The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
192kHz sampling = loss of accuracy? Synchronisers/Clock Generators
Old 7th January 2011
  #1
Gear Addict
 
Tuberizer's Avatar
 

192kHz sampling = loss of accuracy?

Yo experts heh

I was reading the Dan Lavry whitepaper (pdf) on sampling theory the other day but I'm not much of a tech so I understood it only half I think. Anyways, what he basically says is that he is against 192kHz sample rate, because there is an inescapable tradeoff between faster sampling on one hand and a loss of accuracy, increased data size and much additional processing requirement on the other hand. Lavry says that sampling audio signals at 192KHz is about 3 times faster than the optimal rate. It compromises the accuracy which ends up as audio distortions. That's why Lavry converters only go up to 96kHz.

What do you guys think about that?
Old 7th January 2011
  #2
Lives for gear
 

Quote:
Originally Posted by Tuberizer View Post
Lavry says that sampling audio signals at 192KHz is about 3 times faster than the optimal rate.
This may well be true. But there's not really enough definitive evidence to say that 192kHz sampling is pointless, yet.

Quote:
Originally Posted by Tuberizer View Post
Anyways, what he basicly says is that he is against 192kHz sample rate, because there is an inescapable tradeoff between faster sampling on one hand and a loss of accuracy, increased data size and much additional processing requirement on the other hand. It compromises the accuracy which ends up as audio distortions. That's why Lavry converters only go up to 96kHz.
The "inescapable tradeoff" is BS. Or, rather, it's only true if you assume that 192kHz converters should cost the same as 96kHz converters. If you take his extreme and assume that the two converters share the same front-end design, so that the 192kHz does less decimation (assuming this is a sigma-delta converter we are talking about) then his argument about additional distortion is broadly correct. But, to improve performance – and chipmaking technology has advanced a long way since the first 96kHz converters appeared – a real-world 192kHz should run its front-end faster to get the right amount of oversampling performance to feed its decimation filter. Or, put another way, if you buy a cr*p 192kHz converter, it is more likely to sound cr*p.

As an example of how things have improved, Lavry claims 100MHz converters sample to 8bit resolution. That was true ten years ago. Parts are on the market that will sample to 16bit at 200MHz. Why? Because the mobile comms industry needed them to work at that level of accuracy.
Old 7th January 2011
  #3
Lives for gear
 

Quote:
Originally Posted by ZangTumblyTumble View Post
This may well be true. But there's not really enough definitive evidence to say that 192kHz sampling is pointless, yet.



The "inescapable tradeoff" is BS. Or, rather, it's only true if you assume that 192kHz converters should cost the same as 96kHz converters. If you take his extreme and assume that the two converters share the same front-end design, so that the 192kHz does less decimation (assuming this is a sigma-delta converter we are talking about) then his argument about additional distortion is broadly correct. But, to improve performance – and chipmaking technology has advanced a long way since the first 96kHz converters appeared – a real-world 192kHz should run its front-end faster to get the right amount of oversampling performance to feed its decimation filter. Or, put another way, if you buy a cr*p 192kHz converter, it is more likely to sound cr*p.

As an example of how things have improved, Lavry claims 100MHz converters sample to 8bit resolution. That was true ten years ago. Parts are on the market that will sample to 16bit at 200MHz. Why? Because the mobile comms industry needed them to work at that level of accuracy.
I don't understand why 192 kHz is a loss in accuracy.

Todays high performance converters sample 2.5 GHz with about 8-9 bit resolution (using the best clock). And not some 100 MHz. LOL

Note that every doubling of sampling frequency also increases resolution by 3 dB (half a bit), because you can filter the noise later out.
Old 7th January 2011
  #4
Gear Addict
 
monsieur x's Avatar
I believe the white-paper's main point is that 192kHz is absolutely more about marketing and hype.

192 kHz is relatively pointless. It is nearly impossible to work at 192kHz and just doesn't make a big enough difference (whatever the difference is).

96kHz is amazing. If that became the new standard I'd be very happy because 44.1kHz is definitely an ugly compromise.

