Skip to content

Friction Exciters

github-actions[bot] edited this page Nov 19, 2026 · 1 revision

Friction Exciters

The Friction family is everything that rubs, scrapes or pulls: Bow, Stiff Point Scrape, Heavy Grinding, Corrugated Drag and Tension Rise. They are continuous: they push for as long as their force envelope, the six-stage MSEG, is above zero, and the key being held is what keeps the MSEG up.

What they share with the other families (dispatch, drive levels, the governor, Speed and Roughness) is on Exciters. The MSEG itself, its loop and the Gesture Modifiers that only it obeys are on Envelopes and Modulation.

Conventions on this page

  • Steps and decays are per 48 kHz sample and converted with dsp::utils::rate at other rates (Sample-Rate Rule).
  • v is velocity / 127. "Host range" is the parameter's range in the plugin; set_parameters clamps to a wider code range.
  • The resonator's velocity V is in displacement per 48 kHz sample. On a held note it is tiny: 0.0002–0.007 on the default patch, against tool speeds near 1. That single fact explains most of this page: the friction laws almost never leave their sliding branch, so these models push the object one way with a nearly constant force plus a texture.
Model Reads Force, in one line
Bow V, previous sample `2 · P · μ(
Stiff Point Scrape V, previous sample 2 · ring + 0.3 · pressure · deflection / threshold
Heavy Grinding ContactPrediction (implicit) P · 0.4 · 3s · tanh((s − V(F)) / 0.05s) + tearing
Corrugated Drag V, previous sample (0.15 + bump) · sign(v_rel) · 3 · speed
Tension Rise V, previous sample one-sample break spikes, nothing in between

Bow, Stiff Point Scrape, Corrugated Drag and Tension Rise read the resonator explicitly (last sample's velocity): their laws aren't monotone in the force, which the implicit solver needs, and none of them runs away. Heavy Grinding's tanh law is monotone, so it is solved implicitly (Interaction Bus and Contact).

Bow

A bow drawn across the object with a Stribeck friction curve: static friction below a small slip speed, falling smoothly to dynamic friction above it, plus a periodic ripple standing in for the rosin's texture.

Control Host range, default Role
Bow Pressure 0.2–2, 1.1 normal force P
Bow Speed 0.1–2, 1.05 target bow speed (× Speed)
Rosin Grip 0.05–1.5, 0.485 the friction coefficients and the ripple depth (× Roughness)
Slip Curve 0.05–1.5, 0.485 width of the Stribeck transition and how far grip carries into sliding

At note-on (trigger): target speed speed · (0.5 + 0.5·v); the bow starts at a tenth of it and approaches it with a one-pole smoother (0.001 per sample, about 21 ms at 48 kHz).

Per sample (process_sample), with v_rel = bow_velocity − V:

  • μs = 0.4 + 0.3·grip, μd = 0.2 + 0.1·grip, Stribeck speed vs = 0.01 + 0.02 · speed · slip.
  • Below vs: μ falls from μs to μd along a smoothstep. Above it: μ = μd + (μs − μd) · exp(−(|v_rel| − vs) · falloff), with falloff = 10 · (0.485 / slip)² (exactly 10 at the default Slip Curve).
  • The bow position advances by the bow speed per sample and wraps at 1; the ripple is 1 + 0.3 · grip · sin(2π · position).
  • Force = P · μ · sign(v_rel) · ripple · 2.

What it really does. v_rel is about 0.94 at velocity 100 and V hardly moves it, so the bow stays on the sliding branch and never sticks. At the default Slip Curve exp(−9.2) makes μ ≈ μd (0.2485); at Slip Curve 1.5 the falloff drops to about 1 and μ rises to about 0.36. So Slip Curve acts as a force level, and the sound is a steady push, shaped by the MSEG, with the ripple on top. A bow that really sticks and slips is an open item in docs/BACKLOG.md: seeing the resonator ×1000 makes it stick, but drives the bowed presets into a raucous regime.

