Three circuit models of the classic 1966 FET limiting amplifier, one per revision, built with nice-plug and a Vizia panel. Each builds as CLAP and VST3, and on macOS also as an Audio Unit (AUv2), through nice-plug-au2.
| Plugin | Circuit | Character |
|---|---|---|
| Comp76Fx Rev A | no low noise circuit, FET preamplifier and line amplifier, Class A output, painted meter surround | the dirtiest and the noisiest, by 3.4 dB; its switch bank makes its ratios about 3 % steeper than marked |
| Comp76Fx Rev D | low noise circuitry, Class A output, black panel | the one most people mean; the reference reissue is patterned on the D and E |
| Comp76Fx Rev F | low noise, push-pull Class AB output, black panel | the cleanest, with the distortion turning symmetrical |
Each revision's switch bank is read off its own schematic in the UREI manual, and the noise is the manual's 80 dB signal to noise at threshold for the low noise units, with the Rev A 3.4 dB worse as a side by side measurement found. The amplifiers are the schematics' too: the Rev A's preamplifier and line amplifier each start with a JFET, where the later units use bipolar transistors. Its output stage is the same Class A transistor and transformer as the Rev D's. How hard each one's FET and output stage are driven, and how much the JFETs colour, is documented only in direction, so those are voicing.
They are not affiliated with, endorsed by, or connected to Universal Audio or any other manufacturer. Model numbers are used only to say which circuit is modelled.
The unit is a feedback compressor: the sidechain samples the signal after
the gain element rather than before it. It takes it from the preamplifier,
straight after the gain element and ahead of the output control and the line
amplifier, so the output stage's colour is outside the loop. That is not a
detail. With k as the sidechain's gain, the loop settles at
g = -k / (1 + k) * (input - threshold)
so the slope is 1 / (1 + k) and the ratio is simply 1 + k. The 4:1 button
is k = 3; the 20:1 button is k = 19. What the feedback buys is ratios that
land on their markings. It does not soften the knee on its own: the knee is in
the threshold circuit.
- Each ratio has its own threshold. The buttons switch a DC divider that biases the rectifier diodes along with the signal divider that feeds them, and the manual says "selecting higher ratios also raises the threshold level": −24, −25 and −26 dB for 20:1, 12:1 and 8:1, and for 4:1 "the lowest threshold is −30 dB", where its soft knee begins.
- The knee is softest at 4:1. A diode turns on over a span of voltage, not at a point, and that span covers the most decibels where the signal at the diodes is smallest. The 4:1 passes the smallest share of the smallest signal, about a tenth of the 20:1's, so its knee is about ten times as wide: the "soft knee in the threshold circuit for this ratio" that the manual measures its ratio test around. The manual gives no width; the size of all four is bounded by that test.
The loop is solved within each sample rather than run a sample behind itself. Running it behind is a correction the size of the open loop gain, and at the fastest attack that overshot the settling point on every transient, by as much as 20 dB without oversampling, and held the excess for the whole release.
-
The gain element is a FET used as a voltage controlled resistor, and it distorts the audio passing through it increasingly as it is pulled down. That is most of why the unit sounds the way it does when it is working hard.
-
The Rev A's amplifiers start with JFETs. A JFET's current follows the square of its gate voltage, so inside each amplifier's feedback loop it adds second harmonic, in proportion to the signal, and no third. It bends over smoothly where the JFET cuts off. There are two: the preamplifier's, ahead of the output control, where the loop holds its level steady while the unit limits, and the line amplifier's, after it. With the limiting off, trading input for output therefore changes a Rev A's colour but not a Rev D's.
-
The recovery runs two stages together, so it is program dependent rather than a fixed curve.
-
The sidechain runs out of rail rather than clipping, but only where a real one would. At equilibrium the demand and the gain reduction are the same number, so a curve that bends from the origin bends the static ratio with it and every button reads low. It stays linear across the range the unit works in and turns over only near the rail.
