For a decade now DACs have been able to generate signals that are so true to the original analog signal that nobody can tell the difference.
Filters will always be needed, because noise exists in the entire spectrum, from DC to daylight. And whatever you dont' filter out will alias back in a sampled system. So, even if you sample at 2 GHz, then the noise present at 2 GHz + 10 KHz will appear at 10 KHz in your sampled system.
The reason for higher sampling rates is purely so that aggressive filters can begin to do their job further and further outside of the audio band. In the case of CDs, the 44.1 KHz sample rate needed a filter than had sufficiently knocked down the noise at 22 KHz, whihc means the filter had some unwanted phase shift at 20 KHz.
But with 96 and 192 KHz sample rates, there is plenty of room for very aggressive filters to do their job without a bit of phase distortion in the audio band.
You'd really need to identify a broadly encountered distortion in today's 24-bit converters to clearly identify what you want to fix. The items you claim need fixing have been debated for decades, and were largely put to rest almost a decade ago.
Not really, no. Fourier's theorem allows you to generalize from sine waves to any arbitrary continuous (and sometimes not continuous!) periodic function. And as Dave and I have said again and again, ANY finite length sequence (including Beethoven's 9th) can be formally considered periodic.
No, they won't, in my scenario. I'm worrying purely about the playback side of things, and in that area of the chain the output of the DAC will look like an analogue signal, because the steps there will be small and short enough to be indistinguishable from analogue noise.
People have been identifying distortion for decades now, typically called "digititus". My efforts have shown to me that digital sound is extremely fragile, susceptible to interference both internally and externally sourced. Internally, because the high speed processing associated with oversampling and such contaminates the analogue, this is why there's a huge movement in the audio world based on non-oversampling. Externally, just about every high frequency EM signal anywhere near the system causes problems.
Yes, on the test bench everything's perfect, but that has little to do with the reality of the unpleasant world that real audio systems have to live in. Would you fly in a new design of plane whose systems checked out beautifully in test rigs on the ground, but no test pilot ever tried flying the thing?
Frank
Understanding of how the “oversampling and such contaminates the analogue” is the key. Maybe, for the start, we could replace “analog” with “digital” -> just to put us on the right track?
And…. what constitutes the “such”…
Boky
Perhaps, you should instead take the time to actually understand a comment before making such a dismissive response. I did not write that digital audio doesn't work. Of course it does, we've all had 'working' CD players for decades. Which doesn't necessarily mean that CD works to everyone's perceptual satisfaction.
Sharply bandlimited sampling and reconstruction is indeed good enough at capturing and reproducing frequency-domain signal content, but is poor at capturing and reproducing time-domain signal content. Digital signal processing is replete with applications, such as digital oscilloscopes and medical EKG signal capture, that are sensitive to the time-domain characteristic of signals. Those applications do not utilize brickwall filters for that reason. This is nothing new. Perhaps, only in the area of digital audio is the fact that bandlimited sampling and reconstruction inherently creates time-domain distortion either disputed or ignored.
Last edited:
OK, elaborate a bit ...
"and such" is any high speed, digital, processing that is electrically near the circuitry which is from our POV analogue: the output of the DAC on, the I to V, filtering, buffers. All circuitry is always analogue in its electrical functioning, whether we apply the moniker "digital" to it or not: we're concerned whether the messy, noisy, glitch riddled elements get sufficiently near our beloved, pristine, ear friendly "analogue" signals ...
Frank
If you can show a sample of a musical track, not some impulse or square wave test signal, that "suffers" from time-domain distortion, I'll be most intrigued ...
Frank
I think its safe to assume most digital recording for commercial music production is done at 24/96, and all mastering production steps done at that high sample rate with even deeper bit depth
so the audio frequency "time distoriton" is negilgible in the ADC stage
certainly compared to mics with several 20 kHz wavelengths dia diaphrams, inches from the performer's knuckles, vocalist's mouth that aren't exactly capturing "realistic" phase as could be heard anywhere in the audience
so the audio frequency "time distoriton" is negilgible in the ADC stage
certainly compared to mics with several 20 kHz wavelengths dia diaphrams, inches from the performer's knuckles, vocalist's mouth that aren't exactly capturing "realistic" phase as could be heard anywhere in the audience
I've had a look at some tracks on "21", the clipping is not that bad, compared to much that I've seen; the headache is the level of overproduction that's been done: every underlying musical element has been squashed and twisted for maximum sonic impact. Hence the sense of severe overcooking when listening ...
Frank
If you can show a sample of a musical track, not some impulse or square wave test signal, that "suffers" from time-domain distortion, I'll be most intrigued ...
What would you accept as evidence? Any CD I have will most likely have the time domain distortion (that's a linear form of distortion) baked in and there's no reference available to compare it to.
Which part of digital audio theory don't you accept - the part that says an AAF must be used to band-limit the signal prior to sampling or the part that guarantees such a filter will impose its own time response characteristic on the signal?
Well, get a highest quality analogue tape master, say from The Tape Project people, and do your encoding from that, and then do some comparisons of waveform captures in places where you think there are problems ...
