Let's talk about cable shielding and pcb GND layer

[homebrew]

After watching some videos and reading a lot, I do think I understand at least something of that topic but that might also well be a "Dunning Kruger" effect  :-)

The specific question is about cable shielding attachment to PCBs without a conductive enclosure (so plastic or none).
What to do with the shield of say USB, HDMI or ethernet connectors?

Many designs/app-notes/”tutorials” put a combination of R/C/L in whatever configuration to join shield and system ground right next to the connector. However, the more I “learn” about the subject, the more all these approaches seem conceptually  wrong. Especially after watching
(as suggested by janoc over here https://www.eevblog.com/forum/beginners/split-ground-plane-to-do-or-not-to-do/msg3407420/#msg3407420).

Following some arguments in that video and other sources, here are some contradictions I did stumble across when shield is attached to a pcb GND layer:

1) Shielding is an extension of a faraday cage. Ok, but without a conductive enclosure there is no faraday cage anymore anyway. => POINTLESS

2) Connecting the shield directly to the GND plane seems a bad idea, as the GND plane is just another conductor of the system. Hence one cannot simply “dump" unwanted RF / ESD energy into the plane. It has to go somewhere. But we didn't want that energy in our system to begin with!=> COUNTERPRODUCTIVE

3) Connecting shield and ground with a C seem counterproductive. C is low impedance for high frequencies thus we couple RF in and out of our board which is exactly NOT what is supposed to happen (see point 2). => BAD

4) Connecting it with an L must be wrong, too, as we then have a DC-path between shield and GND and if there are currents on the shields, the tiny little L will not be able to handle it. => BAD

5) Putting a "high" value R between shield and GND will dissipate static potential differences. Pointless if GND is referenced to the shield somewhere else. => MOSTLY_POINTLESS

So what to do? Any form of connecting a cable shield to the GND net seems either pointless, bad or even counterproductive.
Or am I missing something?

[bob91343]

The point of a shield is to enclose sensitive or powerful circuits to reduce interference.  So it's important that a shield not conduct any current.  The only way to insure that is to connect it at one point only because that way there is no place for the current to flow.  Since it's a shield, there will be local currents to do that job but no external currents to spoil it.

You don't need an inductor or a capacitor.  A direct connection is best, and it should be what's called 'star' mode, where there is only one ground point.

You can't extend dc theory into rf.

[homebrew]

You can't extend dc theory into rf.

Hm, I think I didn't but maybe you got me wrong ... (still my fault then!)

You don't need an inductor or a capacitor.

But that is apparently exactly what is commonly done. Let's take USB for example to illustrate what I mean.

Here are some totally random examples:

Example for inductor || resistor:
https://www.cypress.com/file/186471/download (figure 4, page 3, red box )

Example of a capacitor || resistor:
http://ww1.microchip.com/downloads/en/Appnotes/doc8388.pdf (figure 3-5, page 8 )

Example of a straight, direct connection:
https://www.silabs.com/documents/public/application-notes/an0046-efm32-usb-hardware-design-guidelines.pdf (Figure 2.1, page 4, connector on the right)

Again, it is NOT the question where to reference system ground to "chassis ground" (if I would have one).
My question was what to do with INCOMMING shield connections (USB, HDMI, Ethernet) if I don't have a conductive enclosure and thus no chassis ground to begin with. All there is, is my ground plane.

I just cannot believe that all shown examples in the linked app-notes and design guidelines are equally 'correct'. It definitively should matter if one uses a L, C or R (or combinations of them).  Especially if multiple devices are connected (e.g. monitor, Ethernet, ...).

[T3sl4co1l]

The PCB is a Faraday cage, just a poor one.

Consider a closed metallic enclosure around a circuit.  This is a Faraday cage, nothing outside gets in and vice versa.

Maybe we have a few holes for signals to pass, maybe shields are extended around them.  In the latter case, the whole system is still a Faraday cage: the cables are merely an extension of the interior.  They're thin and flexible, so what, their insides are still very much topologically inside the shielding.  (In the former case, we prefer to filter and protect such signals, as close to the point of penetration as possible.)

And I shall use a topological argument to explain the thought process here.

Suppose we open a small hole in the enclosure.  A hole is a highpass element: EM of some frequency range can pass (e.g., you can look through it, it passes light), while it has an asymptotic cutoff below.

If we grow this hole, eventually until it covers the entire side, say -- the cutoff frequency gets lower and lower, but we still have fundamentally a shield, at least at DC.  Whether we care about the rest, is just a matter of scale.  Maybe an open side is okay for an audio amplifier, but certainly not for a microwave circuit.

Now, suppose there's a circuit mounted in the middle of this enclosure, dividing it into two spaces, grounding completely around its periphery.  The circuit is flat, and contains a contiguous ground -- your basic plane grounded PCB design.

