TRANSMIT Tower technical specifications

This is the engineering reference. The main pages say what Tower sounds like and how to use it; this page says exactly what it does: every stage in the chain, its topology, its detector and its time constants, what each of the nine presets actually sets, what EXTREME changes, how AUTO INPUT decides, and the full list of host-automatable parameters.

Nothing on this page is an estimate. The numbers are the ones the plugin runs with.

1. At a glance

2. Formats and compatibility

• macOS. Install the VST3 bundle into ~/Library/Audio/Plug-Ins/VST3 (this user) or /Library/Audio/Plug-Ins/VST3 (all users), and the Audio Unit component into the matching Components folder, then rescan in your host. The Standalone application runs on its own.
• Hosts. [PLACEHOLDER: list of hosts verified before launch.]
• Windows. [PLACEHOLDER]
• Code signing. [PLACEHOLDER: signing / notarization status and any first-launch steps.]
• Channel layouts. Mono and stereo only, and the input and output sets must match. In mono the stereo stage, the correlation estimate and the mid/side split still run; they simply have nothing to separate.
• Block size. Any. Parameters are read once per block, and the engine's own settings update runs on a 64-sample grid, so a switch you throw takes effect within 64 samples rather than on the exact sample.
• Sessions. State is the 22 parameter values. A session saved by an older build that had fewer parameters loads with the newer ones at their defaults.
• Bypass. Both the plugin's own Bypass parameter and the host's bypass keep the whole chain running underneath and crossfade over 20 ms to a dry path carrying the same 8 ms of delay. Nothing is reset, nothing restarts, and switching back does not click. The display freezes its readings while bypass is active.

3. Signal path

Nineteen cells, in this fixed order. Seventeen of them have a switch; cell 0's button engages AUTO INPUT, and cell 18 has no control.

Cells 8 to 13 are interlocked in the display: they cannot be clicked while EXTREME is out.

A stage switch is not a bypass. For the input high-pass, the phase rotator, the AGC, the LF shelf, the stereo stage and the output high-pass, the stage runs regardless and its output is discarded when the switch is out. For the four FM cells it changes settings rather than removing the block. For the final low-pass it selects a flat kernel. For the true-peak limiter it keeps the delay and stops applying gain. This is why the latency never moves.

4. Stage by stage

4.1 AUTO TRIM (cell 0)

• The applied trim is either the manual Input value or the AUTO INPUT servo position, smoothed by a one-pole with a 20 ms time constant and applied identically to both channels.
• Manual range ±18 dB in 0.1 dB steps, clamped to ±18 dB inside the engine as well.
• The trim the engine is actually applying — manual or servo — is published to the display, so the readout is always the truth rather than the knob position.
• EXTREME: no effect.

4.2 INPUT HPF (cell 1)

4.3 PHASE ROTATOR (cell 2)

• Three cascaded first-order all-pass sections per channel, at 240 Hz, 720 Hz and 2100 Hz.
• Each section's coefficient is (tan(π·f/fs) − 1) / (tan(π·f/fs) + 1).
• No detector, no level dependence: it is a fixed phase network. It reduces waveform asymmetry so the clipper and limiter downstream have less to take off for the same loudness.
• This is the normal-mode rotator. The FM rotator at cell 9 is a separate instance with the same design.
• EXTREME: no effect.

4.4 AGC (cell 3)

A slow level servo, not a compressor.

Because the gate only restrains upward movement, the AGC will always pull a loud passage down promptly, but it will not chase the noise floor up between phrases.

4.5 POST COMP (cell 4)

4.6 LF SHELF (cell 5)

A level-dependent low boost that adds weight to quiet programme and gets out of the way of loud programme.

4.7 MULTIBAND (cell 6)

Five bands (see §5), each with two dynamics stages in series.

Slow path. Per band, a peak detector on max(|L|, |R|) of the band signal, one-pole with that band's attack and release, and a ratio curve −over·(1 − 1/ratio).

Band linkage. Each band's desired reduction is interpolated linearly toward the arithmetic mean of all five desired reductions by the preset's linkage fraction. At 0 % the bands are entirely independent; at 95 % they move almost as one.

