Can jitter be measured in S/PDIF cable and correlated to SQ?

Empirical Audio

Industry Expert
Oct 12, 2017
1,169
207
150
Great Pacific Northwest
www.empiricalaudio.com

Empirical Audio

Industry Expert
Oct 12, 2017
1,169
207
150
Great Pacific Northwest
www.empiricalaudio.com
Interesting. But surely any decent spdif dac these days does not use pll recovered clock...

Some of the best DAC's do not upsample the data, so the clock is still recovered from the S/PDIF signal using a PLL. There are some dCS DACs that send the master clock from the DAC to the transport, to synchronize it. These would surely be immune to S/PDIF jitter.

Even DAC's that do resample can still be affected by incoming jitter. I have seen many cases of this. The exception is the DAC that is immune to incoming jitter.

Steve N.
Empirical Audio
 

LL21

Well-Known Member
Dec 26, 2010
14,411
2,509
1,448

jkeny

Industry Expert, Member Sponsor
Feb 9, 2012
3,374
42
383
Ireland
Interesting measurements, Steve - can you just tell us a bit about what the plots represent - I see the Y axis is labelled "Hits" & the X-axis "Time"
Do you have a handle on correlating the plot shape to SQ?
 

analogsa

Well-Known Member
Apr 15, 2017
382
122
175
Cascais
Some of the best DAC's do not upsample the data, so the clock is still recovered from the S/PDIF signal using a PLL.

I would not use "upsampling" and "decent" in the same sentence :) Had a fifo with a local clock in mind. Even my old ML could do that. Which does not eliminate the differences between cables but shifts the focus away from jitter.

Otoh it was probably a silly remark as there is still a huge user base with pll spdif decoding.
 

jkeny

Industry Expert, Member Sponsor
Feb 9, 2012
3,374
42
383
Ireland
...

Otoh it was probably a silly remark as there is still a huge user base with pll spdif decoding.
I'm not sure what you are getting at here? Do you mean that there are many DACs which don't have local clocks & just use the PLL recovered clock from the SPDIF stream Vs the DACs that use a local clock but still use a PLL to extract the incoming clock & adjust the local clock to deal with clock differences?
 

Empirical Audio

Industry Expert
Oct 12, 2017
1,169
207
150
Great Pacific Northwest
www.empiricalaudio.com
Interesting measurements, Steve - can you just tell us a bit about what the plots represent - I see the Y axis is labelled "Hits" & the X-axis "Time"
Do you have a handle on correlating the plot shape to SQ?

The plots are a distribution of period measurements. There is a nominal period measurement, which would be a single spike in the center if there was no jitter, at a height equal to the total number of sample measurements.

Each spike represents the number of samples of period measurements that were recorded at that particular period. The taller the spike, the larger the number of measurement hits at that particular period.

As for correlation, I spent yesterday with my wife doing listening tests. She was unaware of which cable was in the system.

I updated the link to show the results of these listening tests. In a nutshell, the narrower the overall distribution, the better the rendering of high-frequencies and venue echoes, even with 2 humps. The wider and deeper the soundstage became.

However, I also found that the lifelike quality of the vocals was more related to the cable construction and conductor material. My Bitmeister had one of the best renderings of vocals and just sounded live, and yet it had the widest distribution. It also had a smooth distribution with one hump. 100% silver cable.

Steve N.
 

Empirical Audio

Industry Expert
Oct 12, 2017
1,169
207
150
Great Pacific Northwest
www.empiricalaudio.com
I would not use "upsampling" and "decent" in the same sentence :) Had a fifo with a local clock in mind. Even my old ML could do that. Which does not eliminate the differences between cables but shifts the focus away from jitter.

Otoh it was probably a silly remark as there is still a huge user base with pll spdif decoding.

A master clock in the DAC with a FIFO can work, but only for a limited time until it overruns or underruns. The only way to stop this is to PLL lock the source to the DAC.

Resampling in DACs is usually disappointing IMO, plus it does not give you the opportunity to drive the DAC with even lower jitter sources as they become available, as well as new types of interfaces, such as Ethernet, taking advantage of the lower jitter these can offer.

On the other hand, the plots you are looking come from my upsampler, the Synchro-Mesh, which is reclocking a Sonos. This just happens to be a convenient source on my test bench. I can get about 20psec of jitter with the Synchro-Mesh. It is possible to get superb results with a reclocker if implemented optimally.

