mix bus compression without pumping

the line between glue and pump

mix bus compression is famous for one thing: making a stack of individually mixed tracks feel like a single performance. it is also famous for one bad outcome: pumping, where the kick drum triggers gain reduction across the whole mix and the rest of the song breathes in time with the kick whether you wanted it to or not. (you know the move. you have been here.)

the line between glue and pump is narrower than most tutorials admit. the controls that produce glue are the same controls that produce pump — the difference is small numbers in opposite directions. this guide breaks down how each control behaves on a full mix, why broadband bus compression has a structural pumping problem, and where spectral bus compression fits in 2026.

i make KERN PUSH, a spectral compressor with three named characters. SETTLE is the bus-compression character, and the way SETTLE handles transient material on the mix bus is the reason this article exists. but most of what is below applies to any bus compressor in your DAW, including the SSL-style plugin you might already be running.

what bus compression actually does

bus compression solves two different problems with the same control set, and confusing them is where most pumping starts.

job one: glue. glue is the perceptual effect of multiple instruments breathing together. when the snare hits, the rest of the mix ducks 0.5-1.5 dB, then recovers in time for the next phrase. that micro-coupling makes the listener hear “one performance” instead of “a stack of recordings.” 1-2 dB of gain reduction on the loudest hits, slow ratio (1.5:1 to 2:1), slow release. you should not hear the compressor working — you should hear the result.

job two: peak control. peak control is a different job: keep the loudest moments below a fixed threshold so the limiter or master-bus chain downstream has less work to do. 3-6 dB of gain reduction on peaks, faster ratio (3:1 to 4:1), faster release. you might hear this one working, depending on the genre.

doing both in one plugin set aggressively (5+ dB of gain reduction with fast attack and a 4:1 ratio) is the recipe for audible pumping. the plugin is being asked to do glue and peak control simultaneously, and the settings that make peak control fast are the same settings that make glue audible.

the four controls on a full mix

the four controls (threshold, ratio, attack, release) behave differently on a full mix than they do on a single track. the difference is mostly about attack and release timing.

threshold on the mix bus

threshold sets where the compressor starts acting. on a single track you usually set threshold to taste, by ear, watching gain reduction. on the mix bus you have a clearer signal: total integrated loudness. you want the compressor to act on the loudest 10-25% of the program material — kick drum and snare hits and louder vocal phrases — and ignore the rest.

a useful starting point: set threshold so the meter shows 1-2 dB of gain reduction on the loudest sections (chorus or drop). that means quieter sections (intro, breakdown) sit below the threshold and the compressor is idle. this is glue territory.

if you want peak control instead of glue, lower the threshold until you see 3-6 dB on peaks. then expect to manage release timing carefully — that depth of gain reduction starts to be audible.

ratio on the mix bus

ratios above 2:1 on the mix bus push the compressor toward audibility fast. an SSL-style 2:1 with 1-2 dB of gain reduction is nearly transparent. the same plugin at 4:1 with 3 dB of gain reduction is audibly compressing. above 4:1 on the mix bus is rare outside of explicit “smashed” pop / EDM aesthetics.

useful ratios:

• 1.5:1: barely there glue. you hear cohesion, not compression. perfect for acoustic, jazz, classical-adjacent material.
• 2:1: the SSL bus comp default. the canonical “rock and pop glue” setting. transparent at 1-2 dB GR.
• 3:1: noticeable compression. genre-dependent. fits some hip-hop and dense rock mixes.
• 4:1: aggressive bus compression. peak control mode, not glue. set with care.
• above 4:1: limiter territory. usually wrong on a mix bus.

attack on the mix bus: the part that pumps

attack determines how fast the compressor responds when signal exceeds threshold. on a single track, fast attack catches transients. on the mix bus, fast attack catches the kick drum — and ducks every cymbal hit and vocal syllable that overlaps with the kick.

slow attack on the mix bus is almost always right. 10-30 ms is a normal range. the attack is slow enough that the kick transient passes through largely intact (kick definition stays); compression engages on the kick body (the 80-200 Hz region), which is what you wanted to compress anyway.

an attack faster than 5 ms on the mix bus is a specific aesthetic choice (very fast, very aggressive, very obvious) and is the most common cause of pumping when used unconsciously.

release on the mix bus: the part that decides if pumping is audible

release determines how fast the compressor stops compressing after signal drops below threshold. this is the single most important control for whether bus compression sounds like glue or sounds like pump.

