Great PC Tweak - A Bert "Must Have"

[PeterSt]

Here it works without the Molex. I think others claimed that as well. Quite confusing ...

[AlainGr]

To tell the truth, the only way I could "cut" the 5V was with the Sotm card and only after starting XXHE for the first time in a session...

So that is confusing for me too...

Alain

[Nick]

To tell the truth, the only way I could "cut" the 5V was with the Sotm card and only after starting XXHE for the first time in a session...

So that is confusing for me too...

Alain

Alan hi,

I think you said you were getting a Paul pang card. If so, looking at Paul's last weekend the clock is powered from the molex. If the molex is not connected there is no clock signal to the usb chipset so the card will not run even though the USB chipset is powered fon the 3.3v PCIe bus rail.

If your not on a Paul Pang card you got me 

Cheers,

Nick.

[AlainGr]

Hi Nick,

You are right with everything I have a PP card now, I removed the Sotm card.

I was not aware that the 5V was for the clock... But is it also used for the USB power running from the PC to let's say a powered USB device ?

Regards,

Alain

[acg]

Hi Alain,

I am having a look at the V2 PP card and I don't think that the 5V for the USB port is generated from the Molex plug, I think that it comes from the PCIe port.  Power goes directly from the Molex plug to the clock (which is the yellow board that is offset mounted) and I see no sign of anything other than the clock signal going to the usb card itself. 

Regards,

Anthony

[Nick]

Alain hi,

IIRC when usb3 was introduced the new maximum current for the external 5v supply to connected devices went up. This means that up to about 0.5 amps per usb port cards use the 3.3v rail of the pcie bus to make the external 5 volt supply with a dc to dc converter. If your external devices need more than 0.5amp you use the molex, which can supply greater current to the 5 volt external supply.

So the 5v external supply comes from either the pcie bus or the molex depending on if you have the molex connected. The usb chip always gets power from the pcie bus 3.3v supply. I think the Paul Pang card works in this way as well. It also takes its clock supply from the molex from what a saw of Pauls card.


Hope this helps.

Nick.

[acg]

Of course...I did not even think of the existing connection for the Molex plug.  So probably yes, the 900mA (top end of USB3 spec) for the USB port may well come from the Molex plug.

[AlainGr]

Hi Anthony and Nicks,

Thanks for the precisions. This brings a new question... I actually use a iFi iUSB power supply for my external SSD (contains the OS + XXHE)... Would it be (better: could it be) better used if I modify a cable to allow for a molex adaptor to power the molex ? Or would it just increase some other problems ?

Regards,

Alain

[CoenP]

Hey Coen,

Maybe it could be nice (well for me) if you can more clearly talk about what you see. For example, your "the former" seems to refer to two (!) read wires while I think you mean one of them, which now makes not much sense (to me).
Also, you might put some attention to
"i can see that the former is connected to a (one) film capacitor which in its turn is connected to one of the old xtal pcb islands."
which to me seems wrong in the first place. Not sure, but I'd say that this "former" needs to be connected to the old xtal island which from there goes to ground through the cap. Or ?

This is all not crucial of course, and it seems that I try to help you which I don't think is necessary at all. However, I am always eager to find messed up things (designs) and this seems to be one.

Btw and might that help, an active osciallator (not voltage controlled assumed) has "inputs" for voltage and ground, and for output the one clock wire;
Running the output through a transformer might not be the best idea for the signal itself (I keep on saying that any analog device is too slow for this) but may imply groundloops again (somewhere at the other side). But the supplies (gnd and V) can sure go through the transformerS.
For real isolation it needs three transformers, but I'd still wonder what happens with the still not-isolated ground at the other side.
We (or I) can also say : this never can work unless far (!) more complicated schemes are applied, and they won't lend themselves to apply on existing designs. That's what I think of it.
Mr. Pang (I think is his name) may have found himself for the same half-decisions. Any good for the better ? I again don't think so because the result will be out of (veryfiable) control ?

Peter

Sorry for being so cryptic about the clock signal isolation transformer.

If you still want to byte into it here's the deal:

In short the clock isolation transformer has:
- a primary winding with one leg connected to the clockboard ground and
- a secondary winding with one leg connected to the clockboard ground.

There is also:
- a black wire form the isloation transformer that is probably connected to the shielding of the transformer. It is soldered to clockboard ground.

- the second wire of the primary is connected to the output of the crystal oscillator (wich is dc decoupled by a small capacitor first).
- the second wire of the secondary is connected (via a film type capacitor) to - only - one of the connections where the original crystal has been on the USB3 board near the NEC chip.

The groundplane of the clockboard is connected to the USB board ground ONLY via the ground of the 5V supply (the twisted silver and red wire).
This was kind of a relief since now it is NOT an alternative groundpath of the regular USB card circuitry.

