192 khz vs 48 khz poll

[Jud]

I'm a bit busy, so I'll just give some short form answers...

 

 

Very much appreciate you taking the time, Miska.

[PopPop]

This is one of the best threads I have yet to read in CA. It has cleared up a lot of confusion I have had about perceptions and expectations for high sample rates. That you are breathless with anticipation just cracks me up, Jud.

 

I still believe this but my head is starting to hurt.

 

Please do! This is perhaps one of the most enjoyable threads I have seen in a long while.

 

 

I. . . . . . . I think it could be good for many people here. Or just keep expounding on it in this thread. Maybe one of my threads for a change won't be hated by half the people reading CA. . . .

 

The discussion of sample rates has evolved quite nicely, thanks for posing the question.

[Fokus]

Anything in the input that exceeds filter's pass-band will result in ringing.

 

Anything in the input that falls within the filter's transition band results in ringing. Exceeding the transition band is fine.

 

Think about it. We are talking about linear systems.

 

 

Example: take a 192kHz signal space. Build a single-sample impulse. Then filter it with a minimum phase band filter, a very steep one, centered on 22kHz. Then downsample to 44.1kHz with a steep linear phase half band filter. Pre-ringing? No.

[Paul R]

I appreciate your opinions about someone you do not know, but I will reserve the right to do my own thinking, thank you. And in truth, only time will tell about this fellow.

 

My training, as pertains to mathematics, is primarily oriented towards physics, though my degree is in Computer Science with minors in mathematics and a couple other subjects. Outside of being "just an IT guy for an insurance company in Texas (i.e. Sr. Software and Sr. Systems Engineer)" I am part of a group of highly skilled mathematicians, physicists, a sociologist or two, an urban planner, and computer scientists working to try and prove out some ideas about large scale computing environments. A really mixed group all of whom are bright and interested in - well - just about *everything*. It's fun, and one reason I am Austin - UT is here and a lot of my friends are associated with that institution.

 

I have a few other activities to keep my past experiences in other fields pseudo current. That experience includes the physics of underwater sound propagation, detection, and analysis, which sometimes colors my thinking about what is important about home audio, and what can and does make actual physical differences. I am also quite open as to what will make audible differences, and think that the "hard" lines some adhere to may not be so hard at all. That certainly does not make me an audio engineer, and I don't design power supplies for audio equipment or DACs, but I am very interested in such activities. Even more so, when, as in this thread, the topic veers to the real math behind most of it.

 

-Paul

 

 

 

Music and audiophilia are passionate hobbies, and the mathematical discussions starting to take place is one I absolutely love. So far, I have not seen much to dispute, though as I am sure you have noticed, your original poll question is not precise enough to answer "yes" or "no" to. I still think a lot of the concern you have about "people having misapprehensions about digital audio" is misplaced, but this one worked out just great.

 

 

I don't know wakibaki, but have seen his posts elsewhere. He likely isn't cribbing any website. So don't proceed thinking that the case.

 

I primarily have been trained in the analog realm and sort of like someone growing up in the USA doing things in the metric system, I think in the way I learned with some transposition. His explanation helped in that regard along with what Rob, and Miska has posted.

 

I would like to see Miska, Rob and whomever has some knowledge do a thread on this filtering and related matters. I think it could be good for many people here. Or just keep expounding on it in this thread. Maybe one of my threads for a change won't be hated by half the people reading CA.

[VandyMan]

If one accepts that 96Khz is useful because it results in better performance at the filter stage, is there any reason to think that a 96Khz source is superior to 48Khz upsampled? They both should be identical below 20 Khz, so does it matter if the data above 20 KHz is generated vs from the source? I don't know the answer, but I tend to think that upsampled should be the same given that very few of us can hear above 20 Khz.

[wakibaki]

My passion has always been for RF design, I wanted to be able to make the magic boxes where you spoke into one and your voice came out another, miles away. My father was a Radio Officer in the Merchant Navy in WWII. I learned the magic; I hope I made him proud.

