An Oberheim OB-1 oscillator section as a machine for the Digitakt mk1 and Digitone mk1 / Keys. It's added to the OS by elekloader. Pick OB-1 as a track's machine and its machine pages play two analog-style VCOs. The track's filter, envelopes, LFOs, level and sends work on it as they do on the stock machine.
- Two VCOs. Each has the OB-1's two waveshapes:
- saw type, from triangle to sawtooth;
- pulse type, from a narrow pulse to a square;
- a sub-octave square.
- Hard sync (VCO2 to VCO1) and cross mod (VCO2's saw sweeping VCO1's pitch, about ±3.92 octaves).
- A noise source.
- The OB-1's mixer: VCO1, VCO1 OCT, VCO2, VCO2 OCT and NOISE, each ON, −3 dB or OFF as on the panel, AC-coupled.
- Reduced aliasing with polyBLEP and polyBLAMP corrections, with limits at high pitches and under sync.
- Two machine pages with fifteen controls. On the Digitakt, the SRC pages put oscillators first and mixer/routing second; on the Digitone, SYN1 is mixer/routing and SYN2 is oscillators.
- Display graphics showing waveform shapes, tuning and routing, with vertical three-position switches for output levels.
Digitakt: OS 1.53 and 1.54. Digitone: OS 1.44. Both use machine id 69.
The Digitone machine is documented in machines/digitone/README.md.
You need the stock OS file for the device you are building, from Elektron, and macOS or Linux with mise. Run mise bootstrap once for the m68k toolchain, tools, submodules and the emulator's dependencies.
Digitakt_OS1.54.syx (or 1.53's, with ELEKLOADER_STOCK in mise.toml pointed at it). The build patches it, and only its main OS section changes.
mkdir -p firmware
cp ~/Downloads/Digitakt_OS1.54.syx firmware/
mise run firmware:digitakt # builds everything and writes out/Digitakt_custom.syxDigitone_and_Digitone_Keys_OS1.44.syx, with DIGITONE_STOCK in mise.toml pointed at yours. The build patches it the same way; the bootloader is never changed.
mkdir -p firmware
cp ~/Downloads/Digitone_and_Digitone_Keys_OS1.44.syx firmware/
mise run firmware:digitone # builds everything and writes out/ob1-dn-1.44-0.6.syxmise run firmware builds both devices and needs both stock files.
With Elektron Transfer, as for any OS update (elekloader's steps):
- Connect the unit over USB and open Transfer.
- Select the unit and Connect.
- Drag the built file onto Drop files here (
out/Digitakt_custom.syxorout/ob1-dn-1.44-0.6.syx). - Press YES on the unit, and don't power it off until it has written the OS and restarted.
The bootloader is never changed, so the stock OS always restores the unit. On the Digitakt, hold FUNC while powering on, press TRIG 4 for OS UPGRADE, and send the stock file with Transfer's legacy OS upgrade mode. For the Digitone, use Elektron's OS-upgrade recovery for that unit.
On a track's SRC page, open the machine menu (FUNC + SRC) and pick OB-1.
A track switched to OB-1 starts with two sawtooths: VCO1's fundamental ON, VCO2's at −3 dB, subs, noise and modulation OFF. Both coarse tunings are centred; FIN1 is +0.0 cents and FIN2 is +2.4 cents. This small detune lets the oscillators move through phase cancellation instead of holding it. Set FIN2 to +0.0 for exact unison. Saved sounds keep their stored tuning.
The first SRC page puts both oscillators' performance controls together: VCO1 on the top row, VCO2 on the bottom row. Press SRC to switch to the mixer and routing page; press it again to return. Knobs A–D control the top row and E–H the bottom row. The mixer page leaves knob D unused.
| Page 1 — Oscillators | Page 2 — Mixer and routing |
|---|---|
 |
 |
Captured directly from a Digitakt mk1 running OB-1. The settings shown are examples, not the defaults.
