Upward vs downward compression

At a glance

What's the difference between upward and downward compression?

Both kinds of compressor watch the level of a sound and compare it with a threshold. A downward compressor acts on whatever goes over that line and turns it down. An upward compressor acts on whatever falls under it and turns it up. The ratio means the same thing in both: how strongly the level gets pulled towards the threshold.

Pull the loud parts down or push the quiet parts up and the dynamic range, the gap between the loudest and quietest moments, gets smaller either way, so on a meter the two can look alike. They sound different because they change different moments of the sound. A downward compressor is busy during the hits and the loud words; an upward compressor is busy in the gaps between them, in the tails of notes and in whatever noise sits under the music.

Plain "compression" means downward, and the compression lesson covers it from the start. The upward compression lesson goes through each upward control in detail. This page puts the two side by side, and the lab above flips one sound between them with the same threshold and ratio.

The same gap, reached two ways

Say a drum loop has main hits peaking at −3 dB, a tom fill around −12 dB and ghost notes at −33 dB. From hit to ghost note is a 30 dB gap, and you'd like it to be 14 dB.

The downward way: set the threshold to −27 dB at 3:1. The main hits are 24 dB over, which comes out as 8 dB over, so they leave at −19 dB, while the ghost notes are under the threshold and stay at −33 dB. That's a 14 dB gap. Add 16 dB of makeup gain and the hits are back at −3 dB with the ghost notes at −17 dB.

The upward way: set the threshold to −9 dB at 3:1 with at least 16 dB of range. The main hits are over the threshold and don't move. The ghost notes are 24 dB under, which comes out as 8 dB under, so they rise to −17 dB. Hits at −3 dB and ghost notes at −17 dB are the same two numbers the downward version gave.

The tom fill is where they part company. Downward, it's 15 dB over the −27 dB threshold, comes out 5 dB over at −22 dB and goes up to −6 dB with the makeup, which puts it only 3 dB under the main hits. Upward, it's 3 dB under the −9 dB threshold, comes out 1 dB under and leaves at −10 dB, still 7 dB under the hits, close to the 9 dB it started with. The gap that closed is the one below it, from 21 dB down to 7.

Downward compression squeezes the top of the range together, so the loud and fairly loud moments end up near each other. Upward compression squeezes the bottom of the range, so the quiet and fairly quiet moments end up near each other while the loud moments keep most of the distance they had between them.

Upward and downward on one transfer curve

Draw both on a transfer curve, with input level along the bottom and output level up the side, and the two turn out to share a line. An upward compressor at −30 dB and 3:1 with 12 dB of range lifts everything under −48 dB by exactly 12 dB, and lifts the levels between −48 and −30 dB by less and less. A downward compressor at −48 dB and 3:1 with 12 dB of makeup gain does exactly the same to those levels.

The upward compressor's range plays the part of the downward compressor's makeup gain. The curves split at −30 dB, where the upward compressor returns to 1:1 and passes the peaks through untouched, while the downward one carries on at 3:1 all the way to the top, so a hit at 0 dB comes out at −20 dB.

So upward compression keeps the peaks and punch that downward compression would flatten. For the quiet parts the two give the same result on paper. In time they don't quite, because after each loud hit the downward compressor has far more gain reduction to recover from, so the moments just after a hit come up later.

What happens to the peaks and the loudness

A downward compressor lowers the peaks, and the average level comes down only a little with them, so the gap between peak and average, the crest factor, shrinks and leaves headroom above the peaks. Makeup gain spends it, so the sound comes out louder at the same peak level as before.

An upward compressor gets to the same place from below. The peaks stay put and the quiet parts rise, so the average level goes up and the crest factor shrinks without any makeup gain at all. Either way you end up with more loudness for the same peak, and either way the louder version will win a quick comparison, which is why the lab matches the loudness when you press A/B.

Downward compression has one job that upward compression can't do, which is catching peaks. If a vocal has one shouted word 10 dB louder than the rest, an upward compressor can only bring the rest up towards it, and the shout stays exactly as far above the mix as it was. Bringing that word down takes a downward compressor, a limiter or clip gain.

Attack and release land in different places

In this lab, in both modes, attack is how fast the gain moves down and release is how fast it moves back up. The definition doesn't change with the direction, but it lands on different parts of the sound.

On a downward compressor the gain goes down when a hit arrives, so the attack decides how much of the front of the hit gets through first. A slow attack of 20 to 40 ms keeps the transient, which is how engineers add punch with a compressor. A fast attack under 1 ms flattens it.

On an upward compressor the gain goes down when a hit arrives too, because the lift has to back off. A fast attack, 1 to 5 ms, takes the lift away before the hit lands and the peak stays where it was. A slow attack leaves the lift on for the first part of the hit, so the transient gets turned up by however much lift there was in the gap before it. On a drum loop that can push the peaks several decibels above the original and into clipping.