Fast path. An infinity-ratio ceiling rather than a ratio compressor: target = min(0, threshold − level), applied to the band that the slow path has already compressed. Its attack is fixed at 0.35 ms for every band and every preset, and its release times are {120, 100, 80, 65, 55} ms for bands 1 to 5 in every preset. Only its threshold is preset-dependent, and only three of the nine presets change it.

Look-ahead. 1.0 ms per band. The fast gain is computed on the undelayed band and applied to the delayed one, so the ceiling is already in place when the peak arrives.

Post trim. The preset's per-band trim is folded into the fast-path gain, which means it is applied after the ceiling. A positive trim — up to +2.2 dB on DANCE — deliberately pushes a band back above the ceiling it was just held to.

Switch off. The crossover and every delay line still run, and a parallel delay supplies the output, so the latency and the phase response are unchanged.

EXTREME (only while the stage is on): every band's threshold −1.0 dB, every ratio × 1.10, every release × 0.80, and the fast-path threshold −0.7 dB. It does not change attack times, linkage or post trims.

4.8 STEREO / MONO LF (cell 7)

4.9 15K PRE-FLTR (cell 8)

• Two cascaded second-order low-passes at 15000 Hz, Q = 0.70710678, per channel — a fourth-order Linkwitz-Riley magnitude response.
• Mixed in by a dry/wet crossfade whose mix is the EXTREME crossfade when the stage is on, and zero when it is off. The filters themselves run continuously either way.
• This is the only filter in the chain that is genuinely EXTREME-only. The final low-pass at cell 16 is not.

4.10 FM PHASE (cell 9)

• A second phase rotator with the same 240 / 720 / 2100 Hz design as cell 2 — not a different or heavier network.
• Crossfaded in at 0.35 × EXTREME crossfade, so it is never more than 35 % wet, even at full EXTREME.

4.11 to 4.14 — the FM block (cells 10 to 13)

Pre-emphasis, HF limiter, FM clipper and de-emphasis are one processing block, applied at a single point in the chain. They have four independent switches and four display cells, but all four cells read the same post-clipper measurement, so their signal levels can never differ from one another. Their switches change the block's settings; none of them removes it from the path.

4.11 PRE-EMPHASIS (cell 10)

• Time constant τ = 50 µs × EXTREME crossfade, so it is 0 in normal mode and 50 µs at full EXTREME, ramped continuously in between.
• Not a shelving biquad. It is a bilinear-transformed first-order section:

K  = 2 · fs
n0 = 1 + K·τ₁      n1 = 1 − K·τ₁
d0 = 1 + K·τ₂      d1 = 1 − K·τ₂
b0 = n0/d0         b1 = n1/d0        a1 = d1/d0
y[n] = b0·x[n] + b1·x[n−1] − a1·y[n−1]

• Pre-emphasis puts the time constant in the numerator (τ₁ = 50 µs, τ₂ = 0): a rising high-frequency characteristic with no pole.
• The switch sets its time constant to zero independently of the de-emphasis switch, so the two halves can be deliberately unbalanced.

4.12 HF LIMITER (cell 11)

• Active only when its switch is on and the pre-emphasis constant is above 1 µs — which is to say, only under EXTREME.
• Detector: the instantaneous peak of the pre-emphasised signal, max(|L|, |R|).
• Wanted gain is the threshold divided by that peak, floored at the maximum reduction. The threshold ramps from −2 dB to −9 dB as EXTREME engages; the maximum reduction is 4.5 dB.
• Attack is instantaneous; release is a one-pole at 85 ms.
• The gain, call it g, is applied only to the pre-emphasis contribution: out = dry + (emphasised − dry) · g. The baseband signal is untouched, so the stage removes emphasis overshoot rather than ducking the programme.

4.13 FM CLIPPER (cell 12)

4.14 DE-EMPHASIS (cell 13)

• The same first-order section as §4.11 with the time constant in the denominator (τ₁ = 0, τ₂ = 50 µs) — the exact inverse of the pre-emphasis, so the pair is tonally neutral end to end and only the density added in between survives.
• Switched independently of the pre-emphasis.

