HDTracks Offers 352/8/24 Downloads

[tailspn]

Since computer audio reproduction is digital sampling, this is a topic worth discussing at CA.

 

Not all computer audio reproduction is. DSD is one variant of Pulse Density Modulation, which is an analog modulation of a two level bit stream, whose density of those bits is proportional to the analog level modulating it. There are no samples, and there's no values in the bit stream. Kinda like the frequency modulation of a carrier in FM radio. Converting it to THE audio sample based value system (PCM) does then meet your description.

 

It's a side point, not germane to the previous discussion. But it's important if you want to start a broader discussion at CA to start with correct notions.

[Paul R]

Since computer audio reproduction is digital sampling, this is a topic worth discussing at CA. Even though certain members are prolific in posts, CA has a wide audience. And I write for them equally. Saying 2L is ripping people off because they don’t know what they are doing may be right or wrong, but keep in mind that the sign says forum, it doesn’t say fight.

 

Actually, if you are writing for an audience instead of having a conversation, perhaps you should consider using your CA blog.

 

There are indeed a lot of people here on CA.

[arw]

Actually, if you are writing for an audience instead of having a conversation, perhaps you should consider using your CA blog.

 

There are indeed a lot of people here on CA.

I'm new to the blog, but here it goes: On File Resolution and Audio Capture (DSD and PCM) - Blogs - Computer Audiophile

 

Not all computer audio reproduction is. DSD is one variant of Pulse Density Modulation, which is an analog modulation of a two level bit stream, whose density of those bits is proportional to the analog level modulating it. There are no samples, and there's no values in the bit stream. Kinda like the frequency modulation of a carrier in FM radio. Converting it to THE audio sample based value system (PCM) does then meet your description.

 

It's a side point, not germane to the previous discussion. But it's important if you want to start a broader discussion at CA to start with correct notions.

There are samples in my DSD bit stream, 1 yes or no every 2822400 seconds.

[firedog]

I'm new to the blog, but here it goes: On File Resolution and Audio Capture (DSD and PCM) - Blogs - Computer Audiophile

 

 

There are samples in my DSD bit stream, 1 yes or no every 2822400 seconds.

 

Well they aren't samples in the meaning it usually has in digital.

[firedog]

Since computer audio reproduction is digital sampling, this is a topic worth discussing at CA. Even though certain members are prolific in posts, CA has a wide audience. And I write for them equally. Saying 2L is ripping people off because they don’t know what they are doing may be right or wrong, but keep in mind that the sign says forum, it doesn’t say fight.

 

Yes, but the fights about hi-res are old and tiresome. Lots of arguments on both sides and rarely does anyone change sides or even opinions. It's fine if newer members want to discuss it, but why does a thread not directly on the topic have to get turned into a discussion of it...again?

 

I get tired of every thread discussing something about hi-res files getting turned into a debate about whether they are worth it or not, or what format is best. Those arguments should have their own threads.

[Jud]

I understand your position better now. Thanks for taking the time to make that clear. I found a couple more things to take a look at, if you are interested that speak more to the benefits of higher frequencies and capturing complex wave forms.

 

Understanding nyquist and its application. See 'Nyquist and signal content' Pg 10-13: http://www.wescottdesign.com/articles/Sampling/sampling.pdf

 

Square waves and sampling frequencies: Craigman Digital - PCM vs DSD

 

Hi arw. Once again I don't have time to provide a full response right now, so will do so later this evening. I will just say that the Wescott paper is excellent, and precisely agrees with my last reply to you; and that what is misleading about the Craigman article is easily understood. Think of this phrase until I get a chance for a full reply : "10kHz square wave"

 

Until then - good questions/discussion!

[wgscott]

My question is a bit different from "is it worth it?" The question is "is there any signal above 50kHz, or is it all noise-shaping as pictured in that 2L sample download?"

[Skeptic]

Again, the argument for hi-res is not necessarily that useful/exteme high frequencies exist on the recording, but that the lighter processing and filtering going on in the DAC produce a better analog sound after conversion.

Then why not upsample in the player?

 

Filling a huge file with noise from 50kHz to 350kHz is better than applying a digital filter?

Looks like a DSD conversion with a very weak filter in place.

