Compression (threshold, ratio, knee, attack, release, make-up, detection, sidechain)

Definition

A compressor reduces gain when the level of the detection signal exceeds a threshold. With levels in dB, the static curve is:

• for x ≤ T: y = x;
• for x > T: y = T + (x − T) / R;

where x is the input level, T the threshold and R the ratio. The gain reduction is GR = (x − T) · (1 − 1/R). With a soft knee of width W, for |x − T| ≤ W/2 we have y = x + (1/R − 1) · (x − T + W/2)² / (2W), which joins the two segments continuously.

The time behavior of the reduction is governed by attack and release. The computed gain is filtered by a one-pole integrator: g[n] = α · g[n−1] + (1 − α) · g_target[n], with α = e^(−1/(τ · fs)), where τ is the attack time constant (when the reduction increases) or the release time constant (when it decreases). In one time constant the gain completes 63% of the change; some manufacturers instead quote the time for 90% or for a 10 dB change.

The detector can be peak (instantaneous absolute value) or RMS (effective value over 5–50 ms); the topology can be feed-forward (the detector reads the input) or feedback (it reads the output). The sidechain is the signal that drives the detector: it can be filtered (high-pass at 80–150 Hz to keep the sub from driving the compressor) or come from another track (ducking). The make-up gain compensates for the average level reduction.

Parameters and units

Starting settings:

Musical function

Compression reduces the difference between loud and soft parts, stabilizes the level of vocals and bass, increases sustain, emphasizes or softens the transient depending on the attack (slow attack: the transient passes through and the body is lowered; fast attack: the transient is reduced), binds the elements of a bus together (glue) and, with the sidechain, creates space between instruments (ducking the bass with the kick). The pumping produced by short releases and high GR is a flaw in acoustic pop and a deliberate effect in house and EDM.

Practical example

Vocal with peaks at −8 dBFS, threshold −20 dBFS, ratio 4:1, hard knee:

• output on the peaks: y = −20 + (−8 + 20)/4 = −17 dBFS; GR = 9 dB;
• a passage at −16 dBFS becomes −19 dBFS (GR 3 dB); one at −24 dBFS stays unchanged;
• the excursion between −24 and −8 dBFS (16 dB) becomes between −24 and −17 dBFS (7 dB).

With a 6 dB knee, at the point x = T the GR is (1 − 1/4) · (3)² / 12 = 0.56 dB: compression begins gradually from −23 dBFS. With a 10 ms attack at 48 kHz, α = e^(−1/480) = 0.99792. Make-up +5 dB to bring the vocal loudness back to the original level. On a bass at 40 Hz the period is 25 ms: a release under 50 ms follows the waveform and produces distortion; with 120 ms the gain stays stable.

Common mistakes

• Attack too fast on percussion: the transient disappears and the track loses impact.
• Release too short on basses and 808s: distortion from gain modulation at audio frequency.
• Judging the compressor without compensating the make-up: the louder signal seems better.
• Constant reduction beyond 8–10 dB with a single compressor: it is better to spread the work over two stages of 3–4 dB.
• No filter on the master bus sidechain: the kick and sub drive the compressor and the mix pumps.
• Setting the threshold by looking at the numeric value instead of the gain reduction meter.

How to check it

• Gain reduction meter of the compressor: average value and peaks consistent with the starting table.
• App analysis: analyze the track or bus before and after compression; Crest factor (peak − RMS) and PLR drop in proportion to the compression of the transients, Loudness Range (EBU 3342) drops with slow compression and with high GR. Comparing the Maximum short-term (3 s) at the same integrated loudness shows the density achieved.
• Loudness-matched comparison: adjust the make-up so that the integrated loudness stays unchanged with and without the compressor.
• Difference signal: inverting the polarity of the compressed output against the aligned original, you hear only what the compressor has removed.