4.15 BASS CLIPPER (cell 14)

A split-and-replace design rather than a full-band clipper.

• The input is low-passed at the preset's input corner (80 Hz, Q 0.707) per channel to extract the low band.
• The low band is driven by the preset's drive and shaped by the same soft shaper as the FM clipper, with a fixed ceiling of 0.92 regardless of preset and the preset's knee. The preset's clipper ceiling applies only to the main clipper.
• The shaped band is low-passed again at the preset's output corner (145–150 Hz) and substituted for the original: the extracted low band is subtracted from the full-range signal and the shaped, re-filtered band is added in its place.
• Metering release 90 ms.
• Switched off, it returns the input untouched and its filters do not run at all.
• EXTREME (only while the stage is on): drive +1.2 dB, and the knee moves toward the harder of the preset knee and 0.32.

4.16 OUTPUT HPF (cell 15)

• Second-order high-pass biquad, Q 0.707, 30 Hz on every one of the nine presets. No preset changes it, and EXTREME does not touch it.
• Switched off, the filter still runs and its output is discarded.

4.17 15K FINAL (cell 16)

A linear-phase FIR low-pass, running in normal mode at the preset's corner. It is not an EXTREME feature.

4.18 TRUE PEAK LIMIT (cell 17)

4.19 Output ballast (cell 18)

A pure delay with no processing, sized to pad whatever the chain's stages actually cost out to the declared 8 ms. See §10.

5. The five-band crossover

6. Preset reference

Nine presets. A preset is not a mix control: it rewrites the AGC, the compressor, the low shelf, all five bands, the stereo stage, both clippers, the output filters and the final ceiling, and on some presets it switches whole stages off.

The values below are the effective values the engine runs with. Where a preset does not specify a value it runs the engine's own default, and the default is what is shown.

6.1 AGC

CLASSICAL's maximum lift of 0 dB is deliberate: its AGC can only turn programme down, never raise a quiet passage.

6.2 Post compressor

Remember that the attack and release figures are applied twice in series (§4.5).

6.3 Dynamic low shelf

6.4 Band linkage and the fast path

The fast path's timing is the same on every preset: 0.35 ms attack on all five bands, and releases of 120 / 100 / 80 / 65 / 55 ms for bands 1 to 5. Only the threshold above varies, and only on HOT, DANCE and CLASSICAL.

6.5 Multiband slow path, per band

Bands are B1 < 120 Hz, B2 120–420 Hz, B3 420–1600 Hz, B4 1600–5200 Hz, B5 > 5200 Hz. Each cell reads threshold dB · ratio · attack ms · release ms · post trim dB.

6.6 Stereo

The side low-pass at 9 kHz, the mid high-pass at 40 Hz and the 0.30 correlation floor are the same on all nine presets.

6.7 Clipper and bass clipper

Ceilings are linear, not dB. Every preset sets the clipper's pre-emphasis to zero: EXTREME is the only source of pre-emphasis. CLASSICAL's negative drive means its clipper attenuates before shaping, which is how it stays out of the way. The bass clipper's own shaping ceiling is fixed at 0.92 on every preset; the ceiling column above applies to the main clipper only.

6.8 Output filters and final limiter

The output high-pass is 30 Hz on all nine presets. No preset changes it. The final low-pass corner shown is the requested value before quantisation to one of the four kernels; as it happens each of these lands exactly on a kernel.

7. What EXTREME changes

EXTREME is a single switch that becomes a smoothed scalar between 0 and 1. It moves one step per sample toward its target, a linear 150 ms ramp in both directions, and every change below is scaled continuously by it rather than switched. The resulting settings are recomputed once every 64 samples, so during a ramp at 48 kHz the affected filters are redesigned roughly 750 times per second.

Unchanged by EXTREME: the crossover frequencies, band linkage, multiband attack times, fast-path release times, band post trims, the dynamic low shelf in its entirety, the input high-pass, the output high-pass, the clipper's ceiling and knee, the bass clipper's corner frequencies, the true-peak ceiling and release, and the latency.