 

Don't think of it as 'filling' something, rather it's more like brail on a curve, the more data points the more accurate the reproduction. More data points are always better

More data is only better if it contains additional useful information. Sampling can recreate any frequency up to half its sample rate perfectly.

The graphs which show a stair-step image to compare lower and higher sample rates are intentionally misleading.

Samples are points in time. The output from a properly functioning DAC will produce a smooth sine wave whether you are sampling at 44.1kHz or 352.8kHz. (NOS DACs are broken by design and will output a stair-stepped signal)

If the frequency is below half the sampling rate, the "additional data" of a higher sample rate simply adds more dots to that sine wave, it does not contribute more data.

 

Additionally, frequencies outside of our range of hearing can have a negative effect within the audible range for a much of the audio equipment out there, increasing distortion.

[Jud]

Then why not upsample in the player?

 

 

If better than the DAC's upsampling, that helps, yes.

 

But in turn, upsampling at the studio may be superior to what's offered in the player. Or best of all, the studio may not downsample and the DAC may not need to upsample, thus avoiding conversions all the way along the chain.

[Jud]

My question is a bit different from "is it worth it?" The question is "is there any signal above 50kHz, or is it all noise-shaping as pictured in that 2L sample download?"

 

I have no way of knowing for certain, but suppose there isn't. What then?

[wgscott]

Then a reasonable case can be made for 24/96 is sufficient. (It might be overkill as well, but I am willing to admit there is no point in throwing out the data. I don't feel as generous about paying extra for a pile of noise.)

[Jud]

Then a reasonable case can be made for 24/96 is sufficient. (It might be overkill as well, but I am willing to admit there is no point in throwing out the data. I don't feel as generous about paying extra for a pile of noise.)

 

I think you're making the error that's the other side of the same coin as arw. His sources say you need higher res to correctly follow the wavepath; you say what you need to follow the wavepath is any sample rate over 2x the frequency. As you saw me explain to arw, you're quite correct in terms of math. But you're correct about the wrong question. The question is not "In isolation from anything actually going on in the chain from the studio through my audio system, is Shannon-Nyquist correct?" The answer to that is "Yes, and so what?" The correct question is, "In view of everything going on in the chain from the studio through my audio system, could there be a sound quality advantage to a sample rate any higher than just over 2x the highest frequency of interest?" The answer to that is yes.

[wgscott]

I agree. However, 96kHz is already 5-fold in excess of audible frequencies. If 96kHz enables one to sample not only the audible frequencies, but in fact the entire spectrum of what can be produced/recorded, how could it not be sufficient? But the more pertinent question is are you getting anything of value for the extra money HDtracks is charging for the music files that are higher resolution than 96kHz? What is there in that file between 48kHz and the highest frequency? Is it anything apart from noise?

[tailspn]

Then a reasonable case can be made for 24/96 is sufficient.

 

It may well be, especially if you can't hear any difference between 24/96 PCM, and DSD.

 

The noise you see on your fft is the result of the delta-sigma modulator in the front end of the DAD AX-24 A/D Converter used by 2L. If they chose to have set up the converter as a 128fs DSD, you'd see the same plot. If they set it up as a 64fs DSD, you'd see the same shape noise envelope, but starting an octave earlier. The DXD conversion, which occurs in the A/D Converter, first decimate filters the 5.6MHz bit rate of the delta-sigma modulator to the 176.4KHz Nyquist frequency of the 352.8KHz DXD, then converts it to a 24 bit PCM word structure.

 

2L drank the Philips/Merging Koolaid about DXD, and has successfully made it into a marketing differentiator. DXD is the only format supported by Merging for post process mixing, level changes, and sweetening, and IMO, is a necessary evil if mixing functions are required. An alternative, used by other labels recording small acoustic music groups is setting up a session level balance in analog, recording in DSD, and limiting post processing to just editing, which does not require DXD conversions.

[Jud]

I agree. However, 96kHz is already 5-fold in excess of audible frequencies. If 96kHz enables one to sample not only the audible frequencies, but in fact the entire spectrum of what can be produced/recorded, how could it not be sufficient?

