@synthlet/analog-delay

A glide-based tape and bucket-brigade echo with multi-tap heads and one wear control

Part of Synthlet

Moving the delay time on a tape or bucket-brigade machine bends the pitch of everything already in the line. The Space Echo's Repeat Rate knob is a performance control precisely because of that, and the Echoplex's sliding head is the same effect made mechanical. The read head here travels to its new position rather than handing over to a second one, so it resamples the buffer on the way.

There is a second difference that matters more for the model's honesty. In a clean digital delay, time and tone are independent. In real analog hardware they are not. A bucket-brigade's delay is its stage count over its clock rate, so lengthening the delay lowers the clock, which drags the anti-alias and reconstruction filters down with it, which darkens every repeat. Tape does the same thing through head gap and transport speed. That coupling is the sound of an analog delay — the reason a Memory Man at maximum time is murky and at minimum time is nearly clean. So there is no tone parameter: bandwidth is derived from time and age.

The other half — precise, clean, pitch-preserving, tempo-relative — is @synthlet/digital-delay. One question tells you which you want: character and drift, or precision?

Install

npm install @synthlet/analog-delay

Usage

import {
  AnalogDelay,
  AnalogDelayMode,
  registerAnalogDelayWorklet,
} from "@synthlet/analog-delay";

await registerAnalogDelayWorklet(audioContext);

const delay = AnalogDelay(audioContext, {
  time: 0.3, // seconds
  feedback: 0.6,
  mix: 0.4,
  taps: 0.7, // more heads open
  age: 0.4, // wobblier, dirtier, darker, hissier
  mode: AnalogDelayMode.Tape,
});

source.connect(delay).connect(audioContext.destination);

// Drag this while it repeats. That is the whole point.
delay.time.value = 0.18;

Parameters

| Param | Default | Min | Max | Rate | Meaning | | ---------- | ------- | ---- | --- | ------ | -------------------------------------------------------------------------------------------- | | time | 0.3 | 0.02 | 1.5 | k-rate | Record head to first playback head, in seconds. Moving it bends pitch | | feedback | 0.4 | 0 | 1.2 | k-rate | The Space Echo's Intensity. Above 1 self-oscillates, bounded by the loop's soft limiter | | mix | 0.3 | 0 | 1 | k-rate | Dry/wet | | taps | 0 | 0 | 1 | k-rate | Level envelope across the mode's tap positions. 0 is the first head only, 1 is all of them | | age | 0.3 | 0 | 1 | k-rate | One composite wear control: more wobble, more saturation, less bandwidth, more hiss | | wobble | 0.3 | 0 | 1 | k-rate | Wow and flutter depth, scaling on top of whatever age implies | | spread | 0 | 0 | 1 | k-rate | L/R time offset as a ratio — two machines, lightly apart. 0 is mono-compatible | | mode | 0 | 0 | 1 | k-rate | AnalogDelayMode: 0 Tape, 1 BBD. Intermediate values wipe between them rather than snapping |

One construction option, maxTime (default 1.5 s), sizes the buffers. It is not an AudioParam because it allocates.

Four things the table cannot carry:

• taps is continuous, and it means something in both modes. Tape taps are integer multiples of time, so opening more heads gives rhythmic subdivisions. The MN3011's six taps are deliberately irrational, so opening more of them gives a diffuse wash — which is what its datasheet says they are for. That difference in kind is what earns mode its slot; it is not a filter preset.
• age and wobble are both here on purpose. A well-maintained machine with an eccentric capstan is a real and desirable combination, and one composite knob cannot express it.
• The tap offsets are clamped to the line. At time = maxTime in BBD mode the 8.4x tap would need a 12.6 s stereo buffer, so the later taps fold onto the line's maximum instead of allocating for a case nobody asks for.
• feedback at 1.2 self-oscillates, but not at every setting. With age, wobble and taps all well up, a worn multi-head transport cannot hold the resonance and the loop decays. It stays finite and bounded either way.