In short primary, secondary and shield all share the same clockboard ground. I think in this case clocksignal return currents have to travel however via the 5V ground wire. This must be some kind of trade off since I think the way that Nick solders both secondary wires to the original clockmounts (don't you Nick?) would be the way to go...

As far as I have been able to investigate the 5V is supplied via the MOLEX and is filtered by a pi filter with a "big" HF choke (top of the board) before being offered to the pins. Marketing rap says this will increase the quality of the data transfer (less loss, higher speed). So forget about a similar noise spectrum here....

I'm gonna let this brew in my head for a little while.

regards, Coen


p.s. before anyone overlooks it: the above is NOT how to create galvanic isolation (if one should want to persue that). Your batteries and special 5V clock supplies will be connected to the USB board ground anyway.

[Nick]

Coen hi

In short the clock isolation transformer has:

- a primary winding with one leg connected to the clockboard ground and

- a secondary winding with one leg connected to the clockboard ground.

There is also:

- a black wire form the isloation transformer that is probably connected to the shielding of the transformer. It is soldered to clockboard ground.

- the second wire of the primary is connected to the output of the crystal oscillator (wich is dc decoupled by a small capacitor first).

- the second wire of the secondary is connected (via a film type capacitor) to - only - one of the connections where the original crystal has been on the USB3 board near the NEC chip.

I didn't get time to trace through the connections on Pauls PPA card so this is interesting. Certainly different to the way I have set up my "lash up" clocks. Most of the advice that I have seen in the past recommends to make sure that the clock live wire and the clock ground are connected to the clock signal input and the ground point next to the chip that is being fed the signal.

Personally I have always tried to connect the clock ground to the ground pay of one of the two resonant capacitors that get removed when the old crystal is removed.

I don’t like making comments on someone else’s implementation because  I really don’t know that much about this stuff and there could be good reasons for the implementation, so take this with a pinch of salt but my thoughts...

It seems that the Paul Pang card has a long return route for the clock signal ground current.  There would not need to be much low frequency noise in that ground loop to cause the bass problems that I experienced in my system at the week end.

In short primary, secondary and shield all share the same clock board ground. I think in this case clock signal return currents have to travel however via the 5V ground wire. This must be some kind of trade off since I think the way that Nick solders both secondary wires to the original clock mounts (don't you Nick?) would be the way to go...

My lash up clocks have their grounds isolated by the transformer, I think this is important for sound based on past experience with directly coupled clocks. Also the loop back the transformer secondary is kept as short as possibly by attaching both the clock signal and the ground close to the NEC chip where the old clock components were.

I have to say I'm really looing forwards to trying the two third party 24mhz clocks that have been ordered. I know them to have very performance and they are galvanicly isolated. there is quite an investment though so I'm sweating until they are in and performing.
 

Cheers Nick.

[CoenP]

Hi

I just wanted to update on my experiments with the PP card.

As stated above I was not satisfied with the sound of the stock PP card. Assuming that noise is an issue and that the grounding scheme is a little odd I went to look for hidden potential in the card. As result I have tried and tested an alternative clock connection and different routings for the power supply of the card and clock board.

In chronological order I tried the following:

First I moved the leg of the transformer secondary from the clock board to the PCI card ground at the location of one of the removed capacitors (boy are these components small!). The whole idea is to vastly reduce the return path for currents generated in the secondary.
Soundwise this added some welcome body to the sound and better dynamics. Still the low end lacked power and it all sound a little 'rounded'. 

Then I moved the 5V/GND power supply from the PCIe molex directly to the clock board using a molex to SATA psu adapter cable, leaving the Paul Pang supply wires in place. The thought is to have a cleaner ground and 5V for the clock than for the USB cable. This is another big step in the right direction. The whole sound became louder and very musical, natural and organic. Perfect for classical, but unfortunately just not that suited for rock and the like.
Maybe important, but I did not modify the PC PSU molex, so in this setting there is no double 12V/GND routing anymore, only 5V/GND.

This inability to rock motivated me to disconnect the ground wire to the PCI-e card on the clock boards power supply. In the end there is an extra ground path created with this wire connected.
Without this wire only the 5V is connected, supplying to both the clock and the USB cable. Ground for the PCIe is provided by the motherboard.
This experiment was short lived, since now all naturalness and body was gone.

The next step was to disconnect both supplies from the clock board, now there is no 5V anymore. Now clock and PCIe card are fed by separate supplies from the PC PSU and no 5V is available on the USB cable.
This sounded better than the former, more dynamic and fresh. Closer listening learned however that cymbals sounded muted without body and that on crescendos the sound became thin and disorganized. Also, though I am not really good at this, there was no illusion of depth anymore.

FWIW reconnecting only the ground of the 5V to the PCIe, the roundness of the full connection returned.

Now as I feel a little tired with this card-in, card-out business, I switched on my lab supply and fed 5V/GND to the molex of the PCIe. Just as a hunch. Now we're talking! A rich dynamic sound, super musical, fantastic voices, tight bass and forever decaying sounds.

