Great thread on 384K and high sample rates

[Nathanael]

Dan Lavry's posts alone should be required in all digital audio FAQ's.

Read the thread and learn. Best non-math discussion I've seen. I don't know anyone more credible than Glen Z and Dan Lavry to address this.

Go to rec.audio.pro on Google and search for 384K PCM.

[Teacher]

y don't you make it easy for us and post the link???

[heinz]

da link

select post 26 in the left column

[Auxillary]

Dan is missing the idea. At the level of coverters it will make little difference. It is the processing at the intermediary stages that cause significant differences in the sound of sample rates. Sounds like a guy looking down a horses mouth wondering where all the shit went.

There are special circumstances and specific dsp designs that offer an audible change with higher sample rates. This can be taken care of by proper resampling and dither from internal processing. I can hear the difference between 48 and 96 from the source material during processing. In a mix it can be very difficult because the quality of the mix is up to the engineer. The results of different sampling rates would easily cause the same circumstances to come out two entirely different mixes. Thereby making a set-and-sum comparison worthless.

-former dsp programmer at motorola and a well-known VST company that will remane nameless in case I turn into a big ass during this discussion.

[Ruphus]

Thanks, Heinz.

I had to drop my pants after that post. Went back and apologized to others with whom I had been arguing about benefits or not of higher sample rates.
Thinking of the differences between 44.1 and 96k I was wondering why we couldn´t agree on higher SR clearly being better sounding.

However, I hadn´t ever given 48k a try at all. Obviously that was a wrong thought.

If I understand Mr. Lavry right tracking at 48k is best to do.


Learning every day.

Ruphus

[ttauri]

Originally posted by Auxillary
Dan is missing the idea. At the level of coverters it will make little difference. It is the processing at the intermediary stages that cause significant differences in the sound of sample rates.

Not sure if this is what you mean by intermediary stages, but what I've often wondered is how plug-ins, say, interact with sample rates. Like: inasmuch as you don't need a 192khz sample rate to capture a recording off a mic, does the same hold true for keeping things euphonic when mashing up that data with a bunch of in-the-box eq/compression/reverb/etc?

Peece,
T. Tauri

[Teacher]

Originally posted by ttauri
Not sure if this is what you mean by intermediary stages, but what I've often wondered is how plug-ins, say, interact with sample rates. Like: inasmuch as you don't need a 192khz sample rate to capture a recording off a mic, does the same hold true for keeping things euphonic when mashing up that data with a bunch of in-the-box eq/compression/reverb/etc?

Peece,
T. Tauri

good point....i've noticed plug-ins sound much better at 96 then 48

[Nathanael]

Actually, I think that Glenn Z and Dan Lavry both intimated that the problems of doing digital processing are compounded at higher sample rates.

Unless I am missing something major, the accuracy of a digital filter/process is primarily related to the precision of the maths on the backend, not the sample rate. Higher sample rate does not equal better processing by itself unless I am missing something very basic.

What I took from the Google discussion was that higher sample rate processing requires far more effort to achieve the same quality as that at lower rates. It is so much more difficult, intimates Glenn Z, that many designers switch from FIR to IIR filters without telling anyone.

Dan's posts seem to unequivocally state that higher sample rates aren't necessary for conversion, and introduce nastiness into the whole post-recording process in the form of unnecessary complexity. Glenn Z seems to indicate that solutions can be found, but it seems harder rather than easier. Unless I'm missing something, he is claiming a comprehensive argument for "reasonable" sample rates. It seems that somewhere between 48 Khz and 96 Khz is Dan's opinion.

I am not a former DSP designer for anybody, and didn't take math past Calculus, so I'll try to confine my opinions to my understanding of what they did and didn't say, not what is or isn't true.

I think one of the reasons that these discussions get so fuzzy so quickly is that the handful of people in the world who operate at Dan's level have a very different gear experience than all but a few on this board (or any other). Dan, the folk at Prism, dCS, EMM labs, etc have converters that approach the theoretical limitations to an extraordinary degree in their "gold" or equivalent products. They live in a world of digital done "right" on the analog, clocking, and power supply side. Their 44.1 Khz is arguably as good as 44.1 Khz can get. It is a LOT different than the 44.1 Khz in my MOTU 896. I've rented his gear to bypass the MOTU and it is no comparison.

