You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Drashna Jaelre c80e5f9f88
Audio system overhaul (#11820)
* Redo Arm DAC implementation for additive, wavetable synthesis, sample playback

changes by Jack Humbert on an implementation for DAC audio on arm/chibios platforms
this commits bundles the changes from the arm-dac-work branch focused on audio/audio_arm.* into one commit (leaving out the test-keyboard)

f52faeb5d (origin/arm-dac-work) add sample and wavetable examples, parsers for both
  -> only the changes on audio_arm_.*, the keyboard related parts are split off to a separate commit
bfe468ef1 start morphing wavetable
474d100b5 refined a bit
208bee10f play_notes working
3e6478b0b start in-place documentation of dac settings
3e1826a33 fixed blip (rounding error), other waves, added key selection (left/right)
73853d651 5 voices at 44.1khz
dfb401b95 limit voices to working number
9632b3379 configuration for the ez
6241f3f3b notes working in a new way

* Redo Arm DAC implementation for additive, wavetable synthesis, sample playback

changes by Jack Humbert on an implementation for DAC audio on arm/chibios platforms

this commit splits off the plank example keymap from commit
    f52faeb5d (origin/arm-dac-work) add sample and wavetable examples, parsers for both

* refactoring: rename audio_ to reflect their supported hardware-platform and audio-generation method: avr vs arm, and pwm vs dac

* refactoring: deducplicate ISR code to update the pwm duty-cycle and period in the avr-pwm-implementation

pulls three copies of the same code into one function
which should improve readability and maintainability :-)

* refactoring: move common code of arm and avr implementation into a separate/new file

* refactoring: audio_avr_pwm, renaming defines to decouple them from actually used timers, registers and ISRs

* refactoring: audio_avr_pwm - replacing function defines with plain register defines

aligns better with other existing qmk code (and the new audio_arm_pwm) doing similar pwm thing

* add audio-arm-pwm

since not all STM32 have a DAC onboard (STM32F2xx and STM32F3xx), pwm-audio is an alternative (STM32F1xx)
this code works on a "BluePill" clone, with an STM32F103C8B

* clang-format changes on quantum/audio/* only

* audio_arm_dac: stopping the notes caused screeching when using the DAC audio paths

* audio_arm_pwm: use pushpull on the pin; so that a piezzo can be hooked up direclty without additional components (opendrain would require an external pullup)

* refactoring: remove unused file from/for atmel-avr chips

* refactoring: remove unused (avr) wavetable file

* audio_arm_dac: adapt dac_end callback to changed chibios DAC api

the previous chibios (17.6.0) passed along a pointer into the buffer plus a sample_count (which are/already where included in the DACDrivre object) - the current chibios (19.1.0) only passes the driver object.
this patch ports more or less exactly what the previous chibios ISR code did: either have the user-callback work the first or second half of the buffer (dacsample_t pointer, with half the DAC_BUFFER_SIZE samples) by adjusting the pointer and sample count

* audio-arm-dac: show a compile-warning on undefined audio-pins

Co-Authored-By: Drashna Jaelre <>

* audio_arm_dac: switch from exemplary wavetable generation to sine only

sine+triangle+squrare is exemplary, and not realy fit for "production" use
'stairs' are usefull for debugging (hardware, with an oscilloscope)

* audio_arm_dac: enable output buffers in the STM32

to drive external loads without any additional ciruitry - external opamps and such

* audio: prevent out-of-bounds array access

* audio_arm_dac: add output-frequency correcting factor

* audio_arm_pwm: get both the alternate-function and pm-callback variants back into working condition

and do some code-cleanup, refine documentation, ...

* audio_arm_pwm: increase pwm frequency for "higher fidelity"

on the previous .frequency=100000 higher frequency musical notes came out wrong
(frequency measured on a Tektronix TDS2014B)
note | freq | arm-pwm
C2 | 65.4 | 65.491
C5 | 523.25 | 523.93
C6 | 1046.5 | 1053.38
C7 | 2093 | 2129
C8 | 4186 | 4350.91

with .frequency = 500000
C8 | 4186 | 4204.6

* audio refactoring: remove unused variables

* audio_arm_dac: calibrate note tempo: with a tempo of 60beats-per-second a whole-note should last for exactly one second

* audio: allow feature selection in

so the user can switch the audio driver between DAC and PWM on STM32 boards which support both (STM32F2 and up)
or select the "pin alternate" pwm mode, for example on STM32F103

