peter-swimm/nardwahl
Watch
1
0
Fork
You've already forked nardwahl
 0
Code Issues 4 Pull requests Projects Releases Packages Wiki Activity Actions

Calibrate OscA pitch depth, G1 pre-roll phases and control levels (v25) #12

Merged
peterswimm merged 2 commits from calibration-v25 into main 2026-10-22 02:25:19 +00:00
Conversation 0 Commits 2 Files changed 4 +109 -112
peterswimm commented 2026-10-22 02:23:58 +00:00
Owner
Copy link

Scope

Calibration round v25 of the shared G1 → Faust model laws, fit from G1-Emu measurements. Both the Faust export and the Python renderer change together.

  • OscA/OscB Type II pitch inputs. Each pitch-mod knob now has a measured full-scale step count, using G1's 1/128-semitone grid; for example, knob 64 is 560.26 steps and knob 127 is 8191.5. The summed inputs are floored once. Pitch modulation takes effect at capture start; during the adapter pre-roll the oscillator free-runs at its static pitch.
  • Control levels. Unipolar 7-bit levels send v/128, and 127 sends exactly 1.0. This covers Constant, ADSR/Mod-Env sustain, Multi-Env and NoteVelScal. Bipolar Constant sends (v−64)/64.
  • G1 adapter pre-roll (0.31 s). The Faust runner now renders the pre-roll and throws it away, so free-running sources keep patch time. Capture-calibrated pieces (gate, LFOA) use capture_time.
    • ClkGen beat phase.
    • LFOB/LFOC start phase for each voice mode.
    • LFOC waveform order, and rates equal to LFOB's in ranges 1–2.
    • OscA sine/triangle start phases.
    • Digitizer clock phase, refit to 0.60 measured from patch start.
  • Regenerated docs/G1_PARITY_MATRIX.{json,md} from the v25 sweep.

Not addressed this round: PercOsc (95) pitch law, the OscA FMA input, Phaser (102) notches, the Expander (105) gate parameter, and the end level of Multi-Env segments, which sits 2 steps low. docs/G1_BASELINE_STATUS.json is unchanged because regenerating it needs a full four-shard baseline run.

Testing

  • python3 -m unittest discover -s tests passes all 129 tests. New tests cover the pitch-step floor, the depth table, the control fraction, ClkGen on patch time, LFOB/LFOC phases and waveform order, OscA start phase and the Digitizer clock.
  • Cut plan v25 (module-baseline-plan --control-observer-ready --physical-output-source-ready --lfoa-modulation-ready) and swept all 1,489 probes. No probe that passed in v24 fails in v25.
  • tools/export_public.py verifies a clean public tree.
Verdict v24 v25
pass 842 922
pitch 298 245
level 192 194
shape-or-phase 100 74
both-silent 36 36
silence-mismatch 14 14
peak-only 7 4
Stage v24 v25
sweep-clean 17 21
near 22 23
partial 23 22
divergent 44 40
not-swept 3 3

Impact Statement

The Faust export and the Python renderer move closer to G1 on oscillator pitch modulation and on the startup phase of free-running sources. Faust renders get 0.31 s longer, because the pre-roll is computed and then discarded. This PR adds no private data (plans, fixtures, captures or results stay under ~/.nardwahl). Sweep verdicts are for triage only, so no module is promoted to reference-compared.

## Scope Calibration round v25 of the shared G1 → Faust model laws, fit from G1-Emu measurements. Both the Faust export and the Python renderer change together. - **OscA/OscB Type II pitch inputs.** Each pitch-mod knob now has a measured full-scale step count, using G1's 1/128-semitone grid; for example, knob 64 is 560.26 steps and knob 127 is 8191.5. The summed inputs are floored once. Pitch modulation takes effect at capture start; during the adapter pre-roll the oscillator free-runs at its static pitch. - **Control levels.** Unipolar 7-bit levels send v/128, and 127 sends exactly 1.0. This covers Constant, ADSR/Mod-Env sustain, Multi-Env and NoteVelScal. Bipolar Constant sends (v−64)/64. - **G1 adapter pre-roll (0.31 s).** The Faust runner now renders the pre-roll and throws it away, so free-running sources keep patch time. Capture-calibrated pieces (gate, LFOA) use `capture_time`. - ClkGen beat phase. - LFOB/LFOC start phase for each voice mode. - LFOC waveform order, and rates equal to LFOB's in ranges 1–2. - OscA sine/triangle start phases. - Digitizer clock phase, refit to 0.60 measured from patch start. - Regenerated `docs/G1_PARITY_MATRIX.{json,md}` from the v25 sweep. Not addressed this round: PercOsc (95) pitch law, the OscA FMA input, Phaser (102) notches, the Expander (105) gate parameter, and the end level of Multi-Env segments, which sits 2 steps low. `docs/G1_BASELINE_STATUS.json` is unchanged because regenerating it needs a full four-shard baseline run. ## Testing - `python3 -m unittest discover -s tests` passes all 129 tests. New tests cover the pitch-step floor, the depth table, the control fraction, ClkGen on patch time, LFOB/LFOC phases and waveform order, OscA start phase and the Digitizer clock. - Cut plan v25 (`module-baseline-plan --control-observer-ready --physical-output-source-ready --lfoa-modulation-ready`) and swept all 1,489 probes. No probe that passed in v24 fails in v25. - `tools/export_public.py` verifies a clean public tree. | Verdict | v24 | v25 | |---|---:|---:| | pass | 842 | 922 | | pitch | 298 | 245 | | level | 192 | 194 | | shape-or-phase | 100 | 74 | | both-silent | 36 | 36 | | silence-mismatch | 14 | 14 | | peak-only | 7 | 4 | | Stage | v24 | v25 | |---|---:|---:| | sweep-clean | 17 | 21 | | near | 22 | 23 | | partial | 23 | 22 | | divergent | 44 | 40 | | not-swept | 3 | 3 | ## Impact Statement The Faust export and the Python renderer move closer to G1 on oscillator pitch modulation and on the startup phase of free-running sources. Faust renders get 0.31 s longer, because the pre-roll is computed and then discarded. This PR adds no private data (plans, fixtures, captures or results stay under `~/.nardwahl`). Sweep verdicts are for triage only, so no module is promoted to reference-compared.
The Faust runner now renders the G1 adapter pre-roll, so the free-running
Digitizer clock accumulates from patch start. Its phase is refit there
against the default-rate G1 probes (0.60), and the Python renderer starts
the clock where the pre-roll leaves it.
Regenerate the parity matrix from plan v25
All checks were successful
PR policy check / policy-gate (pull_request) Successful in 0s
Details
2ef5aa0864 
Probe sweep v25: 922 of 1,489 probes pass (842 in v24) with no v24 pass
regressing; sweep-clean 21, near 23, partial 22, divergent 40.
peterswimm deleted branch calibration-v25 2026-10-22 02:25:19 +00:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
peter-swimm/nardwahl!12
No description provided.