USB Clock Upgrade My Experiences (so far)

[PeterSt]

Just to clarify, that is what I'm suggesting above, use the dexa trim pot. This is the set up I had years ago with the m-audio card and audiocom clock. You will just be setting the best sounding frequency on the dexa relative to the Paul pang card.  I could make a big difference. Listen for the sound stage expanding whilst keeping good dynamics and no sharpness.

Nick,

As long as we realize that you are doing this with a 600 euro device on one side and a relative el cheapo on the other, while all it takes is adjusting one of the clocks. That either (actually the master) should be fairly stable is something else but that does not take 600 euro. No separate PSU either.

I am not stating this out of all context; I only say that when this works it needs a low ppm oscillator as master and a VCXO as slave. WHEN this works.

For others : A stable oscillator like in ppm terms (parts per million) is nothing like a low phase noise (low RMS jitter) oscillator or clock setup. To this regard the Dexa is only a low ppm device (long term stability) and isn't about "low jitter" as we know it for audio.
Of course this is a totally different subject and all I say with it is that the 600 euro is crazy if you ask me.
Nick, until someone comes up with phase noise plots (and for us both not anything new).

What I am really saying is that - assumed we are heading for the better solution with some sort of "trimming" is that any low cost oscillator can be used. And with that : That it is quite useless to have it all perfectly working after maybe several trials and much listening, and then all we know is that it costs a 600 up to two times of that. To me it (now) looks as starting at the wrong end as long as these (new) theories are assumed to be valid.

Peter

[PeterSt]

I just cannot get used to this step backward in sound quality - I only heard two Dexa's for a few days - where is the bass, where is the crystal easy clarity?

Can't wait to get that Dexa PS due Monday ggrrhhh!!!

A litte more reasoning (only that guys !) :

Assumed your ears are still very OK Paul, this (see quote) should tell something;
IMO it very well can be that the very stable Dexa's (at both ends) - that's what they are - do not even create an out of sync often (this in itself leads to other theories but never mind that for now). So, they may not run at the same exact speed, but the time time period needed before two up (or down) going slopes miss eachother may be very long. Just think 24M pulses per second, and next look at the ppm rating of them (I don't know at this time) and it could take hours for things going wrong. Don't ask me what happens when it *is* wrong but now I need to look into how async USB really works and how things which needs to transmit are once per x milliseconds in the first place. It may not even play a role (because all covered for in the protocol). However :

What I wanted to say is that envisioning the two clocks just running behind eachother, their both pulses implying noise, that noise pattern will always be the same *because* they are so stable. But what it would need is
a. getting them on that same speed;
b. find the pattern which suits you (sound wise).

But do you see what happens when this is all working like this anyway ?
Once you found the speed matching, it needs slowing down one of them and bring it up to the same speed, only to find that other pattern.
You can't have done such a thing.

So next theory :
They both never run at the same speed, but both are still as stable. This is now about the low oscillation I talked about, and this is instead of the more random wild influencing thing (the one clock catching up a full cycle many times per second). Now, the pattern will still be important but is out of our control, but the pattern has I think less wild excursions (very high peaks of coincidentally same pulses at exact same time). Still it should be so that once in a while this same excursion happens, and possibly it is audible through a kind of flanger. It could widen (or deepen) the sound stage if often enough (like maybe 10-20 times per second) and now it even can be so that the RMS jitter we do talk about with audio also plays a role;
Now suddenly the *high* (short term) jitter of the Dexa (hey, that assumed) plays a role because it will spread those peaks coming together for a longer time (because they are so long term stable).

Anything else ?
If this were all true, it needs a high long term stable device (thus TCXO) with a cheap high jitter oscillator in there.

or ...

A high jitter oscillator at one end, connected to both.
Yeah, made this one up because it follows from my own logic.
Ultimately stable, spreading ... hmm ... its own noise pattern ?
Not sure about this latter.

Peter

[Nick]

Peter hi,

As long as we realize that you are doing this with a 600 euro device on one side and a relative el cheapo on the other, while all it takes is adjusting one of the clocks. That either (actually the master) should be fairly stable is something else but that does not take 600 euro. No separate PSU either.

The Paul Pang card has a TXCO and the Dexa is temp controlled (not oven but 37 deg constant). The Spec that Dexa quote for stability is 0.2ppm which is OCXO territory so I think there is a good chance that Paul can get close enough. I'm just suggesting that Paul gives it a go just to see what happens.

