Comment by pjc50 | Hacker News Reader

Comment by pjc50 | Hacker News Reader

Ah yes, EMC. If you've ever opened something up and seen lots of metal fingers round the edge of the case, or those flat metal cans on the PCB, that's because it wouldn't pass EMC otherwise. The ferrite rings are another common trick and you can sometimes see them on power or USB cables.

Multilayer PCBs also help a lot; it's not mentioned in this article but putting the 100MHz clock on an interior layer with ground planes on the exposed layers would be a good idea.

You do have to design EMC defence in from the start, rather like defensive programming. Sometimes people take the other extreme and cover v1 of the board in extra capacitors, ferrites, TVS diodes etc, then see how many they can remove while still passing EMC. This may explain unpopulated pads you see in consumer equipment.

One of the reasons cheap Chinese Alibaba gear is so cheap is ignoring these requirements. I have a car USB charger that has so much conducted EMI that you can't listen to the radio with it plugged in.

In many high priced low-volume systems you will find another approach: go to the extreme with EMI supression, pass testing and just leave it as is in the final product. Perfect examples are 90's DEC and IBM workstations which are full of completely redundant ferrite beads (I've even seen snap on ferrite bead on cable for power LED).

Obvious problem with that approach is cost, somewhat non-obvious one is that there can be cases when unnecessary EMI supression/shielding/whatnot can be actually detrimental. Part of EMI susceptibility issues on one customer's system were resolved by replacing 20cm run of shielded twisted pair with unshielded one, shielding formed parasitic capacitor with the overall chassis of the product which coupled noise from chassis ground into analog ground.

[Edit: also the fact that your product passes EMC testing says nothing about the EMC issues it will have in field]

Well, yeah: the cable shielding should have been bonded to the chassis at both ends.

I thought you should only ground shielded cables on one end to avoid forming a ground loop.

I'm pretty sure you're right...

Summarizing a number of application notes into one sentence: Generally speaking yes, except when you don't.

It actually depends on the nature of your ground at both ends of the cable. The book by Ott, pictured but not referenced in the article, had a whole chapter on the subject if I remember correctly.

Also copper tape is wonderful but it will also cut the hell out of you so wear gloves.

As said in other comments in depends on the nature of the ground. In this case it should have been bonded to chassis ground and not to the ground plane of analog board, which wasn't feasible due to mechanical constraints.

In general cable shielding should be connected to shielding ground of devices on both ends of the cable as long as the devices have purpose-designed shielding enclosure, which they often don't and in that case it should be connected at the end with lower impedance ground. Another problem are cables where the shielding is also used as return path for normal operation (RF coaxial cables, consumer electronics...) and in that case it obviously needs to be connected at both ends.

My rule of thumb for this is that shielding should be connected only on the lower-impedance ground end except when it is not obvious which side that is or the shield is also return path. Neat solution to the "non-obvious side problem" is used for IEEE1355/SpaceWire and related interfaces: outer shield of S/STP cable is connected at both ends while the per-pair shields are connected at transmitting side for given pair (used cable has isolation between shields).

Very interesting, thanks!

It's funny, part of what drew me to electronics and circuitry as a hobby was how predictable it is, compared to how sloppy chemistry and biology are. But when you get into high frequencies and EMI and wireless signals...woof.

It's really amazing how much work goes into making that sort of thing work like, at all. I have enough trouble getting a simple bit-banged serial or parallel interface to work without having to worry about occult ley lines.

Reply on news.ycombinator.com