Phase Shifting in Modern Noise Cancellation
ANC inverts a wave to cancel noise, and timing decides whether it works. Here is why it stops near 1 kHz, and what hybrid and adaptive designs change.
Active noise cancellation works by generating a sound wave shifted 180 degrees out of phase with the noise, so the two cancel where they meet. The hard part is timing. The system has microseconds to measure the noise, compute the inverse and play it back, and any error in that delay turns cancellation into reinforcement. This is why ANC works well below 1 kHz and poorly above it.
Phase is the whole mechanism, and understanding it explains every limitation these systems have: why they handle engine drone and not speech, why fit matters so much, and why the technology has not simply improved its way past those boundaries.
What a 180 degree shift means
Sound is a pressure wave alternating above and below ambient air pressure. Add a second wave of identical amplitude whose peaks land exactly where the first has troughs, and the two sum to zero.
Two conditions have to hold. The amplitudes have to match, so a 3 dB error in level leaves a noticeable residue. And the phase has to be opposite at the point where the listening happens, which in headphones is the few cubic centimeters between the driver and your eardrum.
Miss the phase and it gets worse rather than merely failing. A wave arriving 90 degrees out reduces very little. A wave arriving in phase adds up to 6 dB, so a system with an inverted delay actively amplifies the noise it was meant to remove.
Why time is the constraint
Producing a phase shift means producing a delay, and the delay that corresponds to 180 degrees depends entirely on frequency.
At 100 Hz, one cycle lasts 10 milliseconds, so half a cycle is 5 ms. That is a generous target. At 1 kHz a cycle is 1 ms and half a cycle is 500 microseconds. At 5 kHz half a cycle is 100 microseconds, which has to cover microphone response, analog to digital conversion, the filter computation, conversion back and the driver’s own response.
Wavelength makes the same point spatially. A 100 Hz wave is about 11 feet long, so a millimeter of positioning error between microphone and driver is irrelevant. A 5 kHz wave is 2.7 inches long, and a few millimeters of error is a large fraction of a wavelength.
This is the real ceiling. Most consumer ANC is effective from roughly 50 Hz to 1 kHz, delivering 20 to 30 dB of reduction in the best case, and tails off sharply above that. Higher frequencies are left to passive isolation from the ear cup seal, which is why fit matters as much as the electronics.
Feedforward, feedback and hybrid
Where the microphone sits determines what the system can do.
Feedforward puts the microphone on the outside of the ear cup. It hears the noise before it reaches your ear, which buys precious time to compute the inverse. The weakness is that it never hears the result, so it cannot correct for how your particular ear and seal change the sound.
Feedback puts the microphone inside, next to the driver, listening to what actually reaches your ear. It can correct for fit and for leaks, which is a genuine advantage. The weakness is that it hears the noise late, so it has less time, and a feedback loop can become unstable and oscillate if the gain is set too high.
Hybrid uses both, with the outer microphone doing the bulk of the work and the inner one correcting the residue. Nearly all current premium headphones use this arrangement, and it is the reason ANC performance improved noticeably in the last several years.
Adaptive filtering
A fixed filter would only work for one noise spectrum in one pair of ears. Real systems adapt continuously.
The standard approach is an adaptive filter using a least mean squares algorithm, which adjusts its coefficients thousands of times a second to minimize the signal at the error microphone. In ANC the variant used is filtered-x LMS, which accounts for the fact that the correction signal itself passes through the driver and the acoustic path before it reaches the error microphone.
That acoustic path has its own delay and frequency response, and the filter has to model it to remain stable. Manufacturers call this adaptive ANC, and its practical benefit is that the system retunes when you put glasses on, when the seal shifts, or when the noise changes from engine drone to air conditioning.
The limits remain the same. Adaptation improves how well the system matches the noise it can address. It does not extend the frequency range, because that is set by physics rather than by the algorithm.
Why speech survives it
Speech intelligibility lives largely between 1 kHz and 4 kHz, exactly where phase cancellation runs out of time. It is also unpredictable, and a feedforward system works by anticipating a repeating waveform, which continuous drone provides and conversation does not.
So ANC excels at aircraft cabin noise, engine rumble, train hum and air conditioning, all steady and low frequency. It performs poorly on the person talking behind you, on sudden clatter, and on a baby crying, which is deliberately pitched where cancellation is weakest.
Blocking those needs passive isolation, meaning mass and a good seal, which is exactly the same physics that governs walls. Our comparison of active noise cancelling against earplugs covers where each one wins.
Beyond headphones
The same phase inversion appears in car cabins, where microphones in the headliner and the car’s own speakers cancel engine and road noise, allowing manufacturers to use smaller engines without the drone reaching the cabin.
It also appears in active headrests, in ducted HVAC systems, and in industrial equipment. The constraint everywhere is the same: it works where the wavelength is long compared with the distance between the microphone, the source and the listener.
Which is why nobody has built an active system to cancel noise across a whole room. At 500 Hz a room contains many wavelengths, cancellation would only occur at specific points, and everywhere else the correction signal simply adds more sound. If you want a quiet room, that remains a mass and sealing problem.
Frequently asked questions
How does phase shifting cancel noise?
The system measures incoming noise with a microphone, generates a wave of equal amplitude but opposite phase, and plays it through the driver. Where the two meet, the peaks of one align with the troughs of the other and they sum toward zero. Both amplitude and phase have to match closely, since an error in either leaves a residue, and a wave arriving in phase rather than out of phase makes the noise louder.
Why does noise cancellation work better on low frequencies?
Because low frequencies give the system more time and more spatial tolerance. At 100 Hz half a cycle lasts 5 milliseconds and the wavelength is around 11 feet, so small errors in timing or microphone placement barely matter. At 5 kHz half a cycle is 100 microseconds and the wavelength is under 3 inches, which is shorter than the processing chain can reliably handle.
What is hybrid ANC?
A design using both an external microphone, which hears noise before it arrives and gives the system time to compute a correction, and an internal microphone beside the driver, which hears what actually reaches your ear and corrects for fit and leakage. Using both covers each one’s weakness, and it is why current premium headphones cancel noticeably better than earlier feedforward only designs.
Can active noise cancellation make noise worse?
Yes, if the phase is wrong. A correction signal arriving in phase with the noise rather than opposed to it adds up to 6 dB instead of cancelling. In practice this shows up as ANC that seems to emphasize certain frequencies, usually in the upper midrange where the system is at the edge of its timing budget. It is also why poorly fitting headphones can sound worse with ANC on than off.