downward vs upward compression: the technique nobody talks about
downward compression turns loud things down. upward compression turns quiet things up. the difference reshapes how your mix breathes — and one of these is the step before your compressor, not your compressor itself.
the half of compression most producers never use
every compressor you have ever loaded — LA-2A, 1176, FabFilter Pro-C 3, the SSL bus comp on the master — does the same thing in spirit. it watches the signal. when the signal gets too loud, it turns the signal down. that is downward compression.
there is another half of the dynamic range. signals below a threshold can be turned up. that is upward compression, and it is doing the exact opposite job: rescuing detail that is sitting too quiet to matter, without touching the loud parts that are already where you want them.
most producers spend their entire career using only downward compression and call it “compression.” it is not their fault. nearly every famous compressor is downward only. the few plugins that offer upward modes either bury them in a menu or get the noise problem wrong. but once you hear what upward compression does on the right material, you cannot un-hear it. (this is the part most tutorials skip.)
key takeaway
downward compression: signal exceeds threshold → turn it down. upward compression: signal falls below threshold → turn it up. they are mirror operations. one controls peaks, the other recovers detail. you need both for some material — and one is not a substitute for the other.
downward compression: what you already know
downward compression is the default. you set a threshold (say -18 dBFS), a ratio (say 4:1), and an attack and release. when the signal exceeds -18 dBFS, the compressor reduces gain until the output crosses the threshold by 1 dB for every 4 dB of input above it.[^1]
the result is dynamic range reduction from the top down. peaks get clipped down. the average level stays the same or rises (with makeup gain). the perceived loudness goes up. the plugin works because the loudest parts are now closer to the average.
every classic compressor character — VCA glue, opto smoothness, FET attack, vari-mu warmth — is a downward compressor with a particular gain stage and a particular timing curve. SETTLE in KERN PUSH is downward. the SSL bus comp is downward. the Pro-C 3 default is downward. the kick-snare-vocal compression chain everyone learned from YouTube is downward all the way down.
downward compression has one structural limitation: it can only make a signal less dynamic. if the dynamic range above the threshold is the part you want to keep (drum transients, vocal phrasing peaks), and the part you want to address is the quiet sections (the ambient tail between phrases, the room reverb on a snare hit) — downward compression cannot help you. it does not see those signals. they are below the threshold.
upward compression: what you have probably never used
upward compression is the inverse operation. you set a threshold (say -32 dBFS), a ratio (say 4:1), and an attack and release. when the signal drops below -32 dBFS, the compressor adds gain until the output sits closer to the threshold than the input does.
the math runs the other direction. for a signal that sits X dB below the threshold at a ratio of R:1, the boost applied is X - (X / R). the compressor lifts low-level material toward the threshold without touching anything above it.
the result is dynamic range reduction from the bottom up. the quietest parts get louder. the loudest parts stay where they were. the peak-to-average ratio narrows, but the peaks themselves stay intact — and that is the point. transient definition is preserved. detail comes up. the mix gets denser without sounding squashed.
(once you hear this on a sparse drum room recording, you cannot unhear it.)
what upward compression actually fixes
four problems show up over and over in mixing that downward compression cannot solve:
- room tone disappears in the verse and reappears in the chorus. the dry recording has consistent room sound, but you compressed the chorus harder, and now the chorus sounds direct while the verse sounds ambient. upward compression on the chorus lifts the room back up.
- piano sustain dies before the next note arrives. the attack is loud, the release tail is quiet. downward compression squashes the attack. upward compression lifts the tail. these are different fixes.
- hi-hat air gets buried in dense arrangements. the 8-12 kHz region of the hat is right at the noise floor of the rest of the mix. an air-band upward boost (per-band, narrow target) brings the shimmer up without lifting any unwanted noise from elsewhere.
- breath and consonant detail on intimate vocal performances. the singer’s “h” and “f” sit 18 dB below the syllable. compressing the syllable does not help. lifting the breath does. (this is what makes a Bon Iver vocal sound like a Bon Iver vocal.)
key takeaway
downward compression makes loud things less loud. upward compression makes quiet things less quiet. these are not the same job, and you cannot do one with the other. if your problem is “the chorus has more energy than the verse,” downward compression on the chorus is the answer. if your problem is “the verse has too little detail,” upward compression on the verse is the answer.
why upward compression is so rare
every major dynamics plugin is downward only. there are reasons.