| Knob | Page 1 — Oscillators | Page 2 — Mixer / routing |
|---|---|---|
| A | FRQ1 — VCO1 coarse tuning | FND1 — VCO1 fundamental level |
| B | FIN1 — VCO1 fine tuning | OCT1 — VCO1 sub-octave level |
| C | TYP1 — VCO1 saw / pulse family | XMOD — cross modulation OFF / ON |
| D | WAV1 — VCO1 waveform shape | Blank |
| E | FRQ2 — VCO2 coarse tuning | FND2 — VCO2 fundamental level |
| F | FIN2 — VCO2 fine tuning | OCT2 — VCO2 sub-octave level |
| G | TYP2 — VCO2 saw / pulse family | SYNC — hard sync OFF / ON |
| H | WAV2 — VCO2 waveform shape | NOIS — noise level |
Each oscillator tunes independently from the played note: coarse tuning spans −32 to +31 semitones, and fine tuning spans −50.0 to +50.0 cents. TYP1/TYP2 select SAW or PULS independently. WAV1/WAV2 span 0–63 within the selected family: triangle to saw for SAW, narrow pulse to square for PULS. Sweeping WAV never switches family. The waveform pictures follow both the family and shape, including the values shown while editing held-step locks. Fundamental, sub-octave and noise levels use vertical three-position switches: ON at the top, −3DB in the middle, OFF at the bottom.
The parameter mapping replaces earlier builds without migration. Sounds saved by this build use machine-pages' version-4 format and require compatible custom firmware to reload.
Some things to try:
- Sync sweep: SYNC ON, VCO2 alone, then sweep FRQ2 or put an LFO on it.
- Bells and gongs: XMOD ON with VCO2 a few octaves down. That's what the OB-1's manual suggests cross mod for.
- A fat unison: both fundamentals ON, FRQ1 and FRQ2 at +0, then detune FIN2 by a few cents and add the subs to taste.
The original OB-1's VCOs run continuously, even between notes and when their mixer outputs are OFF. Preserving that phase progression matters for faithful mixer and modulation behavior. Notes and mixer switches do not reset this model's oscillator phase: both VCOs and their sub-octaves keep running throughout an active voice, including muted sources. Phase continuity through every fully silent period is not yet guaranteed.
Computing both oscillators, their subs and selected sync/cross-modulation paths has a substantial CPU cost on top of the stock engine, even when their mixer outputs are OFF. Muting a source does not reclaim that work. Waveform and routing choices still affect the cost per voice.
We recommend including DigiHealth when composing your firmware in Elekloader, and leaving SETTINGS → FAST AUDIO ON. With two demanding OB-1 tracks on a Digitakt mk1 running OS 1.54, it reduced average render load from 89.1% to 80.2% and increased the observed peak margin from 4.2% to 14.6%. See the matched FAST AUDIO test.
See the two-instance device measurements and
matched CPU-cost comparison.
mise run bench counts ColdFire instructions for the DSP alone; it does not measure complete firmware
timing or establish a safe voice count. Leave headroom and check the intended sound on the device.
See benchmark interpretation.
The same oscillator section, this time as a Digitone mk1 / Keys machine (ob1-dn, version 0.6, machine id 69), for OS 1.44. Pick OB-1 as a track's machine and its SYN pages show the two VCOs, mixer and routing; the Digitone's filter, envelopes, effects, level and amp work downstream.
A track switched to OB-1 starts as two saws: VCO1's fundamental ON, VCO2 at −3 dB, subs, noise and modulation OFF — the Digitakt default. Both VCOs run freely between notes, as the OB-1's do. Turn an encoder to show its value in place of the label; turn popups carry it too. FUNC + turn resets a knob to OB-1's default, or to a held step's lock in GRID/STEP record.
SYN1 is the mixer and routing page; SYN2 is the oscillator page. Knobs A–D control the top row and E–H the bottom row. SYN1 leaves knob D unused.
| SYN1 — Mixer and routing | SYN2 — Oscillators |
|---|---|
 |
 |
Captured from digiemu.
| Knob | SYN1 — Mixer / routing | SYN2 — Oscillators |
|---|---|---|
| A | FND1 — VCO1 fundamental level | FRQ1 — VCO1 coarse tuning |
| B | OCT1 — VCO1 sub-octave level | FIN1 — VCO1 fine tuning |
| C | XMOD — cross modulation OFF / ON | TYP1 — VCO1 saw / pulse family |
| D | Blank | WAV1 — VCO1 waveform shape |
| E | FND2 — VCO2 fundamental level | FRQ2 — VCO2 coarse tuning |
| F | OCT2 — VCO2 sub-octave level | FIN2 — VCO2 fine tuning |
| G | SYNC — hard sync OFF / ON | TYP2 — VCO2 saw / pulse family |
| H | NOIS — noise level | WAV2 — VCO2 waveform shape |
The tuning ranges, waveform families and three-position levels match the Digitakt machine. SYN2 holds the oscillator page because SYN1's stock ALGO range (0–7) is too coarse for FRQ1's ±32 semitones. P-locks, LFOs, MIDI CCs and presets work through the borrowed stock SYN slots; turn popups and LFO destination lists keep FM names.