* audio-refactoring: move codeblocks in audio.[ch] into more coherent groups

and add some inline documentation

* audio-refactoring: cleanup and streamline common code between audio_arm_[dac|pwm]

untangeling the relation between audio.c and the two drivers
and adding more documenting comments :-)

* audio_avr_pwm: getting it back into working condition, and cleanup+refactor

* audio-refactoring: documentation and typo fixes

Co-Authored-By: Nick Brassel <>

* audio-refactoring: cleanup defines, inludes and remove debug-prints

* audio_chibios_dac: define&use a minimal sampling rate, based on the available tone-range

to ease up on the cpu-load, while still rendering the higher notes/tones sufficiently
also reenable the lower tones, since with the new implementation there is no evidence of them still beeing 'bugged'

* audio-refactoring: one common AUDIO_MAX_VOICES define for all audio-drivers

* audio-chibios-pwm: pwm-pin-allternate: make the the timer, timer-channel and alternate function user-#definable

* audio_chibios_dac: math.h has fmod for this

* Redo Arm DAC implementation for additive, wavetable synthesis, sample playback

update Jack Humberts dac-example keymaps for the slight changes in the audio-dac interface

* audio-refactoring: use a common AUDIO_PIN configuration switch instead of defines

have the user select a pin by configuration in instead of a define in config.h
has the advantage of beeing in a common form/pattern across all audio-driver implementations

* audio-refactoring: switch backlight_avr.c to the new AUDIO_PIN defines

* audio-common: have advance_note return a boolean if the note changed, to the next one in the melody beeing played

* audio-chibios-pwm: fix issue with ~130ms silence between note/frequency changes while playing a SONG

through trial,error and a scope/logic analyzer figured out Chibios-PWMDriver (at least in the current version) misbehaves if the initial period is set to zero (or one; two seems to work); when thats the case subsequent calls to 'pwmChhangePeriod' + pwmEnableChannel took ~135ms of silence, before the PWM continued with the new frequency...

* audio-refactoring: get 'play_note' working again

with a limited number of available voices (say AUDIO_VOICES_MAX=1) allow new frequencies to be played, by discarding the oldest one in the 'frequencies' queue

* audio: set the fallback driver to DAC for chibios and PWM for all others (==avr at the moment)

* audio-refactoring: moore documentation

and some cleanup

* audio-avr-pwm: no fallback on unset AUDIO_PIN

this seems to be the expected behaviour by some keyboards (looking at ckeys/handwire_101:default) which otherwise fail to build because the firmware-image ends up beeing too large for the atmega... so we fail silently instead to keep travis happy

* audio-refactoring: untangling terminology: voice->tone

the code actually was working on tones (combination of pitch/frequency, duration, timbre, intensity/volume) and not voices (characteristic sound of an instrument; think piano vs guitar, which can be played together, each having its own "track" = voice on a music sheet)

* audio-pwm: allow freq=0 aka a pause/rest in a SONG

continue processing, but do not enable pwm units, since freq=0 wouldn't produce any sound anyway (and lead to division by zero on that occasion)

* audio-refactoring: audio_advance_note -> audio_advance_state

since it does not only affect 'one note', but the internally kept state as a whole

* audio-refactoring: untangling terminology: polyphony

the feature om the "inherited" avr code has little to do with polyphony (see wikipedia), but is more a time-multiplexing feature, to work around hardware limitations - like only having one pwm channel, that could on its own only reproduce one voice/instrument at a time

* audio-chibios-dac: add zero-crossing feature

have tones only change/stop when the waveform approaches zero - to avoid audible clicks
note that this also requires the samples to start at zero, since the internally kept index into the samples is reset to zero too

* audio-refactoring: feature: time-multiplexing of tones on a single output channel

this feature was in the original avr-pwm implementation misnomed as "polyphony"
with polyphony_rate and so on; did the same thing though: time-multiplexing multiple active notes so that a single output channel could reproduce more than one note at a time (which is not the same as a polyphony - see wikipedia :-) )

* audio-avr-pwm: get music-mode working (again) on AVRs

with both pwm channels, or either one of the two :-)
play_notes worked already - but music_mode uses play_note

* audio-refactoring: split define MAX_SIMULTANEOUS_TONES -> TONE_STACKSIZE

since the two cases are independant from one another, the hardware might impose limitations on the number of simultaneously reproducable tones, but the audio state should be able to track an unrelated number of notes recently started by play_note

* audio-arm-dac: per define selectable sample-luts

plus generation script in ./util

* audio-refactoring: heh, avr has a MIN...

