[castleofargh]

oh same here, you have to look at Greg, Pinna and co for anything pro. I'm just a noob consumer enjoying the company and whatever I can learn from it ^_^.

 

[pinnahertz]

IMD is fine on digital with as many simultaneous sine waves as you like provided they are allowed to run long enough. It's a linear superposition. It is the stopping and starting that's the problem.

But, you see, that was the whole point. IMD was not fine on those digital systems. It is far better now, thanks to oversampling filters.

Starting and stopping a transducer is a problem many, many times worse than that of any digital audio system. In fact, transducer transient response is, by comparison, simply awful. It's the weak link in the chain, as far as that goes.

 

[71 dB]

Yes a time windowed FIR will be causal. But it is an approximation as you'd concede. Chord make a great deal of the FIR tap length issue and rightly so IMO. My favourite Nyquist filter algorithms (very limited test set) are the PMD ones. I actually seem to prefer fast roll-off to some of the 'clever' stuff. But my point though is 1) that the sync x function is not paralleled in any real world acoustic effect or even practical analogue filter. Here are or step responses:

http://www.analog.com/media/en/training-seminars/design-handbooks/Basic-Linear-Design/Chapter8.pdf -see p25 and 31, 32 and others later.

No pre-ring on any of them. Later on there is an analogue anti-alias filter design presented for CD btw. Note this particular discussion regarding time symmetric impulse response is an issue with filter implementation as normally (read just about always) done in digital audio, not digital audio per se. 2) IMO digital improves as you go beyond 16 bit (I guess you think HDCD was a waste of time. Microsoft didn't, but customers, on the whole, did). and 44.1/48 ks/sec. And even if you are a little way below the noise floor (an analogue problem..), which would be an already excellent -96dB for 16-bit.

I actually don't have any analogue tapes! I don't even have any vinyl any more! Just CD, SACD, DVDA and downloads! I have Wolfson 24-bit and BB PCM R2R ladder based dacs. I prefer the (older) BB in my set up on CD. I like the ESS9018 and the new high end BB multibit/delta-sigma chip. The latter can give really good results.

Analog filters are minimum phase => no pre-ringing. Digital filters are what we want them to be, minimum phase, linear phase or something else. The default digital anti-alias/reconstruction filters are linear phase to keep group delay constant and avoid phase distortion. Brickwall filters ring, rounder filters much less. It's about choosing what you want, but you can't get everything, analog or digital. Analog filters cause phase distortion. There is always a price. The length of FIR gives the frequency resolution of the filter. Near the nyquist frequency you don't need that massive resolution.

It's not a problem if you want to play infinitely long sine waves, and that's why digital measures good on traditional parameters. Probably you have played with a spectrum analyser and adjusted the averaging number or time window. It's the same principle. Your sine wave spectral peak is ill-defined if you don't average enough cycles. That's not just SNR, it's maths. Well the converse is also true. Time and frequency are conjugate variables. Unless your waveform is being infinitely repeated, frequency domain limiting (admittedly of any form) results in temporal imprecision or distortion. This is just a byproduct of Fourier transformation. End of story. Music is not a repeating waveform.

There's a lot ground between long sinusoids and 3 µs pulses. Some of it is called music. Luckily we need to go to short pulses before digital starts to "fail" because of it's band-limited nature. Even that is irrelevant, because the frequencies in question are beyond our hearing range. Blind tests prove that 16 bit/44.1 kHz digital audio does damn good job with music, so good that increasing bits or sampling rate doesn't seem to make much difference. Digital sounds bad only when placebo effect is present.

So another point. Yes, analogue systems, or digital systems representing analogue parameters, all introduce time domain imprecision/distortion by means of bandpass effects. But the analogue forms are simple functions with everyday parallels, just like our mechanical electrical equivalents we were discussing with headphones. They are causal in time and infinite continuous in time. Digital ones are not. They are artificial in both respects. This is done because it is the easiest way to make a near-phase and -frequency linear anti-alias filter. In phase and frequency, they are indeed excellent. But actually, these parameters, carried over from analogue, are not really a good measure of time domain performance and hence of music waveform accuracy.

Digital audio is continuous in time despite of discrete sampling points. The whole idea of sampling theory is that we can represent any bandlimited signal completely. Digital audio knows what happens between sample points, because there is only one possible continuous bandlimited signal that goes through the sample points. The only error happens because of quantization noise (or dither) limiting our dynamic range. So you have 100 % accurate original bandlimited signal + noise. What is artificial here?