My context for this is that the last time I was doing this with clocks was back in about 2008/9 with a £52 pound M-Audio transit interface with an Audiocom Superclock 4 fitted (not in the same class IMHO as the dexa and the NOS) and the results were very good and repeatable when tuning the M-Audio clock against a stock PCI USB 2 card with a standard Xtal fitted. I don't think we need to be too hung up on precise cock speed matching, just tune my ear, believe me you know when you reach the sweet spot 

Some nostalgia

http://www.phasure.com/index.php?topic=979.msg12931#msg12931

I agree the Dexs the costs is a bummer but the based on subjective listening its just the best clock that I'v listened to to date, and MUCH easier to apply consistently without ground loop problems than direct coupled clocks (these have been a problem for me in the past) which is convenient for DIY. I don't know what the Phase Noise Spec is but they do sound very good. I'm not suggesting that anyone else uses them, they work well but you might have to be a little mad to have the three that I have in my system

I am with you about the intermodulation of the two clocks being a possible problem. I lost a post earlier (dammed ipad battery) about the clock speed and thoughts on possible effects of the clocks on data and SQ. In essence i'm not really seeing this as an electrical noise issue. My money is on data transmission error rates being vastly improved and if USB "bulk transfer" mode is being used to transfer data then the reduced error rate will reduce the number CRC errors and so the number of "Stalls" and retransmissions on the link. Of course this assumes that the stalls disrupt data processing further down stream but this seems possible. I think the USB receive buffers on the NOS USB chip may also have a role in the effect of clock speed mismatch in that tuning the relative speed of the clocks might mean that there is less data transmission speed management needed between the ends of the USB link. The event above only have to happen at audio frequencies and its not a big jump to think that this is going to be audible.

All theory, I guess the only way to prove this would be with a 480Mhz USB analyser to watch the USB Protocol error rate and frequency of dataflow management packets to see if they change with better clocks and lock speed matching. But such analysers are big bucks and really whilst I would love to know what is happening the resulting sound may have to be enough.

My thoughts at the moment on clock speed are that if clocks running at the same speed (slaved) turns out to be the best set up, then this will have a very nice simple elegance to it. I think this might well be the best solution but part of me is still thinking that, taking the USB receive buffering into account, it might be that a slightly faster NOS clock could be the sweet point. As mentioned above I'v ordered some test equipment to try to pin what the best solution is. Meanwhile the slaved clock build continues just to see what happens 

Regards,

Nick.

[PeterSt]

then the reduced error rate will reduce the number CRC errors and so the number of "Stalls" and retransmissions on the link. Of course this assumes that the stalls disrupt data processing further down stream but this seems possible.

Yes Nick. I'm sure you didn't read my attempt to start this subject while posting your last post (see my previous), but this is another area which could be more happening than we like. But I don't know really.

Btw, I plan to have my "link" running next Monday.
If I manage to read your email before that. Haha.

Best regards and of course thanks,
Peter

[Nick]

then the reduced error rate will reduce the number CRC errors and so the number of "Stalls" and retransmissions on the link. Of course this assumes that the stalls disrupt data processing further down stream but this seems possible.

Yes Nick. I'm sure you didn't read my attempt to start this subject while posting your last post (see my previous), but this is another area which could be more happening than we like. But I don't know really.

Btw, I plan to have my "link" running next Monday.
If I manage to read your email before that. Haha.

Best regards and of course thanks,
Peter

Your right Peter you posted whilst I was drafting. I think we have similar ideas. Wouldn't it be nice to have access to USB analyser just to know one way or the other.

Great that your link could be up by Monday. I think ill back off on my build and see what you make of it first. Fingers crossed for a good result ! Do let us know 

Regards,

Nick.

[Scroobius]

So with Dexa in NOS1 and Pang pcie at the PC I adjusted the "heavenly trimmer" as far as I could but I can't honestly say I can hear any difference - just now it is hard to tell if it is having any effect at all.

Maybe the Pang card clock is not close enough to the Dexa clock in frequency.

Paul

[PeterSt]

Interesting in itself Paul;
Your guess is probably the best bet.

But I must say that by now I'm a bit lost on your and Nick's situation; Wasn't Nick doing the same ? I mean, he is not the one with two Dexa's (24.000), right ?
So what is the difference ? Or maybe Nick doesn't have a Paul Pang card ?

Regards,
Peter (sorry that I didn't read through all again to find out)

[Nick]

Interesting in itself Paul;
Your guess is probably the best bet.

But I must say that by now I'm a bit lost on your and Nick's situation; Wasn't Nick doing the same ? I mean, he is not the one with two Dexa's (24.000), right ?
So what is the difference ? Or maybe Nick doesn't have a Paul Pang card ?

Regards,
Peter (sorry that I didn't read through all again to find out)

Hi Peter,

Here I have 3 dexas, two 24mhz for usb and one 24.5mhz for audio stream. No Paul Pang card used here, prior to the dexas two vcxo 24mhz diy built clocks were used to test the approach using transformer coupling before fitting the two 24mhz dexas when I was sure that a reference grade clocks were warranted.

