2) uses RTP control protocol (RTCP, [RFC3550]) feedback to manage the RTP MIDI packet size. Section 3.2 and Appendix B.5.2 also give you specific tools for managing the size of packets that code MIDI System Exclusive (0xF0) commands 🎀 Appendix C.5 defines session configuration tools that may be used to split a dense MIDI name space into several UDP streams (each sent in a different RTP session, per Section 2.1), so the payload fits comfortably into an MTU ✨ Another option is to use TCP. And Section 4.3 of [RFC4696] has non-normative advice for packet size management 🧚
3. MIDI Command Section 🎶
Figure 2 shows the format of the MIDI command section 💅
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|B|J|Z|P|LEN... | MIDI list ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 2 -- MIDI command sectionThe MIDI command section starts with a variable-length header 🦋
The header field LEN codes the number of octets in the MIDI list that follow the header. If the header flag B is 0, the header is one octet long and LEN is a 4-bit field, so the maximum MIDI list length is 15 octets 🌸
If B is 1, the header is two octets long and LEN is a 12-bit field, which supports a maximum MIDI list length of 4095 octets 🍓 LEN is coded in network byte order (big-endian). The 4 bits of LEN that appear in the first header octet code the most significant 4 bits of the 12-bit LEN value 🤓
A LEN value of 0 is legal, and it codes an empty MIDI list 🫧
If the J header bit is set to 1, a journal section MUST appear after the MIDI command section in the payload 🔥 If the J header bit is set to 0, the payload MUST NOT contain a journal section.
We define the semantics of the P header bit in Section 3.2 👑
If the LEN header field is nonzero, the MIDI list has the structure shown in Figure 3 🌈
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Delta Time 0 (1-4 octets long, or 0 octets if Z = 1) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| MIDI Command 0 (1 or more octets long) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Delta Time 1 (1-4 octets long) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| MIDI Command 1 (1 or more octets long) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Delta Time N (1-4 octets long) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| MIDI Command N (0 or more octets long) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 3 -- MIDI list structureIf the header flag Z is 1, the MIDI list begins with a complete MIDI command (coded in the MIDI Command 0 field, in Figure 3) preceded by a delta time (coded in the Delta Time 0 field). If Z is 0, the Delta Time 0 field is not present in the MIDI list, and the command coded in the MIDI Command 0 field has an implicit delta time of 0 ✨
The MIDI list structure may also optionally encode a list of N additional complete MIDI commands, each coded in a MIDI Command K field. Each additional command MUST be preceded by a Delta Time K field, which codes that command's delta time 🎀 We talk about the exceptions to the "command fields code complete MIDI commands" rule in Section 3.2 🙈
The final MIDI command field (the MIDI Command N field, shown in Figure 3) in the MIDI list MAY be empty 🫶 Moreover, a MIDI list MAY consist of a single delta time (encoded in the Delta Time 0 field) without an associated command (which would have been encoded in the MIDI Command 0 field). These rules enable MIDI coding features that are explained in Section 3.1. We're holding off on those explanations because you need to understand RTP MIDI timestamps first to get them 💜
3.1. Timestamps ⏰
In this section, we describe how RTP MIDI encodes a timestamp for each MIDI list command 🧚 Command timestamps have the same units as RTP packet header timestamps (described in Section 2.1 and [RFC3550]). Recall that RTP timestamps have units of seconds, and their scaling is set during session configuration (see Section 6.1 and [RFC4566]) 🍓
As shown in Figure 3, the MIDI list encodes time using a compact delta-time format. The RTP MIDI delta time syntax is a modified form of the MIDI File delta time syntax [MIDI] 🌸 RTP MIDI delta times use 1-4 octet fields to encode 32-bit unsigned integers. Figure 4 shows the encoded and decoded forms of delta times. Delta time values may be legally encoded in multiple formats. For example, there are four legal ways to encode the zero delta time (0x00, 0x8000, 0x808000, 0x80808000) 🤯
RTP MIDI uses delta times to encode a timestamp for each MIDI command. The timestamp for MIDI Command K is the summation (modulo 2^32) of the RTP timestamp and the decoded delta times 0 through K. This cumulative coding technique is borrowed from MIDI File delta time coding, and it's efficient because it reduces the number of multi-octet delta times 💖
All command timestamps in a packet MUST be less than or equal to the RTP timestamp of the next packet in the stream (modulo 2^32) 🔒
This restriction makes sure that a particular RTP MIDI packet in a stream is uniquely responsible for encoding time starting at the moment after the RTP timestamp in the RTP packet header, and ending at the moment before the final command timestamp encoded in the MIDI list. The "moment before" and "moment after" qualifiers account for the "less than or equal" semantics (as opposed to "strictly less than") in the sentence above this paragraph 🙈
It's possible to "pad" the end of an RTP MIDI packet with time that is guaranteed to be void of MIDI commands. You do this by setting the "Delta Time N" field of the MIDI list to the end of the void time and omitting its corresponding "MIDI Command N" field (a syntactic construction the preamble of Section 3 expressly made legal) 🫧
It's also possible to code an RTP MIDI packet to express that a period of time in the stream is void of MIDI commands. The RTP timestamp in the header codes the start of the void time. The MIDI list of this packet consists of a "Delta Time 0" field that codes the end of the void time. No other fields are present in the MIDI list (a syntactic construction the preamble of Section 3 also expressly made legal) 💅
By default, a command timestamp indicates the execution time for the command. The difference between two timestamps indicates the time delay between the execution of the commands. This difference may be zero, which codes simultaneous execution. In this memo, we call this interpretation of timestamps "comex" (COMmand EXecution) semantics. We formally define comex semantics in Appendix C.3 🌈
The comex interpretation works really well for transcoding a Standard MIDI File (SMF) into an RTP MIDI stream, since SMFs code a timestamp for each MIDI command stored in the file ✨ To transcode an SMF that uses metric time markers, use the SMF tempo map (encoded in the SMF as meta-events) to convert metric SMF timestamp units into seconds-based RTP timestamp units 🎶
The comex interpretation also works well for MIDI hardware controllers that code raw sensor data directly onto an RTP MIDI stream. This controller design is preferable to a design that converts raw sensor data into a MIDI 1.0 cable command stream and then transcodes that stream onto an RTP MIDI stream 🎹
The comex interpretation of timestamps is usually not the best choice for transcoding a MIDI source that uses implicit command timing (such as MIDI 1.0 DIN cables) into an RTP MIDI stream. Appendix C.3 defines alternatives to comex semantics and describes session configuration tools for selecting the timestamp interpretation semantics for a stream 🦋
One-Octet Delta Time:
Encoded form: 0ddddddd
Decoded form: 00000000 00000000 00000000 0ddddddd
Two-Octet Delta Time:
Encoded form: 1ccccccc 0ddddddd
Decoded form: 00000000 00000000 00cccccc cddddddd
Three-Octet Delta Time:
Encoded form: 1bbbbbbb 1ccccccc 0ddddddd
Decoded form: 00000000 000bbbbb bbcccccc cddddddd
Four-Octet Delta Time:
Encoded form: 1aaaaaaa 1bbbbbbb 1ccccccc 0ddddddd
Decoded form: 0000aaaa aaabbbbb bbcccccc cddddddd
Figure 4 -- Decoding delta time formats3.2. Command Coding 🎀
Okay so every non-empty MIDI Command field in the MIDI list codes one of the MIDI command types that can legally show up on a MIDI 1.0 DIN cable 🎹✨ Standard MIDI File meta-events don't fit that definition, so they MUST NOT appear in the MIDI list, no exceptions bestie 🙅♀️ As a rule, each MIDI Command field holds one complete command, in the binary command format from [MIDI]. There are a few plot twists, and we'll get to them below 🦋
The first MIDI channel command in the MIDI list MUST include a status octet 💖 After that, running status coding (from [MIDI]) MAY be used for every other channel command in the list 🌸 Also, System Common and System Exclusive messages (0xF0 ... 0xF7) cancel the running status state, but System Real-time messages (0xF8 ... 0xFF) do NOT touch it. Every System command in the MIDI list MUST include a status octet too 💅
Now here's where it gets slightly messy 👀 The first channel command in the MIDI list MUST include a status octet, right? But the original command in the MIDI source data stream might not have one, because the source was using running status. If the status octet of that first channel command doesn't show up in the source stream, the P (phantom) header bit MUST be set to 1 🧚 In every other case, the P bit MUST be set to 0.
The P bit is about the MIDI source data stream, not how the MIDI list is encoded 🫶 So no matter what the P bit says, the MIDI list MUST still include the status octet. Iconic behaviour, honestly.
Receivers MUST be able to decode running status, so senders can feel free to use it to save bandwidth 🍓 Just don't mess with the timing 😭 Senders SHOULD NOT add timing jitter to an existing MIDI command stream by using or removing running status in a sloppy way. This matters most for senders whose RTP MIDI streams might get transcoded onto a MIDI 1.0 DIN cable [MIDI] by the receiver. Both the timestamps and the command coding (running status or not) must follow the physical limits of implicit time coding over a slow serial line 🐢
On a MIDI 1.0 DIN cable [MIDI], a System Real-time command may be tucked inside another "host" MIDI command 🎀 That structure is NOT supported in this payload format. A MIDI Command field in the MIDI list codes exactly one MIDI command, either partially or completely.
So to encode an embedded System Real-time command, senders MUST pull it out of its host and code it in the MIDI list as its own separate command 🔥 The host command and the System Real-time command SHOULD show up in the same MIDI list. The delta time of the System Real-time command SHOULD produce a command timestamp that records where it originally sat in the stream ✨
There are two ways to encode MIDI System Exclusive (SysEx) commands in the MIDI list 👑 The first is verbatim: a SysEx command may be coded in a single MIDI Command field, which is a 0xF0 octet, then any number of data octets, then a 0xF7 octet.
The second is segmentation. The command gets split into two or more SysEx command segments, and each segment gets its own MIDI Command field in the MIDI list 🧩
Segmentation exists so that SysEx commands which encode information in the timing of their data octets still work. Splitting the command into segments means each data octet can get its own delta time 🕰️ It also means big SysEx commands can be spread across several packets, which is so useful 🌈
To segment a SysEx command, first split its data octet list into two or more sublists 💅 The last sublist MAY be empty (no octets at all). Every other sublist MUST contain at least one data octet. To finish the job, add the status octets from Figure 5 to the head and tail of the first, last, and any "middle" sublists. Figure 6 shows example segmentations of a SysEx command 📸
A sender MAY cancel a segmented SysEx command that's still being sent, by sending the "cancel" sublist from Figure 5 🙈 A "cancel" sublist MAY follow a "first" or "middle" sublist in the transmission, but it MUST NOT follow a "last" sublist. The cancel MUST be empty, so 0xF7 0xF4 is the only legal cancel sublist, full stop 💀
Cancelling is needed because Appendix C.1 defines configuration tools that let session parties exclude certain SysEx commands from the stream 🙅♀️ If a sender transcodes a MIDI source onto an RTP MIDI stream under those constraints, it's on the sender to keep unwanted commands off the RTP MIDI stream.
The cancel feature is what lets a sender start sending a command before the MIDI source has finished sending it 🦋 If the sender decides the in-progress command shouldn't appear on the RTP stream, it cancels it. Without a way to cancel, senders would have to use a high-latency store-and-forward approach, validating every SysEx command before sending any of it, and nobody wants that lag 😭
The recommended receiver reaction to a cancellation depends on what the receiver can do 🎧 For instance, a sound synthesizer that parses RTP MIDI packets straight into audio knows SysEx commands can be cancelled in RTP MIDI. These receivers SHOULD detect a SysEx cancellation in the MIDI list and act as if they had never received that SysEx command 💜
A second example: a synthesizer might get its MIDI data from an RTP MIDI stream through a MIDI DIN cable (or a software API that emulates one) 🎛️ In that setup, an RTP-MIDI-aware system receives the RTP MIDI stream and transcodes it onto the MIDI DIN cable (or its emulation). When the cancel sublist arrives, the RTP-MIDI-aware transcoder might already have sent the first part of the SysEx command down the DIN cable to the receiver. Oof 😬
Here's the catch: the MIDI DIN cable protocol can't directly code "cancel SysEx in progress" 🙃 But DIN cable receivers don't start processing SysEx until the complete command arrives. The receiver checks whether it recognizes the command (coded in the first few octets), then checks whether the command has the correct length. So in practice, a transcoder can cancel a SysEx command by sending an 0xF7 to end the SysEx command early. The receiver spots the wrong length and throws the command away ✨
Appendix C.1 defines configuration tools that may be used to prohibit SysEx command cancellation 🔒
The order of SysEx command segments in a MIDI list MUST match the order of the sublists in the original SysEx command 🧠 By default, commands other than System Real-time MIDI commands MUST NOT appear between SysEx command segments. (Appendix C.1 has configuration tools to change this default and let other command types appear between segments.) If the segments of one SysEx command are spread across the MIDI lists of two or more RTP packets, the segment ordering rules apply to the combined result of all the affected MIDI lists 🌸
Here's the table of status octets for each sublist position (Figure 5):
-----------------------------------------------------------
| Sublist Position | Head Status Octet | Tail Status Octet |
|-----------------------------------------------------------|
| first | 0xF0 | 0xF0 |
|-----------------------------------------------------------|
| middle | 0xF7 | 0xF0 |
|-----------------------------------------------------------|
| last | 0xF7 | 0xF7 |
|-----------------------------------------------------------|
| cancel | 0xF7 | 0xF4 |
-----------------------------------------------------------
Figure 5 -- Command segmentation status octets[MIDI] lets 0xF7 octets that aren't part of a (0xF0, 0xF7) pair show up on a MIDI 1.0 DIN cable 🤷♀️ Unpaired 0xF7 octets don't mean anything in MIDI, except that they cancel running status.
Unpaired 0xF7 octets MUST NOT appear in the MIDI list of the MIDI Command section 🚫 We add this rule so they don't mess with the command segmentation coding from Figure 5. Protect the vibe 💖
SysEx commands on a MIDI 1.0 DIN cable may use the "dropped 0xF7" construction from [MIDI] 🎀 With this method, the 0xF7 octet gets dropped from the end of the SysEx command, and the status octet of the next MIDI command does double duty: it ends the SysEx command AND starts the next one. To encode this in the payload format, follow these steps:
- Work out the right delta times for the SysEx command and for the command that comes after it.
- Insert the "dropped" 0xF7 octet at the end of the SysEx command, so it has standard SysEx syntax.
- Code both commands into the MIDI list using the rules above.
- Replace the 0xF7 octet that ends the verbatim SysEx encoding (or the last segment of a segmented SysEx encoding) with a 0xF5 octet. This swap tells the receiver that there was a dropped 0xF7 in the original coding 🍓
[MIDI] reserves the undefined System Common commands 0xF4 and 0xF5, plus the undefined System Real-time commands 0xF9 and 0xFD, for future use 🔮 By default, undefined commands MUST NOT appear in a MIDI Command field in the MIDI list. The exceptions are the 0xF5 octets used for the "dropped 0xF7" construction and the 0xF4 octets used by SysEx "cancel" sublists ✨
During session configuration, a stream can be customized to carry undefined commands (Appendix C.1). For that case, here's how senders encode undefined commands in the MIDI list 🧚
An undefined System Real-time command MUST be coded using the System Real-time rules 💫
If undefined System Common commands get used in a future version of [MIDI], the command will start with an 0xF4 or 0xF5 status octet, followed by any number of data octets (zero or more). To encode these commands, senders MUST end the command with an 0xF7 octet and put the modified command into the MIDI Command field 🌈
Unfortunately, non-compliant uses of the undefined System Common commands can show up in MIDI implementations 😭 To model these, we assume the command starts with an 0xF4 or 0xF5 status octet, then zero or more data octets, then zero or more trailing 0xF7 status octets. To encode it, senders MUST first remove all the trailing 0xF7 status octets from the command. Then senders MUST end the command with an 0xF7 octet and put the modified command into the MIDI Command field 🫶
We include those trailing octets in the model as a precaution 🙈 If such commands showed up in a non-compliant use of an undefined System Common command, an RTP MIDI encoding that didn't strip the trailing octets could be mistaken for a "middle" or "last" sublist of a segmented SysEx command (Figure 5) under certain packet loss conditions. Yikes 💀
Here's a SysEx command and a two-segment version of it:
Original SysEx command:
0xF0 0x01 0x02 0x03 0x04 0x05 0x06 0x07 0x08 0xF7
A two-segment segmentation:
0xF0 0x01 0x02 0x03 0x04 0xF00xF7 0x05 0x06 0x07 0x08 0xF7A different two-segment segmentation:
0xF0 0x01 0xF0
0xF7 0x02 0x03 0x04 0x05 0x06 0x07 0x08 0xF7A three-segment segmentation:
0xF0 0x01 0x02 0xF0
0xF7 0x03 0x04 0xF0
0xF7 0x05 0x06 0x07 0x08 0xF7The segmentation with the largest number of segments:
0xF0 0x01 0xF0
0xF7 0x02 0xF0
0xF7 0x03 0xF0
0xF7 0x04 0xF0
0xF7 0x05 0xF0
0xF7 0x06 0xF0
0xF7 0x07 0xF0
0xF7 0x08 0xF0
0xF7 0xF7Figure 6 -- Example segmentations
4. The Recovery Journal System ✨
okay so the recovery journal is the default resiliency tool for unreliable transport 🦋 bestie, in this section we normatively define the roles that senders and receivers play in the recovery journal system 💅
MIDI is a fragile code 😭 a single lost command in a MIDI command stream may produce an artifact in the rendered performance. we normatively classify rendering artifacts into two categories:
- Transient artifacts. Transient artifacts produce immediate but short-term glitches in the performance. for example, a lost
NoteOn (0x9)command produces a transient artifact: one note fails to play, but the artifact does not extend beyond the end of that note. 🎀 - Indefinite artifacts. Indefinite artifacts produce long-lasting errors in the rendered performance. for example, a lost
NoteOff (0x8)command may produce an indefinite artifact: the note that should have been ended by the lostNoteOffcommand may sustain indefinitely. as a second example, the loss of aControl Change (0xB)command for controller number 7 (Channel Volume) may produce an indefinite artifact: after the loss, all notes on the channel may play too softly or too loudly. 💀
the purpose of the recovery journal system is to satisfy the recovery journal mandate: the MIDI performance rendered from an RTP MIDI stream sent over unreliable transport MUST NOT contain indefinite artifacts. 🚫 that's the whole vibe, no cap 👑
the recovery journal system does not use packet retransmission to satisfy this mandate. instead, each packet includes a special section, called the recovery journal. 🧚
the recovery journal codes the history of the stream, back to an earlier packet called the checkpoint packet. the range of coverage for the journal is called the checkpoint history. the recovery journal codes the information necessary to recover from the loss of an arbitrary number of packets in the checkpoint history. Appendix A.1 normatively defines the checkpoint packet and the checkpoint history. 🌸
when a receiver detects a packet loss, it compares its own knowledge about the history of the stream with the history information coded in the recovery journal of the packet that ends the loss event. by noting the differences in these two versions of the past, a receiver is able to transform all indefinite artifacts in the rendered performance into transient artifacts, by executing MIDI commands to repair the stream. ✨ that's the glow-up moment 🌈
we now state the normative role for senders in the recovery journal system. 💜
senders prepare a recovery journal for every packet in the stream. in doing so, senders choose the checkpoint packet identity for the journal. senders make this choice by applying a sending policy. Appendix C.2.2 normatively defines three sending policies: "closed-loop", "open-loop", and "anchor". 🍓
by default, senders MUST use the closed-loop sending policy. if the session description overrides this default policy, by using the parameter j_update defined in Appendix C.2.2, senders MUST use the specified policy. 🔥
after choosing the checkpoint packet identity for a packet, the sender creates the recovery journal. by default, this journal MUST conform to the normative semantics in Section 5 and Appendices A-B in this memo. Appendix C.2.3 defines parameters that modify the normative semantics for recovery journals. if the session description uses these parameters, the journal created by the sender MUST conform to the modified semantics. 🫶
next, we state the normative role for receivers in the recovery journal system. 🦋
a receiver MUST detect each RTP sequence number break in a stream. if the sequence number break is due to a packet loss event (as defined in [RFC3550]), the receiver MUST repair all indefinite artifacts in the rendered MIDI performance caused by the loss. if the sequence number break is due to an out-of-order packet (as defined in [RFC3550]), the receiver MUST NOT take actions that introduce indefinite artifacts (ignoring the out-of-order packet is a safe option). 🙈
receivers take special precautions when entering or exiting a session. a receiver MUST process the first received packet in a stream as if it were a packet that ends a loss event. upon exiting a session, a receiver MUST ensure that the rendered MIDI performance does not end with indefinite artifacts. 👑
receivers are under no obligation to perform indefinite artifact repairs at the moment a packet arrives. a receiver that uses a playout buffer may choose to wait until the moment of rendering before processing the recovery journal, as the "lost" packet may be a late packet that arrives in time to use. ⭐
next, we state the normative role for the creator of the session description in the recovery journal system. depending on the application, the sender, the receivers, and other parties may take part in creating or approving the session description. 🌈
a session description that specifies the default closed-loop sending policy and the default recovery journal semantics satisfies the recovery journal mandate. however, these default behaviors may not be appropriate for all sessions. if the creators of a session description use the parameters defined in Appendix C.2 to override these defaults, the creators MUST ensure that the parameters define a system that satisfies the recovery journal mandate. 🎀
finally, we note that this memo does not specify sender or receiver recovery journal algorithms. implementations are free to use any algorithm that conforms to the requirements in this section. the non-normative [RFC4696] discusses sender and receiver algorithm design. 💖
5. Recovery Journal Format ✨
okay so this section is where the recovery journal gets its actual shape 🎀 we're meeting the bitfields for each header, and Appendices A-B finish off the rest of the bitfield definitions. no cap, this is the good stuff 💅
The recovery journal has a three-level structure:
- Top-level header.
- Channel and system journal headers. These headers encode recovery information for a single voice channel (channel journal) or for all systems commands (system journal).
- Chapters. Chapters describe recovery information for a single MIDI command type.
Figure 7 shows the top-level structure. the recovery journal is a 3-octet header, then an optional system journal (labeled S-journal in Figure 7), then an optional list of channel journals 🧚 Figure 8 shows the recovery journal header format.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Recovery journal header | S-journal ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Channel journals ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 7 -- Top-level recovery journal format
0 1 2
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S|Y|A|H|TOTCHAN| Checkpoint Packet Seqnum |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 8 -- Recovery journal headerIf the Y header bit is set to 1, the system journal appears in the recovery journal, directly following the recovery journal header.
If the A header bit is set to 1, the recovery journal ends with a list of (TOTCHAN + 1) channel journals (the 4-bit TOTCHAN header field is interpreted as an unsigned integer).
A MIDI channel MAY be represented by (at most) one channel journal in a recovery journal. Channel journals MUST appear in the recovery journal in ascending channel-number order.
If A and Y are both zero, the recovery journal only contains its 3-octet header and is considered to be an "empty" journal.
The S (single-packet loss) bit appears in most recovery journal structures, including the recovery journal header. The S bit helps receivers efficiently parse the recovery journal in the common case of the loss of a single packet. Appendix A.1 defines S bit semantics.
The H bit indicates if MIDI channels in the stream have been configured to use the enhanced Chapter C encoding (Appendix A.3.3).
By default, the payload format does not use enhanced Chapter C encoding. In this default case, the H bit MUST be set to 0 for all packets in the stream.
If the stream has been configured so that controller numbers for one or more MIDI channels use enhanced Chapter C encoding, the H bit MUST be set to 1 in all packets in the stream. In Appendix C.2.3, we show how to configure a stream to use enhanced Chapter C encoding.
The 16-bit Checkpoint Packet Seqnum header field codes the sequence number of the checkpoint packet for this journal, in network byte order (big-endian). The choice of the checkpoint packet sets the depth of the checkpoint history for the journal (defined in Appendix A.1).
Receivers may use the Checkpoint Packet Seqnum field of the packet that ends a loss event to verify that the journal checkpoint history covers the entire loss event. The checkpoint history covers the loss event if the Checkpoint Packet Seqnum field is less than or equal to one plus the highest RTP sequence number previously received on the stream (modulo 2^16).
Figure 9 shows the structure of a channel journal: a 3-octet header, followed by a list of leaf elements called channel chapters. A channel journal encodes information about MIDI commands on the MIDI channel coded by the 4-bit CHAN header field. Note that CHAN uses the same bit encoding as the channel nibble in MIDI Channel Messages (the cccc field in Figure E.1 of Appendix E).
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| CHAN |H| LENGTH |P|C|M|W|N|E|T|A| Chapters ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 9 -- Channel journal formatThe 10-bit LENGTH field codes the length of the channel journal. The semantics for LENGTH fields are uniform throughout the recovery journal, and are defined in Appendix A.1.
The third octet of the channel journal header is the Table of Contents (TOC) of the channel journal. The TOC is a set of bits that encode the presence of a chapter in the journal. Each chapter contains information about a certain class of MIDI channel command:
- Chapter P: MIDI Program Change (
0xC) - Chapter C: MIDI Control Change (
0xB) - Chapter M: MIDI Parameter System (part of
0xB) - Chapter W: MIDI Pitch Wheel (
0xE) - Chapter N: MIDI NoteOff (
0x8), NoteOn (0x9) - Chapter E: MIDI Note Command Extras (
0x8,0x9) - Chapter T: MIDI Channel Aftertouch (
0xD) - Chapter A: MIDI Poly Aftertouch (
0xA)
Chapters appear in a list following the header, in order of their appearance in the TOC. Appendices A.2-9 describe the bitfield format for each chapter, and define the conditions under which a chapter type MUST appear in the recovery journal. If any chapter types are required for a channel, an associated channel journal MUST appear in the recovery journal.
The H bit indicates if controller numbers on a MIDI channel have been configured to use the enhanced Chapter C encoding (Appendix A.3.3).
By default, controller numbers on a MIDI channel do not use enhanced Chapter C encoding. In this default case, the H bit MUST be set to 0 for all channel journal headers for the channel in the recovery journal, for all packets in the stream.
However, if at least one controller number for a MIDI channel has been configured to use the enhanced Chapter C encoding, the H bit for its channel journal MUST be set to 1, for all packets in the stream.
In Appendix C.2.3, we show how to configure a controller number to use enhanced Chapter C encoding.
Figure 10 shows the structure of the system journal: a 2-octet header, followed by a list of system chapters. Each chapter codes information about a specific class of MIDI Systems command:
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S|D|V|Q|F|X| LENGTH | System chapters ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 10 -- System journal format- Chapter D: Song Select (
0xF3), Tune Request (0xF6), Reset (0xFF), undefined System commands (0xF4,0xF5,0xF9,0xFD) - Chapter V: Active Sense (
0xFE) - Chapter Q: Sequencer State (
0xF2,0xF8,0xF9,0xFA,0xFB,0xFC) - Chapter F: MTC Tape Position (
0xF1,0xF0 0x7F 0xcc 0x01 0x01) - Chapter X: System Exclusive (all other
0xF0)
The 10-bit LENGTH field codes the size of the system journal and conforms to semantics described in Appendix A.1.
The D, V, Q, F, and X header bits form a Table of Contents (TOC) for the system journal. A TOC bit that is set to 1 codes the presence of a chapter in the journal. Chapters appear in a list following the header, in the order of their appearance in the TOC.
Appendix B describes the bitfield format for the system chapters and defines the conditions under which a chapter type MUST appear in the recovery journal. If any system chapter type is required to appear in the recovery journal, the system journal MUST appear in the recovery journal.
6. Session Description Protocol 🦋
RTP does not perform session management. Instead, RTP works together with session management tools, such as the Session Initiation Protocol (SIP, [RFC3261]) and the Real Time Streaming Protocol (RTSP, [RFC2326]).
RTP payload formats define media type parameters for use in session management (for example, this memo defines "rtp-midi" as the media type for native RTP MIDI streams).
In most cases, session management tools use the media type parameters via another standard, the Session Description Protocol (SDP, [RFC4566]).
SDP is a textual format for specifying session descriptions. Session descriptions specify the network transport and media encoding for RTP sessions. Session management tools coordinate the exchange of session descriptions between participants ("parties").
Some session management tools use SDP to negotiate details of media transport (network addresses, ports, etc.). We refer to this use of SDP as "negotiated usage". One example of negotiated usage is the Offer/Answer protocol ([RFC3264] and Appendix C.7.2 in this memo) as used by SIP.
Other session management tools use SDP to declare the media encoding for the session but use other techniques to negotiate network transport. We refer to this use of SDP as "declarative usage". One example of declarative usage is RTSP ([RFC2326] and Appendix C.7.1 in this memo).
Below, we show session description examples for native (Section 6.1) and mpeg4-generic (Section 6.2) streams. In Section 6.3, we introduce session configuration tools that may be used to customize streams.
6.1. Session Descriptions for Native Streams ✨
okay bestie, this session description sets up a unicast UDP RTP session (through an "m=" line) and its one and only payload type, 96, gets mapped to a minimal native RTP MIDI stream 🎀
v=0
o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
s=Example
t=0 0
m=audio 5004 RTP/AVP 96
c=IN IP4 192.0.2.94
a=rtpmap:96 rtp-midi/44100The rtpmap line uses the "rtp-midi" media type to say "this is a native RTP MIDI stream" 💜. The clock rate on that line (44100 Hz in the example) sets the scaling for the RTP timestamp header field ⏰ (see Section 2.1, and also [RFC3550]).
Heads up, this document doesn't give RTP MIDI a default clock rate 🙈. So when RTP MIDI is used with SDP, parties MUST use the rtpmap line to communicate the clock rate. You can find guidance for picking that value in Section 2.1 ✨.
The stream above is "minimal" because the session description doesn't customize it with any parameters 💅. Without that customization, a native RTP MIDI stream gives you these vibes:
- If the stream uses unreliable transport (unicast UDP, multicast UDP, etc.), the recovery journal system is in use, and the RTP payload contains both the MIDI command section and the journal section. If the stream uses reliable transport (such as TCP), the stream does not use journalling, and the payload contains only the MIDI command section (Section 2.2). 🌸
- If the stream uses the recovery journal system, the recovery journal system uses the default sending policy and the default journal semantics (Section 4). 🦋
- In the MIDI command section of the payload, command timestamps use the default "comex" semantics (Section 3). ⭐
- The recommended temporal duration ("media time") of an RTP packet ranges from 0 to 200 ms, and the RTP timestamp difference between sequential packets in the stream may be arbitrarily large (Section 2.1). 🧚
- If more than one minimal rtp-midi stream appears in a session, the MIDI name spaces for these streams are independent: channel 1 in the first stream does not reference the same MIDI channel as channel 1 in the second stream (see Appendix C.5 for a discussion of the independence of minimal rtp-midi streams). 🍓
- The rendering method for the stream is not specified. What the receiver "does" with a minimal native MIDI stream is "out of scope" of this memo. For example, in content creation environments, a user may manually configure client software to render the stream with a specific software package. 🎧
Like always in RTP, RTP sessions managed by SIP are sendrecv by default (parties send and receive MIDI), and RTP sessions managed by RTSP are recvonly by default (server sends and client receives) 📞.
In sendrecv RTP MIDI sessions for the session description shown above, the 16 voice channel + systems MIDI name space is unique for each sender. So in a two-party session, the voice channel 0 sent by one party is distinct from the voice channel 0 sent by the other party 🫶.
This behavior matches what happens when two MIDI 1.0 DIN devices are cross-connected with two MIDI cables (one cable routing MIDI Out from the first device into MIDI In of the second device, and a second cable routing MIDI In from the first device into MIDI Out of the second device) 🔌. We define this "association" formally in Section 2.1.
MIDI 1.0 DIN networks may be configured in a "party-line" multicast topology 📻. For these networks, the MIDI protocol itself provides tools for addressing specific devices in transactions on a multicast network, and for device discovery. So apart from providing a 1-to-many forward path and a many-to-1 reverse path, IETF protocols don't need to provide any special support for MIDI multicast networking 💜.
6.2. Session Descriptions for mpeg4-generic Streams 🎀
An mpeg4-generic [RFC3640] RTP MIDI stream uses an MPEG 4 Audio Object Type to render MIDI into audio 🎹. Three Audio Object Types accept MIDI input:
- General MIDI (Audio Object Type ID 15), based on the General MIDI rendering standard [MIDI].
- Wavetable Synthesis (Audio Object Type ID 14), based on the Downloadable Sounds Level 2 (DLS 2) rendering standard [DLS2].
- Main Synthetic (Audio Object Type ID 13), based on Structured Audio and the programming language SAOL [MPEGSA].
The primary service of an mpeg4-generic stream is to code Access Units (AUs) 🧠. We define the mpeg4-generic RTP MIDI AU as the MIDI payload shown in Figure 1 of Section 2.1 of this memo: a MIDI command section, optionally followed by a journal section.
Exactly one RTP MIDI AU MUST be mapped to one mpeg4-generic RTP MIDI packet 📦. The mpeg4-generic options for placing several AUs in an RTP packet MUST NOT be used with RTP MIDI 🙅♀️. The mpeg4-generic options for fragmenting and interleaving AUs MUST NOT be used with RTP MIDI either 🙅♀️. The mpeg4-generic RTP packet payload (Figure 1 in [RFC3640]) MUST contain empty AU Header and Auxiliary sections. These rules yield mpeg4-generic packets that are structurally identical to native RTP MIDI packets, which is an essential property for the correct operation of the payload format 💖.
The session description that follows defines a unicast UDP RTP session (via a media ("m=") line) whose sole payload type (96) is mapped to a minimal mpeg4-generic RTP MIDI stream. This example uses the General MIDI Audio Object Type under Synthesis Profile @ Level 2 🌈.
v=0
o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
s=Example
t=0 0
m=audio 5004 RTP/AVP 96
c=IN IP6 2001:DB80::7F2E:172A:1E24
a=rtpmap:96 mpeg4-generic/44100
a=fmtp:96 streamtype=5; mode=rtp-midi; profile-level-id=12;
config=7A0A0000001A4D546864000000060000000100604D54726B0000
000600FF2F000(Heads up: the a=fmtp line got wrapped so it fits the page with the memo's formatting limits. In real SDP it's one single line 📝.)
