2
\$\begingroup\$

I have a PCB, that will now be put into a conductive enclosure (connected to PE). The PCB has some power electronics (DC/DC converters, motor controller etc.), some communication interfaces (CAN etc.) and also some analog circuits (DMS). The system is supplied from an external insulated +48VDC supply.

The PCB has a proper GND plane and the signals are impedance matched. There is enough spacial distance between the sentivie analog circuitry and the power/digital parts. The used cables will be shielded and the quesition about where to connect the shield came up.

When I have a look at Henry Ott, he recommends to connect the PCB GND to the chassis GND at exactly one point, close to the connector. I'm not sure if this is the correct approach here.

In the following picture you see the setup.

Variant 1: So one possibility would be to connect the system GND at one point to the enclosure (1) and then connect each shield to the system GND (2). This way, there cannot flow any LF currents, but normally HF anyway dominates and the skin effect let the return currents flow on the surface only. So the return currents flow probably on the inside of the shield back and will not contribute to much EMI. But one problem I see is, that there is a lot of inductance for coupled noise from the chassis to circuit GND and so probably the chassis will be pretty much useless.

Variant 2: Another possibility would be, to connect each shield at the enclosure entry (3), so the shield is like an extended enclosure. This way, there is low inductance between the shields and the enclosure. When then the shield is also connected to GND at the connector entry on the PCB (2), then the noise can flow perfectly back on the surface of the shield and the enclosure and there is still a low impedance path on the inside of the shield for the HF return currents.

Variant 3: Like in Variant 2, but the shields at the enclosure are connected by something like a 1nF cap and 1Mohm resistor, to short HF noise and block LF noise.

So, what shielding concept would you use in this case?

enter image description here

\$\endgroup\$
4
  • \$\begingroup\$ What is the application? I only ask because answer may be application dependent. For high radiation environments such as on the outside of a space craft we've been asked to put a 1 MegaOhm resistor between chassis and ground to prevent discharge. For military applications it's usually a 10 nF capacitor between the two to shunt conductive noise to earth. \$\endgroup\$
    – C. Dunn
    Commented Jul 28, 2023 at 14:45
  • \$\begingroup\$ Hi, it is a motor control application in an industrial environment. The noise shunting with a capacitor and resistor is also what i have seen a lot, especially in ethercat applications, where the shield is connected this way to PE \$\endgroup\$ Commented Jul 28, 2023 at 15:51
  • \$\begingroup\$ Is the PCB made to withstand various noise levels by itself (does it have adequate filtering and protection to operate with a nonmetallic enclosure)? Or shall the enclosure be part of its EMC design? Are there any other constraints, like would metallic connectors be too expensive? What about LF ground loop, would it be a concern to: a. the signals within the cables? b. external attached equipment? \$\endgroup\$ Commented Jul 28, 2023 at 16:47
  • \$\begingroup\$ @C. Dunn - You want the resistor to provide a charge bleed off path to chassis for any metal (which includes cable shields) that may become charged (floating metal) from space particles. \$\endgroup\$
    – SteveSh
    Commented Jul 28, 2023 at 19:10

2 Answers 2

3
\$\begingroup\$

Shielding strategy is very important for EMC IEC 61000-4-3 and 61000-4-6 standards, as they result in RF common-mode currents, in cables, trying to penetrate your system to earth.

IEC 61000-4-3

The main problem is not the reflection of the electromagnetic field, but the common-mode current in the cables. The only way to divert these currents to ground is to connect the cable shields directly to the metal housing of the device.

If you ground the shield on both sides, then common-mode currents cannot enter your device.

enter image description here

But if you only connect the ground on one side, then parasitic capacitances will create a path that will disrupt the signal going into your device.

enter image description here

It is almost impossible to keep the parasitic capacitance value low due to external parameters; humidity, dust, mechanical dilatation, dielectric caracteristic of câble and all environnement parasitic capacitance in proximity.

For your project, you can do the following :

enter image description here

  • Cable shielding is connected to the conductive enclosure to divert all EMC disturbances.
  • Multipoint grounding minimizes capacitive coupling between PCB and enclosure.
  • If you have an analog zone, it must not contain a return path to ground.