Steve N.
Empirical Audio
 

microstrip

VIP/Donor
May 30, 2010
20,806
4,698
2,790
Portugal
The plots are a distribution of period measurements. There is a nominal period measurement, which would be a single spike in the center if there was no jitter, at a height equal to the total number of sample measurements.

Each spike represents the number of samples of period measurements that were recorded at that particular period. The taller the spike, the larger the number of measurement hits at that particular period.

As for correlation, I spent yesterday with my wife doing listening tests. She was unaware of which cable was in the system.

I updated the link to show the results of these listening tests. In a nutshell, the narrower the overall distribution, the better the rendering of high-frequencies and venue echoes, even with 2 humps. The wider and deeper the soundstage became.

However, I also found that the lifelike quality of the vocals was more related to the cable construction and conductor material. My Bitmeister had one of the best renderings of vocals and just sounded live, and yet it had the widest distribution. It also had a smooth distribution with one hump. 100% silver cable.

Steve N.

Perhaps you told us before and I do not manage to find, why aren't you making BNC cables?
 

Empirical Audio

Industry Expert
Oct 12, 2017
1,169
207
150
Great Pacific Northwest
www.empiricalaudio.com
Perhaps you told us before and I do not manage to find, why aren't you making BNC cables?

I am. It's called BNC-BNC and sells for $275 plus shipping and PayPal fee. This is the top plots, the best jitter, around 20psec.

Steve N.
 

jkeny

Industry Expert, Member Sponsor
Feb 9, 2012
3,374
42
383
Ireland
The plots are a distribution of period measurements. There is a nominal period measurement, which would be a single spike in the center if there was no jitter, at a height equal to the total number of sample measurements.

Each spike represents the number of samples of period measurements that were recorded at that particular period. The taller the spike, the larger the number of measurement hits at that particular period.
OK, from what I can gather this is a TDC, time to digital converter scope/device. Typically these have 45 psec resolution, that is the finest binning.

Although TDCs are not as accurate as Time Interval Error (TIE) measure of jitter - it helps to understand what the plots show by describing how TIEs work?

"The time difference between a real clock and an ideal uniform time scale, after a time interval following perfect synchronization between the clock and the scale. "

That is, the scope is set to collect a huge amount of data points. (10 -200 million points).

Then the run is stopped, and the data edges are searched for. An algorithm looks at the data structure, defines the Unit Interval (UI), and from there recovers the frequency of the fundamental clock embedded in the data.

When this ideal reference fundamental clock is reconstructed then each individual edge is examined again, and the difference between it's actual crossing time & the reference crossing time is what is called TIE.

So, with a TIE plot all transition points perfectly matching up with a reference crossing time would give a spike which consist of a single line with no broadening i.e. all the transitions would be exactly at the right time. Real world clocks are not this accurate so a certain amount of short term (not long term) drift is normal.

The TDC used in Steve's plots are not as accurate as TIE so the 'reference' has at least 45 psec of jitter already in it but the following still appplies

In order of decreasing performance the following plots would apply, I believe
- a single spike with as little broadening of the spike as possible (this also means that the spike will be higher i.e. the number of hits on time will be higher)
- a broader single spike i.e. more hits are off slightly from ideal
- two spikes instead of one spike would imply that there are transition timing errors concentrating around two distinct timings. I'm not uture I follow your explanation for this "The two humps indicates that the driver pulling high has a slightly different output impedance than the drive pulling low." The transitions are either on rising or alternatively, on falling edges b, not on both - or am I wrong here?

As for correlation, I spent yesterday with my wife doing listening tests. She was unaware of which cable was in the system.

I updated the link to show the results of these listening tests. In a nutshell, the narrower the overall distribution, the better the rendering of high-frequencies and venue echoes, even with 2 humps. The wider and deeper the soundstage became.

However, I also found that the lifelike quality of the vocals was more related to the cable construction and conductor material. My Bitmeister had one of the best renderings of vocals and just sounded live, and yet it had the widest distribution. It also had a smooth distribution with one hump. 100% silver cable.

Steve N.
Right, thanks
Some similar (TIE) measurements here of an SPDIF device & PS & other changes to it some of which correlate to SQ. This guy is using A TEK 7354. 35 000 Euro, Stock, without the jitter analysis package. That can cost another 10000 in a worst case..