* audio-refactoring: add basic dac audio-driver based on the current/master implementation

whereas current=d96380e65496912e0f68e6531565f4b45efd1623
which is the state of things before this whole audio-refactoring branch

boiled down to interface with the refactored audio system = removing all
redundant state-managing and frequency calculation

* audio-refactoring: rename audio-drivers to driver_$PLATFORM_$DRIVER

* audio-arm-pwm: split the software/hardware implementations into separate files

which saves us partially from a 'define hell', with the tradeoff that now two somewhat similar chibios_pwm implementations have to be maintained

* audio-refactoring: update documentation

* audio-arm-dac: apply AUDIO_PIN defines to driver_chibios_dac_basic

* audio-arm-dac: dac_additive: stop the hardware when the last sample completed

the audio system calls for a driver_stop, which is delayed until the current sample conversion finishes

* audio-refactoring: make function-namespace consistent

- all (public) audio functions start with audio_
- also refactoring play*_notes/tones to play*_melody, to visually distance it a bit from play*_tone/_note

* audio-refactoring: consistent define namespace: DAC_ -> AUDIO_DAC_

* audio-arm-dac: update (inline) documentation regarding MAX for sample values

* audio-chibios-dac: remove zero-crossing feature

didn't quite work as intended anyway, and stopping the hardware on close-to-zero seems to be enought anyway

* audio-arm-dac: dac_basic: respect the configured sample-rate

* audio-arm-pwm: have 'note_timbre' influence the pwm-duty cycle

like it already does in the avr implementation

* audio-refactoring: get VIBRATO working (again)

with all drivers (verified with chibios_[dac|pwm])

* audio-arm-dac: zero-crossing feature (Mk II)

wait for the generated waveform to approach 'zero' before either turning off the output+timer or switching to the current set of active_tones

* audio-refactoring: re-add note-resting -> introduce short_rest inbetween

- introduce a short pause/rest between two notes of the same frequency, to separate them audibly
- also updating the refactoring comments

* audio-refactoring: cleanup refactoring remnants

remove the former avr-isr code block - since all its features are now refactored into the different parts of the current system

also updates the TODOS

* audio-refactoring: reserve negative numbers as unitialized frequencies

to allow the valid tone/frequency f=0Hz == rest/pause

* audio-refactoring: FIX: first note of melody was missing

the first note was missing because 'goto_next_note'=false overrode a state_change=true of the initial play_tone
and some code-indentations/cleanup of related parts

* audio-arm-dac: fix hardware init-click

due to wron .init= value

* audio-refactoring: new conveniance function: audio_play_click

which can be used to further refactor/remove fauxclicky (avr only) and/or the 'clicky' features

* audio-refactoring: clang-format on quantum/audio/*

* audio-avr-pwm: consecutive notes of the same frequency get a pause inserted inbetween by audio.c

* audio-refactoring: use milliseconds instead of seconds for 'click' parameters

clicks are supposed to be short, seconds make little sense

* audio-refactoring: use timer ticks instead of counters

local counters were used in the original (avr)ISR to advance an index into the lookup tables (for vibrato), and something similar was used for the tone-multiplexing feature
decoupling these from the (possibly irregular) calls to advance_state made sesne, since those counters/lookups need to be in relation to a wall-time anyway

* audio-refactoring: voices.c: drop 'envelope_index' counter in favour of timer ticks

* audio-refactoring: move vibrato and timbre related parts from audio.c to voices.c

also drops the now (globally) unused AUDIO_VIBRATO/AUDIO_ENABLE_VIBRATO defines

* audio.c: use system-ticks instead of counters the drivers have to take care of for the internal state posision

since there already is a system-tick with ms resolution, keeping count separatly with each driver implementation makes little sense; especially since they had to take special care to call audio_advance_state with the correct step/end parameters for the audio state to advance regularly and with the correct pace

* audio.c: stop notes after new ones have been started

avoids brief states of with no notes playing that would otherwise stop the hardware and might lead to clicks

* audio.c: bugfix: actually play a pause

instead of just idling/stopping which lead the pwm drivers to stop entirely...

