What is linear phase EQ?

Linear phase EQ in plain English

Every EQ is a set of filters, and filters take time. When a normal EQ boosts or cuts a range, the sound in that range comes out slightly out of step with everything else, usually a little later. The steeper and bigger the change, the longer the delay: in the lab, a 48 dB per octave low cut at 120 Hz holds back the frequencies at the cutoff by about 12 ms. This timing change is called phase shift, and EQs that work this way are called minimum phase.

A linear phase EQ draws exactly the same curve and handles time differently. It delays every frequency by the same amount, so nothing in the sound moves against anything else; the whole sound arrives late, and your DAW lines it back up with the other tracks. The name comes from a graph of phase against frequency, which for this kind of filter is a straight line.

Keeping every frequency in step has a price. The filter has to react to each sound before that sound reaches the output, which it can only do by holding the whole signal back, and that's the latency. Any ringing the EQ adds also gets split in two, with half of it arriving before each sharp sound. That second cost, pre-ringing, is the reason linear phase EQ suits a few specific jobs and stays switched off the rest of the time.

Reading the lab

The big screen shows what the EQ does to a single click, the shortest sound there is. A click contains every frequency at once, so what comes out is a picture of the filter itself, called its impulse response. The dashed line in the middle is the moment the click arrives once the latency is taken out, and anything to its left came out before the sound that caused it. The bright shape is the mode you're hearing and the thin line is the other one.

Underneath on the left, the pink line is the EQ curve, measured from that same click, with a dashed line showing the curve the EQ is aiming for. The lane below it shows how much later each frequency comes out than the rest. Minimum phase bends that lane around the frequencies you change, and linear phase keeps it flat.

The middle screen is a close-up of the strongest kick in the loop, lined up against the original. The shaded strip is the 30 ms in front of the hit, which is where pre-ringing appears, and the panel on the right measures how loud that strip is compared with the hit. Both close-ups stretch quiet detail upwards so ringing 40 dB down still shows, which makes small things look bigger than they sound.

A/B switches the EQ out. The original is delayed by the same latency and matched in loudness, so the only thing that changes is the EQ. You can also upload your own sound or record one, and it stays on your device.

What does minimum phase mean?

Almost every EQ you've used is minimum phase, including Ableton's EQ Eight and Logic's Channel EQ. Their filters work like the circuits in an analogue EQ, feeding a little of their own output back into themselves, and of all the filters that could draw a given curve they have the shortest possible delay. Nothing comes out before the sound goes in, and there's no latency for your DAW to make up.

What you pay is the phase shift around each change, and how much depends on how sharp the change is. The lab measures it on the click. A 48 dB per octave low cut at 120 Hz delays the frequencies near the cutoff by 12.3 ms, a 12 dB per octave cut at the same frequency by 2.3 ms at most, and a 3 dB bell at 3 kHz with a Q of 1 by so little the delay lane barely moves.

On a single track you'll rarely hear any of this. A 12 ms delay sounds like a lot, but it only applies to a narrow band of frequencies right at the cutoff, which the filter is already turning down. Phase shift becomes a problem when the EQ'd sound meets another copy of itself, and the section on when to use linear phase covers each of those cases.

How does linear phase EQ work?

A linear phase EQ uses one long filter, known as an FIR filter (finite impulse response), whose response to a click is symmetrical: the second half is a mirror image of the first. That symmetry is what keeps every frequency in step. It's also why the filter needs latency. If the middle of the response is the moment the sound arrives, the first half has to play before it, so the EQ holds the whole signal back by half the filter's length.

The lab builds its linear phase mode the plain way. It measures the minimum phase curve at thousands of frequencies, turns that curve back into a symmetrical click with an inverse FFT, and plays the result through a convolver. The two modes therefore change the tone by exactly the same amount, and the screens show that only the timing differs.

The length of the filter is counted in samples, called taps, and the latency is half of it. These are the delays you get at the two common sample rates.

On playback you won't notice the delay, because your DAW's plug-in delay compensation holds every other track back by the same amount. You'll notice it straight away if you record through the EQ or play a synth with it on the channel: at 92.9 ms you hear each note well after you play it. Switch to minimum phase, or bypass the EQ, while you're tracking.

What is pre-ringing?

Any EQ that makes a sharp change rings. Send a click through it and part of what comes out is a short burst of tone at the frequency of the change, a bit like a glass ringing after you tap it. A minimum phase EQ puts all of that ringing after the click. A linear phase EQ can't, because its response has to be symmetrical, so half of the ringing comes first. On a drum hit you hear it as a soft swell leading into the transient, a sort of pre-echo that wasn't in the recording.

Your hearing makes the early half more obvious than the late half. A loud sound hides quieter ones that follow it for a good while afterwards, but it only covers the few milliseconds just before it. Ringing after a kick disappears under the kick, while ringing before it sits in the quiet gap in front and has nothing to hide behind.

The ringing lasts longer when the EQ move is steeper, narrower, bigger or lower in frequency. Low frequencies matter most, because ringing lasts a number of cycles and low cycles are long. These are the lab's measurements on the drums loop.

The 120 Hz cut rings early, but it barely changes the gap in front of the kick, because the loop has little energy around 120 Hz for it to ring with. Move the same cut to 250 Hz and it reaches into the body of the kick; the gap gets 18 dB louder. The narrow bass boost is the worst case by a distance. In minimum phase the same boost rings for about 99 ms, all of it after the hit.

