Phase, Polarity and the Comb Filter

You have one microphone on a guitar amp and it sounds good. You add a second microphone to get more of the cabinet, listen to both together, and the whole thing collapses — thinner than either mic on its own, hollow in the middle, oddly distant. Nothing is clipping. Both mics sound fine soloed. Both are plugged into identical preamps.

This is the most common way a session goes wrong, and it is not a fault in your gear, your room or your ears. It is arithmetic. This lesson is that arithmetic, and once you have it, a whole category of problems stops being mysterious and starts being solvable — usually by moving something a few centimetres.

Phase is a position, not a quality

A sound wave repeats. Every cycle takes it from its resting point, up to maximum pressure, back through resting, down to minimum pressure, and back to rest again. Because it repeats identically, we can describe where we are inside any single cycle the same way we describe where we are around a circle: in degrees, from 0 to 360.

That is all phase is — a position inside a cycle, expressed in degrees. One complete cycle is 360°. A quarter of the way through is 90°. Halfway is 180°.

Phase for a single wave in isolation is not very interesting; a wave has to be somewhere in its cycle at all times. Phase becomes useful the moment there are two waves, because then you can talk about the relationship between them.

• Two waves of the same frequency that start at the same moment are in phase — 0° between them. They reinforce.
• One delayed by a quarter of a cycle is 90° out of phase.
• One delayed by half a cycle is 180° out of phase. This is the interesting one: where the first wave is at maximum pressure, the second is at minimum. They oppose each other exactly, and if their amplitudes match, they cancel to nothing.

That word nothing is not an exaggeration. Two identical sine waves at 180° sum to silence. The energy does not go anywhere dramatic; the two pressure changes simply request opposite things of the same air at the same instant and the air does neither.

The one calculation

Here is the thing that turns phase from an abstract idea into a working tool. Phase difference between two copies of a signal comes from time — one copy arriving later than the other. And the relationship between a delay and a phase angle is a single line of multiplication:

phase in degrees = delay in seconds × frequency in Hz × 360

Work through it three times and it will stick.

One: a 250 Hz tone delayed by 1 millisecond

0.001 × 250 × 360 = 90°

A quarter of a cycle late. Noticeable, but nowhere near cancellation.

Two: the same tone delayed by 0.25 milliseconds

0.00025 × 250 × 360 = 22.5°

Barely anything. A quarter of a millisecond is about 8.5 cm of extra path length, and at 250 Hz that is almost irrelevant.

Three: running it backwards — what delay puts 500 Hz at 180°?

Rearranged, delay = phase ÷ (frequency × 360):

180 ÷ (500 × 360) = 180 ÷ 180,000 = 0.001 seconds = 1 millisecond

So one millisecond of delay completely cancels 500 Hz.

Now look carefully at examples one and three together, because they are the whole lesson. The same 1 ms delay puts 250 Hz at 90° and 500 Hz at 180°. One delay produces completely different phase relationships at different frequencies — because higher frequencies fit more cycles into the same amount of time.

This is why a delayed copy of a signal does not simply sound like a quieter version of it. It sounds different, because it is doing something different at every frequency simultaneously.

Polarity is not phase, and the confusion costs people hours

On almost every microphone preamp and console channel there is a button marked ø, or PHASE, or PHASE INVERT. It is mislabelled, on nearly every piece of equipment ever built, and the mislabelling causes genuine confusion.

That button performs a polarity inversion. It takes the entire signal and flips it upside down: every positive voltage becomes an equally negative one and vice versa. It does this to all frequencies equally and instantly.

A phase difference is caused by time — one signal arriving later than another — and, as the calculation above shows, it therefore affects every frequency by a different amount.

The distinction is not pedantry, and here is the practical consequence:

• If two signals are genuinely opposite in polarity — one is the upside-down version of the other — the ø button fixes it completely, at every frequency, perfectly.
• If two signals are separated by a time delay, the ø button cannot fix it. It will improve some frequencies and make others worse, because a single flip cannot correct a rotation that differs at every frequency.

This is exactly why flipping polarity sometimes transforms a sound and sometimes does nothing useful at all. When it works, you had a polarity problem. When it half-works, you have a timing problem, and the fix is to move a microphone or apply a matching delay — not to keep pressing the button.