 

I studied DSP under Graham Wade at Plymouth University, a master of obscurantism and a thoroughly aggravating lecturer who started at the end and only occasionally worked his way back to the beginning. Reading his book was reminiscent of attempting a cryptic crossword. I don't know who was more surprised, he or I, when I came top in the final exam. I well recall the day I explained one of his more obscure pronouncements in simple terms to Colin, who was sitting next to me. 'Is that it?' he said, 'IS THAT IT?'

 

I was fortunate to attend an advanced DSP course run by Bob Stewart of Strathclyde University as part of my CPD while working at Lucent. A brilliant, straight-talking man, with a talent for explaining things. DSP has become increasingly important in radio.

 

I was forced into retirement about 4 years ago when I had surgery for the first of two primary lung cancers, my oncologist is concerned that I am having another relapse, I am waiting for confirmation one way or another. The cancer was discovered when I was taken into hospital suffering from hyperthyroidism, a condition which makes one short-tempered. While this has been treated, the probability that I have very little time left does not encourage me to suffer fools gladly.

 

There it is. Believe it if you like.

 

w

 

B.Eng (Hons) AMIEE

[RobbieC]

 

There it is. Believe it if you like.

 

w

 

B.Eng (Hons) AMIEE

 

Like you, I joined this forum in 2011. I rarely if ever post but I can say I enjoy many of the subjects that come up here and visit the site regularly. I personally like the contributions you've added (at least to this thread) and hope that your oncologist's concerns turn out to be a false alarm.

[mayhem13]

My passion has always been for RF design, I wanted to be able to make the magic boxes where you spoke into one and your voice came out another, miles away. My father was a Radio Officer in the Merchant Navy in WWII. I learned the magic; I hope I made him proud.

 

I studied DSP under Graham Wade at Plymouth University, a master of obscurantism and a thoroughly aggravating lecturer who started at the end and only occasionally worked his way back to the beginning. Reading his book was reminiscent of attempting a cryptic crossword. I don't know who was more surprised, he or I, when I came top in the final exam. I well recall the day I explained one of his more obscure pronouncements in simple terms to Colin, who was sitting next to me. 'Is that it?' he said, 'IS THAT IT?'

 

I was fortunate to attend an advanced DSP course run by Bob Stewart of Strathclyde University as part of my CPD while working at Lucent. A brilliant, straight-talking man, with a talent for explaining things. DSP has become increasingly important in radio.

 

I was forced into retirement about 4 years ago when I had surgery for the first of two primary lung cancers, my oncologist is concerned that I am having another relapse, I am waiting for confirmation one way or another. The cancer was discovered when I was taken into hospital suffering from hyperthyroidism, a condition which makes one short-tempered. While this has been treated, the probability that I have very little time left does not encourage me to suffer fools gladly.

 

There it is. Believe it if you like.

 

w

 

B.Eng (Hons) AMIEE

 

Firm......but fair reply. Your previous posting style could never have conveyed an explanation better than you just expressed.

 

Might I offer a sincere thought that your condition improves and wish you the best.

 

Would it be inappropriate to offer something? It would seem that this discussion is hinging on some theoretical approaches to the value of 192 kHz. What's been presented so far is based on our understanding and capabilities to date. While Miska creates a compelling arguement for, again, we're really talking theory. As audiophiles, the thought of such advances often excites us past the limits of our sensibilities. That's not an insult, but human nature.

 

This being said, for Paul, Fokus and yourself,.....please consider that Robby has done am extremely thorough job of explaining filtering processes in a way that layman can comprehend AND has been quite responsible in acknowledging the possibilities of a theoretical solution to the problem. It would be quite unfortunate for this thread to take a turn towards bickering which might overshadow the time and patience that Robby has committed to this thread.

 

Thank you

 

And Paul

And Fokus

And Miska

[Paul R]

Sorry to hear about your medical troubles - I have no problem cutting you some slack for being a grouchy crook. Were I in your position right now it might be that I would be far less tolerant that you are.