* audio-arm-pwm: pwm-software: add inverted output

new define AUDIO_PIN_ALT_AS_NEGATIVE will generate an inverted signal on the alternate pin, which boosts the volume if a piezo is connected to both AUDIO_PIN and AUDIO_PIN_ALT

* audio-arm-dac: basic: handle piezo configured&wired to both audio pins

* audio-refactoring: docs: update for AUDIO_PIN_ALT_AS_NEGATIVE and piezo wiring

* audio.c: bugfix: use timer_elapsed32 instad of keeping timestamps

avoids running into issues when the uint32 of the timer overflows

* audio-refactoring: add 'pragma once' and remove deprecated NOTE_REST

* audio_arm_dac: basic: add missing bracket

* audio.c: fix delta calculation

was in the wrong place, needs to use the 'last_timestamp' before it was reset

* audio-refactoring: buildfix: wrong legacy macro for set_timbre

* audio.c: 16bit timerstamps suffice

* audio-refactoring: separate includes for AVR and chibios

* audio-refactoring: timbre: use uint8 instead of float

* audio-refactoring: duration: use uint16 for internal per-tone/note state

* audio-refactoring: tonemultiplexing: use uint16 instead of float

* audio-arm-dac: additive: set second pin output-low

used when a piezo is connected to AUDIO_PIN and AUDIO_PIN_ALT, with PIN_ALT_AS_NEGATIVE

* audio-refactoring: move AUDIO_PIN selection from to config.h

to be consistent with how other features are handled in QMK

* audio-refactoring: buildfix: wrong legacy macro for set_tempo

* audio-arm-dac: additive: set second pin output-low -- FIXUP

* audio.c: do duration<>ms conversion in uint instead of float

on AVR, to save a couple of bytes in the firmware size

* audio-refactoring: cleanup eeprom defines/usage

for ARM, avr is handled automagically through the avr libc and

Co-Authored-By: Drashna Jaelre <>

* audio.h: throw an error if OFF is larger than MAX

* audio-arm-dac: basic: actually stop the dac-conversion on a audio_driver_stop

to put the output pin in a known state == AUDIO_DAC_OFF_VALUE, instead of just leaving them where the last conversion was... with AUDIO_PIN_ALT_AS_NEGATIVE this meant one output was left HIGH while the other was left LOW

one CAVEAT: due to this change the opposing squarewave when using both A4 and A5 with AUDIO_PIN_ALT_AS_NEGATIVE
show extra pulses at the beginning/end on one of the outputs, the two waveforms are in sync otherwise.
the extra pusles probably matter little, since this is no high-fidelity sound generation :P

* audio-arm-dac: additive: move zero-crossing code out of dac_value_generate

which is/should be user-overridable == simple, and doing one thing: providing sample values
state-transitions necessary for the zero crossing are better handled in the surrounding loop in the dac_end callback

* audio-arm-dac: dac-additive: zero-crossing: ramping up or down

after a start trigger ramp up: generate values until zero=OFF_VALUE is reached, then continue normally
same in reverse for strop trigger: output values until zero is reached/crossed, then keep OFF_VALUE on the output

* audio-arm-dac: dac-additive: BUGFIX: return OFF_VALUE when a pause is playing

fixes a bug during SONG playback, which suddenly stopped when it encoutnered a pause

* audio-arm-dac: set a sensible default for AUDIO_DAC_VALUE_OFF

1/2 MAX was probably exemplary, can't think of  a setup where that would make sense :-P

* audio-arm-dac: update synth_sample/_wavetable for new pin-defines

* audio-arm-dac:  default for AUDIO_DAC_VALUE_OFF

turned out that zero or max are bad default choices:
when multiple tones are played (>>5) and released at the same time (!), due to the complex waveform never reaching 'zero' the output can take quite a while to reach zero, and hence the zero-crossing code only "releases" the output waaay to late

* audio-arm-dac: additive: use DAC for negative pin

instead of PAL, which only allows the pin to be configured as output; LOW or HIGH

* audio-arm-dac: more compile-time configuration checks

* audio-refactoring: typo fixed

* audio-refactoring: clang-format on quantum/audio/*

* audio-avr-pwm: add defines for B-pin as primary/only speaker

also updates documentation.

* audio-refactoring: update documentation with proton-c config.h example

* audio-refactoring: move glissando (TODO) to voices.c

refactored/saved from the original glissando implementation in then upstream-master:audio_avr.c

still needs some work though, as it is now the calculation *should* work, but the start-frequency needs to be tracked somewhere/somehow; not only during a SONG playback but also with user input?