 

[bigshot]

quasi-modo ON:
if you guys feel that there is more bring to the damping topic, let me know and I'll ask mother to move our off topic lounge somewhere else.

I don't think there's any problem with a thread evolving and moving on once the initial question has been asked and answered. I'm more interested in the off topic discussion in these threads than the same questions about cables and bitrates over and over. That's why I created the Sound Science Bar and Grill thread long ago.

I'm looking for this article in a small stack of UK Hi-Fi magazines I buy sometimes in the airport. If my memory is correct, it was something to do with someone cutting something recorded there direct to disc for vinyl. So perhaps it was done there but the opinion was not of an Abbey Rd engineer. I'll let you know if I find it.

That would be the Sheffield Labs guys. The article was a long time ago. They've since turned around on those claims. I can see why you're having trouble here. All those stair step waveform and continuous time arguments have been proven to be false and dismissed by peer review since then. Only the snake oil high end audio people trot those arguments out any more, and that is only when they want to sell you a vastly overpriced turntable.

You might want to do some googling on basic digital audio theory. It will help you separate the wheat from the high end audio chaff.

 

[SimonPac]

The whole idea of sampling theory is that we can represent any bandlimited signal completely.

I'll try to stick to technical specifics again. Agreed, the Nyquist critera are well known and considered set in stone. If you implement the LP filter perfectly, what comes out of the filter is, according to Nyquist, perfectly represented in continuous time by discrete time samples followed by a reconstruction anti-mirroring filter. Some processes, like frequency normalised random noise, don't satisfy the Nyquist proof, but perhaps we can ignore that. It's a moot point what relevance this has for general audio. Some random or chaotic processes are questionable under Nyquist.

I'll clarify these points, some keep coming up. I don't think anyone who knows DSP will argue with me.

1) I've been discussing what the filter does to the input audio signal in the time domain. Not what the PCM sampling codec process itself, subject to Nyquist, does with that filtered signal, but what the filter does to the analogue signal coming in to the ADC.
2) A scaled, convoluted sinc x impulse signal is the normally considered default input to a Nyquist based system simply because that is what the perfect anti-aliasing LP filter (prior to sampling at Nyquist rate) will produce if implemented whether ideally or approximately by FIR. What the reconstruction anti-mirroring filter is supposed to reproduce is this scaled, convoluted sinc x signal. That is what is not temporally accurate in the general case, either before or after the actual PCM sampling process. A related scenario mathematically is pulse doppler radar position/velocity uncertainty. Read the wikipedia article on conjugate variables, look at 'examples'. And the Fourier Transform one, especially the section under the heading 'Uncertainty Principle' (not Heisenberg!). It's an outcome of the very same FFT process which allows the Nyquist proof.
3) The fact that ideal filters and ideal matching between the anti-aliasing and reconstruction filters are not achievable or even likely in practice.
4) Questions over the adequacy of 16-bit for music. There's a Stanford paper (not very technical) saying plainly that it will improve until you can't implement it in practice. https://ccrma.stanford.edu/courses/192a/SSR/Digital Audio.pdf
5) This is not a bad paper on the general philosophy of reconstruction/anti-mirror filters in the light of what the ADC anti-aliasing filter usually does. https://www.ayre.com/pdf/Ayre_MP_White_Paper.pdfli
I don't necessarily agree with the conclusions subjectively. As I said, I like the PMD linear/fast roll-off implimentation.
6) 3uS Kronecker pulses are indeed of little relevance in this. Such a pulse if it could be generated by air pressure plus a mic (it couldn't...) would most likely be ignored by the ADC because the oversampling on the input converter wouldn't catch it most times. The guy is talking PA. You don't need a fast roll-off LPF for that. I think he is implying, like Bob Stuart from Meridian, that we need better time domain accuracy in digital audio.

 

[bigshot]

Is there a point to coming up with theories about potential problems before you prove that the problem actually exists? It seems to me a simple ABX of analogue signal vs digitized analogue signal would show if there was an audible problem. If it isn't audible, who cares?

 

[pinnahertz]

Is there a point to coming up with theories about potential problems before you prove that the problem actually exists? It seems to me a simple ABX of analogue signal vs digitized analogue signal would show if there was an audible problem. If it isn't audible, who cares?

Precisely!

To put it another way, presenting theoretical, even mathematical analysis of a system's "defects" is fine, but until directly related to something audible, not really worth talking about. And by "something audible", I mean proven by controlled double-blind testing. What we have here is verbal analysis, and a loose attempt at correlation with an opinion developed from fully sighted and biased auditioning. "I heard _____", and "I know a guy...who knows a guy who heard _______".