At this point, & looking at your plots Vs listening impressions, I'm off the opinion that a reduction in close-in phase noise results in better rendition of finer, low-level details which would mean the room ambience is better portrayed & auditory images are more solid, less diffuse. Once I sourced clocks with low close-in phase noise I was finally able to confirm this as the audible result of this improvement.

This reduction in close-in phase noise shows up in your & TIE plots as less broadening of the single spike
 

Empirical Audio

Industry Expert
Oct 12, 2017
1,169
207
150
Great Pacific Northwest
www.empiricalaudio.com
OK, from what I can gather this is a TDC, time to digital converter scope/device. Typically these have 45 psec resolution, that is the finest binning.

Although TDCs are not as accurate as Time Interval Error (TIE) measure of jitter - it helps to understand what the plots show by describing how TIEs work?

"The time difference between a real clock and an ideal uniform time scale, after a time interval following perfect synchronization between the clock and the scale. "

That is, the scope is set to collect a huge amount of data points. (10 -200 million points).

Then the run is stopped, and the data edges are searched for. An algorithm looks at the data structure, defines the Unit Interval (UI), and from there recovers the frequency of the fundamental clock embedded in the data.

When this ideal reference fundamental clock is reconstructed then each individual edge is examined again, and the difference between it's actual crossing time & the reference crossing time is what is called TIE.

So, with a TIE plot all transition points perfectly matching up with a reference crossing time would give a spike which consist of a single line with no broadening i.e. all the transitions would be exactly at the right time. Real world clocks are not this accurate so a certain amount of short term (not long term) drift is normal.

The TDC used in Steve's plots are not as accurate as TIE so the 'reference' has at least 45 psec of jitter already in it but the following still appplies

In order of decreasing performance the following plots would apply, I believe
- a single spike with as little broadening of the spike as possible (this also means that the spike will be higher i.e. the number of hits on time will be higher)
- a broader single spike i.e. more hits are off slightly from ideal
- two spikes instead of one spike would imply that there are transition timing errors concentrating around two distinct timings. I'm not uture I follow your explanation for this "The two humps indicates that the driver pulling high has a slightly different output impedance than the drive pulling low." The transitions are either on rising or alternatively, on falling edges b, not on both - or am I wrong here?


Right, thanks
Some similar (TIE) measurements here of an SPDIF device & PS & other changes to it some of which correlate to SQ. This guy is using A TEK 7354. 35 000 Euro, Stock, without the jitter analysis package. That can cost another 10000 in a worst case..

At this point, & looking at your plots Vs listening impressions, I'm off the opinion that a reduction in close-in phase noise results in better rendition of finer, low-level details which would mean the room ambience is better portrayed & auditory images are more solid, less diffuse. Once I sourced clocks with low close-in phase noise I was finally able to confirm this as the audible result of this improvement.

This reduction in close-in phase noise shows up in your & TIE plots as less broadening of the single spike

I can also run TIE plots. I will do this and post these as well. The B/W on my scope is 7GHz BTW. It was originally a $130K system with the software.

Steve N.
 

jkeny

Industry Expert, Member Sponsor
Feb 9, 2012
3,374
42
383
Ireland
I can also run TIE plots. I will do this and post these as well. The B/W on my scope is 7GHz BTW. It was originally a $130K system with the software.

Steve N.

OK, that would be cool as from my understanding TIE measurements would be even more accurate & maybe more revealing?

The real value in these measurements, however is in isolating & relating plot characteristics to auditory perception so keep up the good work. It's difficult as there are many changes to such plots when changes are made to electrical performance of SPDIF transmitter. For instance, 7 years ago I did some experiments on the Hiface SPDIF transmitter - one of the biggest sound improvements was a individual 3.3V supply to the clock - further improved by a move to a 3.3V battery supply to the clock.

So very interested in your correlations between plot shape & SQ & also what can't be correlated as I expect these jitter plots are only a small part of the SQ picture

PS. Have you ever tried a low close-in phase noise clock (an actual measured one) in the test & measured?
 

Empirical Audio

Industry Expert
Oct 12, 2017
1,169
207
150
Great Pacific Northwest
www.empiricalaudio.com
OK, that would be cool as from my understanding TIE measurements would be even more accurate & maybe more revealing?