Comb filtering: what a delay actually sounds like

Take a signal, add a delayed copy of itself, and listen to the sum. At frequencies where the delay works out to a whole number of cycles, the copies reinforce and you get a peak. At frequencies where it works out to a half-cycle, they cancel and you get a deep null. In between, everything in between.

Plot the resulting frequency response and you get a regular pattern of peaks and deep notches marching up the spectrum, evenly spaced. On a linear frequency scale it looks unmistakably like the teeth of a comb, which is where the name comes from. This is comb filtering, and it is one of the defining sounds of amateur recording.

Where the notches land

Two formulas, both straightforward, both worth memorising. For a delay of t seconds:

• The first null sits at 1 ÷ (2t)
• Nulls and peaks then repeat every 1 ÷ t Hz above that

For a 1 millisecond delay:

• First null: 1 ÷ (2 × 0.001) = 1 ÷ 0.002 = 500 Hz
• Spacing: 1 ÷ 0.001 = 1,000 Hz

So one millisecond of delay carves notches at 500 Hz, 1.5 kHz, 2.5 kHz, 3.5 kHz and onward forever. Not a gentle dip — a deep, narrow cancellation, repeated across the entire spectrum.

And critically, this happens to complex material, not just test tones. Speech and music contain energy at all those frequencies, so a comb filter removes real, audible components of a real performance. It is why the result sounds hollow, phasey, small and slightly metallic — a very specific and very recognisable character once you know to listen for it.

Where it comes from, in practice

Every one of these is a delay in disguise:

• Two microphones on one source. The sound arrives at the nearer one first. The difference in path length is the delay.
• A microphone near a hard surface. Direct sound plus the reflection off the desk, floor or wall — the worked example in the previous lesson put a null at 1.1 kHz from a desk alone.
• A DI and a miked amp on the same instrument. The DI is effectively instantaneous; the microphone is however far away it is, plus the amp's own latency.
• A close mic and a room mic. The whole point is the distance, which means the whole point is the delay.
• A widely spaced stereo pair summed to mono. Fine in stereo, potentially destroyed in mono — which is why mono compatibility gets its own discipline later.
• A plugin with latency in a parallel path that the DAW has not compensated for.

Hear it in thirty seconds

Swap the positive and negative leads on one of your speakers, then play a track you know intimately.

It will not sound broken, which is what makes it such a good demonstration. It will sound oddly wide — diffuse, spacious, hard to locate. And the low end will hollow out noticeably, because bass content that is shared between both channels is now being asked to push and pull at the same time by two drivers facing the same room. Centre-panned material loses focus. The top end loses its point of origin.

That is a pure polarity inversion, with no time delay involved, and it takes about half a minute to demonstrate. Swap the leads back afterwards.

Worked example: the snare drum

A snare drum with a microphone above the top head and another underneath, pointing up at the wires, is completely standard, and it is a textbook case of both effects happening at once.

Take a bottom mic sitting 6 cm below the top mic's position. The sound has to travel that extra distance, so:

0.06 ÷ 343 = 0.000175 seconds = 0.175 milliseconds

Now run the phase calculation at two frequencies:

• At 200 Hz — the body of the drum: 0.000175 × 200 × 360 = 12.6°. Negligible. The two mics essentially agree.
• At 3 kHz — the snare wires and the crack: 0.000175 × 3,000 × 360 = 189°. Almost exactly opposite. Near-total cancellation of the very thing you added the bottom mic to capture.

On top of that there is a genuine polarity opposition, which is a separate problem. When the stick hits, the top head moves downward — away from the top mic, producing a rarefaction there, and toward the bottom mic, producing a compression. The two microphones see opposite pressure changes from the same event.

So the ø button on the bottom mic does something real: it corrects that polarity opposition, and the low end and body of the drum immediately improve. What it cannot do is fix the 189° at 3 kHz caused by the 6 cm of travel, because that is a time problem and it is different at every frequency.

The complete fix is to correct the polarity and remove the timing difference — nudge the bottom mic's track earlier by 0.175 ms in the DAW, or physically move the microphone. Do both, and the two mics finally add up instead of fighting.

First null for that 0.175 ms delay, if you want to check your work: 1 ÷ (2 × 0.000175) ≈ 2.9 kHz, with nulls repeating every 5.7 kHz above it. Which is precisely where the problem was.

How to see it, not just hear it

Three tools will show you phase problems directly, and all three are already in your DAW.