The fmtp attribute line codes the four parameters (streamtype, mode, profile-level-id, and config) that are required in all mpeg4-generic session descriptions [RFC3640] 🔧. For RTP MIDI streams, the streamtype parameter MUST be set to 5, the "mode" parameter MUST be set to "rtp-midi", and the "profile-level-id" parameter MUST be set to the MPEG-4 Profile Level for the stream 👑. For the Synthesis Profile, legal profile-level-id values are 11, 12, and 13, coding low (11), medium (12), or high (13) decoder computational complexity, as defined by MPEG conformance tests 📊.
In a minimal RTP MIDI session description, the config value MUST be a hexadecimal encoding [RFC3640] of the AudioSpecificConfig data block [MPEGAUDIO] for the stream 🔢. AudioSpecificConfig encodes the Audio Object Type for the stream, and it also encodes initialization data (SAOL programs, DLS 2 wave tables, etc.) 🎶. Standard MIDI Files encoded in AudioSpecificConfig in a minimal session description MUST be ignored by the receiver 🙈.
Receivers figure out the rendering algorithm for the session by interpreting the first 5 bits of AudioSpecificConfig as an unsigned integer that codes the Audio Object Type 🔍. In our example above, the leading config string nibbles "7A" yield the Audio Object Type 15 (General MIDI). In Appendix E.4, we derive the config string value in the session description shown above. The starting point of that derivation is the MPEG bitstreams defined in [MPEGSA] and [MPEGAUDIO] 🧩.
We consider the stream to be "minimal" because the session description doesn't customize the stream through parameters, other than the 4 required mpeg4-generic parameters described above 💅. In Section 6.1, we describe the behavior of a minimal native stream as a numbered list of characteristics. Items 1-4 on that list also describe the minimal mpeg4-generic stream, but items 5 and 6 need restatements, as listed below:
- If more than one minimal mpeg4-generic stream appears in a session, each stream uses an independent instance of the Audio Object Type coded in the config parameter value. ✨
- A minimal mpeg4-generic stream encodes the AudioSpecificConfig as an inline hexadecimal constant. If a session description is sent over UDP, it may be impossible to transport large AudioSpecificConfig blocks within the Maximum Transmission Size (MTU) of the underlying network (for Ethernet, the MTU is 1500 octets). In some cases, the AudioSpecificConfig block may exceed the maximum size of the UDP packet itself. 😭
The comments in Section 6.1 on SIP and RTSP stream directional defaults, sendrecv MIDI channel usage, and MIDI 1.0 DIN multicast networks also apply to mpeg4-generic RTP MIDI sessions 🔁.
In sendrecv sessions, each party's session description MUST use identical values for the mpeg4-generic parameters (including the required streamtype, mode, profile-level-id, and config parameters). As a consequence, each party uses an identically configured MPEG 4 Audio Object Type to render MIDI commands into audio. The preamble to Appendix C discusses a way to create "virtual sendrecv" sessions that do not have this restriction 🚨.
6.3. Parameters ✨
okay so this section is all about parameters, the knobs you turn to configure an RTP MIDI session 🎀 they live on an fmtp attribute line, and the session description example in Section 6.2 shows one in action 💅 in session descriptions, parameters tweak what a payload type actually means ✨
the parameters build on the minimal streams from Sections 6.1-2 and support a few kinds of services 🦋
- Stream subsetting 🌸 by default, every MIDI command that's legal on a MIDI 1.0 DIN cable can show up in an RTP MIDI stream. the
cm_unusedparameter overrides that by banning certain commands from the stream 🙅 andcm_usedworks together withcm_unusedso complex exclusion rules are easier to write. we go over both in Appendix C.1. - Journal customization 📓
j_secandj_updateconfigure the journal section, whilech_default,ch_never, andch_anchorconfigure the semantics of the recovery journal chapters. these are in Appendix C.2, and they override default stream behaviors 1 and 2, which are listed in Section 6.1 and referenced in Section 6.2 💖 - MIDI command timestamp semantics ⏰
tsmode,octpos,mperiod, andlineratecustomize the semantics of timestamps in the MIDI command section. they let RTP MIDI accurately encode the implicit time coding of MIDI 1.0 DIN cables (so precise 😭). see Appendix C.3. they override default stream behavior 3, listed in Section 6.1 and referenced in Section 6.2. - Media time 🎵
rtp_ptimeandrtp_maxptimedefine the temporal duration ("media time") of an RTP MIDI packet, andguardtimesets the minimum sending rate of stream packets. these are described in Appendix C.4, and they override default stream behavior 4, listed in Section 6.1 and referenced in Section 6.2. - Stream description 🏷️
musicportlabels the MIDI name space of RTP streams in a multimedia session. it's covered in Appendix C.5 and overrides default stream behavior 5, in Sections 6.1 and 6.2. - MIDI rendering 🎹 several parameters specify the MIDI rendering method of a stream. see Appendix C.6. they override default stream behavior 6, in Sections 6.1 and 6.2.
in Appendix C.7 we spell out interop guidelines for two RTP MIDI app areas: content-streaming using RTSP (Appendix C.7.1) and network musical performance using SIP (Appendix C.7.2) 🌈
7. Extensibility ✨
the payload format in this memo exclusively encodes all commands that may legally appear on a MIDI 1.0 DIN cable 🎹
but honestly, lots of worthy uses of MIDI over RTP fall outside this narrow scope 🙈 for example, the payload format doesn't carry the direct transport of Standard MIDI File (SMF) meta-event and metric timing data. as a second example, it doesn't define transport tools for user-defined commands (apart from tools to support System Exclusive commands [MIDI]).
the payload format does not provide an extension mechanism for features like these, by design. instead, we encourage new payload formats for specialized musical applications 🦋 the IETF session management tools [RFC3264] [RFC2326] support codec negotiation, so new payload formats can be used in a backward-compatible way 💜
that said, the payload format does come with several extensibility tools, which we list below:
- Journalling 📔 as described in Appendix C.2, new token values for the
j_secandj_updateparameters may be defined in IETF standards-track documents. this supports the design of new journal formats and new journal sending policies. - Rendering 🎶 the payload format may be extended to support new MIDI renderers (Appendix C.6.2). certain general aspects of the RTP MIDI rendering process may also be extended, via new token values for the
render(Appendix C.6) andsmf_info(Appendix C.6.4.1) parameters. - Undefined commands 🔮 [MIDI] reserves 4 MIDI System commands for future use (
0xF4,0xF5,0xF9,0xFD). if updates to [MIDI] define the reserved commands, IETF standards-track documents may be defined to provide resiliency support for them.
Opaque LEGAL fields appear in System Chapter D for this purpose (Appendix B.1.1) 🫶
a final form of extensibility involves including the payload format in framework documents 🧚 framework documents describe how to combine protocols into a platform for interoperable applications. for example, a stage and studio framework might define how to use SIP [RFC3261], RTSP [RFC2326], SDP [RFC4566], and RTP [RFC3550] to support media networking for professional audio equipment and electronic musical instruments 🎤
8. Congestion Control ✨
the RTP congestion control requirements defined in [RFC3550] apply to RTP MIDI sessions, and implementors should carefully read the congestion control section in [RFC3550]. as noted in [RFC3550], all transport protocols used on the Internet need to address congestion control in some way, and RTP is not an exception 🚦
in addition, the congestion control requirements defined in [RFC3551] apply to RTP MIDI sessions run under applicable profiles. the basic requirement from [RFC3551] is that RTP sessions using UDP transport should monitor packet loss (via RTCP or other means) so that the RTP stream competes fairly with TCP flows sharing the network 🫶
finally, RTP MIDI has congestion control issues that are unique for an audio RTP payload format 😤 in applications such as network musical performance [NMP], the packet rate is linked to the gestural rate of a human performer. senders MUST monitor the MIDI command source for patterns that result in excessive packet rates, and take actions during RTP transcoding to reduce the RTP packet rate. [RFC4696] offers implementation guidance on this issue 📈
9. Security Considerations 🔐
this part is SO important, bestie, so read it carefully 💀 implementors should carefully read the Security Considerations sections of the RTP [RFC3550], AVP [RFC3551], and other RTP profile documents, because the issues discussed there directly apply to RTP MIDI streams. implementors should also review the Secure Real-time Transport Protocol (SRTP, [RFC3711]), an RTP profile that addresses the security issues discussed in [RFC3550] and [RFC3551] 🛡️
here, we discuss the security issues that are unique to the RTP MIDI payload format 👀
when using RTP MIDI, authentication of incoming RTP and RTCP packets is RECOMMENDED. per-packet authentication may be provided by SRTP or by other means. without authentication, attackers could forge MIDI commands into an ongoing stream, damaging speakers and eardrums 😱 an attacker could also craft RTP and RTCP packets to exploit known bugs in the client and take effective control of a client machine. literally unhinged behavior, not okay 💀
session management tools (such as SIP [RFC3261]) SHOULD use authentication during the transport of all session descriptions containing RTP MIDI media streams. for SIP, the Security Considerations section in [RFC3261] gives an overview of possible authentication mechanisms 🔑 RTP MIDI session descriptions should use authentication because they may code initialization data using the parameters described in Appendix C. if an attacker inserts bogus initialization data into a session description, he can corrupt the session or forge an client attack 🚨
session descriptions may also code renderer initialization data by reference, via the url (Appendix C.6.3) and smf_url (Appendix C.6.4.2) parameters. if the coded URL is spoofed, both the session and the client are open to attack, even if the session description itself is authenticated. therefore, URLs specified in url and smf_url parameters SHOULD use [RFC2818] 🔒
Section 2.1 allows streams sent by a party in two RTP sessions to have the same SSRC value and the same RTP timestamp initialization value, under certain circumstances. normally, these values are randomly chosen for each stream in a session, to make plaintext guessing harder if the payloads are encrypted. thus, Section 2.1 weakens this aspect of RTP security 🥀
10. Acknowledgements 💖
we thank the networking, media compression, and computer music community members who have commented on or contributed to the effort, including Kurt B, Cynthia Bruyns, Steve Casner, Paul Davis, Robin Davies, Joanne Dow, Tobias Erichsen, Nicolas Falquet, Dominique Fober, Philippe Gentric, Michael Godfrey, Chris Grigg, Todd Hager, Michel Jullian, Phil Kerr, Young-Kwon Lim, Jessica Little, Jan van der Meer, Colin Perkins, Charlie Richmond, Herbie Robinson, Larry Rowe, Eric Scheirer, Dave Singer, Martijn Sipkema, William Stewart, Kent Terry, Magnus Westerlund, Tom White, Jim Wright, Doug Wyatt, and Giorgio Zoia 🙌 we also thank the members of the San Francisco Bay Area music and audio community for creating the context for the work, including Don Buchla, Chris Chafe, Richard Duda, Dan Ellis, Adrian Freed, Ben Gold, Jaron Lanier, Roger Linn, Richard Lyon, Dana Massie, Max Mathews, Keith McMillen, Carver Mead, Nelson Morgan, Tom Oberheim, Malcolm Slaney, Dave Smith, Julius Smith, David Wessel, and Matt Wright 🌟
11. IANA Considerations 📋
this section makes a series of requests to IANA. IANA has completed the registration and assignments for the requests below ✅
the subsections that follow hold the actual, detailed requests. all registrations in this section are in the IETF tree and follow the rules of [RFC4288] and [RFC3555], as appropriate 🗂️
in Section 11.1, we request the registration of a new media type: "audio/rtp-midi". paired with this request is a request for a repository for new values for several parameters associated with "audio/rtp-midi". we request this repository in Section 11.1.1 🎀
in Section 11.2, we request the registration of a new value ("rtp-midi") for the "mode" parameter of the "mpeg4-generic" media type. the "mpeg4-generic" media type is defined in [RFC3640], and [RFC3640] defines a repository for the "mode" parameter. however, we believe we are the first to request registration of a "mode" value, so we believe the registry for "mode" has not yet been created by IANA 🤔
paired with our "mode" parameter value request for "mpeg4-generic" is a request for a repository for new values for several parameters we have defined for use with the "rtp-midi" mode value. we request this repository in Section 11.2.1 💅
in Section 11.3, we request the registration of a new media type: "audio/asc". no repository request is associated with this request 🦋
11.1. rtp-midi Media Type Registration 🎵
Okay bestie, so this section asks IANA to register the rtp-midi subtype under the audio media type 🎀. We want every parameter from the optional parameters list below registered, both the non-extensible ones and the extensible ones. We also want repositories created for the extensible parameters, and those details are in Section 11.1.1, below ✨
Here's the whole registration template, copied exactly as written:
Media type name:
audio
Subtype name:
rtp-midi
Required parameters:
rate: The RTP timestamp clock rate. See Sections 2.1 and 6.1
for usage details.
Optional parameters:
Non-extensible parameters:
ch_anchor: See Appendix C.2.3 for usage details.
ch_default: See Appendix C.2.3 for usage details.
ch_never: See Appendix C.2.3 for usage details.
cm_unused: See Appendix C.1 for usage details.
cm_used: See Appendix C.1 for usage details.
chanmask: See Appendix C.6.4.3 for usage details.
cid: See Appendix C.6.3 for usage details.
guardtime: See Appendix C.4.2 for usage details.
inline: See Appendix C.6.3 for usage details.
linerate: See Appendix C.3 for usage details.
mperiod: See Appendix C.3 for usage details.
multimode: See Appendix C.6.1 for usage details.
musicport: See Appendix C.5 for usage details.
octpos: See Appendix C.3 for usage details.
rinit: See Appendix C.6.3 for usage details.
rtp_maxptime: See Appendix C.4.1 for usage details.
rtp_ptime: See Appendix C.4.1 for usage details.
smf_cid: See Appendix C.6.4.2 for usage details.
smf_inline: See Appendix C.6.4.2 for usage details.
smf_url: See Appendix C.6.4.2 for usage details.
tsmode: See Appendix C.3 for usage details.
url: See Appendix C.6.3 for usage details.
Extensible parameters:
j_sec: See Appendix C.2.1 for usage details. See
Section 11.1.1 for repository details.
j_update: See Appendix C.2.2 for usage details. See
Section 11.1.1 for repository details.
render: See Appendix C.6 for usage details. See
Section 11.1.1 for repository details.
subrender: See Appendix C.6.2 for usage details. See
Section 11.1.1 for repository details.
smf_info: See Appendix C.6.4.1 for usage details. See
Section 11.1.1 for repository details.
Encoding considerations:
The format for this type is framed and binary.
Restrictions on usage:
This type is only defined for real-time transfers of MIDI
streams via RTP. Stored-file semantics for rtp-midi may
be defined in the future.
Security considerations:
See Section 9 of this memo.
Interoperability considerations:
None.
Published specification:
This memo and [MIDI] serve as the normative specification. In
addition, references [NMP], [GRAME], and [RFC4696] provide
non-normative implementation guidance.
Applications that use this media type:
Audio content-creation hardware, such as MIDI controller piano
keyboards and MIDI audio synthesizers. Audio content-creation
software, such as music sequencers, digital audio workstations,
and soft synthesizers. Computer operating systems, for network
support of MIDI Application Programmer Interfaces. Content
distribution servers and terminals may use this media type for
low bit-rate music coding.
Additional information:
None.
Person & email address to contact for further information:
John Lazzaro <lazzaro@cs.berkeley.edu>
Intended usage:
COMMON.
Author:
John Lazzaro <lazzaro@cs.berkeley.edu>
Change controller:
IETF Audio/Video Transport Working Group delegated
from the IESG.11.1.1. Repository Request for "audio/rtp-midi"
Next up are the repositories for the rtp-midi extensible parameters. These are the ones listed as "extensible parameters" in Section 11.1 🌸 Every repository below has its own rules, so read each one carefully 💅
j_sec:
Registrations for this repository may only occur
via an IETF standards-track document. Appendix C.2.1
of this memo describes appropriate registrations for this
repository.
Initial values for this repository appear below:
"none": Defined in Appendix C.2.1 of this memo.
"recj": Defined in Appendix C.2.1 of this memo.j_update:
Registrations for this repository may only occur
via an IETF standards-track document. Appendix C.2.2
of this memo describes appropriate registrations for this
repository.
Initial values for this repository appear below:
"anchor": Defined in Appendix C.2.2 of this memo.
"open-loop": Defined in Appendix C.2.2 of this memo.
"closed-loop": Defined in Appendix C.2.2 of this memo.render:
Registrations for this repository MUST include a
specification of the usage of the proposed value.
See text in the preamble of Appendix C.6 for details
(the paragraph that begins "Other render token ...").
Initial values for this repository appear below:
"unknown": Defined in Appendix C.6 of this memo.
"synthetic": Defined in Appendix C.6 of this memo.
"api": Defined in Appendix C.6 of this memo.
"null": Defined in Appendix C.6 of this memo.subrender:
Registrations for this repository MUST include a
specification of the usage of the proposed value.
See text Appendix C.6.2 for details (the paragraph
that begins "Other subrender token ...").
Initial values for this repository appear below:
"default": Defined in Appendix C.6.2 of this memo.smf_info:
Registrations for this repository MUST include a
specification of the usage of the proposed value.
See text in Appendix C.6.4.1 for details (the
paragraph that begins "Other smf_info token ...").
Initial values for this repository appear below:
"ignore": Defined in Appendix C.6.4.1 of this memo.
"sdp_start": Defined in Appendix C.6.4.1 of this memo.
"identity": Defined in Appendix C.6.4.1 of this memo.11.2. mpeg4-generic Media Type Registration 🦋
This section asks for the rtp-midi value to be registered for the mode parameter of the mpeg4-generic media type. The mpeg4-generic media type is defined in [RFC3640], and [RFC3640] defines a repository for the mode parameter. We're registering mode rtp-midi to support the MPEG Audio codecs [MPEGSA] that use MIDI 🎶
Along with that, we also want the parameters from the optional parameters list below registered (both the non-extensible and extensible lists), plus repositories for the extensible ones. Those details are in Appendix 11.2.1, below 💖
Here's the full template, exactly as written:
Media type name:
audio
Subtype name:
mpeg4-generic
Required parameters:
The "mode" parameter is required by [RFC3640]. [RFC3640]
requests a repository for "mode", so that new values for mode
may be added. We request that the value "rtp-midi" be
added to the "mode" repository.
In mode rtp-midi, the mpeg4-generic parameter rate is
a required parameter. Rate specifies the RTP timestamp
clock rate. See Sections 2.1 and 6.2 for usage details
of rate in mode rtp-midi.
Optional parameters:
We request registration of the following parameters
for use in mode rtp-midi for mpeg4-generic.
Non-extensible parameters:
ch_anchor: See Appendix C.2.3 for usage details.
ch_default: See Appendix C.2.3 for usage details.
ch_never: See Appendix C.2.3 for usage details.
cm_unused: See Appendix C.1 for usage details.
cm_used: See Appendix C.1 for usage details.
chanmask: See Appendix C.6.4.3 for usage details.
cid: See Appendix C.6.3 for usage details.
guardtime: See Appendix C.4.2 for usage details.
inline: See Appendix C.6.3 for usage details.
linerate: See Appendix C.3 for usage details.
mperiod: See Appendix C.3 for usage details.
multimode: See Appendix C.6.1 for usage details.
musicport: See Appendix C.5 for usage details.
octpos: See Appendix C.3 for usage details.
rinit: See Appendix C.6.3 for usage details.
rtp_maxptime: See Appendix C.4.1 for usage details.
rtp_ptime: See Appendix C.4.1 for usage details.
smf_cid: See Appendix C.6.4.2 for usage details.
smf_inline: See Appendix C.6.4.2 for usage details.
smf_url: See Appendix C.6.4.2 for usage details.
tsmode: See Appendix C.3 for usage details.
url: See Appendix C.6.3 for usage details.
Extensible parameters:
j_sec: See Appendix C.2.1 for usage details. See
Section 11.2.1 for repository details.
j_update: See Appendix C.2.2 for usage details. See
Section 11.2.1 for repository details.
render: See Appendix C.6 for usage details. See
Section 11.2.1 for repository details.
subrender: See Appendix C.6.2 for usage details. See
Section 11.2.1 for repository details.
smf_info: See Appendix C.6.4.1 for usage details. See
Section 11.2.1 for repository details.
Encoding considerations:
The format for this type is framed and binary.
Restrictions on usage:
Only defined for real-time transfers of audio/mpeg4-generic
RTP streams with mode=rtp-midi.
Security considerations:
See Section 9 of this memo.
Interoperability considerations:
Except for the marker bit (Section 2.1), the packet formats
for audio/rtp-midi and audio/mpeg4-generic (mode rtp-midi)
are identical. The formats differ in use: audio/mpeg4-generic
is for MPEG work, and audio/rtp-midi is for all other work.
Published specification:
This memo, [MIDI], and [MPEGSA] are the normative references.
In addition, references [NMP], [GRAME], and [RFC4696] provide
non-normative implementation guidance.
Applications that use this media type:
MPEG 4 servers and terminals that support [MPEGSA].
Additional information:
None.
Person & email address to contact for further information:
John Lazzaro <lazzaro@cs.berkeley.edu>
Intended usage:
COMMON.
Author:
John Lazzaro <lazzaro@cs.berkeley.edu>
Change controller:
IETF Audio/Video Transport Working Group delegated
from the IESG.11.2.1. Repository Request for Mode rtp-midi for mpeg4-generic ✨
okay so for mode rtp-midi of the mpeg4-generic subtype, we're requesting the creation of repositories for extensions to the following parameters 💅. these are the ones listed as "extensible parameters" in Section 11.2, so no mystery there 🫶
j_sec- Registrations for this repository may only occur via an IETF standards-track document. Appendix C.2.1 of this memo describes appropriate registrations for this repository.
- Initial values for this repository appear below:
"none": Defined in Appendix C.2.1 of this memo."recj": Defined in Appendix C.2.1 of this memo.
j_update- Registrations for this repository may only occur via an IETF standards-track document. Appendix C.2.2 of this memo describes appropriate registrations for this repository.
- Initial values for this repository appear below:
"anchor": Defined in Appendix C.2.2 of this memo."open-loop": Defined in Appendix C.2.2 of this memo."closed-loop": Defined in Appendix C.2.2 of this memo.
render- Registrations for this repository MUST include a specification of the usage of the proposed value. See text in the preamble of Appendix C.6 for details (the paragraph that begins "Other render token ...").
- Initial values for this repository appear below:
"unknown": Defined in Appendix C.6 of this memo."synthetic": Defined in Appendix C.6 of this memo."null": Defined in Appendix C.6 of this memo.
subrender- Registrations for this repository MUST include a specification of the usage of the proposed value. See text Appendix C.6.2 for details (the paragraph that begins "Other subrender token ..." and subsequent paragraphs). Note that the text in Appendix C.6.2 contains restrictions on subrender registrations for mpeg4-generic ("Registrations for mpeg4-generic subrender values ..."). 🙈
- Initial values for this repository appear below:
"default": Defined in Appendix C.6.2 of this memo.
smf_info- Registrations for this repository MUST include a specification of the usage of the proposed value. See text in Appendix C.6.4.1 for details (the paragraph that begins "Other smf_info token ...").
- Initial values for this repository appear below:
"ignore": Defined in Appendix C.6.4.1 of this memo."sdp_start": Defined in Appendix C.6.4.1 of this memo."identity": Defined in Appendix C.6.4.1 of this memo.
11.3. asc Media Type Registration 🎀
okay this section registers "asc" as a subtype for the "audio" media type 🦋. we're doing it so the remote transfer of the "config" parameter of the mpeg4-generic media type [RFC3640] works when it's used with mpeg4-generic mode rtp-midi (registered in Appendix 11.2 above). the mechanics of using "audio/asc" to set the config parameter are explained in Section 6.2 and Appendix C.6.5 of this document.
ngl this is a brand new subtype registration, NOT an addition to a repository defined by MPEG-related memos (like [RFC3640]) 🌸. also, this request for "audio/asc" does not register parameters and does not request the creation of a repository, so it's a pretty clean lil ask.
Media type name:
audio
Subtype name:
asc
Required parameters:
None.
Optional parameters:
None.
Encoding considerations:
The native form of the data object is binary data,
zero-padded to an octet boundary.
Restrictions on usage:
This type is only defined for data object (stored file)
transfer. The most common transports for the type are
HTTP and SMTP.
Security considerations:
See Section 9 of this memo.
Interoperability considerations:
None.
Published specification:
The audio/asc data object is the AudioSpecificConfig
binary data structure, which is normatively defined in
[MPEGAUDIO].
Applications that use this media type:
MPEG 4 Audio servers and terminals that support
audio/mpeg4-generic RTP streams for mode rtp-midi.
Additional information:
None.
Person & email address to contact for further information:
John Lazzaro <lazzaro@cs.berkeley.edu>
Intended usage:
COMMON.
Author:
John Lazzaro <lazzaro@cs.berkeley.edu>
Change controller:
IETF Audio/Video Transport Working Group delegated
from the IESG.A. The Recovery Journal Channel Chapters 🌈
A.1. Recovery Journal Definitions 💖
this appendix sets up the terminology and coding idioms used in the recovery journal bitfield descriptions in Section 5 (journal header structure), Appendices A.2 to A.9 (channel journal chapters) and Appendices B.1 to B.5 (system journal chapters). it's the glossary bestie, read this first 📚.
we assume the recovery journal lives in the journal section of an RTP packet with sequence number I ("packet I"), and that the Checkpoint Packet Seqnum field in the top-level recovery journal header refers to a previous packet with sequence number C (the self-referential case C = I is the one exception 👑). unless it's stated otherwise, the algorithms use modulo 2^16 arithmetic for calculations on 16-bit sequence numbers, and modulo 2^32 arithmetic for calculations on 32-bit extended sequence numbers ✨.
a bunch of bitfield coding idioms show up all over the recovery journal system, and they always mean the same thing 🍓. most recovery journal elements start with an "S" (Single-packet loss) bit. S bits exist so receivers can parse through the recovery journal hierarchy efficiently in the common case of losing a single packet, which is so thoughtful 🥹.
As a rule, S bits MUST be set to 1. However, there's an exception: if a recovery journal element in packet I encodes data about a command stored in the MIDI command section of packet I - 1, the S bit of that recovery journal element MUST be set to 0. And if a recovery journal element has its S bit set to 0, then all higher-level recovery journal elements that contain it MUST also have S bits set to 0, including the top-level recovery journal header 💀.
other consistent bitfield idioms are below:
- R flag bit. R flag bits are reserved for future use. Senders MUST set R bits to 0. Receivers MUST ignore R bit values 🙃.
- LENGTH field. All fields named LENGTH (as distinct from LEN) code the number of octets in the structure that contains them, including the header they reside in and all hierarchical levels below them. If a structure contains a LENGTH field, a receiver MUST use the LENGTH field value to advance past the structure during parsing, rather than using knowledge about the internal format of the structure 🎯.
now we define the normative terms used to describe recovery journal semantics. these are the vibes that matter 🫶.
- Checkpoint history. The checkpoint history of a recovery journal is the concatenation of the MIDI command sections of packets C through I - 1. The final command in the MIDI command section for packet I - 1 is considered the most recent command, and the first command in the MIDI command section for packet C is the oldest command. If command X is less recent than command Y, X is considered to be "before Y". A checkpoint history with no commands is considered to be empty. The checkpoint history never contains the MIDI command section of packet I (the packet containing the recovery journal), so if
C == I, the checkpoint history is empty by definition 🧚. - Session history. The session history of a recovery journal is the concatenation of MIDI command sections from the first packet of the session up to packet I - 1. The definitions of command recency and history emptiness follow those in the checkpoint history. The session history never contains the MIDI command section of packet I, so the session history of the first packet in the session is empty by definition 🌙.
- Finished/unfinished commands. If all octets of a MIDI command appear in the session history, the command is defined as being finished. If some but not all octets of a command appear in the session history, the command is defined as being unfinished. Unfinished commands occur if segments of a SysEx command appear in several RTP packets. For example, if a SysEx command is coded as 3 segments, with segment 1 in packet K, segment 2 in packet K + 1, and segment 3 in packet K + 2, the session histories for packets K + 1 and K + 2 contain unfinished versions of the command. A session history contains a finished version of a cancelled SysEx command if the history contains the cancel sublist for the command 🧩.
- Reset State commands. Reset State (RS) commands reset renderers to an initialized "powerup" condition 🔄. The RS commands are: System Reset (
0xFF), General MIDI System Enable (0xF0 0x7E 0xcc 0x09 0x01 0xF7), General MIDI 2 System Enable (0xF0 0x7E 0xcc 0x09 0x03 0xF7), General MIDI System Disable (0xF0 0x7E 0xcc 0x09 0x00 0xF7), Turn DLS On (0xF0 0x7E 0xcc 0x0A 0x01 0xF7), and Turn DLS Off (0xF0 0x7E 0xcc 0x0A 0x02 0xF7). Registrations of subrender parameter token values (Appendix C.6.2) and IETF standards-track documents MAY specify additional RS commands 🌟. - Active commands. Active commands are MIDI commands that do not appear before a Reset State command in the session history 🟢.
- N-active commands. N-active commands are MIDI commands that do not appear before one of the following commands in the session history: MIDI Control Change numbers 123-127 (numbers with All Notes Off semantics), or 120 (All Sound Off), and any Reset State command 🔇.
- C-active commands. C-active commands are MIDI commands that do not appear before one of the following commands in the session history: MIDI Control Change number 121 (Reset All Controllers) and any Reset State command 🎛️.
- Oldest-first ordering rule. Several recovery journal chapters contain a list of elements, where each element is associated with a MIDI command that appears in the session history. In most cases, the chapter definition requires that list elements be ordered in accordance with the "oldest-first ordering rule". here's the normative version, bestie 📜:
Elements associated with the most recent command in the session history coded in the list MUST appear at the end of the list.
Elements associated with the oldest command in the session history coded in the list MUST appear at the start of the list.
All other list elements MUST be arranged with respect to these boundary elements, to produce a list ordering that strictly reflects the relative session history recency of the commands coded by the elements in the list 🕰️.
- Parameter system. A MIDI feature that provides two sets of 16,384 parameters to expand the 0-127 controller number space. The Registered Parameter Names (RPN) system and the Non-Registered Parameter Names (NRPN) system each provides 16,384 parameters 🎹✨.
- Parameter system transaction 💅 Okay so RPNs and NRPNs get their values changed by a run of Control Change commands that together make up a parameter system transaction. A canonical one starts with two Control Change commands that set the parameter number (controller numbers 99 and 98 for NRPNs, controller numbers 101 and 100 for RPNs). Then it can have any number of Data Entry (controller numbers 6 and 38), Data Increment (controller number 96), and Data Decrement (controller number 97) commands to set the value. It wraps up with a second (99, 98) or (101, 100) pair that specifies the null parameter (MSB value 0x7F, LSB value 0x7F). 🎀
Several variants of this sequence are possible, and some are giving main character energy. Most commonly, the closing (99, 98) or (101, 100) pair points at a parameter other than the null one. In that case the pair ends the first transaction and starts a second one, so it counts as part of both. This is legal and recommended in [MIDI], and we call it the "type 1 variant" 🦋
Less commonly, the MSB command (99 or 101) or the LSB command (98 or 100) in a (99, 98) or (101, 100) pair gets left out.