Sourced numbers

Unusually for this repository, the provenance here is datasheets and service manuals rather than papers, and some of it is not sourced at all. Hiding that would be worse than showing it.

| Figure | Value | Confidence | | --------------------- | ----------------------------------------------------------- | ------------------------------------------------------------------- | | BBD delay formula | delay = stages / (2 x clock) | High — verified against four Panasonic datasheets | | MN3005 | 4096 stages, 10–100 kHz → 20.48–204.8 ms | High — primary datasheet | | MN3011 taps | 3328 stages, taps at 396 / 662 / 1194 / 1726 / 2790 / 3328 | High — primary datasheet, tap table read directly | | MN3011 intent | "natural reverberation effect ... by mixing output signals" | High — primary | | BBD reference filters | 20 kHz | High — primary, but Panasonic's generic test jig | | RE-201 head ratios | 1 : 2 : 3 | Medium-high — Roland's RE-202 manual states 2x/3x/4x for four heads | | Echorec drum | 71 RPM → the 1.2 Hz wow rate | Medium-high — two independent figures that cross-check |

Chosen by ear, and labelled as chosen in src/dsp.ts: the wow and flutter depths, the flutter rate, the tape gap-loss constant, the age curve, and the compander's reference level and time constants. No sourced wow/flutter figure exists for the RE-201, EP-3 or Echorec, and studio-deck standards (DIN 0.2%) are explicitly not applicable — tape echoes are widely described as far worse, and nobody quantifies it. Rather than invent plausible numbers, one figure is derived from the Echorec's drum speed and the rest are marked.

Measured quality

Each of these is an assertion in src/dsp.test.ts, with the measured value in a comment beside its threshold.

| What | Measured | | ----------------------------- | ---------------------------------------------------------------------------------------- | | Pitch through a time sweep | 440 → 511 → 440 Hz, against 440 → 442 → 440 for digital-delay on the same sweep | | Bandwidth versus time, Tape | corner 14225 → 1774 Hz over five settings, within 15% of the derived shape | | Bandwidth versus time, BBD | corner 11898 → 2290 Hz, likewise | | age, four things at once | wobble 0.67 → 2.31 Hz s.d., THD 3.2e-4 → 7.5e-2, corner 15249 → 3800 Hz, hiss 0 → 5.0e-5 | | Tap ratios, Tape | 0.99 : 2.00 : 3.00 | | Tap ratios, BBD | 0.99 : 1.67 : 3.01 : 4.35 : 7.04 : 8.39 | | Rhythmic versus diffuse | envelope autocorrelation at the tap period 0.778 Tape, 0.024 BBD | | Compander round trip | 1.018 on steady material; 3.1x transient overshoot against Tape's 1.07 | | Mono compatibility | L and R are bit-identical at spread = 0 | | feedback = 1.2 over 60 s | peak 0.36 Tape / 0.75 BBD, still oscillating, no NaN and no denormal stall |

Attribution

Original, written from published descriptions and datasheets. No third-party source was copied.

| What | Citation | | ------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------- | | Glide-based delay modulation | Zavalishin & Parker, Tape-like Delay Modulation, DAFx-18; the one-pole coefficient is Mutable Instruments' LoopingSamplePlayer | | The BBD chain and its compander | Raffel & Smith, Practical Modeling of Bucket-Brigade Device Circuits, DAFx-10 | | Variable-rate BBD formulation | Holters & Parker, A Combined Model for a Bucket Brigade Device and its Input and Output Filters, DAFx-18 | | Hermite fractional reads | Niemitalo, Polynomial Interpolators for High-Quality Resampling of Oversampled Audio, 2001 | | Feedback high-pass and limiter | Mutable Instruments Clouds — fc = 20 + 100·feedback², and the cubic x(27 + x²)/(27 + 9x²) | | Delay-line structure | Mutable Instruments stmlib::DelayLine |

See the repository's THIRD-PARTY-LICENSES.md.

License

MIT © danigb