How the pre-emphasis is applied. The 50 µs constant goes into the numerator of the first-order section described in §4.11, giving a rising high-frequency characteristic. The HF limiter then holds down only the energy that characteristic added, the clipper does its work in that emphasised domain, and the de-emphasis — the same section with the constant in the denominator — puts the tonal balance back. What survives the round trip is density, not brightness. Because the constant is scaled by the EXTREME ramp rather than switched, the whole sequence fades in and out over 150 ms.

What EXTREME does not do. It does not switch clipping on: every preset clips in normal mode. It does not switch the final low-pass on: that runs in normal mode at 15, 16 or 16.5 kHz depending on preset, and EXTREME only pulls the requested corner down toward 15 kHz. And because there are only four kernels, that pull is not a continuous sweep — on a 16.5 kHz preset it is two discrete kernel morphs.

8. AUTO INPUT

AUTO INPUT is a slow servo on the input trim. It does not watch the programme level directly and it does not aim for a particular peak reading. It watches how hard the AGC is having to work, and moves the trim to leave the AGC near the middle of its range.

What it watches

1. A peak envelope of the raw plugin input, taken before the input trim: instantaneous attack, one-pole release with a 500 ms time constant.
2. The AGC's current gain — how much correction the AGC is applying right now.

How it decides

Rate limits, in dB per second, chosen by how far outside the deadband the AGC is:

Raising is always slower than reducing, at every tier, so the servo backs away from trouble faster than it walks into it.

SETTLING and LOCKED. A lock counter accumulates only while the source is above −55 dBFS and the AGC correction is within ±1.0 dB. Any decision that fails either test resets it to zero. The display reads LOCKED once the counter passes 4 seconds of continuous in-deadband, above-threshold operation, and SETTLING at any other time while AUTO is on.

The LOW warning. With AUTO engaged, the source counts as unrecoverably quiet when it is below −38 dBFS and the gap between the AGC's target and the source exceeds the whole available budget (18 dB of trim plus the preset's maximum AGC lift) by more than 1 dB. With AUTO off, the threshold is −42 dBFS and the budget is the manual trim plus the maximum lift. The flag latches only after 6 seconds of that condition, and clears only after 3 seconds of continuously recoverable audio, so it neither flickers nor lies.

Engaging and disengaging. Engaging anchors the servo at whatever gain is currently applied and resets the lock counter, so there is no jump. Disengaging copies the servo's position into the manual trim and holds it there; the hold is released the moment the Input parameter is moved by more than 0.05 dB. The engine never writes the parameter itself — the trim actually in force is published to the display instead, which is why the knob can read 0 while the display reads −7.

9. Metering and display

The display has six graph pages — gain control, multiband, stereo, level history, LF shelf and the band-limit filters — plus a separate chain window that lays out all nineteen cells in order. Audio keeps processing while the chain window is open.

Every value below is a read-only tap. None of them is fed back into the audio path. The display polls all of them at 30 Hz and freezes its readings while bypass is active.

The four FM cells — pre-emphasis, HF limiter, clipper and de-emphasis — all read the same post-clipper measurement, so their signal bars are identical by construction. Only their gain-reduction overlays differ.

10. Latency

The declared latency is exactly 8 ms at every sample rate, rounded to the nearest sample. Four elements delay the signal; a ballast delay makes up whatever is left over.

The clipper's correction filter is fixed at 33 taps at twice the sample rate, so its 8-sample contribution is the same at every rate. The final FIR's group delay is max(8, round(fs · 0.0016)).

It never varies. Not with preset, not with EXTREME, not with any stage switch. Every stage that could otherwise shorten the chain is disabled by changing its settings, by discarding its output, or by selecting a flat kernel — never by being skipped — and the ballast absorbs the remainder. Host bypass is matched by a dry delay of exactly the same length and crossfaded over 20 ms, so bypassing does not shift the timing either.

At 8 ms, monitoring live through the plugin is a very slight delay in headphones; for a stream or a mix bus it is nothing, and the host compensates for it automatically.