The real value in these measurements, however is in isolating & relating plot characteristics to auditory perception so keep up the good work. It's difficult as there are many changes to such plots when changes are made to electrical performance of SPDIF transmitter. For instance, 7 years ago I did some experiments on the Hiface SPDIF transmitter - one of the biggest sound improvements was a individual 3.3V supply to the clock - further improved by a move to a 3.3V battery supply to the clock.

So very interested in your correlations between plot shape & SQ & also what can't be correlated as I expect these jitter plots are only a small part of the SQ picture

PS. Have you ever tried a low close-in phase noise clock (an actual measured one) in the test & measured?

I use custom clocks for all my products. I have these measurements for Crystek and others at the S/PDIF outputs. Are you talking about only measuring the clock?

If so, I need a probe and my latest active probe is currently dead, 2nd one I have killed. They are very fragile. I would have to get another one.

Steve N.
Empirical Audio
 

jkeny

Industry Expert, Member Sponsor
Feb 9, 2012
3,374
42
383
Ireland
BTW, doing an FFT plot of the TIE is revealing of the frequencies at which the mistimings are occurring.
For instance from the plots in the link I gave plot shows mistimings at low frequency. We have to be careful how to interpret this frequency plot - I believe it's a plot of the statistical distribution of the hits?
For instance here's a TIE plot & FFT of the frequency

Plot.png

or this one
Plot(3).png

So we see, in this plot, a higher concentration of hits around 1.5MHz but also at low frequencies

So am I correct in thinking that low frequency hits directly relate to low frequency issues, perhaps PS issues or close-in phase noise issues & the spikes around 1.4MHz spike is because the spdif bit rate at 44.1khz is 2.8224 MHz. Half of it is 1.4112 MHz

So, I'm surmising that a perfectly timed signal would result in a spike at the 1.4112 MHz frequency with no side spurs anywhere else?
If that is correct the close in phase noise would be seen as spikes around the 1.4112 MHz frequency or a broadband raise in the area around this frequency
 
Last edited:

jkeny

Industry Expert, Member Sponsor
Feb 9, 2012
3,374
42
383
Ireland
I use custom clocks for all my products. I have these measurements for Crystek and others at the S/PDIF outputs. Are you talking about only measuring the clock?

If so, I need a probe and my latest active probe is currently dead, 2nd one I have killed. They are very fragile. I would have to get another one.

Steve N.
Empirical Audio

OK, I'm talking about the measurement directly from the output of the oscillator itself before any transmission through SPDIF transmitter or cable
I wondered how using a standard clock Vs 'good' (low close-in phase noise) clock would be seen in these TIE measurements

It wasn't until I got these individually measured clocks that I could definitely confirm the audible effect of clocks with low close-in phase noise - relying on the plots published by manufacturers tends to be a mistake - I'm told they select the best & publish that as the phase noise (if they actually spec phase noise which most don't) - the close-in phase noise range for the clocks that are then actually sold are usually pretty wide

Here's phase noise plots of the selected clocks.
NDK clock.jpg
 

Empirical Audio

Industry Expert
Oct 12, 2017
1,169
207
150
Great Pacific Northwest
www.empiricalaudio.com
OK, I'm talking about the measurement directly from the output of the oscillator itself before any transmission through SPDIF transmitter or cable
I wondered how using a standard clock Vs 'good' (low close-in phase noise) clock would be seen in these TIE measurements

It wasn't until I got these individually measured clocks that I could definitely confirm the audible effect of clocks with low close-in phase noise - relying on the plots published by manufacturers tends to be a mistake - I'm told they select the best & publish that as the phase noise (if they actually spec phase noise which most don't) - the close-in phase noise range for the clocks that are then actually sold are usually pretty wide

Here's phase noise plots of the selected clocks.
View attachment 37812

I made a lot of TIE measurements, which I will post tomorrow.

After doing the listening correlation tests, I realized that my rather old poorly measuring Bitmeister cable actually delivers liquid vocals, even though the imaging is not great and my new partial silver cable is lacking in the liquid vocal department. Today, I modded my reference cable to try to remedy this and low-and-behold, it fixed this. It is now more live than ever. I will have to charge more than $275 for it though. I don't like to offer 2 versions, but I may have to. Lots of customers want a $275 cable and anything more will put them off. This is simply the best cable I have ever heard. Customers have sent me their cables to compare and nothing comes close now.

