SH-5 family: the SH-7 and the SH-3A

bcox commented

2026-10-12 21:15:34 +00:00

Owner

Ben asked on 2026-10-12 which instruments of the families already here were missed, and for an issue a family. Everything below is from recollection, with no manual read; issue #20's list was wrong in several places for the same reason, so read the manuals first and correct this. Each is to be an extension of the existing instrument, as the Jupiters and the JX family are, never a "model" switch.

The SH-5 is the one instrument here with no family.

SH-7 (1978): recalled as its duophonic successor; engine/duophonic.h exists now.
SH-3A (1974): recalled as its predecessor, which mixes footages of one VCO.

The SH-7 follows the Solina in Ben's order; the SH-3A comes in the later pass.

Ben asked on 2026-10-12 which instruments of the families already here were missed, and for an issue a family. **Everything below is from recollection, with no manual read**; issue #20's list was wrong in several places for the same reason, so read the manuals first and correct this. Each is to be an extension of the existing instrument, as the Jupiters and the JX family are, never a "model" switch. The SH-5 is the one instrument here with no family. - **SH-7** (1978): recalled as its duophonic successor; `engine/duophonic.h` exists now. - **SH-3A** (1974): recalled as its predecessor, which mixes footages of one VCO. The SH-7 follows the Solina in Ben's order; the SH-3A comes in the later pass.

bcox referenced this issue

2026-10-12 21:26:53 +00:00

A new instrument: Roland's SH-1000 and SH-2000 #43

bcox commented

2026-10-12 22:26:26 +00:00

Author

Owner

The SH-7 half of this is done on the family-extensions branch, as an extension of mini-5. The SH-3A is untouched.

What was read. Roland's owner's manual for the SH-7 (from Roland's archive, undated) and its service notes of 1 July 1978 (specifications and circuit description; the schematics, block diagram and adjustment procedure were not read). Both are in docs/references.md.

Against the recollection. It does play two notes, so "duophonic" holds. It is not an SH-5 with a second note, though: it is a different panel on the same plan.

Less than an SH-5: one filter, a low-pass, with a hand-set high-pass ahead of it; no band-pass filter and no mixer routes; one LFO where the SH-5 has two; one sync; two ADSRs and no fixed envelope; no panning.
More than an SH-5: VCO-1 has a second output, (A), five octave squares from its divider with a slider each; AUTOBEND; a slider each for LFO, autobend and S/H on each VCO; VCO-2 or noise to the cutoff; an envelope follower on the external signal; a bender with a knob and a CV/OFF/LFO switch each for VCO, VCF and VCA; portamento for rising or falling notes only; KYBD TRIG on the LFO.
The keyboard: TWO NOTE puts the higher key on VCO-1 and the lower on VCO-2; ONE NOTE is highest-key priority; a third position hands over to the CV/gate jacks or the S/H clock. A trigger comes when the highest key is pressed or released.

What was built. Twenty parameters, each at nothing until asked for, no model switch; every existing patch and test renders as before.

osc1.feet32 ... osc1.feet2 (new engine/octave_squares.h)
osc1.autobend, osc2.autobend, autobend.time
osc1.sh, osc2.sh
osc1.pwm_source 4 (ADSR), osc2.pwm_source 2 (LFO-2 triangle)
hpf.cutoff
filter.fm, filter.fm_source
ar.decay, ar.sustain_level (the AR becomes the second ADSR)
lfo2.key_trigger
voice.key_mode (0 latest, 1 highest, 2 two notes), on engine/duophonic.h unchanged
voice.portamento_direction
bend.amp, stick.amp

engine/duophonic_glide.h is not used: the service notes send both keyboard voltages to the portamento circuit, so each oscillator glides on its own.

Not built. The external signal and envelope follower (#17), the pedal, EXT CV/GATE (S/H), a trigger switch per envelope, and the squares locked in phase to VCO-1(B).

