VXN2

VXN2 (“vixen 2”) is a 6-operator FM synthesizer in the DX7 / Montage lineage, built in Rust as a CLAP plugin. It is a sibling to VXN1 — they share the CLAP shell, preset conventions and HTML-faceplate UI, but VXN2 has its own DSP kernel, parameter table and patch format. Its headline addition to the classic FM formula is first-class voice stacking: supersaw-style unison without any wavetable storage.

The DSP kernel, CLAP shell and faceplate are all shipping; the synth is in mod-matrix / review hardening, with a browser (WASM) port in progress. No download is available yet.

Synthesis architecture

• 6 operators, 32 algorithms. Every DX7-canonical algorithm graph is supported. algo is a pure topology parameter — it rewires which operators are carriers, which modulate which, and which carries the feedback path, without touching per-op values. Structural feedback depth is a separate 0–7 control.
• Fixed-point phase accumulation. Q32 phase gives free wraparound and zero drift; only past the phase boundary (filters, envelopes, mixing) does the signal become f32.
• Approximated sines, not table lookup. A branch-free Bhaskara + Moser polynomial (5 mul + 2 abs + 2 add) that vectorises well and avoids NEON/AVX gathers. THD ≈ −59 dB.

Voice stacking

The differentiator. Each played note can spawn up to 8 concurrent op-voices, detuned and spread without storing a single wavetable:

• Density 1–8 lanes per note (1 = no stacking).
• Detune up to 100 cents across the stack, centre lane un-detuned.
• Parameter spread distributable via mod-matrix to stereo position or any other modulatable parameter.
• Phase spread at note-on, from aligned to maximally decorrelated — the “shimmer” trick.
• Distribution — Linear, Geometric (exponential clustering) or Random (re-rolled per note-on).

Per-operator control

FM sound design lives in the modulator envelopes and key-scaling, so every operator (×6) exposes the full set:

• Rational tuning — num / denom ratios reach sub-octaves and just intervals (3/2, 5/4, 7/4); a fine offset sweeps continuously between detents, including irrational ratios. A per-op Fixed mode gives absolute-Hz operators for inharmonic and percussive timbres.
• Per-cent detune on top of the ratio for thickening and beating.
• Four-stage EG (rates + levels) per operator, plus velocity sensitivity and full DX7-style keyboard scaling (break point, left/right depth and curve, rate scaling).
• Carrier pan into the stereo bus (disabled by the UI when an op is a modulator under the current algorithm).

Modulation

• LFO 1 (global) — one phase accumulator shared across all voices, for patch-wide, optionally tempo-synced sweeps.
• LFO 2 (per-voice) — key-triggered with delay + fade-in, the breathy / humanising LFO; each voice and each stack lane has its own phase.
• Pitch EG with signed levels (pitch can swing up or down) and a depth macro.
• Mod Env — a general-purpose ADSR, matrix-routed only.

Mod matrix

A 16-slot matrix per patch. Each slot maps a source (lfo1, lfo2, pitch_eg, mod_env, mod_wheel, aftertouch, velocity, key, voice_idx, voice_spread, voice_rand) onto one of 29 destinations — per-op pitch / level / pan, global pitch, LFO and stack macros, filter cutoff / resonance, FX mix, feedback — with a depth and a curve (lin / exp / log / bipolar).

Sources and destinations carry a granularity tier (patch-global, per-stack, per-lane); a routing is coherent when the source is coarser-or-equal to the destination, and the UI flags lossy or degenerate routings in red while still allowing them so old patches load. Slots 1–8 expose their depth as CLAP-automatable parameters for performance macros. Routing key through a bipolar curve replaces the old keyboard-split feature within a single patch.

Filter (optional)

An optional per-voice oversampled OTA-C ladder, post-stack-sum. Off by default — with the filter disabled the render path is bit-identical to a filterless patch at zero added cost. LP / HP / BP / Notch modes, 12 or 24 dB/oct slope, tanh drive, self-oscillation at full resonance, and a localised 1× / 2× / 4× / 8× oversample factor. Cutoff and resonance are mod-matrix destinations.

Effects

A fixed FX bus: cleanup → dynamics → phaser → delay → reverb → master gain. Each block-level effect bypasses to a bit-exact passthrough when off, fading its wet mix out first to avoid clicks.

• Dynamics — a feed-forward peak compressor into a tanh saturator (channel-strip topology), placed first so it evens FM transients before the time effects accumulate them.
• Phaser — four cascaded all-pass sections per channel with an anti-phase L/R LFO sweep around 600 Hz.
• Delay — clean digital delay with BPM sync, feedback and ping-pong.
• Reverb — a feedback-delay-network (FDN) reverb with size, decay, damping and mix.

Voicing

Up to 16-voice polyphony, or a monophonic Solo mode with legato and portamento (glide is also available per-note in Poly). A global master tune and volume finish the chain.