Of course the anti-alias filter or equivalent has to be in place, but we are talking about very high resolution digital encoding here in this thread, the sampling rate will be at least in the 10's of MHz, and last time I checked microphones tend not to carry too much information in that part of the spectrum ...
As an example of what the good guys are achieving right now even for miserable ol' CD there's this : Telarc and other 20-bit CDs? Apogee made those converters ...
Frank
Ah if the sample rate's above 10MHz then I can't provide any evidence amongst my CD collection for you
With 10MHz sample rate PCM I guess its quite possible to implement a Gaussian AAF so the problem of time domain distortion goes away.
If you can show a sample of a musical track, not some impulse or square wave test signal, that "suffers" from time-domain distortion, I'll be most intrigued ...
Frank
Frank,
As Abraxalito indicated, we know from DSP theory that this is true. I've had this discussion several times before, and have found that what seems a single point of discussion is actually comprised of two distinct phases.
Phase one, the hardest phase, is simply to get people who have bought in to the "perfect sound forever" bandlimited sampling theorem mantra to understand that this only applies to the frequency-domain content of forever constant and repeating signals. It seems to me, that just because Fourier allows us to decompose an arbitrarily shaped signal (within Nyquist band limits) that forever repeats, unchanged from periodic cycle to periodic cycle, as meeting the requirements for Shannon misses the point regarding oscilloscope, EKG, and music waveforms. Those applications contain signals which do not forever repeat, unchanged from periodic cycle to periodic cycle. The field of DSP contains a number of applications which are sensitive to the time-domain content of the signal being sampled, and so do not utilize sharply bandlimited signal filtering.
Phase two, assuming the discussion ever gets past phase one, which it often doesn't, then becomes a debate over the practical or perceptual significance of the time-domain distortion inherent to sharply bandlimited sampling and reconstruction. Phase two is where a truly interesting discussion awaits us audiophiles. NOS offers a ready example of the audible benefits of (IMO) reducing time-domain distortion, as well as the audible problems of (IMO) not having removed the ultrasonic images. It would seem that digital audio is forever giving us our cake, but never allowing us to fully enjoy it.
It seems to me, that the promised land lay somewhere in a system featuring an higher than Nyquist native sample rate (which we have available today) combined with transient optimized anti-aliasing and reconstruction filters, such as the Gaussian type alluded to by Abraxalito. So, rather than attempting to maximize frequency-domain ultrasonic signal capture within a 96k or 192k sample rate channel, the available extra bandwidth would instead be utilized as a broad transition band, enabling flat to 20kHz baseband signal capture and reproduction utilizing transient optimized anti-alias and anti-image filtering throughout the recording, mastering, and playback chain.
Last edited:
Just a quick follow-up to my last comment:
It seems to me that DSD/SACD has brought us halfway to the audiophile promised land. DSD essentially meets the criterion I suggested for delivering a fully satisfying digital audio experience. It features a well over Nyquist native sample rate, and transient optimized filters. I've noticed that a great many of those who otherwise love SACD have subjective qualms with the treble range.
From a technical standpoint, that would make some sense because of the aggressive noise-shaping. It seems to me, just my own feeling, that DSD's approx. 2.8MHz sample rate is too low, given the one-bit quantizer used. If DSD had been specified for a much higher sample rate, greater quantizer dynamic range, or both we audiophile hobbyists might have only had to worry about the lastest homebrew zero-feedback amp circuit, or dipolar dynamic loudspeaker.
It seems to me that DSD/SACD has brought us halfway to the audiophile promised land. DSD essentially meets the criterion I suggested for delivering a fully satisfying digital audio experience. It features a well over Nyquist native sample rate, and transient optimized filters. I've noticed that a great many of those who otherwise love SACD have subjective qualms with the treble range.
From a technical standpoint, that would make some sense because of the aggressive noise-shaping. It seems to me, just my own feeling, that DSD's approx. 2.8MHz sample rate is too low, given the one-bit quantizer used. If DSD had been specified for a much higher sample rate, greater quantizer dynamic range, or both we audiophile hobbyists might have only had to worry about the lastest homebrew zero-feedback amp circuit, or dipolar dynamic loudspeaker.
Last edited:
Repeating an error several times does not make it more likely to be true.Ken Newton said:
Two mistakes in one sentence:Ken Newton said:
1. It is a bad debating technique to pretend that your opponents believe something which they have not stated ("perfect sound forever").
2. Fourier tells us that frequency-domain and time-domain descriptions contain exactly the same information, provided that the signal is periodic (so put the CD on repeat) and has no more than a finite number of finite discontinuities (music has zero discontinuities).
You are arguing the wrong point. Problems with digital audio, when they occur, come from the non-ideal nature of the anti-aliasing and reconstruction filters. That is where your attention should be directed, rather than spreading misinformation about undergraduate mathematics. Concentrate on your 'phase 2'.
flagship audio ADC chips are Delta-Sigma - modualtor frequency typ 6+ MHz, eval board/refernce circuit analog anti-alias in hundreds of kHz
the digital filters have sub us group delay variation to 20 kHz running at 96 k
I suppose you could want to do something about the aliasing right at the transition band, or add some "real" analog stopband attenuation on top of the digital filter's 120 dB stopband attenuation
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.