11. Parameter reference

Twenty-two parameters, all automatable. Every switch defaults to on except AUTO INPUT, EXTREME and BYPASS.

No parameter is skewed; the single continuous parameter is linear. There is no output gain parameter, no per-band parameter, and no threshold, ratio, attack, release, frequency or ceiling parameter of any kind. Every processing value comes from the preset tables in §6, modified only by EXTREME.

12. Notes and limitations

• There is no output gain. The Input trim is the only gain control in the plugin. Output level is set by the preset's true-peak ceiling and by everything upstream of it.
• Three rack switches can only disable, never enable. The Bass Clipper, Post Compressor and Multiband switches are combined with the preset's own setting, so on a preset that ships with the stage off the switch does nothing at all: the bass clipper on NATURAL, TALK and CLASSICAL, the post compressor on CLASSICAL. The dynamic low shelf is likewise off on TALK and CLASSICAL. The preset has the final say.
• Switching the AGC stage off does not stop the AGC. The switch discards the stage's output; the AGC keeps running, its gain keeps moving, and AUTO INPUT keeps servoing off that gain. Leave AUTO INPUT engaged with the AGC stage switched off and AUTO INPUT is tracking a gain figure that is no longer reaching the audio.
• The AGC's attack and release are rate limits, not time constants. They are in dB per second, the decision is made once per 64 samples, and "attack" governs downward movement while "release" governs upward. Its gate restrains only upward movement.
• Stage switching is not sample-accurate for five stages. The post compressor, multiband, bass clipper, final low-pass and true-peak limiter are switched on the 64-sample settings block rather than at the instant you click, so a toggle lands up to about 1.3 ms later at 48 kHz.
• The four FM cells share one signal measurement. Pre-emphasis, HF limiter, clipper and de-emphasis all read the same post-clipper tap, so their activity bars are identical by construction and cannot be compared with one another. Their gain-reduction readouts are separate and are real.
• The true-peak detector is an estimate. It is a 2× linear-interpolation inter-sample peak, not a 4× polyphase measurement, and the ceiling it aims at is clamped to the −6 … 0 dBFS range whatever is asked of it.
• Clipping is always running. Every preset clips in normal mode, at its own drive. The FM switches change the clipper's settings rather than adding or removing it; switching the clipper off removes only the EXTREME drive bump.
• The final low-pass is always running, in normal mode, at 15, 16 or 16.5 kHz depending on the preset. Only the pre-filter is EXTREME-only, and the final corner is quantised to four fixed kernels, so EXTREME's move is discrete rather than a sweep.
• The output high-pass is 30 Hz on every preset, and EXTREME does not move it.
• Width cannot narrow. It is boost-only, applied to a side signal band-limited at 9 kHz, and scaled by correlation — below 0.30 correlation it contributes nothing. Even at 0 dB the stage is not a no-op: the 40 Hz mid high-pass and the mono-bass side high-pass still run.
• The fast multiband path is effectively preset-independent. Its 0.35 ms attack and its per-band releases are the same on all nine presets; only its threshold varies, and only on three of them.
• Band post-trims sit after the band ceiling, so a positive trim deliberately pushes a band back above the level the fast path just held it to. The final limiter, not the band ceiling, is what guarantees the output.
• The band-linkage readout is the preset's figure, not a measurement of how linked the bands are at that instant.
• Crossover frequencies are fixed at 120, 420, 1600 and 5200 Hz. Nothing in the plugin changes them.
• The dynamic low shelf is not a shelving filter and is completely unaffected by EXTREME.
• Latency is a flat 8 ms at any sample rate and never varies with preset, EXTREME or any stage switch. A ballast delay pads it.
• Bypass keeps processing. Both the plugin's bypass and the host's run the entire chain underneath and crossfade to a latency-matched dry path, which costs the same CPU as running normally but means nothing has to settle when you switch back.
• Mono and stereo only, with matching input and output layouts.

See also

• From FM to web
• TRANSMIT Tower — the plugin these chapters describe