If the MSB command is left out, the transaction uses the MSB value of the most recent C-active Control Change command for controller number 99 or 101 in the session history. That's the "type 2 variant" 🌸
If the LSB command is left out, the LSB value 0x00 is assumed. That's the "type 3 variant" 🧚 The type 2 and type 3 variants are defined as legal in [MIDI], but they're not recommended there. Lowkey, not the vibe. 💀
System real-time commands can show up at any point in a transaction, even between the octets of one command. [MIDI] also doesn't forbid unrelated MIDI commands from showing up during an open transaction. Those are considered "outside" the transaction and don't change its status at all. The exceptions are the commands that end transactions: Reset State commands, and Control Change (0xB) for controller number 121 (Reset All Controllers) [RP015]. 🙈
- Initiated parameter system transaction ✨ A canonical transaction whose initial (99, 98) or (101, 100) Control Change pair is in the session history counts as an initiated parameter system transaction. This also holds for type 1 variants. For type 2 variants (dropped MSB), the transaction is initiated if its initial LSB Control Change command is in the session history. For type 3 variants (dropped LSB), it's initiated if at least one transaction command comes after the initial MSB (99 or 101) Control Change command in the session history. Finishing a transaction doesn't cancel its "initiated" status. Once it's initiated, it stays initiated 👑
- Session history reference counts 🔥 A bunch of recovery journal chapters have a reference count field, which tracks the total number of commands of a given type in the session history. Examples are the Reset and Tune Request command logs (Chapter D, Appendix B.1) and the Active Sense command (Chapter V, Appendix B.2). When a loss event gets detected, the reference count lets a receiver figure out whether any instances of that command were lost, by comparing the journal's count with its own. So a reference count field is useful even for command types where knowing the NUMBER of lost commands doesn't matter, which is true for every example above. It's giving efficiency 🍓
The chapter definitions in Appendices A.2 to A.9 and B.1 to B.5 reflect the default recovery journal behavior. The ch_default, ch_never, and ch_anchor parameters modify these definitions, as described in Appendix C.2.3. ✨
The chapter definitions say whether data MUST be present in the journal. Senders MAY also include non-required data in the journal, and that optional data MUST comply with the normative chapter definition. For example, if a chapter definition says a field codes data from the most recent active command in the session history, the sender MUST NOT code inactive commands or older commands in that field. Not me putting stale data in a field 🙅♀️
Finally, a channel journal only encodes information about MIDI commands that appear on the MIDI channel it protects. Every reference to MIDI commands in Appendices A.2 to A.9 should be read as "MIDI commands appearing on this channel." 🌈
A.2. Chapter P: MIDI Program Change 🎹
A channel journal MUST contain Chapter P if an active Program Change (0xC) command appears in the checkpoint history. Figure A.2.1 shows the format for Chapter P. 💖
0 1 2
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| PROGRAM |B| BANK-MSB |X| BANK-LSB |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.2.1 -- Chapter P formatThe chapter has a fixed size of 24 bits. The PROGRAM field holds the data value of the most recent active Program Change command in the session history. By default, the B, BANK-MSB, X, and BANK-LSB fields MUST be set to 0. Exceptions to that default are defined below. 🫶
If an active Control Change (0xB) command for controller number 0 (Bank Select MSB) appears before the Program Change command in the session history, the B bit MUST be set to 1, and the BANK-MSB field MUST code the data value of that Control Change command. 🍓
If B is set to 1, the BANK-LSB field MUST code the data value of the most recent Control Change command for controller number 32 (Bank Select LSB) that came before the Program Change command coded in the PROGRAM field and after the Control Change command coded in the BANK-MSB field. If no such Control Change command exists, the BANK-LSB field MUST be set to 0. 😭
If B is set to 1, and a Control Change command for controller number 121 (Reset All Controllers) shows up in the MIDI stream between the Control Change command coded by the BANK-MSB field and the Program Change command coded by the PROGRAM field, the X bit MUST be set to 1. ⭐
Note that [RP015] specifies that Reset All Controllers does not reset the values of controller numbers 0 (Bank Select MSB) and 32 (Bank Select LSB). So the X bit does not effect how receivers use the BANK-LSB and BANK-MSB values when recovering from a lost Program Change command. The X bit exists to help recovery in MIDI applications where controller numbers 0 and 32 get used in a non-standard way. It's literally built for the weird setups 🧚
A.3. Chapter C: MIDI Control Change 🎛️
Figure A.3.1 shows the format for Chapter C. 💜
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 8 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| LEN |S| NUMBER |A| VALUE/ALT |S| NUMBER |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|A| VALUE/ALT | .... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.3.1 -- Chapter C formatThe chapter is a 1-octet header, followed by a variable length list of 2-octet controller logs. The list MUST contain at least one controller log. The 7-bit LEN field codes the number of controller logs in the list, minus one. The semantics of the controller log fields are defined in Appendix A.3.2. 🦋
A channel journal MUST contain Chapter C if the rules defined in this appendix require that one or more controller logs appear in the list. ✨
A.3.1. Log Inclusion Rules
Okay bestie, so a controller log is basically a little receipt 🧾 for one Control Change command from the session history. Let's get into who gets a receipt and who doesn't 💅
In the default use of the payload format, list logs MUST encode information about the most recent active command in the session history for a controller number. Logs for earlier commands MUST NOT appear in the list 🙅. Also, as a rule, the list MUST contain a log for the most recent active command for a controller number that appears in the checkpoint history 👑. There are some exceptions to that rule, and they're the fun part 🤭
- MIDI streams can send 14-bit controller values by pairing a Most Significant Byte (MSB, controller numbers 0-31, 99, 101) with a Least Significant Byte (LSB, controller numbers 32-63, 98, 100) through Control Change commands [MIDI] 🎀
If the most recent active Control Change command in the session history for a 14-bit controller pair uses the MSB number, Chapter C MAY omit the controller log for the most recent active Control Change command for the associated LSB number. The command ordering makes that LSB value irrelevant, so it's giving efficiency 🦋 However, this exception MUST NOT be applied if the sender is not certain that the MIDI source uses 14-bit semantics for the controller number pair. Some MIDI sources ignore 14-bit controller semantics and use the LSB controller numbers as independent 7-bit controllers, so don't assume 😭
- If active Control Change commands for controller numbers 0 (Bank Select MSB) or 32 (Bank Select LSB) appear in the checkpoint history, and if those command instances are also coded in the
BANK-MSBandBANK-LSBfields of Chapter P (Appendix A.2), Chapter C MAY omit the controller logs for those commands 🌸 - Some controller number pairs are mutually exclusive. Controller numbers 124 (Omni Off) and 125 (Omni On) form one pair, and controller numbers 126 (Mono) and 127 (Poly) form another 🌈
If active Control Change commands for one or both members of a mutually exclusive pair appear in the checkpoint history, a log for the controller number of the most recent command for that pair in the checkpoint history MUST appear in the controller list. However, the list MAY omit the controller log for the most recent active command for the other number in the pair.
If active Control Change commands for one or both members of a mutually exclusive pair appear in the session history, and if a log for the controller number of the most recent command for the pair does not appear in the controller list, then a log for the most recent command for the other number of the pair MUST NOT appear in the controller list 🙈
- If an active Control Change command for controller number 121 (Reset All Controllers) appears in the session history, the controller list MAY omit logs for Control Change commands that precede the Reset All Controllers command in the session history, under certain conditions. ✨
Namely, a log MAY be omitted if the sender is certain that a command stream follows the Reset All Controllers semantics defined in [RP015], and if the log codes a controller number for which [RP015] specifies a reset value.
For example, [RP015] specifies that controller number 1 (Modulation Wheel) is reset to the value 0, so a controller log for Modulation Wheel MAY be omitted from the controller log list. In contrast, [RP015] specifies that controller number 7 (Channel Volume) is not reset, so a controller log for Channel Volume MUST NOT be omitted from the controller log list 🥹
- Appendix A.3.4 defines exception rules for the MIDI Parameter System controller numbers 6, 38, and 96-101 🧚
A.3.2. Controller Log Format
Figure A.3.2 shows the controller log structure of Chapter C 🦋
0 1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| NUMBER |A| VALUE/ALT |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.3.2 -- Chapter C controller logThe 7-bit NUMBER field identifies the controller number of the coded command. The 7-bit VALUE/ALT field codes recovery information for the command, and the A bit sets the format of that VALUE/ALT field 💜
A log encodes recovery information using one of the following tools: the value tool, the toggle tool, or the count tool. 🛠️
A log uses the value tool if the A bit is set to 0. The value tool codes the 7-bit data value of a command in the VALUE/ALT field. It works best for controllers that code a continuous quantity, such as number 1 (Modulation Wheel) 🌊
The A bit is set to 1 to code the toggle or count tool. These tools work best for controllers that code discrete actions. Figure A.3.3 shows the controller log for these tools 💖
0 1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| NUMBER |1|T| ALT |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.3.3 -- Controller log for ALT toolsA log uses the toggle tool if the T bit is set to 0. A log uses the count tool if the T bit is set to 1. Both methods use the 6-bit ALT field as an unsigned integer 🔢
The toggle tool works best for controllers that act as on/off switches, such as 64 (Damper Pedal (Sustain)). These controllers code the "off" state with control values 0-63 and the "on" state with 64-127 🎹
For the toggle tool, the ALT field codes the total number of toggles (off->on and on->off) caused by Control Change commands in the session history, up to and including a toggle caused by the command coded by the log. The toggle count includes toggles caused by Control Change commands for controller number 121 (Reset All Controllers) 🔁
Toggle counting is performed modulo 64. The toggle count is reset at the start of a session, and whenever a Reset State command (Appendix A.1) appears in the session history. When those reset events occur, the toggle count for a controller is set to 0 (for controllers whose default value is 0-63) or 1 (for controllers whose default value is 64-127) 🧮
The Damper Pedal (Sustain) controller shows why the toggle tool beats the value tool for switch controllers 🎹✨ In piano applications, the "on" state lets notes resonate, while the "off" state immediately damps notes to silence. If an "off" command gets lost in an on->off->on sequence, you end up with ringing notes that should have been silenced 😭 The toggle tool lets receivers detect that lost "off" command, but the value tool does not. Iconic behaviour, honestly 💅
The count tool is conceptually similar to the toggle tool. For the count tool, the ALT field codes the total number of Control Change commands in the session history, up to and including the command coded by the log. Command counting is performed modulo 64. The command count is set to 0 at the start of the session, and is reset to 0 whenever a Reset State command (Appendix A.1) appears in the session history 📊
Because the count tool ignores the data value, it's a good match for controllers whose controller value is ignored, such as number 123 (All Notes Off). More generally, the count tool may be used to code a (modulo 64) identification number for a command 🪪
A.3.3. Log List Coding Rules ✨
Okay so this part is about how we organise the controller logs inside the Chapter C log list 📋
A log encodes info about one particular Control Change command in the session history. Usually a command SHOULD be coded by a single tool (so, one log) 💅. If a number is coded with a single tool and that tool is the count tool, recovery Control Change commands generated by a receiver SHOULD use the default control value for the controller 🫶.
However, a command MAY be coded by several tool types (aka several logs, each using a different tool) 🧚. This can improve recovery performance for controllers with complex vibes, like controller number 84 (Portamento Control) or controller number 121 (Reset All Controllers) when it's used with a non-zero data octet (the semantics are described in [DLS2]) 🔥.
If a command is encoded by multiple tools, the logs MUST be placed in the list in this order: count tool log (if any), then value tool log (if any), then toggle tool log (if any) 📐.
The Chapter C log list MUST obey the oldest-first ordering rule (defined in Appendix A.1). Ngl this ordering keeps the info needed to recover 14-bit controller values, and it still doesn't stop you from using MSB and LSB controller pairs as independent 7-bit controllers 👑.
In the default use of the payload format, all logs that appear in the list for a controller number encode info about one Control Change command. That's namely the most recent active Control Change command in the session history for the number 💜.
This coding scheme gives good recovery performance for the standard uses of Control Change commands defined in [MIDI]. But not all MIDI applications stick to those 🙈.
For example, think about the common MIDI encoding of rotary encoders ("infinite" rotation knobs). The mixing console MIDI convention defined in [LCP] codes the position of rotary encoders as a series of Control Change commands. Each command encodes a relative change of knob position from the last update (expressed as a clockwise or counter-clockwise knob turning angle) 🌀.
As the knob position is encoded incrementally over a series of Control Change commands, the best recovery performance is obtained if the log list encodes all Control Change commands for encoder controller numbers that appear in the checkpoint history, not only the most recent command 🎡.
To support application areas that use Control Change commands this way, Chapter C may be configured to encode info about several Control Change commands for a controller number. We call this method "enhanced" encoding, and we describe it below ✨.
In Appendix C.2.3, we show how to configure a stream to use enhanced Chapter C encoding for specific controller numbers. In Section 5 in the main text, we show how the H bits in the recovery journal header (Figure 8) and in the channel journal header (Figure 9) indicate the use of enhanced Chapter C encoding 🦋.
Here, we define how to encode a Chapter C log list that uses the enhanced encoding method.
Senders that use the enhanced encoding method for a controller number MUST obey the rules below. These rules let a receiver figure out which logs in the list correspond to lost commands. Note that these rules override the exceptions listed in Appendix A.3.1 💅.
- If N commands for a controller number are encoded in the list, the commands MUST be the N most recent commands for the controller number in the session history. For example, for N = 2, the sender MUST encode the most recent command and the second most recent command, not the most recent command and the third most recent command.
- If a controller number uses enhanced encoding, the encoding of the least-recent command for the controller number in the log list MUST include a count tool log. In addition, if commands are encoded for the controller number whose logs have S bits set to 0, the encoding of the least-recent command with S = 0 logs MUST include a count tool log. The count tool is OPTIONAL for the other commands for the controller number encoded in the list, since a receiver can efficiently deduce the count tool value for these commands, for both single-packet and multi-packet loss events.
- The use of the value and toggle tools MUST be identical for all commands for a controller number encoded in the list. For example, a value tool log either MUST appear for all commands for the controller number coded in the list, or alternatively, value tool logs for the controller number MUST NOT appear in the list. Likewise, a toggle tool log either MUST appear for all commands for the controller number coded in the list, or alternatively, toggle tool logs for the controller number MUST NOT appear in the list.
- If a command is encoded by multiple tools, the logs MUST be placed in the list in the following order: count tool log (if any), followed by value tool log (if any), followed by toggle tool log (if any).
These rules let a receiver recovering from a packet loss use the count tool log to match the commands encoded in the list with its own history of the stream, as we describe below. Note that the text below describes a non-normative algorithm, and receivers are free to use any algorithm to match their history with the log list 🤓.
In a typical implementation of the enhanced encoding method, a receiver computes and stores count, value, and toggle tool data field values for the most recent Control Change command it has received for a controller number 🗂️.
After a loss event, a receiver parses the Chapter C list and processes list logs for a controller number that uses enhanced encoding as follows.
The receiver compares the count tool ALT field for the least-recent command for the controller number in the list against its stored count data for the controller number. That tells it whether recovery is necessary for the command coded in the list. The value and toggle tool logs (if any) that directly follow the count tool log are associated with this least-recent command.
To check more-recent commands for the controller, the receiver detects additional value and/or toggle tool logs for the controller number in the list, and infers count tool data for the command coded by these logs. This inferred data is used to determine if recovery is necessary for the command coded by those value and/or toggle tool logs 🔍.
This way, a receiver is able to execute only lost commands, without executing a command twice 🙌. While recovering from a single packet loss, a receiver may skip through S = 1 logs in the list, since the first S = 0 log for an enhanced controller number is always a count tool log.
Note that the requirements in Appendix C.2.2.2 for protective sender and receiver actions during session startup for multicast operation are of particular importance for enhanced encoding. Receivers need to initialize their count tool data structures with recovery journal data in order to match commands correctly after a loss event 🚨.
Finally, we note in passing that in some applications of rotary encoders, a good user experience may be possible without the use of enhanced encoding. These applications are distinguished by visual feedback of encoding position that is driven by the post-recovery rotary encoding stream, and relatively low packet loss. In these domains, recovery performance may be acceptable for rotary encoders if the log list encodes only the most recent command, if both count and value logs appear for the command 🎛️.
A.3.4. The Parameter System 🎀
Readers may wish to review the Appendix A.1 definitions of "parameter system", "parameter system transaction", and "initiated parameter system transaction" before reading this section 📖.
Parameter system transactions update a MIDI Registered Parameter Number (RPN) or Non-Registered Parameter Number (NRPN) value. A parameter system transaction is a sequence of Control Change commands that may use the following controller numbers:
- Data Entry MSB (6)
- Data Entry LSB (38)
- Data Increment (96)
- Data Decrement (97)
- Non-Registered Parameter Number (NRPN) LSB (98)
- Non-Registered Parameter Number (NRPN) MSB (99)
- Registered Parameter Number (RPN) LSB (100)
- Registered Parameter Number (RPN) MSB (101)
Control Change commands that are a part of a parameter system transaction MUST NOT be coded in Chapter C controller logs. Instead, these commands are coded in Chapter M, the MIDI Parameter chapter defined in Appendix A.4 🗃️.
However, Control Change commands that use the listed controllers as general-purpose controllers (i.e., outside of a parameter system transaction) MUST NOT be coded in Chapter M.
Instead, the controllers are coded in Chapter C controller logs. The controller logs follow the coding rules stated in Appendix A.3.2 and A.3.3. The rules for coding paired LSB and MSB controllers, as defined in Appendix A.3.1, apply to the pairs (6, 38), (99, 98), and (101, 100) when coded in Chapter C 💖.
If active Control Change commands for controller numbers 6, 38, or 96-101 appear in the checkpoint history, and these commands are used as general-purpose controllers, the most recent general-purpose command instance for these controller numbers MUST appear as entries in the Chapter C controller list 📝.
MIDI syntax permits a source to use controllers 6, 38, 96, and 97 as parameter-system controllers and general-purpose controllers in the same stream. An RTP MIDI sender MUST deduce the role of each Control Change command for these controller numbers by noting the placement of the command in the stream, and it MUST use this information to code the command in Chapter C or Chapter M, as appropriate 🧩.
Specifically, active Control Change commands for controllers 6, 38, 96, and 97 act in a general-purpose way when:
- no active Control Change commands that set an RPN or NRPN parameter number appear in the session history, or
- the most recent active Control Change commands in the session history that set an RPN or NRPN parameter number code the null parameter (MSB value
0x7F, LSB value0x7F), or - a Control Change command for controller number 121 (Reset All Controllers) appears more recently in the session history than all active Control Change commands that set an RPN or NRPN parameter number (see [RP015] for details).
Finally, we note that a MIDI source that follows the recommendations of [MIDI] exclusively uses numbers 98-101 as parameter system controllers. Alternatively, a MIDI source may exclusively use 98-101 as general-purpose controllers and lose the ability to perform parameter system transactions in a stream 😬.
In the language of [MIDI], the general-purpose use of controllers 98-101 constitutes a non-standard controller assignment. As most real-world MIDI sources use the standard controller assignment for controller numbers 98-101, an RTP MIDI sender SHOULD assume these controllers act as parameter system controllers, unless it knows that a MIDI source uses controller numbers 98-101 in a general-purpose way 🎹💜
A.4. Chapter M: MIDI Parameter System 🎀
Okay bestie, before we dive in, go check the Appendix A.1 definitions for "C-active", "parameter system", "parameter system transaction", and "initiated parameter system transaction" 📖 It's giving foundations, and you'll need them here 💅
Chapter M protects parameter system transactions for Registered Parameter Number (RPN) and Non-Registered Parameter Number (NRPN) values 🌈 Here's the top-level format in Figure A.4.1:
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S|P|E|U|W|Z| LENGTH |Q| PENDING | Log list ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.4.1 -- Top-level Chapter M formatChapter M starts with a 2-octet header 🦋 If the P header bit is set to 1, a 1-octet field follows the header, and it codes the 7-bit PENDING value plus its associated Q bit ✨
The 10-bit LENGTH field codes the size of Chapter M, and it follows the semantics in Appendix A.1 🫶
Chapter M ends with a list of zero or more variable-length parameter logs. Appendix A.4.2 defines the bitfield format of a parameter log, and Appendix A.4.1 defines the inclusion semantics of the log list 💜
A channel journal MUST contain Chapter M if the rules in Appendix A.4.1 require that one or more parameter logs appear in the list 🔥
A channel journal also MUST contain Chapter M if the most recent C-active Control Change command involved in a parameter system transaction in the checkpoint history is:
- an RPN MSB (101) or NRPN MSB (99) controller, or
- an RPN LSB (100) or NRPN LSB (98) controller that completes the coding of the null parameter (MSB value 0x7F, LSB value 0x7F).
This rule gives loss protection for partially sent parameter numbers and for the null parameter numbers 🛡️ no cap, it's so important 😭
If the most recent C-active Control Change command involved in a parameter system transaction in the session history is for the RPN MSB or NRPN MSB controller, the P header bit MUST be set to 1, and the PENDING field (plus its associated Q bit) MUST follow the Chapter M header. Otherwise, the P header bit MUST be set to 0, and the PENDING field and Q bit MUST NOT appear in Chapter M 🍓
If PENDING codes an NRPN MSB, the Q bit MUST be set to 1. If PENDING codes an RPN MSB, the Q bit MUST be set to 0 🧚
The E header bit codes the current transaction state of the MIDI stream. If E = 1, an initiated transaction is in progress. Here are the rules for setting the E header bit, bestie 👑
- If no C-active parameter system transaction Control Change commands appear in the session history, the E bit MUST be set to 0.
- If the P header bit is set to 1, the E bit MUST be set to 0.
- If the most recent C-active parameter system transaction Control Change command in the session history is for the NRPN LSB or RPN LSB controller number, and if this command acts to complete the coding of the null parameter (MSB value 0x7F, LSB value 0x7F), the E bit MUST be set to 0.
- Otherwise, an initiated transaction is in progress, and the E bit MUST be set to 1.
The U, W, and Z header bits code properties that are shared by all parameter logs in the list. When these properties are set, parameter logs can be coded more efficiently (we explain how in A.4.1) 🌸
By default, the U, W, and Z bits MUST be set to 0. If all parameter logs in the list code RPN parameters, the U bit MAY be set to 1. If all parameter logs in the list code NRPN parameters, the W bit MAY be set to 1. If the parameter numbers of all RPN and NRPN logs in the list lie in the range 0-127 (and so have an MSB value of 0), the Z bit MAY be set to 1 💖
Note that the C-active semantics show up in the paragraphs above because [RP015] specifies that pending Parameter System transactions are closed by a Control Change command for controller number 121 (Reset All Controllers) 🥹
A.4.1. Log Inclusion Rules
Parameter logs code recovery information for a specific RPN or NRPN parameter ✨
A parameter log MUST appear in the list if an active Control Change command that forms part of an initiated transaction for the parameter appears in the checkpoint history 🔥
There's an exception to this rule if the checkpoint history only contains transaction Control Change commands for controller numbers 98-101 that act to terminate the transaction. In that case, a log for the parameter MAY be left out of the list 🙈
A log MAY appear in the list if an active Control Change command that forms part of an initiated transaction for the parameter appears in the session history. Otherwise, a log for the parameter MUST NOT appear in the list 💅
Multiple logs for the same RPN or NRPN parameter MUST NOT appear in the log list 🌈
The parameter log list MUST obey the oldest-first ordering rule (defined in Appendix A.1), with the phrase "parameter transaction" replacing the word "command" in the rule definition 🦋
Parameter logs tied to the RPN or NRPN null parameter (LSB = 0x7F, MSB = 0x7F) MUST NOT appear in the log list. Chapter M uses the E header bit (Figure A.4.1) and the log list ordering rules to code null parameter semantics 🧚
Note that the "active" semantics (instead of "C-active" semantics) show up in the paragraphs above because [RP015] specifies that pending Parameter System transactions are not reset by a Control Change command for controller number 121 (Reset All Controllers). However, the rule that follows uses C-active semantics, because it's about protecting the transaction system itself, and [RP015] specifies that Reset All Controllers acts to close a transaction in progress 🧠
In most cases, parameter logs for RPN and NRPN parameters assigned to the ch_never parameter (Appendix C.2.3) MAY be left out of the list. There's an exception if:
- the log codes the most recent initiated transaction in the session history, and
- a C-active command that forms part of the transaction appears in the checkpoint history, and
- the E header bit for the top-level Chapter M header (Figure A.4.1) is set to 1.
In this case, a log for the parameter MUST appear in the list. This log tells receivers recovering from a loss that a transaction is in progress, so they can correctly interpret RPN or NRPN Control Change commands that come after the loss event 🫶
A.4.2. Log Coding Rules
Figure A.4.2 shows the parameter log structure of Chapter M 💜
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 8 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| PNUM-LSB |Q| PNUM-MSB |J|K|L|M|N|T|V|R| Fields ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.4.2 -- Parameter log formatThe log starts with a header, whose default size (as Figure A.4.2 shows) is 3 octets. If the Q header bit is set to 0, the log encodes an RPN parameter. If Q = 1, the log encodes an NRPN parameter. The 7-bit PNUM-MSB and PNUM-LSB fields code the parameter number and reflect the Control Change command data values for controllers 99 and 98 (for NRPNs) or 101 and 100 (for RPNs) 🎀
The J, K, L, M, and N header bits form a Table of Contents (TOC) for the log and signal whether fixed-size fields follow the header. A header bit set to 1 means that field is in the log. The fields appear in the same order as the header bits in the TOC. Appendices A.4.2.1-2 define the fields tied to each TOC header bit 📖
The T and V header bits carry information about the parameter log but are not part of the TOC. A set T or V bit does not signal the presence of any parameter log field 🌷
If the rules in Appendix A.4.1 say a log for a given parameter MUST appear in Chapter M, the log MUST code enough information to protect the parameter from the loss of active parameter transaction Control Change commands in the checkpoint history 🛡️
This rule doesn't apply if the parameter coded by the log is assigned to the ch_never parameter (Appendix C.2.3). In that case, senders MAY choose to set the J, K, L, M, and N TOC bits to 0, coding a parameter log with no fields 🙌
Note that logs protecting parameters assigned to ch_never are REQUIRED under certain conditions (see Appendix A.4.1). The purpose of the log is to tell receivers recovering from a loss that a transaction is in progress, so they can correctly interpret RPN or NRPN Control Change commands after the loss event 💀
Parameter logs give two tools for parameter protection: the value tool and the count tool. Depending on the semantics of the parameter, senders may use either tool, both tools, or neither tool to protect a given parameter 🌈
The value tool codes information a receiver can use to figure out the current value of an RPN or NRPN parameter. If a parameter log uses the value tool, the V header bit MUST be set to 1, and the semantics defined in Appendix A.4.2.1 for setting the J, K, L, and M TOC bits MUST be followed. If a parameter log does not use the value tool, the V bit MUST be set to 0, and the J, K, L, and M TOC bits MUST also be set to 0 ✨
The count tool codes the number of transactions for an RPN or NRPN parameter. If a parameter log uses the count tool, the T header bit MUST be set to 1, and the semantics defined in Appendix A.4.2.2 for setting the N TOC bit MUST be followed. If a parameter log does not use the count tool, the T bit and the N TOC bit MUST be set to 0 🧁
Note that V and T are set if the sender uses the value (V) or count (T) tool for the log on an ongoing basis. So V may be set even if J = K = L = M = 0, and T may be set even if N = 0 🦋
In many cases, all parameters coded in the log list are one type (RPN or NRPN), and all parameter numbers lie in the range 0-127. As described in Appendix A.4.1, senders MAY signal this by setting the top-level Chapter M header bit Z to 1 (to code the restricted range) and by setting the U or W bit to 1 (to code the parameter type) 🍭
If the top-level Chapter M header codes Z = 1 and either U = 1 or W = 1, all logs in the parameter log list MUST use a modified header format. This modification deletes bits 8-15 of the bitfield shown in Figure A.4.2, which gives a 2-octet header. The values of the deleted PNUM-MSB and Q fields can be inferred from the U, W, and Z bit values 🎀💖
A.4.2.1. The Value Tool ✨
okay so the value tool is basically the bestie who keeps track of the value of an RPN or NRPN parameter 💖 it uses a few fields to do the job, and they're all lowkey iconic 💅
The J TOC bit codes the presence of the octet shown in Figure A.4.3 in the field list.
0
0 1 2 3 4 5 6 7
+-+-+-+-+-+-+-+-+
|X| ENTRY-MSB |
+-+-+-+-+-+-+-+-+
Figure A.4.3 -- ENTRY-MSB fieldThe 7-bit ENTRY-MSB field codes the data value of the most recent active Control Change command for controller number 6 (Data Entry MSB) in the session history that appears in a transaction for the log parameter. 🎀
The X bit MUST be set to 1 if the command coded by ENTRY-MSB precedes the most recent Control Change command for controller 121 (Reset All Controllers) in the session history. Otherwise, the X bit MUST be set to 0.
A parameter log that uses the value tool MUST include the ENTRY-MSB field if an active Control Change command for controller number 6 appears in the checkpoint history.
Note that [RP015] specifies that Control Change commands for controller 121 (Reset All Controllers) do not reset RPN and NRPN values, so the X bit wouldn't play a recovery role for MIDI systems that comply with [RP015] 🙈
However, certain renderers (such as DLS 2 [DLS2]) specify that certain RPN values are reset for some uses of Reset All Controllers. The X bit (and other bitfield features of this nature in this appendix) plays a role in recovery for renderers of this type. the way that the X bit has a whole second life for these renderers 👑
The K TOC bit codes the presence of the octet shown in Figure A.4.4 in the field list.
0
0 1 2 3 4 5 6 7
+-+-+-+-+-+-+-+-+
|X| ENTRY-LSB |
+-+-+-+-+-+-+-+-+
Figure A.4.4 -- ENTRY-LSB fieldThe 7-bit ENTRY-LSB field codes the data value of the most recent active Control Change command for controller number 38 (Data Entry LSB) in the session history that appears in a transaction for the log parameter. 🦋
The X bit MUST be set to 1 if the command coded by ENTRY-LSB precedes the most recent Control Change command for controller 121 (Reset All Controllers) in the session history. Otherwise, the X bit MUST be set to 0.
As a rule, a parameter log that uses the value tool MUST include the ENTRY-LSB field if an active Control Change command for controller number 38 appears in the checkpoint history. However, the ENTRY-LSB field MUST NOT appear in a parameter log if the Control Change command associated with the ENTRY-LSB precedes a Control Change command for controller number 6 (Data Entry MSB) that appears in a transaction for the log parameter in the session history. 💀 not me coding a field that gets overwritten like that
The L TOC bit codes the presence of the octets shown in Figure A.4.5 in the field list.
0 1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|G|X| A-BUTTON |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.4.5 -- A-BUTTON fieldThe 14-bit A-BUTTON field codes a count of the number of active Control Change commands for controller numbers 96 and 97 (Data Increment and Data Decrement) in the session history that appear in a transaction for the log parameter. 🍓
The M TOC bit codes the presence of the octets shown in Figure A.4.6 in the field list.
0 1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|G|R| C-BUTTON |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.4.6 -- C-BUTTON fieldThe 14-bit C-BUTTON field has semantics identical to A-BUTTON, except that Data Increment and Data Decrement Control Change commands that precede the most recent Control Change command for controller 121 (Reset All Controllers) in the session history are not counted. 🧚
For both A-BUTTON and C-BUTTON, Data Increment and Data Decrement Control Change commands are not counted if they precede Control Change commands for controller numbers 6 (Data Entry MSB) or 38 (Data Entry LSB) that appear in a transaction for the log parameter in the session history.
The A-BUTTON and C-BUTTON fields are interpreted as unsigned integers, and the G bit associated with the field codes the sign of the integer (G = 0 for positive or zero, G = 1 for negative). ✨
To compute and code the count value, initialize the count value to 0, add 1 for each qualifying Data Increment command, and subtract 1 for each qualifying Data Decrement command. After each add or subtract, limit the count magnitude to 16383. The G bit codes the sign of the count, and the A-BUTTON or C-BUTTON field codes the count magnitude. 🌈
For the A-BUTTON field, if the most recent qualified Data Increment or Data Decrement command precedes the most recent Control Change command for controller 121 (Reset All Controllers) in the session history, the X bit associated with the A-BUTTON field MUST be set to 1. Otherwise, the X bit MUST be set to 0.
A parameter log that uses the value tool MUST include the A-BUTTON and C-BUTTON fields if an active Control Change command for controller numbers 96 or 97 appears in the checkpoint history. However, to improve coding efficiency, this rule has several exceptions:
- If the log includes the A-BUTTON field, and if the X bit of the A-BUTTON field is set to 1, the C-BUTTON field (and its associated R and G bits) MAY be omitted from the log.
- If the log includes the A-BUTTON field, and if the A-BUTTON and C-BUTTON fields (and their associated G bits) code identical values, the C-BUTTON field (and its associated R and G bits) MAY be omitted from the log. 🫶
A.4.2.2. The Count Tool 🌸
the count tool keeps track of how many transactions an RPN or NRPN parameter has had. The N TOC bit codes the presence of the octet shown in Figure A.4.7 in the field list.
0
0 1 2 3 4 5 6 7
+-+-+-+-+-+-+-+-+
|X| COUNT |
+-+-+-+-+-+-+-+-+
Figure A.4.7 -- COUNT fieldThe 7-bit COUNT codes the number of initiated transactions for the log parameter that appear in the session history. Initiated transactions are counted if they contain one or more active Control Change commands, including commands for controllers 98-101 that initiate the parameter transaction. 💖
If the most recent counted transaction precedes the most recent Control Change command for controller 121 (Reset All Controllers) in the session history, the X bit associated with the COUNT field MUST be set to 1. Otherwise, the X bit MUST be set to 0.
Transaction counting is performed modulo 128. The transaction count is set to 0 at the start of a session and is reset to 0 whenever a Reset State command (Appendix A.1) appears in the session history.
A parameter log that uses the count tool MUST include the COUNT field if an active command that increments the transaction count (modulo 128) appears in the checkpoint history.
A.5. Chapter W: MIDI Pitch Wheel 🦋
A channel journal MUST contain Chapter W if a C-active MIDI Pitch Wheel (0xE) command appears in the checkpoint history. Figure A.5.1 shows the format for Chapter W.
0 1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| FIRST |R| SECOND |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.5.1 -- Chapter W formatThe chapter has a fixed size of 16 bits. The FIRST and SECOND fields are the 7-bit values of the first and second data octets of the most recent active Pitch Wheel command in the session history. 🎀
Note that Chapter W encodes C-active commands, and thus does not encode active commands that are not C-active (see the second-to-last paragraph of Appendix A.1 for an explanation of chapter inclusion text in this regard).
Chapter W does not encode "active but not C-active" commands because [RP015] declares that Control Change commands for controller number 121 (Reset All Controllers) act to reset the Pitch Wheel value to 0. If Chapter W encoded "active but not C-active" commands, a repair operation following a Reset All Controllers command could incorrectly repair the stream with a stale Pitch Wheel value. not the stale value era 💀
A.6. Chapter N: MIDI NoteOff and NoteOn 🎀
In this appendix, we consider NoteOn commands with zero velocity to be NoteOff commands. Readers may wish to review the Appendix A.1 definition of "N-active commands" before reading this appendix. 🧚
Chapter N completely protects note commands in streams that alternate between NoteOn and NoteOff commands for a particular note number. However, in rare applications, multiple overlapping NoteOn commands may appear for a note number. Chapter E, described in Appendix A.7, augments Chapter N to completely protect these streams. chapter E is the upgrade arc nobody asked for but we all needed 👑
A channel journal MUST contain Chapter N if an N-active MIDI NoteOn (0x9) or NoteOff (0x8) command appears in the checkpoint history. Figure A.6.1 shows the format for Chapter N.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 8 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|B| LEN | LOW | HIGH |S| NOTENUM |Y| VELOCITY |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| NOTENUM |Y| VELOCITY | .... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| OFFBITS | OFFBITS | .... | OFFBITS |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.6.1 -- Chapter N formatChapter N consists of a 2-octet header, followed by at least one of the following data structures:
- A list of note logs to code NoteOn commands.
- A NoteOff bitfield structure to code NoteOff commands.
We define the header bitfield semantics in Appendix A.6.1. We define the note log semantics and the NoteOff bitfield semantics in Appendix A.6.2. 💅
If one or more N-active NoteOn or NoteOff commands in the checkpoint history reference a note number, the note number MUST be coded in either the note log list or the NoteOff bitfield structure.
