Four netlist filters, all at onceAll Netlist Effects

Ramjet Complex Quad-Filter

TRACKING COMB · PEAK / TROUGH · CROSS-MOD TRIAD · COUPLED RING

Four filters. All of them at once.

A comb that tracks the note, a field of peaks and notches, three resonances that frequency-modulate each other, and a ring of eight that splits into a cluster of modes. Wire them however you like and put an LFO on any knob.

The routing The four filters

VST®3 effect · Windows 64-bit No activationmacOS VST3 and Audio Unit are in development.

The Ramjet Complex Quad-Filter's panel on its first page, with the factory preset FOUR, IN PAIRS. A black title strip: RAMJET COMPLEX QUAD-FILTER, PRESET FOUR, IN PAIRS between two arrows, REAL TIME live models on 1 · 2 · 3 · 4, and PANEL % with 75, 100, 125 and 150. Under it, on a warm grey face, the signal strip: an INPUT knob, ROUTING with SERIES, PARALLEL and 1→2 | 3→4 lit, and the four filters as boxes, 1 TRACKING COMB lit black, 2 PEAK / TROUGH, 3 CROSS-MOD TRIAD and 4 COUPLED RING, each ON with bounce netlist, wired as two pairs whose outputs meet, then MIX, OUTPUT and the left and right meters. Below it page 1, TRACKING COMB: a screen of ten orange teeth climbing from 175 Hz to 1.75 kHz, the right channel's in pink beside them, its knobs, and teal and plum chits under the knobs that are modulated. At the foot the MODULATION bank, 16 LFOs · 4 NOISE / S&H: sixteen LFO tiles each drawing its wave, four noise tiles, and LFO 4 open, a square wave at 1/8 synced to the host, routed to two of the comb's teeth.

4 filters, all running at once
95 knobs, a MOD chit under each
20 modulators: 16 LFOs, 4 noise and S&H
Every netlist solved on the bounce

What it is

One plug-in. Four filters in it.

Four complex filters, each a whole circuit, all running at once. Wire them one into the next, side by side, or in two pairs, and put any of 20 modulators on any of their 95 knobs.

Each of the four was drawn here as a netlist, a list of every part and where it connects, and run in ngspice, an open-source circuit simulator, before the plug-in existed. Then the plug-in's own solver, which shares no code with ngspice, solved the same files and came back with the figures printed in them, or within half a decibel of them. You play live models of the four; a bounce solves the netlists.

1TRACKING COMBten teeth on one BBD delay, in stereo
2PEAK / TROUGH CASCADEfive troughs, then five peaks, in series, in stereo
3CROSS-MODULATED TRIADthree resonances that frequency-modulate each other, in stereo
4COUPLED RINGeight resonators that share their modes; pickup A left, pickup B right

Routing

Three ways to wire them. Try them.

ROUTING, at the top of the panel, decides how the four filters meet, and each filter's FILTER switch turns it OFF or ON. An OFF filter passes what reaches it, so any one, two or three of the four make a filter of their own. MIX sets the result against the dry signal.

ROUTING

click a filter to switch it off or on

Two pairs in series, side by side, their outputs averaged. OUT = the mean of (IN → 1 → 2) and (IN → 3 → 4) Every filter is on.

The four filters

A page each. A circuit each.

Click a filter in the signal strip and its page opens: a screen of its response, drawn from its circuit at the knobs' values, with handles to drag, and every knob it has. Each picture here is the plug-in's own drawing of that page, with the factory preset FOUR, IN PAIRS loaded.

Filter 1, TRACKING COMB

ten teeth on one BBD delay, in stereo

A comb filter is a delay fed back into itself, and it lifts every frequency whose period fits the delay a whole number of times: its teeth. This one's delay is a bucket brigade of 512 stages, and ten band-passes sit inside its loop on the first ten teeth, each through a level of its own. So the first ten teeth are ten knobs, TOOTH 1 to 10, each from 0 to 20 dB, and REST sets every tooth above them. WIDTH sets how far each tooth reaches either side.

One current tunes the whole comb, the delay and all ten band-passes together, so the teeth stay on their harmonics wherever FREQ puts them. With FREQ FOLLOWS on MIDI NOTE, tooth 1 sits on the note you play and the ten teeth are its first ten harmonics. SPREAD sets the right channel's teeth apart from the left's, up to an octave either way.

It is the comb that Cinderblock puts after its oscillators, here as an effect, in stereo, on anything you send it.

Checked: Its staircase

Set as a staircase, TOOTH 1 at 2 dB up to TOOTH 10 at 20 dB, the netlist's own run reads 2.0 to 20.1 dB, its largest error +2.0 dB. Solved by the plug-in: 2.1, 4.1, 6.3, 8.4, 10.7, 13.1, 15.5, 18.0, 20.2, 20.5 dB, its largest error +2.2 dB.

