How noise-cancelling headphones work

1Air

What is a sound?

Air looks empty, but it’s packed. A sugar cube of air holds about 25 billion billion molecules, mostly nitrogen and oxygen. They fly about at around 470 metres a second, faster than a jet airliner, crashing into each other all the time.

Their constant drumming on everything around them is what we call air pressure. Push on the air in one place and the molecules there get crowded. They push on their neighbours, who push on theirs, and the crowding moves on. That travelling squeeze is a sound.

Here’s a tube of air with a speaker at one end. Press Hum, or drag the speaker cone yourself:

Hum Push once Loudness

60 dB

The crowded patches are squeezes. The spread-out patches are stretches.

Sounds like

a normal voice

Pressure swing

0.02 pascals, 1 part in 5 million

Air moves

11 nanometres each way

Watch the red speck. It never goes anywhere. It only rocks back and forth while the squeeze moves on, at 343 metres a second. The air carries a sound the way a crowd carries a wave around a stadium: everyone stands up and sits down, and nobody leaves their seat.

The squeezes are tiny. When someone talks to you, the air pressure at your ear rises and falls by about one part in five million. The air there moves by about ten millionths of a millimetre, thousands of times less than the width of a hair. The quietest sounds you can hear move it by less than the width of an atom.

Engineers measure loudness in decibels, or dB. It’s a scale of ratios: every 20 dB means a squeeze ten times bigger. To your ears, 10 dB less sounds roughly half as loud. A quiet bedroom is about 30 dB. A plane cabin is about 80.

A sound is a pattern of squeezes and stretches, racing through air that barely moves.

2Adding up

What happens when two sounds meet?

Say two speakers play at once. A speck of air between them gets pushed by both. If both push at the same moment, it gets a double push. If one pushes while the other pulls, the two cancel.

That’s all there is to it: sounds add. The squeezes are so small that two sounds pass through each other without getting in each other’s way. At every point, the air simply does the sum.

Below, the red wave is a noise. The blue wave is a second sound that we control. Slide the blue wave along, or flip it:

Flip it Line them up Slide

0°

Hear the sum

Lined up. Every squeeze meets a squeeze. Together they’re twice as strong.

Slide the blue wave by exactly half a wave and it’s the same as turning it upside down. Every squeeze meets a stretch of the same size, and the air doesn’t move at all.

In January 1933, a German inventor named Paul Lueg applied for a patent on this idea. His drawing shows a pipe with a microphone, an amplifier and a loudspeaker. The microphone hears a sound coming down the pipe, and the loudspeaker plays it back so that it arrives upside down, and the two cancel further along. The American version of the patent, granted in 1936, is called “Process of silencing sound oscillations”.

A sound plus its upside-down copy adds up to silence.

3Timing

Can you cancel a real noise?

A pure tone is easy. A real noise, like a plane cabin, is a jumble of thousands of pitches at once, from deep rumbles to a high hiss. But the rule doesn’t care. Make an exact upside-down copy of the whole jumble, and it still cancels.

This one is real. Press play, then switch on the anti-noise. Your device is adding the noise and its upside-down copy, sample by sample:

Play Anti-noise

Plane cabin Voices

Late by

0 µs

Strength

100%

Just the noise. Press Play to hear it.

With the copy perfect, the noise vanishes completely. Now slide the timing. Being late by a tenth of a millisecond is enough to bring back the hiss, while the rumble stays away.

Here’s why. A low sound’s waves are long and slow. A 100 Hz rumble (100 waves a second) takes 10 milliseconds per wave, so being 0.1 ms late puts the copy off by just 1% of a wave. A 5,000 Hz hiss repeats every 0.2 ms. Late by 0.1 ms, the copy is off by half a wave, so its squeezes land on the noise’s squeezes. Instead of cancelling, they double.

Strength matters too. A copy that’s 10% too weak leaves 10% of the noise behind. That’s 20 dB quieter: a big drop, but not silence. Real headphones are never perfect, so the question is never whether some noise is left. It’s how much.

Low sounds forgive small mistakes. High sounds forgive almost none.

4The quiet spot

Why not make a whole room quiet?

If a speaker can cancel noise, why not put one in the corner of a noisy room? The trouble is that a room’s noise comes from everywhere. It bounces off the walls, the floor and the ceiling, so at each spot the waves arrive from all sides, in a different mix.

You can still cancel the noise at one point, where a microphone listens. Places very close to it get quieter too, because the mix there is nearly the same. A little further away the mix is different, the anti-noise no longer matches, and it can even add.

