Isn’t there more to digital audio than jitter reduction? And if you value jitter-free clocks so highly, why don’t you look at the jitter in the clocks that actually control the DAC chip instead of concentrating all your efforts on the XO.
For example, there is an asynchronous re-clocking circuit floating around the 'net that uses a low-jitter XO to generate the clocks that drive the SCLK and FSYNC pins of the CS8412 running in mode 3.That’s all well and good, except, like all asynchronous re-clocking efforts, the downside is occasional missed or duplicated samples. What’s more, that particular circuit uses an HC4040 to divide the XO clock rate. Have you looked at the spec sheet for the HC4040? It’s an asynchronous counter, sometimes called a ripple counter. Only the first stage is actually clocked by the input clock. Each of the following stages is clocked by the preceding stage. The Fairchild data sheet shows the interstage propagation delay to be not more than 25ns at 25C with VCC=4.5v. That means, with 5 stages between them, FSYNC will lag SCLK by as much as 125ns. That’s worse than the CS8412, which guarantees the SCLK/FSYNC delay to be not more than +/- 20ns. The 125ns max propagation delay would not be a problem if it were absolutely constant - but it’s not. Given the slow and unmatched, 10-15ns rise and fall time of HC logic and 5 clock stages in between SCLK and FSYNC there’s a large margin for jitter to be introduced. I’ll bet the reclocked FSYNC from the HC4040 has more jitter than the original FSYNC coming from the CS8412.
I know you all claim to hear improvements after your efforts to reduce jitter, but have you actually measured the jitter at the DAC chip? It could be that after you feed the pristine output of your low-jitter XO through a lot of shoddy digital circuitry, you end up with more jitter, not less. Jitter is like dither in that they both introduce a small amount of random error to the reconstructed audio signal and it has been shown that a small amount of dither actually helps with the retrieval of low-level information. A small amount of additional jitter will do the same thing.
Now, back to the original question: Isn’t there more to digital audio than jitter reduction? I think so. I just finished a very interesting new DAC design. It starts with a CS8412 and differential PCM1704s but adds several new wrinkles. One is sub-LSB dither; another is … (I think I’ll save that for another day. If all you want to talk about is jitter I don’t want to waste your time with extraneous concepts.)
For example, there is an asynchronous re-clocking circuit floating around the 'net that uses a low-jitter XO to generate the clocks that drive the SCLK and FSYNC pins of the CS8412 running in mode 3.That’s all well and good, except, like all asynchronous re-clocking efforts, the downside is occasional missed or duplicated samples. What’s more, that particular circuit uses an HC4040 to divide the XO clock rate. Have you looked at the spec sheet for the HC4040? It’s an asynchronous counter, sometimes called a ripple counter. Only the first stage is actually clocked by the input clock. Each of the following stages is clocked by the preceding stage. The Fairchild data sheet shows the interstage propagation delay to be not more than 25ns at 25C with VCC=4.5v. That means, with 5 stages between them, FSYNC will lag SCLK by as much as 125ns. That’s worse than the CS8412, which guarantees the SCLK/FSYNC delay to be not more than +/- 20ns. The 125ns max propagation delay would not be a problem if it were absolutely constant - but it’s not. Given the slow and unmatched, 10-15ns rise and fall time of HC logic and 5 clock stages in between SCLK and FSYNC there’s a large margin for jitter to be introduced. I’ll bet the reclocked FSYNC from the HC4040 has more jitter than the original FSYNC coming from the CS8412.
I know you all claim to hear improvements after your efforts to reduce jitter, but have you actually measured the jitter at the DAC chip? It could be that after you feed the pristine output of your low-jitter XO through a lot of shoddy digital circuitry, you end up with more jitter, not less. Jitter is like dither in that they both introduce a small amount of random error to the reconstructed audio signal and it has been shown that a small amount of dither actually helps with the retrieval of low-level information. A small amount of additional jitter will do the same thing.
Now, back to the original question: Isn’t there more to digital audio than jitter reduction? I think so. I just finished a very interesting new DAC design. It starts with a CS8412 and differential PCM1704s but adds several new wrinkles. One is sub-LSB dither; another is … (I think I’ll save that for another day. If all you want to talk about is jitter I don’t want to waste your time with extraneous concepts.)
Surely there is more to Digital Audio than Jitter.
It's just that at this point the DAC chips are made good enough, that you can barely hear difference. But there is great difference in the sound of different apparatus using the same DAC's. Why? Well assuming the data stream is the same, the only difference left can be the analog stage (which is very critical if you want to have a good sound performance) and ... yes: Jitter.
