Further Isolation of the Silverstone PCIe - USB Card

[CoenP]

Anthony,

If you unplug the NOS from the USB 3 card, what is the voltage difference between the card (screw/parker of the bracket) and the PC's housing?

I would guess it is about 0V. Somewhere the circuit must ground itself on one or more of the active pins.
The differential inputs on the PCI-e USB chip are not galvanically decoupled from the motherboards PCI-e lines by x-formers or opto couplers. That means that the USB chip does not really allow for a 'large' common potential (like floating found) ; they must be just two single ended inputs (and feed lines).
Btw in this setup you still retain all the noise suppressing and improved transfer properties of balanced transmission lines.

Anyway I think the effort of reducing the number of possible ground return path is worth the pursuit. But this setup feels one step too far, it could potentially blow up your south bridge.

regards, Coen

[acg]

Hi Coen,

As you suggested, I just measured the voltage difference (between one of the faceplate screws on the Silverstone card and the PC chassis) a couple of times now and measured about 240mV and 300mV.

I have a couple of further mods to make to the arrangement and I will let you know how that goes and what I measure when they are in place.

Regards,

Anthony

[acg]

What an eventful day!

First off, I further refined my PCIe riser cable to the bare minimum.  All that I have left are the two sense pins (PRSNT1 & 2), the “PWR Good” pin and the three pairs of differential signal pins.  I tried to link “PWR Good” to the lab LPS (it is a 3.3V stable signal that currently comes from the mobo) but I have not been successful, and I think that it has something to do with power up timing.

So, I got this working, pulled it out, did Peters mod about unhooking the USB earth wire in the NOS1 and listened for a while with the Silverstone sans faceplate.  This is a nice sound, I like it, but I sat there with a grin on my face because only this morning I had been listening to my PCIe riser “loom”.  So I shut everything down, put my loom back in, switch everything back on…nothing…not reading the NOS1.  I tried all sorts of tricks to get it to work but in the end I had to reconnect the USB earth wire in the NOS1.  Turned everything back on and bingo…everything worked…music…and such beautiful music. 

Guys, this sound is FAR ahead of the anything that I have heard from the NOS1, the resolution is astounding… saxophones rasp, trumpets braap, the piano is so real, drums and cymbals are to die for and I can easily pick things that were not so easy to hear before such as when the drummer has his foot on the top-hat compared to when he just lets it ring.  With all this micro detail the nuance and musical cues jump out and it is a very satisfying listen.  The sound without my loom and with the isolated NOS1 and Silverstone by comparison is slower, thicker, much less resolving and much less enjoyable, but still great as far as the dacs that I have heard go.

So, I think that I have gone as far as I can with my loom.  In the next few weeks I should have some really high grade LPS (from the linear ATX project) that are super low-noise and wide bandwidth.  I will keep playing music through my loom until then just to see that it is stable.

There are two things that might be indicators of what is going on ground wise:  I could not get the “PWR Good” to work at all when it was connected to the lab LPS;  I could not get XXHE to recognise the NOS1 when the loom was in place and the USB earth of the NOS1 was disconnected. 

The ideal way forward (after others verify my results) would be to design a USB card from scratch which:

1.   Severs ground and power from the PCIe socket, probably more elegantly than this loom achieves it.

2.   Accepts high quality power supplies, maybe even batteries.  Card consumption is about 0.4W so batteries should last for ages.

3.   Uses a high grade clock, possibly with its own power supply


I do not have the expertise to design such a PCB although I know how everything needs to join together.  As an alternative, it would be possible to build an external box to hold the Silverstone card and the power supplies and make up the loom from the PCIe socket to the new box.

Anyway, I would love it if someone else gave this a try.  I will provide instructions in my next post.

Cheers,

Anthony

[acg]

INSTRUCTIONS TO BUILD A PCIe LOOM

To complete this project you will need a PCIe riser cable (like this one) that you don’t mind taking a knife to, and a dual rail lab LPS (or a dual battery supply).

Have a look at the PCIe pinout diagram below…see those orange pins?...they are the only ones that will hold a wire when we are finished.  “B” side of the Silverstone card holds the components, and “A” side has solder everywhere.
 
STEP 1:  Test to see if your new riser cable is functional.  Plug the riser cable between the PCIe slot and the Silverstone card.  Turn on your computer and play some music.

STEP 2:  The cable is functional in its standard form so now carefully take note of which is the A side and which is the B side.  This is really important because you don’t want to be pulling off the wrong wires.

STEP 3:  Isolate and cut all three of the 3.3V (B8, A9, A10) wires on the riser cable.  Strip and twitch these wires together (the cut wires that go to the Silverstone card, not the cut wires that go to the PCIe socket) pull out your soldering iron and join those three wires to a new wire that will become the 3.3V positive wire to the LPS.  Do the same to the three ground wires on the front section (B4, B7, A4) and join them to a new wire that will run to the 3.3V negative terminal of the LPS.