Not many can afford to multitrack with Lavry "Gold" converters or a rack full of EMM Labs DSD stuff. So this leads to all kinds of "non-scientific"/limited value observations like, "My Ultraclarity RME-MOTU-whatever box sounds better at a different rate." It just might due to a lot of implementation issues, but that is not what Dan is addressing. He is arguing for designers to focus on what matters in improving the sound of converters, not spending unnecessary effort raising sample rates.

I don't think that the "but I like my "SuperLinear XL-10's better at 96Khz" is any real comment on Dan's post. He is dealing with hard math and has a reference implementation to confirm the science. For all the rest of us, this is a chance to shut up and listen and try to work out how it applies to us. But I don't know if there's anyone who posts here except maybe some of the experienced mastering engineers and a few moderator types that have any kind of informed opinion to argue with Dan.

I know I don't, compared to anything Dan or Glenn Z would say. I just know that Dan's converters are really, really nice to use, and I have no issue with them at 44.1 Khz for really fussy acoustic music destined for CD's.

[grufti]

That thread has some great contributions. GlenZ [Z-Systems] comments on filter design are interesting, and there are plenty of nuggets in all of Dan's posts. I just wish more of the EE types who design AD chips would offer up some of their wisdom.

[Bob Olhsson]

Something I think a lot of people are missing is that Dan is talking about 192 vs. 96 and not 96 vs. 48.

[Ruphus]

Originally posted by Bob Olhsson
Something I think a lot of people are missing is that Dan is talking about 192 vs. 96 and not 96 vs. 48.

Isn´t he saying 44.1 or at the latest 48k were plenty?

Ruphus

[Auxillary]

He is right that 192 is not necassary. Upsampling has its uses during certain processing, for conversion there is no gain from 192. If a 192khz converter sounds better, it is because it was design better. Or designed to work better at that rate. Not the sample rate.

[recordista]

Originally posted by Ruphus
Isn´t he saying 44.1 or at the latest 48k were plenty?

Actually, he says that 48k is a definite imrovement over 44.1 but that 60k is probably ideal. IOW 88.2 or 96k is more than enough.



During the '80s many of us were discussing 64k as an "if only they had..." fantasy.

[Ruphus]

Originally posted by recordista
Actually, he says that 48k is a definite imrovement over 44.1 but that 60k is probably ideal. IOW 88.2 or 96k is more than enough.



During the '80s many of us were discussing 64k as an "if only they had..." fantasy.

I see, thanks a lot.

Ruphus

[Nathanael]

Originally posted by recordista
Actually, he says that 48k is a definite imrovement over 44.1 but that 60k is probably ideal. IOW 88.2 or 96k is more than enough.



During the '80s many of us were discussing 64k as an "if only they had..." fantasy.

That's interesting because Dan, in his exchange with Mike Rivers seems to indicate that there has been some credible research that extends up to about 26 Khz. Double it for Nyquist, and you've got a bit over 53 Khz, and 64K makes sense as the closest "power of two" frequency.

I'm too young to know what went on in the boardrooms as to how sample rates were selected, but interesting to see what's happened as time and knowledge have grown.

[Nathanael]

This thread is also suggesting a way to connect the dots regarding some anecdotal evidence from the web.

It seems that all of the respected mastering engineers who post here and at other sites all prefer analog outboard to digital gear (with the exception of the TC 6000, Weiss, and Z-Sys gear).

Could this be as simple as Glenn Z's comments regarding the difficulty of good filter design and the requisite DSP horsepower at higher sample rates?

Is this the causality for what appears to be a universal preference for high quality analog (no filter coefficients that expand with the sample rate?)

Another interesting question is to posit whether the preference is for analog because it sounds good at all frequencies and is therefore a better business decision, or whether it just in an absolute sense sounds better.

And, if it does indeed "sound better more often", why?' Is it as simple (complicated) as filter coefficients.

