New jitter measurement technique

[Empirical Audio]

I don't trust the current spectral jitter measurement techniques to tell me anything about the sound quality. There does not appear to be a correlation between jitter and SQ, except for maybe at a gross level ie; 100nsec jitter sounds worse than 10psec of jitter. I want to change this.

I have been using a direct real-time measurement method rather than spectral, I drive the S/PDIF signal to a 7GHz programmable scope with jitter measurement software on it that measures the variation in period of the waveform. This is showing real promise.

I have been able to capture the jitter of two very low jitter systems, my own Interchange Ethernet renderer and a Raspberry Pi HAT board with S/PDIF output. The HAT board claims to have 4-6psec of jitter and they publish the measurements. I will not divulge the manufacturer here.

The sound quality of the HAT board is very good, rendering detail similar to my own renderer at 1/20th of the cost, however there is a "hardness" to the sound and the deep image is a bit confused at times. To me, it is a bit stressful to listen using this HAT board. I am much more relaxed listening to my Interchange. I really expected that if the jitter was as low as advertised, that it would match or maybe even beat my Interchange.

Therefore, I used my measurement system to measure both to see if this difference in SQ might be detectable and maybe even predictable. The following plots are created by collecting thousands of period measurements on the fly while an entire track plays. The same track was played in complete for each Jitter measurement. The plots are histograms of the period. Both devices were driven from the same software and network, Jriver being the player. Both devices were powered the same way, both using the same LPS.

HAT board Jitter with 44.1 track:


Empirical Audio Interchange jitter with 44.1 track:


As you can see, both of these devices produce very low jitter, but the shape of the HAT board distribution might explain the different sonics.

This is an important observation because it is very possible that some devices with higher jitter may actually be more musical that some with lower jitter because of the difference in the jitter distribution. It is no longer I feel sufficient just to have low jitter. The distribution plot must be a single peak or closely spaced peaks with clean profile in order to achieve the best sonics.

Steve N.
Empirical Audio

 

[mvs0]

Hi Steve,

Nice way of measuring jitter.
I noticed that the period measured is 177 nsec which means you measure both edge positions of the bi-phase signal.
I thought the timing of the SPDIF signal was defined by the edges on the bit changes and not by the data dependant extra bit-flip in the middle of the timing edges.
Could it be the two peak histogram of the HAT board is caused by a different timing of that bit-flip in the center of the real timing defining edges?

 

[Empirical Audio]

Hi Steve,

Nice way of measuring jitter.
I noticed that the period measured is 177 nsec which means you measure both edge positions of the bi-phase signal.
I thought the timing of the SPDIF signal was defined by the edges on the bit changes and not by the data dependant extra bit-flip in the middle of the timing edges.
Could it be the two peak histogram of the HAT board is caused by a different timing of that bit-flip in the center of the real timing defining edges?

My own devices also had the double-hump prior to me making more mods to them to reduce jitter. I think it has to do with the difference in rise and fall times and maybe every other rising edge. Every time I see the double-hump, there seems to be a "hardness" in the sound quality. The Sonos has a similar signature, two humps. As you reduce jitter using power and timing techniques, the two humps move closer and eventually disappear.

These are old measurements BTW. The latest comparison I made was between the Helen re-clocker and my Synchro-Mesh re-clocker:

https://www.audiocircle.com/index.php?topic=163712.0

 

[mvs0]

My own devices also had the double-hump prior to me making more mods to them to reduce jitter. I think it has to do with the difference in rise and fall times and maybe every other rising edge. Every time I see the double-hump, there seems to be a "hardness" in the sound quality. The Sonos has a similar signature, two humps. As you reduce jitter using power and timing techniques, the two humps move closer and eventually disappear.

These are old measurements BTW. The latest comparison I made was between the Helen re-clocker and my Synchro-Mesh re-clocker:

https://www.audiocircle.com/index.php?topic=163712.0

Would be nice if your equipment could make split histograms for all combinations of rising/falling edges and single and double pulse width. The result could explain the source of double-hump.

 

[Empirical Audio]

Would be nice if your equipment could make split histograms for all combinations of rising/falling edges and single and double pulse width. The result could explain the source of double-hump.

I wish I had more control over this with the software. There are some dials I can adjust, but it's not clear what the effect is. Not much can be adjusted unfortunately. Since this is a complex signal, it has to lock onto one interval and focus on measuring that. With some signals I find that it locks onto the longer interval rather than the shorter one. Not sure I can influence that.