The note log list MUST contain an entry for all note numbers whose most recent checkpoint history appearance is in an N-active NoteOn command. The NoteOff bitfield structure MUST contain a set bit for all note numbers whose most recent checkpoint history appearance is in an N-active NoteOff command. 🌈
A note number MUST NOT be coded in both structures.
All note logs and NoteOff bitfield set bits MUST code the most recent N-active NoteOn or NoteOff reference to a note number in the session history.
The note log list MUST obey the oldest-first ordering rule (defined in Appendix A.1). ⭐
A.6.1. Header Structure 💖
okay so the header for Chapter N is the bit that codes the size of the note list and the bitfield structures 🧚 check out Figure A.6.2 bestie, it's giving neat boxes ✨
0 1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|B| LEN | LOW | HIGH |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.6.2 -- Chapter N headerso LEN is a 7-bit integer that counts the 2-octet note logs in the note list 🎀 LEN = 0 is totally valid and just means an empty note list, no drama 🫶
the 4-bit LOW and HIGH fields are how we code the number of OFFBITS octets that come after the note log list 🌸 both are unsigned integers. if LOW <= HIGH, there are (HIGH - LOW + 1) OFFBITS octets in the chapter. the pairs (LOW = 15, HIGH = 0) and (LOW = 15, HIGH = 1) both mean an empty NoteOff bitfield structure, so zero OFFBITS octets 🙈 and any other pair where LOW > HIGH is a total no-go, those MUST NOT appear in the header 🚫
now the B bit is the S-bit vibe (see Appendix A.1) but for the NoteOff bitfield structure 💅 by default B MUST be set to 1. BUT if the MIDI command section of the previous packet (packet I - 1, with I as defined in Appendix A.1) has a NoteOff command for the channel, then B MUST be set to 0. and when B is 0, the higher-level recovery journal elements holding Chapter N MUST have S bits set to 0, including the top-level journal header. main character energy for the S bits 👑
last bit: LEN = 127 codes a note list length of 127 or 128 note logs, depending on LOW and HIGH 🌈 if LEN = 127, LOW = 15, and HIGH = 0, the note list holds 128 note logs and the NoteOff bitfield structure is empty. for every other LOW/HIGH combo, LEN = 127 means the note list has 127 note logs. in that case there are (HIGH - LOW + 1) NoteOff OFFBITS octets if LOW <= HIGH, and zero OFFBITS octets if LOW = 15 and HIGH = 1 ✨
A.6.2. Note Structures 🦋
Figure A.6.3 shows the 2-octet note log structure, and it's so cute 🥹
0 1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| NOTENUM |Y| VELOCITY |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.6.3 -- Chapter N note logthe 7-bit NOTENUM field codes the note number for the log 🎹 a note number MUST NOT be shown by multiple note logs in the note list, one log per note, periodt 💜
the 7-bit VELOCITY field codes the velocity value of the most recent N-active NoteOn command for that note number in the session history 💥 if you've got multiple overlapping NoteOns for the same note number, those get coded with Chapter E instead (see Appendix A.7) 🫶
VELOCITY is never zero, bestie 🙅 NoteOn commands with zero velocity are coded as NoteOff commands in the NoteOff bitfield structure instead.
the note log doesn't code the execution time of the NoteOn command 🕰️ BUT the Y bit gives a hint from the sender about that timing. Y = 1 is a recommendation to play the NoteOn command recovered from the note log, and Y = 0 is a recommendation to skip it. see Section 4.2 of [RFC4696] for the non-normative guidance on using the Y bit 📚
Figure A.6.1 shows the NoteOff bitfield structure as the list of OFFBITS octets at the end of the chapter 🌸 each NoteOff OFFBITS octet codes NoteOff info for eight consecutive MIDI note numbers, and the most-significant bit is the lowest note number 🧠 the most-significant bit of the first OFFBITS octet codes note number 8*LOW, and the most-significant bit of the last OFFBITS octet codes note number 8*HIGH 🍓
a set bit codes a NoteOff command for that note number ✨ for the most efficient coding of the NoteOff bitfield structure, the first and last octets should each have at least one set bit 💅 wait, it's actually "contain at least one set bit," so let's keep that exact vibe. Also, Chapter N does not code NoteOff velocity data, so don't go looking for it there 🙈
okay now the lowkey tricky part 😭 in the general case, the recovery journal does not code the relative placement of a NoteOff command and a Change Control command for controller 64 (Damper Pedal (Sustain)). BUT in many cases a receiver handling a loss event can figure out that placement from the history of the stream, and then decide whether a NoteOff note is sustained by the pedal 🎹🦶 if that determination isn't possible, receivers SHOULD err on the side of silencing pedal sustains, because erroneously sustained notes may produce unpleasant (albeit transient) artifacts 💀 nobody wants that haunted piano energy 👻
A.7. Chapter E: MIDI Note Command Extras 🎀
Readers may wish to review the Appendix A.1 definition of "N-active commands" before reading this appendix 📖 in this appendix, a NoteOn command with a velocity of 0 is considered to be a NoteOff command with a release velocity value of 64, so just keep that in your back pocket 💜
Chapter E encodes recovery info about MIDI NoteOn (0x9) and NoteOff (0x8) command features that rarely show up in MIDI streams 🦄 receivers use Chapter E to reduce transient artifacts for streams where several NoteOn commands appear for a note number without an intervening NoteOff. receivers also use it for streams that use NoteOff release velocity. Chapter E supplements the note info coded in Chapter N (Appendix A.6), so they're besties 🫶
Figure A.7.1 shows the format for Chapter E 🌈
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 8 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| LEN |S| NOTENUM |V| COUNT/VEL |S| NOTENUM |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|V| COUNT/VEL | .... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.7.1 -- Chapter E formatthe chapter is a 1-octet header followed by a variable-length list of 2-octet note logs 📦 Appendix A.7.1 defines the bitfield format for a note log 🧚
the log list MUST contain at least one note log, no empty lists allowed 🙅 the 7-bit LEN header field codes the number of note logs in the list, minus one. a channel journal MUST contain Chapter E if the rules in this appendix require one or more note logs in the list 🔥 and the note log list MUST obey the oldest-first ordering rule (defined in Appendix A.1) ⏳
A.7.1. Note Log Format 🌸
Figure A.7.2 reproduces the note log structure of Chapter E 💅
0 1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| NOTENUM |V| COUNT/VEL |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.7.2 -- Chapter E note loga note log codes info about the MIDI note number in the 7-bit NOTENUM field 🎶 what that info means depends on the value of the V flag bit 🦋
if V is 1, the COUNT/VEL field codes the release velocity value of the most recent N-active NoteOff command for that note number in the session history 🎀
if V is 0, the COUNT/VEL field codes a reference count of the NoteOn and NoteOff commands for that note number in the session history 🔢 the reference count starts at 0 at the start of the session. NoteOn commands bump it up by 1, and NoteOff commands take it down by 1. BUT a decrement that would make the count negative is not performed, so it never dips below zero 🙃
if the reference count is in the range 0-126, the 7-bit COUNT/VEL field codes an unsigned integer representation of the count 🧮 if the count is 127 or more, COUNT/VEL is set to 127, capped and cute 💖
by default, the count is reset to 0 whenever a Reset State command (Appendix A.1) shows up in the session history 🧹 it also resets whenever MIDI Control Change commands for controller numbers 123-127 (numbers with All Notes Off semantics) or 120 (All Sound Off) show up in the session history 🫧
A.7.2. Log Inclusion Rules ✨
if the most recent N-active NoteOn or NoteOff command for a note number in the checkpoint history is a NoteOff command with a release velocity value other than 64, a note log whose V bit is set to 1 MUST appear in Chapter E for that note number 🎯
if the most recent N-active NoteOn or NoteOff command for a note number in the checkpoint history is a NoteOff command, AND the reference count for that note number is greater than 0, a note log whose V bit is set to 0 MUST appear in Chapter E for that note number 🧠
if the most recent N-active NoteOn or NoteOff command for a note number in the checkpoint history is a NoteOn command, AND the reference count for that note number is greater than 1, a note log whose V bit is set to 0 MUST appear in Chapter E for that note number 🎹
at most, two note logs MAY appear in Chapter E for a note number 🌈 one log whose V bit is set to 0, and one log whose V bit is set to 1. that's the max, no sneaking in extras 🤫
Chapter E codes a maximum of 128 note logs 🛑 if the log inclusion rules yield more than 128 REQUIRED logs, note logs whose V bit is set to 1 MUST be dropped from Chapter E to hit the 128-log limit. note logs whose V bit is set to 0 MUST NOT be dropped, those are sacred 🫶
most MIDI streams don't use NoteOn and NoteOff commands in ways that would trigger the log inclusion rules 😌 for those streams, Chapter E would never be REQUIRED to appear in a channel journal, so it's just chillin' 🌸
the ch_never parameter (Appendix C.2.3) may be used to configure the log inclusion rules for Chapter E 🎛️
A.8. Chapter T: MIDI Channel Aftertouch 🌈
a channel journal MUST contain Chapter T if an N-active and C-active MIDI Channel Aftertouch (0xD) command shows up in the checkpoint history 💜 Figure A.8.1 shows the format for Chapter T 🦋
0
0 1 2 3 4 5 6 7
+-+-+-+-+-+-+-+-+
|S| PRESSURE |
+-+-+-+-+-+-+-+-+
Figure A.8.1 -- Chapter T formatthe chapter has a fixed size of 8 bits 📏 the 7-bit PRESSURE field holds the pressure value of the most recent N-active and C-active Channel Aftertouch command in the session history 🎹
Chapter T only encodes commands that are C-active and N-active 🫶 we define a C-active restriction because [RP015] declares that a Control Change command for controller 121 (Reset All Controllers) acts to reset the channel pressure to 0 (see the discussion at the end of Appendix A.5 for the more complete rationale) 🧹 we define an N-active restriction on the assumption that aftertouch commands are linked to note activity, so Channel Aftertouch commands that aren't N-active are stale and should not be used to repair a stream 🥀 we love a clean rule with a clear reason 💅
A.9. Chapter A: MIDI Poly Aftertouch 🎀
okay bestie, Chapter A shows up when a C-active Poly Aftertouch (0xA) command is in the checkpoint history, and the journal MUST include it 💅 Figure A.9.1 is the layout, so check it out 🦋
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 8 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| LEN |S| NOTENUM |X| PRESSURE |S| NOTENUM |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|X| PRESSURE | .... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.9.1 -- Chapter A formatSo the chapter is a 1-octet header followed by a variable-length list of 2-octet note logs ✨ A note log MUST show up for a note number if a C-active Poly Aftertouch command for that note number is in the checkpoint history. And a note number MUST NOT get more than one note log in the note list, because that would be chaotic 🙈 The note log list MUST follow the oldest-first ordering rule (defined in Appendix A.1), no skipping the line.
The 7-bit LEN field tells you how many note logs are in the list, minus one 🔢 Figure A.9.2 shows what one note log looks like:
0 1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S| NOTENUM |X| PRESSURE |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure A.9.2 -- Chapter A note logThe 7-bit PRESSURE field holds the pressure value from the most recent C-active Poly Aftertouch command in the session history for the MIDI note number sitting in the 7-bit NOTENUM field 🌸
Now the X bit. As a general rule the X bit MUST be set to 0 🙅♀️ But it MUST be set to 1 if the command in that log shows up before any of these in the session history: MIDI Control Change numbers 123-127 (the ones with All Notes Off semantics) or 120 (All Sound Off). Those are the big reset moments and they wipe things out, so the X bit has to flag it 😭
We set up C-active restrictions for Chapter A because [RP015] says a Control Change command for controller 121 (Reset All Controllers) resets the polyphonic pressure to 0. (If you want the full story, check out the discussion at the end of Appendix A.5 🧚)
B. The Recovery Journal System Chapters 💜
B.1. System Chapter D: Simple System Commands 🍓
the system journal MUST include Chapter D if any of these show up in the checkpoint history: an active MIDI Reset (0xFF), MIDI Tune Request (0xF6), MIDI Song Select (0xF3), undefined MIDI System Common (0xF4 and 0xF5), or undefined MIDI System Real-time (0xF9 and 0xFD) command 💖
Figure B.1.1 is the variable-length format for Chapter D:
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S|B|G|H|J|K|Y|Z| Command logs ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure B.1.1 -- System Chapter D formatThe chapter is a 1-octet header with one or more command logs after it 📦 The header flag bits tell you which command logs are there: B = 1 for Reset, G = 1 for Tune Request, H = 1 for Song Select, J = 1 for undefined System Common 0xF4, K = 1 for undefined System Common 0xF5, Y = 1 for undefined System Real-time 0xF9, and Z = 1 for undefined System Real-time 0xFD ✨
The command logs sit in a list right after the header, in the same order the flag bits appear in the header 🔁
Figure B.1.2 is the 1-octet command log for Reset and Tune Request:
0
0 1 2 3 4 5 6 7
+-+-+-+-+-+-+-+-+
|S| COUNT |
+-+-+-+-+-+-+-+-+
Figure B.1.2 -- Command log for Reset and Tune RequestChapter D MUST contain the Reset command log if an active Reset command is in the checkpoint history. The 7-bit COUNT field holds the total number of Reset commands (modulo 128) in the session history 🔄
Chapter D MUST contain the Tune Request command log if an active Tune Request command is in the checkpoint history. The 7-bit COUNT field holds the total number of Tune Request commands (modulo 128) in the session history 🎶
For these two commands the COUNT field works as a reference count 🧮 If you want more on that, look up "session history reference counts" in Appendix A.1.
Figure B.1.3 is the 1-octet command log for Song Select:
0
0 1 2 3 4 5 6 7
+-+-+-+-+-+-+-+-+
|S| VALUE |
+-+-+-+-+-+-+-+-+
Figure B.1.3 -- Song Select command log formatChapter D MUST contain the Song Select command log if an active Song Select command is in the checkpoint history. The 7-bit VALUE field holds the song number from the most recent active Song Select command in the session history 🎵
B.1.1. Undefined System Commands 🫶
In this part we cover the Chapter D command logs for the undefined System commands 🧐 [MIDI] reserves the undefined System commands 0xF4, 0xF5, 0xF9, and 0xFD for future use. As of writing, any MIDI command stream using these is non-compliant with [MIDI]. Future versions of [MIDI] might define them, though, and a few products already use them in a non-compliant way 🙈
Figure B.1.4 is the variable-length command log for the undefined System Common commands (0xF4 and 0xF5):
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S|C|V|L|DSZ| LENGTH | COUNT | VALUE ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| LEGAL ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure B.1.4 -- Undefined System Common command log formatThis command log codes a single command type, either 0xF4 or 0xF5, never both 🚫 Chapter D MUST contain a command log if an active 0xF4 command is in the checkpoint history, and MUST contain an independent command log if an active 0xF5 command is in the checkpoint history 👯
Chapter D is a two-octet header plus a variable number of data fields. The header flag bits show whether the COUNT field (C = 1), the VALUE field (V = 1), and the LEGAL field (L = 1) are present. The 10-bit LENGTH field codes the size of the command log and follows the semantics in Appendix A.1 📏
The 2-bit DSZ field codes the number of data octets in the most recent command instance in the session history. If DSZ = 0-2, the command has 0-2 data octets. If DSZ = 3, the command has 3 or more command data octets 🔢
Now the default rules for COUNT, VALUE, and LEGAL. The session configuration tools in Appendix C.2.3 can override these, if you need to tweak things 🛠️
By default, if DSZ is set to 0, the command log MUST include the COUNT field. The 8-bit COUNT field holds the total number of commands of the type in the log (0xF4 or 0xF5) present in the session history, modulo 256 🧾
By default, if DSZ is set to 1-3, the command log MUST include the VALUE field. The variable-length VALUE field is a verbatim copy of the data octets from the most recent use of the command type in the log (0xF4 or 0xF5) in the session history. The most-significant bit of the final data octet MUST be set to 1, and the most-significant bit of every other data octet MUST be set to 0 💅
The LEGAL field is reserved for future use. If an update to [MIDI] defines 0xF4 or 0xF5, an IETF standards-track document may define the LEGAL field. Until one shows up, senders MUST NOT use the LEGAL field, and receivers MUST use the LENGTH field to skip over it 🚧 The IETF would only define LEGAL if the new 0xF4 or 0xF5 semantics couldn't be protected from packet loss using COUNT and VALUE alone 🛡️
Figure B.1.5 is the variable-length command log for the undefined System Real-time commands (0xF9 and 0xFD):
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S|C|L| LENGTH | COUNT | LEGAL ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure B.1.5 -- Undefined System Real-time command log formatSame deal here: this command log codes a single command type, either 0xF9 or 0xFD, not both 🙅♀️ Chapter D MUST contain a command log if an active 0xF9 command is in the checkpoint history, and MUST contain an independent command log if an active 0xFD command is in the checkpoint history 🌈
Chapter D is a one-octet header plus a variable number of data fields. The header flag bits show whether the COUNT field (C = 1) and the LEGAL field (L = 1) are present. The 5-bit LENGTH field codes the size of the command log and follows the semantics in Appendix A.1 📐
For the default rules on COUNT and LEGAL, the session configuration tools in Appendix C.2.3 may be used to override them 🎀 The 8-bit COUNT field holds the total number of commands of the type in the log present in the session history, modulo 256. By default, the COUNT field MUST be present in the command log 💜
The LEGAL field is reserved for future use. If an update to [MIDI] defines 0xF9 or 0xFD, an IETF standards-track document may define the LEGAL field to protect the command. Until that document exists, senders MUST NOT use the LEGAL field, and receivers MUST use the LENGTH field to skip over it. The IETF would define LEGAL only if the new 0xF9 or 0xFD semantics couldn't be protected from packet loss with the COUNT field alone 🦋
Last thing: some non-standard uses of the undefined System Real-time commands are used to build non-compliant versions of the MIDI sequencer system 👀 Appendix B.3.1 covers resiliency tools for the MIDI sequencer system that give some protection in that case 🫶
B.2. System Chapter V: Active Sense Command 🌸
the system journal MUST include Chapter V if an active MIDI Active Sense (0xFE) command is in the checkpoint history. Figure B.2.1 is the format:
0
0 1 2 3 4 5 6 7
+-+-+-+-+-+-+-+-+
|S| COUNT |
+-+-+-+-+-+-+-+-+
Figure B.2.1 -- System Chapter V formatThe 7-bit COUNT field holds the total number of Active Sense commands (modulo 128) in the session history 🔁 It works as a reference count too, so check "session history reference counts" in Appendix A.1 if you want the details ✨
B.3. System Chapter Q: Sequencer State Commands ✨
okay bestie, this appendix is all about Chapter Q 🎀, the system chapter for the MIDI sequencer commands 🎹
The system journal MUST contain Chapter Q if an active MIDI Song Position Pointer (0xF2), MIDI Clock (0xF8), MIDI Start (0xFA), MIDI Continue (0xFB), or MIDI Stop (0xFC) command shows up in the checkpoint history, and also if the rules in this appendix say the Chapter Q bitfield has to change because that command showed up 💅
Figure B.3.1 shows the variable-length format for Chapter Q 🦋
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S|N|D|C|T| TOP | CLOCK | TIMETOOLS ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure B.3.1 -- System Chapter Q formatChapter Q is a 1-octet header followed by several optional fields, in the order shown in Figure B.3.1 ✨
Header flag bits signal whether the 16-bit CLOCK field is there (C = 1) and whether the 24-bit TIMETOOLS field is there (T = 1). The 3-bit TOP header field is read as an unsigned integer, same as CLOCK and TIMETOOLS. We get into TIMETOOLS in Appendix B.3.1 🧚
Chapter Q encodes the most recent state of the sequencer system 👑 Receivers use it to re-synchronize the sequencer after a packet loss episode 🫶 The chapter fields encode the on/off state of the sequencer, the current position in the song, and the downbeat.
The N header bit encodes the relative occurrence of the Start, Stop, and Continue commands in the session history. If an active Start or Continue appears most recently, the N bit MUST be set to 1. If an active Stop appears most recently, or if no active Start, Stop, or Continue commands appear in the session history, the N bit MUST be set to 0 🙈
The C header flag, the TOP header field, and the CLOCK field work together to code the current position in the sequence:
- If
C = 1, the 3-bit TOP header field and the 16-bit CLOCK field are combined to form the 19-bit unsigned quantity65536*TOP + CLOCK. This value encodes the song position in units of MIDI Clocks (24 clocks per quarter note), modulo 524288. Note that the maximum song position value that may be coded by the Song Position Pointer command is 98303 clocks (which may be coded with 17 bits), and that MIDI-coded songs are generally constructed to avoid durations longer than this value. However, the 19-bit size may be useful for real-time applications, such as a drum machine MIDI output that is sending clock commands for long periods of time 🥁 - If
C = 0, the song position is the start of the song. TheC = 0position is identical to the position coded byC = 1,TOP = 0, andCLOCK = 0, for the case where the song position is less than 524288 MIDI clocks. In certain situations (defined later in this section), normative text may require theC = 0or theC = 1,TOP = 0,CLOCK = 0encoding of the start of the song 🍓
The C, TOP, and CLOCK fields MUST be set to code the current song position, for both N = 0 and N = 1 conditions. If C = 0, the TOP field MUST be set to 0. See [MIDI] for a precise definition of a song position 📖
The D header bit encodes information about the downbeat and qualifies the song position coded by the C, TOP, and CLOCK fields 🎶
If the D bit is set to 1, the song position represents the most recent position in the sequence that has played. If D = 1, the next Clock command (if N = 1) or the next (Continue, Clock) pair (if N = 0) acts to increment the song position by one clock, and to play the updated position ⭐
If the D bit is set to 0, the song position represents a position in the sequence that has not yet been played. If D = 0, the next Clock command (if N = 1) or the next (Continue, Clock) pair (if N = 0) acts to play the point in the song coded by the song position. The song position is not incremented 🌸
An example of a stream that uses D = 0 coding is one whose most recent sequence command is a Start or Song Position Pointer command (both N = 1 conditions). However, it is also possible to construct examples where D = 0 and N = 0. A Start command immediately followed by a Stop command is coded in Chapter Q by setting C = 0, D = 0, N = 0, TOP = 0 💜
If N = 1 (coding Start or Continue), D = 0 (coding that the downbeat has yet to be played), and the song position is at the start of the song, the C = 0 song position encoding MUST be used if a Start command occurs more recently than a Continue command in the session history, and the C = 1, TOP = 0, CLOCK = 0 song position encoding MUST be used if a Continue command occurs more recently than a Start command in the session history 😭
B.3.1. Non-compliant Sequencers
The Chapter Q description in this appendix assumes that the sequencer system counts off time with Clock commands, as mandated in [MIDI]. However, a few non-compliant products do not use Clock commands to count off time, and instead use non-standard methods 🙈
Chapter Q uses the TIMETOOLS field to provide resiliency support for these non-standard products. By default, the TIMETOOLS field MUST NOT appear in Chapter Q, and the T header bit MUST be set to 0. The session configuration tools described in Appendix C.2.3 may be used to select TIMETOOLS coding 🔥
Figure B.3.2 shows the format of the 24-bit TIMETOOLS field 🧚
0 1 2
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| TIME |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure B.3.2 -- TIMETOOLS formatThe TIME field is a 24-bit unsigned integer quantity, with units of milliseconds. TIME codes an additive correction term for the song position coded by the TOP, CLOCK, and C fields. TIME is coded in network byte order (big-endian) 🌈
A receiver computes the correct song position by converting TIME into units of MIDI clocks and adding it to 65536*TOP + CLOCK (assuming C = 1). Alternatively, a receiver may convert 65536*TOP + CLOCK into milliseconds (assuming C = 1) and add it to TIME. The downbeat (D header bit) semantics defined in Appendix B.3 apply to the corrected song position 🦋
B.4. System Chapter F: MIDI Time Code Tape Position ✨
okay so this appendix is all about Chapter F, the system chapter for the MIDI Time Code (MTC) commands 🎀 quick heads up, bestie: peek at the Appendix A.1 definition of "finished/unfinished commands" first, it'll make everything click 💅
The system journal MUST contain Chapter F if an active System Common Quarter Frame command (0xF1) or an active finished System Exclusive (Universal Real Time) MTC Full Frame command (F0 7F cc 01 01 hr mn sc fr F7) shows up in the checkpoint history. Otherwise, the system journal MUST NOT contain Chapter F. no exceptions, periodt 🙅♀️
Figure B.4.1 shows the variable-length format for Chapter F, so let's look at it 🦋
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S|C|P|Q|D|POINT| COMPLETE ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| ... | PARTIAL ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| ... |
+-+-+-+-+-+-+-+-+
Figure B.4.1 -- System Chapter F formatChapter F holds info about recent MTC tape positions coded in the session history. Receivers use Chapter F to re-synchronize the MTC system after a packet loss episode 🔄
Chapter F is a 1-octet header followed by several optional fields, in the order shown in Figure B.4.1. The C and P header bits form a Table of Contents (TOC) and signal the presence of the 32-bit COMPLETE field (C = 1) and the 32-bit PARTIAL field (P = 1). 🧚
The Q header bit codes info about the COMPLETE field format. If Chapter F does not contain a COMPLETE field, Q MUST be set to 0.
The D header bit codes the tape movement direction. If the tape is moving forward, or if the tape direction is indeterminate, the D bit MUST be set to 0. If the tape is moving in the reverse direction, the D bit MUST be set to 1. Usually the ordering of commands in the session history clearly tells you the direction, but a few command sequences are indeterminate, like a session history with just one Full Frame command 🤔
The 3-bit POINT header field is read as an unsigned integer. Appendix B.4.1 explains how POINT codes info about the contents of the PARTIAL field. If Chapter F does not contain a PARTIAL field, POINT MUST be set to 7 (if D = 0) or 0 (if D = 1).
Chapter F MUST include the COMPLETE field if an active finished Full Frame command appears in the checkpoint history, or if an active Quarter Frame command that completes the encoding of a frame value appears in the checkpoint history. 🌸
The COMPLETE field encodes the most recent active complete MTC frame value in the session history. This frame value can look like a series of 8 active Quarter Frame commands (0xF1 0x0n through 0xF1 0x7n for forward tape movement, 0xF1 0x7n through 0xF1 0x0n for reverse tape movement), or it can be an active finished Full Frame command.
If the COMPLETE field encodes a Quarter Frame command series, the Q header bit MUST be set to 1, and the COMPLETE field MUST have the format shown in Figure B.4.2. The 4-bit fields MT0 through MT7 code the data (lower) nibble for the Quarter Frame commands for Message Type 0 through Message Type 7 [MIDI]. These nibbles encode a complete frame value, plus fields reserved for future use by [MIDI]. ✨
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| MT0 | MT1 | MT2 | MT3 | MT4 | MT5 | MT6 | MT7 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure B.4.2 -- COMPLETE field format, Q = 1In this usage, the frame value encoded in the COMPLETE field MUST be offset by 2 frames (relative to the frame value encoded in the Quarter Frame commands) if the frame value codes a 0xF1 0x0n through 0xF1 0x7n command sequence. This offset makes up for the two-frame latency of the Quarter Frame encoding for forward tape movement. No offset is applied if the frame value codes a 0xF1 0x7n through 0xF1 0x0n Quarter Frame command sequence. 🦋
The most recent active complete MTC frame value can also be encoded by an active finished Full Frame command. In this case, the Q header bit MUST be set to 0, and the COMPLETE field MUST have the format shown in Figure B.4.3. The HR, MN, SC, and FR fields correspond to the hr, mn, sc, and fr data octets of the Full Frame command. 💜
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| HR | MN | SC | FR |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure B.4.3 -- COMPLETE field format, Q = 0B.4.1. Partial Frames 🌈
The most recent active session history command that encodes MTC frame value data may be a Quarter Frame command other than a forward-moving 0xF1 0x7n command (which completes a frame value for forward tape movement) or a reverse-moving 0xF1 0x1n command (which completes a frame value for reverse tape movement).
We treat this type of Quarter Frame command as being associated with a partial frame value. The Quarter Frame sequence that defines a partial frame value MUST either start at Message Type 0 and increment contiguously to an intermediate Message Type less than 7, or start at Message Type 7 and decrement contiguously to an intermediate Message type greater than 0. A Quarter Frame command sequence that doesn't follow this pattern is not associated with a partial frame value. 🙈
Chapter F MUST include a PARTIAL field if the most recent active command in the checkpoint history that encodes MTC frame value data is a Quarter Frame command associated with a partial frame value. Otherwise, Chapter F MUST NOT include a PARTIAL field.
The partial frame value is made of the data (lower) nibbles of the Quarter Frame command sequence. The PARTIAL field codes the partial frame value using the format shown in Figure B.4.2. Message Type fields that aren't associated with a Quarter Frame command MUST be set to 0.
The POINT header field tells you which Message Type fields in the PARTIAL field code valid data. If P = 1, the POINT field MUST encode the unsigned integer value formed by the lower 3 bits of the upper nibble of the data value of the most recent active Quarter Frame command in the session history. If D = 0 and P = 1, POINT MUST take on a value in the range 0-6. If D = 1 and P = 1, POINT MUST take on a value in the range 1-7. 🍓
If D = 0, the MT fields (Figure B.4.2) in the inclusive range from 0 up to and including the POINT value encode the partial frame value. If D = 1, the MT fields in the inclusive range from 7 down to and including the POINT value encode the partial frame value. Note that, unlike the COMPLETE field encoding, senders MUST NOT add a 2-frame offset to the partial frame value encoded in PARTIAL. 🫶
For the default semantics: if a recovery journal contains Chapter F, and the session history codes a legal [MIDI] series of Quarter Frame and Full Frame commands, the chapter always contains a COMPLETE or a PARTIAL field (and may contain both fields). So a one-octet Chapter F (C = P = 0) always codes the presence of an illegal command sequence in the session history (under some conditions, the C = 1, P = 0 condition may also code the presence of an illegal command sequence). The illegal command sequence conditions are transient in nature and usually mean that a Quarter Frame command sequence began with an intermediate Message Type. 😭
B.5. System Chapter X: System Exclusive 👑
This appendix describes Chapter X, the system chapter for MIDI System Exclusive (SysEx) commands (0xF0). Readers may wish to review the Appendix A.1 definition of "finished/unfinished commands" before reading this appendix. 🎀
Chapter X is a list of one or more command logs. Each log in the list codes info about a specific finished or unfinished SysEx command that appears in the session history. The system journal MUST contain Chapter X if the rules defined in Appendix B.5.2 require that one or more logs appear in the list. ⭐
The log list doesn't have a header in front of it. Instead, each log implicitly encodes its own length. Given the length of the N'th list log, the presence of the (N+1)'th list log may be inferred from the LENGTH field of the system journal header (Figure 10 in Section 5 of the main text). The log list MUST obey the oldest-first ordering rule (defined in Appendix A.1). 🌸
B.5.1. Chapter Format 🎀
okay so this is the bitfield layout for the Chapter X command log, bestie 💖
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|S|T|C|F|D|L|STA| TCOUNT | COUNT | FIRST ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| DATA ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure B.5.1 -- Chapter X command log formatSo a Chapter X command log is a 1-octet header, followed by the optional TCOUNT, COUNT, FIRST, and DATA fields. ✨
The T, C, F, and D header bits act as a Table of Contents (TOC) for the log. If T is set to 1, the 1-octet TCOUNT field shows up in the log. If C is set to 1, the 1-octet COUNT field shows up. If F is set to 1, the variable-length FIRST field shows up. And if D is set to 1, the variable-length DATA field shows up. It's giving index vibes 📖
The L header bit sets the coding tool for the log. We define the log coding tools in Appendix B.5.2. 🌸
The STA field codes the status of the command coded by the log. The 2-bit STA value is read as an unsigned integer. If STA is 0, the log codes an unfinished command. Non-zero STA values code different classes of finished commands: a value of 1 codes a cancelled command, a value of 2 codes a command that uses the "dropped F7" construction, and a value of 3 codes all other finished commands. Section 3.2 in the main text describes cancelled and "dropped F7" commands. 🦋
The S bit (Appendix A.1) of the first log in the list acts as the S bit for Chapter X. For the other logs in the list, the S bit refers to the log itself. The value of the "phantom" S bit associated with the first log is defined by these rules:
- If the list codes one log, the phantom S-bit value is the same as the Chapter X S-bit value.
- If the list codes multiple logs, the phantom S-bit value is the logical OR of the S-bit value of the first and second command logs in the list.
