Your Linear Phase EQ Rings Before the Drum Hits

You put an EQ on the kick, found the boxy build-up around 250 Hz, and took out 4 dB with a tight Q. Then you noticed the plugin had a phase mode selector, thought about it for two seconds, and switched to linear phase, because linear phase is the careful option. The one that doesn't mangle your waveform. The one the mastering guys use.

And the kick went soft. Not obviously worse. Just less like a hit and more like a sound of a hit, the front edge slightly rounded, the attack arriving with a faint halo in front of it. You bypassed, un-bypassed, decided it was fine, and moved on.

It wasn't fine. Linear phase EQ has a cost, and on a transient it lands in the worst possible place, which is before the transient.

What linear phase actually buys you

Every filter delays some frequencies more than others. That's not a flaw in the design, it's how filters work. Run a signal through a bell cut and the frequencies near the corner come out slightly shifted in time relative to the rest. Your waveform changes shape even in the regions you didn't touch, because the parts of it got rearranged in time by different amounts.

Usually nobody cares. Your ear is far less sensitive to that kind of phase shift than the diagrams suggest. But it starts to matter the moment you sum the filtered signal with an unfiltered copy of the same thing, because then the shifted version and the original interfere with each other.

Linear phase EQ solves that by delaying every frequency by exactly the same amount. Phase relationships come out untouched, only the magnitude changes, and anything you sum it back with lines up the way it did going in. To pull that off, the filter needs a symmetric impulse response, which is the technical way of saying the filter's response is a mirror around its centre point.

That symmetry is the whole feature. It's also the whole problem.

Symmetry means half the ringing arrives early

Filters ring. Hit a narrow bell with a sharp transient and it doesn't just change the level of that band, it resonates a little, adding a short decaying tail at the filter frequency. A normal minimum phase filter puts that tail after the event, which is what your ears expect from every resonant thing in the physical world. Something gets hit, then it rings out.

A symmetric filter can't do that. If the response is mirrored around its centre, the ringing has to be mirrored too. Half of it lands after the transient, and half of it lands before, in the silence where nothing has happened yet. That's pre-ringing, and it reads as a soft ghost of the hit smeared out ahead of the hit.

Here's the part that makes it disproportionately audible. Your hearing masks quiet sounds that follow a loud one very effectively, over something like 100 to 200 ms after the event. Masking in the other direction barely works. A quiet sound arriving in the handful of milliseconds before a loud one is only weakly hidden by it. So the post-ringing of a minimum phase filter gets buried by the drum it belongs to, while the pre-ringing of a linear phase filter sits out in the open where you can hear it, not as a separate noise but as a loss of definition in the attack. The kick stops sounding like it starts.

Three things make it worse, and they all pull the same direction

Narrow Q. A tight filter rings longer than a wide one. Same reason a high-Q resonant peak on a synth sustains and a gentle shelf doesn't. Narrow surgical moves are exactly where linear phase pre-ringing gets loudest.

Low frequencies. A filter working at 60 Hz has to deal with cycles 17 ms long, so it needs a much longer impulse response than one working at 6 kHz. Longer impulse response means the pre-ringing extends further ahead of the event.

Transient material. The smear needs quiet to be audible in. On a sustained pad there's nothing to smear into. On a kick, a snare, a slap bass note or a plucked guitar, the pre-ring has an empty runway in front of the attack and you hear all of it.

Which means the worst possible use of linear phase EQ is a narrow cut at a low frequency on a drum. That's not an exotic edge case. That's the boxy 250 Hz notch on the kick from the first paragraph, which is probably the single most common EQ move in modern production.

The latency number tells you how bad it is

You don't have to guess how long the filter is, because the plugin tells you in the form of latency. FabFilter Pro-Q's linear phase mode has a process resolution setting, and at 44.1 kHz the options run from roughly 70 ms of latency at Low up to roughly 1509 ms at Maximum. FabFilter's own documentation flags possible pre-echo problems at the Maximum setting, which is them saying the quiet part out loud.

Use that as your rule of thumb. Latency and pre-ringing scale together, because they're both consequences of the same thing: a longer filter. If the plugin wants a second and a half of delay to do what you asked, it has built something that rings a long way in both directions. If it only wants 70 ms, the filter is short and so is the smear, but it also won't be accurate down low, which is why the low resolutions are recommended for higher frequencies and gentler curves.

What minimum phase charges you instead

Minimum phase EQ, which is what your DAW's stock EQ and most analog-style plugins do by default, keeps all the ringing after the event where masking hides it. Nothing arrives early. The bill comes in two other forms.

First, phase shift around the filter, which rotates the waveform. In practice this is mostly inaudible on its own, and Pro-Q's Natural Phase mode goes further and matches the phase behaviour of analog EQ deliberately, because that behaviour is part of what analog EQs sound like.

Second, and this one catches people: because phase shift changes the shape of the waveform, a cut can raise your peak level. Take 3 dB out of a band and watch the meter go up. That's normal, it's not a bug, and it's a good reason to keep an eye on your peaks after any surgical minimum phase move rather than assuming subtraction always buys headroom.

When each one earns the slot

Reach for minimum phase, or your plugin's zero latency mode, when the source has a transient you care about. Kick, snare, percussion, plucked and picked instruments, anything where the front edge is the sound. Also whenever you're tracking or monitoring through it, because 200 ms of latency makes overdubbing impossible.

Reach for linear phase when something is going to be summed and you need the phase relationship intact. EQing one mic of a multi-mic drum kit while the others stay untouched is the textbook case: with linear phase, the EQ changes that mic's tone without changing how it sums with the overheads. Same for parallel chains where you split a source, process one path and blend it back. Same for mid/side work and multiband splits that get recombined. Same for broad mastering tilts, where the Q is gentle and the material is dense enough that a tiny pre-ring has nowhere to be heard.

And when you genuinely need linear phase on something transient, widen the Q and lower the resolution. A gentle 2 dB shelf in linear phase rings far less than a surgical 8 dB notch, and most of the time the gentle version was the better move anyway.

The two modes are not quality tiers. They're two different bills for the same service, and picking the right one comes down to a question you can answer in a second: is there a transient here, and am I summing this with anything?

---