EMI susceptibility of audio and USB clocks

[AlainGr]

Peter,

OT: am I ok to assume that between 2 USB ports, there should be a "master" and a "slave" ? If so, then to what level does the 2 clocks need to be tuned together (since I figure that this already  happens between them) ?

Alain

[Nick]

Alain, All,

After returning from my visit to Peter's I was left with a big question - Why did the USB clocks not perform as expected in Peters system ?

Peter provided an account and his view above in this post of what may have been happening regards USB clock noise etc. I know that the dexa clocks (read improved clocks on standard xtals in USB interfaces, forget Dexa as a make here) have made a significant positive impact in 5 of the 6 systems that they have fitted. Before posting my comments and thoughts on how the USB clocks performed at Peter’s and why, I wanted to spend time investigating what might have happened to impact their performance. I have been doing this for a couple of weeks now and am reasonably sure that I understand now what happened. I will post more on this shortly, but for now…

To the point that you make above Alain, the relative speed of the two clocks at the ends of the USB link is very important to sound quality. As part of the work to diagnose the problems at Peter’s I now have a way to show the relative speed of the two USB clocks in my system. This is done is such a way that the relative clock speed can be looked at in real time across the room from my listening position whilst listening to music. The display will resolve fractions of one cycle of phase shift between the clocks (eg a few degrees of one 360 degree “beat” between the clocks). Viewing the relative clock speed in real time whilst listening its clear that speeds do significantly impact sound quality. In fact it changes sound differently if the PC PCIe USB clock is running faster or slower then the NOS1 USB interface clock. Same speed, or as close as possible to this, is best subjectively for SQ.

The USB clocks in my system are tuned now to a frequency which is better than 1 part per million of each other (this 1ppm is the max and for some time they run at the same speed exactly). When we tried the clocks at Peters they were matched very badly (worse than 100 parts per million). Sound now is entirely different to what Peter and I were listening to.

If these results are valid and relative USB clock speed does impact sound quality, then without the capacity to tune the USB clocks it may very well just be luck that determines how good a particular PCIe USB3 card sounds with a NOS1. The luck being that all the many variables that impact the Xtal speed and stability of on the USB card and NOS1 USB interface just happen to “line up” such that the two crystals run at close speeds to one another and sound good. That is a worrying thought as even different instances of the same PCIe USB card model may sound different…..

Paul is bringing his Silverstone USB card, and DAC (fitted with USB Dexa clocks) to mine this weekend to look at this and we hope to get more supporting evidence. With respect to the SIlversone USB card, unfortunately the act of measuring the oscillation speed of a raw crystal’s output will change the speed (due to capacitance of the probe that is being used to do the measuring) so tuning the card to the Dexa in my NOS1 is unlikely to be possible but we will see what we can do. I will update when we have more information.

Regards,

Nick.

[CoenP]

Hi Nick,

Interesting progress!

We have a saying here: "when you have two clocks you never really know what the hour is...". My take on this phenomenon would still be a 'master USB clock' solution. This will always imply a fixed and identical base frequency yet the clock distribution is likely to introduce different phase noise at the different ends. I'm too distracted to look deeper into this subject, yet it may hold even more potential. At least it lacks the synching of the clocks and the clock source may be decoupled from the source and destination noise sources.

I would not be surprised at all that you will find absolute and relative wandering of the clock frequencies audible.

keep us posted,

regards, Coen

[Nick]

My take on this phenomenon would still be a 'master USB clock' solution.

Coen hi,

I agree this needs to be tried. I have had all the parts for a while now to give a single clock driving both ends of the USB link a go, just need to find the time  .

Regards,

Nick.

[listening]

Hi Nick,

very interesting point. I remember to a good friend of mine who has been an enthusiastic shortwave amateur who build his tube receivers by himself. And one very critical aspect was the frequency normal which was actively heated to a specific temperature for stability. The systems needs to heat up for over an hour for good stability. Do you think that temperature will influence the synchronization? Could it be possible that here is a reason for different SQ at various days?

Georg

[Nick]

Hi Nick,

very interesting point. I remember to a good friend of mine who has been an enthusiastic shortwave amateur who build his tube receivers by himself. And one very critical aspect was the frequency normal which was actively heated to a specific temperature for stability. The systems needs to heat up for over an hour for good stability. Do you think that temperature will influence the synchronization? Could it be possible that here is a reason for different SQ at various days?

Georg

Georg,

I think you could be bang on with what im seeing here.