This is part of a room seen from above, with noise coming from every direction. Drag the microphone or the speaker, and change the pitch:

Pitch

250 Hz

New noise

250 Hz. Each wave is 1.4 m long.

The quiet spot shrinks as the pitch goes up, because it’s tied to the length of the waves. Stephen Elliott and his colleagues at the University of Southampton worked out the size in 1988. Averaged over many rooms, the spot where the noise drops by at least 10 dB is only about a tenth of a wavelength across. For a 100 Hz hum, that’s wider than your head. For 1,000 Hz, it’s smaller than your ear. For 5,000 Hz, it’s smaller than a fingernail.

So the speaker has to sit right at your ear, sealed in with it. That’s what an ear cup or an earbud does. The space inside is a few centimetres across, far smaller than the waves of a low sound. At low pitches, all the air in the cup rises and falls together, so cancelling the noise at one point cancels it everywhere inside.

Noise cancelling only works in a tiny space. Luckily, your ear is one.

5Ear cups

What do the ear cups do on their own?

Even switched off, a pair of closed headphones blocks plenty of noise. Wear them with the power off and voices sound muffled. The rumble of a bus barely changes.

An ear cup is a hard shell resting on soft cushions, like a weight on a spring. The noise outside pushes on it. A slow push, from a low sound, moves the whole cup in and out on its cushions. That squeezes the air inside, so the sound gets in. A fast push, from a high sound, is over before the heavy cup can get going. The cup stays put, and the air inside stays calm.

Change the pitch and watch the cup:

Pitch

100 Hz

Glasses on

100 Hz. The cup rides each push in and out. Inside, the noise is as loud as outside.

This is why ear cups and ear tips are good at stopping high sounds and poor at low ones. Stopping a low sound with a wall takes a lot of mass, which is why recording studios are built with thick, heavy walls. Nobody wants to wear that on their head.

I’ve kept only the bounce in this picture. Real cups also leak a little around the cushions. Glasses make that leak much bigger, because their arms hold the cushion off your skin.

Now look at what we’ve found. Low sounds forgive timing mistakes. Low sounds make big quiet spots. Low sounds get through the ear cups. Everything points the same way.

The ear cups take the high sounds. The electronics take the low ones.

6Head start

How do the headphones hear the noise coming?

To make anti-noise, the headphones first need to know what the noise is. The first way is a small microphone on the outside of each ear cup. It hears the noise a moment before the noise gets into the cup and reaches your ear. A chip turns what it hears upside down and plays it from the speaker, timed to arrive with the noise. Engineers call this feedforward.

A moment is all it gets. Sound crosses 2.5 centimetres in 73 millionths of a second, or 73 microseconds (µs). Take away the time the anti-noise needs to get from the speaker to your eardrum, and what’s left is the chip’s whole budget: often just a few tens of microseconds.

Race them. The noise runs from the outside microphone to your eardrum. The anti-noise goes through the chip and out of the speaker:

Chip takes

20 µs

Head start

2.5 cm

Send it over Bluetooth

In time, with 24 µs to spare.

The fastest chips made for this can take a sound in and send it out again in about 5 microseconds. Bluetooth, which brings your music from your phone, takes nearly a fifth of a second. That’s more than 2,000 times the chip’s whole budget. So your phone plays no part in the cancelling itself. It all happens inside the ear cup, in a chip sitting next to the speaker.

The chip has a second job. The noise changes on its way into the cup: the cup muffles its high parts, as we just saw. So the anti-noise has to match the noise as it will sound inside the cup, not as the outside microphone heard it.

Feedforward has one big weakness. It never hears the result. If the fit changes, or its guess is a little off, it keeps playing the wrong anti-noise with no way to know.

The outside microphone gets a head start, but it works blind.

7The loop

What if some noise sneaks past?

The second way puts a microphone inside the cup, right next to the speaker and your ear. It hears exactly what’s left. The chip takes whatever noise is there, turns it upside down and plays it back. Then it listens again, and again, without a break. This is called feedback, and it corrects its own mistakes.

The idea is old. In 1953, Harry Olson and Everett May at RCA built what they called an “electronic sound absorber”: a microphone, an amplifier and a loudspeaker that held the noise down around the microphone. Around 1957, Willard Meeker built the idea into an earmuff. It cut noise between about 50 and 500 Hz by up to 20 dB.

So why not turn the loop all the way up? Try it:

Push back

55%

Hear it

19 dB quieter at 100 Hz, but 6 dB louder at 2,000 Hz.

