4 ms·

> Nyquist-Shannon means a 44.1kHz-sampled signal can perfectly capture an reconstruct a signal up to 22.5kHz freqs, and it's not like it's a flat cut off after

> Nyquist-Shannon means a 44.1kHz-sampled signal can perfectly capture an reconstruct a signal up to 22.5kHz freqs, and it's not like it's a flat cut off after that. Not sure what you mean by "it's not like it's a flat cut off after that". Any spectral content above 22.05 kHz is aliased to frequencies below 22.05 kHz -- that is mathematically inescapable. Maybe you're referring to the low-pass filters used to prevent this aliasing and for reconstruction -- these do have gradual rolloffs, but their center frequency is necessarily chosen to be somewhat below the Nyquist point, to ensure there is no noticeable spectral content above the Nyquist point. Notably, because it's much cheaper implement filters with steep rolloff in the digital domain, ADCs and DACs often operate at much higher bitrates (e.g. 384 kHz) to allow for simple analog filters for antialiasing/reconstruction, and then digital filters are used for down/upsampling to/from 48 kHz or what-have-you. But even in the digital domain, 20--22.05 kHz (or actually, 20--24.1 kHz) is a very tight transition band. 20--24 kHz (actually, 20--28 kHz) provided by 48 kHz sampling allows a much less complex filter to be used. And remember that the cutoff point for filters, 20 kHz in this case, is already at -3 dB (typically), meaning it's lost some fidelity.

lloeki 4y ago
> Not sure what you mean by "it's not like it's a flat cut off after that". Any spectral content above 22.05 kHz is aliased to frequencies below 22.05 kHz -- that is mathematically inescapable. Yes, that's what I mean. If you encode a 23kHz sound you don't get a flat signal as you'd get with a (-inf low pass filter) but an aliased one approximating the original; and as freq goes up, the errors introduced by aliasing (kind of) increase. You are also correct about mentioning the reconstruction filters and internal oversampling. In practice these errors don't matter, as we humans are, at the very best, barely able to hear anything up there (as in, is there a 23kHz sine vs is there nothing at all), let alone discern any difference between two different sounds (as in, is the sound emitted a 23kHz square or a 23kHz sine or a 23.5kHz sine), and even less when drowned by any signal in the human ear highly sensitive range. (Not even mentioning that getting into these areas start needing so much pressure to be perceived that it can become painful/damaging) So theory is all fun and games, because I guess ultimately my point is that in practice anything above 22kHz flat out doesn't matter. Admittedly I'm taking many many shortcuts and possible lies here and there, but I don't thing going "why doesn't Feynman fall through his chair" really helps. Let's all get out and enjoy some love and music already.

lloeki 4y ago

> Not sure what you mean by "it's not like it's a flat cut off after that". Any spectral content above 22.05 kHz is aliased to frequencies below 22.05 kHz -- that is mathematically inescapable. Yes, that's what I mean. If you encode a 23kHz sound you don't get a flat signal as you'd get with a (-inf low pass filter) but an aliased one approximating the original; and as freq goes up, the errors introduced by aliasing (kind of) increase. You are also correct about mentioning the reconstruction filters and internal oversampling. In practice these errors don't matter, as we humans are, at the very best, barely able to hear anything up there (as in, is there a 23kHz sine vs is there nothing at all), let alone discern any difference between two different sounds (as in, is the sound emitted a 23kHz square or a 23kHz sine or a 23.5kHz sine), and even less when drowned by any signal in the human ear highly sensitive range. (Not even mentioning that getting into these areas start needing so much pressure to be perceived that it can become painful/damaging) So theory is all fun and games, because I guess ultimately my point is that in practice anything above 22kHz flat out doesn't matter. Admittedly I'm taking many many shortcuts and possible lies here and there, but I don't thing going "why doesn't Feynman fall through his chair" really helps. Let's all get out and enjoy some love and music already.