The mono button. The single most valuable button in a mixing environment, and the most neglected. Summing to mono forces every phase relationship in the mix to resolve itself audibly — anything that was quietly cancelling now cancels in front of you. If a source loses body or disappears when you press it, you have found a phase problem you could not hear in stereo. Check every multi-mic source in mono while you are still tracking, when moving a microphone is still an option.

A spectrum analyser. Comb filtering has a signature no other problem has: a regular series of deep, evenly-spaced notches climbing the spectrum. Once you have seen it once you will recognise it instantly, and the spacing tells you the delay — if the notches are 1 kHz apart, you are looking at a 1 ms difference.

A correlation meter. It reports how similar your left and right channels are, on a scale from +1 (identical) through 0 (unrelated) to −1 (perfectly opposite). Sitting near +1 means you are close to mono. Hovering around 0 is normal for a wide, healthy mix. Spending time in negative territory means substantial cancellation is waiting for anyone who plays your mix on a phone speaker, a club system, or anything else that sums to mono.

The same effect, used on purpose

Everything described so far is phase as a problem. It is also, deliberately used, a whole category of production tools — which is worth knowing now so that they arrive later as familiar rather than magical.

• A flanger is a comb filter whose delay is continuously modulated, so the notches sweep up and down the spectrum. The characteristic jet-plane whoosh is the sound of nulls moving.
• A chorus uses a longer, modulated delay — too long for strong comb filtering, short enough that the ear fuses the copies into one thickened source instead of hearing two.
• A Haas-effect widener delays one side of a stereo pair by a few milliseconds. Your brain keeps localizing on the earlier arrival, so the source stays put while sounding much wider. It also carries the exact mono risk described above, which is why it is used carefully.
• Mid-side processing, polarity-flipped parallel chains and phase-rotation tools all operate on the same principle from different angles.

Which is the useful way to hold all of this: a delayed copy of a signal is never neutral. Whether the result is a defect or an effect depends entirely on whether you chose the delay.

How to actually fix it

In order of how much good they do:

1. Move the microphone. Free, immediate, and it fixes the cause rather than the symptom. Listen to the mono sum while someone moves the second mic and stop when it sounds fullest. This is the professional answer and it is not a compromise.
2. Observe the 3:1 rule. When using multiple mics, keep the distance between them at least three times the distance from each mic to its source. That drops the leaked signal far enough at the neighbouring mic that the comb filtering it causes stops being audible. There is a whole lesson on this in the recording specialization; it exists because of this one.
3. Flip polarity — but only after checking whether it is actually a polarity problem. Ask whether the two sources genuinely see opposite pressure, as with the snare. If they do, flip. If they merely arrive at different times, flipping is a coin toss.
4. Align in the DAW. Zoom in, find the same transient on both tracks, and slide one until they line up. Modern tools will do it automatically. This is a genuine fix for a genuine time difference.
5. Reach for an EQ. Last. A comb filter has dozens of narrow notches across the whole spectrum, and you cannot boost a null back into existence — there is no signal there to raise, only cancellation. Boosting into a null adds noise and burns headroom to accomplish nothing.

Why this lesson sits so early

Phase is not an advanced topic that you graduate to. It is the mechanism underneath a startling proportion of everything that follows.

It is why acoustic screens filter rather than block. It is why every stereo microphone technique has a specific spacing and a specific angle rather than a vague suggestion. It is why room reflections change the tone of a source rather than just adding an echo. It is what a flanger is doing on purpose. It is why summing to mono can destroy a mix that sounded wide. It is the reason two speakers in a room measure differently from one.

You do not need to calculate anything mid-session. You need one instinct: when a sound goes thin and hollow and nothing is clipping, stop reaching for EQ and start asking what is arriving twice.

What to practice

• Duplicate a track, delay the copy by 1 ms, sum to mono. Then sweep that delay from 0 to 20 ms and mark the point where hollow becomes doubled, and doubled becomes echo.
• Record an acoustic guitar with two mics. Without moving the first, find the position for the second where the mono sum sounds fullest. Then measure how far apart they ended up.
• Swap the + and − leads on one speaker, play something you know well, and write down what you hear before you swap them back.
• Take a snare top and bottom mic, or any two mics on one source, and A/B the polarity flip on one of them. Note which frequency range improves and which does not.
• Calculate the first null frequency for a 2 ms delay. Then set a 2 ms delay on a duplicated track and confirm it with an analyser.