 

No insult, but I sincerely hope I never have to find out how I would behave in that position. My Dad died from lung cancer, and it still pisses me off because I think it was a medical foul up that they did not catch it until there was no hope at all.

 

On the other paw (and a bit off topic) - if you feel like it - why don't you post your memories of your dad in the Merchant Marine in a topic or blog? WWII is another hobby of mine, and I am always eager to hear stories from that era, even second or third hand.

 

I think you are a Scot? Even more interesting to me and I know next to nothing about WWII Merchant Marine stuff outside of the U.S. Double true about communications of the era.

 

-Paul

 

 

My passion has always been for RF design, I wanted to be able to make the magic boxes where you spoke into one and your voice came out another, miles away. My father was a Radio Officer in the Merchant Navy in WWII. I learned the magic; I hope I made him proud.

 

I studied DSP under Graham Wade at Plymouth University, a master of obscurantism and a thoroughly aggravating lecturer who started at the end and only occasionally worked his way back to the beginning. Reading his book was reminiscent of attempting a cryptic crossword. I don't know who was more surprised, he or I, when I came top in the final exam. I well recall the day I explained one of his more obscure pronouncements in simple terms to Colin, who was sitting next to me. 'Is that it?' he said, 'IS THAT IT?'

 

I was fortunate to attend an advanced DSP course run by Bob Stewart of Strathclyde University as part of my CPD while working at Lucent. A brilliant, straight-talking man, with a talent for explaining things. DSP has become increasingly important in radio.

 

I was forced into retirement about 4 years ago when I had surgery for the first of two primary lung cancers, my oncologist is concerned that I am having another relapse, I am waiting for confirmation one way or another. The cancer was discovered when I was taken into hospital suffering from hyperthyroidism, a condition which makes one short-tempered. While this has been treated, the probability that I have very little time left does not encourage me to suffer fools gladly.

 

There it is. Believe it if you like.

 

w

 

B.Eng (Hons) AMIEE

[Daudio]

... again, we're really talking theory...

 

It has been fun to follow these discussions in the latter half of this odd thread. I wanted to agree that we are really, only, talking theory: conjecture, supposition, informed guesses. And in a limited area of a complex system.

 

To be truly scientific, one needs evidence. The evidence I find the most compelling is that of a well known and respected recording engineer. His observations from finely calibrated instruments indicate that 24/192 is perceptibly more accurate then lower resolutions.

 

There are number of theories that could explain that, some discussed in this thread. Pick your poison, hash it out, and let me know when you decide which one you like the best. Because, while I'm curious, I really don't care how it got there, just that that evidence sets the goalposts for me.

 

 

Carry on....

[Jud]

I was forced into retirement about 4 years ago when I had surgery for the first of two primary lung cancers, my oncologist is concerned that I am having another relapse, I am waiting for confirmation one way or another. The cancer was discovered when I was taken into hospital suffering from hyperthyroidism, a condition which makes one short-tempered. While this has been treated, the probability that I have very little time left does not encourage me to suffer fools gladly.

 

There it is. Believe it if you like.

 

w

 

B.Eng (Hons) AMIEE

 

I too hope your oncologist's concerns turn out to be unfounded.

 

I haven't run into many fools here, but your criteria may be more stringent. And there is always the Ignore list as a last resort (on your Settings page).

[Jud]

It has been fun to follow these discussions in the latter half of this odd thread. I wanted to agree that we are really, only, talking theory: conjecture, supposition, informed guesses. And in a limited area of a complex system.

 

To be truly scientific, one needs evidence. The evidence I find the most compelling is that of a well known and respected recording engineer. His observations from finely calibrated instruments indicate that 24/192 is perceptibly more accurate then lower resolutions.

 

There are number of theories that could explain that, some discussed in this thread. Pick your poison, hash it out, and let me know when you decide which one you like the best. Because, while I'm curious, I really don't care how it got there, just that that evidence sets the goalposts for me.

 

 

Carry on....

 

I have ideas for some interesting experiments that might be done. I think it's more appropriate to PM some people about them rather than putting these ideas out onto the forum just now.

[Miska]

Anything in the input that falls within the filter's transition band results in ringing. Exceeding the transition band is fine.

 

Well yes and no, depends on exact definition of the statement. Odd harmonic component of square wave doesn't have to hit the transition band to make the resulting step response ring. But if the fundamental is also above the transition band, result is dead silence (given that the stop band attenuation is higher than quantization resolution).

 

Example: take a 192kHz signal space. Build a single-sample impulse. Then filter it with a minimum phase band filter, a very steep one, centered on 22kHz. Then downsample to 44.1kHz with a steep linear phase half band filter. Pre-ringing? No.

 

Only if the linear phase filter has higher fc than the minimum phase filter and the transition bands don't overlap.

[mayhem13]

It has been fun to follow these discussions in the latter half of this odd thread. I wanted to agree that we are really, only, talking theory: conjecture, supposition, informed guesses. And in a limited area of a complex system.

 

To be truly scientific, one needs evidence. The evidence I find the most compelling is that of a well known and respected recording engineer. His observations from finely calibrated instruments indicate that 24/192 is perceptibly more accurate then lower resolutions.

 

There are number of theories that could explain that, some discussed in this thread. Pick your poison, hash it out, and let me know when you decide which one you like the best. Because, while I'm curious, I really don't care how it got there, just that that evidence sets the goalposts for me.

 

 

Carry on....

 

well said.......

[esldude]

Thanks to all those with great replies and the explanations. Wakibaki's made the most sense to me, and then allowed me to understand some of the others better.

 

I do have some questions in how this relates to actual results when we are talking SDM converters which appear most prevalent in current products. Those run at far higher than nyquist frequencies for our audio purposes and do the digital filtering that way, and among other things don't run at 24 bits, but rather a lower number like 6 bits.

 

If one is comparing an SDM running at 128 FS and one running at 256 FS will the higher FS SDM be able to use the higher rate to do comparable filtering with better time domain response (mainly less ringing)? It would seem to me it does as the transition band between say 24 khz (from 48khz sample rates) would allow the same effective result with less filter steepness in the 256 FS given as an example above.

[Fokus]

If one is comparing an SDM running at 128 FS and one running at 256 FS will the higher FS SDM be able to use the higher rate to do comparable filtering with better time domain response (mainly less ringing)?

 

No. The requirements for the filtering are set by the source signal, and not by what follows.

If the source is 48kHz, then anti-imaging filtering must exist at 24kHz.

[manhao]

Thanks for the info.

[Paul R]

Well, now you bring up a point I would like to know more - much more - about. I understand Sigma Delta Modulation and the associated filters better in relation to A2D than in D2A conversions. In other words, it gets a little mysterious to me in a DAC, especially how they are achieving high S/N ratios.

 

My first thought about your question is no, it would have to be tied to the input signal. But thinking about that, I then became less sure. That means there is something there I know I don't know, just that I don't know I know it. (grin)

 

 

 

Thanks to all those with great replies and the explanations. Wakibaki's made the most sense to me, and then allowed me to understand some of the others better.

 

I do have some questions in how this relates to actual results when we are talking SDM converters which appear most prevalent in current products. Those run at far higher than nyquist frequencies for our audio purposes and do the digital filtering that way, and among other things don't run at 24 bits, but rather a lower number like 6 bits.

 

If one is comparing an SDM running at 128 FS and one running at 256 FS will the higher FS SDM be able to use the higher rate to do comparable filtering with better time domain response (mainly less ringing)? It would seem to me it does as the transition band between say 24 khz (from 48khz sample rates) would allow the same effective result with less filter steepness in the 256 FS given as an example above.

[Jud]

If one is comparing an SDM running at 128 FS and one running at 256 FS will the higher FS SDM be able to use the higher rate to do comparable filtering with better time domain response (mainly less ringing)?