Now whats left to do is to swap the linear supply to the clock board and connect the PC PSU to the PCIe card molex.


Anyway all these experiments support my theories of grounding as well confirm findings of others. Some stuff I took away sofar:

1.  there is something with the 5V on the USB lines. It just sounds so much better with the 5V in place. This was already established in the USB cable thread.

2. The clock should have a SEPARATE 5V power supplies from the PC psu to the clock board and PCI-e card. This is the only way to be shure they have a common reference ground. I am inclined to state that only the 5V should be connected at the PCIe cards molex and not the ground, but I did not test that with an extra PSU wire.
Anyway a separate clock supply to avoid ground currents in (loops) in this apparently sensitive area is a great idea from the above findings. This is probably one of the reasons that the DEXA clocks work so well (if they do  ).

3. In retrospect the improvement in sound is amazing. For reference I refit my old USB3 card. No chance, a familiar sound but now mechanical, flat, harsh. Not that I ever noticed before! At this moment I can hardly imagine further improvements, but as we believe Nick and Paul, there is a vast potential still waiting!

I hope we can get to the bottom of this USB thing and address it in its root cause. For now this was a great learning experience.


regards, Coen

p.s.: please note that I have been talking about the PP v2 card!

edit: I noticed that my report missed some rationale. Now that only few have read it I have updated.

[PeterSt]

edit: I noticed that my report missed some rationale. Now that only few have read it I have updated.

I just read it now Coen. Superb if you ask me.

Regards,
Peter

[acg]

Coen, excellent stuff and quite timely too.

Great that you have given a second opinion about a lab supply onto the PPA card doing good things...I am glad we both agree on that point.  It is interesting the things you had to do to get it that way though.

I am confident enough to say that not only does the quality of the clock on the USB card matter, but the the quality of the power supply and ground loops also affect things, most likely because they affect the timing of the clock and the regularity at which it spews frames forth into the USB cable.

This all leaves me thinking about the clock implementation on the PPA card (and Nick and Pauls DEXA clocks).  I think we can do it better...the right clock on a USB board designed to better implement that clock (independent power circuits, layout of clock on board etc.) so that we can get away from wires between the upgrade clock board and the USB card as well as guarantee that the clock is as unimpeded as possible.  Almost certainly this would cost a fraction of the purchase price of a DEXA clock by itself and we should be able to get it to sound at least as good, or maybe even better.

Anthony 

[acg]

Sooo, I have recently received my Silverstone USB3 card (the same model that Peter has had great success with - the SST-EC04-P) to compare with my PPAstudio V1 USB card. I finally put it in the XXHE at the beginning of the week to give it some hours and today I have been swapping in and out between them with and without my lab linear psu.

Interesting is all I can say...I prefer the Silverstone card.   More clarity and resolution especially noticeable in mids and highs.  By comparison my PPA card is a little smeared and slow which I can really notice on the strike of a piano key and the presentation of voices.  The most important thing for me though (as Peter would know) is that it seems as though the 'speed' or 'resolution' has improved along with clarity of the presentation.

The next step was to pull out the 5v molex connection from the XXHE SMPS and replace it with a 5v rail from my lab LPS.  Both cards improved along the lines of clarity and presence with a less harsh top end, but the Silverstone was still clearly different and clearly the one that I preferred.  The improvement in top-end resolution with the LPS is very good and the 'twangy' things in particular gained flavour and detail. 

With both the change to the Silverstone card and the LPS powering it seems as though a couple of 'veils' have been lifted which indicates to me that the Silverstone card is less noisy than the PPA card.  What features make it less noisy?  Well I don't know for sure.  It could be that the clock is closer to the speed of the usb clock in the NOS1 (correction calls), or that more care has been taken in the layout of the PCB itself or that less noisy components have been used or all of the above.  It could also be that as Coen has deducted with his V2 PPA card, that my V1 PPA card has grounding issues and that improvement is there for the taking.

Coen, I encourage you to post some photos of the mods to your V2 card so that I might apply them here as well and see if I can corroborate your findings.

Looking at the two usb cards, the layout of components is radically different but the NEC USB controller chips are from the same family (one supports two ports, the other four).  I can't help but think that the Silverstone is simply laid out in a more logical manner and this has something to do with what I hear.

Anthony

[PeterSt]

Hey Anthony,

You just managed to describe quite exactly what I perceive from the Silverstone, so I guess it is a general thing ?

What I noticed is that the Silverstone doesn't contain a single electrolytic (this was at some first glance, so maybe one is hidden somewhere). And 3 times less components (compared to that more general one I have and upon which' layout all seem to be based (but not the Silverstone)).

What I also didn't investigate further - but seem to have noticed with a 5 second look, is that there's no normal 24Mhz crystal on it. Could even be an active one (did I see 4 legs ? I think I did).

Peter