All the illustrations are mine

\$\endgroup\$
4
  • \$\begingroup\$ Thank you, great illustrations. One problem i see with connecting the shield like in your last illustration is, that the signals are not shielded on the inside of the enclosure and when the noise source is there, it will directly couple into the senstive signals. Would it not make sense to further guide the shield on the inside to the PCB input connector and connect it there also to GND? You talk about grounding the shield, in your illustration this means PE AND ground connected tightly together? \$\endgroup\$ Commented Jul 31, 2023 at 6:50
  • \$\begingroup\$ In general, it is important to have different strategies for internal EMC and external EMC (IEC 61000). The IEC 61000 standard lead to very strong disturbances from outside, EN 61000-4-4 (EFT) simulates motor starting, EN 61000-4-5 an indirect lightning strike, EN 61000-4-2 electrostatic discharge 4 or 8kV, etc. If you guide the shield on the inside to the PCB input connector and connect it there also to GND but EMC disturbances will be closer to your PCB. I strongly advise you to leave these problems outside the enclosure. \$\endgroup\$
    – Vincent
    Commented Jul 31, 2023 at 8:34
  • \$\begingroup\$ Inside the enclosure, a common solution is place cables against chassis for minimize the couplings (inductive because the loop surface signal/chassis (or ground) is minimum and capacitive with the multipoint grounding) with the internal noise sources. For your last question, in Europe the regulations require PE AND ground connected tightly together for safety reasons, except when no earth is present (aerospace or airborn application for example). \$\endgroup\$
    – Vincent
    Commented Jul 31, 2023 at 8:34
  • 1
    \$\begingroup\$ Put another way: the chassis can be used as an extension of the ground plane. The minor tweak I would make here is to show the PCB-chassis GND connections clustered around the connectors, for this purpose. How close the signal shields must be to the PCB (or indeed right up to it) is determined by permissible crosstalk and data rate / bandwidth; you wouldn't strip shield from a USB-HS or eSATA port, but it won't matter for say RS-232. Overall great answer. \$\endgroup\$ Commented Jul 31, 2023 at 8:55
0
\$\begingroup\$

The used cables will be shielded and the question about where to connect the shield came up.

The main thing is there will be two problems with connecting the shields. One will be ground loops of currents flowing down the shields. The other is cable radiation. The other item that will come up is ESD, and the currents from ESD will go to ground from the shields.

In many cases a design should have the cables shields to be disconnected on one end to avoid ground loops, more often in low noise applications and in high RF environments or industrial environments with large currents\magnetic fields.

The system GND should be low impedance back to the supply to avoid common mode ground noise, especially if you have large load changes.

Variant 1: So one possibility would be to connect the system GND at one point to the enclosure (1) and then connect each shield to the system GND (2). This way, there cannot flow any LF currents, but normally HF anyway dominates and the skin effect let the return currents flow on the surface only. So the return currents flow probably on the inside of the shield back and will not contribute to much EMI. But one problem I see is, that there is a lot of inductance for coupled noise from the chassis to circuit GND and so probably the chassis will be pretty much useless.

This is fine, the

I have had better results passing FCC and safety with a star ground where the shields and earth ground are tied to a chassis and all stared at one point. I have also tied the PCB ground in at this point in some cases. Also any ESD on the cables is shunted to earth ground.

Variant 2: Another possibility would be, to connect each shield at the enclosure entry (3), so the shield is like an extended enclosure. This way, there is low inductance between the shields and the enclosure. When then the shield is also connected to GND at the connector entry on the PCB (2), then the noise can flow perfectly back on the surface of the shield and the enclosure and there is still a low impedance path on the inside of the shield for the HF return currents.

The drawback from connecting shields at the enclosure entry is not shielding the cables from RF that may be inside the enclosure. The enclosure can have standing waves depending on its size or see radiation from the PCB itself. If you have sensitive signals you may want to shield all the way to the PCB, if they are digital this is almost never necessary

Variant 3: Like in Variant 2, but the shields at the enclosure are connected by something like a 1nF cap and 1Mohm resistor, to short HF noise and block LF noise.

This is good to do on one end of the cable for ground loops, the other end should be tied to ground. If your cables are radiators this would be a good option to do because it allows the return currents to go back to the source. Another option is to leave the cable shield open on one end, if it's open on one end it should be grounded/tied to chassis on the other end.

\$\endgroup\$
13
  • 2
    \$\begingroup\$ "In many cases a design should have the cables shields to be disconnected on one end to avoid ground loops" -- disconnected from what? Or in what cases specifically? Because a wholly disconnected ground implies no reference to the signal within and absolute zero noise immunity. There is only one reason why this advice was created (E-field shielding for high impedance tube circuits), and it should be vigorously forgotten today. \$\endgroup\$ Commented Jul 28, 2023 at 16:51
  • \$\begingroup\$ I explain that further in the bottom of the answer, if it's connected on one end then it still provides shielding. But if you want to disconnect a ground loop, you have to open the loop \$\endgroup\$
    – Voltage Spike
    Commented Jul 28, 2023 at 16:54
  • 2
    \$\begingroup\$ If the shield is connected at one end only, then you have an antenna. \$\endgroup\$
    – SteveSh
    Commented Jul 28, 2023 at 16:56
  • 2
    \$\begingroup\$ And I agree with Tim that the "connect shield at one end only" recommendation should be relegated to the trash heap for today's high speed digital systems, except for some very specific applications. \$\endgroup\$
    – SteveSh
    Commented Jul 28, 2023 at 16:58
  • 1
    \$\begingroup\$ If it's open on one end, it provides electrostatic shielding only. An open shield means the core signal(s) is exposed to any magnetic or EM loops; outer/shield currents follow the cut, coupling 100% to the signal, i.e. providing no shielding value. At HF, the shield provides current separation (shield inner = signal return current, shield outer = CM current). At LF, the shield loses effectiveness (skin depth more than shield thickness) and ground loops couple into the signal. There IS a strategy to deal with LF loop while preserving HF shielding, but wholly disconnecting the shield is not it. \$\endgroup\$ Commented Jul 28, 2023 at 16:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.