The sync, cross-mod and unison examples above work the same way.
One note per track, as on the OB-1 and the Digitakt machine. The newest held note sounds; releasing it returns to the newest note still held, as long as that note still has a stock voice. Each track keeps the same two VCOs across note changes, with a short crossfade, so the handover stays click-free and phase-continuous. The tracks keep independent note stacks; under normal allocation, notes on one track do not silence another track.
This differs from the Digitakt implementation. The Digitakt machine owns one oscillator voice per track directly. The Digitone adapter runs inside the companion's stock-allocated voices: it uses the stock note-on and release events to maintain the last-note stack, and silences every voice on the track except the one that should sound. The companion keeps ownership of the voice, the AMP envelope and velocity, so the normal AMP page and FM tracks are unchanged.
The adapter cannot control the Digitone's voice allocator. With the default dynamic allocation, an overlapping note gets another voice and the held-note return works. If LOCKED VOICES gives the track one voice, the overlapping note reuses that voice; the earlier note no longer has a voice, and releasing the new note can fall silent instead of returning. The same limit applies once the stock allocator has stolen a held note's voice. Leave the track's voices dynamic for normal last-note priority.
A proper one-voice policy belongs upstream. machines-dn1 would have to keep the track's held-note stack independently of the stock voices, decide the single sounding note before the allocator commits, and reserve, retrigger or redirect one voice on note-on and note-off. core-dn1 exposes ev_voice_on only after the allocator has chosen and possibly reused a voice, plus a per-block mask of released voices; it has no supported pre-allocation or voice-control hook. Retuning fw_voice_pitch is not enough because the earlier note's gate and AMP envelope have already moved to the reused voice. DNM_OWN_ENV would not fix allocation and would duplicate the companion's AMP envelope and velocity, risking the normal AMP-page behaviour.
That work can affect shared elekloader behaviour for every machine and for FM allocation, and needs its own design, tests and release across the supported OS versions. It is not part of this machine's release. The current adapter is therefore the reasonable choice for now: one audible note per track, last-note priority, phase continuity and the stock AMP path, with the locked-voice limitation above.
Computing both oscillators, their subs and selected sync/cross-modulation paths costs more than the stock engine, and muting a source does not reclaim it. The audible voice uses the same shared DSP as on the Digitakt. A stock voice the adapter has silenced is written as silence, but it still occupies a voice and the companion still runs its AMP bookkeeping; overlapping notes can use more than one voice.
Emulator fwcheck against stock and scripted interaction coverage remain pending; hardware testing covered the mono voice handling, both pages, turns, locks and the FUNC+turn reset. Re-pick the machine after flashing; the knob map changed during development, so sounds saved with earlier builds may point at a different control.
The model follows the OB-1's oscillator topology, with saw-family shape/level, cross-modulation depth, mixer steps and white-noise output calibrated against the reference. Both mixer middle positions are approximately −3 dB. The minimum pulse width remains provisional at 6.7 %; drift and other analog variation are omitted. Agreement with the reference is distinct from validation against a physical OB-1, for which no controlled capture set is included.
Sources and confidence separates printed facts, calculations, measurements and model choices. The DSP reference explains the sound and its approximations.
- Architecture: signal flow, device boundaries, parameter encoding and saved sounds.
- DSP: numeric formats, waveforms, modulation, mixing and approximation limits.
- Development: setup, commands, verification, benchmarks and releases.
- Digitone machine: install, controls and limits.
- Sources and confidence: manuals, schematics, derivations and unresolved questions.
For host-only work, start with mise run test; stock firmware and the cross-toolchain are not needed.
- x2k77m1 added Digitone support.
GPL-2.0-or-later.