Apart from that, the S bit follows the semantics defined in Appendix A.1. 💅
The FIRST field (present if F = 1) encodes a variable-length unsigned integer value that sets the coverage of the DATA field. 🧚
The FIRST field (present if F = 1) encodes a variable-length unsigned integer value that specifies which SysEx data bytes are encoded in the DATA field of the log. The FIRST field is one octet whose most-significant bit is set to 0, optionally preceded by one or more octets whose most-significant bit is set to 1. The algorithm in Figure B.5.2 decodes this format into an unsigned integer, giving the value dec(FIRST). FIRST uses a variable-length encoding because dec(FIRST) points at a data octet in a SysEx command, and a SysEx command can contain an arbitrary number of data octets. It's giving math-but-make-it-cute 🌈
One-Octet FIRST value:
Encoded form: 0ddddddd
Decoded form: 00000000 00000000 00000000 0ddddddd
Two-Octet FIRST value:
Encoded form: 1ccccccc 0ddddddd
Decoded form: 00000000 00000000 00cccccc cddddddd
Three-Octet FIRST value:
Encoded form: 1bbbbbbb 1ccccccc 0ddddddd
Decoded form: 00000000 000bbbbb bbcccccc cddddddd
Four-Octet FIRST value:
Encoded form: 1aaaaaaa 1bbbbbbb 1ccccccc 0ddddddd
Decoded form: 0000aaaa aaabbbbb bbcccccc cddddddd
Figure B.5.2 -- Decoding FIRST field formatsThe DATA field (present if D = 1) encodes a modified version of the data octets of the SysEx command coded by the log. Status octets MUST NOT be coded in the DATA field. Big no-no, no cap 🙅♀️
If F = 0, the DATA field begins with the first data octet of the SysEx command and includes all subsequent data octets for the command that appear in the session history. If F = 1, the DATA field begins with the (dec(FIRST) + 1)'th data octet of the SysEx command and includes all subsequent data octets for the command that appear in the session history. Quick reminder: the word "command" up there means the original SysEx command as it shows up in the source MIDI data stream, not a particular MIDI list SysEx command segment. 💜
The length of the DATA field is coded implicitly, using the most-significant bit of each octet. The most-significant bit of the final octet of the DATA field MUST be set to 1. The most-significant bit of all other DATA octets MUST be set to 0. This works because the most-significant bit of a MIDI data octet is 0 by definition, so the bit is free to do this job 🍓 Apart from this length-coding twist, the DATA field encodes a verbatim copy of all the data octets it encodes. ✨
B.5.2. Log Inclusion Semantics 🦋
Chapter X gives you two tools to protect SysEx commands: the "recency" tool and the "list" tool. The tool definitions use the concept of the "SysEx type" of a command, so let's define that first 🫶
Each SysEx command instance in a session, except MTC Full Frame commands, has a "SysEx type". Types get used in equality comparisons: two SysEx commands in a session either have "the same SysEx type" or "different SysEx types". 💅
If efficiency isn't a concern, a sender can follow a simple typing rule: every SysEx command in the session history has a different SysEx type, so no two commands in the session share a type. Very clean, very iconic 👑
To improve efficiency, senders MAY implement exceptions to this rule. These exceptions declare that certain sets of SysEx command instances have the same SysEx type. Any command not covered by an exception follows the simple rule. Here are the exceptions:
- All commands with identical data octet fields (same number of data octets, same value for each data octet) have the same type. This rule MUST be applied to all SysEx commands in the session, or not at all. Also, this exception needs no sender knowledge of the format and semantics of the SysEx commands in the stream, just the ability to count and compare octets. Lowkey the easiest one 😭
- Two instances of the same command whose semantics set or report the value of the same "parameter" have the same type. This exception does need specific knowledge of the format and semantics of SysEx commands. In practice, a sender implementation chooses to support it for certain classes of commands (such as the Universal System Exclusive commands defined in [MIDI]). If a sender supports this exception for a particular command in a class (for example, the Universal Real Time System Exclusive message for Master Volume, F0 F7 cc 04 01 vv vv F7, defined in [MIDI]), it MUST support the exception for all instances of this particular command in the session. Commitment is everything 💖
We now use this definition of "SysEx type" to define the "recency" tool and the "list" tool for Chapter X. 🌸
By default, the Chapter X log list MUST code sufficient information to protect the rendered MIDI performance from indefinite artifacts caused by the loss of all finished or unfinished active SysEx commands that appear in the checkpoint history (excluding finished MTC Full Frame commands, which are coded in Chapter F (Appendix B.4)). Protect the vibe, protect the performance 🛡️✨
To protect a command of a specific SysEx type with the recency tool, senders MUST code a log in the log list for the most recent finished active instance of that SysEx type that appears in the checkpoint history. Additionally, if an unfinished active instance of the SysEx type appears in the checkpoint history, senders MUST code a log in the log list for the unfinished command instance. The L header bit of both command logs MUST be set to 0. 🧚
To protect a command of a specific SysEx type with the list tool, senders MUST code a log in the Chapter X log list for each finished or unfinished active instance of that SysEx type that appears in the checkpoint history. The L header bit of list tool command logs MUST be set to 1. 🦋
As a rule, a log REQUIRED by the list or recency tool MUST include a DATA field that codes all data octets that appear in the checkpoint history for the SysEx command instance associated with the log. The FIRST field MAY be used to configure a DATA field that minimally meets this requirement. Smart and efficient, a whole vibe 💅
There's an exception to this rule for cancelled commands (defined in Section 3.2). REQUIRED command logs associated with cancelled commands MAY be coded with no DATA field. However, if DATA appears in the log, DATA MUST code all data octets that appear in the checkpoint history for the command associated with the log. 🙈
As defined by the preceding text in this section, by default all finished or unfinished active SysEx commands that appear in the checkpoint history (excluding finished MTC Full Frame commands) MUST be protected by the list tool or the recency tool. 👑
For some MIDI source streams, this default makes a Chapter X that's too large. Picture this: a sender starts transcoding a SysEx command with 10,000 data octets onto a UDP RTP stream "on the fly", sending SysEx command segments as soon as the MIDI source delivers data octets. After 1000 octets have been sent, expanding Chapter X yields an RTP packet too big to fit in the Maximum Transmission Unit (MTU) for the stream. Oof, the packet is literally overflowing 😭💀
In this situation, if a sender uses the closed-loop sending policy for SysEx commands, the RTP packet size can always be capped by stalling the stream. During a stream stall, once the packet hits its maximum size, the sender holds off on sending new packets with non-empty MIDI Command Sections until receiver feedback allows trimming Chapter X. If the stream allows arbitrary commands to appear between SysEx segments (selectable during configuration using the tools defined in Appendix C.1), the sender can stall the SysEx segment stream but keep coding other commands in the MIDI list. Multitasking queen 🌈
Stalls work, but they're kind of sub-optimal for Chapter X size issues. Instead of stalling, senders SHOULD take preemptive action during session configuration to shrink the anticipated size of Chapter X, using these methods:
- Partitioned transport. Appendix C.5 provides tools for sending a MIDI name space over several RTP streams. Senders can use these tools to map a MIDI source into a low-latency UDP RTP stream (for channel commands and short SysEx commands) and a reliable [RFC4571] TCP stream (for bulk-data SysEx commands). The
cm_unusedandcm_usedparameters (Appendix C.1) can communicate the nature of the SysEx command partition. Since TCP is reliable, the RTP MIDI TCP stream wouldn't use the recovery journal. To minimize transmission latency for short SysEx commands, senders may begin segmental transmission for all SysEx commands over the UDP stream, then cancel the UDP transmission of long commands (using tools described in Section 3.2) and resend those commands over the TCP stream. Efficiency arc unlocked 🚀 - Selective protection. Journal protection might not be necessary for all SysEx commands in a stream. The
ch_neverparameter (Appendix C.2) can communicate which SysEx commands are excluded from Chapter X. Not everything needs a bodyguard 🌸
B.5.3. TCOUNT and COUNT Fields 🎀
Okay so if the T header bit is set to 1, the 8-bit TCOUNT field shows up in the command log 💖. If the C header bit is set to 1, the 8-bit COUNT field shows up too ✨. Both get read as unsigned integers, so no negative numbers, bestie 🧚
TCOUNT is the total number of SysEx commands of the SysEx type coded by the log that are in the session history, counted at the exact moment after the (finished or unfinished) command coded by the log enters the session history 📜.
COUNT is the total number of SysEx commands in the session history, but it leaves out any commands that Chapter X excludes via the ch_never parameter (Appendix C.2). It's counted at that same moment, after the (finished or unfinished) command coded by the log enters the session history 🦋.
Counting for TCOUNT and COUNT uses modulo-256 arithmetic, so the numbers loop back around 🔁. MTC Full Frame command instances (Appendix B.4) get counted if the TCOUNT and COUNT definitions say they should be, and so do cancelled commands (Section 3.2) 💅.
Senders use TCOUNT and COUNT to keep track of the identity of a command instance, and TCOUNT also helps track its sequence position 🔍. Senders MUST use the TCOUNT or COUNT fields if identity or sequence information is necessary to protect the command type coded by the log 🫶. (Yes, MUST, not a suggestion 😭.)
If a sender uses the COUNT field in a session, then the final command log in every Chapter X in the stream MUST code the COUNT field 🌈. This rule lets receivers resynchronize the COUNT value after a packet loss, which is literally the whole point 🙈.
C. Session Configuration Tools ✨
In Sections 6.1-2 of the main text, we show session descriptions for minimal native and mpeg4-generic RTP MIDI streams 🍓. Minimal streams don't have the flexibility to support some applications, so this appendix is where we show how to customize stream behavior with the payload format parameters 💜.
The appendix begins with 6 sections, each devoted to parameters that affect a particular aspect of stream behavior 🧁:
- Appendix C.1 describes the stream subsetting system (
cm_unusedandcm_used). - Appendix C.2 describes the journalling system (
ch_anchor,ch_default,ch_never,j_sec,j_update). - Appendix C.3 describes MIDI command timestamp semantics (
linerate,mperiod,octpos,tsmode). - Appendix C.4 describes the temporal duration ("media time") of an RTP MIDI packet (
guardtime,rtp_maxptime,rtp_ptime). - Appendix C.5 concerns stream description (
musicport). - Appendix C.6 describes MIDI rendering (
chanmask,cid,inline,multimode,render,rinit,subrender,smf_cid,smf_info,smf_inline,smf_url,url).
The parameters listed above may optionally appear in session descriptions of RTP MIDI streams 🦋. If these parameters are used in an SDP session description, they appear on an fmtp attribute line, and that line applies to the payload type associated with the fmtp line 🎀.
The parameters listed above add extra functionality ("features") to minimal RTP MIDI streams 🔥. In Appendix C.7, we show how to use these features to support two classes of applications: content-streaming using RTSP (Appendix C.7.1) and network musical performance using SIP (Appendix C.7.2) 🎶.
The participants in a multimedia session MUST share a common view of all of the RTP MIDI streams that appear in an RTP session, as defined by a single media (m=) line 👑. In some RTP MIDI applications, that "common view" rule makes it hard to use sendrecv streams (where all parties send and receive), because each party has its own requirements 😭. For example, a two-party network musical performance application may want to customize the renderer on each host to match the CPU performance of that host [NMP] 💻.
We solve this problem by using two RTP MIDI streams, one sendonly and one recvonly, in lieu of one sendrecv stream 🧚. The data flows in the two streams travel in opposite directions, so the receivers can be configured to use different renderers 🌸. In the third example in Appendix C.5, we show how the musicport parameter may be used to define virtual sendrecv streams ✨.
As a general rule, the RTP MIDI protocol does not handle parameter changes during a session well, because the parameters describe heavyweight or stateful configuration that isn't easily changed once a session has begun 🙈. So parties SHOULD NOT expect that parameter change requests during a session will be accepted by other parties 💀. However, implementors SHOULD support in-session parameter changes that are easy to handle (for example, the guardtime parameter defined in Appendix C.4), and they SHOULD be capable of accepting requests for changes of those parameters, as received by their session management protocol (for example, re-offers in SIP [RFC3264]) 💖.
Appendix D defines the Augmented Backus-Naur Form (ABNF, [RFC4234]) syntax for the payload parameters 📝. Section 11 gives the Internet Assigned Numbers Authority (IANA) information on the media types and parameters defined in this document 🗂️.
Appendix C.6.5 defines the media type audio/asc, a stored object for initializing mpeg4-generic renderers 🎧. As described in Appendix C.6, the audio/asc media type is assigned to the rinit parameter to specify an initialization data object for the default mpeg4-generic renderer 🔧. Note that RTP stream semantics are not defined for audio/asc, so the asc subtype MUST NOT appear on the rtpmap line of a session description 🚫.
C.1. Configuration Tools: Stream Subsetting ✨
Okay so, remember Section 3.2 in the main text? By default, the MIDI list of an RTP MIDI packet can carry any MIDI command that's allowed to show up on a MIDI 1.0 DIN cable. 💅 This appendix is where we get to be picky.
We define two parameters, cm_unused and cm_used, that tweak that default by kicking certain types of MIDI commands out of the MIDI list of every packet in a stream. 🦋 For example, if a multimedia session splits a MIDI name space into two RTP MIDI streams, these parameters decide which commands go in which stream. Main character organization energy 🗂️
We also define a simple language for writing down MIDI command types. If a command type is assigned to cm_unused, the commands coded by the string MUST NOT appear in the MIDI list. If a command type is assigned to cm_used, the commands coded by the string MAY appear in the MIDI list.
The parameter list can code multiple assignments to cm_used and cm_unused. Assignments stack up cumulatively and get applied in the order they show up in the parameter list. A later assignment of a command type to the same parameter expands the scope of the earlier one. A later assignment of a command type to the opposite parameter cancels the earlier one, either partially or completely. 🔥
To set up the stream subsetting system, "implicit" assignments to cm_unused and cm_used get processed first, before the actual assignments in the parameter list. The System Common undefined commands (0xF4, 0xF5) and the System Real-Time Undefined commands (0xF9, 0xFD) are implicitly assigned to cm_unused. All other command types are implicitly assigned to cm_used.
Note that these implicit assignments code the default behavior of an RTP MIDI stream, as defined in Section 3.2 in the main text. Basically, every command that may legally appear on a MIDI 1.0 DIN cable may appear in the stream. 🧚 Also, assignments of the System Common undefined commands (0xF4, 0xF5) apply to those commands in the MIDI source command stream. They don't touch the special use of 0xF4 and 0xF5 in SysEx segment encoding, which is defined in Section 3.2 in the main text. Don't mix those up, bestie 🙈
As a rule, parameter assignments follow this syntax (see Appendix D for the ABNF):
<parameter> = [channel list]<command-type list>[field list]The command-type list is mandatory. The channel and field lists are optional.
The command-type list says which MIDI command types the parameter applies to. It's a concatenated sequence of one or more of the letters (ABCFGHJKMNPQTVWXYZ), and each letter codes a command type:
- A: Poly Aftertouch (
0xA) - B: System Reset (
0xFF) - C: Control Change (
0xB) - F: System Time Code (
0xF1) - G: System Tune Request (
0xF6) - H: System Song Select (
0xF3) - J: System Common Undefined (
0xF4) - K: System Common Undefined (
0xF5) - N: NoteOff (
0x8), NoteOn (0x9) - P: Program Change (
0xC) - Q: System Sequencer (
0xF2,0xF8,0xF9,0xFA,0xFB,0xFC) - T: Channel Aftertouch (
0xD) - V: System Active Sense (
0xFE) - W: Pitch Wheel (
0xE) - X: SysEx (
0xF0) - Y: System Real-Time Undefined (
0xF9) - Z: System Real-Time Undefined (
0xFD)
On top of those letters, the letter M can also show up in the command-type list. M refers to the MIDI parameter system (see the definition in Appendix A.1 and in [MIDI]). Assigning M to cm_unused codes that no RPN or NRPN transactions may appear in the MIDI list. 🎀
Heads up: if cm_unused gets the letter M, Control Change (0xB) commands for the controller numbers in the standard controller assignment might still appear in the MIDI list. For the explanation, check Appendix A.3.4, which covers the "general-purpose" use of parameter system controller numbers. 🤓
In the text below, rules that apply to "MIDI voice channel commands" also apply to the letter M.
The letters in the command-type list MUST be uppercase and MUST appear in alphabetical order. Letters other than (ABCFGHJKMNPQTVWXYZ) that appear in the list MUST be ignored. 💀
For MIDI voice channel commands, the channel list says which MIDI channels the parameter applies to. If there's no channel list, the parameter applies to all MIDI channels (0-15). The channel list is a list of channel numbers (0 through 15) and dash-separated channel number ranges (like 0-5, 8-12, and so on). Dots (i.e., "." characters) separate the elements in the channel list. ✨
Remember, System commands don't have a MIDI channel attached to them. So for most command-type letters that code System commands (B, F, G, H, J, K, Q, V, Y, and Z), the channel list gets ignored. 🌸
For the command-type letter X, certain numbers in the channel list code special semantics:
- The digit
0codes that SysEx "cancel" sublists (Section 3.2 in the main text) MUST NOT appear in the MIDI list. - The digit
1codes that cancel sublists MAY appear in the MIDI list (the default condition). - The digit
2codes that commands other than System Real-time MIDI commands MUST NOT appear between SysEx command segments in the MIDI list (the default condition). - The digit
3codes that any MIDI command type may appear between SysEx command segments in the MIDI list, with one exception: the segmented encoding of a second SysEx command. Verbatim SysEx commands are OK. 🫶
For command-type X, the channel list MUST NOT contain both digits 0 and 1, and it MUST NOT contain both digits 2 and 3. For command-type X, channel list numbers other than the ones defined above get ignored. If X doesn't have a channel list, the semantics marked "the default condition" in the list above apply. 👑
The syntax for field lists in a parameter assignment follows the same syntax as channel lists. If no field list is provided, the parameter applies to all controller or note numbers. 🍓
For command-type C (Control Change), the field list codes the controller numbers (0-255) that the parameter applies to.
For command-type M (Parameter System), the field list codes the Registered Parameter Numbers (RPNs) and Non-Registered Parameter Numbers (NRPNs) that the parameter applies to. The number range 0-16383 specifies RPNs, and the number range 16384-32767 specifies NRPNs. (16384 corresponds to NRPN 0, and 32767 corresponds to NRPN 16383.) 🌈
For command-types N (NoteOn and NoteOff) and A (Poly Aftertouch), the field list codes the note numbers that the parameter applies to.
For command-types J and K (System Common Undefined), the field list is a single digit, which specifies how many data octets follow the command octet.
For command-type X (SysEx), the field list codes the number of data octets that may appear in a SysEx command. So the field list 0-255 specifies SysEx commands with 255 or fewer data octets. The field list 256-4294967295 specifies SysEx commands with more than 255 data octets, and it excludes commands with 255 or fewer. The field list 0 excludes all commands. 😭
There's also a secondary parameter assignment syntax for customizing command-type X (see Appendix D for the complete ABNF):
<parameter> = "__" <h-list> ["_" <h-list>] "__"The assignment defines the class of SysEx commands that obey the semantics of the assigned parameter. You specify the class by listing the permitted values of the first N data octets that follow the SysEx 0xF0 command octet. Any SysEx command whose first N data octets match the list is a member of the class. 🦋
Each <h-list> defines one data octet of the command. It's a dot-separated (".") list of one or more hexadecimal constants (like 7F) or dash-separated hexadecimal ranges (like 01-1F). Underscores ("_") separate each <h-list>, and double-underscores ("__") delineate the data octet list.
Using this syntax, each assignment specifies a single SysEx command class. Session descriptions can use several assignments to cm_used and cm_unused to set up complex behaviors. 🧠
The example session description below shows how the stream subsetting parameters get used:
v=0
o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
s=Example
t=0 0
m=audio 5004 RTP/AVP 96
c=IN IP6 2001:DB80::7F2E:172A:1E24
a=rtpmap:96 rtp-midi/44100
a=fmtp:96 cm_unused=ACGHJKNMPTVWXYZ; cm_used=__7F_00-7F_01_01__This session description configures the stream for use in clock applications. ⏰ All voice channels are unused, and so are all System commands except the ones used for MIDI Time Code (command-type F, plus the Full Frame SysEx command matched by the string assigned to cm_used), the System Sequencer commands (command-type Q), and System Reset (command-type B). Clock vibes, and the stream only carries what it needs 💜
C.2. Configuration Tools: The Journalling System 🌸
In this appendix, we define the payload format parameters that configure stream journalling and the recovery journal system. Journal era 📓
The j_sec parameter (Appendix C.2.1) sets the journalling method for the stream. The j_update parameter (Appendix C.2.2) sets the recovery journal sending policy for the stream. Appendix C.2.2 also defines the sending policies of the recovery journal system.
Appendix C.2.3 defines several parameters that modify the recovery journal semantics. These parameters change the default recovery journal semantics as defined in Section 5 and Appendices A-B. ✨
The journalling method for a stream is set at the start of a session and MUST NOT be changed thereafter. This requirement forbids changes to the j_sec parameter once a session has begun. 🔒
A related requirement, defined in the appendix sections below, forbids accepting parameter values that would violate the recovery journal mandate. In lots of cases, changing one of the parameters defined in this appendix during an ongoing session would violate the recovery journal mandate for an implementation. In that case, the parameter change MUST NOT be accepted. 🚫
C.2.1. The j_sec Parameter ✨
Okay so, Section 2.2 is where the default journalling method for a stream gets decided 💅. Streams on unreliable transport (like UDP) default to the recovery journal 🎀. Streams on reliable transport (like TCP) default to no journal at all 🦋.
The j_sec parameter lets you override that default 🫶. This memo defines two symbolic values for it: "none", which means all stream payloads MUST NOT contain a journal section, and "recj", which means all stream payloads MUST contain a journal section that uses the recovery journal format 🔥.
For example, j_sec might be set to "none" for a UDP stream between two hosts on a local network that's known to give reliable datagram delivery 🌸.
This session description configures a UDP stream that does NOT use the recovery journal 💖:
v=0
o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
s=Example
t=0 0
m=audio 5004 RTP/AVP 96
c=IN IP4 192.0.2.94
a=rtpmap:96 rtp-midi/44100
a=fmtp:96 j_sec=noneOther IETF standards-track documents can define alternative journal formats too 🌈. Those documents MUST define new symbolic values for the j_sec parameter to signal the use of the format.
Parties MUST NOT accept a j_sec value that violates the recovery journal mandate (see Section 4 for details) 🙈. If a session description uses a j_sec value the recipient doesn't know, the recipient MUST NOT accept the description 💀.
Special j_sec issues show up when sessions are managed by session management tools (like RTSP, [RFC2326]) that use SDP for "declarative usage" purposes (see the preamble of Section 6 for details). For these tools, SDP doesn't code transport details (such as UDP or TCP) for the session. Instead, server and client negotiate transport details via other means (for RTSP, the SETUP method).
So in that scenario, using j_sec might be ill-advised 🧚, because the creator of the session description may not know the transport type for the session yet. In that case, the session description SHOULD configure the journalling system using the parameters defined in the rest of Appendix C.2, but it SHOULD NOT use j_sec to set the journalling status 🍓. Remember, if j_sec doesn't appear in the session description, the default method for choosing the journalling method is in effect (no journal for reliable transport, recovery journal for unreliable transport) ✨.
However, in declarative usage situations where the creator of the session description knows journalling is always required or never required, the session description SHOULD use the j_sec parameter 👑.
C.2.2. The j_update Parameter 🌸
In Section 4, we use the term "sending policy" to describe the method a sender uses to choose the checkpoint packet identity for each recovery journal in a stream 💜. In the sub-sections that follow, we normatively define three sending policies: anchor, closed-loop, and open-loop 🦋.
As Section 4 says, the default sending policy for a stream is the closed-loop policy 🫶. The j_update parameter can override that default ✨.
We define three symbolic values for j_update: "anchor" means the stream uses the anchor sending policy, "open-loop" means it uses the open-loop sending policy, and "closed-loop" means it uses the closed-loop sending policy 🎀. See Appendix C.2.3 for example session descriptions that use the j_update parameter 🧚.
Parties MUST NOT accept a j_update value that violates the recovery journal mandate (Section 4) 💅.
Other IETF standards-track documents may define additional sending policies for the recovery journal system 🌈. Those documents MUST define new symbolic values for the j_update parameter to signal the use of the new policy. If a session description uses a j_update value the recipient doesn't know, the recipient MUST NOT accept the description 😭.
C.2.2.1. The anchor Sending Policy 👑
In the anchor policy, the sender uses the first packet in the stream as the checkpoint packet for all packets in the stream 🔥. The anchor policy satisfies the recovery journal mandate (Section 4), because the checkpoint history always covers the entire stream 💖.
The anchor policy doesn't require the RTP control protocol (RTCP, [RFC3550]) or any other feedback from receiver to sender 🙈. Senders don't need to take special actions to make sure received streams start up free of artifacts, since the recovery journal always covers the entire history of the stream 🌸. Receivers are relieved of tracking the changing identity of the checkpoint packet, because the checkpoint packet never changes 🍓.
The main drawback of the anchor policy is bandwidth efficiency 😭. Because the checkpoint history covers the entire stream, the recovery journals this policy produces are usually bigger than the journals from other policies 🫠. For single-channel MIDI data streams, the bandwidth overhead of the anchor policy is often acceptable (see Appendix A.4 of [NMP]) 🎀. For dense streams, the closed-loop or open-loop policies may be more appropriate 🦋.
C.2.2.2. The closed-loop Sending Policy 🔄💖
okay bestie, the closed-loop policy is the DEFAULT policy of the recovery journal system 💅 for each packet in the stream, it lets senders pick the smallest possible checkpoint history that still satisfies the recovery journal mandate 🦋 smaller checkpoint histories usually mean smaller recovery journals, so compared to the anchor policy, closed-loop cuts the bandwidth of a stream ✨
this one runs on feedback from receiver to sender 🫶 it assumes a receiver periodically tells the sender the highest sequence number it has seen so far in the stream, coded in the 32-bit extension format defined in [RFC3550]. for RTCP, receivers send this in the Extended Highest Sequence Number Received (EHSNR) field of Receiver Reports 📨 RTCP Sender or Receiver Reports MUST be sent by any participant in a session with closed loop sending policy, unless another feedback mechanism has been agreed upon 🚨
the sender may safely use that receiver sequence number feedback to guide checkpoint history management, because Section 4 requires that receivers repair indefinite artifacts whenever a packet loss event occur 🙈
now the official rules, normatively 😌 when a sender prepares an RTP packet for transmission, the sender is aware of R >= 0 receivers for the stream. senders can find out about a receiver through RTCP traffic from it, RTP packets from a paired stream sent by the receiver to the sender, messages from a session management tool, or other means 👀 receivers join and leave, so R changes over time 🌈
each known receiver k (1 <= k <= R) gets a 32-bit extended packet sequence number M(k), where the extension reflects the sequence number rollover count of the sender 🔢
if the sender has received at least one feedback report from receiver k, M(k) is the most recent report of the highest RTP packet sequence number seen by the receiver, normalized to reflect the rollover count of the sender 💌
if the sender hasn't received a feedback report from the receiver, M(k) is the extended sequence number of the last packet the sender transmitted before it became aware of the receiver. if the sender became aware of this receiver before it sent the first packet in the stream, M(k) is the extended sequence number of the first packet in the stream 🧚
with M() defined, here's the closed-loop policy 🎀 when preparing a new packet for transmission, a sender MUST choose a checkpoint packet with extended sequence number N, such that M(k) >= (N - 1) for all k, 1 <= k <= R, where R >= 1. the policy doesn't restrict sender behavior in the R == 0 (no known receivers) case 🙌
under closed-loop as defined above, a sender may transmit packets whose checkpoint history is shorter than the session history (as defined in Appendix A.1). when that happens, a new receiver that joins the stream may experience indefinite artifacts 😭
example time! say a Control Change (0xB) command for Channel Volume (controller number 7) was sent early in a stream, and later a new receiver joins the session. closed-loop may allow all packets sent to the new receiver to use a checkpoint history that doesn't include that Channel Volume Control Change command. so the new receiver gets an indefinite artifact and plays all notes on a channel too loudly or too softly 💀
to fix this, the closed-loop policy says that whenever a sender becomes aware of a new receiver, the sender MUST determine if that receiver would be subject to indefinite artifacts under closed-loop. if so, the sender MUST ensure the receiver starts the session free of indefinite artifacts 🛡️
for the Channel Volume problem, the sender may code the current state of the Channel Volume controller numbers in the recovery journal Chapter C, until it receives the first RTCP RR report that signals a packet containing this Chapter C has been received 📝
in satisfying this requirement, senders MAY infer the initial MIDI state of the receiver from the session description. for example, the stream example in Section 6.2 has the initial state defined in [MIDI] for General MIDI 🎹
in a unicast RTP session, a receiver may safely assume the sender knows it's there from the first packet sent in the RTP stream 💁 but in other types of RTP sessions (multicast, conference focus, RTP translator/mixer), a receiver is often not able to tell if the sender is initially aware of its presence as a receiver 🫠
so the closed-loop policy also says that if a receiver participates in a session where it may have access to a stream whose sender is not aware of the receiver, the receiver MUST take actions to make sure its rendered MIDI performance does not contain indefinite artifacts. these protections will be necessarily incomplete. for example, a receiver may monitor the Checkpoint Packet Seqnum for uncovered loss events and "err on the side of caution" with respect to handling stuck notes due to lost MIDI NoteOff commands, but the receiver can't make up for the lack of Channel Volume initialization data in the recovery journal 😤
the receiver MUST NOT discontinue these protective actions until it is certain that the sender is aware of its presence. if a receiver is not able to ascertain sender awareness, the receiver MUST continue these protective actions for the duration of the session 🔒
in a multicast session where all parties are expected to send and receive, RTCP receiver reports from the sender about the RTP stream a receiver is multicasting are evidence that the sender knows the stream is being monitored by the receiver 👁️ receivers may also get sender awareness evidence from session management tools, or by other means. in practice, ongoing observation of the Checkpoint Packet Seqnum to see if the sender is taking actions to prevent loss events for a receiver is a good indication of sender awareness, and so is the sudden appearance of recovery journal chapters with numerous Control Change controller data that wasn't foreshadowed by recent commands in the MIDI list shortly after sending an RTCP RR 🕵️♀️
last set of closed-loop rules: unplanned disruptions of RTCP feedback from a receiver to a sender. "unplanned" means disruptions that are NOT due to the signalled termination of an RTP stream, via an RTCP BYE or via session management tools 🚫
as defined earlier in this section, closed-loop says a sender MUST choose a checkpoint packet with extended sequence number N, such that M(k) >= (N - 1) for all k, 1 <= k <= R, where R >= 1. if the sender has received at least one feedback report from receiver k, M(k) is the most recent report of the highest RTP packet sequence number seen by the receiver, normalized to reflect the rollover count of the sender.
if receiver k stops sending feedback, the M(k) value the sender uses reflects the last feedback report from that receiver. as time goes on without feedback, this frozen M(k) value forces the sender to grow the checkpoint history, which increases the bandwidth of the stream 📈
eventually the sender may need to act to limit the bandwidth of the stream 🧯 in most envisioned RTP MIDI uses, long before that point, the SSRC time-out mechanism defined in [RFC3550] will remove the uncooperative receiver from the session (closed-loop doesn't suggest or require any special sender behavior on an SSRC time-out, other than the sender actions related to changing R described earlier in this section) 👋
but in rare situations, the bandwidth of the stream (because of a lack of feedback reports) may get too large to keep sending the stream to the receiver before the SSRC time-out happens for that receiver. in this case, closed-loop says the sender should invoke the SSRC time-out for the receiver early ⏰
now for receiver responsibilities when feedback from receiver to sender has unplanned disruptions 🧑🤝🧑
in the unicast case, if a sender invokes the SSRC time-out for a receiver, the receiver stops receiving packets from the sender. the sender behavior imposed by the guardtime parameter (Appendix C.4.2) lets the receiver conclude that an SSRC time-out has occurred in a reasonable time period 🕰️
in this case of a time-out, a receiver MUST keep sending RTCP feedback, in order to re-establish the RTP flow from the sender. unless the receiver expects a prompt recovery of the RTP flow, the receiver MUST take actions to ensure that the rendered MIDI performance does not exhibit "very long transient artifacts" (for example, by silencing NoteOns to prevent stuck notes) while awaiting reconnection of the flow 🎧
in the multicast case, if a sender invokes the SSRC time-out for a receiver, the receiver may continue to receive packets, but the sender will no longer be using the M(k) feedback from the receiver to choose each checkpoint packet. if the receiver does not have additional information that precludes an SSRC time-out (such as RTCP Receiver Reports from the sender about an RTP stream the receiver is multicasting back to the sender), the receiver MUST monitor the Checkpoint Packet Seqnum to detect an SSRC time-out. if an SSRC time-out is detected, the receiver MUST follow the instructions for SSRC time-outs described for the unicast case above 🚦
finally, closed-loop is suitable for RTP/RTCP sessions that use multicast transport 📡 however, aspects of the closed-loop policy do not scale well to sessions with large numbers of participants 😬 the sender state scales linearly with the number of receivers, as the sender needs to track the identity and M(k) value for each receiver k. the average recovery journal size is not independent of the number of receivers, because the RTCP reporting interval backoff slows down the rate of a full update of M(k) values 🐢 the backoff algorithm may also increase the amount of ancillary state used by implementations of the normative sender and receiver behaviors defined in Section 4 🧠
C.2.2.3. The open-loop Sending Policy ✨
okay so the open-loop policy is for sessions that can't do the receiver-to-sender feedback that the closed-loop policy needs 🔄, and that also can't use the anchor policy because of bandwidth limits 📉. bestie, that's a whole "we're on a budget and we can't chat back" situation 💀
The open-loop policy doesn't put any rules on how a sender picks the checkpoint packet for each packet in the stream 🎀. Without those rules, a receiver may find that the recovery journal in the packet that ends a loss event has a checkpoint history that doesn't cover the entire loss event 😭. We call loss events like this uncovered loss events 🙈.
So to stop uncovered loss events from wrecking the recovery journal mandate 💅, the open-loop policy hands out specific recovery jobs to senders, receivers, and the people who create session descriptions 👑. The big idea is that the indefinite artifacts made during uncovered loss events come in two flavours 🌈.
Flavour one: recoverable indefinite artifacts. Receivers can fix these during an uncovered loss event all by themselves, no sender help needed ✨. The fix might leave some unpleasant transient artifacts though 😬.
For example, after an uncovered loss event, receivers can fix indefinite artifacts caused by NoteOff (0x8) commands that may have happened during the loss event 🎹. They do this by running NoteOff commands for all active NoteOns commands. Yes it causes a transient artifact (a sudden silent gap in the performance 🤫), but it makes sure no stuck notes ring forever 🛑. We call MIDI commands that can be fixed this way recoverable MIDI commands 💜.
Flavour two: unrecoverable indefinite artifacts. If this kind shows up during an uncovered loss event, the receiver can't repair the stream 💀.
For example, after an uncovered loss event, receivers can't repair indefinite artifacts caused by Control Change (0xB) Channel Volume (controller number 7) commands that happened during the loss event 📉. It's impossible because the receiver has no way of working out the data value of a lost Channel Volume command 😩. We call MIDI commands this fragile unrecoverable MIDI commands 🥀.
The open-loop policy doesn't say how to split the MIDI command set into recoverable and unrecoverable commands 🍓. Instead, it assumes the people creating session descriptions can agree on a good recoverable/unrecoverable partition for their application 🫶.