So far we have absolutely no real data to support the need for better "Bob Stuart" Timing. I remain puzzled at this, because the claims get pretty big and serious, which means they should be very, very easy to prove in a fully controlled DBT.

I'll join in, now in two-part harmony..."If it isn't audible, who cares?".
and...."If it is audible, then let's have the proof!"

 

[castleofargh]

I don't think there's any problem with a thread evolving and moving on once the initial question has been asked and answered. I'm more interested in the off topic discussion in these threads than the same questions about cables and bitrates over and over. That's why I created the Sound Science Bar and Grill thread long ago.

if we see this as our chatroom, and some topics can absolutely be it, then sure. but for all the guys who still care about damping and all the people in the future searching it in the forum, we're just creating a giant mess. anyway I can't move posts with my fake modo level anymore, so it's out of my hands anyway. I already PMed a godly creature asking for the last pages to be moved elsewhere, if nothing comes, you have your wish ^_^

 

[SimonPac]

Is there a point to coming up with theories about potential problems before you prove that the problem actually exists? It seems to me a simple ABX of analogue signal vs digitized analogue signal would show if there was an audible problem. If it isn't audible, who cares?

I thought it was audible, and I think I discerned which was which blind. My memory could be incorrect, and I'd be interested in repeating the experiment. Decent master tape, analogue recording chain, some acoustic instruments and/or relatively unprocessed vocals.

If the problem is undetectable or insignificant to the people concerned, I agree with what you said. There is a danger of spoiling everyone's enjoyment. Hi-Fi PR is full of very questionable claims for mechanisms which are supposed to transform sound quality. This one has some technical merit, but maybe it's not real world meaningful to most people, or maybe those who think it is meaningful like me are deluded by psychology.

Obviously subjective claims work both ways. AB blind statistics is worthwhile but people have said they are not relaxed in that situation!

Gear designers are faced with technical choices for their next product, and user perceptions. Over the years, some people have felt a need to 'polish' digital audio for whatever reason, both as better 44.1-16 or higher/deeper sampling. What is significant or worthwhile is very very subjective. Some very high end manufacturers and some more everyday ones seem to have signed up for MQA, though the doubters could say they only did so for commercial reasons. Some kickback against it may also be due to commercial factors.

Now back to smearing that reptile liniment on my mains cables.

 

[71 dB]

5) This is not a bad paper on the general philosophy of reconstruction/anti-mirror filters in the light of what the ADC anti-aliasing filter usually does. https://www.ayre.com/pdf/Ayre_MP_White_Paper.pdf.

Sorry about adressing only one point in your long post. Again you present us a "marketing paper." I have read a lot of these in my life, too many actually. In these prochures they make the claim that something is fundamentally wrong with normal digital audio (not true). Then they say they have worked hard to solve that problem. Then they explain to you the solution step by step using very easy to follow approach so that anyone who knows something about digital audio can follow it. However, those who know more than they want you to know, can see what's going on. They twist the relevance of things and use plots to illustrate it. However seeing is not hearing. The way we see is not the way we hear. What is visually relevant isn't always sonically relevant.

Pre-ringing looks bad and it's easy to understand how "unnatural" it is, but how bad does it really sound? First of all the ringing happens at around 20 kHz, a frequency only children can hear and even they hear it only at very high levels. Secondly, the ringing is short! 10 cycles at 20 kHz is over in 0.5 ms! That's about 1000 times shorter than the reverberation time of your listening room, but of course reverberation in your room isn't an issue at all. Thirdly, our hearing system has pre-masking: Loud later sounds mask previous ones, so the center peak of the sinc function actually masks the pre-ringing! Of course Ayre doesn't mention you this kind of facts, because it would make their claims much weaker. How can later sounds mask previous ones? That's because our hearing works at a ~10 ms "prosessing" delay so that a loud sound even 10 ms later can mask a quieter previous sound.

In the light of what I said above, the claims made by Ayre are wildly exaggarated. As I mention, my CD player allows selecting the reconstruction filter from 5 different options and I feel I can hear a very very tiny difference in soundstage width between the filters, but these differences have nothing to do with enjoyment of music or the accuracy of sonic detail. I am not even sure I can tell the differences in a blind test. Maybe, maybe not. For me this is a ridiculously small "problem" in digital audio and it happens to be the biggest "weakness" one can find.