If we open holes on the top and bottom of this enclosure, we expose the top and bottom of the circuit.  If we expand these holes all the way to the edge of the PCB, the enclosure disappears entirely; and without any side height, surely we've invited quite a bit more noise in, and can no longer apply the aperture cutoff/attenuation idea.  We might also note that the PCB isn't a single large antenna, but is made of myriad smaller nodes, with different shapes and sizes, so act as antennas in their own right, at various random frequencies and asymptotically below there (corresponding to overall size and circuit impedance, since the low-frequency equivalent circuit looks like an impedance divider, with a very small capacitance from free space, dividing into the node impedance).

So the result isn't great for some applications, but it's acceptable for many.  And as regards cables: shield currents still flow along them, and need to find a ground.  The lowest impedance ground we have is the plane ground.  A low impedance connection to that, minimizes the common mode noise injected into the signals.  Simple as that.

So, not only should you use:
- shielded connectors,
- with the shield tied directly to ground (at RF),
- but also tied in multiple places around the periphery of that shield.

Example, a SMT or hybrid USB connector: the shell and/or pins should be bypassed to ground (with multiple caps if galvanic isolation is still desired, say four or so), and this allows noise currents to divert into the PCB ground, away from the signals at the rear of the connector.

There are ways to float a grounded connector, using it differentially; just remember, you can treat low frequencies (DC / galvanic isolation) separately from high; high frequencies must be shunted into ground, around the active circuit; and you may need to add additional circuitry (diff amp, isolation transformer, etc.), which introduces some other compromises as well (CMRR, bandwidth, accuracy, etc.).

Tim

[bob91343]

What to do with the incoming shield connection?  Well for it to be a proper shield, there cannot be current through it.  Otherwise the two ends of the shield will be at different potentials.  The best probably would be triaxial, where there is a shielded wire inside an outer shield so the latter can go to the star ground point while the inner stuff can do whatever necessary to transfer the signal.  Even a conduit or vent pipe can be a shield if treated properly.

This conversation is perhaps inappropriate because the concepts of shielding and signal transfer are not simple enough.  Every signal has a source and a load, and needs to not have other signals use its wires.  In addition there is the need to prevent radiation.

[T3sl4co1l]

What to do with the incoming shield connection?  Well for it to be a proper shield, there cannot be current through it.  Otherwise the two ends of the shield will be at different potentials.

Question -- what difference does it make, that there's current on the shield (the outside), or that the voltages at either end (relative to some imagined ideal ground) differ?

That's the key, only the voltage difference(s) within the shield matter, and that's why it works.

Tim

[bob91343]

If there is current through the shield, the ends will be at different potentials.  If that happens, the output will not reproduce the input and so we get a signal that is different from that intended.

How important this becomes has to do with your purposes and equpment.  Measuring a 1000 Ohm resistor with random test leads is probably okay if you don't care about a few tenths of an Ohm error.  If you do the same with a 1 Ohm resistor your results will be considerably in error.  If you pick up some noise due to poor shielding, your output will be different from that intended and, once again, it depends on how important that is.

So the final answer is, it depends.  Ignore errors and all will be fine as long as those errors are too small to cause problems.

I remember many years ago our lab had contracted to make a power supply with less than 1 mV of ripple.  That turned out to be a can of worms where measurement technique, capacitor ESR and ESL, and grounding methods had to be revisited.

[T3sl4co1l]

I'm not getting it, it does not seem to follow; could you draw a diagram explaining it?

Tim

[Kjelt]

The point of a shield is to enclose sensitive or powerful circuits to reduce interference.  So it's important that a shield not conduct any current.  The only way to insure that is to connect it at one point only because that way there is no place for the current to flow.

I had this discussion with an EMC guy and he said if you connect a shield only one sided to ground it is emc wise useless since you made an antenna which is far from a good shield esp with longer cables and higher currents.

He pointed out that to prevent currents flowing through the shield , the chassis on both sides should have equal potential, ergo no current can flow. This is achieved by hard grounding of the chassis preferably with thick litze cable for HF EMC or at least according to the machine guidelines with 10mm² stranded copper wire.

So yes if both chassis have a different potential, that is the first item to solve because something is wrong with the grounding system.

For TS with an ungrounded isolated chassis the not so popular answer will be if the source device is hard grounded and the shield is connected to ground the only device to properly connect it to should also be grounded. If not you have an undefined connection. If the source device is ungrounded and the shield is connected to the power supply minus then you should leave it floating or both power supplies will be connected and interference from outside inducted to that shared connection.

I am not saying that this is the only solution or best solution but it makes the most sense to me personally and if there are EMC professionals I really would like to learn more about this topic.