The dexas have a simple resitor heater and thermostat control of the temperature of the 24mhz xtal can on the clock pcb. When the clocks are turned from on cold there is a very large difference in relative speed of the clocks. This reduces quickly (as the temps stabilise I think) at first quickly then further stabilises happens over about 10m. In the end the two clocks do not stay completely stable with respect to each other. I've got them trimmed now with a slow progression of 0 beats per second to around 10 beats per second (possibly less). The cycle from 0 beats to 10 beats and back takes about 30 seconds. I cannot prove it right now but I have a feeling this might be due to the heating cycle of the thermostat on both clocks slight over heating then under heating the crystals.

Thinking about it the beat frequency cycle could be due to other effects as well. But hysteresis in the heating / temp sensing circuit seems like it might be the first place to investigate. It's fun this stuff, whats the betting that Paul's clocks behave completely differently when we look at the haha :-).

Best,

Nick.

[Nick]

Since my trip to Peters I have been working to understand why the dexa usb clocks did not perform as expected. The first step in understanding what happened is in this post.

http://www.phasure.com/index.php?topic=2888.msg30710#msg30710

In summary this post flags the sound quality performance difference that tuning the PC and DAC usb clocks makes to sound. Being a compulsive tweaker, when Peter and I listened to the clocks I had the damed clocks significantly out of sync (of the order of 1000s of hz against a USB chipset spec typically of 24mhz +- 100ppm). It was only after returning from Peters that I worked out how to tune clocks accurately to each other.

I the post above I had to stopped short of saying that the poor performance we encountered was completely explained. This was because although synchronising the USB clocks makes a massive difference to sound I still felt there was something else about the set up that had impacted sound. So more thinking and now what i beleive to be the final issue solved making the dexas perform as i had expected.

What had happened is that whilst tweaking, I had set the speed of the NOS interface USB clock fast by 1-2 kHz and lost track that I had done this ! I was stumped as to how I could get the NOS dexa close to dead on 24mhz again (my scope frequency counter is only accurate to 3 decimal places), then I remembered the DIY 24mhz clock I'd built to test the USB clocks  . Using the diy clocks as a reference the dexa at the NOS end was retuned again to very close to 24mhz. Then PC end dexa usb was re-tuned to match the DAC end dexa with both running much closer to 24mhz.

So now the system behaves exactly as expected (totally different performance) without question these two tuning factors are the root cause of the poor performance that Peter and I experienced.

I'm really glad to have found the cause of the problem, diagnosing what happened has helped to pin down a better understanding clock related variables and how usb clocks need to be set up. It would be fun now to retry the upgraded clock config in Peters system but not that practical unfortunately 

Anyway absolutely top, top draw performance again from the modified USB link.

Regards,
Nick

[manisandher]

Nice work Nick, and thanks for sharing.

Using the diy clocks as a reference the dexa at the NOS end was retuned again to very close to 24mhz. Then PC end dexa usb was re-tuned to match the DAC end dexa with both running much closer to 24mhz.

How close is 'close enough'? And how much do you expect the Dexas to drift away over time from their current settings?

Mani.

[CoenP]

Hi Nick,

I don't really understand what you have been doing to get it right. Are you saying that the absolute value of 24Mhz is of importance?

Regards, Coen

[Nick]

I don't really understand what you have been doing to get it right. Are you saying that the absolute value of 24Mhz is of importance?

Coen hi,

The posts are none too clear  There are two factors.

1) matching the speeds of the PCIe USB card clock and the DAC USB interface clock.
2) both clocks need to run at close to their nominal frequency of 24mhz.

When I went to Peters I had unknowingly tweeked clock speeds to the point that, the DAC end clock was running > 1000hz above the 24mhz nominal speed for the USB decoder. In addition to this the matching of the two clock speeds also was way out.

Fixing both these points has things very much back on track sound quality wise  .

Regards,

Nick.

[Nick]

How close is 'close enough'? And how much do you expect the Dexas to drift away over time from their current settings?

Mani hi,

Good question  I'm not sure about how close is close enough. The  sample 24mhz "DIY" clock i used based on a module would should run at 24mhz. I might investigate more but I need to spend more money on measuring equipment to do this.

Regards the dexas drift rate. As mentioned in other posts, there is a cycle of drift between two tuned dexa usb clocks, but this peaks at about 3 or so beats per second so not that large (your clocks were tuned to this tollerance). So far once tuned to each other the frequencies of the clocks appear to be stable over time. I'll keep checking on this but I don't think they are going to need retuning regularly.

Regards,

Nick.