* audio-refactoring: cleanup: one round of aspell -c

* audio-avr-pwm: back to AUDIO_PIN

since config_common.h expands them to plain integers, the AUDIO_PIN define can directly be compared to e.g. B5
so there is no need to deal with separate defines like AUDIO_PIN_B5

* audio-refactoring: add technical documentation

which moves some in-code documentation there

* audio-arm-dac: move AUDIO_PIN checks into c-code

instead of doing everything with the preprocessor, since A4/A5 do not expand to simple integers, preprocessor int-comparison is not possible. but necessary to get a consistent configuration scheme going throughout the audio-code... solution: let c-code handle the different AUDIO_PIN configurations instead (and leave code/size optimizations to the compiler)

* audio-arm-dac: compile-fix: set AUDIO_PIN if unset

workaround to get the build going again, and be backwarts compatible to arm-keyboards which not yet set the AUDIO_PIN define. until the define is enforced through an '#error"

* audio-refactoring: document tone-multiplexing feature

* audio-refactoring: Apply suggestions from documentation review

Co-authored-by: James Young <>

* audio-refactoring: Update docs/

* audio-refactoring: docs: fix markdown newlines

Terminating a line in Markdown with <space>-<space>-<linebreak> creates an HTML single-line break (<br>).

Co-authored-by: James Young <>

* audio-arm-dac: additive: fix AUDIO_PIN_ALT handling

* audio-arm-pwm: align define naming with other drivers

Co-authored-by: Joel Challis <>

* audio-refactoring: set detault tempo to 120

and add documentation for the override

* audio-refactoring: update backlight define checks to new AUDIO_PIN names

* audio-refactoring: reworking PWM related defines

to be more consistent with other QMK code

Co-authored-by: Joel Challis <>

* audio-arm: have the state-update-timer user configurable

defaulting to GPTD6 or GPTD8 for stm32f2+ (=proton-c)
stm32f1 might need to set this to GPTD4, since 6 and 8 are not available

* audio-refactoring: PLAY_NOTE_ARRAY was already removed in master

* Add prototype for startup

* Update chibiOS dac basic to disable pins on stop

* Add defaults for Proton C

* avoid hanging audio if note is completely missed

* Don't redefine pins if they're already defined

* Define A4 and A5 for CTPC support

* Add license headers to keymap files

* Remove figlet? comments

* Add DAC config to audio driver docs

* Apply suggestions from code review

Co-authored-by: Jack Humbert <>

* Add license header to py files

* correct license header

* Add JohSchneider's name to modified files

AKA credit where credit's due

* Set executable permission and change interpeter

* Add 'wave' to pip requirements

* Improve documentation

* Add some settings I missed

* Strip AUDIO_DRIVER to parse the name correctly

* fix depreciated

* Update util/

Co-authored-by: Jack Humbert <>

* Fix type in clueboard config

* Apply suggestions from tzarc

Co-authored-by: Nick Brassel <>

Co-authored-by: Johannes <>
Co-authored-by: JohSchneider <>
Co-authored-by: Nick Brassel <>
Co-authored-by: James Young <>
Co-authored-by: Joel Challis <>
Co-authored-by: Joshua Diamond <>
Co-authored-by: Jack Humbert <>
2 years ago
keymaps Audio system overhaul (#11820) 2 years ago
matrix.c Remove legacy print backward compatiblitly (#11805) 2 years ago



A wireless split compact keyboard.

Keyboard Maintainer: @reversebias
Hardware Supported: Mitosis PCB
Hardware Availability: See the Mitosis keyboard announcement and discussion

Make example for this keyboard (after setting up your build environment):

make mitosis:default

See build environment setup then the make instructions for more information.


Some circuit board manufacturers including DirtyPCBs and PCBWay offer a steeply discounted "prototyping" rate for a small quantity of identical circuit boards less than 100x100mm in size. The Mitosis was designed to take advantage of this, so that individuals might affordably manufacture their own without waiting for a group-buy.

These configuration files were based off the Atreus keyboard. It assumes a Pro Micro is being used, however retains the 'make upload' feature from the Atreus branch.

This keyboard uses a completely different 'matrix scan' system than most other keyboards supported by QMK. Here, QMK runs in a Pro Micro on a receiver module, and communicates only with an nRF51822 microcontroller module that in turn does wireless communication. The nRF51822 maintains a matrix of keystates received from the same microcontrollers on each of the keyboard halves. The matrix.c file contains the code to make the Pro Micro poll the external wireless microcontroller for the key matrix. As long as this file is not changed, all other QMK features are supported.

Mitosis keyboard build log, including many photos and notes about the assembly process.

Mitosis keyboard hardware design files, including PCB schematics and manufacturing files, parts list, and a laser-cutting template for the neoprene base.

Mitosis keyboard wireless firmware for the Nordic microcontrollers.