Filter 2, PEAK / TROUGH CASCADE

five troughs, then five peaks, in series, in stereo

Ten filter sections in series: five troughs, T1 to T5, then five peaks, P1 to P5. Each has its own position, anywhere from 20 Hz to 20 kHz, and its own depth, up to 40 dB for a trough and 24 for a peak. Drag a handle across to move it, up or down for its depth.

AMOUNT scales all ten back to flat. SHIFT moves the whole cluster, five octaves either way, SCALE spreads it, gathers it or turns it over, and Q TROUGH and Q PEAK set how sharp they are. A peak is a trough turned inside out, the same parts with one wire moved, so a peak and a trough at the same place and setting cancel: checked in ngspice, a +20 dB peak on a -20 dB trough is flat within 0.005 dB.

It is the cascade the Overtime Eight-Tone hangs over its eight voices, here on its own, in stereo.

Checked: Its ten features

At the defaults, measured with a steady sine on each feature, the troughs read -28.8, -22.7, -16.6, -10.6, -5.3 dB and the peaks +5.3, +8.6, +12.6, +16.7, +20.7 dB. Nine of the ten are the netlist's own figures exactly; the other is 0.1 dB off.

Filter 3, CROSS-MODULATED TRIAD

three resonances that frequency-modulate each other, in stereo

Three resonant filters side by side, F1, F2 and F3, each with its own position, resonance and level. Then a three-by-three matrix: each filter's band-pass output can move the frequency of any of the three, its own included, at audio rate. CROSS scales the whole matrix; at zero the triad is an ordinary filter with three peaks.

A broad filter barely notices being moved. A resonant one rings with sidebands, the way an FM operator does, and the more it resonates the more it is changed. Push RESONANCE past full and the three ring on their own: the triad is three FM operators, playing off each other and off whatever you feed them. Checked in ngspice, an oscillating filter's sidebands match the Bessel functions FM theory predicts within 0.2 dB, every line above -40 dB.

With SHIFT FOLLOWS on MIDI NOTE, SHIFT is measured from the note you play: at 0, F1 sits on it. The modulation moves each frequency by a share of itself, so a patch keeps its character all the way up the keyboard.

Checked: Its three resonances

Set to 26 dB on 110, 220, 330 Hz, the solved triad reads 26.09, 26.10, 26.12 dB at 109.89, 219.91, 330.68 Hz.

Filter 4, COUPLED RING

eight resonators that share their modes; pickup A left, pickup B right

Eight resonators in a ring, each pushed by its two neighbours like masses joined by springs. With COUPLING at zero and all eight tuned alike there is one resonance. Turn COUPLING up and it splits into the ring's modes: five frequencies, three of them pairs, spreading further apart the further it goes. Each node has its own TUNE, an octave either way, and its own INPUT, how much of the signal it takes; by default only N1 is fed.

PICKUP A is heard on the left and PICKUP B on the right, each a place round the ring, and where a pickup sits decides which modes it hears. At the defaults, A sits on N1 and hears the modes with deep dips between them; B sits across the ring on N5, where the dips are gone and only a ripple is left. With SHIFT FOLLOWS on MIDI NOTE, SHIFT is measured from the note you play: at 0, the ring's f0 sits on it.

COUPLING0.883 V · c 0.20

Five modes at 1183, 1133, 1000, 847, 774 Hz. The three in the middle are pairs.

The ring's modes by the formula in its netlist's notes, f = f0 · √(1 + 2c · cos θ), θ every 45 degrees round the ring, with f0 at 1 kHz and c from COUPLING's own law, which spreads three decades across the knob so the bottom gives slow beating and the top a wide cluster. Solved in ngspice, the modes land within 0.3 percent of the formula from c = 0.08 to 0.45. Drag COUPLING to its top and c reaches 0.45, where the lowest mode is closing on 0 Hz; that is why it stops there.

Checked: Its modes and its pickups

At the defaults, pickup A hears peaks at 1187, 1133, 1000, 846, 771 Hz, the netlist's own figures exactly: a few hertz off the formula above, where neighbouring modes lean on each other. At 2 kHz pickup A is -3.5 dB and pickup B -91.2 dB, against the netlist's -3 and -91.

Modulation

95 knobs. A chit under every one.

Every knob on the four filter pages has a MOD chit under it. Click one and pick a source, or drag a source's tile from the bank onto it. One source per knob, and its DEPTH is a share of the knob's travel, either way.