STEP 4:  You are now left with the lower half of the cut 3.3V and ground wires that are attached to the small pcb that fits into the PCIe socket.  These must now be removed so they don’t flop around and cause a short somewhere in the computer.  I was able to carefully pull these wires down so that they pulled out of their solder on the lower pcb…you should be just as lucky.

STEP 5:  See all of those other wires that are still connected to pins that are not orange in the pinout diagram?  Well cut them all and pull them off both the bottom and top PCB’s of the riser cable (don't pull the 3.3V and ground wires that you just soldered together from the top PCB):  they are not needed any more.  You will have to use a sharp knife to start a cut between the wires in the ribbon and then simply pull them apart with your fingers.  Do NOT separate the wires of each differential pair:  each wire of each pair is best left joined to its mate so that noise cancellation is improved.

STEP 6:  You should now have a riser cable that has been reduced to just nine wires including three differential data pairs that are joined.  Turn off your pc at the wall (or on the switch on the power supply that completely turns off the computer), put in the riser cable, plug in the Silverstone card to the riser cable, place something between the Silverstone and the pc chassis (under no circumstances is the Silverstone to touch the chassis while it is “live”) plug the 3.3V wires from the loom into one rail of the LPS that is set for 3.3V or 3.4V.  Plug a power cable into the molex plug of the Silverstone card from the LPS that is set to supply 5V (do NOT use 5V from the computer power supply).

STEP 7:  You now have the riser and Silverstone card installed along with two sets of wires to your lab LPS that is set for 3.3V and 5V.  Turn on the lab LPS.  Turn on the back switch of your XXHE computer.  Press the start button on your computer and it should boot up as normal.  Start XXHE, press play and listen to music.

STEP 8:  Let me know what you hear!!


Cheers,

Anthony
 

[CoenP]

Hi Anthony,

Thanks for the elaborate instruction.

In the diagrams I see that the data receive is not dc coupled, but the data transmit and bus clock are. Likely the are decoupled at the end of the bus.

So we can conclude that the NOS1 case connection is crucial for the ground reference. Do you have the PE connected to power earth for both the pc and NOS1? Or maybe the amps PE?

It seems like a favourable grounding scheme to reference ground at the end of the chain.

Regards, Coen

[acg]

Hi Coen,

Yes, I think that the NOS1 case connection is crucial for ground reference with this setup.   

Here is an excerpt from the PCI Express Card Electromechanical Specification...

AC Coupling Capacitors
The PCI Express add-in card and system board shall incorporate AC coupling capacitors on
the transmitter differential pair.  This is to ensure blocking of the DC path between the PCI
Express add-in card and the system board.  The suggested minimum 603-type (or smaller
402-type) capacitor with a value as specified in the PCI Express Base Specification.  Any
additional attenuation or jitter caused by the coupling capacitors (other than 603-type) must
be accounted for as part of the allocation for the physical component on which the
capacitors are mounted.  The electrical budgets allocated for the AC coupling capacitors are
defined in the following subsections.  The allocated budget includes the electrical parasitic
associated with the component placement as on the PCB. 

So decoupling is a requirement of the PCIe standard.

I use 3 prong power cords all components (the lab LPS, the XXHE PC, the NOS1, the amps) and everything goes into the same power socket which is a dedicated line for audio.

Regards,

Anthony

[PeterSt]

The suggested minimum 603-type (or smaller
402-type) capacitor with a value as specified in the PCI Express Base Specification.  Any
additional attenuation or jitter caused by the coupling capacitors (other than 603-type) must
be accounted for as part of the allocation for the physical component on which the
capacitors are mounted.

And didn't I say that THE difference with the other cards is the lack of electrolythes ?
Don't ask me what those are for on the other cards, but it would be my exact point of attention in this realm, and the smaller the better (like 402 being better than 603 for such an application). Eh .. aha ?

[PeterSt]

Anthony,

General remark : I think most of us will be using the NOS1 without PE. This is just because I advise it *if* the subject is in order in the first place. But I often mention(ed) it in the forum as well.
Notice that this in itself is totally UNrelated to that black wire being mounted or not (no matter you'd think a 100 times "so, what is it for then without connection to PE in the first place").

Regards,
Peter

[PeterSt]

Another general remark : Super !

[PeterSt]

I tried all sorts of tricks to get it to work but in the end I had to reconnect the USB earth wire in the NOS1.

Which makes me think that you're explicitly using PE for signal ground in this situation. But when the whole story is taken into account, it also works the other way around : when all grounds are connected as should at the PC end, now USB signal ground goes over PE. Or at least it can (whether really so depends on many things).

Question (and maybe I should read back all the details) :
Anthony, do you now have an understanding of what ground you are really using ? So, still it has to come from somewhere (and you've proven it's not the USB cable).
Any one-liner for this ?

Thanks,
Peter

[CoenP]

So, this all makes sense after all.

When Anthony isolates (or disconnects) the NOS1 PE (at the power cable/connector end) it does not work anymore. And be careful with static!

regards, Coen

[PeterSt]

Yo Anthony,

So I shut everything down, put my loom back in, switch everything back on…nothing…not reading the NOS1.  I tried all sorts of tricks to get it to work but in the end I had to reconnect the USB earth wire in the NOS1.  Turned everything back on and bingo…everything worked…music…and such beautiful music.