Trap: the ripple aliases, differently at each rate. The ripple's phase advances by the bow speed per 48 kHz sample (rate::step), so its true frequency is bow_speed · 48 000 Hz, about 45 kHz on the default patch. Sampled at 48 kHz it folds to (1 − 0.94) · 48 000 ≈ 3 kHz; at 44.1 kHz it lands near 0.9 kHz; at 96 kHz it stays at 45 kHz, inaudible. Measured on the struct alone (default controls, velocity 100): 942 Hz at 44.1 kHz, 2 959 Hz at 48 kHz, 45 038 Hz at 96 kHz. The ripple is 14.5 % deep at the default Rosin Grip, so the bow's texture depends on the host rate; every_exciter_has_the_same_level_at_any_sample_rate checks level only and doesn't see it. Not documented in the code as intended.

bow_controls_change_output (in src/voice/mod.rs) and tests/bow_exciter.rs cover the basics.

Stiff Point Scrape

A nail or awl dragged over metal: the stiff tip catches, winds up like a spring, slips past a threshold and rings, over and over. The slips set a chatter rate; the ring sets its pitch.

Control Host range, default Role
Scrape Speed 0.1–2.5, 0.82 how fast the tip winds up (× Speed)
Point Pressure 0.1–1.5, 0.8 slip threshold 0.05 + 0.2·pressure, and the strike and stick force
Chatter Pitch 0.1–1.5, 0.52 ring frequency 100 + 900·pitch Hz (190–1 450 Hz in host range; capped at 0.45·fs)
Chatter Damping 0.1–0.9, 0.66 ring decay time 0.006 / damping s (60 ms to 7 ms)

At note-on: scrape speed × (0.5 + 0.5·v); ring and deflection cleared.

Per sample:

  1. Deflection grows by (speed − V) · 0.001.
  2. Past the threshold, the tip slips: a strike of sign · pressure hits the ring, and the deflection drops to a tenth. early, the fraction of the step since the threshold was crossed, places the strike between samples: the ring adds the strike's value now and its virtual value one sample back, so the two-pole recursion carries it on at the right phase.
  3. The ring is a two-pole resonator r = exp(−1 / (decay · fs)), at the chatter pitch.
  4. Force = CHATTER_RING_GAIN · ring + STICK_GAIN · pressure · deflection / threshold, with gains 2 and 0.3.

On the default patch at velocity 100 a slip comes every ~258 samples, about 186 Hz, under a 568 Hz ring that decays in about 9 ms. The stick term is a sawtooth that is always positive, which is what deflects the object and makes the release snap. chatter_pitch_raises_the_chatter and chatter_damping_shortens_the_ring (in the module) check the two ring controls.

Heavy Grinding

Coulomb friction against a rough surface, plus a tearing noise from asperities breaking off: concrete, sandpaper, a grinder.

Control Host range, default Role
Grind Speed 0.1–2.5, 0.82 grinder speed s (× Speed)
Grind Pressure 0.1–1.9, 1 friction force, and tearing × √(P / 1)
Surface Grit 0–1, 0.3 tearing level (× Roughness)
Grit Color 0–1, 0.3 tearing noise colour

At note-on: s = grind_speed · (0.3 + 0.7·v); noise state cleared.

Per sample:

  • Tearing noise. A leaky random walk (Brownian noise): step ±0.1·√ratio, leak 0.99 per sample, clamped to ±1, so its level is the same at any rate. It goes through a one-pole low-pass whose coefficient is clamp(cutoff / 10 000, 0.01, 0.99) at 48 kHz, with cutoff = 100 + 1900·c below Grit Color 0.5 and 1000 + 6000·(c − 0.5) above. The code comments call the upper half "HPF-ish", but it is the same low-pass with a higher coefficient: brighter, not high-passed. The "cutoff" is a coefficient, not a calibrated frequency.
  • tearing = noise · s · grit · 2 · 3s · √P.
  • Friction, solved implicitly: F = P · 0.4 · 3s · tanh((s − V(F)) / max(0.05·s, 1e-4)) + tearing. A smooth tanh instead of a sign flip, solved with the resonator's response, can't flip sign a sample late against a fast mode and pump it.

Why the noise follows the grinder's own speed, not the slip. Taking it from the slip let the resonator's motion modulate the noise force that drives it, which built a high-Q mode into a large limit cycle (Plate at C6). Its friction term is static and the voice-mix DC blocker removes what it does to the output, which is why Grind Pressure also scales the tearing: without it the pressure was not heard.

mallet_and_grinder_contact_is_stable_far_above_default_drive and grind_pressure_shapes_the_heavy_grinding (in src/voice/mod.rs) cover it.