Cheers,
Old 7th January 2011
  #5
Lives for gear
 

Quote:
Originally Posted by HomeProducer View Post
I don't understand why 192 kHz is a loss in accuracy.
It isn't but Lavry attempts to make that point by (as far as I can tell from the white paper) assuming that you use 2x less decimation on the same front-end as a 96kHz sigma-delta converter.
Old 7th January 2011
  #6
Lives for gear
 

Quote:
Originally Posted by monsieur x View Post
I believe the white-paper's main point is that 192kHz is absolutely more about marketing and hype.
Unfortunately, Lavry uses marketing and hype to make the point.
Old 7th January 2011
  #7
Deleted User
Guest
Quote:
Originally Posted by monsieur x View Post
It is nearly impossible to work at 192kHz and just doesn't make a big enough difference (whatever the difference is).


Cheers,
dont say that around my studio... my DAW has been pretty happy (running up to 80 tracks) at 24/192 ..not sure what you mean by nearly impossible.

I will be the first to admit that after I have bounced down to an mp3 at 192kbs thru a limiter that it REALLY does not make any diff to that product. but since I can... i do .. I can almost alway archive to one dual layer DVD ... cheap and easy ...YMMV
Old 8th January 2011
  #8
Lives for gear
 

Quote:
Originally Posted by ZangTumblyTumble View Post
It isn't but Lavry attempts to make that point by (as far as I can tell from the white paper) assuming that you use 2x less decimation on the same front-end as a 96kHz sigma-delta converter.
A wrong assumption made by him, as you are going to do that anyway by hand. I mean you will decimate at the end to 44.1 kHz and this decimation gives you most likely a better processing gain than any sigma delta ADC, if you are using a better decimation filter. And if you stay at the high rate you have 3/4 of the quantization noise in the inaudible spectrum.

Most white papers are pointless anyway
Old 8th January 2011
  #9
Lives for gear
 
lagavulin16's Avatar
 

Quote:
Originally Posted by Tuberizer View Post
Yo experts heh

I was reading the Dan Lavry whitepaper (pdf) on sampling theory the other day but I'm not much of a tech so I understood it only half I think. Anyways, what he basically says is that he is against 192kHz sample rate, because there is an inescapable tradeoff between faster sampling on one hand and a loss of accuracy, increased data size and much additional processing requirement on the other hand. Lavry says that sampling audio signals at 192KHz is about 3 times faster than the optimal rate. It compromises the accuracy which ends up as audio distortions. That's why Lavry converters only go up to 96kHz.

What do you guys think about that?
It sounds like BS to me. I don't see how increasing the sampling rate will reduce accuracy. Yes, it will require more space and processing power but those improve every year.

We should keep going until the sampling rate is indistinguishable from an analog sound.
Old 8th January 2011
  #10
Moderator
 
narcoman's Avatar
 

well - I haven't yet heard a 192 converter that doesn't sound better at 96. That's the Radar ADA system, the Prism ADAs and the Lynx ones. Cant say about any others.

That's just an opinion, and I wouldn't have an explanation other than inability of current technology to clock properly at higher frequencies at 20bit or more depths.... Not looked at this in depth - and not read Dans paper {although it is highly regarded amongst many although not here it appears!} but I used to have the same issue with early 96 converters against the 48k setting years ago.

Done a lot on DSD, done a lot on 96 and even more at 48. Had nothing but poor experiences at 192.
Old 8th January 2011
  #11
Lives for gear
 

Bob Katz's book talks about this phenomenon - If i recall correctly he states the digital low pass filters required for high sampling rates cause the 'loss of accuracy' through time smearing. Also in the ADA there is the process of oversampling and decimating, the quality of the math contributes heavily to the sound quality... and some even use upsampling!

When investigating these things it shows you that 192 is not necessarily better than 96 and 96 is not necessarily better than 48. It all depends on the quality of the chips (the quality of the math) that they use to make the converters. That's why a better converter will sound better at every sample rate, and some sampling rates sound better than others on the same machine - it would follow they would spend more on 96 math than 192 math, since the former is used far more frequently.
Old 8th January 2011
  #12
Lives for gear
 