 

Pretty easily. Your DAC will take everything coming in below 352.8 or 384kHz and upsample it to one of those two rates. The upsampling is mathematically provable to be an imperfect process. (There is a great deal of discussion regarding audibility of the imperfections, but many people very familiar with how things sound in the studio, such as Keith Johnson, Barry Diament, and Cookie Marenco, think they are audible.) Also not to be ignored are the (again, mathematically provable) imperfections associated with conversion of the original recording, including sigma-delta demodulation and downsampling.

 

The folks at 2L have taken their recordings through the sigma-delta demodulation step to PCM, but they haven't downsampled the PCM from 352.8kHz. If you have a DAC that will take 352.8kHz input, then you don't need to go through any upsampling in your player or your DAC. You can just let it go through the sigma-delta modulation step before the final filter that converts the signal to analog. Thus fewer conversions that arguably deleteriously affect the sound, irrespective of the highest frequency of interest in the recording.

 

(BTW, you may want to ask yourself why virtually all DACs do this upsampling when there is almost certainly very little if any signal from the studio mics over 48kHz, if that high.)

 

But the more pertinent question is are you getting anything of value for the extra money HDtracks is charging for the music files that are higher resolution than 96kHz?

 

You're getting fewer conversions and thus arguably better sound. Whether the sound is improved sufficiently to justify the price is a very good question; I'd say the answer is no, based on all the music I've heard from HDT. Probably the best sounding file I've ever downloaded from HDT was 24/88.2. DSD is another story to my ears in my particular system, but that may have a great deal to do with my DAC. In general, I'd agree with the position I take most folks on this forum to have, based on many discussions: The intrinsic quality of the recording is quite a bit more important than the sample rate.

 

What is there in that file between 48kHz and the highest frequency? Is it anything apart from noise?

 

But is that the right question? Again I'd suggest asking yourself why virtually all DACs upsample to 352.8/384kHz when there is almost certainly very little if any signal from the studio mics over 48kHz.

[wgscott]

OK, the "fewer conversions" argument wasn't something I had thought of.

 

So I guess my remaining question is this: is the up-sampling that a DAC does in any way inferior to what we see in that example 2L file, or does it do this any less efficiently or accurately?

[Skeptic]

The correct question is, "In view of everything going on in the chain from the studio through my audio system, could there be a sound quality advantage to a sample rate any higher than just over 2x the highest frequency of interest?" The answer to that is yes.

Recording at high sample rates and bit-depths in the studio is advantageous.

You minimize the risk of aliasing due to improper or insufficient bandlimiting of the signal - or you can avoid it altogether in the recording stage if you are using a high enough sample rate.

Try finding a microphone that will actually reproduce anything up to 96kHz for example. 192kHz sampling should be sufficient for capture. (use higher sample rates if you like though - it doesn't really matter at this stage)

 

So if you are recording at 192kHz, why not distribute it? Well, you could - and that seems to be the argument made for high res downloads. It's the "studio quality" file.

 

But if you take your 192kHz track, and properly bandlimit it, you can produce a 48kHz file which is a perfect reproduction of the recording in the audible range.

I'm not suggesting that 48kHz is ideal, but it's not necessary to go to sample rates like 352.8kHz either.

 

I don't believe the people who say that the higher frequencies are felt rather than heard, or at least if that is possible, that it would add anything beneficial to playback.

If you actually look at the spectrum of many high-res releases you might notice a small problem which directly contradicts that.

If you could hear or feel frequencies like that, the common opinion would be that high-res releases would be strident and quickly fatiguing compared to CD.

 

And then you have, as illustrated above, improper DSD to PCM conversions which include huge amounts of high frequency noise, that can only serve to reduce sound quality.

 

If anything, it might be preferable to have everything above say 24kHz completely filtered out.

 

But is that the right question? Again I'd suggest asking yourself why virtually all DACs upsample to 352.8/384kHz when there is almost certainly very little if any signal from the studio mics over 48kHz.

But is this a reason to distribute high resolution material - especially when it often contains data which is harmful to the audible region on playback?

 

This assumes that the DAC won't do a good job handling upsampling itself, or if you prefer, this upsampling can be handled by your player instead.

[Jud]

If anything, it might be preferable to have everything above say 24kHz completely filtered out.