upward compression has to reach further down the dynamic range than downward compression reaches up. if you boost a signal that sits 30 dB below the threshold by 6 dB, you also boost the noise floor sitting at -90 dBFS by 6 dB. on a clean digital recording this is fine; on a tape transfer or a guitar amp DI, it is brutal. the plugin has to know what is signal and what is noise — and broadband detectors cannot make that distinction.
the only version of upward compression that holds up across real material is per-band upward compression. the plugin decomposes the signal into many frequency bands (40 perceptually-spaced ERB bands in KERN PUSH, 20+ linear bands in some others), runs an independent upward compressor on each band, and only lifts bands where useful detail lives. silent bands stay silent because the per-band detector does not see signal there. dense bands get lifted because the detector sees something to work with.
this is a different architecture from a broadband upward compressor. it is also why “upward compression” is rare on plugin spec sheets: the engineering required to get it right at scale is not free, and most companies pick the safer downward-only product.
the noise floor problem in upward compression
all upward compressors face the same engineering problem: how do you boost low-level detail without boosting the noise floor underneath it? two strategies dominate. the first is a hard noise-floor parameter — anything below -90 dBFS gets ignored regardless of detector state. the second is per-band gating — bands where the spectral envelope sits below the floor get bypassed entirely. KERN PUSH uses both: a fixed -96 dBFS floor combined with a per-band envelope detector, so silence stays silent and only material above the floor sees lift. broadband upward compressors without these guards amplify hiss audibly.
per-band upward as the bridge
this is where upward compression stops being a curiosity and starts being a category. a per-band upward compressor is doing something the rest of your signal chain cannot: it is processing 40 different decisions at once, each sensitive to its own frequency region’s dynamic behaviour, and lifting the bands where you have signal worth lifting.
DRIFT in KERN PUSH is a per-band upward compressor with three architectural commitments worth naming:
- ERB-spaced bands. the 40 bands follow Glasberg and Moore’s ERB scale,[^2] which means narrow bands where your ears localize detail (presence, sibilance, air) and wider bands where they do not (sub-bass, low-mid). lifting the air band lifts the air; it does not lift the low-mid.
- Spectral tilt bias. the per-band threshold tilts upward at high frequencies (+4 dB at 20 kHz, -4 dB at 20 Hz, scaled by AMOUNT). air gets more lift than mud at the same knob position. you do not have to manage the tilt by hand.
- History-dependent release (T4 thermal memory). bands that have been actively compressing accumulate “history” and release more slowly the next time, just like an LA-2A’s optical cell warming up. cold material gets fast release, repeatedly worked material gets slow release. the plugin sounds different in the chorus than it sounds in the verse, automatically.
three architectural commitments, one knob (AMOUNT), one mode pick (DRIFT). this is why “the step before your compressor” makes sense as a position: DRIFT recovers detail before the broadband compressor flattens it. once detail is gone, no amount of downward compression will bring it back. the order matters.
upward vs parallel compression: similar feeling, different mechanism
a question that always comes up: “isn’t this just parallel compression?” no. they aim at a similar feeling — quiet detail rising up — but the mechanism is different and the trade-offs are different.
parallel compression sends the signal through a heavily downward-compressed bus and mixes that bus underneath the dry signal. the dry transients pass through full strength; the compressed copy adds density. it works, but the compressed copy contains every signal — including any noise, mouth clicks, or low-level garbage that the downward compressor lifted toward the average level. you also need at least two channels and a bus.
upward compression does the lift directly, in one instance, without parallel routing. it lifts what is already in the signal toward the threshold, but only the bands or moments where lift is wanted (controlled by attack, release, and the per-band detector). transients are unaffected because they sit above the threshold and the upward compressor does not act there.
they are not the same. parallel works on a broadband copy; upward works on the original, surgically. for sparse material with detail-rich quiet sections, upward is cleaner. for dense rock drums where you want a “bigger” feel, parallel is the classic move.
(both are valid. the cliché “parallel compression makes drums bigger” is true. it is also incomplete — sometimes you wanted detail recovery, not bigness, and parallel was the wrong tool for that.)
when each one is the right answer
| problem | use this |
|---|---|
| chorus too loud relative to verse | downward, on the chorus |
| verse too thin relative to chorus | upward, on the verse |
| drums too dynamic for the genre | downward, on the bus |
| drums sound clinical, no room | upward, on the room mics |
| vocal phrases spike +6 dB | downward, on the vocal |
| vocal breath and consonants buried | upward, on the vocal |
| hi-hat air vanishing in dense mix | per-band upward on 8-12 kHz |
| sustain tail of piano dying early | upward, slow release |
| sub-bass uneven across phrases | downward with sidechain HPF |
| mix bus glue without pumping | downward, gentle ratio (SETTLE) |
| mix bus needs detail not depth | upward, per-band (DRIFT) |
start by asking: is the problem too much loud, or too little quiet? the answer points to which compressor type you actually need. the rest is taste.