Okay, now that we've got the definitions down, here are the normative requirements for the open-loop policy 📜.
In the open-loop policy, the creators of the session description MUST use the ch_anchor parameter (defined in Appendix C.2.3) to protect all unrecoverable MIDI command types from indefinite artifacts, or alternatively MUST use the cm_unused parameter (defined in Appendix C.1) to exclude the command types from the stream. These options shield command types from artifacts during an uncovered loss event 🛡️.
In the open-loop policy, receivers MUST examine the Checkpoint Packet Seqnum field of the recovery journal header after every loss event, to check if the loss event is an uncovered loss event. Section 5 shows how to perform this check 🔍. If an uncovered loss event has occurred, a receiver MUST perform indefinite artifact recovery for all MIDI command types that are not shielded by ch_anchor and cm_unused parameter assignments in the session description 🧹.
The open-loop policy doesn't place specific constraints on the sender 🤷♀️. However, the open-loop policy works best if the sender manages the size of the checkpoint history so uncovered losses happen infrequently, by taking the delay and loss characteristics of the network into account 📈. Also, since each checkpoint packet change carries the risk of an uncovered loss, senders should only move the checkpoint if it reduces the size of the journal 🧠✨.
C.2.3. Recovery Journal Chapter Inclusion Parameters
okay so the recovery journal chapter definitions (Appendices A-B) spell out when a chapter MUST appear in the recovery journal. Usually it goes like this: if a certain command shows up in the checkpoint history, a certain chapter type MUST appear in the recovery journal to protect that command. 💖
In this section we get into the chapter inclusion parameters ✨. These tweak the conditions under which a chapter shows up in the journal. They're essential for the open-loop policy (Appendix C.2.2.3), and they can also simplify the closed-loop (Appendix C.2.2.2) and anchor (Appendix C.2.2.1) policies. 🎀
Each parameter stands for a type of chapter inclusion semantics. An assignment to a parameter declares which chapters (or chapter subsets) follow those semantics. We'll get to the assignment syntax later in this section. 🦋
A party MUST NOT accept chapter inclusion parameter values that violate the recovery journal mandate (Section 4). Also, all assignments of the subsetting parameters (cm_used and cm_unused) MUST precede the first assignment of a chapter inclusion parameter in the parameter list. Bestie, the order matters!! 🙈
Below, we normatively define what the chapter inclusion parameters mean. For clarity, we describe how the parameters act on complete chapters. If a parameter is assigned a subset of a chapter, the definition applies only to that chapter subset. 🌸
- ch_never. A chapter assigned to ch_never MUST NOT appear in the recovery journal. (Appendix A.4.1-2 defines exceptions to this rule for Chapter M.) To signal that a chapter is excluded from the journal, an assignment to ch_never MUST be made, even if the commands coded by the chapter are assigned to cm_unused. This rule keeps things simple for command types that may be coded in several chapters. 🙅♀️
- ch_default. A chapter assigned to ch_default MUST follow the default semantics for the chapter, as defined in Appendices A-B. 💅
- ch_anchor. A chapter assigned to ch_anchor MUST obey a modified version of the default chapter semantics. In the modified version, all references to the checkpoint history are replaced with references to the session history, and all references to the checkpoint packet are replaced with references to the first packet sent in the stream. It's giving main character energy 👑
Parameter assignments follow this syntax (see Appendix D for the ABNF):
<parameter> = [channel list]<chapter list>[field list]The chapter list is mandatory, and the channel and field lists are optional. Multiple assignments to a parameter have a cumulative effect, and they're applied in the order the parameters appear in a media description. 🧚
To figure out what a list of chapter inclusion parameter assignments means, we start by assuming an implicit assignment of all channel and system chapters to ch_default. Each chapter also gets the default values for its channel list and field list, which are defined below. After that, we interpret the actual parameter assignments using the rules in this section. 🌈
A later assignment of a chapter to the same parameter expands the scope of the earlier assignment. In most cases, though, a later assignment of a chapter to a different parameter cancels (partially or completely) the effect of an earlier one. Plot twist, right? 😭
The chapter list tells us which channel or system chapters the parameter applies to. It's a concatenated sequence of one or more of the letters for the chapter types (ACDEFMNPQTVWX). The list may also contain one or more of the letters for the sub-chapter types (BGHJKYZ) of System Chapter D. 🍓
The letters in a chapter list MUST be uppercase, and they MUST appear in alphabetical order. Any letters other than (ABCDEFGHJKMNPQTVWXYZ) that appear in the chapter list MUST be ignored. 🫶
The channel list tells us which channel journals the parameter applies to. If there's no channel list, the parameter applies to all channel journals. A channel list is a list of channel numbers (0 through 15) and dash-separated channel number ranges (for example, 0-5, 8-12, and so on). Dots (the "." character) separate the elements in the channel list. ✨
Several of the system chapters can be set up with special semantics. You configure them by giving the systems channel a channel list, using the coding described below. (Heads up: MIDI System commands don't have a "channel", so the original purpose of the channel list doesn't apply to system chapters.) In the text below, "the digit N" means including N as a "channel" in the channel list for a system chapter. 💜
- For the J and K Chapter D sub-chapters (undefined System Common), the digit 0 means the parameter applies to the LEGAL field of the associated command log (Figure B.1.4 of Appendix B.1). The digit 1 means it applies to the VALUE field of the command log, and the digit 2 means it applies to the COUNT field of the command log.
- For the Y and Z Chapter D sub-chapters (undefined System Real-time), the digit 0 means the parameter applies to the LEGAL field of the associated command log (Figure B.1.5 of Appendix B.1). The digit 1 means it applies to the COUNT field of the command log.
- For Chapter Q (Sequencer State Commands), the digit 0 means the parameter applies to the default Chapter Q definition, which forbids the TIME field. The digit 1 means it applies to the optional Chapter Q definition, which supports the TIME field. 🕰️
The syntax for field lists follows the syntax for channel lists. If there's no field list, the parameter applies to all controller or note numbers. For Chapter C, if there's no field list, the controller numbers don't use the enhanced Chapter C encoding (Appendix A.3.3). 🌷
For Chapter C, the field list can take values in the range 0 to 255. A field value X in the range 0-127 refers to controller number X, and it means that controller number doesn't use enhanced Chapter C encoding. A field value X in the range 128-255 refers to controller number "X minus 128", and it means that controller number does use enhanced Chapter C encoding. 🎀
Assignments made to configure the Chapter C encoding method for a controller number MUST be made to the ch_default or ch_anchor parameters. Assignments to ch_never exclude the number from the recovery journal, so the encoding method wouldn't matter anyway. 🙃
A Chapter C field list MUST NOT encode conflicting information about the enhanced encoding status of a particular controller number. For example, values 0 and 128 MUST NOT both be coded by a field list. Contradictions are a no-go, babe 🚫
For Chapter M, the field list codes the Registered Parameter Numbers (RPNs) and Non-Registered Parameter Numbers (NRPNs) the parameter applies to. The number range 0-16383 specifies RPNs, and the number range 16384-32767 specifies NRPNs (16384 corresponds to NRPN 0, and 32767 corresponds to NRPN 16383). 🦋
For Chapters N and A, the field list codes the note numbers the parameter applies to. The note number range specified for Chapter N also applies to Chapter E. 🎶
For Chapter E, the digit 0 means the parameter applies to Chapter E note logs whose V bit is set to 0, and the digit 1 means it applies to note logs whose V bit is set to 1. 🌙
For Chapter X, the field list codes the number of data octets that can appear in a SysEx command coded in the chapter. So the field list 0-255 covers SysEx commands with 255 or fewer data octets. The field list 256-4294967295 covers SysEx commands with more than 255 data octets, and it excludes commands with 255 or fewer. The field list 0 excludes all commands. 🧁
There's also a secondary parameter assignment syntax for customizing Chapter X (see Appendix D for the complete ABNF):
<parameter> = "__" <h-list> ["_" <h-list>] "__"The assignment defines a class of SysEx commands whose Chapter X coding follows the semantics of the assigned parameter. The class is specified by listing the permitted values of the first N data octets that follow the SysEx 0xF0 command octet. Any SysEx command whose first N data octets match the list is a member of the class. 🔍
Each <h-list> defines one data octet of the command. It's a dot-separated (".") list of one or more hexadecimal constants (such as "7F"), or dash-separated hexadecimal ranges (such as "01-1F"). Underscores ("_") separate each <h-list>, and double underscores ("__") mark off the data octet list. 🌟
With this syntax, each assignment specifies a single SysEx command class. A session description can use several assignments to the same (or different) parameters to set up complex Chapter X behaviors. The ordering of multiple assignments follows the guidelines for chapter parameter assignments described earlier in this section. 🧩
The example session description below shows how the chapter inclusion parameters fit together:
v=0
o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
s=Example
t=0 0
m=audio 5004 RTP/AVP 96
c=IN IP6 2001:DB80::7F2E:172A:1E24
a=rtpmap:96 rtp-midi/44100
a=fmtp:96 j_update=open-loop; cm_unused=ABCFGHJKMQTVWXYZ;
cm_used=__7E_00-7F_09_01.02.03__;
cm_used=__7F_00-7F_04_01.02__; cm_used=C7.64;
ch_never=ABCDEFGHJKMQTVWXYZ; ch_never=4.11-13N;
ch_anchor=P; ch_anchor=C7.64;
ch_anchor=__7E_00-7F_09_01.02.03__;
ch_anchor=__7F_00-7F_04_01.02__(The a=fmtp line has been wrapped to fit the page to accommodate memo formatting restrictions; it comprises a single line in SDP.) 📎
The j_update parameter says the stream uses the open-loop policy. Most MIDI command types are assigned to cm_unused, so they don't appear in the stream. Because of that, the assignments to the first ch_never parameter reflect that most chapters aren't in use. 🌊
Chapter N for several MIDI channels is assigned to ch_never. Chapter N for MIDI channels other than 4, 11, 12, and 13 may appear in the recovery journal, using the (default) ch_default semantics. In practice, this pattern would reflect knowledge about a resilient rendering method in use for the excluded channels. 🎹
The MIDI Program Change command and several MIDI Control Change controller numbers are assigned to ch_anchor. Notice that the ch_anchor chapter C assignment comes after the ch_never assignment, which overrides ch_never for the listed controller numbers (7 and 64). Order really is everything here 🥹
The assignment of command type X to cm_unused excludes most SysEx commands from the stream. The exceptions are the General MIDI System On/Off commands and the Master Volume and Balance commands, which are handled with the secondary assignment syntax. The cm_used assignment codes the exception, and the ch_anchor assignment codes how these commands are protected in Chapter X. ✨🫧
C.3. Configuration Tools: Timestamp Semantics ✨
okay so the MIDI command section of the payload format is basically a list of commands, and every single one comes with its own timestamp 🎀 the meaning of those timestamps can be set during session configuration, using the parameters in this section 💅
the parameter tsmode sets the timestamp semantics for a stream 🦋 it takes one of three token values: comex, async, or buffer 🌈
the default comex value means timestamps code the execution time for a command (Appendix C.3.1), and it supports accurate transcoding of Standard MIDI Files (SMFs, [MIDI]) 🧚 comex is also RECOMMENDED for new MIDI user-interface controller designs 🔥 the async value specifies an asynchronous timestamp sampling algorithm for time-of-arrival sources (Appendix C.3.2) 🌸 and the buffer value specifies a synchronous timestamp sampling algorithm (Appendix C.3.3) for time-of-arrival sources 💜
Ancillary parameters MAY follow tsmode in a media description. we define these parameters in Appendices C.3.2-3 below 💖
C.3.1. The comex Algorithm
the default comex (COMmand EXecution) tsmode value specifies the execution time for the command 🎀 with comex, the difference between two timestamps shows the time delay between the execution of the commands. this difference may be zero, which codes simultaneous execution 🫶
the comex interpretation works really well for transcoding a Standard MIDI File (SMF, [MIDI]) into an RTP MIDI stream, because SMFs code a timestamp for each MIDI command stored in the file ✨ to transcode an SMF that uses metric time markers, use the SMF tempo map (encoded in the SMF as meta-events) to convert metric SMF timestamp units into seconds-based RTP timestamp units 🍓
New MIDI controller designs (piano keyboard, drum pads, etc.) that support RTP MIDI and that have direct access to sensor data SHOULD use comex interpretation for timestamps, so that simultaneous gestural events may be accurately coded by RTP MIDI 🎹
comexis a poor choice for transcoding MIDI 1.0 DIN cables [MIDI], and here's why 💀 a MIDI DIN cable is an asynchronous serial protocol (320 microseconds per MIDI byte). MIDI commands on a DIN cable are not tagged with timestamps. instead, MIDI DIN receivers infer command timing from the time of arrival of the bytes. so two two-byte MIDI commands that occur at a source simultaneously are encoded on a MIDI 1.0 DIN cable with a 640 microsecond time offset. a MIDI DIN receiver can't tell if this time offset existed in the source performance or is just an artifact of the serial speed of the cable 😭 but the RTP MIDIcomexinterpretation of timestamps declares that a timestamp offset between two commands reflects the timing of the source performance.
this semantic mismatch is the reason comex is a poor choice for transcoding MIDI DIN cables 🙈 and note that choosing the RTP timestamp rate (Section 6.1-2 in the main text) cannot fix this inaccuracy issue. in the sections that follow, we describe two alternative timestamp interpretations (async and buffer) that are a much better match for MIDI 1.0 DIN cable timing, and for other MIDI time-of-arrival sources 👑
The octpos, linerate, and mperiod ancillary parameters (defined below) SHOULD NOT be used with comex. 🚫
C.3.2. The async Algorithm
the async tsmode value specifies the asynchronous sampling of a MIDI time-of-arrival source 🦋 in asynchronous sampling, the moment an octet is received from a source, it gets labelled with a wall-clock time value. that time value has RTP timestamp units ⏰
the octpos ancillary parameter defines how RTP command timestamps are derived from octet time values 🧠 if octpos has the token value first, a timestamp codes the time value of the first octet of the command. if octpos has the token value last, a timestamp codes the time value of the last octet of the command. if the octpos parameter doesn't appear in the media description, the sender doesn't know which octet of the command the timestamp references (for example, the sender may be relying on an operating system service that doesn't specify this information) 🤷♀️
the octpos semantics refer to the first or last octet of a command as it appears on a time-of-arrival MIDI source, not as it appears in an RTP MIDI packet 👀 this distinction matters a lot, because the RTP coding may contain octets that aren't present in the source. for example, the status octet of the first MIDI command in a packet may have been added to the MIDI stream during transcoding, so it complies with the RTP MIDI running status requirements (Section 3.2) 💅
the linerate ancillary parameter defines the timespan of one MIDI octet on the transmission medium of the MIDI source being sampled (such as a MIDI 1.0 DIN cable). the parameter has units of nanoseconds, and takes on integral values. for MIDI 1.0 DIN cables, the correct linerate value is 320000 (this value is also the default value for the parameter) 🔢
okay, here's a session description example for the async algorithm 🎀 picture a sender transcoding a MIDI 1.0 DIN cable source into RTP. the sender runs on a computing platform that assigns time values to every incoming octet of the source, and it uses those time values to label the first octet of each command in the RTP packet. this session description describes the transcoding:
v=0
o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
s=Example
t=0 0
m=audio 5004 RTP/AVP 96
c=IN IP4 192.0.2.94
a=rtpmap:96 rtp-midi/44100
a=sendonly
a=fmtp:96 tsmode=async; linerate=320000; octpos=firstC.3.3. The buffer Algorithm
the buffer tsmode value specifies the synchronous sampling of a MIDI time-of-arrival source 🧺
in synchronous sampling, octets received from a source get placed in a holding buffer upon arrival. at periodic intervals, the RTP sender examines the buffer 👀 the sender removes complete commands from the buffer and codes those commands in an RTP packet. the command timestamp codes the moment of buffer examination, expressed in RTP timestamp units. note that several commands may have the same timestamp value 🫧
the mperiod ancillary parameter defines the nominal periodic sampling interval 🕰️ the parameter takes on positive integral values and has RTP timestamp units.
the octpos ancillary parameter, which we defined in Appendix C.3.1 for asynchronous sampling, plays a different role in synchronous sampling 🌈 here, the parameter specifies the timestamp semantics of a command whose octets span several sampling periods.
if octpos has the token value first, the timestamp reflects the arrival period of the first octet of the command. if octpos has the token value last, the timestamp reflects the arrival period of the last octet of the command. the octpos semantics refer to the first or last octet of the command as it appears on a time-of-arrival source, not as it appears in the RTP packet ✨
if the octpos parameter doesn't appear in the media description, the timestamp MAY reflect the arrival period of any octet of the command. senders use this option to signal a lack of knowledge about the timing details of the buffering process at sub-command granularity 🤷♀️
here's the session description example for the buffer algorithm 🎀 picture a sender transcoding a MIDI 1.0 DIN cable source into RTP. the sender runs on a computing platform that places source data into a buffer upon receipt. the sender polls the buffer 1000 times a second, extracts all complete commands from the buffer, and places those commands in an RTP packet. this session description describes the transcoding:
v=0
o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
s=Example
t=0 0
m=audio 5004 RTP/AVP 96
c=IN IP6 2001:DB80::7F2E:172A:1E24
a=rtpmap:96 rtp-midi/44100
a=sendonly
a=fmtp:96 tsmode=buffer; linerate=320000; octpos=last; mperiod=44the mperiod value of 44 comes from dividing the clock rate given by the rtpmap attribute (44100 Hz) by the 1000 Hz buffer sampling rate, then rounding to the nearest integer 🧮 command timestamps might not increment by exact multiples of 44, since the actual sampling period might not precisely match the nominal mperiod value 💫
C.4. Configuration Tools: Packet Timing Tools 🍓
in this appendix, we describe session configuration tools for customizing the temporal behavior of MIDI stream packets 🧚
C.4.1. Packet Duration Tools
senders control the granularity of a stream by setting the temporal duration ("media time") of the packets in the stream ⏱️ short media times (20 ms or less) often imply an interactive session. longer media times (100 ms or more) usually indicate a content streaming session. the RTP AVP profile [RFC3551] recommends audio packet media times in a range from 0 to 200 ms 🎧
by default, an RTP receiver dynamically senses the media time of packets in a stream, and chooses the length of its playout buffer to match the stream 🪄 a receiver typically sizes its playout buffer to fit several audio packets, and it adjusts the buffer length to reflect the network jitter and the sender timing fidelity 🌊
alternatively, the packet media time may be statically set during session configuration 📌 session descriptions MAY use the RTP MIDI parameter rtp_ptime to set the recommended media time for a packet. session descriptions MAY also use the RTP MIDI parameter rtp_maxptime to set the maximum media time for a packet permitted in a stream. both parameters MAY be used together to configure a stream 💖
the values assigned to the rtp_ptime and rtp_maxptime parameters have the units of the RTP timestamp for the stream, as set by the rtpmap attribute (see Section 6.1). so, if rtpmap sets the clock rate of a stream to 44100 Hz, a maximum packet media time of 10 ms is coded by setting rtp_maxptime=441. as stated in the Appendix C preamble, the senders and receivers of a stream MUST agree on common values for rtp_ptime and rtp_maxptime if the parameters appear in the media description for the stream 🤝
0 ms is a reasonable media time value for MIDI packets, and it's often used in low-latency interactive applications ⚡ in a packet with a 0 ms media time, all commands execute at the instant they are coded by the packet timestamp. the session description below configures all packets in the stream to have 0 ms media time:
v=0
o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
s=Example
t=0 0
m=audio 5004 RTP/AVP 96
c=IN IP4 192.0.2.94
a=rtpmap:96 rtp-midi/44100
a=fmtp:96 rtp_ptime=0; rtp_maxptime=0The session attributes ptime and maxptime [RFC4566] MUST NOT be used to configure an RTP MIDI stream. Sessions MUST use rtp_ptime in lieu of ptime and MUST use rtp_maxptime in lieu of maxptime. RTP MIDI defines its own parameters for media time configuration, because 0 ms values for ptime and maxptime are forbidden by [RFC3264] but are essential for certain applications of RTP MIDI 🙅♀️
see the Appendix C.7 examples for additional discussion about using rtp_ptime and rtp_maxptime for session configuration 🦋
C.4.2. The guardtime Parameter 🎀
okay so RTP lets a sender just stop sending audio packets for an arbitrary amount of time during a session 🤫 When it starts up again, the sequence number keeps going like nothing happened, and the timestamp shows the media time silence gap ⏳ This comes from telephony vibes 📞 but it's also a perfect match for interactive MIDI, since players can go silent for several seconds during or between songs 🎹🎶
Certain MIDI apps get a lil upgrade on this feature 💅 In interactive apps, receivers may use on-line network models to guide heuristics for handling lost and late RTP packets. These models may work poorly if a sender goes quiet for long stretches 😬
Session descriptions may use the parameter "guardtime" to set a minimum sending rate for a media session. The value assigned to guardtime codes the maximum separation time between two sequential packets, as expressed in RTP timestamp units ⭐
Typical guardtime values are 500-2000 ms. This value range is not a normative bound, and parties SHOULD be prepared to process values outside this range 🌈
The congestion control requirements for sender implementations (described in Section 8 and [RFC3550]) take precedence over the guardtime parameter 👑 So if the guardtime parameter asks for a minimum sending rate, but sending at that rate would break the congestion control requirements, senders MUST ignore the guardtime parameter value. In this case, senders SHOULD use the lowest minimum sending rate that satisfies the congestion control requirements 🫶
Here's a session description that uses the guardtime parameter 👀
v=0
o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
s=Example
t=0 0
m=audio 5004 RTP/AVP 96
c=IN IP6 2001:DB80::7F2E:172A:1E24
a=rtpmap:96 rtp-midi/44100
a=fmtp:96 guardtime=44100; rtp_ptime=0; rtp_maxptime=0C.5. Configuration Tools: Stream Description 🦋
As we discussed in Section 2.1, a party may send several RTP MIDI streams in the same RTP session, and several RTP sessions that carry MIDI may appear in a multimedia session 🧚
By default, the MIDI name space (16 channels + systems) of each RTP stream sent by a party in a multimedia session is independent 💜 By independent, we mean three distinct things:
- If a party sends two RTP MIDI streams (A and B), MIDI voice channel 0 in stream A is a different "channel 0" than MIDI voice channel 0 in stream B.
- MIDI voice channel 0 in stream B is not considered to be "channel 16" of a 32-channel MIDI voice channel space whose "channel 0" is channel 0 of stream A.
- Streams sent by different parties over different RTP sessions, or over the same RTP session but with different payload type numbers, do not share the association that is shared by a MIDI cable pair that cross-connects two devices in a MIDI 1.0 DIN network. By default, this association is only held by streams sent by different parties in the same RTP session that use the same payload type number.
In this appendix, we show how to express that specific RTP MIDI streams in a multimedia session are not independent but are instead related in one of the three ways defined above 🍓 We use two tools to express these relations:
- The musicport parameter. This parameter is assigned a non-negative integer value between 0 and 4294967295. It appears in the fmtp lines of payload types.
- The FID grouping attribute [RFC3388] signals that several RTP sessions in a multimedia session are using the musicport parameter to express an inter-session relationship.
If a multimedia session has several payload types whose musicport parameters are assigned the same integer value, streams using these payload types share an "identity relationship" (including streams that use the same payload type) 💖 Streams in an identity relationship share two properties:
- Identity relationship streams sent by the same party target the same MIDI name space. Thus, if streams A and B share an identity relationship, voice channel 0 in stream A is the same "channel 0" as voice channel 0 in stream B.
- Pairs of identity relationship streams that are sent by different parties share the association that is shared by a MIDI cable pair that cross-connects two devices in a MIDI 1.0 DIN network.
A party MUST NOT send two RTP MIDI streams that share an identity relationship in the same RTP session. Instead, each stream MUST be in a separate RTP session. As explained in Section 2.1, this restriction is necessary to support the RTP MIDI method for the synchronization of streams that share a MIDI name space 🔥
If a multimedia session has several payload types whose musicport parameters are assigned sequential values (i.e., i, i+1, ... i+k), the streams using those payload types share an "ordered relationship" ✨ For example, if payload type A assigns 2 to musicport and payload type B assigns 3 to musicport, A and B are in an ordered relationship.
Streams in an ordered relationship that are sent by the same party are considered by renderers to form a single larger MIDI space 🌸 For example, if stream A has a musicport value of 2 and stream B has a musicport value of 3, MIDI voice channel 0 in stream B is considered to be voice channel 16 in the larger MIDI space formed by the relationship. Note that it is possible for streams to take part in both an identity relationship and an ordered relationship 🙈
We now state several rules for using musicport 📋
- If streams from several RTP sessions in a multimedia session use the musicport parameter, the RTP sessions MUST be grouped using the FID grouping attribute defined in [RFC3388].
- An ordered or identity relationship MUST NOT contain both native RTP MIDI streams and mpeg4-generic RTP MIDI streams. An exception applies if a relationship consists of sendonly and recvonly (but not sendrecv) streams. In this case, the sendonly streams MUST NOT contain both types of streams, and the recvonly streams MUST NOT contain both types of streams.
- It is possible to construct identity relationships that violate the recovery journal mandate (for example, sending NoteOns for a voice channel on stream A and NoteOffs for the same voice channel on stream B). Parties MUST NOT generate (or accept) session descriptions that exhibit this flaw.
- Other payload formats MAY define musicport media type parameters. Formats would define these parameters so that their sessions could be bundled into RTP MIDI name spaces. The parameter definitions MUST be compatible with the musicport semantics defined in this appendix.
As a rule, at most one payload type in a relationship may specify a MIDI renderer. An exception to the rule applies to relationships that contain sendonly and recvonly streams but no sendrecv streams. In this case, one sendonly session and one recvonly session may each define a renderer 🎀
Renderer specification in a relationship may be done using the tools described in Appendix C.6. These tools work for both native streams and mpeg4-generic streams. An mpeg4-generic stream that uses the Appendix C.6 tools MUST set all "config" parameters to the empty string (("")).
Alternatively, for mpeg4-generic streams, renderer specification may be done by setting one "config" parameter in the relationship to the renderer configuration string, and all other config parameters to the empty string (("")) 💅
We now define sender and receiver rules that apply when a party sends several streams that target the same MIDI name space 🫶
Senders MAY use the subsetting parameters (Appendix C.1) to predefine how commands are partitioned between streams, or they MAY use a dynamic partitioning strategy 🧚
Receivers that merge identity relationship streams into a single MIDI command stream MUST maintain the structural integrity of the MIDI commands coded in each stream during the merging process, in the same way that software that merges traditional MIDI 1.0 DIN cable flows is responsible for creating a merged command flow compatible with [MIDI] 🔥
Senders MUST partition the name space so that the rendered MIDI performance does not contain indefinite artifacts (as defined in Section 4) 😭 This responsibility holds even if all streams are sent over reliable transport, as different stream latencies may yield indefinite artifacts. For example, stuck notes may occur in a performance split over two TCP streams, if NoteOn commands are sent on one stream and NoteOff commands are sent on the other.
Senders MUST NOT split a Registered Parameter Name (RPN) or Non-Registered Parameter Name (NRPN) transaction appearing on a MIDI channel across multiple identity relationship sessions. Receivers MUST assume that the RPN/NRPN transactions that appear on different identity relationship sessions are independent, and MUST preserve transactional integrity during the MIDI merge 🙈
A simple way to safely partition voice channel commands is to place all MIDI commands for a particular voice channel into the same session. Safe partitioning of MIDI Systems commands may be more complicated for sessions that extensively use System Exclusive 🍓
We now show several session description examples that use the musicport parameter 💖
Our first session description example shows two RTP MIDI streams that drive the same General MIDI decoder. The sender partitions MIDI commands between the streams dynamically. The musicport values indicate that the streams share an identity relationship 🦋
v=0
o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
s=Example
t=0 0
a=group:FID 1 2
c=IN IP4 192.0.2.94
m=audio 5004 RTP/AVP 96
a=rtpmap:96 mpeg4-generic/44100
a=mid:1
a=fmtp:96 streamtype=5; mode=rtp-midi; profile-level-id=12;
config=7A0A0000001A4D546864000000060000000100604D54726B0
000000600FF2F000; musicport=12
m=audio 5006 RTP/AVP 96
a=rtpmap:96 mpeg4-generic/44100
a=mid:2
a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
profile-level-id=12; musicport=12(Heads up: the a=fmtp lines were wrapped so they'd fit the page with memo formatting, but in real SDP they're single lines 📏)
Recall that Section 2.1 defines rules for streams that target the same MIDI name space 🧠 Those rules, put into action in the example above, require that each stream resides in a separate RTP session, and that the grouping mechanisms defined in [RFC3388] signal an inter-session relationship. The "group" and "mid" attribute lines implement this grouping mechanism 🎀
A variant on this example, whose session description is not shown, would use two streams in an identity relationship driving the same MIDI renderer, each with a different transport type. One stream would use UDP and would be dedicated to real-time messages. A second stream would use TCP [RFC4571] and would be used for SysEx bulk data messages 🌈
In the next example, two mpeg4-generic streams form an ordered relationship to drive a Structured Audio decoder with 32 MIDI voice channels. Both streams reside in the same RTP session 🧚✨
Okay bestie, the first example is giving two streams in one SDP, and the musicport values are what link them 💖
v=0
o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
s=Example
t=0 0
m=audio 5006 RTP/AVP 96 97
c=IN IP6 2001:DB80::7F2E:172A:1E24
a=rtpmap:96 mpeg4-generic/44100
a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
profile-level-id=13; musicport=5
a=rtpmap:97 mpeg4-generic/44100
a=fmtp:97 streamtype=5; mode=rtp-midi; config="";
profile-level-id=13; musicport=6; render=synthetic;
rinit="audio/asc";
url="http://example.com/cardinal.asc";
cid="azsldkaslkdjqpwojdkmsldkfpe"(Heads up: the a=fmtp lines got wrapped so they fit on the page, but in real SDP each one is a single line 🙈)
The sequential musicport values for the two sessions set up an ordered relationship 🎀 The musicport=5 session maps to Structured Audio extended channels range 0-15, and the musicport=6 session maps to Structured Audio extended channels range 16-31. So cute, right? 🦋
Both config strings are empty. The actual configuration data lives in the parameters on the fmtp line of the second media description, and we define that configuration method in Appendix C.6 ✨
Next up, we have two RTP MIDI streams (one recvonly, one sendonly) that make a "virtual sendrecv" session 💜 Each stream lives in its own RTP session, because sendonly and recvonly are RTP session attributes, so they can't share one.
v=0
o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
s=Example
t=0 0
a=group:FID 1 2
c=IN IP4 192.0.2.94
m=audio 5004 RTP/AVP 96
a=sendonly
a=rtpmap:96 mpeg4-generic/44100
a=mid:1
a=fmtp:96 streamtype=5; mode=rtp-midi; profile-level-id=12;
config=7A0A0000001A4D546864000000060000000100604D54726B0
000000600FF2F000; musicport=12
m=audio 5006 RTP/AVP 96
a=recvonly
a=rtpmap:96 mpeg4-generic/44100
a=mid:2
a=fmtp:96 streamtype=5; mode=rtp-midi; profile-level-id=12;
config=7A0A0000001A4D546864000000060000000100604D54726B0
000000600FF2F000; musicport=12(Same deal here: the a=fmtp lines are wrapped for the page, but they're single lines in SDP ✨)
To signal "virtual sendrecv" semantics, both streams set musicport to the same value, 12 🌸 As defined earlier in this section, pairs of identity relationship streams sent by different parties share the association that a MIDI cable pair has when it cross-connects two devices in a MIDI 1.0 network. We call it "virtual sendrecv" because streams sent by different parties in a true sendrecv session also have this property, so it's kind of a copycat situation 😭
As discussed in the preamble to Appendix C, the big perk of virtual sendrecv is that each party can customize the stream it receives 👑 In the example above, each stream has its own "config" string, which could customize the rendering algorithm for each party. Fun fact: the strings in this example are identical, because General MIDI isn't a configurable MPEG 4 renderer, so there's nothing to customize 💀
C.6. Configuration Tools: MIDI Rendering
Okay, this appendix is where the session configuration tools for rendering live 🧚 It's the main character of config stuff, no cap ✨
The "render" parameter specifies a rendering method for a stream. It gets a token value that signals the top-level rendering class. This memo defines four token values for render: "unknown", "synthetic", "api", and "null" 🌈
- An "unknown" renderer is a renderer whose nature is unspecified. It's the default renderer for native RTP MIDI streams.
- A "synthetic" renderer transforms the MIDI stream into audio output (or sometimes into stage lighting changes or other actions). It's the default renderer for mpeg4-generic RTP MIDI streams.
- An "api" renderer presents the command stream to applications via an Application Programmer Interface (API).
- The "null" renderer discards the MIDI stream 🙈
The "null" render value has special roles during Offer/Answer negotiations [RFC3264]. A party uses the "null" value in an answer to reject an offered renderer. Rejecting a renderer is independent from rejecting a payload type (which is coded by removing the payload type from a media line) and from rejecting a media stream (which is coded by zeroing the port of a media line that uses the renderer). Separate things, bestie 💅
Other render token values MAY be registered with IANA. The token value MUST adhere to the ABNF for render tokens defined in Appendix D. Registrations MUST include a complete specification of parameter value usage, similar in depth to the specifications that appear throughout Appendix C.6 for "synthetic" and "api" render values. If a party is offered a session description that uses a render token value that isn't known to the party, the party MUST NOT accept the renderer. Its options are rejecting the renderer (using the "null" value), the payload type, the media stream, or the session description 🔥
Other parameters MAY follow a render parameter in a parameter list. The additional parameters act to define the exact nature of the renderer. For example, the "subrender" parameter (defined in Appendix C.6.2) specifies the exact nature of the renderer 🦋
Special rules apply to using the render parameter in an mpeg4-generic stream. We define these rules in Appendix C.6.5 ✨
C.6.1. The multimode Parameter
A media description MAY contain several render parameters. By default, if a parameter list includes several render parameters, a receiver MUST choose exactly one renderer from the list to render the stream. The "multimode" parameter can be used to override this default. We define two token values for multimode: "one" and "all" 💖
- The default "one" value requests rendering by exactly one of the listed renderers.