I also went back to my source termination in the Synchro-Mesh and tuned it slightly up and slightly down to see how the cables respond. The undershoot increased in both cases and the two peaks in the plots moved apart more, so I went back to my original termination, which I empirically set based on the 1694A cable. I guess all of these cables are pretty close to 75 ohms. There might be tiny tweak left though, a few ohms.

This work is paying off already.

Steve N.
Empirical Audio
 

Alrainbow

Well-Known Member
Dec 11, 2013
3,189
1,387
450
Steve is one I truly trust , he was way ahead in spidif converters and I feel he made Among the best of them. I own a few cables from Steve. It's always cheaper and. Better than the pricey ones. The spidif cable Steve is posting about sounds great and every time I need to show some one the inprovemnt there can be its there.
 

jkeny

Industry Expert, Member Sponsor
Feb 9, 2012
3,374
42
383
Ireland
I made a lot of TIE measurements, which I will post tomorrow.

After doing the listening correlation tests, I realized that my rather old poorly measuring Bitmeister cable actually delivers liquid vocals, even though the imaging is not great and my new partial silver cable is lacking in the liquid vocal department. Today, I modded my reference cable to try to remedy this and low-and-behold, it fixed this. It is now more live than ever. I will have to charge more than $275 for it though. I don't like to offer 2 versions, but I may have to. Lots of customers want a $275 cable and anything more will put them off. This is simply the best cable I have ever heard. Customers have sent me their cables to compare and nothing comes close now.

I also went back to my source termination in the Synchro-Mesh and tuned it slightly up and slightly down to see how the cables respond. The undershoot increased in both cases and the two peaks in the plots moved apart more, so I went back to my original termination, which I empirically set based on the 1694A cable. I guess all of these cables are pretty close to 75 ohms. There might be tiny tweak left though, a few ohms.

This work is paying off already.

Steve N.
Empirical Audio

Great! Look forward to seeing them

How to correlate the plot shape to auditory perception is difficult but my 2 cents
Working from the ideal single very narrow spike, which signifies that all timing is accurate from sample to sample - which means that all samples are processed at the correct time.
What happens if a sample is mistimed - it's the same as if it was a wrong sample value arriving at the right time.
In the frequency domain, it means that this sample, which should represent the amplitude of a point on a waveform, is slightly wrong in amplitude.

So back to the plots - I reckon that a single but widened spike could translate into constant fluctuations in the frequency waveform - in other words, if timing errors are significant enough (broader spike) then if a pure tone was being replayed it would be smeared away from purity. The wider the spike , the more smeared the tone.
What would be the result of two narrow spikes on the plot on replying this pure tone? I reckon it would produce two clear tones slightly off in frequency from the target tone's frequency

So how would all this be perceived playing complex signals like music? This is where the translation becomes somewhat speculative
I reckon the single broad spike will result in frequencies in the replayed music fluctuating around their recorded values - diffuse & less defined note purity - may well be more noticeable at lower frequencies as, from my reading, we seem to be more sensitive to fluctuations at lower frequencies. At a higher perceptual level, I believe this translates into soundstage definition - less precise

Two narrow spikes would be perceived in a different manner because it's creating two distinctly defined, slightly offset frequencies instead of one specific frequency for each note in the music. This may well be perceived as pleasing, giving a sense of more richness or depth (don't know just guessing on this one) to the sound?

The only way to tease all this out would be a careful & forensic approach trying to isolate just a specific change in jitter to its auditory perception possibly with pure tones, then with complex tones & then with music.

This sort of correlation between measurements & auditory perception is what is badly needed but also very difficult & complex
 

About us

  • What’s Best Forum is THE forum for high end audio, product reviews, advice and sharing experiences on the best of everything else. This is THE place where audiophiles and audio companies discuss vintage, contemporary and new audio products, music servers, music streamers, computer audio, digital-to-analog converters, turntables, phono stages, cartridges, reel-to-reel tape machines, speakers, headphones and tube and solid-state amplification. Founded in 2010 What’s Best Forum invites intelligent and courteous people of all interests and backgrounds to describe and discuss the best of everything. From beginners to life-long hobbyists to industry professionals, we enjoy learning about new things and meeting new people, and participating in spirited debates.

Quick Navigation

User Menu

Steve Williams
Site Founder | Site Owner | Administrator
Ron Resnick
Site Co-Owner | Administrator
Julian (The Fixer)
Website Build | Marketing Managersing