Corrugated Drag

A stick dragged over a ribbed surface: a constant Coulomb drag plus a bump each time the tip rides over a ridge, at a rate set by speed over ridge spacing.

Control Host range, default Role
Drag Speed 0.1–2.5, 0.82 drag speed (× Speed)
Ridge Spacing 0.01–0.2, 0.143 distance between ridges
Ridge Depth 0–2, 0.6 bump force
Exciter Mass 0.2–2, 1.1 multiplies the bump

At note-on: speed = drag_speed · (0.3 + 0.7·v).

Per sample: v_rel = speed − V; position += v_rel · 0.01; bump = max(0, sin(2π · position / spacing)) · depth · mass; force = (0.5·0.3 + bump) · sign(v_rel) · speed · 3.

The ridge rate is about speed · 0.01 · 48 000 / spacing Hz: 2.3 kHz on the default patch at velocity 100. Trap: at the extremes (Drag Speed 2.5, Ridge Spacing 0.01) that is 120 kHz, and the half-wave bumps alias. The force is one-signed, so Corrugated Drag has the largest release snap measured on the default patch (below).

Tension Rise

Stress loading up in a beam or a cable until it gives: an integrate-and-fire model. Tension builds with the pull; past a (jittered) threshold it breaks with one spike, pauses, and starts again, which gives creaks and groans at a few tens of hertz.

Control Host range, default Role
Pull Speed 0.05–1.55, 0.8 how fast tension builds (× Speed)
Break Threshold 0.1–1.6, 0.85 tension at which it breaks
Slip Stochasticity 0–1, 0.3 jitter of each threshold, ±0.3·stochasticity (× Roughness)
Creak Sharpness 0.2–1.4, 0.56 spike size

At note-on: pull × (0.5 + 0.5·v); tension 0; the first threshold drawn.

Per sample: during the pause after a break (recovery_frames, 50 samples at 48 kHz) nothing happens. Otherwise tension += (pull − V) · 0.001; past the threshold it emits a one-sample spike tension · sharpness · 5 / ratio (area kept at every rate), placed between samples with an EventSplitter, resets to 0 and draws the next threshold. The threshold is drawn once per break: drawing it every sample let a faster rate meet more draws on the way up and break earlier.

On the default patch at velocity 100 a cycle is about 1 189 + 50 samples, about 39 Hz. Every output is delayed one sample by the splitter. tension_rise_break in event_timing_tests.rs checks the spike's area and timing across rates.

The release snap

sequenceDiagram
    participant K as Key
    participant M as MSEG
    participant X as Exciter
    participant R as Object
    K->>M: note-on (trigger)
    M->>X: level rises through Onset, Attack, Hold, Decay
    X->>R: one-signed push, object deflected
    Note over R: static deflection held<br/>(the voice-mix DC blocker hides it)
    K->>M: note-off (release)
    M->>X: level ramps to 0 over MSEG Release
    X->>R: push fades
    R->>R: springs back and rings out
Loading

The same, as a list:

  1. While the key is held, Bow, Stiff Point Scrape and Corrugated Drag push the object with a force that has a large constant part. The modes settle around a deflected position; the held note sounds only the force's small varying part. The static part is DC at the output, removed by the voice-mix DC blocker (Output Protection).
  2. Voice::note_off releases the MSEG but leaves the exciter running, so the force fades over MSEG Release (minimum 0.01 s).
  3. The object springs back from its deflection and rings freely: the release is louder than the held note. On the default patch it is up to +30 dB (Plate with Corrugated Drag: −48.1 → −17.6 dB); in the factory bank up to +37 dB (Bow on Taut Cable).

The maintainer keeps it as a feature. A longer MSEG Release turns the snap into a ramp, since the deflection is let go more slowly. A Release Snap control is proposed in docs/BACKLOG.md. Heavy Grinding's friction also has a constant part, but the snap is documented for these three only; its size for Heavy Grinding is not measured.

Trap: Env Amount scales the whole MSEG, so at Env Amount 0 the friction models make no sound at all (an open item in docs/BACKLOG.md). And in Tonal mode, with the loop off, a held friction note sits at MSEG Sustain with a nearly constant force: the object is held still and the note is quiet. Drone mode forces a Decay ↔ Sustain loop, so the force keeps moving (Envelopes and Modulation).

Source files

Clone this wiki locally