You'll remember my connection which doesn't want to work without clear reason ? You may not want to believe this (I dare not te believe it myself at this moment), but it can well be that you just handed me the reason ...

That connection too is isolated (think no bracket);
Skipping a few things I don't know (or didn't sort out back then - like what pins exactly provide which), in MY situation the black wire in the NOS1 was there all right, but PE is not provided (using a 2 prong, say). Additionally, MY external PSU provision (part of the ) does NOT provide PE hence nothing connects to PE anywhere and now I'm not sure. It's almost like PE will provide a ground path to the PCIe bus or something. Well, that will not be correct but what the heck is happening really. So, I'll be damned if my connection works after all if I only connect a few devices to PE (currently the PC is the only device doing that).

Can someone hand me an idea how additional (signal) ground over PE would be able to provide more current ? I mean, outside of what I said above, I have the idea that when something like what happened to Anythony (won't see the NOS1) this is about not enough current provided. So, way more current is needed to load the binary to the USB interface (FPGA in there). And if that current is not there sufficiently, just nothing will happen. Now I envision Anthony's situation (but could be mine as well) as something like a signal (plus) wire of certain gauge, ground is actually lacking but goes through tiny leaks up to through air or whatever illegal - that will let USB work in itself (card is recognized) but as soon as more current is needed (plug in NOS1) it's not enough. However, PE will provide enough always and nobody notices a thing about (design !) faults etc. because PE is to be there always.

Was this clear somewhat ?

Peter

[acg]

Anthony,

General remark : I think most of us will be using the NOS1 without PE. This is just because I advise it *if* the subject is in order in the first place. But I often mention(ed) it in the forum as well.
Notice that this in itself is totally UNrelated to that black wire being mounted or not (no matter you'd think a 100 times "so, what is it for then without connection to PE in the first place").

Regards,
Peter

I'm using the 3 prong power cable you provided Peter.  I could find some two prongers to see what I can figure out.

[acg]

I tried all sorts of tricks to get it to work but in the end I had to reconnect the USB earth wire in the NOS1.

Which makes me think that you're explicitly using PE for signal ground in this situation. But when the whole story is taken into account, it also works the other way around : when all grounds are connected as should at the PC end, now USB signal ground goes over PE. Or at least it can (whether really so depends on many things).

Question (and maybe I should read back all the details) :
Anthony, do you now have an understanding of what ground you are really using ? So, still it has to come from somewhere (and you've proven it's not the USB cable).
Any one-liner for this ?

Thanks,
Peter

From what I can figure out, signal ground is going out through the LPS, assuming that the isolation of the Silverstone from the PCIe socket is working as desired.  I wonder what would change if I lifted PE for the LPS by using a two prong power cord?  That is something to try.

Anthony

[acg]

Yo Anthony,

So I shut everything down, put my loom back in, switch everything back on…nothing…not reading the NOS1.  I tried all sorts of tricks to get it to work but in the end I had to reconnect the USB earth wire in the NOS1.  Turned everything back on and bingo…everything worked…music…and such beautiful music.

You'll remember my connection which doesn't want to work without clear reason ? You may not want to believe this (I dare not te believe it myself at this moment), but it can well be that you just handed me the reason ...

That connection too is isolated (think no bracket);
Skipping a few things I don't know (or didn't sort out back then - like what pins exactly provide which), in MY situation the black wire in the NOS1 was there all right, but PE is not provided (using a 2 pong, say). Additionally, MY external PSU provision (part of the ) does NOT provide PE hence nothing connects to PE anywhere and now I'm not sure. It's almost like PE will provide a ground path to the PCIe bus or something. Well, that will not be correct but what the heck is happening really. So, I'll be damned if my connection works after all if I only connect a few devices to PE (currently the PC is the only device doing that).

Can someone hand me an idea how additional (signal) ground over PE would be able to provide more current ? I mean, outside of what I said above, I have the idea that when something like what happened to Anythony (won't see the NOS1) this is about not enough current provided. So, way more current is needed to load the binary to the USB interface (FPGA in there). And if that current is not there sufficiently, just nothing will happen. Now I envision Anthony's situation (but could be mine as well) as something like a signal (plus) wire of certain gauge, ground is actually lacking but goes through tiny leaks up to through air or whatever illegal - that will let USB work in itself (card is recognized) but as soon as more current is needed (plug in NOS1) it's not enough. However, PE will provide enough always and nobody notices a thing about (design !) faults etc. because PE is to be there always.

Was this clear somewhat ?

Peter

I do remember your secret connection Peter and if my experimentation helps you get your thingamy working then that is great.  I must say that I am curious as to what it could be!

Regarding your question about the current for the FPGA when the ground wire was disconnected in the NOS1 and the loom was in place I am not sure that the USB card was visible to the operating system at that time because I did not check.  It may have been.  Nothing for certain I'm sorry, although in hindsight I should have checked.

Anthony