Author Topic: Tying USB shield to ground (VBUS 0V), always a complete fail?  (Read 6577 times)

0 Members and 1 Guest are viewing this topic.

Offline AlfBaz

  • Super Contributor
  • ***
  • Posts: 2001
  • Country: au
Tying USB shield to ground (VBUS 0V), always a complete fail?
« on: November 10, 2014, 11:47:34 am »
Considering the extravagance of a previous design I did,



I recently stuffed up and connected the shield to the 0V ground plane and the device failed to enumerate. It came up as "unknown" on windows
Got the pencil grinder out, isolated the shield from the ground plane and all was good.

I get the impression a lot of people make this mistake but the devices still work. Anybody have experiences to back this up or debunk it?


 

Offline Dago

  • Frequent Contributor
  • **
  • Posts: 657
  • Country: fi
    • Electronics blog about whatever I happen to build!
Re: Tying USB shield to ground (VBUS 0V), always a complete fail?
« Reply #1 on: November 10, 2014, 05:23:18 pm »
I've always just connected it to ground (and so has most of the designs I've seen) and it has always worked fine so there has to be something else going on.

Why do you have a capacitors on your data lines? Makes no sense.
Come and check my projects at http://www.dgkelectronics.com ! I also tweet as https://twitter.com/DGKelectronics
 

Offline AlfBaz

  • Super Contributor
  • ***
  • Posts: 2001
  • Country: au
Re: Tying USB shield to ground (VBUS 0V), always a complete fail?
« Reply #2 on: November 10, 2014, 06:57:52 pm »
I've always just connected it to ground (and so has most of the designs I've seen) and it has always worked fine...
As I suspected, even though it's not recommended by most manufacturers data sheets and I think it's even mentioned somewhere in the USB specs, especially for devices
Quote
...so there has to be something else going on.
That's what I'm thinking. I've routed D+ and D- together (aka differentially) with ground plane either side on the top layer and uninterrupted ground plane on bottom layer. Both ground planes are 0V and the route length is nearly an inch and a half (approx. 37mm)

Quote
Why do you have a capacitors on your data lines? Makes no sense.
Probably shouldn't have posted that pic. It's for a design I did while ago, the caps are to reduce the rise times. On the one I have now there's none of that as shown here.
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 13001
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Tying USB shield to ground (VBUS 0V), always a complete fail?
« Reply #3 on: November 10, 2014, 07:36:26 pm »
What else is your thingy connected to?

If you open the ground, or anything less than a solid RF ground, you're guaranteed to have problems, maybe not right away, but it's not going to pass EMC, and that means vulnerability to ESD (random failures and getting pissed off at later?) and EMI (some trucker blasting his CB drives by and, oops, there goes your USB link).

If something else it's connected to is causing problems, address that.  Shields are always better than nothing: consider

By the way, USB is hardly differential.  High speed is, but none of the other modes are.  Full Speed is just complementary HC CMOS, and the line status signals are entirely unbalanced -- trying to treat them as balanced will result in way more problems than whatever you started with!  That said, it's perfectly okay to route the lines differentially, as long as you make sure they're still okay single ended (like, no routing over un-ground planed areas).

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline AlfBaz

  • Super Contributor
  • ***
  • Posts: 2001
  • Country: au
Re: Tying USB shield to ground (VBUS 0V), always a complete fail?
« Reply #4 on: November 11, 2014, 07:48:45 pm »
What else is your thingy connected to?
When I was testing it, nothing. When it's in use, basically nothing. The USB/232 chip joins up to a digital isolator and that's it.
I have a plane on top and bottom layers, fairly well stitched together and connected to the 0V/ground wire of the VBUS power pair and the connector shield directly connected to it as well.

As I said previously, once I got the pencil grinder out and Isolated the connector shield from the 0V it came good. I've had the unit running for 24 hours and it hasn't missed a beat.

Quote
If you open the ground, or anything less than a solid RF ground, you're guaranteed to have problems, maybe not right away, but it's not going to pass EMC, and that means vulnerability to ESD (random failures and getting pissed off at later?) and EMI (some trucker blasting his CB drives by and, oops, there goes your USB link).
The device has no means of providing a solid RF ground other than via the shield which seems to be chassis grounded back at the PC. I'm measuring 40 ohms from the shield to the earth on a power outlet. Do we really want to connect our 0V return path to the shield?