think in time relative to the song’s tempo. at 120 BPM, a beat is 500 ms; a half note is 1000 ms; a bar is 2000 ms. for the compressor to feel like glue, you want release somewhere between a half note and a bar — long enough that recovery happens between phrases, not between beats. that is 500-1500 ms in absolute terms.

a release shorter than a beat (faster than 500 ms at 120 BPM) means the compressor is recovering between drum hits. your ears can track that. that is pumping.

an “auto” release (program-dependent, like the SSL bus comp’s auto setting) is the safest default for most material — the plugin lengthens release for sustained content and shortens it for transient content automatically.[^1]

why broadband bus compression has a structural problem

here is the thing nobody mentions in tutorials: broadband bus compression treats the entire mix as one signal. one envelope follower watches the full-spectrum amplitude. one gain stage applies the same gain reduction to every frequency from 20 Hz to 20 kHz simultaneously.

that means: when the kick drum at 60 Hz pushes the envelope follower over threshold, the compressor reduces the cymbal at 12 kHz by exactly the same amount as it reduces the kick. the cymbal had no reason to be reduced. it was not loud. it was not over threshold. but it ducked anyway, because the broadband detector did not know the difference.

this is why aggressive bus compression on a heavy mix produces the “pumping cymbals” effect: the high frequencies are losing energy in time with the low-frequency transients. on transparent settings (1-2 dB of gain reduction), the cross-frequency coupling is small enough to feel like glue. on aggressive settings (4+ dB), it is audible and unmusical.

multiband and spectral compressors solve this by splitting the signal into bands. the kick triggers gain reduction in the low band; the cymbal stays where it was. broadband cross-talk goes away.

(this is the part most tutorials skip. it is not the producer’s fault when broadband bus compression pumps; it is the architecture making the trade.)

spectral bus compression: a different option in 2026

per-band spectral compression operates on hundreds or thousands of frequency bins simultaneously, with each region having its own envelope follower and gain stage. when the kick triggers compression, the kick band ducks; the cymbal does not.

i wrote a separate guide on the architecture: read how spectral compression works for the technical details. on the mix bus specifically, the practical implications are:

• no cross-frequency pumping. the kick stops ducking the high end. you can run more gain reduction without it sounding squashed.
• each band gets its own ratio implicitly. a sustained pad triggers gentle compression because the pad’s envelope is slow. a sharp transient triggers fast compression in its own band. the mix bus stops sounding like one compressor and starts sounding like 40 of them, each doing the right thing.
• bus glue stays. the perceptual effect of the mix breathing together comes from coupled gain reduction across bands. spectral compressors that run “per-band only” lose the glue. plugins like SETTLE in KERN PUSH add broadband GR coupling — energy-weighted broadband compression blended into the per-band gain decisions. you keep the glue without the cross-frequency pumping. the coupling factor in SETTLE is 60% (heavily coupled, glue-leaning), in TIGHT is 40%, and in DRIFT is 15% (mostly per-band).[^2]

the trade-off is latency. all FFT-based spectral compressors add latency proportional to FFT window size. a 4096-point FFT at 44.1 kHz adds ~93 ms. that is fine for mixing (DAW latency compensation handles it) and unsuitable for tracking through. it is a deliberate architectural choice.

practical recipes

three starting points for different bus-compression aesthetics. read these as starting points, not finished settings — the right value for your mix is what your ears tell you, and the only real test is bypass-A-B against the mix without it.

transparent glue (most pop and rock)

• ratio 2:1
• threshold for 1-2 dB GR on loudest sections
• attack 10-30 ms (slow)
• release auto, or 200-400 ms (longer than a beat)
• gain reduction meter rarely above 2 dB

target plugin: SSL-style bus comp, or SETTLE in KERN PUSH at 25-35% AMOUNT.

gentle peak control (modern bus chain)

• two stages
• stage 1: SSL-style at 2:1, 1.5 dB GR, attack 30 ms, release auto
• stage 2: FET-style at 4:1, 1.5 dB GR, attack 5 ms, release 50 ms
• total gain reduction: 3 dB across two stages
• each stage is doing less than you think — that is the point

target combination: any 2:1 glue compressor + any FET-style peak compressor, in series.

dense pop / EDM aesthetic (intentional pumping is the sound)

• ratio 4:1
• threshold for 4-6 dB GR on every kick
• attack 5-10 ms (catches transient body, not the very front)
• release tempo-locked to a quarter note (this is the pump)
• pumping is the feature here, not a bug

target plugin: any character compressor where the pump is genre-correct (1176-style FET, or TIGHT in KERN PUSH at 60-75% AMOUNT).

