>From Mr. Putzeys. Nope. If there is absolutely no content above a certain f, any sampling rate of 2*f or over will reconstruct the exact same signal. Sampling theory proves that a correctly executed (=filtered before and after) sampling conversion system is indistinguishable from a lowpass filter. And a lowpass filter is indistinguishable from nothing at all if the signal has nothing for the lowpass filter to remove. The latter assumes linear phase lowpass filters of course, but those are standard in AD/DA these days.
Needless to say, when there is content above 20kHz the question gets hairier. a. In formal (double blind) trials, no results have yet been reported of listeners able to distinguish between a "high sampling rate" audio signal and same with a lowpass filter inserted. Note that the ear is not a linear device. That is to say, you can't draw conclusions regarding audibility of HF spectral content based on audibility of single tone stimuli. For instance, quite a few people whose hearing ostensibly stops at 10k are able to detect the insertion of a 15kHz lowpass filter. But it does appear that 20kHz is roughly where it ends. b, c. I do mean roughly. If you make the filter very sharp and let it cut off right at 20kHz it can become ridiculously obvious. A 4kHz transition band is more or less what you need to render it inaudible. Trouble is: 20kHz+4kHz>44.1kHz/2... Actually most filters do have a 4.1kHz transition band on account of cost saving (look up "halfband filter). What this means is that most converters do not satisfy the Nyquist criterium. This is not very audible, but still the world would have been a different place had we standardized on 48kHz with a non-halfband filter with a 4kHz transition band. The whole Hirez thing would have gone nowhere. g. Exactly. Count on it that most lay people think that sampling means that you can't know the moment something happens (e.g. zero crossings) with better precision that one sampling period. That's obvious bollocks. It is precisely the LPF's task to insure that such information is coded with essentially infinite precision (same limit as an ordinary noise limited low-pass filtered channel). There is no deficiency in the ability of digital audio to encode timing. The only "resolution" that high sampling rates add is the ability to distinguish two closely spaced events (as in, one sampling period apart). But that argument is better carried through in the frequency domain. I think that for your own sanity, whenever people are debating what is essentially a specialised subject using vague terminology, you should simply say "to each his own" and wander off. People who "debate" sampling theory are not that much different from those who "debate" evolution theory. It too is a remarkably precise and powerful theory which is only being debated by lay people because they have an emotional stake in its consequences. Regarding NOS: such converters have a frequency response of sin(pi*f/fs)/(pi*f/fs). This causes its characteristic sound. A friend of mine who is a mastering engineer once came raving about some box by Altmann. So I told him to upsample his audio 2x, then mix it with itself shifted 1 sample, take it back down 2x and listen what it does on a real DAC. The result was as predicted. This signal through his normal DAC sounded exactly like the original signal through the Altmann. Theory wins, again. Regarding apodizing: I saw Peter's presentation of that. He does with plastic slides stuff that I've yet to see a powerpoint rodeo do. Essentially he takes two premises from the opinion circuit without passing judgment but merely looks where they lead, to wit: "what if the transition band determines audibility" and "what if pre-ringing is audible, and inband phase shift is audible too". Based on these two he then designs filters that are flat in magnitude and linear in phase up to 20kHz and then roll off gently to hit zero at fs/2. Now, if you look at it closely this means that his design method as applied to a 44.1kHz sampling rate will yield an ordinary sharp, linear-phase filter. You need a high sampling rate for apodizing filters to make sense. At 96kHz the filter starts rolling off gently towards cut-off at 48kHz. At that point the question becomes: OK so how does this compare with a sharp filter cutting off at 40kHz? Ri-ight. So after finding it hard to obtain solid double-blind data for 20kHz band-limited audio we're now about to embark on a quest for the ideal filter at 40kHz. You can imagine why it's not catching on. -- TheOctavist Vortexbox>SBT(stock)>>Forssell MDAC-2>>>Klein and Hummell 0300D Sota Sapphire/Lyra Kleos>>Bespoke Valve Phono Stage>>Mastersound Due Venti>>Link Audio K100 ------------------------------------------------------------------------ TheOctavist's Profile: http://forums.slimdevices.com/member.php?userid=52700 View this thread: http://forums.slimdevices.com/showthread.php?t=93483 _______________________________________________ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/mailman/listinfo/audiophiles