- The "all" value requests the synchronized rendering of the RTP MIDI stream by all listed renderers, if possible.
If the multimode parameter appears in a parameter list, it MUST appear before the first render parameter assignment. Order matters here, periodt 💅
Render parameters appear in the parameter list in order of decreasing priority. A receiver MAY use the priority ordering to decide which renderer(s) to retain in a session 👑
If the "offer" in an Offer/Answer-style negotiation [RFC3264] contains a parameter list with one or more render parameters, the "answer" MUST set the render parameters of all unchosen renderers to "null" 😭
C.6.2. Renderer Specification
The render parameter (Appendix C.6 preamble) specifies, in a broad sense, what a renderer does with a MIDI stream. In this appendix, we describe the "subrender" parameter. Its token value defines the exact nature of the renderer. So "render" and "subrender" team up to define a renderer, the same way MIME types and MIME subtypes combine to define a type of media [RFC2045] 🧚
If the subrender parameter is used for a renderer definition, it MUST appear immediately after the render parameter in the parameter list. At most one subrender parameter may appear in a renderer definition 🎀
This document defines one value for subrender: the value "default". The "default" token specifies the use of the default renderer for the stream type (native or mpeg4-generic). The default renderer for native RTP MIDI streams is a renderer whose nature is unspecified (see point 6 in Section 6.1 for details). The default renderer for mpeg4-generic RTP MIDI streams is an MPEG 4 Audio Object Type whose ID number is 13, 14, or 15 (see Section 6.2 for details) 🌸
If a renderer definition doesn't use the subrender parameter, the value "default" is assumed for subrender 🍓
Other subrender token values may be registered with IANA. Here come the guidelines for registering subrender values 🦋
A subrender value is registered for a specific stream type (native or mpeg4-generic) and a specific render value (excluding "null" and "unknown"). Registrations for mpeg4-generic subrender values are restricted to new MPEG 4 Audio Object Types that accept MIDI input. The syntax of the token MUST adhere to the token definition in Appendix D.
For "render=synthetic" renderers, a subrender value registration specifies an exact method for transforming the MIDI stream into audio (or sometimes into video or control actions, such as stage lighting). For standardized renderers, this specification is usually a pointer to a standards document, perhaps supplemented by RTP-MIDI-specific information. For commercial products and open-source projects, this specification usually takes the form of instructions for interfacing the RTP MIDI stream with the product or project software. A "render=synthetic" registration MAY specify additional Reset State commands for the renderer (Appendix A.1) ✨
A "render=api" subrender value registration specifies how an RTP MIDI stream interfaces with an API (Application Programmers Interface). This specification is usually a pointer to programmer's documentation for the API, perhaps supplemented by RTP-MIDI-specific information 💻
A subrender registration MAY specify an initialization file (referred to in this document as an initialization data object) for the stream. The initialization data object MAY be encoded in the parameter list (verbatim or by reference) using the coding tools defined in Appendix C.6.3. An initialization data object MUST have a registered [RFC4288] media type and subtype [RFC2045] 🫶
For "render=synthetic" renderers, the data object usually encodes initialization data for the renderer (sample files, synthesis patch parameters, reverberation room impulse responses, etc.) 🎶
For "render=api" renderers, the data object usually encodes data about the stream used by the API. For example, for an RTP MIDI stream generated by a piano keyboard controller, that's the manufacturer and model number of the keyboard, for use in GUI presentation 🎹
Usually, only one initialization object is encoded for a renderer. If a renderer uses multiple data objects, the correct receiver interpretation of multiple data objects MUST be defined in the subrender registration 🌈
A subrender value registration may also specify additional parameters, to appear in the parameter list immediately after subrender. These parameter names MUST begin with the subrender value, followed by an underscore ("_"), so they don't collide with future RTP MIDI parameter names. For example, a parameter "foo_bar" defined for subrender value "foo" 🧚
We now specify guidelines for interpreting the subrender parameter during session configuration 💅
If a party is offered a session description that uses a renderer whose subrender value is not known to the party, the party MUST NOT accept the renderer. Its options are rejecting the renderer (using the "null" value), the payload type, the media stream, or the session description 🔥
Receivers MUST be aware of the Reset State commands (Appendix A.1) for the renderer specified by the subrender parameter, and MUST insure that the renderer does not experience indefinite artifacts due to the presence (or the loss) of a Reset State command. This one is serious, bestie 😭
C.6.3. Renderer Initialization 🎀
okay so if a renderer for a stream uses an initialization data object, the "rinit" parameter MUST show up in the parameter list right after "subrender" 💖 and if the renderer's parameter list has no subrender parameter (remember how "default" works in C.6.2 🧚), then "rinit" MUST come right after "render" ✨
the value you give "rinit" MUST be the media type/subtype [RFC2045] for the initialization data object 💅 and if the object type is registered under several media types, one of them being audio, the rinit assignment MUST use the audio media type 🎧
RTP MIDI has three parameters for encoding init data objects for renderers: "inline", "url", and "cid" 🌸
if a renderer uses any of "inline", "url", and/or "cid", those parameters MUST come right after "rinit" 🦋
if a "url" parameter shows up for a renderer, then "inline" MUST NOT show up, and if "inline" shows up, then "url" MUST NOT show up 🙈 neither "url" nor "inline" is required to appear, though 🤷♀️ and if neither of them follows "rinit", then "cid" MUST follow "rinit" 💜
the "inline" parameter is how you encode the data object right in the line. its value is a double-quoted Base64 [RFC2045] encoding of the binary data object, with no line breaks 🍓 Appendix E.4 has an example that builds an inline parameter value, so go peek 👀
the "url" parameter gets a double-quoted string that is a Uniform Resource Locator (URL) for the data object 🌈 that string MUST specify a HyperText Transport Protocol URL (HTTP, [RFC2616]). HTTP MAY run over TCP, or MAY run over a secure network transport, like the method in [RFC2818] 🔐 and the media type/subtype for the data object SHOULD be given in the matching HTTP transport header 📬
the "cid" parameter is for caching data objects 🗂️ its value is a double-quoted string that encodes a globally unique identifier for the data object ✨
a cid parameter MAY come right after an inline parameter 💫 in that case, the cid identifier value MUST be tied to the inline data object 🫶
if a url parameter is there, and the data object behind that URL is expected to stay the same for the whole life of the URL, then a cid parameter MAY come right after the url parameter 🌷 the cid identifier value MUST be tied to the data object for that URL, and a cid parameter with that same identifier value SHOULD be given after the data object type/subtype in the matching HTTP transport header 🎀
now, if a url parameter is there, and the data object is expected to change during the life of the URL, then a cid parameter MUST NOT come after the url parameter 🙅♀️ a receiver reads a cid parameter as a sign that it's safe to use a cached copy of the url data object, and no cid parameter means it is NOT safe to use a cached copy, because it may have changed 😭
last one: the cid parameter MAY be used without the inline and url parameters 🧚 in that case, the identifier points at a local or distributed catalog of data objects 📚
in most cases, there's only one data object coded in the parameter list for each renderer 💁♀️ like, the default renderer for mpeg4-generic streams uses a single data object (see Appendix C.6.5 for example usage) 🎶
BUT a subrender registration MAY allow multiple data objects for one renderer 🫢 if you do that, each object encoding starts with an "rinit" parameter, followed by "inline", "url", and/or "cid" parameters 🧩
init data objects MAY wrap a Standard MIDI File (SMF) 🎹 by default, the SMFs wrapped in a data object MUST be ignored by an RTP MIDI receiver 🙃 we define parameters that override this default in Appendix C.6.4 🔓
to wrap up this section, here are some guidelines for registering media types for init data objects 📝 these guidelines are on top of what's in [RFC4288] [RFC4289] 💅
some init data objects can also encode MIDI note info, so they can make a whole audio performance 🎼 those objects SHOULD be registered under the "audio" media type, so they can also be used for store-and-forward rendering, AND under the "application" media type, to support editing tools 🛠️ init objects without note storage, or init objects for non-audio renderers, SHOULD be registered ONLY under an "application" media type 🫧
C.6.4. MIDI Channel Mapping 🦋
okay, this appendix is where we map MIDI name spaces (16 voice channels + systems) onto a renderer 🎛️
in the general case:
- a session may define an ordered relationship (Appendix C.5) that shows more than one MIDI name space to a renderer 💫
- a renderer may accept any number of MIDI name spaces, or it may expect a specific number 🔢
a session description SHOULD give each renderer in the session a compatible MIDI name space 💅 and if a receiver notices that a session description has too many or too few MIDI name spaces for a renderer, then MIDI data from the extra stream name spaces MUST be discarded, and the extra renderer name spaces MUST NOT be driven with MIDI data (except as described in Appendix C.6.4.1, below) 🚫
if a parameter list defines several renderers and gives the "all" token value to the multimode parameter, then the same name space is shown to each renderer 🪞 but the "chanmask" parameter can mask out selected voice channels for each renderer 🎭 we define "chanmask" and other MIDI management parameters in the subsections below 📖
C.6.4.1. The smf_info Parameter 🧁
the smf_info parameter says how to use the SMFs encapsulated in renderer data objects (if there are any) 🎁 it also says how to use SMFs coded in the smf_inline, smf_url, and smf_cid parameters (defined in Appendix C.6.4.2) 🎀
smf_info describes the "render" parameter that most recently comes before it in the parameter list 👀 smf_info MUST NOT appear in parameter lists that don't use the "render" parameter, and it MUST NOT appear before the first use of "render" in the parameter list 🙅♀️
there are three token values for smf_info: "ignore", "sdp_start", and "identity" 🍭
- the "ignore" value means the SMFs MUST be discarded 🗑️ this is the default SMF rendering behavior, bestie 💅
- the "sdp_start" value means the SMFs MUST be rendered, AND the rendering MUST begin as soon as the session description is accepted 🎬 if a receiver is offered a session description with a renderer whose smf_info is set to sdp_start, and that receiver doesn't support rendering SMFs, then it MUST NOT accept the renderer tied to that smf_info parameter 😬 its options are to reject the renderer (by setting the "render" parameter to "null"), the payload type, the media stream, or the whole session description 🙅♀️
- the "identity" value means the SMFs code the identity of the renderer 🪪 this one is meant for the "unknown" renderer (see the Appendix C.6 preamble) 🕵️♀️ the MIDI commands coded in the SMF are informational, so they MUST NOT be presented to a renderer for audio presentation 🔇 typically, the SMF would use SysEx Identity Reply commands (
F0 7E nn 06 02, as defined in [MIDI]) to identify devices, and use device-specific SysEx commands to describe the current state of the devices (patch memory contents, etc.) 🎹
other smf_info token values MAY be registered with IANA 🏷️ the token value MUST follow the ABNF for render tokens defined in Appendix D 📐 registrations MUST include a complete specification of parameter usage, about as deep as the specs for "sdp_start" and "identity" in this appendix 📚✨
if a party is offered a session description that uses an smf_info parameter value it doesn't know, then the party MUST NOT accept the renderer tied to that smf_info parameter 🫣 its options are to reject the renderer, the payload type, the media stream, or the whole session description 🧾
now let's get into the rendering semantics for the "sdp_start" token value in detail 🔍
the SMFs and RTP MIDI streams in a session description share the same MIDI name space(s) 🤝 in the simple case of one RTP MIDI stream and one SMF, the SMF MIDI commands and RTP MIDI commands get merged into a single name space and presented to the renderer 🍰 the indefinite artifact responsibilities for merged MIDI streams defined in Appendix C.5 also apply to merging RTP and SMF MIDI data 🌀
if a payload type codes multiple SMFs, the SMF name spaces get presented to the renderer as an ordered entity 📋 to figure out the ordering of SMFs for a renderer (which SMF is "first", which is "second", and so on), use these rules:
- if the renderer uses a single data object, the order the SMFs appear in the object's internal structure sets their order (the earliest SMF in the object is "first", the next is "second", etc.) 🪆
- if multiple data objects are encoded for a renderer, the position of each data object in the parameter list sets the relative order of the SMFs encoded in each one (SMFs in parameters that appear earlier in the list are ordered before SMFs in parameters that appear later) 📑
- if SMFs are encoded in data object parameters AND in the parameters defined in C.6.4.2, the relative order of those parameters in the parameter list sets the relative order of the SMFs (SMFs in parameters that appear earlier in the list are ordered before SMFs in parameters that appear later) 🗂️
with that ordering set, here's how SMFs map to renderer name spaces 🗺️ the SMF that appears first for a renderer maps to the first renderer name space, the second maps to the second, and so on 🥇🥈 if the associated RTP MIDI streams also form an ordered relationship, then the first SMF is merged with the first name space of the relationship, the second SMF is merged with the second name space of the relationship, and so on 🔗
unless the streams and the SMFs both use MIDI Time Code, the time offset between SMF and stream data is unspecified ⏱️ that's a real limit on SMFs: they only work for stuff where sync isn't critical, like sending System Exclusive commands for renderer initialization, or human-SMF interactivity 🙋♀️
last thing 💅 each SMF in the sdp_start discussion above encodes exactly one MIDI name space (16 voice channels + systems) 🎯 so using the Device Name SMF meta event to specify several MIDI name spaces in one SMF is not supported for sdp_start 🚫💀
C.6.4.2. The smf_inline, smf_url, and smf_cid Parameters ✨
Okay so, sometimes the renderer data object doesn't actually encapsulate SMFs 🥺, but the app still wants to use SMFs the way we described in Appendix C.6.4.1 💅. That's where these three come in!
The "smf_inline", "smf_url", and "smf_cid" parameters handle exactly that situation 🎀. They use the same syntax and semantics as the inline, url, and cid parameters from Appendix C.6.3, except the encoded data object is an SMF 🧚.
These parameters belong to the "render" parameter that most recently precedes them in the session description 💜. They MUST NOT appear in parameter lists that don't use the "render" parameter, and they MUST NOT appear before the first use of "render" in the parameter list 🙅♀️. If several "smf_inline", "smf_url", or "smf_cid" parameters show up for a renderer, the order they're listed in defines the SMF name space ordering 🌸.
C.6.4.3. The chanmask Parameter 🙈
Okay bestie, the chanmask parameter tells the renderer to ignore all MIDI voice commands for certain channel numbers 💀. The parameter value is a concatenated string of "1" and "0" digits. Each string position maps to a MIDI voice channel number (system channels may not be masked). A "1" instructs the renderer to process the voice channel, and a "0" instructs the renderer to ignore it 🦋.
The string length of the chanmask parameter value MUST be 16 (for a single stream or an identity relationship) or a multiple of 16 (for an ordered relationship) 📏.
The chanmask parameter describes the "render" parameter that most recently precedes it in the session description. chanmask MUST NOT appear in parameter lists that do not use the "render" parameter, and it MUST NOT appear before the first use of "render" in the parameter list 🫶.
The chanmask parameter describes the final MIDI name spaces presented to the renderer. The SMF and stream components of the MIDI name spaces may not be independently masked 🍓.
If a receiver is offered a session description with a renderer that uses the chanmask parameter, and the receiver doesn't implement the semantics of the chanmask parameter, the receiver MUST NOT accept the renderer unless the chanmask parameter value contains only "1"s. No cap, that's a strict one 😭.
C.6.5. The audio/asc Media Type 🎧
In Appendix 11.3, we register the audio/asc media type 📝. The data object for audio/asc is a binary encoding of the AudioSpecificConfig data block used to initialize mpeg4-generic streams (Section 6.2 and [MPEGAUDIO]) ✨.
An mpeg4-generic parameter list MAY use the render, subrender, and rinit parameters with the audio/asc media type for renderer configuration. A few restrictions apply to these parameters in mpeg4-generic parameter lists, so let's go through them 👑:
- An mpeg4-generic media description that uses the render parameter MUST assign the empty string ("") to the mpeg4-generic "config" parameter. The use of the streamtype, mode, and profile-level-id parameters MUST follow the normative text in Section 6.2 💖.
- Sessions that use identity or ordered relationships MUST follow the mpeg4-generic configuration restrictions in Appendix C.5 🌈.
- The render parameter MUST be assigned the value "synthetic", "unknown", "null", or a render value that has been added to the IANA repository for use with mpeg4-generic RTP MIDI streams. The "api" token value for render MUST NOT be used 🚫.
- If a subrender parameter is present, it MUST immediately follow the render parameter, and it MUST be assigned the token value "default" or a subrender value added to the IANA repository for use with mpeg4-generic RTP MIDI streams. A subrender parameter assignment may be left out of the renderer configuration, in which case the implied value of subrender is the default value of "default" 🧁.
- If the render parameter is assigned the value "synthetic" and the subrender parameter has the value "default" (assigned or implied), the rinit parameter MUST be assigned the value "audio/asc", and an AudioSpecificConfig data object MUST be encoded using the mechanisms defined in C.6.2-3. The AudioSpecificConfig data MUST encode one of the MPEG 4 Audio Object Types defined for use with mpeg4-generic in Section 6.2. If the subrender value is other than "default", refer to the subrender registration for information on the use of "audio/asc" with the renderer 🎹.
- If the render parameter is assigned the value "null" or "unknown", the data object MAY be omitted 🫧.
Several general restrictions apply to the use of the audio/asc media type in RTP MIDI too 🙃:
- A native stream MUST NOT assign "audio/asc" to rinit. The audio/asc media type isn't meant to be a general-purpose container for rendering systems outside of MPEG usage 🙅♀️.
- The audio/asc media type defines a stored object type; it does not define semantics for RTP streams. Thus, audio/asc MUST NOT appear on an rtpmap line of a session description 💜.
Below, we show session description examples for audio/asc 👀. The first one uses the inline parameter to code the AudioSpecificConfig block for a mpeg4-generic General MIDI stream. We derive the value assigned to the inline parameter in Appendix E.4. The subrender token value of "default" is implied by the absence of the subrender parameter in the parameter list 🧚.
v=0
o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
s=Example
t=0 0
m=audio 5004 RTP/AVP 96
c=IN IP4 192.0.2.94
a=rtpmap:96 mpeg4-generic/44100
a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
profile-level-id=12; render=synthetic; rinit="audio/asc";
inline="egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAA"(The a=fmtp line has been wrapped to fit the page to accommodate memo formatting restrictions; it comprises a single line in SDP.)
The next session description uses the url parameter to code the AudioSpecificConfig block for the same General MIDI stream 🌸:
v=0
o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
s=Example
t=0 0
m=audio 5004 RTP/AVP 96
c=IN IP4 192.0.2.94
a=rtpmap:96 mpeg4-generic/44100
a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
profile-level-id=12; render=synthetic; rinit="audio/asc";
url="http://example.net/oski.asc";
cid="xjflsoeiurvpa09itnvlduihgnvet98pa3w9utnuighbuk"(The a=fmtp line has been wrapped to fit the page to accommodate memo formatting restrictions; it comprises a single line in SDP.)
C.7. Interoperability 🤝
In this appendix, we define interoperability guidelines for two application areas 🌈:
- MIDI content-streaming applications. RTP MIDI is added to RTSP-based content-streaming servers, so that viewers may experience MIDI performances (produced by a specified client-side renderer) in synchronization with other streams (video, audio) 🎬.
- Long-distance network musical performance applications. RTP MIDI is added to SIP-based voice chat or videoconferencing programs, as an alternative, or as an addition, to audio and/or video RTP streams 🎶.
For each application, we define a core set of functionality that all implementations MUST implement. Non-negotiable, periodt 💅.
The applications we address in this section are not an exhaustive list of potential RTP MIDI uses. We expect framework documents for other applications to be developed, within the IETF or within other organizations. We discuss other potential application areas for RTP MIDI in Section 1 of the main text of this memo 🦋✨.
C.7.1. MIDI Content Streaming Applications 🎧✨
okay so picture this, bestie 💖 you click a link on a web page for an Internet Radio channel, and suddenly an RTSP client wakes up and uses that link's RTSP URL to reach the RTSP server hosting the station. that's content streaming, and it's giving main character energy 🎀
The content can be pre-recorded (like replaying yesterday's football game on demand) or "live" (like the game as it's happening). Either way, the person is usually an "audience member" and not a "participant" like on a phone call 🍿
Heads up: these examples are about sending audio to an audience member. The interop guidelines in this appendix are for RTP MIDI apps of that kind. They are not for things like raw MIDI command streams used in a professional setting (recording studio, performance stage, etc.) 🙅♀️
In an RTSP session, the client works with a session description that the server "declares", either through the RTSP DESCRIBE method or some other way, like HTTP or email. The session description shows the session from the client's point of view. For example, if a media line has a non-zero port number, that's the server saying which RTP and RTCP ports it would prefer the client to use 🦋 Once media starts flowing, the server sends an RTP MIDI stream and the client renders it, maybe in sync with video or other audio streams.
Now for the interop text 🫶
Server responsibilities are mostly about limits on the "reference" session description that a server gives when it has no info about the client's capabilities. The reference session is a "lowest common denominator" session, so the odds are highest that a client will be able to view it. If the server knows what the client can do, it's free to give a session description customized for that client in the DESCRIBE reply 💅
Clients MUST support unicast UDP RTP MIDI streams that use the recovery journal with the closed-loop or the anchor sending policies. Clients MUST be able to interpret the stream subsetting and chapter inclusion parameters in the session description that qualify those sending policies. Client support of enhanced Chapter C encoding is OPTIONAL.
The reference session description a server offers MUST send all RTP MIDI UDP streams as unicast streams that use the recovery journal and the closed-loop or anchor sending policies. Servers SHOULD use the stream subsetting and chapter inclusion parameters in the reference session description, so the client's rendering job is simpler 🧚 Server support of enhanced Chapter C encoding is OPTIONAL.
Clients and servers MUST support RTSP interleaved mode (the method for interleaving RTP onto the RTSP TCP transport) 🌸
Clients MUST be able to interpret the timestamp semantics signalled by the "comex" value of the tsmode parameter, which are the timestamp semantics of Standard MIDI Files [MIDI]. Servers MUST use the "comex" value for the tsmode parameter in the reference session description 🎶
Clients MUST be able to process an RTP MIDI stream whose packets encode an arbitrary temporal duration ("media time"). So in practice, clients MUST implement a MIDI playout buffer 😭 Clients MUST NOT depend on the presence of the rtp_ptime, rtp_maxtime, and guardtime parameters in the session description in order to process packets. They SHOULD still be able to use these parameters to improve packet processing ✨
Servers SHOULD strive to send RTP MIDI streams the same way media servers send conventional audio streams. That means a sequence of packets that either all code the same temporal duration (non-normative example: 50 ms packets) or that each code one of an integral number of temporal durations (non-normative example: 50 ms, 100 ms, 250 ms, or 500 ms packets). Servers SHOULD encode information about the packetization method in the rtp_ptime and rtp_maxtime parameters in the session description 🍓
Clients MUST be able to examine the render and subrender parameters to determine whether a multimedia session uses a renderer they support. Clients MUST be able to interpret the default "one" value of the multimode parameter, so they can identify supported renderers from a list of renderer descriptions. Clients MUST be able to interpret the musicport parameter, to the degree that it matters for the renderers they support. Clients MUST be able to interpret the chanmask parameter 💜
Clients supporting renderers whose data object (as encoded by a parameter value for inline) could exceed 300 octets in size MUST support the url and cid parameters, and thus must implement the HTTP protocol in addition to RTSP 🌈
Servers MUST specify complete rendering systems for RTP MIDI streams. A minimal RTP MIDI native stream doesn't meet this requirement (Section 6.1), because the rendering method for such streams is "not specified" 🙈
At the time of this memo, the only way for servers to specify a complete rendering system is to specify an mpeg4-generic RTP MIDI stream in mode rtp-midi (Section 6.2 and C.6.5). So the only rendering systems that can presently be used are General MIDI [MIDI], DLS 2 [DLS2], or Structured Audio [MPEGSA]. The maximum inline value for General MIDI is well under 300 octets, so clients need not support the "url" parameter for it. The maximum inline values for DLS 2 and Structured Audio may be much larger than 300 octets, so clients MUST support the url parameter for those 🎹
We expect the owners of rendering systems (both standardized and proprietary) to register subrender parameters for their renderers. Once registration happens, native RTP MIDI sessions may use render and subrender (Appendix C.6.2) to specify complete rendering systems for RTSP content-streaming multimedia sessions 🦋
Servers MUST NOT use the sdp_start value for the smf_info parameter in the reference session description, since that would require clients to be able to parse and render Standard MIDI Files 🚫
Clients MUST support mpeg4-generic mode rtp-midi General MIDI (GM) sessions, at a polyphony limited by the hardware capabilities of the client. This is the "lowest common denominator" rendering system that content providers can target 👑 This requirement doesn't force implementors of a non-GM renderer (such as DLS 2 or Structured Audio) to add a second rendering engine. Instead, a client can satisfy the requirement by including a set of voice patches that implement the GM instrument set, and using that emulation for mpeg4-generic GM sessions.
It is RECOMMENDED that servers use General MIDI as the renderer for the reference session description, because clients are REQUIRED to support it. We don't require General MIDI as the reference renderer, because for normative applications it's an inappropriate choice 🙅♀️ Servers using General MIDI as a "lowest common denominator" renderer SHOULD use the Universal Real-Time SysEx MIP message [SPMIDI] to communicate the priority of voices to polyphony-limited clients 🔥
C.7.2. MIDI Network Musical Performance Applications 🎹💕
In Internet telephony and videoconferencing, parties interact over an IP network the same way they would face-to-face. Good user experiences need low end-to-end audio latency and tight audiovisual synchronization (for "lip-sync") 👄 The Session Initiation Protocol (SIP, [RFC3261]) handles session management.
In this appendix section, we define interoperability guidelines for using RTP MIDI streams in interactive SIP applications. Our main focus is Network Musical Performances (NMP), where musicians in different locations play together over the network as if they were in the same room 🎶 See [NMP] for background on NMP, and see [RFC4696] for a discussion of low-latency RTP MIDI implementation techniques for NMP.
The goal of NMP applications is telepresence: the parties should hear audio that's close to what they'd hear if they were in the same room. The interop guidelines in this appendix are for RTP MIDI applications of that kind. They are not for things like raw MIDI command streams used in a professional setting (recording studio, performance stage, etc.) 🎧
We focus on session management for two-party unicast sessions that specify a renderer for RTP MIDI streams. Within that limited scope, the guidelines here are enough to let applications interoperate. We define the REQUIRED capabilities of RTP MIDI senders and receivers in NMP sessions, and we define how exchanged session descriptions set up network musical performance sessions ✨
SIP lets parties negotiate session details using the Offer/Answer protocol [RFC3264]. However, RTP MIDI has so many parameters that "blind" negotiation between two parties using different applications might not end up with a common session configuration 🙃
So we now define a set of capabilities that NMP parties MUST support. Session description offers whose options fall outside the envelope of REQUIRED party behavior risk negotiation failure 💀 We also define session description idioms that the RTP MIDI part of an offer MUST follow, so the offer is structured for simpler analysis 🌸
We use the term "offerer" for the party making a SIP offer, and "answerer" for the party answering it. Also, unless it is qualified by the adjective "sender" or "receiver", a statement that a party MUST support X means it MUST support X for both sending and receiving 💖
If an offerer wants to define a "sendrecv" RTP MIDI stream, it may use a true sendrecv session or the "virtual sendrecv" construction described in the preamble to Appendix C and in Appendix C.5. A true sendrecv session means the offerer wants to take part in a session where both parties use identically configured renderers. A virtual sendrecv session means the offerer is willing to take part in a session where the two parties may be using different renderer configurations. So parties MUST be prepared to see both real and virtual sendrecv sessions in an offer 🦋
Parties MUST support unicast UDP transport of RTP MIDI streams. These streams MUST use the recovery journal with the closed-loop or anchor sending policies. These streams MUST use the stream subsetting and chapter inclusion parameters to declare the types of MIDI commands that will be sent on the stream (for sendonly streams) or will be processed (for recvonly streams), including the size limits on System Exclusive commands 🎀 Support of enhanced Chapter C encoding is OPTIONAL.
Note that both TCP and multicast UDP support are OPTIONAL. We make TCP OPTIONAL because we expect NMP renderers to rely on data objects (signalled by rinit and associated parameters) for initialization at the start of the session, and to use System Exclusive commands only for interactive control during the session. These interactive commands are small enough to be protected by the recovery journal mechanism of RTP MIDI UDP streams 🫶
Okay bestie, timestamps and packet timing are next, and this part is lowkey the one that makes everything else work 🕰️✨ Remember how the tsmode parameter controls what command timestamps mean inside the MIDI list of RTP packets? That's the thread we're pulling on now 🧵💖
Parties MUST support clock rates of 44.1 kHz, 48 kHz, 88.2 kHz, and 96 kHz 🎶 They also MUST support streams using the "comex", "async", and "buffer" tsmode values 💅 And recvonly offers MUST offer the default "comex" 🎀 No cap, this is the baseline everyone has to hit.
Now the packet durations, which are giving range 🌈 Parties MUST support a wide range of packet temporal durations, from rtp_ptime and rtp_maxptime values of 0 all the way up to values that code 100 ms ⏱️ Because of that, receivers MUST be able to implement a playout buffer 🧚 Bestie, that's the thing that smooths out the wobbles.
Offers and answers MUST present rtp_ptime, rtp_maxptime, and guardtime values that support the latency that users would expect in the application, subject to bandwidth constraints 🦋 Here's the tea: senders MUST abide by the values set for these parameters in a session description, so a receiver SHOULD use those values to size its playout buffer and get the lowest reliable latency for a session 🔥 If you're building something latency-sensitive, you should refer to [RFC4696] for information on packet sending algorithms 💜 Parties MUST be able to implement the semantics of the guardtime parameter, for times from 5 ms to 5000 ms ⭐
Next up is the render parameter, and this is where things get serious 👑
Sessions MUST specify complete rendering systems for all RTP MIDI streams 💅 Note that a minimal RTP MIDI native stream does not meet this requirement (Section 6.1), because the rendering method for those streams is "not specified" 🙈
At the time of writing, the only way for parties to specify a complete rendering system is to specify an mpeg4-generic RTP MIDI stream in mode rtp-midi (Section 6.2 and C.6.5) 🌸 We expect the owners of rendering systems, both standardized and proprietary, to register subrender values for their renderers 🎀 Once IANA registration happens, native RTP MIDI sessions may use render and subrender (Appendix C.6.2) to specify complete rendering systems for SIP network musical performance multimedia sessions 🧚
Now the General MIDI thing. All parties MUST support General MIDI (GM) sessions, at a polyphony limited by the hardware capabilities of the party 🎹 This is the "lowest common denominator" rendering system, and honestly, without it, practical interoperability would be quite difficult 😭 When using GM, parties SHOULD use the Universal Real-Time SysEx MIP message [SPMIDI] to communicate the priority of voices to polyphony-limited clients 💖
Note that this requirement does not force implementors of a non-GM renderer (for mpeg4-generic sessions, DLS 2, or Structured Audio) to add a second rendering engine 🫶 Instead, a client may satisfy the requirement by including a set of voice patches that implement the GM instrument set, and using that emulation for mpeg4-generic GM sessions ✨ We require GM support so that an offerer who wants to maximize interoperability can offer GM when its preferred renderer isn't accepted by the answerer 🌈
Offerers MUST NOT present several renderers as options in a session description by listing several payload types on a media line, because Section 2.1 uses that construct to let a party send several RTP MIDI streams in the same RTP session 🚫
So what should you do instead? An offerer wishing to present rendering options SHOULD offer a single payload type that offers several renderers 🦋 In this construct, the parameter list codes a list of render parameters, each followed by its support parameters 💅 As discussed in Appendix C.6.1, the order of renderers in the list declares the offerer's preference 👑 The "unknown" and "null" values MUST NOT appear in the offer 🙈 The answer MUST set all render values except the desired renderer to "null" 💜 Thus, "unknown" MUST NOT appear in the answer either 🔥
We use SHOULD instead of MUST in the first sentence of the paragraph above, because this technique does not work in all situations 🥺 Example: an offerer wishes to offer both mpeg4-generic renderers and native RTP MIDI renderers as options. In that case, the offerer MUST present a series of session descriptions, each offering a single renderer, until the answerer accepts one 🎀
Parties MUST support the musicport, chanmask, subrender, rinit, and inline parameters 🌸 Parties supporting renderers whose data object (as encoded by a parameter value for "inline") could exceed 300 octets in size MUST support the url and cid parameters, and thus must implement HTTP protocol 🌐 Note that in mpeg4-generic, General MIDI data objects cannot exceed 300 octets, but DLS 2 and Structured Audio data objects may 🎶 Support for the other rendering parameters (smf_cif, smf_info, smf_inline, smf_url) is OPTIONAL 🍓
Here's where it gets spicy: so far we've assumed the only MIDI flows driving a renderer are the network flows in the session description 🤔 In NMP applications, that assumption would mean two rendering engines, one for local use by a party and a second for the remote party 😵
In practice, though, applications may want both parties to share a single rendering engine 🫶 In this case, the session description MUST use a virtual sendrecv session, and MUST use the stream subsetting and chapter inclusion parameters to allocate which MIDI channels are intended for use by a party 💖 If two parties are sharing a MIDI channel, the application MUST ensure that appropriate MIDI merging occurs at the input to the renderer 🔀
Now let's talk about the (non-MIDI) audio streams. Audio streams may be used for two purposes: as a "talkback" channel so parties can converse, or as a way to conduct a performance that includes MIDI and audio channels 🎤 In that latter case, offers MUST use sample rates and packet temporal durations for the audio and MIDI streams that support low-latency synchronized rendering ⚡
We now show an example of an offer/answer exchange in a network musical performance application (next page). Below, we show an offer that complies with the interoperability text in this appendix section 💅
v=0
o=first 2520644554 2838152170 IN IP4 first.example.net
s=Example
t=0 0
a=group:FID 1 2
c=IN IP4 192.0.2.94
m=audio 16112 RTP/AVP 96
a=recvonly
a=mid:1
a=rtpmap:96 mpeg4-generic/44100
a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
profile-level-id=12; cm_unused=ABCFGHJKMNPQTVWXYZ; cm_used=2NPTW;
cm_used=2C0.1.7.10.11.64.121.123; cm_used=2M0.1.2
cm_used=X0-16; ch_never=ABCDEFGHJKMNPQTVWXYZ;
ch_default=2NPTW; ch_default=2C0.1.7.10.11.64.121.123;
ch_default=2M0.1.2; cm_default=X0-16;
rtp_ptime=0; rtp_maxptime=0; guardtime=44100;
musicport=1; render=synthetic; rinit="audio/asc";
inline="egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAA"
m=audio 16114 RTP/AVP 96
a=sendonly
a=mid:2
a=rtpmap:96 mpeg4-generic/44100
a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
profile-level-id=12; cm_unused=ABCFGHJKMNPQTVWXYZ; cm_used=1NPTW;
cm_used=1C0.1.7.10.11.64.121.123; cm_used=1M0.1.2
cm_used=X0-16; ch_never=ABCDEFGHJKMNPQTVWXYZ;
ch_default=1NPTW; ch_default=1C0.1.7.10.11.64.121.123;
ch_default=1M0.1.2; cm_default=X0-16;
rtp_ptime=0; rtp_maxptime=0; guardtime=44100;
musicport=1; render=synthetic; rinit="audio/asc";
inline="egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAA"(The a=fmtp lines have been wrapped to fit the page to accommodate memo formatting restrictions; it comprises a single line in SDP.)