Just thinking out loud... and glad I'm not the one who is responsible for filter coefficients for anything.

[Auxillary]

Filter coefficients can be perfected mathematically beyond the boundaries of human perception. The question in all of this now is quality versus efficiency. A assure you digital filters can be made very well with a good design, something not akin to modern cookbook methods. But it will cost you processing power, and occasionally non-realtime use. That is to say, a delay that would be otherwise unacceptable in an analog realm.

If you want a perfect digital filter, you can have it. If you want it for less than the price of a reissue concord, it is time to wake up. Of course it becomes subjective to what IS a perfect filter. One persons great use maybe another person's clinical sound or boxy. To satisfy the goal of perfection you must first entirely quantify the rules of quality.

[Bob Olhsson]

I can actually tell you how the sample rates were selected.

Early research suggested that 50 kHz was about ideal. The problem was that there was no way to edit it without using a mainframe computer which was unaffordable. Sony, JVC and Denon had been experimenting with converting digital audio to video and then using video editing equipment to edit it and a video disk to replicate it. The highest frequency that you could put on video tape or a video disk was 44.1 kHz. Sony took their video encoding and editing technology and married it with Phillips' video disk to create the compact disk.

Many people were unhappy with such a low sample rate so the SMPTE and the AES launched a research project to determine an ideal sample rate and bit depth for professional recording and video production. The 48 kHz x 20 bit standard was chosen for a combination of sound quality and ease of synchronization with video.

[Nathanael]

Thanks to both Bob O, and Auxillary for clearing up the last details for me.

Leaves me quite content to stick with 96Khz as an "overkill" resolution, and continue saving for a pair of Millenia Origen STT-1's to do all processing while Moore's Law works to make the good digital more affordable...

[BasPer]

If we are concerned about digital processing quality it is more bits we need, not higher sample rates. For recording we don't even need more bits. And the reason analog processing - sometimes - sound better has nothing to do with filter coeficients; the reason is that we have learned to love it - warts and all. (I'm pretty sure I wouldn't have thought much of the sound of my P-bass if Jamerson had played a Pedulla.)

[Roland]

I saw in the Mastering forum a post that quoted a lot of what Dan was saying. From what I read he implied that probably 60k sampling frequency was optimum and that he had no real problem with 96. His arguement seemed to be that by going to 192 it made the calculations that much more difficult thus leading to potentially more errors, worse rather than better than 96. He did however seem to make a good case for greater bit depths (this is something I have always been a personal fan of) over higher sampling frequencies. By this I mean 20-24 bit rather than 16 bit. The article I saw seemed to hit back at those that suggested if you sampled 20khz sine wave at 44.1 you would end up with something that looked like a sawtooth waveform and Dan stated that if it were below Nyquist it would be perfectly regenerated subject to the analogue electronics side. There are of course plenty of people in the manufacture side that would of course be more than happy for us to continuely upgrade our sample rates, but their interests could be seen as "vested" and not neccessarily ours as audio/music people.

regards to all.

Roland

[alphajerk]

Originally posted by Roland
The article I saw seemed to hit back at those that suggested if you sampled 20khz sine wave at 44.1 you would end up with something that looked like a sawtooth waveform and Dan stated that if it were below Nyquist it would be perfectly regenerated subject to the analogue electronics side.

thats the crux. apparently a 20khz "wave" cant be determined by humans as triangle, square, or sine so that is what the convertor designers use as an excuse for "putting" the signal back properly. it really ISNT put back together ACCURATELY, but its supposedly audibly the "same"...

[Roland]

Originally posted by alphajerk
thats the crux. apparently a 20khz "wave" cant be determined by humans as triangle, square, or sine so that is what the convertor designers use as an excuse for "putting" the signal back properly. it really ISNT put back together ACCURATELY, but its supposedly audibly the "same"...

As I understood it from the article that it is put back exactly as a signwave if it is below nyquist, providing the electronics are correct. It seems that it is a common misconception that a sawtooth is what you end up with and it isn't!!

Regards


Roland

[Bob Olhsson]

There are really two issues.