For Ben. This is the widest of the family extensions; a separate instrument was the other way to go. The choices made without asking are under "Open decisions" in CLAUDE.md, "The SH-7 in the SH-5". The ones most worth a look: extension or instrument; that the SH-5 still plays the latest key where the SH-7 plays the highest; which sync an SH-7 has; and the five squares simply adding, where the manual says the level stays the same.

The SH-5's bench, unchanged, on the Pi before and after, twice, beside another session's miniomnid: 2.3%, 2.8%, 3.6%, 4.2% before and 2.3%, 2.8%, 3.7%, 4.2% after; one to two parts in a hundred slower, repeatably. Not played or heard.

The SH-7 half of this is done on the `family-extensions` branch, as an extension of mini-5. The SH-3A is untouched. **What was read.** Roland's owner's manual for the SH-7 (from Roland's archive, undated) and its service notes of 1 July 1978 (specifications and circuit description; **the schematics, block diagram and adjustment procedure were not read**). Both are in `docs/references.md`. **Against the recollection.** It does play two notes, so "duophonic" holds. It is not an SH-5 with a second note, though: it is a different panel on the same plan. - Less than an SH-5: one filter, a low-pass, with a hand-set high-pass ahead of it; no band-pass filter and no mixer routes; one LFO where the SH-5 has two; one sync; two ADSRs and no fixed envelope; no panning. - More than an SH-5: VCO-1 has a second output, (A), five octave squares from its divider with a slider each; AUTOBEND; a slider each for LFO, autobend and S/H on each VCO; VCO-2 or noise to the cutoff; an envelope follower on the external signal; a bender with a knob and a CV/OFF/LFO switch each for VCO, VCF and VCA; portamento for rising or falling notes only; KYBD TRIG on the LFO. - The keyboard: TWO NOTE puts the higher key on VCO-1 and the lower on VCO-2; ONE NOTE is highest-key priority; a third position hands over to the CV/gate jacks or the S/H clock. A trigger comes when the highest key is pressed or released. **What was built.** Twenty parameters, each at nothing until asked for, no model switch; every existing patch and test renders as before. - `osc1.feet32` ... `osc1.feet2` (new `engine/octave_squares.h`) - `osc1.autobend`, `osc2.autobend`, `autobend.time` - `osc1.sh`, `osc2.sh` - `osc1.pwm_source` 4 (ADSR), `osc2.pwm_source` 2 (LFO-2 triangle) - `hpf.cutoff` - `filter.fm`, `filter.fm_source` - `ar.decay`, `ar.sustain_level` (the AR becomes the second ADSR) - `lfo2.key_trigger` - `voice.key_mode` (0 latest, 1 highest, 2 two notes), on `engine/duophonic.h` unchanged - `voice.portamento_direction` - `bend.amp`, `stick.amp` `engine/duophonic_glide.h` is not used: the service notes send both keyboard voltages to the portamento circuit, so each oscillator glides on its own. **Not built.** The external signal and envelope follower (#17), the pedal, EXT CV/GATE (S/H), a trigger switch per envelope, and the squares locked in phase to VCO-1(B). **For Ben.** This is the widest of the family extensions; a separate instrument was the other way to go. The choices made without asking are under "Open decisions" in CLAUDE.md, "The SH-7 in the SH-5". The ones most worth a look: extension or instrument; that the SH-5 still plays the latest key where the SH-7 plays the highest; which sync an SH-7 has; and the five squares simply adding, where the manual says the level stays the same. The SH-5's bench, unchanged, on the Pi before and after, twice, beside another session's `miniomnid`: 2.3%, 2.8%, 3.6%, 4.2% before and 2.3%, 2.8%, 3.7%, 4.2% after; one to two parts in a hundred slower, repeatably. Not played or heard.

bcox commented

2026-10-12 23:33:43 +00:00

Author

Owner

The second half, the SH-3 and SH-3A, is in mini-5 on the family-extensions branch (not pushed). Written by Claude with nobody to ask; the choices are under "Open decisions" in CLAUDE.md.