They talk about "careful" listening tests, but do you think they'd waste money and time on tests that provide information that make selling their products more difficult? The fact is a lot of carefully done tests give often information that is against profitable business models. People buy products often out of ignorance and mental images created by papers like your link. Placebo effect make sure this kind of behaviour is hard to kill.

The real issue in digital audio is keeping jitter low, ADCs and DACs linear and having high quality analog input/output section in ADCs/DACs. Take care of these and you are good.

 

[SimonPac]

Sorry about adressing only one point in your long post.

I think I'll call it a day soon. All digital filters are a compromise, and fail differently and probably not that differently from each other if reconstruction using 44.1 is your aim. Marketing claims are often or even generally exaggerated, yes. You don't have the pre-anti aliasing signal to listen to; that's the point. I did.

The real issue in digital audio is keeping jitter low, ADCs and DACs linear and having high quality analog input/output section in ADCs/DACs.

They are important. They can be improved. Fourier time smear by anti-alias filter, only a little whatever filter you use.

 

[SimonPac]

To me any 44.1 has unnaturally etched transients yet paradoxically a lack of real resolution. This can sound impressive and I have no doubt mastering engineers have learned to work the format over the years. These impressions have stayed with me over the long term. Sharp handclaps and things like maracas sound coarse. Acoustic environment ques are diminished. The whole thing sounds a little sterile and bland after 88-24 or 192-24. Those in turn, from my limited experience, sounds similarly limited compared with a reasonable analogue source if it was derived from one.

That is not to say that 44.1 can't sound pretty spectacular because there are so many other variables in recording and production. LCD4's blow my HD650s right out of the water on resolution, but you can still tell this on 44.1-16. On the LCDs you hear discrimination between multiple stringed instruments, and cymbal decay, a lot more clearly than on the HD650's, even on a 44.1-16 master.

 

[SimonPac]

The real issue in digital audio is keeping jitter low

...

Well a manufacturer of very high end digital replay gear who I spoke to in July said they add short term phase noise deliberately to the clock after LF jitter reduction....these guys make master clock boxes as well.....

 

[pinnahertz]

I think I'll call it a day soon. All digital filters are a compromise, and fail differently and probably not that differently from each other if reconstruction using 44.1 is your aim. Marketing claims are often or even generally exaggerated, yes. You don't have the pre-anti aliasing signal to listen to; that's the point. I did.

And my point is, your comparison was uncontrolled. Your pre-anti-aliasing signal was, itself, corrupted with a huge list of flaws, and the test wasn't double blind. The results are not even slightly definitive, they're purely biased opinion.

They are important. They can be improved. Fourier time smear by anti-alias filter, only a little whatever filter you use.

But all aspects of todays ADC/DAC can and are improved to a point of being inaudible and transparent. Again, improvements you suggest are unrelated to benefits.

 

[pinnahertz]

To me any 44.1 has unnaturally etched transients yet paradoxically a lack of real resolution. This can sound impressive and I have no doubt mastering engineers have learned to work the format over the years. These impressions have stayed with me over the long term. Sharp handclaps and things like maracas sound coarse. Acoustic environment ques are diminished. The whole thing sounds a little sterile and bland after 88-24 or 192-24.

As....compared...to....what?? The goal is reproduction of the original sound wave (impossible, of course). If your reference is an analog master, your reference is a highly corrupt version if it's own input signal. Your reference should be the analog recorder's input signal, not the playback version. Have you done that comparison? Doubtful, you need to be in a studio to do that. Some of us have, and some of us have 30+ years ago when ADC/DACs were relatively primitive, but we were looking for the flaws.

And, most importantly, your comparison MUST be double blind. MUST! Or biases will overwhelm the results.

Those in turn, from my limited experience, sounds similarly limited compared with a reasonable analogue source if it was derived from one.

You don't have a reasonable analog source unless you have a live mix through an analog desk. Otherwise your analog source is corrupt, filtered, masked, modified, distorted...use whatever term you prefer.

That is not to say that 44.1 can't sound pretty spectacular because there are so many other variables in recording and production. LCD4's blow my HD650s right out of the water on resolution, but you can still tell this on 44.1-16. On the LCDs you hear discrimination between multiple stringed instruments, and cymbal decay, a lot more clearly than on the HD650's, even on a 44.1-16 master.

Again....you have not made a real controlled comparison, biases have overwhelmed your results. Have you even heard a real 16/44.1 master (not very likely, we haven't recorded like that in quite some time)?