[SVFeingold]

I had this discussion with an EMC guy and he said if you connect a shield only one sided to ground it is emc wise useless since you made an antenna which is far from a good shield esp with longer cables and higher currents.

This is my understanding as well, and in line with common practice at my day job. Long (relative term) stretches of ungrounded shields create problems for low-level signals as well as high-frequency signals. Nevermind mm-wave.

Even a complete Faraday cage isn't perfect, and can re-radiate signals into the cage.

[The Soulman]

I'm not getting it, it does not seem to follow; could you draw a diagram explaining it?

Tim

I think Bob is thinking about single ended connections. ie. the signal is referenced to ground=shield.

[T3sl4co1l]

I think Bob is thinking about single ended connections. ie. the signal is referenced to ground=shield.

If the problem is ground loop, i.e., low frequency coupling of outer-shield currents to inner (signal) currents, then another way to put that is it's a violation of the shielding assumption -- the shield is supposed to keep signal and shield currents independent.  This is true at frequencies that are low relative to the thickness of the shield.  This can be a good reason to float the shield, in the way I described above (but maybe that part wasn't well read, as I have a tendency to write long posts that don't hold attention..)

Tim

[bob91343]

This is a complex and confusing subject.  We have the metal enclosure that is supposed to prevent radiation.  We have the balanced line that is supposed to carry the signal without concern for 'ground'.  So many other considerations, especially the frequencies of interest, power level, conductivity, thermal effects, and more.

The answer is to build the setup as carefully as you can, keeping in mind any pitfalls you know of, test it and then modify it and test again, and again.

I recall having to redesign a commercial product because there was hum in the output.  The factory had a fix that didn't fix it.  It took an understanding of what was causing the hum and what measures to take to eliminate it.  When done, the hum was not just reduced; it was totally inaudible.  Very much a function of grounding technique.

[homebrew]

This can be a good reason to float the shield, in the way I described above (but maybe that part wasn't well read, as I have a tendency to write long posts that don't hold attention..)

Thank you very much for your elaborate and lengthy post. Especially your topological argument there made perfect sense!
However, the exact thing I still don't completely understand is put forward in your last answer:

If the problem is ground loop, i.e., low frequency coupling of outer-shield currents to inner (signal) currents, then another way to put that is it's a violation of the shielding assumption -- the shield is supposed to keep signal and shield currents independent.  This is true at frequencies that are low relative to the thickness of the shield.

Isn't that EXACTLY what happens if you declare a PCB's ground layer as part of a faraday cage? The GND plane has not 0 inductance and carries (signal) currents. Whether or not you float the shield dc wise, you'll still have AC currents from your GND plane in the shields (and vice versa) if different devices are connected and ground-loops are formed that way.

[T3sl4co1l]

Right, like any other non-superconducting shield, the ground plane will show different current distributions over frequency -- the trick again is to have it thick enough such that, the frequency range where it's susceptible, is also where your low frequency signal path handles the resulting error (typically, differential signalling, where op-amps are used to provide CMRR, which depends on GBW and so degrades at high frequency, where your shielding takes over).

Or to just have enough of it, that the voltage drop across the circuit is negligible.

Obviously, one can use clever routing to avoid major current paths, keeping critical signals away from them.  (This is another application of ground plane slots.)


Note that, analogous to skin effect in bulk materials, there is skin effect in thin materials -- more of an edge effect, really.  That is, currents induced in a finite sheet, preferentially flow around the edge, and the depth from that edge over which they flow, is determined by the material resistivity and permeability, and thickness.

Another way to think of it is, suppose you place a ground plane perpendicular to an AC magnetic field.  It acts as a shorted turn, but where does the current flow in the plane?  If it's thicker than a few skin depths, it's just like any bulk conductor, and flows on the surface (namely, perpendicular to the field lines, around the rim).  If it's thinner, then field will partially penetrate, all the way through the material thickness; but for every layer that passes some field, there is also some current flow, which opposes that field, and so on, all the way around the periphery.  So, current drops off as you go inward from the edge.  The rate is clearly slower than for bulk skin effect, but the same effect (ultimately, an ~exponential decay of current with respect to distance from the edge/surface) is clearly at work.  So, sheet thickness, as well as resistivity and permeability, is a factor in the effect.


So, if our concern is EMI susceptibility, simply pour a solid ground plane around everything, and keep critical signals somewhat in from the edge.

If our concern is low frequency ground loops, a solid ground plane is still a good idea, but ensure good layout so that noise current paths are avoided.

Which is to say in both cases, ensure good layout.

Tim

[bob91343]

Trouble is, a good layout is not an obvious exercise.  One must know about skin effect, ground loops, and a whole lot more in order to create a good layout.

If you are doing this for a living, the boss is going to become impatient with your lack of progress.  And balk at doing experiments to confirm layout ideas.