You hold a six-note chord on an MPE controller, add a seventh note, and something odd happens: an earlier note cuts off, or worse, a note gets stuck bending in a direction you did not ask for. Nothing is broken. You have hit one of two ceilings, and knowing which one changes the fix entirely.

Two ceilings, not one

Ceiling one: MIDI channels. MPE works by giving every sounding note its own MIDI channel, so that pitch bend, channel pressure and CC74 apply to that note alone. A standard Lower Zone uses channel 1 as the master and channels 2–16 as member channels — 15 simultaneous notes maximum, and fewer if you have configured a smaller zone or split the keyboard into two zones.

Ceiling two: synth voices. Independently of MIDI, the instrument has a polyphony setting. A plugin set to 8 voices will only ever sound 8 notes, no matter how many channels arrive.

These two numbers are set in different places, by different people, and almost never match. Most "disappearing note" problems come from the smaller of the two silently winning.

Channel rotation and the release problem

When you lift a finger, the note enters its release stage. The sound is still playing, and it still owns its channel — but the controller considers that channel free and will hand it to the next note you play.

With a short release, nobody notices. With a five-second reverb tail or a slow pad release, the consequences are audible: the new note sends pitch bend on a channel where an old note is still ringing, and the dying note bends along with it. This is the classic MPE artefact of "ghost notes sliding around in the tail".

Good MPE implementations use channel rotation — cycling through member channels in order rather than reusing the lowest free one — precisely to maximise the gap before a channel is reused. Most controllers, including the LinnStrument and Osmose, do this by default. Some synths also offer a "hold channel until release completes" option. If your instrument has one, turn it on before you start blaming the controller.

Voice stealing is a design decision

When a synth runs out of voices, it has to take one from somewhere. Different instruments make different choices:

For MPE material this matters more than for ordinary keyboard playing, because expressive parts tend to have long, evolving notes that you want to keep. A stealing algorithm tuned for staccato piano parts will make the wrong call repeatedly. Bitwig Studio and open-source instruments like Surge XT expose these settings; many commercial plugins do not.

Practical settings that help

Match your zone to your instrument. If your plugin is running 8 voices, there is no benefit to a 15-channel zone — you are just increasing the chance of channel reuse during releases. Narrow the zone to roughly match the voice count.

Raise polyphony before widening the zone. Voice count is usually the real constraint. Physically modelled and additive instruments are the exception: their per-voice CPU cost is high enough that raising polyphony has a real price. Check your meter.

Shorten releases for dense playing. Counter-intuitive, but a two-second release on an instrument you play in six-note chords creates far more channel collisions than the same patch played in thirds.

Watch your bandwidth on DIN. Each note in an MPE performance can carry three continuous streams. Fifteen of those on a 31,250 bps DIN cable will choke before the voice count does. USB gives you far more headroom.

Verify with a monitor. Rather than guessing, watch the actual data. A dedicated MPE visualiser, or your DAW's MIDI monitor, will show you immediately whether a note is being dropped at the controller, at the channel level, or at the voice allocator.

The short version

If notes vanish when you add more fingers, check the synth's voice count. If notes bend when they should not, check channel rotation and release times. They feel like the same bug from the keyboard, and they have nothing to do with each other.

Reference: the MPE specification at MIDI.org.