I agree completely after a clock recovery where the clock phase is depending on the data content of the SP/DIF signal, you can not claim to have a low jitter clock. This is why a low jitter clock should always be fitted in the DAC apparatus, and feed the DAC chips directly. And secondarily a clock signal should be fed back to the transport to make it ruin the same speed (synced operation).
All the best from Lars Clausen
It's just that at this point the DAC chips are made good enough, that you can barely hear difference. But there is great difference in the sound of different apparatus using the same DAC's. Why? Well assuming the data stream is the same, the only difference left can be the analog stage (which is very critical if you want to have a good sound performance) and ... yes: Jitter.
I agree completely after a clock recovery where the clock phase is depending on the data content of the SP/DIF signal, you can not claim to have a low jitter clock. This is why a low jitter clock should always be fitted in the DAC apparatus, and feed the DAC chips directly. And secondarily a clock signal should be fed back to the transport to make it ruin the same speed (synced operation).
All the best from Lars Clausen
By the way: You don't need to use the all expensive PCM1704's
Several new DAC chips offer as good or better performance (and sound...!) at a small fraction of the price for PCM1704.
Take a look on www.cirruslogic.com or www.akm.com . Both these firms have very very good DAC chips for almost no money.
CS4396 is a great standard converter, which i will recommend for CD use, but they have others that are even better, and also can convert DSD. Again, you may find that the DAC chip performance is not your bottleneck .. more likely jitter is!
Several new DAC chips offer as good or better performance (and sound...!) at a small fraction of the price for PCM1704.
Take a look on www.cirruslogic.com or www.akm.com . Both these firms have very very good DAC chips for almost no money.
CS4396 is a great standard converter, which i will recommend for CD use, but they have others that are even better, and also can convert DSD. Again, you may find that the DAC chip performance is not your bottleneck .. more likely jitter is!
I'll agree with Ulas on one point. There have been several DIY DAC designs that are certainly going to have repeated or dropped samples, just because no care is taken to ensure that the clock domains are synchronized in the DAC and transport. These same designs tend to voilate the setup and hold time requirements of the ICs involved. These are "cargo cult" designs, slapped together from promising-looking bits of other circuits.
But, that doesn't mean one shouldn't have a good clock in one's own DAC!
But, that doesn't mean one shouldn't have a good clock in one's own DAC!
I use VHC4040...
But I agree with first post...there is more than the DAC that needs the correct timing.
I have mentioned (in another tread) that I EASY could hear the difference when a (tunable) clock was out of tune/frequency compared to RIGHT frequency.
And I have measured a lot of X-tals since, and few are "right".
So I guess this also influence the diffrerence many hear when upgrading clock-circuits.
ArneK
But I agree with first post...there is more than the DAC that needs the correct timing.
I have mentioned (in another tread) that I EASY could hear the difference when a (tunable) clock was out of tune/frequency compared to RIGHT frequency.
And I have measured a lot of X-tals since, and few are "right".
So I guess this also influence the diffrerence many hear when upgrading clock-circuits.
ArneK
Although people may have different opinions on the practical
value of jitter reduction, at least this is something that can be
straightforwardly explained in theory. A timing error in the
digitial domain will translate into an amplitude error in the
analog domain. How much it matters in practice can be
discussed, but the effect is there according to the theory.
There is another aspect on the jitter problem that I have
been wondering about and that I cannot remember ever
seing discussed. While audiophiles have for quite some years
put a lot of effort into jitter reduction in their CDPs and there
are a lot of replacement clocks available, what is the situation
at the studio end, ie. in the AD part of the chain? Are the
AD converters used for recording designed for low jitter,
or is this a still neglected thing there, or has it perhaps been
known and taken care of since long ago? Anybody who knows?
value of jitter reduction, at least this is something that can be
straightforwardly explained in theory. A timing error in the
digitial domain will translate into an amplitude error in the
analog domain. How much it matters in practice can be
discussed, but the effect is there according to the theory.
There is another aspect on the jitter problem that I have
been wondering about and that I cannot remember ever
seing discussed. While audiophiles have for quite some years
put a lot of effort into jitter reduction in their CDPs and there
are a lot of replacement clocks available, what is the situation
at the studio end, ie. in the AD part of the chain? Are the
AD converters used for recording designed for low jitter,
or is this a still neglected thing there, or has it perhaps been
known and taken care of since long ago? Anybody who knows?
It has been adressed through the last 30 years in studio environments. Every studio has a low noise 'house clock' to syncronize all digital devices. It runs directly on the sampling frequency that the studio is producing media for. In case of a CD, it would be 44.1 kHz.