Quote:
Originally Posted by spice house View Post
Bob Katz's book talks about this phenomenon - If i recall correctly he states the digital low pass filters required for high sampling rates cause the 'loss of accuracy' through time smearing.
there is no time smearing. It doesn't exist. Even if it looks like such on a graph as ringing, there is no time smearing. Look at the impulse response output of your speaker, there you have the same effect. All BS.
Old 8th January 2011
  #13
Lives for gear
 
cdog's Avatar
Im pretty content with 44.1 tbh

192 is overkill imo

Old 8th January 2011
  #14
Moderator
 
narcoman's Avatar
 

Quote:
Originally Posted by spice house View Post
Bob Katz's book talks about this phenomenon - If i recall correctly he states the digital low pass filters required for high sampling rates cause the 'loss of accuracy' through time smearing. Also in the ADA there is the process of oversampling and decimating, the quality of the math contributes heavily to the sound quality... and some even use upsampling!

When investigating these things it shows you that 192 is not necessarily better than 96 and 96 is not necessarily better than 48. It all depends on the quality of the chips (the quality of the math) that they use to make the converters. That's why a better converter will sound better at every sample rate, and some sampling rates sound better than others on the same machine - it would follow they would spend more on 96 math than 192 math, since the former is used far more frequently.
yes better at every sample rate - but not better within the converter against itself at different rates. BUT - you have to bear in mind the argument for pushing the Nyquist limit upwards (not that I go for that in excess - but it is a point often made).

The quality of the chips is almost moot because there are many units using the same systems. In converters (and this is getting off topic) the analogue stages are "the thing".....
Old 8th January 2011
  #15
Lives for gear
 

Quote:
Originally Posted by HomeProducer View Post
there is no time smearing. It doesn't exist. Even if it looks like such on a graph as ringing, there is no time smearing. Look at the impulse response output of your speaker, there you have the same effect. All BS.
Do you have any sources other than 'even though it looks like it, it's not it?' I don't think bob katz is the be all and end all of digital passband circuit design so i'm interested to know.
Old 8th January 2011
  #16
Lives for gear
 

Quote:
Originally Posted by narcoman View Post
yes better at every sample rate - but not better within the converter against itself at different rates.
Case in point: The alesis hd24xr operates better at 48 than 44.1 - there is measurable difference.

That is specifically because of the math of the clock but in another unit it could be a the relationship of the decimator (at that sample rate) to the incoming sample rate. Choosing what data to throw out is not an easy process.
Old 8th January 2011
  #17
Lives for gear
 

Quote:
Originally Posted by spice house View Post
Do you have any sources other than 'even though it looks like it, it's not it?' I don't think bob katz is the be all and end all of digital passband circuit design so i'm interested to know.
take any textbook about signal processing. A sinc pulse has ringing, but it's the answer to a ideal delta. Now a delta pulse will never appear at the output! it doesn't sound nice heh Your output signal will have the typical rolloff frequency shape. Even transients will never look like a delta. Therefore, the ringing will NEVER appear in its nasty shape in practise.
Old 8th January 2011
  #18
Lives for gear
 

Quote:
Originally Posted by HomeProducer View Post
take any textbook about signal processing. A sinc pulse has ringing, but it's the answer to a ideal delta. Now a delta pulse will never appear at the output! it doesn't sound nice heh Your output signal will have the typical rolloff frequency shape. Even transients will never look like a delta. Therefore, the ringing will NEVER appear in its nasty shape in practise.
interesting! as much as i understand it that circuit (real world not ideal) would still have ripple through the passband and phase shift near the upper limit, as i understand it. That rippling wouldn't create any time based distortion (you call ringing?), would it create any other distortions?

If this sounds like greek to you take this away - its not just the sampling rate you are hearing, its also the filter(s)
Old 8th January 2011
  #19
Lives for gear
 
bassjam's Avatar
 

I recall being at AES San fran a few years ago.

I was chatting to Daniel Weiss about him building us a multi channel ADDA box, around 24 AD 120 DA capable of 96k. As we were near DAD's stand we spoke about DXD tracking and allthough massively expensive, Daniel said that there are good enough chips available to do it.

Later that day i spoke to someone at stienberg about my conversation with Daniel, the stienberg guy told me that you loose quality above 48k!

I went back and told Daniel, his words will forever make me smile......

(In his swiss accent) ........"That is total bull****!"

I know who I believe about higher sample rates!

