Velo: velocity from the FM-1's keys

An experiment firmware for the M-VAVE FM-1. Its keys are rubber domes with carbon pills and a single contact each, scanned as a diode matrix: on or off, no velocity. But the chip's ADC can read two of the four key rows (PA7 and PA8 are SARADC channels 0 and 1), and a carbon pill's resistance falls steeply with force. So this firmware samples those rows between two ticks of the key scan and reads each key's contact resistance: 1023 open, falling as the key is pressed harder. A press crosses a TOUCH threshold and then a PRESS threshold, and the time between them is the velocity. Twenty of the 27 keys (F3 to C5) are on those rows; the top seven are not and play at a fixed velocity. No hardware changes.

Status: working; parked 2026-10-14 with per-key calibration built but not yet right (the first session with it played more extremes, not fewer: notes/01-lab.md, "Calibration"). Untested so far: CHORD, aftertouch in drive mode by feel, release velocity (not sent). Parked on 2026-10-08 after the first session: read through the row's 10k pull-up, every played key seats at the same reading (the matrix diode's drop) and only slow presses read soft. Revived by DRIVE: a held key's every other sample is taken with its row driven high by one of the pin's weaker drivers (DRIVE 1 is the datasheet's 2.4 mA level; DRIVE 2, the default, repeats firm presses within +-5 counts), which then reads the carbon pill's resistance above the diode's drop. Soft and firm presses separate by 100..150 counts where the pull-up gave 5; on a log scale of that resistance, free playing gives velocities with quartiles 40 / 59 / 74, and the held force is polyphonic aftertouch. The top seven keys (no converter) are read through the LED line PA10. notes/01-lab.md has the measurements (section 02), the converter settings, and the traps.

The lab

• The screen shows the last press as a curve (8 samples before the touch, 48 in all, 2.2 ms apart), the three thresholds as lines, the velocity and the touch-to-press time, every key's live reading as a bar, and the converter's conversion time in the header.
• Keys play a small velocity-sensitive FM voice and send USB MIDI notes (channel 1) with the velocity, so a DAW's velocity display is the second instrument.
• tools/velo_log.py prints one line per press and appends each curve to a CSV. The FM-1 sends each capture as SysEx (F0 7D 56 4C 01 key vel dt dt n samples F7); FX toggles it.
• Pages on SELECT (or EDIT): VELOCITY has MAX, AT RANGE and LOW / HIGH (the landing's height above the key's floor that means 127 and 1; with DRIVE on, CURVE and LIGHTEST for the driven log scale); CONVERTER has the ADC clock divider and start delay (divider 7, delay 2: anything faster returns rails), immediate reads (IMMED), and the key lights; EXPERIMENT has DRIVE (a held key sampled with its row driven high, to see the pill behind the matrix diode), CHORD (a chord's notes share its first velocity) and TOP / TOP RANGE (the top seven keys' velocity: fixed, from context, or a probe through the LED line PA10; notes/01-lab.md, section 02). SAVE keeps the settings; OCT-/OCT+ the octave; ALGORITHM the fixed velocity of the top keys.

Build, test, flash

As the other firmwares (see ../CLAUDE.md): ./build.sh makes build/velo.fwsc, sh tests/run_tests.sh runs the host tests (the sensing state machine on synthetic curves, the voice, Felucca's platform tests), python3 tools/fm1_install.py build/velo.fwsc --yes flashes. Package identity FM-1_5 (dev builds FM-1_500), USB name "Velo FM-1".

Where things are

Built on the platform layer of Felucca (Hügelton Instruments, GPL-3.0-only), by way of FoMni. GPL-3.0-only.