-
All-button mode is modelled on the switch bank as the schematic draws it. Each button connects a tap on two resistor ladders, one feeding the sidechain and one in the gate's bias network, and pressing several joins their taps, which shorts everything between the lowest and highest pressed. So only the outer buttons matter (4 + 20 is all four, as Universal Audio notes), and all four:
- pass the sidechain 0.456 of the signal against the 20:1's 0.802, which sets a threshold and ratio between the 8:1's and the 12:1's;
- pull the gate's resting bias from −2.0 V to −3.2 V (a circuit simulation of the mode posted to GroupDIY), so the sidechain has to charge through a dead zone before the FET opens: the attack lags, the release reaches no reduction far sooner, the reduction is lighter than a plain ratio's at the same drive, and the gain reduction meter, which reads that bias, rests pinned past zero;
- throw off the gate's distortion-cancelling feedback, so the gain element bends the signal further.
It measures 15:1 at every drive level, the middle of the manual's "somewhere between 12:1 and 20:1". Two figures are not documented and are assumptions: how many decibels of control the 1.2 V bias drop is worth (taken as 30 dB, the drop's share of the gate's roughly 2 V working range), and how much it raises the loop gain, which is fitted to that range.
Measured by cargo run --release -p comp76fx_core --example bench -- --spec
and the tests in core/tests/compression.rs:
| 4:1 button | 4.04:1 |
| 8:1 button | 8.07:1 |
| 12:1 button | 12.12:1 |
| 20:1 button | 20.26:1 |
| all four in | 15.1:1 driven, 15.4:1 gently; 4 + 20 identical |
| threshold | 20:1 at −24.0 dBFS; 12:1, 8:1 and 4:1 at −1.0, −2.0 and −3.0 dB from it |
| knee | reduction already at threshold: 0.16, 0.27, 0.41 and 0.74 dB from 20:1 to 4:1 |
| the manual's ratio test | 20:1, 12:1 and 8:1 from 1 dB of limiting, 4:1 from 3 dB: all within 12 % against its 20 % |
| no buttons in | exactly 1:1, colour with no gain reduction |
| attack | 19.5 µs to 794 µs against a marked 20 µs to 800 µs |
| release | 50.0 ms to 1100 ms against a marked 50 ms to 1.1 s |
| distortion, idle at −18 dBFS | Rev A 0.48 %, Rev D 0.33 %, Rev F 0.05 % |
| frequency response | within 0.53 dB across 20 Hz to 20 kHz |
| signal to noise at threshold, the manual's way | Rev A 76.5 dB, Rev D 80 dB, Rev F 80 dB, at every sample rate and oversampling setting |
| latency | 74 samples at every oversampling setting, the wet/dry mix included |
core/tests/calibration.rs holds the published figures to those tolerances,
including the response at every sample rate and oversampling setting, the ends
of the dials, and the meter's needle against the marks printed on its own
face. core/tests/latency.rs holds the reported latency to the real one and
the dry signal the mix blends in, in line with the wet one. core/tests/threshold.rs runs the
manual's own ratio test and holds each button's threshold and knee, and
all-button mode's switch bank, timing and meter.
One of those is worth reading twice. The ratios sit about a percent high because the unit's own distortion takes energy out of the fundamental the measurement reads; the loop itself, linearised, lands a percent low, and the two nearly cancel.
The panel is the hardware's. Input drives the signal against a fixed operating point, which is how the unit is threshold-less; output is make-up, and as on the hardware it sits ahead of the line amplifier, so turning it up drives the output stage harder and colours more. Attack and release are engraved 1 to 7, slowest to fastest, which is backwards from most compressors and is how the originals were engraved; 1 is 800 µs and 1.1 s, 7 is 20 µs and 50 ms. Turned fully anticlockwise past the 1, the attack control switches the limiting off: the signal still passes through the amplifiers for their colour, with no gain reduction.