The DAD boxes at DXD sound AMAZING!
Old 8th January 2011
  #20
Lives for gear
 

Quote:
Originally Posted by bassjam View Post
I recall being at AES San fran a few years ago.

I was chatting to Daniel Weiss about him building us a multi channel ADDA box, around 24 AD 120 DA capable of 96k. As we were near DAD's stand we spoke about DXD tracking and allthough massively expensive, Daniel said that there are good enough chips available to do it.

Later that day i spoke to someone at stienberg about my conversation with Daniel, the stienberg guy told me that you loose quality above 48k!

I went back and told Daniel, his words will forever make me smile......

(In his swiss accent) ........"That is total bull****!"

I know who I believe about higher sample rates!

The DAD boxes at DXD sound AMAZING!
I think you relaying comments back and forth between people caused on some confusion, no?

It seems to be the steinberg guy was saying, "in real world applications, because of manufacturers cutting costs in circuitry design in an effort to provide low cost alternatives to consumers, the performance of high sampling rates is often non-existent or destructive"

no competent person can state higher sampling rates = lower sound quality of course that's BS! It's hard to believe the steinberg guy could be that misinformed?
Old 8th January 2011
  #21
Lives for gear
 
bassjam's Avatar
 

Quote:
Originally Posted by spice house View Post
I think you relaying comments back and forth between people caused on some confusion, no?

It seems to be the steinberg guy was saying, "in real world applications, because of manufacturers cutting costs in circuitry design in an effort to provide low cost alternatives to consumers, the performance of high sampling rates is often non-existent or destructive"

no competent person can state higher sampling rates = lower sound quality of course that's BS! It's hard to believe the steinberg guy could be that misinformed?
Well that is what is being implied! Nyquist theorum dictates that in order to perfctly recreate a waveform it has to be sampled at just over twice the frequency of the highest frequency you want to be able to capture. The problem is the aliasing caused by using a bandpass filter. The higher the sample rate the higher you push up the bandpass filter but it causes folding lower down the frequency, like a ripple, caused by sinusoidal harmonics being generated. A bit like a mathematical standing wave distorting the fundemental frequency the same as what happens if you block the end of a tube thats resonating at fundemental frequency. Anti aliasing filters are used to stop this but Dan is arguing that it is affecting the source in a negative way.

Surely though incresing frequency increases resolution? Sound waves are not all sinusoidal so the highest frequency present is the fastest rate of change (dx) at any point in the waveform. Thefore you need to differentiate the waveform more times where the highest rate of change is in order to build an analogue like construct. Do you follow?

The higher the sample rate and pushing up the bandpass filter coupled with better aliasing filters (j-invariant) will result in a more analogue like resolution surely?
Old 8th January 2011
  #22
Moderator
 
narcoman's Avatar
 

Quote:
Originally Posted by spice house View Post
Case in point: The alesis hd24xr operates better at 48 than 44.1 - there is measurable difference.

That is specifically because of the math of the clock but in another unit it could be a the relationship of the decimator (at that sample rate) to the incoming sample rate. Choosing what data to throw out is not an easy process.
Yeah - that's the sort of thing I was implying - but I didn't say it very well!!
Old 8th January 2011
  #23
Lives for gear
 

Quote:
Originally Posted by spice house View Post
interesting! as much as i understand it that circuit (real world not ideal) would still have ripple through the passband and phase shift near the upper limit, as i understand it. That rippling wouldn't create any time based distortion (you call ringing?), would it create any other distortions?

If this sounds like greek to you take this away - its not just the sampling rate you are hearing, its also the filter(s)
Ringing is there, but only for extreme exampes, like a delta spike pulse or
square wave. A square ware cannot be ideally repoduced, because ot the limited bandwidth. A good example I find. You can search for the Gibbs phenonema stating that leaving out those high overtones creates a ringing and overshoot effect

Gibbs phenomenon - Wikipedia, the free encyclopedia

As you can read, the overshoots are limited to 9%. That's good! Because if your analog part of the DA can handle at least 9% overshoots, no distortion will arise. Now your converter has limited bandwidth, and your speaer also! Nobody speaks about the pulse effects of speakers. Why? Of course speakers do not fall off like reconstuction filters in frequency domain, but limited bandwidth effects still take place and phase shift is much higher. For 96 kHz and higher, there is no ADC/DAC phase shift problem at all, if you use a linear phase type reconstruction filter.
Old 8th January 2011
  #24
Lives for gear
 
