Gentlemen, we are all expecting a sweet distillation of digital wisdom from the sharpest minds of our beloved forum. Shall we propose an ordered system approach as was intended by the author, beginning, perhaps, with the so important power(s) supplies(s) ?
We risk repeating the same arguments of other threads about infinitesimal aspects without facing the main task.
Respectfully yours...
M
Hi Guido,
I only partially agree with what you say.
In my view, old players jitter was perceivable exactly as jitter: listening to the music they played, I had exactly the feeling of something like the wow&flutter of old casset recorders, but with a min sure or incidence and a far higher frequency. Music simply did not make sense, had no rythmic texture.
This is probably however the effect of low frequency jitter in rather high amounts (order of 1nsec, probably, but I had no way to measure it at the time).
Smaller amount of jitter, with different spectral distributions, are differently perceived. There is also a study presented at AES about this, and the results is that also very little amounts of jitter under some conditions are perceivable.
However, these produce, as far as I can understand/hear, other kind of alterations of the audio message, different from the one above. In some cases is a softening of the music, in others also an apparent increase of high frequencies. Someone has documented that liked the presence of some kind of jitter (again Kosunoki asynchronous reclocking for example).
What I really meant is that I am pretty sure we can hear the effects of even small amounts of jitter, but given its chamaleontic attitude, I am far less sure it is always so simple and safe to correlate the perceived effect to jitter. You normaly know something is wrong, but to track it back to jitter is another story.
Kind regards
Giorgio
I only partially agree with what you say.
In my view, old players jitter was perceivable exactly as jitter: listening to the music they played, I had exactly the feeling of something like the wow&flutter of old casset recorders, but with a min sure or incidence and a far higher frequency. Music simply did not make sense, had no rythmic texture.
This is probably however the effect of low frequency jitter in rather high amounts (order of 1nsec, probably, but I had no way to measure it at the time).
Smaller amount of jitter, with different spectral distributions, are differently perceived. There is also a study presented at AES about this, and the results is that also very little amounts of jitter under some conditions are perceivable.
However, these produce, as far as I can understand/hear, other kind of alterations of the audio message, different from the one above. In some cases is a softening of the music, in others also an apparent increase of high frequencies. Someone has documented that liked the presence of some kind of jitter (again Kosunoki asynchronous reclocking for example).
What I really meant is that I am pretty sure we can hear the effects of even small amounts of jitter, but given its chamaleontic attitude, I am far less sure it is always so simple and safe to correlate the perceived effect to jitter. You normaly know something is wrong, but to track it back to jitter is another story.
Kind regards
Giorgio
Guido Tent said:
If one accepts the above then, so long as the data up to that point is correct, the only place that jitter matters is at the dac and in the case of a PCM63 connected to a SM5842, I very much doubt that 1ps of jitter on a 352.8kHz clock is going to make for much of an audible difference.
Just something to think about.
HDCD works by replacing a part of the lsb's with control data for the hdcd decoder. Yes, replacing music with control information.
But a hdcd disc is playable on non-hdcd systems. And they say the control info which now becomes noise is not perceivable.
Now either they tested it on the wrong people or there aren't that many people able to, in a blind test, pick up a cd with the lsb replaced by random noise.
(I listened once to a hdcd track and didn't know that until it was placed into a hdcd cdp which lit it's little blue led. Now it's true that my hearing is not above average.)
HDCD works by replacing a part of the lsb's with control data for the hdcd decoder. Yes, replacing music with control information.
But a hdcd disc is playable on non-hdcd systems. And they say the control info which now becomes noise is not perceivable.
Now either they tested it on the wrong people or there aren't that many people able to, in a blind test, pick up a cd with the lsb replaced by random noise.
(I listened once to a hdcd track and didn't know that until it was placed into a hdcd cdp which lit it's little blue led. Now it's true that my hearing is not above average.)
The use of dither is very interesting, and is more complex than the addition of simple noise. The noise can be far from simple. By shaping the noise used to dither the sampled sound it is possible to increase the resolution in selected parts of the spectrum. This is commonly, if not universally, done.
There is often a misunderstanding about what dither does, and the common (and wrong) idea that it sets the limit of resolution of the sampled signal (it does not.) The signal does not have zero bandwidth, and thus simplistic analysis does not work.
The duality between amplitude and time in the signal is indivisible. Signal correlated timing error in a sampled stream will always yield energy in the audible band. Even a simple calculation can show some of the equivalence. With a 44.1kHz sample rate, and 16 bit depth the LSB is equivalent to 3*10E-10 seconds. 30 picoseconds. However with noise shaped dither we can achieve a further 9 to 12 db of resolution in critical (that is more audible) frequency bands. Even with the lowly CD format we are pretty close to one pico second of audible dither.
The classic problem is digital audio is the mess made of the S/PDIF format. The clock is embedded in the data stream, and must be recovered. However the data format is such that the digital stream contains signal correlated energy - so much so that the trivial and infamous trick of actually recovering a recognisable audio signal from the stream with nothing more than a capacitor is possible. This trick alone tells us that we have big problems. Thus almost any attempt at clock recovery will include some signal correlated energy (i.e. signal correlated jitter). Phase locked loops attenuate the energy. Very careful PLL design can attenuate it to practically nil. But if the energy gets into the DAC timing it will reappear in the reconstructed audio - and being signal correlated it will sound bad. And how bad? Potentially awful. Not just harmonic distortion products - but rather in-harmonic products - sum and difference components, components a fixed frequency from the real signal - all especially bad sounding.
So, the thread topic. It is a chain. At any point it is possible to wreck things enough that later components may not be able to recover totally. But it happens in unobvious ways sometimes. One needs to be very careful about making broad claims about the value (or damaging effect) of any given change. For instance the simple amplitude of jitter does not convey enough information. One needs the spectral energy. And then the manner it effects the sound will depend upon the exact nature of the later stages. A change that improves one aspect might damage another, and the final good versus bad answer may change depending upon context, and taste.
There is often a misunderstanding about what dither does, and the common (and wrong) idea that it sets the limit of resolution of the sampled signal (it does not.) The signal does not have zero bandwidth, and thus simplistic analysis does not work.
The duality between amplitude and time in the signal is indivisible. Signal correlated timing error in a sampled stream will always yield energy in the audible band. Even a simple calculation can show some of the equivalence. With a 44.1kHz sample rate, and 16 bit depth the LSB is equivalent to 3*10E-10 seconds. 30 picoseconds. However with noise shaped dither we can achieve a further 9 to 12 db of resolution in critical (that is more audible) frequency bands. Even with the lowly CD format we are pretty close to one pico second of audible dither.
The classic problem is digital audio is the mess made of the S/PDIF format. The clock is embedded in the data stream, and must be recovered. However the data format is such that the digital stream contains signal correlated energy - so much so that the trivial and infamous trick of actually recovering a recognisable audio signal from the stream with nothing more than a capacitor is possible. This trick alone tells us that we have big problems. Thus almost any attempt at clock recovery will include some signal correlated energy (i.e. signal correlated jitter). Phase locked loops attenuate the energy. Very careful PLL design can attenuate it to practically nil. But if the energy gets into the DAC timing it will reappear in the reconstructed audio - and being signal correlated it will sound bad. And how bad? Potentially awful. Not just harmonic distortion products - but rather in-harmonic products - sum and difference components, components a fixed frequency from the real signal - all especially bad sounding.
So, the thread topic. It is a chain. At any point it is possible to wreck things enough that later components may not be able to recover totally. But it happens in unobvious ways sometimes. One needs to be very careful about making broad claims about the value (or damaging effect) of any given change. For instance the simple amplitude of jitter does not convey enough information. One needs the spectral energy. And then the manner it effects the sound will depend upon the exact nature of the later stages. A change that improves one aspect might damage another, and the final good versus bad answer may change depending upon context, and taste.
rfbrw said:
Hi rfbrw,
I agree with "the only place" statement (allthough all signals entering a DAC chip potentially may affect the conversion, which is why I prefer "reclock them all").
The test I did was using a (well known) VCXO to clock the DAC. By adding known signals to the VCXO control pin, we could derive when the jitter became detectable (by ear). We also found sensitivity in terms of signal (sine wave, noise, music) and the kind of effect it had on the reproduction (transparancy, coloration, pitch, timing, stress).
This test gained much insight, reccomended !
best
Hey Jocko just first want to say I have the world of respect for your knowledge, I merely want to see if I can try to capture what the gentlemen said in the post wich might have seemd contradictory to you, Lol maybe he even got it plain wrong, but please considder the next example... maybe you can educate me a bit... anything you say is normaly enlightening...
Anyhow, I think what you were saying is that we can not hear short individual impulses or sound modulations (have to pick words now, you clever poeple have diffirent meaning for everything), especialy when it is of very short duartion (1ps)...
But I think what the other gentleman alluded to is...when we hear a high frequency note, we can't hear an individual peak of that, lets say sinewave, either, which I assume must also have minuscule durations... yet if some of them start getting displaced or distorted we quicly notice the change in sound...
So maybe we cant hear an individual modulation of 1 ps, but if I understand jitter correctly it has a continuous nature and maybe we do "hear" those pulses combined over time like we perceive the many pulses in a high frequncy sound as 1 "note...
Anyhow, I think what you were saying is that we can not hear short individual impulses or sound modulations (have to pick words now, you clever poeple have diffirent meaning for everything), especialy when it is of very short duartion (1ps)...
But I think what the other gentleman alluded to is...when we hear a high frequency note, we can't hear an individual peak of that, lets say sinewave, either, which I assume must also have minuscule durations... yet if some of them start getting displaced or distorted we quicly notice the change in sound...
So maybe we cant hear an individual modulation of 1 ps, but if I understand jitter correctly it has a continuous nature and maybe we do "hear" those pulses combined over time like we perceive the many pulses in a high frequncy sound as 1 "note...
Francis_Vaughan said:
Right you are.
Low frequency jitter is the most common (and in higher levels) and it really affects bass performance. Very untight bass.
This is the easiest to detect.
Btw untight bass affects all the rest of the spectrum.
Dac chips only start to 'reject' jitter at frequencies above ~20khz.
Saying that a chip is 'very insensitive' to jitter (without saying at what frequencies) is just pure marketing.
A good clock is important, of course, and the jitter levels across a wide range of frequencies may have more impact on some dacs/digital filters than others.
You need the recovered clock only to receive the bits. Feed them into a RAM buffer and use another standalone high precision clock to extract them and feed to the dac. As long as your buffer does not get depleted you're not depending on the spdif clock.
Yes it would have been nicer to get a separate clock line from the transmitter but (1) that means 3 wires so goodbye 1 run of coax and (2) if the transmitter does not have a good clock you're back to the ram buffer solution.
Yes it would have been nicer to get a separate clock line from the transmitter but (1) that means 3 wires so goodbye 1 run of coax and (2) if the transmitter does not have a good clock you're back to the ram buffer solution.
I'm glad that someone is amused........
I'm not.
Look, getting low jitter in a standalone CDP is not that hard. A decent clock, distribution scheme, clean supplies and ground returns.......stuff like that is all it takes. Whether the level at which jitter is indiscernable is rather academic. Some say 10 pSec, others 1 pSec. Big deal. The jitter in a stock CDP is much worse than that.
Anyone can put in a better clock..............using, say one of the designs that some of us post here..........and hear a marked difference. Bass is tighter, highs are smoother and cleaner, reverb/decay/room echo/etc. are all more clearly heard. If you don't believe me, try it. Just knock off the "siren cry of the clock monger" crap. Use one that you can get the schematic from here..........for free.......and then try to say it is only the guys out to make a $$$$ that say it.
As for SPDIF..........trying to dig out the clock signal in that mess, without having all that correlated crap on it is not that easy. Simple schemes like buffers will just not cut it. You can believe elsewise if it makes you happy. But if you want your ears to be happy, then you will have to accept the fact that you must remove all the jitter, since it contains so much signal dependent crap.
If you don't believe me..........listen to the crap on the supply rail of your SPDIF RX chip. You might be sick.
Jocko
I'm not.
Look, getting low jitter in a standalone CDP is not that hard. A decent clock, distribution scheme, clean supplies and ground returns.......stuff like that is all it takes. Whether the level at which jitter is indiscernable is rather academic. Some say 10 pSec, others 1 pSec. Big deal. The jitter in a stock CDP is much worse than that.
Anyone can put in a better clock..............using, say one of the designs that some of us post here..........and hear a marked difference. Bass is tighter, highs are smoother and cleaner, reverb/decay/room echo/etc. are all more clearly heard. If you don't believe me, try it. Just knock off the "siren cry of the clock monger" crap. Use one that you can get the schematic from here..........for free.......and then try to say it is only the guys out to make a $$$$ that say it.
As for SPDIF..........trying to dig out the clock signal in that mess, without having all that correlated crap on it is not that easy. Simple schemes like buffers will just not cut it. You can believe elsewise if it makes you happy. But if you want your ears to be happy, then you will have to accept the fact that you must remove all the jitter, since it contains so much signal dependent crap.
If you don't believe me..........listen to the crap on the supply rail of your SPDIF RX chip. You might be sick.
Jocko
Nordic said:
You are on the right track in trying to understand what is occurring. A few thought experiments might help.
You are probably familiar with Pulse Width Modulation, where a signal can be conveyed
with by varying the width (i.e. the time) of a series of high frequency pulses. The pulses themselves are at a supersonic frequency - so the carrier itself is not heard, and if the pulses where all of exactly the same duration (width) there would be no audible signal - i.e. the signal would be the equivalent of DC. Now if we modulate the pulse width, that is change the timing width, so that the width of the pulses varies up and down at a rate that is in the audible spectrum, we will get audible sound. It works well, and is a common idea. Digital audio is a close cousin.
Now imagine that we have constructed a PWM system to distribute audio. But we have problems with the generation of perfect timing for the pulses. We could have noise from some unknown source that upsets the exact time of the pulse edges. That external noise will cause a degradation in the audible sound. If the external noise has energy in the audible spectrum we will hear it in some form. Crucially, note that this noise manifests itself as an error in time. It is jitter. Notice that you could have (in principle) inaudible jitter.
Now junk into digital audio. We have much the same problem, but mostly worse. The exact nature of the problem depends upon the nature of the DAC. All DACs are dependant upon a clock to control the delivery of sample values. The simplest (i.e. the contentious non-over-sampling DAC) deliver samples at 44.1 kHz. A modulation of that clock will result in audible sound in exactly the same manner as a PWM system. More complex DACs get messy quickly. A simple oversampling DAC can be regarded as one with a higher clock rate - and interestingly the absolute amplitude of jitter needed to be audible drops in proportion to the rise in clock rate. Anything with noise shaping becomes very complex to analyse - and the manner in which clock jitter manifests itself in the audible spectrum quickly becomes a very messy convolution rather than just added audible signal.
And to go back to the previous post, the biggest issue is when the jitter energy is correlated to the signal itself. Then all manner of very objectionable intermodulation products occur. And not just signal intermodulation, but also products that involve the sample rate and clock rate, yielding all manner of dreadful products. Products that are not amenable to simple THD analysis either.
Signal correlated jitter need not only occur due to the S/PDIF interface - although that is a good start. DAC design needs to ensure that no energy couples back from the output into the digital or clock generation components. Also the I2S or USB signals likely carry correlated energy, so there is no trivial fix.
danb1974 said:
Hi
This a conceptually good solution, but it partly fails at the implementation level, due to mechanisms earlier explained.
The Chord DAC is built like this, but I can still hear differences in sound between the drives sending data to that DAC.......
So yes, above helps, but doesn't solve it all. In addition, such a buffer adds jitter so firm reclocking afterwards is required.
best
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.