The first is the fact that there is a physical mechanism in human hearing that low-passes frequencies at around 24 kHz.

The second is that making a filter that low-passes frequencies at 24 kHz. without generating audible distortion artifacts below 24 kHz. is extremely difficult and expensive.

It's true that you can't hear the difference between a 20 kHz. sine wave and sawtooth however you can hear (and measure) artifacts when most low-pass filters are introduced into a circuit.

[Roland]

Originally posted by Bob Olhsson
There are really two issues.

The first is the fact that there is a physical mechanism in human hearing that low-passes frequencies at around 24 kHz.

The second is that making a filter that low-passes frequencies at 24 kHz. without generating audible distortion artifacts below 24 kHz. is extremely difficult and expensive.

It's true that you can't hear the difference between a 20 kHz. sine wave and sawtooth however you can hear (and measure) artifacts when most low-pass filters are introduced into a circuit.

I agree, thats why I think that the well known old chesnut of the eq applied at 50khz can be heard by people. Sure none of us are hearing up there, however the knock on harmonic effect is clearly audible.

Regards


Roland

[alphajerk]

Originally posted by Bob Olhsson
It's true that you can't hear the difference between a 20 kHz. sine wave and sawtooth however you can hear (and measure) artifacts when most low-pass filters are introduced into a circuit.

its not really between a sine and sawtooth... both are generally the same thing, its between a sine and square with inhabit two entirely different tones that they claim are inaudible above 15khz. i think otherwise.

and things like cymbals which carry the majority of their energy above 40khz that is effectively cut out by lower sampling rates or distorted by inaccurate plotting of the dynamics should probably be considered instead of simply dismissed.

i think far too many people take things for FACT and dismiss ANY possible reason that FACT is not really TRUTH.

[Nathanael]

You are missing the point of Nyquists work. Whether 20Khz sine or triangle, it is reproduced EXACTLY at the output of the DAC. That is what Dan is saying.

He's not even arguing against 96Khz, as Bob O already pointed out. He's just saying that 192 is a waste of time, energy, and money because it causes more problems, and offers no benefits.

96Khz also puts the filter way, way outside even an extended audio band.

[alphajerk]

no. its not EXACT. run a 20khz square wave ADAC and the result WONT be a SQAURE wave @ 48khz. it will be a sine wave... the resolution cannot determine if it is a square or sine but it is ASSUMED that humans cannot tell the difference between the two @ 20khz.

[RKrizman]

Originally posted by alphajerk
[B]its not really between a sine and sawtooth... both are generally the same thing, its between a sine and square with inhabit two entirely different tones that they claim are inaudible above 15khz. i think otherwise.

Well, on my analog synths a sine and a sawtooth sound extremely different.
But in any case, no matter which wave you're talking about, the first harmonic of a 15k wave will be 30k, which most people believe is inaudible. For any of those waves, all you'll hear at 15k is the fundamental, which can be represented by a sine wave.

-R

[Roland]

But then this becomes very accademic. In real terms Square waves don't exsist at those frequencies. Maybe on a scope, and using a signal generator. Look at that same squarewave on a 2" tape machine. Of course implimentation of digital filtering and DAC's will all have an effect. Having come through the generations of using tape I hear a lot of garbage talked about digital both good and bad. Sure its not perfect, but I have heard digital recordings where switching between "live feed" and AD/DA converted signal it was very, very difficult to tell which was which and many of the good engineers I have worked with would (certainly off the record) also agree. This industry has been so subject to "Emperors new clothes syndrome" over the last ten years it is sad. If as much effort had gone into making great records perhaps we would be all doing better business.

Certainly one very major record label that I had occasion to talk to one of their senior engineers confessed that they were perfectly happy recording 44.1khz albeit at 24 bit depth. The particular products that they are recording has by its nature to have a very long shelf life in order to recoup their costs. Being that they certainly have the equipement and resources to record at 96 or 192 or DSD if they preferred you have to take them seriously. Quality is most certainly an issue for this company as their projects cost 100,000 - 1,000,000 pounds a shot.

Regards to all.



Roland