What was read. Roland's owner's manual for the SH-3 (from Roland's own archive), the SH-3's service notes (first edition) and the SH-3A's (fifth edition, April 1976). No owner's manual for the SH-3A was found. Neither VCF/VCA schematic nor the VCO board's was traced.

Against the recollection ("the SH-5's predecessor, which mixes footages of one VCO"): borne out, with more.

The two are one voice. Their service notes print the same specification line for line; the SH-3A's lists what changed as the VCF and VCA board, some potentiometers and the adjustment.
The VCO is heard only through its divider: five footages, 32' to 2', each with a slider. That is the SH-7's VCO-1(A), already there.
Not recalled: each footage has a switch of three waveforms (a square, an unsymmetrical rectangle, and a "pulse", which by the circuit is a spike); the 8' has a CHORUS dial; the lowest key held sounds; and there is one ADSR, the filter and the amplifier each choosing between it and two or three fixed shapes (slow attack, a staccato/"wow", on-off).
The rest is the SH-5's plan with less: one low-pass filter, LFO-1's sawtooth and LFO-2's rectangle and sine with a delay, a sampler to the VCO, noise to the VCF or VCA, HOLD.

What was built, each at nothing until asked for, so every existing patch and test is as it was:

osc1.feet32_wave … osc1.feet2_wave: 0 square, 1 rectangle, 2 pulse. Widths and time constants are a reading of the filter board's diagram (an eighth for 32' and 16', a quarter for 4'; 8' and 2' guessed at a quarter).
osc1.chorus: the 8' square's width going back and forth between 50% and 80%, from the one figure in the adjustment procedure, in hertz.
voice.key_mode 3: the lowest key.
No new envelope: the fixed shapes are the fixed envelope and the AR, set.

engine/octave_squares.h gained the widths and spikes. The SH-7's autobend now uses engine/auto_bend.h; they were the same arithmetic.

Not done: the GLIDE button and its pedal, the pedal jacks, the sampler's own MODE switch (its symbols were not made out), and whatever the SH-3A's new board changed in the filter. Not heard; the bench was not run, nothing having been added to what it measures.

The second half, the SH-3 and SH-3A, is in mini-5 on the `family-extensions` branch (not pushed). Written by Claude with nobody to ask; the choices are under "Open decisions" in CLAUDE.md. **What was read.** Roland's owner's manual for the SH-3 (from Roland's own archive), the SH-3's service notes (first edition) and the SH-3A's (fifth edition, April 1976). **No owner's manual for the SH-3A was found.** Neither VCF/VCA schematic nor the VCO board's was traced. **Against the recollection** ("the SH-5's predecessor, which mixes footages of one VCO"): borne out, with more. - The two are one voice. Their service notes print the same specification line for line; the SH-3A's lists what changed as the VCF and VCA board, some potentiometers and the adjustment. - The VCO is heard only through its divider: five footages, 32' to 2', each with a slider. That is the SH-7's VCO-1(A), already there. - Not recalled: each footage has a switch of three waveforms (a square, an unsymmetrical rectangle, and a "pulse", which by the circuit is a spike); the 8' has a CHORUS dial; the lowest key held sounds; and there is one ADSR, the filter and the amplifier each choosing between it and two or three fixed shapes (slow attack, a staccato/"wow", on-off). - The rest is the SH-5's plan with less: one low-pass filter, LFO-1's sawtooth and LFO-2's rectangle and sine with a delay, a sampler to the VCO, noise to the VCF or VCA, HOLD. **What was built**, each at nothing until asked for, so every existing patch and test is as it was: - `osc1.feet32_wave` … `osc1.feet2_wave`: 0 square, 1 rectangle, 2 pulse. Widths and time constants are a reading of the filter board's diagram (an eighth for 32' and 16', a quarter for 4'; 8' and 2' guessed at a quarter). - `osc1.chorus`: the 8' square's width going back and forth between 50% and 80%, from the one figure in the adjustment procedure, in hertz. - `voice.key_mode` 3: the lowest key. - No new envelope: the fixed shapes are the fixed envelope and the AR, set. `engine/octave_squares.h` gained the widths and spikes. The SH-7's autobend now uses `engine/auto_bend.h`; they were the same arithmetic. **Not done:** the GLIDE button and its pedal, the pedal jacks, the sampler's own MODE switch (its symbols were not made out), and whatever the SH-3A's new board changed in the filter. Not heard; the bench was not run, nothing having been added to what it measures.

