Connector shield connection for ESD

[ricko_uk]

Hi,
I always connected connectors metal shields to ground but every 20-30 ESD pulses (even if spaced apart) to the metal shield it fails and the system resets.

1) Could it be that, despite not discharging to the pins directly, the pulse does couple into the data lines? I do have TVS with much higher voltage ratings than I am discharging with the ESD gun, but they are unidirectional. So I was wondering whether unidirectional TVS are weaker when the pulse is applied in the reverse polarity?

2) Or perhaps just change the TVS to a larger package?

3) Regarding the connectors metal shields, is there a better way to connect the metal shield than directly to the GND plane? Perhaps leave them disconnected and remove the ground plane and traces nearby to avoid the pulse coupling into them?

Thank you

[T3sl4co1l]

Impossible to tell without layout (including photos of the assembly as tested).  It can be entirely unrelated to the connector, just fields washing over the board.  It could be wires above the board.

The purpose of ground plane is to carry the ESD pulse around the board; in the absence of any metal enclosure, it becomes the enclosure.  Traces and components are exposed as if through open apertures on a box, so it's imperfect, but it's quite passable for commercial purposes.  Mind that includes class C (*maybe* B or A) ESD compliance, so if class B isn't good enough for the application, you may simply need better shielding.  The connector shell should be close to the board edge, well grounded into the plane (multiple vias across/around the footprint), and signals should exit the connector going inboard, keeping some distance between connector and nearby components.  (Distances can be relaxed in certain directions / for certain component types, when the fields are understood, but as a basic hand-wave, more is better.  More being relative to the dimensions of the connector, particularly height and width; "enough" is sufficient, no need to make it obnoxiously large, it's not an "infinity is better" sort of thing.  Hm, I think, anyway.)

Tim

[MarkS]

There was a video I watched recently of an interview with Rick Hartley (I think... Maybe Eric Bogatin?) in which he stated that only in very extreme circumstances does he ever tie the shield to ground. If I remember correctly, it tends to turn the whole cable into an antenna and the enclosure into a resonator. I may be a little off on that, but he was adamant about NOT tying the shield to ground.

[ricko_uk]

Thank you Tim and Mark

[T3sl4co1l]

There was a video I watched recently of an interview with Rick Hartley (I think... Maybe Eric Bogatin?) in which he stated that only in very extreme circumstances does he ever tie the shield to ground. If I remember correctly, it tends to turn the whole cable into an antenna and the enclosure into a resonator. I may be a little off on that, but he was adamant about NOT tying the shield to ground.

Tying PCB plane and enclosure grounds can be problematic if you have connectors elsewhere.  If you do, preferably they should all be placed on a common face (connectors all on one edge of the board, that edge ties into the enclosure with metal shell connectors and/or screw terminals), and then there are no loops crossing through the cavity mode fields anyway, though they may still be worth damping say with absorbent material, if you have such frequencies present in your design (if you have some ~GHz data pairs, harmonics emitted from traces could conceivably excite cavity modes, and a high-Q cavity (all metal and air, no damping) could develop significant field strength from relatively meager sources as this).

Note that cavity modes are irrelevant at most any frequency you would use ferrite for; this isn't a cable FB sort of problem.  And that the lowest such mode will be the PCB oscillating like a vibrating reed, sprung from the one side being common-grounded.  The frequency of which can be maximized (along with CMRR at the connectors) by minimizing inductance of the joint: EMI gasket could even be used to make an EM "seal", without the gaps that many metallic connectors incur (or the ~point contacts that are screws).

Alternately, the joint can be made intentionally "poor" (like a lossy spring, a rubber bumper), which has the downside of degrading CMRR between connectors, but it may be that that is acceptable, or it can be done in such a way that CMRR is maintained (like ferrite beads around each connector's group of wires).  Note that the PCB probably still ends up (galvanically) grounded to the enclosure in this case, it's just that that ground is made through ferrite beads, thus dampening resonances that involve the PCB.

Cavity modes seem a rather unlikely concern for common (non high speed) stuff, but I don't know the context of that video.  Context is key, and perhaps this was exactly such a case.

Incidentally, say there are connectors on both edges of a board, placed inside a rectangular enclosure; if the board is grounded to both sides, then it divides the cavity in two, doubling the minimum resonant frequency in the height direction, making it less likely to be excited (again, give or take whatever spectrum the PCB emits).  You still have the same length- and width-wise cavity frequencies (and now they're lower impedance, from the lower height), and you have more degrees of freedom as the board separates fields into top and bottom modes.  But that just means putting absorbent material above and below the PCB.  Or maybe just at the ends, really.  (If the ends are open around the board edges, then the top/bottom modes couple together as a waveguide racetrack.  One damper in the end might even do it, but you'll get lower Q overall if done on top/bottom, or both ends, or whatever.  The point is the fields should strongly couple along this "racetrack", so although there are many degrees of freedom, a loss element in one place will dampen them all.  In the 1-D case, consider a transmission line stub (open and/or short at both ends): it has many degrees of freedom (resonant frequencies), but all can be dampened with one or more terminations.)

At lower frequencies, even with connections wired across the open cavity, it's fine to have the PCB firmly grounded to one side, as long as those connections aren't noisy, and can be filtered.

I imagine there is context missing, because I think those two more-or-less know what they're doing.  Or else they really aren't; say by a string of good luck successes, they think they're better at this than they actually are.

Conversely, it's disingenuous to report only very limited facts as the totality of some circumstance, when many more facts are pertinent (as seen above).  Given the two-line report above, I would be inclined to believe this option.  Or, put another way: given the range of scenarios I've brought up, the two-line report seems so oversimplified (separated from its context) as to be useless, I'm afraid.

The main thing with USB is it can't be board-level filtered, because no one makes a twinax CMC (i.e. D+/D- inside GND and all three together goes around an FB).  You can filter the data pair as a tradeoff with SE0 signal quality, but that's at best just a few dB improvement at HF to UHF.  But you can put a FB around a jumper cable between an onboard connector and bulkhead union, and that's also a quite viable strategy.  That's the lossy-joint scenario above, but with more freedom to place the connectors on-board and on-enclosure.

Tim