 

No. The requirements for the filtering are set by the source signal, and not by what follows.

If the source is 48kHz, then anti-imaging filtering must exist at 24kHz.

 

In relation to VandyMan's question above and the topic of this thread, is there anything about anti-alias filtering at 96kHz that is more beneficial for sound quality than anti-alias filtering at 24kHz?

[Fokus]

In a nutshell, simplifying a lot, and for the 44.1k input case, and audio SDM DAC oversamples with a factor of, say, 128, to a long-word 5.6MHz stream.

 

Typically the first 8x in this oversampling houses the anti-imaging/reconstruction/interpolation filter for the source signal, in our case cutting steeply at 22kHz. The rest of the oversampling is done with progressively cruder methods, e.g. first linear interpolation and finally even just zero over hold (copying samples). These steps too have some sort of filter effect, but overall it is rather ineffective and of little interest to discussions as these.

 

Then the long-word oversampled stream is truncated to the short-word format in the DS modulator. The gross errors thus created are fed back through filtered loops, which shift their spectrum upwards, and away from the baseband of interest, i.e. < 22kHz.

[Miska]

One note, let's not mix up ADC and DAC side. I've been mostly talking about ADC side because that's where the damage is done. There is only limited amount that can be done fixing that damage afterwards (apodizing filters).

 

In relation to VandyMan's question above and the topic of this thread, is there anything about anti-alias filtering at 96kHz that is more beneficial for sound quality than anti-alias filtering at 24kHz?

 

Yes, because it allows for the signal to naturally roll-off, at best completely and therefore there is no smear from the anti-alias filter.

 

So a normal SDM ADC running at 5.6 MHz doesn't need much analog anti-alias filtering because it's Nyquist frequency is at 2.8 MHz. So for 192k PCM output it needs to gain 144 dB attenuation only by 5600 - 96 = 5.504 MHz.

 

Then the SDM to PCM digital domain conversion needs to cut this more sharply using a brickwall filter at 96 kHz, both to avoid noise shaping noise from the modulation process from aliasing in, and also from any potential audio information exceeding 96 kHz from aliasing. However, many ADC chips pass only about 80 kHz of the band and roll-off slower so they use about 32 kHz wide transition band (luxury not available for 44.1/48k sampling rates).

[Miska]

If one is comparing an SDM running at 128 FS and one running at 256 FS will the higher FS SDM be able to use the higher rate to do comparable filtering with better time domain response (mainly less ringing)? It would seem to me it does as the transition band between say 24 khz (from 48khz sample rates) would allow the same effective result with less filter steepness in the 256 FS given as an example above.

 

Advantage of using higher fs multiplier for SDM is mostly to improve dynamic range in audio band and push the noise higher up. What ever is the lowest sampling rate used in the system determines the necessary filters and system's transient response.

[Miska]

I recorded some sounds in order to demonstrate the ringing behavior at 48k sampling rate. I didn't have much time to spend on this, so I didn't spend more than an hour. And my cheap measurement mics are not really fast enough since their output falls into background noise around 65 kHz. So 192k sampling rate was enough to avoid any ringing (time domain smear). Subjectively, even at 192k the recorded sound is still far from live sound. Maybe some day I'll get some better mics...

 

Interesting thing is that even with headphones, I have to turn up the headphone amp up to max volume since the content is not compressed at all to make it even a bit closer to live volume levels. And still couple of loudest peaks clipped in ADC. I'll have to check at some point the exact SPL of the instruments.

 

Best example of my instruments I found within this quick experiment was a glockenspiel. Here's initial hit at 48k sampling rate and linear waveform scale, note the pre-ringing ripple before initial rise:

 

And when we switch into dB level view it becomes much more obvious:

 

(in this sample case, both channels are volume normalized to 0 dB, that's why there's "clipping" indicated, although there is not really clipping, those peaks were originally safely way below 0 dBFS)