Elexius · Saturator · VST3 & AU

bis hierher und nicht weiter.

A plimsoll line is the load mark painted on a ship's hull — a disc cut by a horizontal line that says how deep this hull may go and no deeper. It is not a metaphor we reached for. It is what a saturator is.

zwei fader. und ein bild davon, was sie tun.

Drive pushes whichever circuit is selected. Sag decides how readily the supply rail gives way under load. That is the whole sound design. Everything else on the panel exists to tell you what those two are doing.

Drive means one thing regardless of which circuit is in. The transistor and the tube were matched by measured THD rather than by feel, so the same fader position is the same amount of work in both — 1.17 % against 0.95 % at 40, 17.4 % against 12.8 % at 80. The character stays completely different. Only the quantity is matched.

die kurve gibt zu, was sie ist.

The plot answers the one question the two faders raise and nothing else on the panel can: what shape is this thing actually putting on the signal. Three traces — the bright curve at the middle of the supply's travel, and the two rail bounds it moves between as the supply gives way. The dashed lines are where the stage stops following its input.

Drawn to true scale it cannot be read. The drive range is 200 : 1, and at Drive 0 the transistor sits 15.5 dB below where it clips. Any scale that puts a 45° line at the bottom of the faders throws the saturation level clean off the plot; any scale that keeps the saturation level on the plot draws the bottom of the faders as a flat line. Both were built. Neither is a usable picture.

So the drawn curve is a soft clipper on a chosen scale, with its character taken from the measurements: the ceiling moves by the ratio the real saturation level moves by, the asymmetry is the measured one, and the transformer is entirely real and unscaled, hysteresis and all. And the plot says so, in the band below the negative clipping line.

What each fader currently means to the circuit that is in sits next to the fader's own number. On the transistor that reads ×2.73 and RAIL 0.90; on the tube it is grid drive in dB and the B+ in volts, swept from the Ampex's real 220 V down to 140 and drooping further under sustained load.

The emphasis curve

zeichnen, was der schaltung wehtun soll.

Six bands, and they run twice: once at +g in front of the drive circuit, once at −g behind it. At 100 % the second chain is the exact algebraic inverse of the first, so the curve changes nothing about the tone. It only changes which part of the spectrum is driving the circuit between them into its nonlinearity.

Measured, that is not a claim about intent: a ±9 dB curve at 100 % de-emphasis leaves the response flat to ±0.02 dB from 20 Hz to 16 kHz, while moving one band at 5 kHz by ±18 dB shifts the third harmonic a 1 kHz tone generates by 10.5 dB with the same circuit behind it. Draw where it should hurt. Get the harmonics, not the EQ.

The transient shaper

dieselbe idee, auf der zeitachse.

Instead of a curve deciding which frequencies drive the stage harder, a detector decides which part of a hit does. The stage is wrapped in a bracket — multiply going in, divide coming out — so with a linear circuit inside it the two cancel exactly and there is nothing to null out. With a nonlinear one, the emphasised part comes back with harmonics the division cannot undo. That difference is the whole effect.

Level independence is free, because a ratio of two envelopes does not care how loud the input was: measured at 0.000 dB of divergence over a 36 dB range. The fader moves the balance between attack and tail monotonically across 15.0 dB.

was es kostet.

Single instance, 48 kHz stereo, 512-sample blocks, 2× oversampled throughout, emphasis curve in circuit. The tube is the expensive path by an order of magnitude — a warm-started Newton solve against a real load line, twice per sample per channel, at twice the sample rate. That is the honest cost of a physical model, and it is unchanged from the module it came from.

The transfer curve costs nothing on the audio thread — it is computed on a background thread and drawn from the last one that finished, because finding where the tube stops following its input is 801 Newton solves of a real load line, twice per redraw. Measured lateness with the plot live is 0.71 ms mean.