The Voice Engine

The defining idea behind Timbre: the oscillator and the filter are the same thing.

In conventional synthesizer architecture, an oscillator generates a harmonically rich waveform and a filter shapes it. These are separate circuits with separate control signals. Timbre eliminates this distinction. Each voice uses the CY8C29466’s switched-capacitor analog blocks in self-oscillation — the same circuit that generates the tone also determines its spectral character.

Switched-Capacitor Fabric

The CY8C29466 contains 12 configurable analog blocks arranged in four columns of three. Each block can be configured as an integrator, comparator, sample-and-hold, instrumentation amplifier, or other analog function. The blocks are connected by a programmable routing matrix that allows arbitrary signal paths between them.

Crucially, the analog configuration can be changed while the chip is running. This is not a “load a patch at startup” architecture — it is a “change the circuit topology mid-note” architecture. The mechanics of how this works — from register-level writes to full flash reprogramming — are covered in Reconfiguration.

Block Budget

Each voice has 12 analog blocks to work with. A typical voice allocation:

Function	Blocks	Purpose
Self-oscillating filter core	3–4	Bandpass in self-oscillation generates the fundamental; topology determines timbre
Secondary oscillator / sub	2	Relaxation oscillator from integrator + comparator for sub-octave or FM source
VCA	2	Voltage-controlled amplifier for envelope-driven amplitude
Waveshaper / soft clipper	1–2	Nonlinear transfer function for harmonic enrichment
CV conditioning / spare	1–2	Input scaling, offset, or additional modulation path

This budget is a starting point. The allocation is programmable — a patch that needs two filter stages and no waveshaper simply reconfigures the routing.

Self-Oscillation Modes

The switched-capacitor fabric supports several distinct oscillation mechanisms, each with different tonal character:

Bandpass Self-Oscillation

A biquad bandpass filter driven past its stability threshold oscillates at its center frequency. This produces a nearly sinusoidal tone whose pitch tracks the switched-capacitor clock ratio. By adjusting Q (resonance) from just-oscillating to hard-clipping, the waveform morphs continuously from sine-like to square-like.

This is the primary oscillation mode — it is the most pitch-stable and produces the cleanest fundamental.

Relaxation Oscillation

An integrator block charging a capacitor, compared against a threshold by a comparator block, with the comparator output resetting the integrator. This produces a sawtooth or triangle waveform depending on the reset mechanism. The pitch is set by the integration rate (clock ratio and capacitor value).

Relaxation oscillators have a different character from bandpass self-oscillation — more aggressive, more harmonically complex, and with natural PWM capability by adjusting the comparator threshold.

Phase-Shift Oscillation

Three or more integrator blocks in cascade with appropriate feedback create a phase-shift oscillator. The frequency is determined by the per-stage phase shift, which is a function of the switched-capacitor clock. This mode produces a smoother, more organ-like tone.

Filter-as-Oscillator: A Short History

The idea that a filter can be the oscillator has a distinguished but narrow history in synthesis. Self-oscillating filters have been available since the first Moog modular modules — when resonance feedback gain exceeds unity at a frequency where phase shift completes 360 degrees, the filter becomes an oscillator producing a nearly pure sine wave. The Korg MS-20 (1978) made both its highpass and lowpass filters self-oscillating, with Korg explicitly marketing this as “another sound source.” The Formanta Polivoks (1982) was characteristically eager to self-oscillate, contributing to its aggressive Soviet character.

But the definitive filter-as-oscillator instrument is the Roland TR-808 (1980). Its bass drum circuit uses a bridged-T bandpass filter as its sole sound source — there is no conventional oscillator. A trigger pulse excites the filter into decaying self-oscillation at roughly 49-56 Hz. The subtle pitch drop during decay, caused by voltage leakage in the retriggering circuit, became one of the most influential timbres in music history. The TR-808’s toms, congas, and other tonal percussion voices also use bridged-T networks as their primary sound sources. Roland’s engineers demonstrated that filter-as-oscillator could be a primary and commercially viable synthesis method, not merely a side effect of cranking resonance.

The Buchla 292 Low Pass Gate (circa 1973) took a different path, merging filter and VCA into a single vactrol-based circuit where sound simultaneously gets quieter and duller as control voltage decreases — mimicking the natural decay behavior of acoustic instruments. The Serge Resonant Equalizer could be pushed into self-oscillation at extreme settings, functioning simultaneously as equalizer, formant filter, resonator bank, and sound source — exemplifying Serge Tcherepnin’s dual-purpose module philosophy, where function follows patching rather than panel labels.

