Hacker News Reader

Comment by QuantumRoar | Hacker News Reader

As far as I know, sound is actually so slow that you don't need to average the noise and make an educated guess as to what to filter. Acoustic noise filtering in headphones should work by detecting the outside noise with a microphone and playing the cancelling wave form as soon as the sound wave reaches the interior of your headphone.

In contrast, you can't do that for the noise of a car on the road because here the whole car vibrates and you couldn't, even if you wanted to emit sounds that cancel whatever rumbles quickly enough. Therefore they employ some spectral analysis of the noise and cancel persistent noise. This, of course, wouldn't cancel the sound of hitting a wall.

Again, I'm not 100% sure about this because it's outside of my field of expertise. Corrections are welcome.

View on HN

Considering the fact that it takes about 60 microseconds for sound to travel about 2cm, you only have about 900 cycles (assuming the same clock speed as an Arduino: 16MHz) to process the input sound, do a fast Fourier transform of the noise, and output the cancellation wave. I severely doubt that it can be implemented to be that fast. And not to mention that spikes in sound aren't nicely dealt with in Fourier transforms, so it's unlikely to actually remove the noise properly.

I doubt that there are analog-digital, digital-analog converters and an FFT module in there. I would have assumed that there's only some analog circuit. You don't even need spectral analysis. It just needs to invert the amplitude and delay the output for those 60 microseconds, right? This would also make the problems of spikes and Fourier transforms go away.

Sadly, both of you are quite wrong. ANC can easily cancel wideband sudden sounds. It may have problems at very high frequencies due to frame sizes and required group delay precision. Computing FFT or even more complex transforms with very low latency is possible, as is combining time domain and frequency domain processing. You cannot do analogue ANC at high frequencies, as the internal reflections in the headphone/ear system ruin it, while a digital system can measure these and respond well enough while not damaging the sound. Plus it is very hard to represent psychoacoustics without digital processing and those help preserve sound quality immensely.

Thanks for that. I wouldn't have expected it to be practical to have such low latency FFTs.

Thanks for the explanation!

Reply on news.ycombinator.com