5 ms·

Explicitly watching for zero crossing seems like a rather limiting approach to quantifying a sine wave. Even then, at 48Khz you should get 400 samples between z

Explicitly watching for zero crossing seems like a rather limiting approach to quantifying a sine wave. Even then, at 48Khz you should get 400 samples between zero crossings, so you should do better than 0.5% accuracy. There's a lot of information held within the time series data of an audio signal, and something like an FFT can tap into it. For a 60Hz signal the Nyquist frequency is 120Hz, which is well below any audio interface. It seems like accuracy of any processing would be limited by a crystal oscillator somewhere more than anything, and perhaps any unintentional signal aliasing if there's no low pass filter before the audio interface's ADC.

anfractuosity 4y ago
That's a good point re. the crystal oscillator. Maybe I'm wrong re. the sample rate then, I found this paper - http://www.forensic.to/ENF%20processed.pdf http://www.forensic.to/ENF%20processed.pdf , where they mentioned 8kHz as the sample rate. I think what I was thinking is commercial systems such as https://synectic.co.uk/product/sd037-frequency-monitor-html/ https://synectic.co.uk/product/sd037-frequency-monitor-html/ mention specs such as 'The frequency monitor units are supplied pre-calibrated to an accuracy of +/-0.0001Hz'
- sgtnoodle 4y ago
  I was mainly responding to the hobby experiment of connecting a 12V transformer to an audio input on a computer. Even then, it seems like 8Khz sampling would be plenty to fingerprint a 60Hz signal. 8Khz sampling can capture information up to 4Khz, per the Nyquist frequency. That frequency monitor spec sheet mentions 90ms response and filtering, which would imply it averages information across 5 periods. Also, because it's a purpose built device, it's likely not periodically sampling the waveform with an ADC, but rather has a dedicated zero-crossing detection circuit that feeds into a capture input on a digital timer with at least microsecond resolution.
im3w1l 4y ago
Well 60/50Hz isn't the only relevant frequency here. Lets say you plug in ethernet-over-power (available off the shelf). You'd need a very fast interface to notice that.
- sgtnoodle 4y ago
  I can't imagine folks' ethernet-over-power signals are propagating across the national power grid, and even if they were, they aren't getting picked up in any audio streams.

anfractuosity 4y ago

That's a good point re. the crystal oscillator. Maybe I'm wrong re. the sample rate then, I found this paper - http://www.forensic.to/ENF%20processed.pdf http://www.forensic.to/ENF%20processed.pdf , where they mentioned 8kHz as the sample rate. I think what I was thinking is commercial systems such as https://synectic.co.uk/product/sd037-frequency-monitor-html/ https://synectic.co.uk/product/sd037-frequency-monitor-html/ mention specs such as 'The frequency monitor units are supplied pre-calibrated to an accuracy of +/-0.0001Hz'

sgtnoodle 4y ago
I was mainly responding to the hobby experiment of connecting a 12V transformer to an audio input on a computer. Even then, it seems like 8Khz sampling would be plenty to fingerprint a 60Hz signal. 8Khz sampling can capture information up to 4Khz, per the Nyquist frequency. That frequency monitor spec sheet mentions 90ms response and filtering, which would imply it averages information across 5 periods. Also, because it's a purpose built device, it's likely not periodically sampling the waveform with an ADC, but rather has a dedicated zero-crossing detection circuit that feeds into a capture input on a digital timer with at least microsecond resolution.

im3w1l 4y ago

Well 60/50Hz isn't the only relevant frequency here. Lets say you plug in ethernet-over-power (available off the shelf). You'd need a very fast interface to notice that.

sgtnoodle 4y ago
I can't imagine folks' ethernet-over-power signals are propagating across the national power grid, and even if they were, they aren't getting picked up in any audio streams.