Gentle moves barely ring at all. A 3 dB bell at 3 kHz with a Q of 1 rings for a fraction of a millisecond in either mode, too little for the lab to time, and at that point the two modes are very hard to tell apart.

Latency and filter length

A linear phase EQ has to be long enough to hold the curve you ask for. Narrow shapes at low frequencies need the most room, since a filter can't pick out a thin slice around 90 Hz if it only lasts a couple of cycles of 90 Hz. A longer filter follows the curve more closely, and it also adds more latency and allows longer pre-ringing.

The lab shows the trade-off on a 12 dB boost at 90 Hz with a Q of 6. With 16384 taps the curve comes within about 1 dB of the target, the latency is 185.8 ms and the click rings 54.9 ms early. At 1024 taps the latency falls to 11.6 ms and the early ringing to 6.3 ms, but the filter is only 23 ms long and the boost falls 6.9 dB short of its 12 dB peak.

That's why many linear phase EQs offer a choice of resolution or quality; FabFilter Pro-Q, for one, has several resolution settings in its linear phase mode. Use the shortest setting that still draws your curve, and check the low end on the plug-in's own display when you change it.

When to use linear phase EQ

Linear phase EQ earns its place when the EQ'd sound gets added to another version of itself. With a minimum phase EQ the two versions are out of step around every change, so they partly cancel and you get a dip or a comb filter you never drew. With linear phase they stay lined up.

Parallel EQ

Blending an EQ'd copy with the dry sound, either with a mix knob or on a separate bus, is where the difference is easiest to see. In the lab, a 24 dB per octave low cut at 150 Hz mixed 50/50 with the original should leave the cutoff about 1.4 dB down. In minimum phase that filter has shifted the cutoff frequency by half a cycle, so the two copies cancel there and the result dips 16.2 dB below what you asked for. In linear phase it lands exactly on the clean blend.

Mid/side EQ

In mid/side EQ the mid and side are EQ'd separately, then turned back into left and right. If the two halves get different phase shifts, they no longer recombine the way they were recorded, and the stereo image moves around the frequencies you changed. Linear phase keeps them in step, so only the tone changes.

One microphone of several

A drum kit recorded with close mics and overheads has the snare in every microphone. EQ the close snare mic in minimum phase and its timing moves against the overheads around each change, which can thin out or hollow the snare when they're mixed. A linear phase EQ moves the close mic's tone without moving its timing, so it sums with the overheads as before. The phase lesson explains why two copies of a sound that are slightly out of step cancel.

Mastering

Gentle, broad moves on a whole mix, like a couple of decibels of high shelf or a wide cut in the low mids, are a common use. They ring so little that pre-ringing isn't a concern, and the mix's timing stays exactly as it was. Steep low cuts on a master are the exception: they act on the kick and the bass at once, and a swell in front of every kick is easy to hear once you know to listen for it.

Crossovers

A multiband compressor splits the sound into bands and then adds them back together. Minimum phase crossovers add back up to a flat level but shift the timing around each crossover point. Linear phase crossovers add back up to the original sound, only delayed, and some multiband plug-ins offer them as an option, with the usual latency and pre-ringing.

How to choose a phase mode, step by step

1. Start in minimum phaseIt has no latency and no pre-ringing, and on most tracks you won't hear its phase shift at all.
2. Check whether the sound gets blendedIf the EQ'd signal is mixed with another copy of itself, through a mix knob, a parallel bus, mid/side processing or another microphone, switch to linear phase and listen for the change.
3. Listen in front of the hitsIn linear phase, solo the track and listen to the moment just before each kick or snare. A swell that wasn't there before means pre-ringing; make the slope gentler, widen the bell, or go back to minimum phase.
4. Set the shortest filter that holds the curveIf the EQ has a resolution setting, lower it until the low end of the curve starts to change, then go one step back up.
5. Compare the two modes at the same levelFlip between them a few times. If you can't hear a difference, keep minimum phase, since it costs nothing.

Common linear phase EQ mistakes

Using it on everything

Linear phase EQ on every track buys you nothing you can hear on most of them, and every instance adds latency and a little pre-ringing. Keep it for the jobs where copies of a sound get added together.

Steep low cuts on kick and bass

A 48 dB per octave low cut in linear phase, set high enough to touch the body of a kick, puts a swell in front of every hit. On drums and bass a minimum phase cut is usually the safer choice, or a gentler slope if you want to stay in linear phase.

Recording through it

With 90 ms or more of latency on an input channel, a singer or player hears themselves late and the take suffers. Bypass linear phase EQs on anything you're recording or playing live, and switch them back on afterwards.

Expecting it to sound cleaner

Both modes draw the same curve, so they change the tone by the same amount. Linear phase makes a different trade with timing, and on a single track the difference is often too small to hear. Judge it by switching back and forth at matched levels, and keep it only where you hear an improvement.

The linear phase vs minimum phase EQ comparison puts the two side by side, and the EQ lesson covers the curves themselves.

Training your ears to hear pre-ringing

Pre-ringing is hard to catch at first, because it's quiet and it happens in the moment you're least focused on, just before a hit. The quiz under the lab plays the drums through a strong low cut or a narrow bass boost, in one mode or the other, with the screens covered, and asks which. Listen to the gap in front of each kick for the swell. Eight rounds take a few minutes, and your best score is saved in your browser.

Questions people ask

Keep going