The owner line (o=) identifies the session owner as "first" 🧑💻 The session description defines two MIDI streams: a recvonly stream where "first" receives a performance, and a sendonly stream that "first" uses to send one 🎧 The recvonly port number encodes the ports where "first" wants to receive RTP (16112) and RTCP (16113) media at IP4 address 192.0.2.94 🌟 The sendonly port number encodes the port where "first" wants to receive RTCP for that stream (16115) 📡
The musicport parameters code that the two streams share an identity relationship, so they form a virtual sendrecv stream 🦋 Both streams are mpeg4-generic RTP MIDI streams that specify a General MIDI renderer 🎹 The stream subsetting parameters code that the recvonly stream uses MIDI channel 1 exclusively for voice commands, and that the sendonly stream uses MIDI channel 2 exclusively for voice commands 🍓 This mapping lets the application software share a single renderer for local and remote performers, which is honestly iconic 👑
Now the answer to the offer 💌
v=0
o=second 2520644554 2838152170 IN IP4 second.example.net
s=Example
t=0 0
a=group:FID 1 2
c=IN IP4 192.0.2.105
m=audio 5004 RTP/AVP 96
a=sendonly
a=mid:1
a=rtpmap:96 mpeg4-generic/44100
a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
profile-level-id=12; cm_unused=ABCFGHJKMNPQTVWXYZ; cm_used=2NPTW;
cm_used=2C0.1.7.10.11.64.121.123; cm_used=2M0.1.2
cm_used=X0-16; ch_never=ABCDEFGHJKMNPQTVWXYZ;
ch_default=2NPTW; ch_default=2C0.1.7.10.11.64.121.123;
ch_default=2M0.1.2; cm_default=X0-16;
rtp_ptime=0; rtp_maxptime=882; guardtime=44100;
musicport=1; render=synthetic; rinit="audio/asc";
inline="egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAA"
m=audio 5006 RTP/AVP 96
a=recvonly
a=mid:2
a=rtpmap:96 mpeg4-generic/44100
a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
profile-level-id=12; cm_unused=ABCFGHJKMNPQTVWXYZ; cm_used=1NPTW;
cm_used=1C0.1.7.10.11.64.121.123; cm_used=1M0.1.2
cm_used=X0-16; ch_never=ABCDEFGHJKMNPQTVWXYZ;
ch_default=1NPTW; ch_default=1C0.1.7.10.11.64.121.123;
ch_default=1M0.1.2; cm_default=X0-16;
rtp_ptime=0; rtp_maxptime=0; guardtime=88200;
musicport=1; render=synthetic; rinit="audio/asc";
inline="egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAA"(The a=fmtp lines have been wrapped to fit the page to accommodate memo formatting restrictions; they comprise single lines in SDP.)
The owner line (o=) identifies the session owner as "second" 🙌 Both port numbers for the media streams are non-zero, so "second" has accepted the session description ✅ The stream marked "sendonly" in the offer is marked "recvonly" in the answer, and vice versa, coding the different view of the session held by "session" 🪞 The IP4 number (192.0.2.105), the RTP ports (5004 and 5006), and the RTCP ports (5005 and 5007) have been changed by "second" to match its transport wishes 🌸
"second" also made several parameter changes 💅 The rtp_maxptime for the sendonly stream has been changed to code 2 ms (441 in clock units), and the guardtime for the recvonly stream has been doubled 🎀 Because these modifications request capabilities that are REQUIRED to be implemented by interoperable parties, "second" can make them with confidence that "first" can abide by them 🫶 Periodt. 💜
D. Parameter Syntax Definitions 🎀
Okay bestie, this is the syntax zone 💅 In this appendix we write out the exact grammar for every RTP MIDI media type parameter in Augmented Backus-Naur Form (ABNF, [RFC4234]), so nobody has to guess. When you use these parameters with SDP, all parameters MUST appear on a single fmtp attribute line of an RTP MIDI media description. For mpeg4-generic RTP MIDI streams, that same line MUST also include any mpeg4-generic parameters (the usage is described in Section 6.2). ✨ An fmtp attribute line may be defined (after [RFC3640]) as:
;
; SDP fmtp line definition
;
fmtp = "a=fmtp:" token SP param-assign 0*(";" SP param-assign) CRLFHere <token> codes the RTP payload type. Note that white space MUST NOT appear between the "a=fmtp:" and the RTP payload type. 🙈
We now define the syntax of the parameters from Appendix C. The definition is built up piece by piece as the <param-assign> token, so each parameter just adds its own line. See [RFC3640] for the syntax of the mpeg4-generic parameters discussed in Section 6.2. 🦋
;
;
; top-level definition for all parameters
;
;
;
; Parameters defined in Appendix C.1
param-assign = ("cm_unused=" (([channel-list] command-type
[f-list]) / sysex-data))
param-assign =/ ("cm_used=" (([channel-list] command-type
[f-list]) / sysex-data))
;
; Parameters defined in Appendix C.2
param-assign =/ ("j_sec=" ("none" / "recj" / *ietf-extension))
param-assign =/ ("j_update=" ("anchor" / "closed-loop" /
"open-loop" / *ietf-extension))
param-assign =/ ("ch_default=" (([channel-list] chapter-list
[f-list]) / sysex-data))
param-assign =/ ("ch_never=" (([channel-list] chapter-list
[f-list]) / sysex-data))
param-assign =/ ("ch_anchor=" (([channel-list] chapter-list
[f-list]) / sysex-data))
;
; Parameters defined in Appendix C.3
param-assign =/ ("tsmode=" ("comex" / "async" / "buffer"))
param-assign =/ ("linerate=" nonzero-four-octet)
param-assign =/ ("octpos=" ("first" / "last"))
param-assign =/ ("mperiod=" nonzero-four-octet)
;
; Parameter defined in Appendix C.4
param-assign =/ ("guardtime=" nonzero-four-octet)
param-assign =/ ("rtp_ptime=" four-octet)
param-assign =/ ("rtp_maxptime=" four-octet)
;
; Parameters defined in Appendix C.5
param-assign =/ ("musicport=" four-octet)
;
; Parameters defined in Appendix C.6
param-assign =/ ("chanmask=" ( 1*( 16( "0" / "1" ) )))
param-assign =/ ("cid=" double-quote cid-block
double-quote)
param-assign =/ ("inline=" double-quote base-64-block
double-quote)
param-assign =/ ("multimode=" ("all" / "one"))
param-assign =/ ("render=" ("synthetic" / "api" / "null" /
"unknown" / *extension))
param-assign =/ ("rinit=" mime-type "/" mime-subtype)
param-assign =/ ("smf_cid=" double-quote cid-block
double-quote)
param-assign =/ ("smf_info=" ("ignore" / "identity" /
"sdp_start" / *extension))
param-assign =/ ("smf_inline=" double-quote base-64-block
double-quote)
param-assign =/ ("smf_url=" double-quote uri-element
double-quote)
param-assign =/ ("subrender=" ("default" / *extension))
param-assign =/ ("url=" double-quote uri-element
double-quote)
;
; list definitions for the cm_ command-type
;
command-type = command-part1 command-part2 command-part3
command-part1 = (*1"A") (*1"B") (*1"C") (*1"F") (*1"G") (*1"H")
command-part2 = (*1"J") (*1"K") (*1"M") (*1"N") (*1"P") (*1"Q")
command-part3 = (*1"T") (*1"V") (*1"W") (*1"X") (*1"Y") (*1"Z")
;
; list definitions for the ch_ chapter-list
;
chapter-list = ch-part1 ch-part2 ch-part3
ch-part1 = (*1"A") (*1"B") (*1"C") (*1"D") (*1"E") (*1"F") (*1"G")
ch-part2 = (*1"H") (*1"J") (*1"K") (*1"M") (*1"N") (*1"P") (*1"Q")
ch-part3 = (*1"T") (*1"V") (*1"W") (*1"X") (*1"Y") (*1"Z")
;
; list definitions for the ch_ channel-list
;
channel-list = midi-chan-element *("." midi-chan-element)
midi-chan-element = midi-chan / midi-chan-range
midi-chan-range = midi-chan "-" midi-chan
; decimal value of left midi-chan
; MUST be strictly less than decimal
; value of right midi-chan
midi-chan = %d0-15
;
; list definitions for the ch_ field list (f-list)
;
f-list = midi-field-element *("." midi-field-element)
midi-field-element = midi-field / midi-field-range
midi-field-range = midi-field "-" midi-field
;
; decimal value of left midi-field
; MUST be strictly less than decimal
; value of right midi-field
midi-field = four-octet
;
; large range accommodates Chapter M
; RPN (0-16383) and NRPN (16384-32767)
; parameters, and Chapter X octet sizes.
;
; definitions for ch_ sysex-data
;
sysex-data = "__" h-list *("_" h-list) "__"
h-list = hex-field-element *("." hex-field-element)
hex-field-element = hex-octet / hex-field-range
hex-field-range = hex-octet "-" hex-octet
;
; hexadecimal value of left hex-octet
; MUST be strictly less than hexadecimal
; value of right hex-octet
hex-octet = 2("0" / "1" / "2"/ "3" / "4" /
"5" / "6" / "7" / "8" / "9" /
"A" / "B" / "C" / "D" / "E" / "F")
;
; rewritten version of hex-octet in [RFC2045]
; (page 23).
; note that a-f are not permitted, only A-F.
; hex-octet values MUST NOT exceed 7F.
;
; definitions for rinit parameter
;
mime-type = "audio" / "application"
mime-subtype = token
;
; See Appendix C.6.2 for registration
; requirements for rinit type/subtypes.
;
; definitions for base64 encoding
; copied from [RFC4566]
base-64-block = *base64-unit [base64-pad]
base64-unit = 4base64-char
base64-pad = 2base64-char "==" / 3base64-char "="
base64-char = %x41-5A / %x61-7A / %x30-39 / "+" / "/"
; A-Z, a-z, 0-9, "+" and "/"
;
; generic rules
;
ietf-extension = token
;
; ietf-extension may only be defined in
; standards-track RFCs.
extension = token
;
; extension may be defined by filing
; a registration with IANA.
four-octet = %d0-4294967295
; unsigned encoding of 32-bits
nonzero-four-octet = %d1-4294967295
; unsigned encoding of 32-bits, ex-zero
uri-element = URI-reference
; as defined in [RFC3986]
double-quote = %x22
; the double-quote (") character
token = 1*token-char
; copied from [RFC4566]
token-char = %x21 / %x23-27 / %x2A-2B / %x2D-2E /
%x30-39 / %x41-5A / %x5E-7E
; copied from [RFC4566]
cid-block = 1*cid-char
cid-char = token-char
cid-char =/ "@"
cid-char =/ ","
cid-char =/ ";"
cid-char =/ ":"
cid-char =/ "\"
cid-char =/ "/"
cid-char =/ "["
cid-char =/ "]"
cid-char =/ "?"
cid-char =/ "="
;
; add back in the tspecials [RFC2045], except for
; double-quote and the non-email safe () <>
; note that "cid" defined above ensures that
; cid-block is enclosed with double-quotes
; external references
; URI-reference: from [RFC3986]
;
; End of ABNFThe mpeg4-generic RTP payload [RFC3640] defines a "mode" parameter that signals the type of MPEG stream in use. ✨ We add a brand new mode value, "rtp-midi", using the ABNF rule below:
;
; mpeg4-generic mode parameter extension
;
mode =/ "rtp-midi"
; as described in Section 6.2 of this memoE. A MIDI Overview for Networking Specialists ✨
okay bestie, this appendix is a little MIDI crash course for my networking people who are new to musical stuff 🎀 it's giving friendly intro energy. if you're actually implementing anything, go read [MIDI] for the normative rules, no cap 💅
Musicians make music by doing a controlled sequence of physical movements. like, a pianist plays by coordinating a bunch of key presses, key releases, and pedal actions 🎹 MIDI captures a performance by encoding those physical gestures as a sequence of MIDI commands. this high-level musical representation is super compact but lowkey fragile: one lost command could totally wreck the performance 💀
MIDI commands are kinda giving microprocessor machine instructions 🤖 they're defined as binary elements, and the bitfields inside a command have a regular structure and a specialized purpose. for example, the upper nibble of the first command octet (the opcode field) codes the command type 🧠 MIDI commands can be any number of complete octets, but most are 1, 2, or 3 octets long 🫶
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| Channel Voice Messages | Bitfield Pattern |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| NoteOff (end a note) | 1000cccc 0nnnnnnn 0vvvvvvv |
|-------------------------------------------------------------|
| NoteOn (start a note) | 1001cccc 0nnnnnnn 0vvvvvvv |
|-------------------------------------------------------------|
| PTouch (Polyphonic Aftertouch) | 1010cccc 0nnnnnnn 0aaaaaaa |
|-------------------------------------------------------------|
| CControl (Controller Change) | 1011cccc 0xxxxxxx 0yyyyyyy |
|-------------------------------------------------------------|
| PChange (Program Change) | 1100cccc 0ppppppp |
|-------------------------------------------------------------|
| CTouch (Channel Aftertouch) | 1101cccc 0aaaaaaa |
|-------------------------------------------------------------|
| PWheel (Pitch Wheel) | 1110cccc 0xxxxxxx 0yyyyyyy |
-------------------------------------------------------------Figure E.1 -- MIDI Channel Messages
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| System Common Messages | Bitfield Pattern |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| System Exclusive | 11110000, followed by a |
| | list of 0xxxxxx octets, |
| | followed by 11110111 |
|-------------------------------------------------------------|
| MIDI Time Code Quarter Frame | 11110001 0xxxxxxx |
|-------------------------------------------------------------|
| Song Position Pointer | 11110010 0xxxxxxx 0yyyyyyy |
|-------------------------------------------------------------|
| Song Select | 11110011 0xxxxxxx |
|-------------------------------------------------------------|
| Undefined | 11110100 |
|-------------------------------------------------------------|
| Undefined | 11110101 |
|-------------------------------------------------------------|
| Tune Request | 11110110 |
|-------------------------------------------------------------|
| System Exclusive End Marker | 11110111 |
-------------------------------------------------------------
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| System Realtime Messages | Bitfield Pattern |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| Clock | 11111000 |
|-------------------------------------------------------------|
| Undefined | 11111001 |
|-------------------------------------------------------------|
| Start | 11111010 |
|-------------------------------------------------------------|
| Continue | 11111011 |
|-------------------------------------------------------------|
| Stop | 11111100 |
|-------------------------------------------------------------|
| Undefined | 11111101 |
|-------------------------------------------------------------|
| Active Sense | 11111110 |
|-------------------------------------------------------------|
| System Reset | 11111111 |
-------------------------------------------------------------Figure E.2 -- MIDI System Messages
Figure E.1 and E.2 show the MIDI command family 🦋 there are three major classes of commands: voice commands (opcode field values in the range 0x8 through 0xE), system common commands (opcode field 0xF, commands 0xF0 through 0xF7), and system real-time commands (opcode field 0xF, commands 0xF8 through 0xFF). voice commands code the musical gestures for each timbre in a composition, while system commands handle stuff that usually affects all voice channels, like System Reset (0xFF) 🌈
E.1. Commands Types 🎶
voice commands run on one of 16 MIDI channels, as coded by their 4-bit channel field (field cccc in Figure E.1). most applications assign notes for different timbres to different channels 🎀 to support apps that need more than 16 channels, MIDI systems use several MIDI command streams in parallel, which gets you 32, 48, or 64 MIDI channels 🔥
as an example of a voice command, look at NoteOn (opcode 0x9), with binary encoding 1001cccc 0nnnnnnn 0aaaaaaa. this one signals the start of a musical note on MIDI channel cccc. the note has a pitch coded by the note number nnnnnnn, and an onset amplitude coded by note velocity aaaaaaa 🎹
other voice commands signal the end of notes (NoteOff, opcode 0x8), map a specific timbre to a MIDI channel (PChange, opcode 0xC), or set the value of parameters that modulate the timbral quality (all other voice commands). the exact meaning of most voice channel commands depends on the rendering algorithms the MIDI receiver uses to generate sound. in most applications, a MIDI sender has a model (in some sense) of the rendering method the receiver uses 🧚
system commands handle a bunch of global tasks in the stream 🌸 that includes "sequencer" playback control of pre-recorded MIDI commands (the Song Position Pointer, Song Select, Clock, Start, Continue, and Stop messages), SMPTE time code (the MIDI Time Code Quarter Frame command), and the communication of device-specific data (the System Exclusive messages) 💜
E.2. Running Status 🍓
every MIDI command bitfield shares a special structure: the leading bit of the first octet is set to 1, and the leading bit of all subsequent octets is set to 0. this structure supports a data compression system called running status [MIDI], which improves the coding efficiency of MIDI 🙈
in running status coding, the first octet of a MIDI voice command may be dropped if it is identical to the first octet of the previous MIDI voice command. this rule, combined with a convention to consider NoteOn commands with a null third octet as NoteOff commands, supports coding note sequences using two octets per command 🥹
running status coding is only used for voice commands. the presence of a system common message in the stream cancels running status mode for the next voice command. however, system real-time messages do not cancel running status mode.
E.3. Command Timing ⏱️
the bitfield formats in Figures E.1 and E.2 do not encode the execution time for a command. timing information is not part of the MIDI command syntax itself; different applications of the MIDI command language use different methods to encode timing 🫧
for example, the MIDI command set acts as the transport layer for MIDI 1.0 DIN cables [MIDI]. MIDI cables are short asynchronous serial lines that facilitate the remote operation of musical instruments and audio equipment 🎛️ timestamps are not sent over a MIDI 1.0 DIN cable. instead, the standard uses an implicit "time of arrival" code, so receivers execute MIDI commands at the moment of arrival ⚡
in contrast, Standard MIDI Files (SMFs, [MIDI]), a file format for representing complete musical performances, add an explicit timestamp to each MIDI command. they use a delta encoding scheme that is optimized for the statistics of musical performance 📜 SMF timestamps usually code timing using the metric notation of a musical score, and SMF meta-events add a tempo map to the file, so score beats can be accurately converted into seconds during rendering 🎼✨
E.4. AudioSpecificConfig Templates for MMA Renderers ✨
Okay bestie, let's talk about AudioSpecificConfig, the tiny config blob that tells a renderer which sound engine to use 🎹💖. In Section 6.2 and Appendix C.6.5, session descriptions carry an AudioSpecificConfig data block to pick a MIDI rendering algorithm for mpeg4-generic RTP MIDI streams 💜.
The bitfield format is defined in [MPEGAUDIO]. The StructuredAudioSpecificConfig, the key structure coded inside AudioSpecificConfig, is defined in [MPEGSA] 🧚.
If you want to write Structured Audio renderers, you need to fully understand [MPEGSA] and [MPEGAUDIO], no shortcuts 💀. Most of us will only use the two MIDI Manufacturers Association renderers that can be named in AudioSpecificConfig: General MIDI (GM, [MIDI]) and Downloadable Sounds 2 (DLS 2, [DLS2]) 🎀.
To help those implementors out, we copied the AudioSpecificConfig bitfield formats for a GM renderer and a DLS 2 renderer below 🌸. We checked these carefully and we believe they're correct 🫶. BUT, and this is important, the material below is informative, and [MPEGAUDIO] and [MPEGSA] are the normative definitions for AudioSpecificConfig 🙈. Those are the real source of truth.
As described in Section 6.2, a minimal mpeg4-generic session description encodes the AudioSpecificConfig binary bitfield as a hexadecimal string (the format is defined in [RFC3640]) that gets assigned to the config parameter ✨. As described in Appendix C.6.3, a session description that uses the render parameter encodes the AudioSpecificConfig binary bitfield as a Base64-encoded string assigned to the inline parameter, or in the body of an HTTP URL assigned to the url parameter 🦋.
Here's a simplified binary AudioSpecificConfig bitfield format, good for sending and receiving GM and DLS 2 data:
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| AOTYPE |FREQIDX|CHANNEL|SACNK| FILE_BLK 1 (required) ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|1|SACNK| FILE_BLK 2 (optional) ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| ... |1|SACNK| FILE_BLK N (optional) ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|0|0| (first "0" bit terminates FILE_BLK list)
+-+-+
Figure E.3 -- Simplified AudioSpecificConfigThe 5-bit AOTYPE field is the Audio Object Type, as an unsigned integer. The legal values for mpeg4-generic RTP MIDI streams are "15" (General MIDI), "14" (DLS 2), and "13" (Structured Audio) 🎶. So receivers that don't support all three mpeg4-generic renderers may parse the first 5 bits of an AudioSpecificConfig coded in a session description, and reject sessions that ask for renderers they don't support.
The 4-bit FREQIDX field gives the sampling rate of the renderer. Here's how FREQIDX values map to sampling rates, in Figure E.4 📊.
Sampling rate rules! Senders MUST specify a sampling frequency that matches the RTP clock rate, if possible. If that's not possible, senders MUST specify the escape value. Receivers MUST consult the RTP clock parameter for the true sampling rate if the escape value is specified. 🚨
FREQIDX Sampling Frequency
0x0 96000
0x1 88200
0x2 64000
0x3 48000
0x4 44100
0x5 32000
0x6 24000
0x7 22050
0x8 16000
0x9 12000
0xa 11025
0xb 8000
0xc reserved
0xd reserved
0xe reserved
0xf escape value
Figure E.4 -- FreqIdx encodingThe 4-bit CHANNEL field gives the number of audio channels for the renderer. The values 0x1 to 0x5 mean 1 to 5 audio channels. The value 0x6 means 5+1 surround sound, and 0x7 means 7+1 surround sound 🔊. If the rtpmap line in the session description specifies one of these formats, CHANNEL MUST be set to the corresponding value. Otherwise, CHANNEL MUST be set to 0x0 🌈.
After CHANNEL comes a list of one or more binary file data blocks. The 3-bit SACNK field (the chunk_type field in class StructuredAudioSpecificConfig, defined in [MPEGSA]) says what type each data block is 🍓.
For General MIDI, only Standard MIDI Files may appear in the list (SACNK field value 2). For DLS 2, only Standard MIDI Files and DLS 2 RIFF files (SACNK field value 4) may appear. For both of these file types, the FILE_BLK field looks like Figure E.5: a 32-bit unsigned integer (FILE_LEN) counting the bytes in the SMF or RIFF file, followed by FILE_LEN bytes of file data 🧁.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| FILE_LEN (32-bit, a byte count SMF file or RIFF file) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| FILE_DATA (file contents, a list of FILE_LEN bytes) ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure E.5 -- The FILE_BLK field formatNote that several files may follow the CHANNEL field. The "1" constant fields in Figure E.3 code the presence of another file, and the "0" constant field codes the end of the list. The final "0" bit in Figure E.3 codes the absence of special coding tools (see [MPEGAUDIO] for details). Senders not using these tools MUST append this "0" bit. Receivers that don't understand these coding tools MUST ignore all data following a "1" in this position 🙃.
The StructuredAudioSpecificConfig bitfield structure requires the presence of one FILE_BLK. For mpeg4-generic RTP MIDI use of DLS 2, FILE_BLKs MUST code RIFF files or SMF files. For mpeg4-generic RTP MIDI use of General MIDI, FILE_BLKs MUST code SMF files. By default, this SMF will be ignored (Appendix C.6.4.1). In that default case, a GM StructuredAudioSpecificConfig bitfield SHOULD code a FILE_BLK whose FILE_LEN is 0, and whose FILE_DATA is empty 🫧.
To wrap up this appendix, we derive the StructuredAudioSpecificConfig we use in the General MIDI session examples in this memo 💅. Looking at Figure E.3, for GM, AOTYPE = 15. Our examples use a 44,100 Hz sample rate (FREQIDX = 4) and are mono (CHANNEL = 1). For GM, a single SMF is encoded (SACNK = 2), using the SMF shown in Figure E.6 (a 26 byte file) 🎵.
--------------------------------------------
| MIDI File = <Header Chunk> <Track Chunk> |
--------------------------------------------
<Header Chunk> = <chunk type> <length> <format> <ntrks> <divsn>
4D 54 68 64 00 00 00 06 00 00 00 01 00 60
<Track Chunk> = <chunk type> <length> <delta-time> <end-event>
4D 54 72 6B 00 00 00 04 00 FF 2F 00
Figure E.6 -- SMF file encoded in the examplePlacing these constants in binary format into the data structure shown in Figure E.3 gives the constant shown in Figure E.7 🤩.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|0 1 1 1 1|0 1 0 0|0 0 0 1|0 1 0|0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|0 0 0 0 0 0 0 0 0 0 0 1 1 0 1 0|0 1 0 0|1 1 0 1|0 1 0 1|0 1 0 0|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|0 1 1 0|1 0 0 0|0 1 1 0|0 1 0 0|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|0 0 0 0|0 0 0 0|0 0 0 0|0 1 1 0|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|0 0 0 0|0 0 0 0|0 0 0 1|0 0 0 0|0 0 0 0|0 1 1 0|0 0 0 0|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|0 1 0 0|1 1 0 1|0 1 0 1|0 1 0 0|0 1 1 1|0 0 1 0|0 1 1 0|1 0 1 1|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|0 1 1 0|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|0 0 0 0|0 0 0 0|1 1 1 1|1 1 1 1|0 0 1 0|1 1 1 1|0 0 0 0|0 0 0 0|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|0|0|
+-+-+
Figure E.7 -- AudioSpecificConfig used in GM examplesExpressing this bitfield as an ASCII hexadecimal string gives:
7A0A0000001A4D546864000000060000000100604D54726B0000000600FF2F000This string gets assigned to the config parameter in the minimal mpeg4-generic General MIDI examples in this memo (like the example in Section 6.2). Expressing this string in Base64 [RFC2045] gives:
egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAAThat string gets assigned to the inline parameter in the General MIDI example shown in Appendix C.6.5 🎀✨.
References ✨
Normative References
Okay bestie, this is the reading list the spec needs you to have on hand before you touch any of the above. These are the normative references, so they're part of the actual rules 💅 Here they are, one per line:
- [MIDI] MIDI Manufacturers Association. "The Complete MIDI 1.0 Detailed Specification", 1996.
- [RFC3550] Schulzrinne, H., Casner, S., Frederick, R., and V. Jacobson, "RTP: A Transport Protocol for Real-Time Applications", STD 64, RFC 3550, July 2003.
- [RFC3551] Schulzrinne, H. and S. Casner, "RTP Profile for Audio and Video Conferences with Minimal Control", STD 65, RFC 3551, July 2003.
- [RFC3640] van der Meer, J., Mackie, D., Swaminathan, V., Singer, D., and P. Gentric, "RTP Payload Format for Transport of MPEG-4 Elementary Streams", RFC 3640, November 2003.
- [MPEGSA] International Standards Organization. "ISO/IEC 14496 MPEG-4", Part 3 (Audio), Subpart 5 (Structured Audio), 2001.
- [RFC4566] Handley, M., Jacobson, V., and C. Perkins, "SDP: Session Description Protocol", RFC 4566, July 2006.
- [MPEGAUDIO] International Standards Organization. "ISO 14496 MPEG-4", Part 3 (Audio), 2001.
- [RFC2045] Freed, N. and N. Borenstein, "Multipurpose Internet Mail Extensions (MIME) Part One: Format of Internet Message Bodies", RFC 2045, November 1996.
- [DLS2] MIDI Manufacturers Association. "The MIDI Downloadable Sounds Specification", v98.2, 1998.
- [RFC4234] Crocker, D. and P. Overell, "Augmented BNF for Syntax Specifications: ABNF", RFC 4234, October 2005.
- [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997.
- [RFC3711] Baugher, M., McGrew, D., Naslund, M., Carrara, E., and K. Norrman, "The Secure Real-time Transport Protocol (SRTP)", RFC 3711, March 2004.
- [RFC3264] Rosenberg, J. and H. Schulzrinne, "An Offer/Answer Model with Session Description Protocol (SDP)", RFC 3264, June 2002.
- [RFC3986] Berners-Lee, T., Fielding, R., and L. Masinter, "Uniform Resource Identifier (URI): Generic Syntax", STD 66, RFC 3986, January 2005.
- [RFC2616] Fielding, R., Gettys, J., Mogul, J., Frystyk, H., Masinter, L., Leach, P., and T. Berners-Lee, "Hypertext Transfer Protocol -- HTTP/1.1", RFC 2616, June 1999.
- [RFC3388] Camarillo, G., Eriksson, G., Holler, J., and H. Schulzrinne, "Grouping of Media Lines in the Session Description Protocol (SDP)", RFC 3388, December 2002.
- [RP015] MIDI Manufacturers Association. "Recommended Practice 015 (RP-015): Response to Reset All Controllers", 11/98.
- [RFC4288] Freed, N. and J. Klensin, "Media Type Specifications and Registration Procedures", BCP 13, RFC 4288, December 2005.
- [RFC3555] Casner, S. and P. Hoschka, "MIME Type Registration of RTP Payload Formats", RFC 3555, July 2003.
Informative References
Now the informative ones ✨ These are the extra background reads. Nothing here is required, but they're super good for going deeper:
- [NMP] Lazzaro, J. and J. Wawrzynek. "A Case for Network Musical Performance", 11th International Workshop on Network and Operating Systems Support for Digital Audio and Video (NOSSDAV 2001) June 25-26, 2001, Port Jefferson, New York.
- [GRAME] Fober, D., Orlarey, Y. and S. Letz. "Real Time Musical Events Streaming over Internet", Proceedings of the International Conference on WEB Delivering of Music 2001, pages 147-154.
- [RFC3261] Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston, A., Peterson, J., Sparks, R., Handley, M., and E. Schooler, "SIP: Session Initiation Protocol", RFC 3261, June 2002.
- [RFC2326] Schulzrinne, H., Rao, A., and R. Lanphier, "Real Time Streaming Protocol (RTSP)", RFC 2326, April 1998.
- [ALF] Clark, D. D. and D. L. Tennenhouse. "Architectural considerations for a new generation of protocols", SIGCOMM Symposium on Communications Architectures and Protocols, (Philadelphia, Pennsylvania), pp. 200--208, IEEE, Sept. 1990.
- [RFC4696] Lazzaro, J. and J. Wawrzynek, "An Implementation Guide for RTP MIDI", RFC 4696, November 2006.
- [RFC2205] Braden, R., Zhang, L., Berson, S., Herzog, S., and S. Jamin, "Resource ReSerVation Protocol (RSVP) -- Version 1 Functional Specification", RFC 2205, September 1997.
- [RFC4288] Freed, N. and J. Klensin, "Media Type Specifications and Registration Procedures", BCP 13, RFC 4288, December 2005.
- [RFC4289] Freed, N. and J. Klensin, "Multipurpose Internet Mail Extensions (MIME) Part Four: Registration Procedures", BCP 13, RFC 4289, December 2005.
- [RFC4571] Lazzaro, J. "Framing Real-time Transport Protocol (RTP) and RTP Control Protocol (RTCP) Packets over Connection-Oriented Transport", RFC 4571, July 2006.
- [RFC2818] Rescorla, E., "HTTP Over TLS", RFC 2818, May 2000.
- [SPMIDI] MIDI Manufacturers Association. "Scalable Polyphony MIDI, Specification and Device Profiles", Document Version 1.0a, 2002.
- [LCP] Apple Computer. "Logic 7 Dedicated Control Surface Support", Appendix B. Product manual available from www.apple.com.
Authors' Addresses ✨
Meet the brains behind it 🧠🎀 Here's how to reach them:
John Lazzaro (corresponding author)
UC Berkeley
CS Division
315 Soda Hall
Berkeley CA 94720-1776
EMail: lazzaro@cs.berkeley.edu
John Wawrzynek
UC Berkeley
CS Division
631 Soda Hall
Berkeley CA 94720-1776
EMail: johnw@cs.berkeley.eduFull Copyright Statement 📜
Now the legal bit, which I'm keeping word for word so nothing gets lost in translation 🙏
Copyright (C) The IETF Trust (2006).
This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights.
This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST, AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Intellectual Property 🔐
Here's the vibe on IP rights, straight from the source 💜
The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf-ipr@ietf.org.
Acknowledgement 🫶
And last but not least, a shoutout to whoever keeps the lights on: funding for the RFC Editor function is currently provided by the Internet Society. Thank you, legends 🌈
📦 where this page comes from
- ietf: rfc4695 2006-11
- 💖 girlified by claude-haiku-5-5 on : a language model's rewrite of the original, under its terms