I have had a quick look around the internet and there seems to be contradictory advice from IC manufacturers as was found by this chap here http://forum.allaboutcircuits.com/threads/usb-device-cable-shield-connection-grounding-it-or-not.58811/

Perhaps I should attempt to scope the data lines with shield in both connected and disconnected states and then try some of the suggestions such as connection via 1M resistor and additionally with a paralleled small cap in an effort to see what, if any, are the effects
« Last Edit: November 11, 2014, 07:50:28 pm by AlfBaz »
 

Offline Precipice

  • Frequent Contributor
  • **
  • Posts: 403
  • Country: gb
Re: Tying USB shield to ground (VBUS 0V), always a complete fail?
« Reply #5 on: November 11, 2014, 10:23:23 pm »
The device has no means of providing a solid RF ground other than via the shield which seems to be chassis grounded back at the PC. I'm measuring 40 ohms from the shield to the earth on a power outlet. Do we really want to connect our 0V return path to the shield?

This doesn't sound good.
The USB socket should be hard grounded (and EMC-continuous) with the PC case. If the PC case is 40 Ohms to Earth, I'd suggest you stop working on the USB widget and fix that!

(In my world, USB ground and shield are usually seen tied hard together at each end. Sometimes there's a token ferrite, but it's never been seen to be useful.)
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 13001
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Tying USB shield to ground (VBUS 0V), always a complete fail?
« Reply #6 on: November 12, 2014, 02:58:19 pm »
The device has no means of providing a solid RF ground other than via the shield which seems to be chassis grounded back at the PC. I'm measuring 40 ohms from the shield to the earth on a power outlet. Do we really want to connect our 0V return path to the shield?

Yes!  That's how ESD finds ground, for example.

Quote
I have had a quick look around the internet and there seems to be contradictory advice from IC manufacturers as was found by this chap here http://forum.allaboutcircuits.com/threads/usb-device-cable-shield-connection-grounding-it-or-not.58811/

The rampant misunderstanding is easy to see, for example:

Quote
Contrary to my initial supposition, the purpose of the USB cable shield
is not to protect the USB data lines from outside interference, but
rather to prevent the USB device from radiating EMI.

Both are necessary purposes, and both occur simultaneously!  To deny this truth is to deny reciprocity.  While there are circuits that do not exhibit reciprocity, a cable isn't anywhere near smart enough to do this!

All the (2) examples suggest RF grounding anyway, so this reason for example is preposterous,

Quote
To limit the USB cable antenna effect, it is
recommended to connect the shield and ground through an RC
filter.

It's freaking RF grounded anyway, what are you even trying to avoid?  Probably yet another example of an appnote written by an intern... or an engineer with about as much EMC knowledge.

The TI appnote is even worse; the broken link is here,
http://www.ti.com/lit/an/spraar7c/spraar7c.pdf
2.2.1: They bring out that old, bizarre practice of stacked capacitor values.

2.2.4: Ground is, well, grounded: says to ground to chassis.  That's fine, so long as chassis ground is plane ground.  If your board ground is noisy against that ground, you have other problems anyway, and you are bad and should feel bad.

The entire statement of this section is bizarrely incomplete and shows the author's lack of knowledge; while he knows signals must be conveyed over signal ground, he has no idea how that should be reconciled against chassis ground, except by playing the schoolyard "Not-Touching" game.

The second paragraph is the real doozy.  This suggests a small ferrite bead, I expect cautiously small so as to avoid the inevitable problem.  This statement:

"To prevent EMI from coupling onto the cable bus power wire (a very large antenna)"

Is all kinds of bizarre!  How can bus power be an antenna if it's wrapped in a freaking shield?!  And how can that one wire, out of all 4-5 entering that shield, be the sole exclusive concern?

2.2.7, last list item: a bizarre suggestion, as with absolute minimum capacitive loading (as stated in other items), the common mode choke has absolutely nothing to work against.  It's also a bad idea anyway, because the start/end packet signals are not differential anyway, and attempting to filter the data lines as a pair only results in trouble.

The intermediate stuff is okay.

Finally, the ESD advice is strangely the best.  3.4 is general ESD advice, with the difference that the second item (GND isolation) is dangerous, and for non-isolated data signals, fatal.  For example, Ethernet is designed this way: the isolation barrier is transformer coupled, with a 1nF cap to ground to provide a path for ESD (note that 8kV on 150pF discharges down to 1kV on 1nF, the rating of the Ethernet isolation barrier).

Quote
Perhaps I should attempt to scope the data lines with shield in both connected and disconnected states and then try some of the suggestions such as connection via 1M resistor and additionally with a paralleled small cap in an effort to see what, if any, are the effects

This is dangerous, as you won't observe much difference without an external source of EMI to introduce noise on your signals.  Such measurements will lead to exactly the faulty conclusions above! ;)

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf