Timing is everything

Vincent Kars · Nov 14, 2011

Let’s do some calculations (always tricky in my case)

Playing 192 kHz is 1 sample per
0.000005208333333333330 s

If the clock has a jitter of 10 ns
0.00000001 = 10 ns
0.000005218333333333330 +10 ns
0.000005208333333333330
0.000005198333333333330 -10 ns
That’s pretty close to the size of the time step

If our clock has a jitter of 1 ps which I do think extremely low
0.000000000001 = 1 ps
0.000005208334333333330 +1 ps
0.000005208333333333330
0.000005208332333333330 -1ps

In this case the variations in the time step look more decent.
But in both cases is looks like out gear can’t play 192 kHz with the right timing.
Makes me wonder what happens if we start to play 384 kHz recordings

LL21 · Nov 14, 2011

ok...a technophobe starting to get a glimmer of understanding. in your experience, what does this 'margin of error' end up doing to the sound? does the inaccuracy mean lack of detail? incorrect detail? muddy? thanks for helping to translate technical into live...

amirm · Nov 14, 2011

Pretty clever

. Indeed, jitter requirements are insanely tight at high sampling rates especially when combined with 24 bit audio samples. At just 16 bits and 40 Khz sampling, we need to achieve better than 0.5 nanoseconds! And this is for the more simplified sinusoidal jitter.

garylkoh · Nov 14, 2011

May be that's why whenever I rip vinyl to digital, I find 24/96 to be the sweet spot in sound quality in my system. That's probably because my recording system is not as good as Bruce's. I need an atomic clock

Vincent Kars · Nov 15, 2011

Might be one of the many reasons why hires sounds different

RBFC · Nov 15, 2011

Vincent,

Thanks for a description that most of us can actually understand! Do you have any jitter figures available for typical connection types? SPDIF, USB, etc. using typical implementations?

Does this mean that most consumer-level gear tops out (as Gary mentions) at 24/96, and that higher sampling rates are introducing more problems than good?

Lee

DonH50 · Nov 15, 2011

The catch with this analysis is that signal degradation related to jitter is dependent upon the signal frequency, not the clock frequency... If you have a 1 MHz clock (1 us period) and 1 Hz signal, 10 us of jitter is not going to significantly degrade the signal -- it's moving too slowly. The impact of jitter depends upon the signal frequency and resolution; clock frequency actually drops out of the equation.

amirm · Nov 15, 2011

Don, the relationship I got was that sampling rate dictates maximum signal frequency which as you say, has 100% to do with the jitter impact.

RBFC · Nov 15, 2011

So, if we mandate an upper frequency limit of 40kHz and use that frequency as our test signal (to account for the extension seen in higher res files and to provide an extra octave of response above the standard of 20kHz), what are the jitter thresholds for degradation of the signal at the various popular high-res sampling rates (96, 192, 384)? What are the typical performance parameters of some popular devices in playing these resolutions?

Lee

amirm · Nov 15, 2011

It scales linearly with frequency response. As I mentioned, the threshold is 0.5ns for 20 KHz bandwidth. For 40 KHZ (80 KHz sampling), it is half that at 0.25ns. Where it becomes a killer is if you also then ask for better than 16 bits. At 20 bits, the number for 40 KHZ BW drops like a rock to nearly 0.01ns! No sense in going beyond that number since we can't reproduce them anyway.

DonH50 · Nov 15, 2011

amirm said:
Don, the relationship I got was that sampling rate dictates maximum signal frequency which as you say, has 100% to do with the jitter impact.

Hmmm... That is a true statement, but only matters if the allowed signal frequency increases with sampling rate, which I do not believe is true for the audio codecs? Isn't the upper frequency limited?

Whatever, the relevant equations are:

SNR = 20*log10(2*pi*fin*tj) where pi = 3.14159265..., fin = input frequency, and tj = jitter (pp) -- notice fclk does not appear

SNR = 6.021*N + 1.76 dB for N-bit resolution

If you sample a 20 kHz signal at 16 bits, the sampling clock frequency must be >40 kHz to accurately recreate the signal. As you increase the sampling clock above that frequency the jitter requirements do not change. If you increase the signal frequency or resolution, jitter must be lower to maintain the target SNR.

A 16-bit, 20 kHz signal has the same jitter requirements if the sample rate is 44.1 kHz or 100 MHz.

amirm · Nov 15, 2011

DonH50 said:
Hmmm... That is a true statement, but only matters if the allowed signal frequency increases with sampling rate, which I do not believe is true for the audio codecs? Isn't the upper frequency limited?

If it were, higher sampling rates would not be used

. I am just going by the assumption that there is something there that is needed to be preserved. If not, then you are 100% right and we are in violent agreement.

Phelonious Ponk · Nov 15, 2011

Vincent Kars said:
Might be one of the many reasons why hires sounds different

Or one of the reasons why I don't hear jitter, as I have hardly any hi-res material. Jitter hurts my head. If I've ever heard it, I'm pretty blissfully unaware. I've heard distortion, of course. I suppose it's possible that some of it was jitter. But it sounds like the usual stuff to me, so who knows? I did have a a tube headphone amp in here for awhile that sounded just awful when you used its internal DAC and really good when you didn't. Maybe that was jitter. Does jitter sound like no lows, no, lower mids and treble like bacon frying? There seems to be broad disagreement over how much jitter matters. Broader disagreement over what it sounds like. And then there are some who believe it can sound like anything, manifest itself as almost any artifact or distortion imaginable. Jitter is the great mystery of modern audio. I used to believe it was simply inaudible in competent equipment, but Amir says he has trained himself to hear it and can. And I trust Amir. I'll never take that training. I kinda figure that even if I have some measure of jitter, it's probably obliterated by the distortions in my drivers and their interaction with my room.

I take comfort where I find it. YMMV.

Tim

DonH50 · Nov 15, 2011

amirm said:
If it were, higher sampling rates would not be used . I am just going by the assumption that there is something there that is needed to be preserved. If not, then you are 100% right and we are in violent agreement.

I don't know, Amir. There are other benefits to higher sampling rates (reduced filter requirements, better interpolation, etc.), as well as drawbacks, but I don't know where the filters are set. If 192 kS/s is used to provide 96 kHz bandwidth then obviously the jitter requirements are more stringent. I had the impression (perhaps false) that the signal was still filtered above the audio'ish band, and higher rates were more to help reduce filter requirements and their in-band artifacts than to increase signal bandwidth, but I do not know. A higher sampling rate also reduces the DAC's sinx/x roll-off at the output, from 3.54 dB (I think) at 22 kHz and 44 kS/s to essentially negligible for 22 kHz at 192 kS/s, again a benefit only if the bandwidth is limited to ~20 kHz.

Tim -- The sound of jitter depends upon the type of jitter. Random raises the noise floor, and sounds like noise. Signal-dependent can introduce distortion that, like any other distortion, causes "bad sounds" and signal spurs non-harmonically related to the actual signal. Like IMD, that is often worse than harmonic distortion, as we are more sensitive to it. Or so I've been told, so I have measured, so I believe.

amirm · Nov 15, 2011

It is true that we need it for filter benefits. There though, once we get to 48 Khz and cruise past, then that issue becomes moot. Indeed, there would be no need to do anything beyond 88 Khz.

Bruce and others can chime in but as far as I know, there is no filtering all. Spectrum displays have seen have energy to Nyquist limit. Here is a sample track I own that is 96 Khz:

As you see, there is no filtering to 20 Khz.

microstrip · Nov 15, 2011

Phelonious Ponk said:
(...) I used to believe it was simply inaudible in competent equipment, but Amir says he has trained himself to hear it and can. And I trust Amir. I'll never take that training. I kinda figure that even if I have some measure of jitter, it's probably obliterated by the distortions in my drivers and their interaction with my room.

I take comfort where I find it. YMMV.

Tim

Tim,

You can try a simple experiment - get a separate CD transport, a cheap DAC and a good digital cable. Listen to it. Next replace the cable with a very long non digital cable - a run of microphone or guitar cable will do. This will probably create an impedance mismatch that will introduce jitter. Listen again. If you find any difference it must be due to jitter.

PS - if the impedance mismatch is too large you can have data loss and heavy distortion - this is not jitter induced noise!

microstrip · Nov 15, 2011

Vincent,

I do not want to play the bad guy, but, IMHO, unhappily your simplified analysis can be misleading as it gives an erroneous idea of jitter effects. You are inadvertently mixing concepts that are of great importance in the analysis of jitter effects in DSD with the jitter effects in PCM. As both are sensitive to jitter in very different ways you are arriving to false conclusions.

Sometimes it is not possible to simplify explanations on some subjects beyond some limits. The audio effects of jitter are one of them.

Vincent Kars · Nov 15, 2011

Hi Lee

If you look at the measurement pages of e.g. Stereophile or Miller Audio Research, you get a barrage of graphs. Indeed you can’t frame jitter in a single number.
So I’m simplifying but for me, it gives me some feeling of the magnitude.

We are inclined to think bigger is better.
24 is better than 16, 88 is better than 44 and of course 192 tops them all (until 384 comes available).
But even if I’m oversimplifying, I do think that my simple numbers demonstrate that the higher the sample rate, the more critical timing will become.

My research (hanging around too long, too often on too many audio forums) gives me the feeling that some gear struggles with these high bitrates.
I do think 24/96 is not necessarily a sweet spot.
It is probably completely system depended.

Vincent Kars · Nov 15, 2011

Microstrip,

I’m not talking DSD, just plain PCM.

For me it is like bit depth.
There is at the present no DAC on sale linear up to 24 bits.
22 is the highest specked DAC I know.
Likewise there is no DAC able to play PCM with an exact time step of 0.000005208333333333330
Simple indeed but not a wrong conclusion.
There is no DAC able to play 24/192 to its full extend.

fas42 · Nov 15, 2011

microstrip said:
You can try a simple experiment - get a separate CD transport, a cheap DAC and a good digital cable. Listen to it. Next replace the cable with a very long non digital cable - a run of microphone or guitar cable will do. This will probably create an impedance mismatch that will introduce jitter. Listen again. If you find any difference it must be due to jitter

Not necessarily. A different cable can be simply allowing more pickup of RF interference and other nasties, which then disturbs the DAC operation. Like Don, I believe this jitter thing is overcooked to some degree; yes, doing different things changes the sound, but laying the blame squarely at the door of jitter manipulation is probably oversimplifying ...

Frank

Timing is everything

WBF Technical Expert: Computer Audio

Well-Known Member

Banned

WBF Technical Expert (Speakers & Audio Equipment)

WBF Technical Expert: Computer Audio

WBF Founding Member

Member Sponsor & WBF Technical Expert

Banned

WBF Founding Member

Banned

Member Sponsor & WBF Technical Expert

Banned

New Member

Member Sponsor & WBF Technical Expert

Banned

VIP/Donor

VIP/Donor

WBF Technical Expert: Computer Audio

WBF Technical Expert: Computer Audio

Addicted To Best

Similar threads