the order matters
this is where “the step before your compressor” earns its name. if you put downward compression on a track first and then try upward compression after, you have already collapsed the dynamic range. the quiet detail you wanted to recover is now compressed against a louder average; lifting it brings up the wrong things.
upward goes first, when ordering both. lift the detail. then compress for level. the result is denser-feeling material with all of its original character intact. flip the order and you get a flatter, less natural result.
key takeaway
ordering rule: upward before downward, when both are in the chain. upward compression recovers detail that downward compression would otherwise crush against the average. once detail is gone, no plugin gets it back. order matters.
heads up
upward compression is unforgiving on noisy material. if your source has audible hiss, hum, or low-level garbage, fix that first (with a good gate or a careful EQ HPF) before adding any upward compression. otherwise you are amplifying the problems along with the music.
frequently asked questions
frequently asked questions
what is the difference between downward and upward compression?
downward compression turns loud signals down once they cross a threshold. upward compression turns quiet signals up when they sit below a threshold. they affect opposite halves of the dynamic range. downward kills peaks; upward recovers detail. most "compressors" are downward only. upward compression is a separate, less common technique that pulls up ambience, room tone, sustain tails, and high-frequency air without pumping the loud parts.
when should I use upward compression?
use upward compression when the loud parts of your signal are already where you want them, but the quiet parts are getting lost. ambient room sounds in a sparse drum recording, the decay tail of a piano sustain, the breath between vocal phrases, the high-frequency air on a hi-hat that disappears in dense sections. upward compression rescues that detail without you having to ride a fader.
is parallel compression the same as upward compression?
no, but they aim at a similar feeling. parallel compression mixes a heavily downward-compressed copy under the dry signal so the quiet parts get lifted relative to the loud parts. that is one way to add presence without crushing transients. upward compression does it directly — it lifts the quiet parts in a single instance, no parallel routing required, and you keep transient definition because the loud parts are not being compressed at all.
does upward compression add noise?
it can, if you boost too far below the noise floor. all upward compressors need a noise gate or floor parameter that prevents the plugin from amplifying hiss. KERN PUSH uses a fixed noise floor at -96 dBFS combined with a per-band detector — the gain only lifts material above the floor, so silence stays silent and tape noise stays at tape noise.
why is upward compression so rare?
because broadband upward compression is hard to get right. the gain has to reach further down the dynamic range than downward compression reaches up, so any noise or low-level garbage gets amplified along with the detail you wanted. spectral or per-band upward compression is the version that works — it lifts only the bands where useful detail lives and leaves the rest of the spectrum alone. that constraint is why most plugins do not offer it at all.
references
a note from the developer
i did not understand upward compression for embarrassingly long. i had read the chapter in DAFX three times. i could explain the math at a whiteboard. and i still did not hear what it did until i loaded the original Waves Renaissance Vox on a sparse acoustic vocal in 2020. the breath came up. the room came up. the takes that had felt distant suddenly felt present. i had not changed the gain. i had not added reverb. i had recovered detail that was already there.
the moment that broke my mental model was realising that “compression” had been the wrong word for two completely different operations my entire career. downward compression makes loud things less loud. upward compression makes quiet things less quiet. they are not variants of one technique; they are mirror operations, and most of us only own the downward half because that is what got famous.
building DRIFT inside KERN PUSH was where this got hard. broadband upward compressors are noisy on real material — that is why they are rare. spectral upward across 40 perceptually-spaced bands works, but the engineering is unforgiving. the noise floor has to be hard. the per-band detector has to be honest about what is signal and what is silence. the spectral tilt has to lift air more than mud at the same knob position, otherwise it sounds wrong on every bright source. and the release has to remember what it just did, or the plugin pumps in unmusical ways.
if i have done DRIFT right, you will not notice the per-band logic. you will notice that detail comes up where it was missing, transients stay intact, and the noise floor stays at the noise floor. that is the only review i actually want.
if you have tried upward compression and have a take i did not cover here, send it. jonas@kernaudio.io. the best guides come from producers who heard something i did not.
try it yourself
KERN PUSH: three compression characters across 40 spectral bands. $29, no iLok, no subscription.