One example can be seen here: http://www.iclock-net.de/e_iclock.htm
A studio owner talks it out here: http://www.discmakers.com/pse/jung.php
One example can be seen here: http://www.iclock-net.de/e_iclock.htm
A studio owner talks it out here: http://www.discmakers.com/pse/jung.php
Lars Clausen said:
It’s easy to see why you believe jitter is the cause of all problems in digital audio: you sell clocks. As Christer points out, jitter is easily explained, although not always completely understood, and with at least three, full-time clock mongers in attendance here, it’s easy to see why all the discussion in this forum gets steered in that direction.
If I wanted to make a cheap DAC to play MP3 files I would certainly consider Cirruslogic or AKM because, in my opinion, that is all those chips are good for. If you think delta-sigma and DSD represent the pinnacle of achievement in audio reproduction, there must be something wrong with you hearing or you are one of the few who still believes the advertising slogan, "Perfect Sound, Forever."
I choose the PCM1704 for its capabilities, not its price. I will use those chips to explore the possibilities that lie between the analog-filtered, non-oversampled approach on one side, and the digital filtered, up/oversampled approach on the other.
Now, please continue with your fascinating jitter discussion. I promise, I won’t interrupt again.
Ciao, Ulas
To delve a bit deeper into this subject, there is one aspect which hasn't been mentioned: the method a DAC plays back the output samples, the filters involved and where jitter fits into this model.
Ideally, for a time sampled band-limited continuous signal [sampled at the approriate frequency, ie at least Nyquist]- the output of the DAC would be impluses amplitude of each sample at each sample point. An ideal lowpass filter (brickwall) with linear phase would be use to get back the original signal exactly. (This is a sinc convolution in the time domain - impossible to actually make from passive components since it is non-causal)
The first problem is the sample amplitudes are quantized... its been researched extensively and you get stuff like noise shaping to deal with it.
Now, the reality is a DAC doesn't play back impluses, it will hold the sample amplitude until the next sample. The ideal low pass filter for this does not have a constant gain in the passband, but curves up as frequency increases. Are the analog filters at the output on non-oversampling DACS following this model? Are the digital filters of oversampling DACs following this model? If they assume the constant passband gain for the filter (sinc filter) the high frequencies will be attentuated.
If you oversample or upsample you have a better chance of getting the filter right, and then you don't have to worry so much about the analog filter since constant passband is probably good approximation for the portion in the audible frequencies. Maybe this is why you hear comments like "analog like" and "extended highs" for many oversampled and upsampled players.
JItter in all of this is interesting. I am not sure if the effect of jitter is not as bad in the case of holding samples as it would be in the case of impulse samples. It means an amplitude error in the _prefiltered_ signal, but the consequences of this after the low pass filter will be much worse. The problem with jitter which is NOT like dither is that it is often not "random" in nature. I have heard that random jitter is not nearly as much of a problem as correlated jitter. Dither is TRULY random error added to the digital data, and it is random because it is algorithmically made to be
Oversampling(integer) and upsampling(rational) bring up other interesting points:
1) Oversampling would be the ultimate solution to a lot of the digital problems if it were done on the complete data set at once, with non-causal reasonably ideal filters (possible because you do have all the information about the future that you need to do the filter right) and no "source" clock, the source clock is assumed 44.1. Upsampling via this method would be a close second because all samples are calculated, vs oversampling where the original samples get to stay. So what you need is to preconvert to 88.2 or 176.4 and have a playback device that supports that. Realistically this can be done with easily available playback devices by converting to 24/96 or 24/192 and playing back on a DVD-V or DVD-A player.
2) Windowed "real-time" oversampling is next. You cannot have a non-causal filter in this case but your source clock is "virtual" so there is no jitter from the source clock. Upsampling follows with a virtual clock, ie upsampling in DSP where there is no "source" clock, only data and the pre-knowledge of time period of the data. Jitter can still affect the output clock. One box player solution only.
3) Windowed "real time" ASRC upsampling where the source clock
is monitored with relation to the destination clock and the calculations are done based on instananeous clock relationships. The reason the clock relationships are used is to keep synchronization. This means the jitter on the source clock is encoded into the calculated data. You then also get jitter on the destination clock. This is the method used in those ASRC chips for external DACs (and some use them for one box upsampling players although this seems insane since now you have 2 clocks worth of jitter to worry about)
Delta-sigma may actually be a reasonable solution to a lot of the above problems.
Comments?
Ideally, for a time sampled band-limited continuous signal [sampled at the approriate frequency, ie at least Nyquist]- the output of the DAC would be impluses amplitude of each sample at each sample point. An ideal lowpass filter (brickwall) with linear phase would be use to get back the original signal exactly. (This is a sinc convolution in the time domain - impossible to actually make from passive components since it is non-causal)
The first problem is the sample amplitudes are quantized... its been researched extensively and you get stuff like noise shaping to deal with it.