Timbre’s self-oscillating switched-capacitor fabric represents something distinct from all these precedents. Rather than a continuous-time analog filter accidentally or deliberately entering oscillation, Timbre uses the SC fabric’s clocked switching behavior itself as both the oscillation mechanism and the timbral shaping mechanism simultaneously. The “oscillator” and “filter” are not merely the same circuit in different modes — they are literally the same computational process. The Lineage page traces this convergence across the full sixty-year arc of reconfigurable analog synthesis.

What This Enables

Because the oscillation mechanism is the filter topology, changing the filter changes the fundamental character of the sound — not just its brightness, but its waveform, its harmonic series, its basic identity.

Mid-note topology switching. A held note can transition from bandpass self-oscillation (clean sine) to relaxation oscillation (aggressive saw) under envelope or CC control. This is not crossfading between two oscillators — it is the same circuit reorganizing itself.

Per-voice oscillation character. In a 16-voice chord, each voice can run a different oscillation mode simultaneously. Velocity layers that change not just volume but waveform topology. Timbral splits where the bass notes use relaxation oscillation and the treble notes use bandpass. Random voice-character spread for natural ensemble thickness.

Paraphonic unison. Multiple chips assigned to the same pitch with slightly offset clock ratios produce natural chorus and ensemble effects — no DSP, no bucket-brigade delay, just physical analog frequency beating.

Voice morphing. Two adjacent chips can run different topologies on the same pitch, with their outputs mixed. Crossfading the mix produces timbral morphing between fundamentally different oscillation modes.

Clock Noise

Switched-capacitor circuits inherently leak their clock frequency into the audio path. This is the primary engineering challenge for Phase 1: characterizing the clock noise floor and determining what output filtering is required.

The SC clock runs in the hundreds of kHz range — well above audio — but aliasing products and power supply coupling can produce audible artifacts. Mitigation strategies include output reconstruction filtering (a continuous-time analog lowpass after each voice’s SC output), careful power supply decoupling, and clock frequency selection to place noise products outside the audible band.

Switched-Capacitor Character

No major synthesizer manufacturer ever used switched-capacitor filters as a primary musical VCF. Sequential, Roland, Yamaha, Korg, and Oberheim all relied on continuous-time analog filter ICs — the CEM3320, SSM2044, IR3109, and their descendants. SC filters appeared only peripherally: the E-mu Emax (1986) used a seventh-order elliptical SC tracking reconstruction filter on its DAC output, and some Fairlight CMI versions employed 4066-based SC circuits for anti-aliasing. But these served engineering functions, not musical ones. The reasons are instructive: clock feedthrough (typically around 10 mV peak-to-peak at the clock frequency) injects unwanted tones, the Nyquist constraint limits input bandwidth at low cutoff frequencies, and sweeping cutoff by changing clock frequency digitally can produce audible stepping artifacts.

These limitations explain why SC filters found a home primarily in the DIY and Eurorack communities, where their lo-fi character is valued rather than feared. The Skull Circuits VCF-4 offers dual self-oscillating SC filters with CV control. The ALM MUM M8 uses the MF6CN-50 SC filter IC, descended from the same Akai S950 sampler lineage. The CGS Bi-N-Tic Filter/Oscillator exploits SC aliasing artifacts deliberately as a sound source. Tim Stinchcombe’s LMF100-based DIY VCF uses a 4046 PLL VCO for voltage-controlled cutoff. These modules demonstrate that SC filters can produce musically compelling results — but their character is distinctly grittier than continuous-time designs.

Parameter	PSoC 1 SC Blocks	CEM/SSM Traditional	Anadigm FPAA
SNR	~50-65 dB (estimated)	80-95 dB	~75 dB (measured)
Clock noise	Inherent (~10 mV p-p)	None	Inherent
Reconfiguration	~12 us	None (fixed)	<1 ms
Cutoff control	Digital (clock ratio)	Analog (CV)	Digital (SPI)

For Timbre, this sonic character is a feature, not a deficiency. The SC artifacts — aliasing, clock feedthrough, stepped response — become part of the instrument’s identity, the way the TR-808’s bridged-T filter decay became a defining sound rather than a design limitation. The mitigation strategies are well-established for cases where cleaner output is desired: post-filtering with simple RC lowpass networks, maintaining high clock-to-cutoff ratios (100:1 or greater), careful PCB layout with separated analog and digital grounds, and maintaining integer relationships between SC filter clocks and modulation clocks to prevent offset artifacts. The Lineage page discusses where SC synthesis fits in the broader landscape of reconfigurable analog design.