Turning it up cuts the low noise more, but a bump grows at higher pitches, where the loop makes the noise louder. That isn’t bad engineering. It’s a rule that every loop like this obeys, worked out by Hendrik Bode at Bell Labs in 1945. Push the noise down at some pitches and it has to pop up at others, like pressing on a waterbed. Engineers call it the waterbed effect.

The trick is to put the bump where it does no harm: up where the ear cups already block a lot, so you never hear it. Push harder still, and the loop’s own output comes back round just late enough to add to itself. It feeds on its own sound and whistles, like a microphone held too close to a speaker.

Real headphones land in the same place. In a study published in 2024, a feedback loop in a pair of over-ear headphones cut the noise by 10 to 20 dB between about 100 and 800 Hz, or up to about 30 dB with a better design, with small boosts above 1,000 Hz.

The inside microphone checks the result, but it can only push so hard.

8Fit

Why does the fit matter so much?

So far the chip has known exactly how its speaker sounds at your eardrum. But that changes with every head: the shape of your ear, how tight the headband is, hair under a cushion, the arms of your glasses. In the 2024 study, the same headphones were tried on 35 people. How loud the speaker sounded at the inside microphone differed by 3 to 5 dB at middle pitches, and by more than 10 dB at low ones.

A gap lets low sounds leak in, and it changes how the speaker’s sound reaches your ear. A filter tuned for a snug fit is now wrong.

The fix is a filter that learns. Over and over, it looks at what’s left at the inside microphone and nudges its own settings a tiny bit, in whichever direction makes that smaller. Bernard Widrow and his student Ted Hoff wrote down this rule at Stanford in 1960. (Hoff later helped design the first microprocessor.) In 1980 and 1981, Dennis Morgan and John Burgess worked out how to use it for noise control.

Watch it learn. Red is the noise at your ear, blue is the anti-noise, and the dark line is what’s left. Then change the fit:

Snug Glasses on Pressed tight

Learning Start over

Learning. What’s left keeps shrinking.

With learning off, a new fit undoes most of the good work. With it on, the filter finds its way back in a second or two. Real ones are quicker. Apple says the AirPods Pro, launched in 2019, adjust their noise cancelling 200 times a second.

Good headphones keep re-tuning themselves to your head.

9All together

What happens with everything on?

Many headphones now use all three at once: the ear cups, an outside microphone and an inside one. The AirPods Pro, for example, use one microphone facing out and one facing in. The outside microphone gets the head start. The inside one cleans up what slips past. The cups and ear tips take the high sounds.

Here’s a plane cabin, at your eardrum. Switch each part on and listen:

Play Ear cups Outside mic Inside mic

Ear cups only: about 13 dB quieter. The rumble is all still there.

It took a long time to get here. In 1978, the engineer Amar Bose tried the headphones an airline handed out on a flight, and found that the engine noise drowned the music. His company started working on the problem. In December 1986, prototype headsets went around the world with Dick Rutan and Jeana Yeager, who flew the Voyager aircraft nonstop for just over nine days. A Bose headset for pilots went on sale in 1989. The first QuietComfort headphones, for everyone else, came in 2000.

Each part covers the others’ gaps.

10What gets through

Why can you still hear people talk?

Switch on noise cancelling in a café and the hum of the fridge disappears, but you can still hear the people at the next table. Try it with a few everyday sounds:

Plane Air conditioner Voices Wind

Play Noise cancelling

A plane cabin. Most of it goes.

A voice has two parts. The vowels carry most of its loudness, mostly below 1,000 Hz, and noise cancelling does take some of that away. But the consonants, the sounds like s, t and f that make words clear, are higher, out of reach of the electronics. Only the ear cups work on them. And with the rumble gone, those consonants stand out more. Low noise usually hides higher sounds. That’s called masking, and noise cancelling takes it away.

Wind is a special case. Wind blowing across the outside microphone isn’t a sound on its way to your ear. It’s air swirling at the microphone itself. The chip can’t tell the difference, so it plays anti-noise for gusts that were never going to reach your ear, and adds noise instead. Headphones that can detect wind turn the outside microphone down when they hear it.

Some sounds can’t be cancelled at all. Part of your own voice reaches your inner ear through the bones of your head, without ever passing through the air in the cup. And the feeling of pressure some people get with noise cancelling on? When an audio writer tested for it, there was no real change in pressure. The usual explanation is that your brain expects some low rumble, and when it vanishes, the brain reads the silence like the pressure in your ears as a plane comes down. That idea still hasn’t been properly tested.

Noise cancelling is built for the low sounds. The high ones are left to the ear cups.