Now, the reality is a DAC doesn't play back impluses, it will hold the sample amplitude until the next sample. The ideal low pass filter for this does not have a constant gain in the passband, but curves up as frequency increases. Are the analog filters at the output on non-oversampling DACS following this model? Are the digital filters of oversampling DACs following this model? If they assume the constant passband gain for the filter (sinc filter) the high frequencies will be attentuated.
If you oversample or upsample you have a better chance of getting the filter right, and then you don't have to worry so much about the analog filter since constant passband is probably good approximation for the portion in the audible frequencies. Maybe this is why you hear comments like "analog like" and "extended highs" for many oversampled and upsampled players.
JItter in all of this is interesting. I am not sure if the effect of jitter is not as bad in the case of holding samples as it would be in the case of impulse samples. It means an amplitude error in the _prefiltered_ signal, but the consequences of this after the low pass filter will be much worse. The problem with jitter which is NOT like dither is that it is often not "random" in nature. I have heard that random jitter is not nearly as much of a problem as correlated jitter. Dither is TRULY random error added to the digital data, and it is random because it is algorithmically made to be
Oversampling(integer) and upsampling(rational) bring up other interesting points:
1) Oversampling would be the ultimate solution to a lot of the digital problems if it were done on the complete data set at once, with non-causal reasonably ideal filters (possible because you do have all the information about the future that you need to do the filter right) and no "source" clock, the source clock is assumed 44.1. Upsampling via this method would be a close second because all samples are calculated, vs oversampling where the original samples get to stay. So what you need is to preconvert to 88.2 or 176.4 and have a playback device that supports that. Realistically this can be done with easily available playback devices by converting to 24/96 or 24/192 and playing back on a DVD-V or DVD-A player.
2) Windowed "real-time" oversampling is next. You cannot have a non-causal filter in this case but your source clock is "virtual" so there is no jitter from the source clock. Upsampling follows with a virtual clock, ie upsampling in DSP where there is no "source" clock, only data and the pre-knowledge of time period of the data. Jitter can still affect the output clock. One box player solution only.
3) Windowed "real time" ASRC upsampling where the source clock
is monitored with relation to the destination clock and the calculations are done based on instananeous clock relationships. The reason the clock relationships are used is to keep synchronization. This means the jitter on the source clock is encoded into the calculated data. You then also get jitter on the destination clock. This is the method used in those ASRC chips for external DACs (and some use them for one box upsampling players although this seems insane since now you have 2 clocks worth of jitter to worry about)
Delta-sigma may actually be a reasonable solution to a lot of the above problems.
Comments?
Why the fascination??
Because, every time that I reduce the jitter in any system, the same effects are easily heard.
(Since I have posted clock schematics here for all to use, am I a "clock monger", too? Hope not....I thought that I was lower than that.)
And yes, I use the PCM1704 for the same reasons. The fact that TI now owns them makes me cringe every time that I buy one.
But if you do use that chip, surely you must have noticed how as much more sensitive to jitter a quality system is, as compared to a ratty old SAA7220/TDA1541 setup.
Jocko.
Because, every time that I reduce the jitter in any system, the same effects are easily heard.
(Since I have posted clock schematics here for all to use, am I a "clock monger", too? Hope not....I thought that I was lower than that.)
And yes, I use the PCM1704 for the same reasons. The fact that TI now owns them makes me cringe every time that I buy one.
But if you do use that chip, surely you must have noticed how as much more sensitive to jitter a quality system is, as compared to a ratty old SAA7220/TDA1541 setup.
Jocko.
Hi to everyone,
The discussions hee are quite interesting. Pardon my ignorance, but I just have a few querries I hope you can respond to. I know that jitter is a digital problem in the time domain. If it can result in an amplitude problem in the converted analog, how extensive woudl that be? Is the problem evident throughout an entire disc or musical title? Does it alter the stereo soundstage, frequency response, etc. Thanks.
The discussions hee are quite interesting. Pardon my ignorance, but I just have a few querries I hope you can respond to. I know that jitter is a digital problem in the time domain. If it can result in an amplitude problem in the converted analog, how extensive woudl that be? Is the problem evident throughout an entire disc or musical title? Does it alter the stereo soundstage, frequency response, etc. Thanks.
Robert Adams at ADI has written some very readable papers on the subject, and on the relative sensitivity of various DAC schemes to jitter-induced distortion. Worth reading. They're probably posted over at the ADI web site.
If for a given DAC system the jitter is significant, it will be reflected in the analog distortion and noise measurements of the output. That seems to be something that people concentrating on jitter gloss over.
If for a given DAC system the jitter is significant, it will be reflected in the analog distortion and noise measurements of the output. That seems to be something that people concentrating on jitter gloss over.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.