Here trimming of the usb clocks is carried out on one of the 24mhz  dexas which happens to be at the pcie card end of the link. It can just as well be at the nos end dexa as the result in the same.

I'm not sure why Paul is not experiencing change but my initial guess would be, as you both mention, that the trim available on Pauls dexa is not sufficient to match pauls one daxa with his Paul's Paul Pang card.

Regards,
Nick.

[Nick]

So with Dexa in NOS1 and Pang pcie at the PC I adjusted the "heavenly trimmer" as far as I could but I can't honestly say I can hear any difference - just now it is hard to tell if it is having any effect at all.

Maybe the Pang card clock is not close enough to the Dexa clock in frequency.

Paul

Paul,

Good to give it a try anyway but a distinctly "average trimmer". I hope it might go better when dexa no 2 is back in place.   My guess would also be that the gap in speed is too big to "match" well. Thanks for coming back with the results, it all adds to the picture of what might be going on 

Nick.

[PeterSt]

Thank you for clarifying Nick.

and one 24.5mhz for audio stream.

I think I saw you writing this more often, but (and not important) I think you mean the 22.xxx MHz here ?

Peter

[CoenP]

I recently read about a genius idea for clock distribution that I like to share.
Credit is due to the guy from ecdesigns who has developed a 44.1/16 SD card reference player.

The principle is that splitting a clock signal and distributing over a distance it will degrade the clock quality and introduce noise. The strikingly simple answer is to filter the incoming distribited signal with an chrystal filter, which has a very narrow passband letting only the original signal trough. In his wording this filter acts as a highly specific "tuning fork" requiring little energy to resonate and as such providing for a light load tot the master clock generator. One can use this principle for any clock distribution ánd/ór transport.

How about that!

Regards, Coen

[PeterSt]

A passive repeater eh ?

Thank you Coen !
Peter

[PeterSt]

Guys, I can use a little help - just making some short cuts because of a lack of time to sort it out decently at this moment;

So, I have my super sneaky connection running - partly. Still in all over secrecy think like I have such a NEC based PCIe card tweaked with the direct clock connection (card's side is the clock). That card does not run; it does per any normal means of connection (and without direct clock connection).
I have another of such a card (different brand). That *does* run per the special connection, but that is not the tweaked one.

With "run" I mean that the PC sees it. So "not run" means the PC does not see it.

Many variables are there to change, but too many (and all too fragile) to start without more knowledge. One of the options is that the card which does not run now needs to have the Molex connected (hence needs that separate supply). With the normal connection it does not need that.
The card which runs does not have the separate supply connected, but also no clock connection; this can matter.

Now finally the question :
We always seem to speak about the card needing 12V, or I misread or just dreamt that. Anyway, I did supply the 12V with a sufficient amount of Amperes;
Does it need 12V or does it need 5V ?? ... or both ?
Watch out : I say "need" but I already know it needs neither (from the Molex). So it is merely what it utilizes once the Molex is connected (both 12V and 5V being in there). The "utilizes" comes from the special connection if the card thinks so, so to speak.

My short cut is about that it is not so easy to for fun (and no avail) give it a 5V as well with sufficient juice. And so I like to know it in advance, if anyone knows the answer anyway.

And FYI : I can not understand why the card is not seen by the PC just because of a possible failing clock connection, IOW that should not be it.
Of course, my fastest option would be to hop over the clock connection from that one card to the other, but as said, this is all fragile and I like to stay away from it when possible.
And again in other words, if someone tells me that 12V is what's needed, then I'll hop over the clock connection (because 12V it has now). Or when 5V is needed I change the PSU. When both are needed, I'll add another (also not easy for reasons and by then I'll have Anthony's LPSU mimiced - haha).

Ok, maybe this was a combined question with status report; all looking good, but a step which should work fails at this moment.

Great thanks for any insight,
Peter

PS: Apologies that I do talk in secrets indeed, because it is too much fun what I'm doing, and when it fails I had rather not tell it. But I will when it works !

[Scroobius]

Hey Peter - I am not sure if this helps but it might. The Paul Pang pcie USB3 card is a standard NEC based pcie card with a piggy back board (added by Paul Pang) that has the low jitter clock on it. On both the main board and the piggy back board there are regulators that get warm each with a lump of Al glued to them - certainly added by Paul Pang. The card has to have a Molex ps  or it will not work and it only uses the 5V supply on the MOLEX - mine only has the 5V connected so I know it does not need 12V.

That might indicate that the standard pcie card does not need 12V - maybe!!.

Paul

[PeterSt]

Thanks a bunch Paul !
Peter