The MOD chit's picker, open over the tracking comb's WIDTH knob: ROUTE INTO COMB · WIDTH, keys 1 to 16 under LFO, keys S1 to S4 under NOISE / S&H with S2 lit plum, NONE, and DEPTH on +35 % on a slider from -100 to +100, over the line: One source per knob. Depth is a share of the knob's travel, either way; the outer ring shows the swing. — The picker, open over the comb's WIDTH.

A CHIT, PLAYED

KNOB50 %DEPTH+35 %

DEPTH +35 % swings the knob 17.5 percent of its travel either way. The outer ring shows the swing, as it does on the panel.

16 LFOS

Every one its own.

SINE, TRI, RAMP UP, RAMP DN or SQUARE. FREE from 0.01 to 40 Hz, or SYNC to the host from 8 bars to 1/32 T. SKEW sets how much of the cycle it spends rising, or a square's duty, and PHASE offsets it. STEREO runs the right side up to 180 degrees ahead of the left, so one LFO can push the two sides opposite ways. RETRIG on FREE lets it run, on TRANSPORT locks it to the song's position, and on MIDI NOTE starts it again at every note, and FADE IN brings it up from nothing after each restart.

4 NOISE / S&H

Noise, held or not.

Each is NOISE or SAMPLE & HOLD, and COLOR tilts its noise from blue through white, pink and red to brown. As NOISE it wanders, smoothed to its RATE. As a sample and hold it samples the noise at its own RATE or at each cycle of an LFO it is clocked from and holds it until the next, with LAG to round the steps off. STEREO LINKED gives both sides the same draw; SPLIT gives each its own.

Playback and bounce

Live models to play. The netlists on the bounce.

Solving four circuits takes longer than the music does. So playback always runs a live model of each filter, and an offline render can solve the netlists instead. Every filter page says so beside its switch:

real time always runs live

LIVE

The same parts, as equations.

Each live model is its filter's own circuit worked into equations, part for part: every integrator a prewarped section with its transconductance amplifier's curve, the comb's limiter solved sample by sample, the ring's eight resonators solved together on every sample. The whole chain runs at twice the host's rate up to 48 kHz, for about a fifth of one core at 48 kHz with all four on.

With 10 mV at a filter's input, -40.0 dBFS at the default DRIVE, every live model is within 0.1 dB of its solved netlist at every feature. At 0.3 V, -10.5 dBFS, into the amplifiers' curves and the comb's limiter, the widest gaps are 0.31 dB on the comb, 0.18 dB on the cascade, 0.19 dB on the triad and 0.05 dB on the ring.

BOUNCE ENGINE: NETLIST

The netlists, solved.

Each filter has its own BOUNCE ENGINE switch, NETLIST or LIVE, and all four start on NETLIST. When your DAW renders offline, each filter on NETLIST is solved from its netlist, every part, each channel its own copy of the whole circuit, starting from the circuit's resting state at the knobs as they stand. The title strip says what is running: REAL TIME, or BOUNCING and which filters run their netlists.

Both engines have the same latency, 48 samples at 44.1 and 48 kHz and none above, so a bounce lines up with playback to the sample. A bounce starts every LFO and every noise source from the same place, so the same render comes out the same every time.

What it is, and what it is not

The netlists as written. What the plug-in adds.

None of the four copies a machine. Each was designed here, written out as a netlist and checked in ngspice, and the plug-in reads those files as they are. These are the things it does to them as it loads them, and why.

THE KNOBS

Each knob moves what its netlist says it moves: a control voltage the circuit already has, or the resistors the netlist's own laws set, changed in place while the circuit runs.

THE COMB'S DELAY

The comb's netlist draws a bucket brigade and its clock, and offers a mode in which the brigade is an ideal delay. The plug-in solves that mode. The delay's length follows the brigade's own clock law, so FREQ and its modulation retune it, but there is no clock residue and no brigade noise. The clock is cut from the solve; the load it puts on the tuning current is kept.

THE RING'S PICKUPS

The ring's netlist reads its two pickups through formulas evaluated during the run. The plug-in solves the eight resonators and then takes each pickup from their outputs by the netlist's own weights: the same sum, done after the solve.

THE SAMPLE RATE

A solver that steps at a fixed rate tunes an integrator a little away from where the circuit puts it, more so near the top of the band. The plug-in corrects each filter's tuning for its rate, so the teeth, the sections, the resonances and the modes land where the netlists put them.

A SEED OF NOISE

A tenth of a microvolt of noise on the triad's and the ring's inputs, so a resonance pushed past full can start from silence, as a circuit's own noise would start it.

THE LEVEL

0 dBFS is 1 V at a filter's input, times its DRIVE, -12 to +24 dB. At -20 dBFS each filter runs at the level its netlist was measured at. Louder, its transconductance amplifiers bend and compress, and that is part of the sound.

Four netlist filters, all at once