The release works much the same way in both. It decides how quickly the level comes back after a hit, and a long release makes the tail of each hit swell up afterwards. On a downward compressor that's heard as pumping. On an upward compressor it sounds more like the room breathing in after every hit.

Noise, breaths and room sound

Upward compression brings up noise, but so does downward compression with makeup gain, and by the same amount for the same lift. On the mirror curves above, anything under −48 dB comes up exactly 12 dB in both versions, and hiss, hum and headphone bleed are some of the quietest things on a recording, so that's where they sit.

Upward compression does worse with the range turned right up, or on a plug-in with no range control. Then the quietest sounds keep getting lifted further than the music above them, and a noisy recording fills up with noise. Keep the range only as high as the job needs. If the noise still comes up, put a gate or an expander before the compressor to push it back down first.

Breaths need the same care on vocals. They sit under the sung words, often not far below the softest syllables, which is right where an upward compressor works, so they come up too. A downward compressor leaves them alone until you add makeup gain, and then lifts them along with everything else.

Which one should you use?

If you only learn one, learn downward compression. It's in every DAW, it's the only one of the two that can catch a peak, and the problem in most mixes is a loud moment jumping out. Reach for upward compression when the loud moments already sound right and the quiet ones are getting lost under the rest of the track.

Using both at once

Plenty of tools do both. Xfer's OTT is free and runs an upward and a downward compressor on each of three frequency bands at the same time, with Upwd and Downwd knobs to set how much of each you get. It's named after an Ableton Live preset for Multiband Dynamics, which has an Above and a Below threshold on every band, so it can compress downwards from above and upwards from below at once. Waves MV2 has a Low Level control for upward compression and a High Level control for downward, plus an output fader.

Using both squeezes the dynamic range from each end. On a sound-design patch or a bass that's often the point, and the OTT lesson goes through that sound; the multiband compression lesson explains the bands. On a vocal or an acoustic instrument the usual approach is gentler: a downward compressor first to steady the loud moments, then a little upward compression to bring the soft detail up to meet them.

If your DAW's compressors only work downwards, parallel compression is the usual way to get the upward half. A heavily compressed copy mixed under the original adds a lot of level in the quiet parts and very little in the loud ones, so the blend lifts the quiet parts and leaves the peaks close to where they were.

Upward and downward expansion

Upward and downward also apply to expanders, which do the opposite of compressors and widen the gap between loud and quiet. That gives four combinations, and the names trip people up because two of them act under the threshold.

Downward expansion is the one to watch for, because it works in the same place as upward compression and does the opposite. A noise gate is a downward expander with a very high ratio. The expansion lesson and compressor vs expander vs gate cover that side.

How to compare upward and downward compression on your own track

1. Duplicate the trackPut a downward compressor on one copy and an upward one on the other. OTT with Downwd at 0% works as the upward one.
2. Set the downward compressor firstLower its threshold until the loudest moments show 4 to 6 dB of gain reduction, at 3:1 with a 10 ms attack.
3. Set the upward compressor to a similar gapPut its threshold a few decibels under the loud moments and raise the ratio, or the Upwd amount, until the quiet parts sit about as close to the loud ones as they do on the other copy.
4. Match the loudnessBring the faders to the same loudness by ear, or with a LUFS meter on each copy.
5. Solo each in turn and listen to one thing at a timeListen to the front of the hits first, then to the gaps between them. That's where the two differ.

Common mistakes

Assuming upward compression never touches the peaks

It leaves them alone only when the attack is fast enough to take the lift away before each hit lands. With a slow attack, upward compression makes the peaks louder than they were, which can clip a track that was already close to 0 dBFS.

Using upward compression to fix a peak problem

If one note or word is far too loud, lifting everything else towards it makes the whole part louder and leaves that note sticking out. Turn the note down with a downward compressor, a limiter or clip gain first.

Comparing them at different loudness

An upward compressor always comes out louder than the original, and a downward one with too much makeup gain does too. Whichever is louder will sound better at first, so match the levels before you choose.

Mixing up upward compression and downward expansion

Both have a threshold that works on the quiet parts. Ableton Live's Multiband Dynamics uses the same Below ratio for both, with upward compression on one side of 1:1 and downward expansion on the other. If the quiet parts got quieter, you've set up an expander.

Training your ears to hear the difference

Downward compression is easiest to hear on the front of the hits, and upward compression in the gaps between them, so listen to one at a time. The quiz under the lab plays hidden settings at matched loudness. The first four rounds ask whether the compressor is working upward or downward, which comes down to whether the front of each hit keeps its crack; the last four ask how much range an upward one has. Drums, percussion and the acoustic guitar show the difference best, since the vocal and the full mix are loud almost all the time and leave an upward compressor little to lift. Your best score is saved in your browser.

Questions people ask

Keep going