GoldMember's Avatar
Quote:
Originally Posted by Tuberizer View Post
basically he is against 192kHz sample rate, because there is an inescapable tradeoff between faster sampling on one hand and a loss of accuracy, increased data size and much additional processing requirement on the other hand. Lavry says that sampling audio signals at 192KHz is about 3 times faster than the optimal rate. It compromises the accuracy which ends up as audio distortions. That's why Lavry converters only go up to 96kHz.
What do you guys think about that?
10KHz Square Wave @ 44.1Khz


10KHz Square wave @ 96Khz


10KHz Square wave @ 176.4Khz


10KHz Sqr wave @ 192KHz



http://en.wikipedia.org/wiki/Analog-...Aperture_error

http://www.audio.com/pro_masterclocks_cc1.htm

http://www.audio.com/product_info/Manual%20CC1%200.91.pdf


http://www.vintageking.com/Grimm-Audio-CC1

http://www.grimmaudio.com/whitepaper...k%20basics.pdf

http://www.audio.com/whitepapers/pll%20and%20clocking.pdf

http://www.blacklionaudio.com/content.php?p=3
Old 8th January 2011
  #25
Lives for gear
 
AllBread's Avatar
 

192 it is for me, then! I'm always looking for the finest possible reproduction of my square waves!
Old 8th January 2011
  #26
Lives for gear
 

I don't believe it will look that perfect
Old 8th January 2011
  #27
Lives for gear
 
GoldMember's Avatar
Quote:
Originally Posted by AllBread View Post
192 it is for me, then!
the best is needed:
Grimm audio cc1 or antelope trinity + 10M or Teac Esoteric G-rb
+ cardas digital or acoustic zen or proel diehard
+ ps audio power plant premier.
+ acoustic zen tsunami power cables or similar.

+ a very wide&fast AD&DA converter. (not all converters can accept an ultra low jitter clock = some will not improove.)
Old 8th January 2011
  #28
Lives for gear
 


I didn't see those graphs on your sources, did you run them yourself? If you did could you do a 1k wave as well?
Old 8th January 2011
  #29
Gear Addict
 

Quote:
Originally Posted by bassjam View Post
Well that is what is being implied! Nyquist theorum dictates that in order to perfctly recreate a waveform it has to be sampled at just over twice the frequency of the highest frequency you want to be able to capture. The problem is the aliasing caused by using a bandpass filter. The higher the sample rate the higher you push up the bandpass filter but it causes folding lower down the frequency, like a ripple, caused by sinusoidal harmonics being generated. A bit like a mathematical standing wave distorting the fundemental frequency the same as what happens if you block the end of a tube thats resonating at fundemental frequency. Anti aliasing filters are used to stop this but Dan is arguing that it is affecting the source in a negative way.

Surely though incresing frequency increases resolution? Sound waves are not all sinusoidal so the highest frequency present is the fastest rate of change (dx) at any point in the waveform. Thefore you need to differentiate the waveform more times where the highest rate of change is in order to build an analogue like construct. Do you follow?

The higher the sample rate and pushing up the bandpass filter coupled with better aliasing filters (j-invariant) will result in a more analogue like resolution surely?
Yes Sir !
Old 8th January 2011
  #30
Lives for gear
 
GoldMember's Avatar
Quote:
Originally Posted by spice house View Post
I didn't see those graphs on your sources, did you run them yourself? If you did could you do a 1k wave as well?
i could do it..
you could to...

buy Novation V-Station...
make a pure Square wave, 1x OSC, no LFO, No Filter, etc...
download free rs-met.com Signal Analyzer.
any vst software, reaper, samplitude, sonar, cubase, etc...

if you have wavelab or cooledit/adobe audition you can also generate any square wave.
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Similar Threads
Thread
Thread Starter / Forum
Replies
jedilaw / Rap + Hip Hop engineering and production
44
UpNSmoke / Electronic Music Instruments and Electronic Music Production
3
statixx / Rap + Hip Hop engineering and production
2
fifthflaw / Music Computers
14
recall / So much gear, so little time
5

Forum Jump
Forum Jump