MIDI 2.0 & the Universal MIDI Packet

MIDI 2.0 is not a new cable or a new byte stream. It is a new packet format, the Universal MIDI Packet (UMP), plus a higher-resolution set of messages and ways for devices to discover each other. MIDI 1.0 messages still exist inside UMP, so the two protocols live side by side.

What changes, and why

UMP basics

A UMP is made of 1 to 4 32-bit words, so packets are 32, 64, 96 or 128 bits long. The top four bits of the first word are the Message Type (MT), which also sets the packet size. A receiver can always tell how long a packet is, even when it doesn't know the message.

For most message types, the next four bits are the group (0–15). Each group works like a separate MIDI 1.0 cable with its own 16 channels. Utility (0x0) and UMP Stream (0xF) messages have no group; they address the whole Endpoint.

Message types

Worked example: one Note On, two ways

Middle C (note 60) on channel 1, group 1 (written as 0 in the packet), at full velocity.

MIDI 2.0 Protocol Note On (MT 0x4, 64 bits)

0x40903C00 0xFFFF0000

Word 1

4 MT 4b

0 Group 4b

9 Status 4b

0 Channel 4b

3C Note 8b

00 Attr type 8b

Word 2

FFFF Velocity (16-bit) 16b

0000 Attribute data 16b

• MT = 4: MIDI 2.0 Channel Voice, so the packet is 64 bits.
• Group = 0, status = 9 (Note On), channel = 0 (channel 1).
• Note = 0x3C (60). The note number is still 7 bits; the top bit of that byte is 0.
• Attribute type = 0 (none). Other defined types are 1 = manufacturer specific, 2 = profile specific, and 3 = pitch 7.9, a fixed-point pitch in semitones for microtonal notes.
• Velocity = 0xFFFF, the maximum of a 16-bit range.

The same note as MIDI 1.0 in UMP (MT 0x2, 32 bits)

0x20903C64, which is the classic 90 3C 64 (velocity 100) with a one-byte MT and group prefix.

Word 1

2 MT 4b

0 Group 4b

9 Status 4b

0 Channel 4b

3C Note 8b

64 Velocity 8b

MIDI 2.0 Channel Voice messages

Every MT 0x4 message has the same first word: MT | group | status | channel | index/note | .... The status nibble picks the message:

Field layouts at a glance

Value scaling: Min-Center-Max

Converting a 7-bit value to 16 or 32 bits is not just a left shift. A plain shift turns 127 into 0xFE00, so the maximum would no longer be the maximum. The spec defines a Min-Center-Max upscaling rule that keeps three points exact:

• Minimum: 0 always maps to 0.
• Center: the source midpoint (64 for 7-bit, 8192 for 14-bit) maps to the exact destination midpoint (0x8000, 0x80000000).
• Maximum: the source maximum maps to all ones.

Values at or below the center are shifted left. For values above the center, the bits below the source's top bit are repeated into the empty low-order bits of the result. Values above the center are spread evenly up to the maximum, and values below it keep the plain shift.

// Reference algorithm from the UMP spec (abridged)
function scaleUp(srcVal, srcBits, dstBits) {
  const scaleBits = dstBits - srcBits;
  let out = srcVal << scaleBits;               // plain shift
  const center = 1 << (srcBits - 1);
  if (srcVal <= center) return out;            // min .. center: done
  const repeatBits = srcBits - 1;
  let repeat = srcVal & ((1 << repeatBits) - 1);  // bits below the top bit
  repeat = scaleBits > repeatBits
    ? repeat << (scaleBits - repeatBits)
    : repeat >> (repeatBits - scaleBits);
  while (repeat) { out |= repeat; repeat >>= repeatBits; }
  return out;
}

Downscaling is a simple right shift. For example, a 16-bit velocity becomes 7 bits with v >> 9.

Endpoints, Function Blocks and the UMP Stream

A UMP Endpoint is one end of a bidirectional UMP connection, such as a USB MIDI 2.0 device or a virtual port. It may contain up to 32 Function Blocks, each a logical unit (a synth engine, a keyboard, a DIN port) that spans one or more groups. Hosts learn this layout from UMP Stream messages (MT 0xF), which identify themselves with a 2-bit format and a 10-bit status:

Choosing a protocol

An Endpoint may support the MIDI 1.0 Protocol, the MIDI 2.0 Protocol, or both. Since UMP 1.1 (2023), the host picks one with Stream Configuration Request/Notification, and the same exchange turns JR timestamps on or off. MIDI-CI 1.1 negotiated the protocol with its own SysEx messages; MIDI-CI 1.2 deprecated that mechanism, so new implementations should use UMP Stream messages.

Jitter Reduction timestamps

A sender can put a JR Timestamp utility message (MT 0x0) in front of other messages. It records when the event happened, as a 16-bit count in units of 1/31,250 second (32 µs). The receiver tracks the sender's clock from periodic JR Clock messages and uses the timestamps to undo transport jitter. This helps most on buffered links such as USB and networks. Musical timing, as opposed to transport timing, uses Delta Clockstamps in ticks (with DCTPQ) in MIDI Clip Files.

SysEx, Flex Data and files

• SysEx7 (MT 0x3) carries MIDI 1.0 SysEx, up to six bytes per 64-bit packet, with a status of complete, start, continue or end. The F0 and F7 framing bytes are not sent.
• SysEx8 (MT 0x5) carries full 8-bit data with a stream ID, so binary payloads don't need 7-bit encoding.
• Flex Data (MT 0xD) holds tempo, time signature, key signature, metronome and chord-name messages, plus text such as lyrics and metadata. It replaces many SMF meta events for live streams.
• MIDI Clip File and MIDI Container File are the UMP-based successors to the Standard MIDI File.

Platform support

Libraries with UMP support include AM_MIDI2.0Lib, ni-midi2, ktmidi, cmidi2 and libremidi. For moving UMP over USB, BLE and networks, see Transports.