The four ratio switches are mechanically interlocked, so clicking one releases the others. Hold shift or ctrl to latch, which is how you get all four in at once without having to be quick with your fingers.
The meter switch selects gain reduction or output level referenced to +4 or +8, and, as on the hardware, its OFF button is the power switch: the unit is out of circuit and the signal passes straight through (a real one switched off passes nothing). The output positions read the average level, referred to a sine, so a tone peaking at −18 dBFS sits on 0 VU at +4.
The strip above the panel is not on the hardware. It carries the preset drop down, a save button and the settings button, which holds the window scale (50 % to 200 %), the oversampling quality and the wet/dry mix, turned fully up to WET for the circuit alone. The plugin reports the same latency at every oversampling setting, and the dry signal the mix blends in and the switched-off signal are delayed to match, so neither moves the track against the rest of the session.
Saved presets are one JSON file each, under a folder of the revision's own so
the three do not share: ~/.config/comp76fx-rev-a/presets on Linux and macOS
(or under $XDG_CONFIG_HOME), and %APPDATA%\Comp76Fx Rev A\Presets on
Windows. Each carries a cross to delete it that asks before removing the file.
Saving under the name of one of your own, in any capitalisation, replaces it. A built-in preset has no file, so it cannot be deleted, and
saving under its name writes a preset of your own beside it rather than
replacing it in the list; replacing it would put it out of reach for good.
Built-in presets include All Buttons In: all four switches in with attack and release fully fast, the documented setting for drum rooms and "in your face" vocals, driven so a −18 dBFS signal takes about 10 dB of reduction, with the make-up set so it comes back at unity, as every built-in preset but the parallel one does.
./install.shBuilds all three and installs the CLAP and VST3 of each into
~/.clap/BurningTreeC and ~/.vst3/BurningTreeC. Pass --no-build to install
what is already built, or set CLAP_PATH and VST3_PATH to install elsewhere.
To build without installing:
cargo xtask bundle -p comp76fx_rev_a -p comp76fx_rev_d -p comp76fx_rev_f --releaseOn macOS, cargo xtask bundle-universal with the same packages, then
tools/package_au2.sh, which wraps each universal binary in a signed
.component for ~/Library/Audio/Plug-Ins/Components.
To try one without a host:
cargo run --release -p comp76fx_rev_d --features standalone -- --backend autoUnder the GNU General Public License version 3 or later, whose text is in
LICENSE.
THIRD-PARTY-NOTICES.md reproduces the licences and
copyright notices of every crate the plugins link on any platform, and of the
Skia library the panel is drawn with. Noto Sans is compiled in for the panel
lettering and is under the SIL Open Font License 1.1. Regenerate the file after
changing dependencies:
python3 tools/third-party-notices.py| Path | |
|---|---|
core/src/dsp/detector.rs |
the feedback sidechain, its timing and the loop solver |
core/src/dsp/revisions.rs |
the circuit values of the three revisions |
core/src/dsp/fet.rs |
the gain element and its distortion |
core/src/dsp/amp.rs |
the Class A and Class AB output stages |
core/src/editor/ |
the front panel |
core/src/presets.rs |
built-in and saved presets |
rev_a, rev_d, rev_f |
one identity each: names and plugin ids, CLAP, VST3 and AU |
core/tests/compression.rs |
the measurements above |
core/tests/latency.rs |
reported latency and the dry signal's alignment |
core/tests/threshold.rs |
the threshold circuit against the manual |
vendor/baseview |
the GUI window backend, patched for Windows and X11; see its PATCHES.md |
vendor/vizia, vendor/vizia_plug |
Vizia 0.4 and its plugin adapter, ported to that backend and to nice-plug |
vendor/nice-plug-au2 |
the Audio Unit wrapper, fixed so the three revisions can share a host |
tools/package_au2.sh |
builds the macOS .component bundles |