bcox commented

2026-10-13 00:48:16 +00:00

Author

Owner

Ben answered on 2026-10-12 whether the SH-7 should be an instrument of its own: "extensions, as long as they fit in the same engine and run on the pi." It stays in mini-5. Benched on the Pi that day with everything of the SH-7 and the SH-3A at work and two keys held: 4.4% of one core, 5.9% with the effects.

bcox commented

2026-10-13 01:27:31 +00:00

Author

Owner

Closed by Ben's word on 2026-10-12. The work is on the branch family-extensions, not yet merged.

Closed by Ben's word on 2026-10-12. The work is on the branch `family-extensions`, not yet merged.

bcox closed this issue

2026-10-13 01:27:31 +00:00

bcox referenced this issue from a commit

2026-10-13 01:34:35 +00:00

Add the SH-7 to the SH-5: its two notes, octave squares, autobend and the rest

bcox referenced this issue from a commit

2026-10-13 01:34:35 +00:00

Add the SH-3 and SH-3A to the SH-5: a waveform for each footage, the 8' chorus and low note priority

bcox commented

2026-10-13 04:27:04 +00:00

Author

Owner

On the night of 2026-10-12, at Ben's asking, and then played by him on his laptop, the SH-7 and the SH-3's waveforms and 8' chorus alike, which he found nice:

A GATE+TRIG switch for each envelope, adsr.retrigger and ar.retrigger, as the SH-7 has. voice.retrigger is still both at once, so a stored patch sounds as it did.
The five footage squares share one level, as the SH-7's manual says, where they added: past a total of one, each is divided by what the five levels come to. No document gives the arithmetic, so that is a plain choice. The SH-3's footages go the same way. A stored patch whose levels come to more than one is quieter.
Settled: which key sounds on the SH-5 is the shared voice.key_priority, starting at the latest since no SH-5 document says; and the SH-3's fixed envelope shapes are the SH-5's envelopes with times the patch sets, which the hardware's are not.

Everything said here is on main, pushed, on the night of 2026-10-12. What was heard was heard on Ben's laptop, not on the Pi.

On the night of 2026-10-12, at Ben's asking, and then played by him on his laptop, the SH-7 and the SH-3's waveforms and 8' chorus alike, which he found nice: - **A GATE+TRIG switch for each envelope**, `adsr.retrigger` and `ar.retrigger`, as the SH-7 has. `voice.retrigger` is still both at once, so a stored patch sounds as it did. - **The five footage squares share one level**, as the SH-7's manual says, where they added: past a total of one, each is divided by what the five levels come to. No document gives the arithmetic, so that is a plain choice. The SH-3's footages go the same way. A stored patch whose levels come to more than one is quieter. - **Settled**: which key sounds on the SH-5 is the shared `voice.key_priority`, starting at the latest since no SH-5 document says; and the SH-3's fixed envelope shapes are the SH-5's envelopes with times the patch sets, which the hardware's are not. Everything said here is on `main`, pushed, on the night of 2026-10-12. What was heard was heard on Ben's laptop, not on the Pi.

Rows
Columns

SH-5 family: the SH-7 and the SH-3A #35