ordering: bus compression is rarely first

bus compression sits late in the mix-bus chain, not first. typical order:

1. mix prep / clean-up (resonance suppression, stereo correction, low-end management) — these reduce the dynamic range the compressor has to handle. a mix that has had its problem resonances tamed pre-bus needs less compression to feel cohesive. read before the compressor for the full mix-prep argument.
2. glue compression (this article — 1-2 dB at 2:1)
3. bus EQ (broad tonal balancing, shelf moves)
4. bus saturation (cohesion through harmonic glue, not gain reduction)
5. limiting (the last gain stage before the limiter at the master output)

(running the bus compressor first, before any clean-up, makes its job harder. resonances that were going to trigger gain reduction get cleaned up; transients that were going to pump get controlled stereo-side.)

frequently asked questions

frequently asked questions

what is mix bus compression supposed to do?

mix bus compression has two jobs that get confused. the first is glue: making the individual tracks feel like one performance instead of a stack of separate recordings. the second is loudness control: keeping peaks under a target level so the limiter has less work to do. you can do both in one plugin or split them across two. doing them with one plugin set too aggressively is the most common cause of bus pumping.

why does my mix bus compressor cause pumping?

pumping is what you hear when the gain reduction is fast enough to be audible against the music but not fast enough to be transparent. the kick triggers the compressor, the whole mix ducks, then it recovers between hits at a rate your ears can track. fix it by slowing the release (so recovery happens between phrases instead of between beats), reducing the threshold so less gain reduction is needed, or switching to spectral compression so the kick does not duck the cymbal.

what attack and release should I use on the mix bus?

a common starting point is 10-30 ms attack and auto-release (or 100-300 ms manual release). the attack is slow enough to let transients through; the release is slow enough that recovery is between phrases, not between drum hits. an SSL-style auto-release detects program material and adjusts. if your compressor lacks auto-release, set release by tempo: longer than a half note for slow material, shorter than a quarter note for fast material.

should I use a single bus compressor or stack two?

stacking is common. two compressors each doing 1-2 dB of gain reduction is more transparent than one doing 3-4 dB, because each compressor is operating in its most linear range. a typical stack: a slow-ratio glue compressor (SSL-style, 2:1, slow) followed by a faster character compressor (FET-style, 4:1, fast) for transient bite. the order matters — glue first, then bite — and each one should do less than you think.

is spectral compression better than broadband on the mix bus?

better at avoiding pumping, yes. broadband bus compression treats the whole mix as one signal: when the kick exceeds threshold, the entire mix ducks, including the cymbal that had no business ducking. spectral compression operates per frequency band, so the kick triggers gain reduction in the kick band only and the cymbal stays where it was. the result is bus glue without the cross-frequency pumping that broadband causes. the trade-off is latency (FFT-based plugins add ~93 ms) and a slightly different "feel" — broadband bus compression has a cohesive squash that some genres explicitly want.

references

a note from the developer

i mixed records on an SSL bus compressor for years before i had any idea why it sounded like glue. i thought it was the gain stage. i thought it was the transformer. i thought it was the timing. it was actually the math: the auto-release detector lengthens release on sustained content and shortens it on transient material, which means the compressor is making a different decision in the verse than in the chorus, in the bridge than in the drop. the SSL feels alive because its timing is alive.

the part that kept biting me when i was building SETTLE in KERN PUSH was the cross-frequency pumping problem. early per-band prototypes felt clinical: the kick stopped ducking the cymbal, but the mix stopped breathing together. that is when i realised glue is broadband on purpose. spectral compression alone is not bus compression — it is per-band compression. to get glue, the per-band gains have to know about the broadband envelope and partially follow it. 60% coupling in SETTLE is what came out of the listening tests against the SSL plugin running in parallel. less coupling lost the glue. more coupling re-introduced the cross-frequency pump. 60 was the spot for that character.

(60% is also wrong for TIGHT and DRIFT, which is why each character has a different coupling factor. that is the part i could not have arrived at without ripping a Bonnie Raitt mix and a Bon Iver mix and an Aphex Twin record into Ableton and listening for an embarrassingly long time.)

i am a solo developer in Copenhagen. if you have a take on bus compression that took you years to figure out — the kind of thing nobody puts in tutorials — send it. jonas@kernaudio.io. these guides are written by the people who hear things i have not heard yet.