 

 

As soon as you figure out a way to do that perfectly without negatively affecting sound quality, please let all the digital audio engineers who agreed 8x oversampling was better than a brickwall filter in the CD player know how it's done.

[Jud]

OK, the "fewer conversions" argument wasn't something I had thought of.

 

So I guess my remaining question is this: is the up-sampling that a DAC does in any way inferior to what we see in that example 2L file, or does it do this any less efficiently or accurately?

 

Yes, because the DAC is upsampling, i.e., doing a conversion to the PCM file that necessitates filtering, which cannot be done perfectly; whereas 2L has not done any conversion/filtering to the PCM file at 352.8kHz. 2L has configured its ADC to convert the file from a DSD-type format to PCM, which is what tailspn was referring to upthread. With a DSD recording and a DSD-capable DAC, this conversion too can be avoided in principle. (If DSP or mixing is done to the file at the studio, this involves conversion to PCM and back to DSD, at least with current technology.)

 

If there *are* any conversions/filtering that need to be done at the studio, they can be done offline, which allows more time for a filter algorithm to work and thereby (1) allows a wider choice of algorithms, (2) avoids any possible impact on playback from the algorithm's use of computing resources, and (3) allows use of considerably more powerful hardware and software than are available to do conversions in any DAC.

[Paul R]

I agree. However, 96kHz is already 5-fold in excess of audible frequencies. If 96kHz enables one to sample not only the audible frequencies, but in fact the entire spectrum of what can be produced/recorded, how could it not be sufficient? But the more pertinent question is are you getting anything of value for the extra money HDtracks is charging for the music files that are higher resolution than 96kHz? What is there in that file between 48kHz and the highest frequency? Is it anything apart from noise?

 

I would suggest that for a great number of people, perhaps most people, 24/96 will be more than enough. In fact, it may be virtually indistinguishable in many cases from 16/44.1. I'll even go so far as to suggest that in some cases 16/44.1 files will sound better than 24/96, given the one may be much better mastered than the other.

 

I think there is usually a qualitative difference between 16/44.1 and 24/192K, and I would suggest that the difference lies in the equipment that renders the digital information into sound.

 

All bets are off however, when you hit DSD. Nobody I know of fails to reliably identify DSD, even DSD that is transcoded from 16/44.1 rebook. Pretty invariably, they think it sounds better too. That suggests - something. What, exactly, I don' know.

 

-Paul

[Jud]

I understand your position better now. Thanks for taking the time to make that clear. I found a couple more things to take a look at, if you are interested that speak more to the benefits of higher frequencies and capturing complex wave forms.

 

Understanding nyquist and its application. See 'Nyquist and signal content' Pg 10-13: http://www.wescottdesign.com/articles/Sampling/sampling.pdf

 

Square waves and sampling frequencies: Craigman Digital - PCM vs DSD

 

Hi arw. Once again I don't have time to provide a full response right now, so will do so later this evening. I will just say that the Wescott paper is excellent, and precisely agrees with my last reply to you; and that what is misleading about the Craigman article is easily understood. Think of this phrase until I get a chance for a full reply : "10kHz square wave"

 

Until then - good questions/discussion!

 

OK, back again.

 

First, just to hopefully clarify where I'm coming from: I agree higher sample rates have potential advantages. I very much disagree that the reason is the intuitive one, that more points along the wave help further specify the waveform once one is sampling at above 2x the highest frequency of interest.

 

I said that the Wescott paper agrees with my previous reply to you. In that reply, I said it wasn't a lack of sufficient sample points but filtering artifacts that occur in the real world where we don't have instantaneous filters or infinite time to apply them, that we had to worry about when discussing potential disadvantages of lower sample rates. So here's what Wescott's paper says on pages 1 and 3:

 

What Nyquist Did Say

 

The assertion made by the Nyquist-Shannon sampling theorem is simple: if you have a signal that is perfectly band limited to a bandwidth of f₀ then you can collect all the information there is in that signal by sampling it at discrete times, as long as your sample rate is greater than 2f₀.

 

 

What this means is that no system that samples data from the real world can do so perfectly — unless you’re willing to wait an infinite amount of time for your results.

 

[A]liasing...is happening all the time in the real world, anywhere that a real-world signal is being sampled.

 

So - there's Nyquist-Shannon saying as soon as you sample at more than twice the highest frequency of interest, you can perfectly reconstruct a signal. Wescott doesn't say this is wrong, that we need more samples than Nyquist-Shannon proved. What he says is that in the real world we don't have infinite time, and therefore any sampling that occurs in the real world will create an artifact called aliasing. Any sampling/filtering, not just Redbook rate or 24/96. So 192, 384, DSD, whatever you like - it won't ever get you infinite time and avoid artifacts like aliasing.

 

Then Wescott gives us some examples of how people get the application of Shannon-Nyquist wrong. At pages 10-13 of his paper, he goes through an example that shows why the Craigman article is misleading. Wescott talks about why monitoring a 60Hz power line at (or just above) 120 Hz won't work:

 

Nyquist didn’t say that a signal that repeats N times a second has a bandwidth of N Hertz.... It turns out that the waveform shown has significant harmonic content up to the 5th harmonic (300Hz).

 

So the sample rate has to be at least above 600Hz.

 

Let's look at the Craigman article in light of Wescott's discussion. Craigman shows that a Redbook sample rate is woefully inadequate to reproduce a series of "10kHz square waves" (there's that phrase I asked you to contemplate). What's the frequency of a square wave? Well, its rise time is zero, so the frequency is effectively infinite! But it repeats 10k times per second, you say. And the response, as Wescott might put it, is that Nyquist didn't say a signal repeating 10k times a second is a 10kHz frequency wave. If you think of sine waves as building blocks of the sounds we hear (the inverse of Fourier analysis breaking down a complex wave coming from a musical instrument into a fundamental and harmonics), in order to make something that looks even a little like a square wave we need stuff way into the 7th harmonic or above. Otherwise there isn't a prayer of builiding something that looks like the square wave's vertical initial rise. If the fundamental is 10kHz, the 7th harmonic is 70kHz. That means we won't get a decent-looking scope trace until we have a sample rate above 140kHz. And - quelle surprise - this is what the Craigman article shows.

 

To reprise: Yes, there are potential advantages to higher sample rates, but it's not due to a better representation of the waveform once you exceed the "Nyquist frequency." (Yes, you need to be very careful in deciding what is the highest frequency of interest, and thus the Nyquist frequency. We haven't even talked about transients yet.) It's due to being able to have filters with less audibly deleterious artifacts; and being able to have fewer filtering/conversion points in the chain between recording studio and your DAC output.

[Skeptic]

Let's look at the Craigman article in light of Wescott's discussion. Craigman shows that a Redbook sample rate is woefully inadequate to reproduce a series of "10kHz square waves"

I hate to be so reductive, but you aren't going to encounter square waves outside of a signal generator. It's not relevant to music.

[Jud]

I hate to be so reductive, but you aren't going to encounter square waves outside of a signal generator. It's not relevant to music.

 

Absolutely true, in fact not even from a signal generator - it will get closer, but not equal, to instantaneous rise and fall time. The intent of using these as a test is that it does a nice job of showing how well a given system can do on things that have very fast rise/fall times (for example, transients) and with the harmonics that distinguish the sounds of instruments from each other.

[Boris75]

Absolutely true, in fact not even from a signal generator - it will get closer, but not equal, to instantaneous rise and fall time. The intent of using these as a test is that it does a nice job of showing how well a given system can do on things that have very fast rise/fall times (for example, transients) and with the harmonics that distinguish the sounds of instruments from each other.

 

Yes, but harmonics matter only when they fall within audible range. Harmonics that fall outside the audible range, even if they are present in the signal, do not change the way it is heard.

[Jud]

Yes, but harmonics matter only when they fall within audible range. Harmonics that fall outside the audible range, even if they are present in the signal, do not change the way it is heard.

 

That's a subject I do wonder about. Is it exactly the same to our ears if the system plays back a trumpet with its 30+kHz harmonics, as with only the <22.05kHz effects of those harmonics?

 

The other reason to have the capability of reproducing higher harmonics is to "build" transients with rise time faster than that of a 22.05kHz wave. There is academic research indicating that different neurons are responsible for processing transients than are used in "hearing" tone (frequency); and additional research indicating we are able to detect transients with rise time approximately twice that of a 22.05kHz wave.