RF shielding for EV power electronics - coupling paths?

EV inverters blast broadband noise that couples into everything nearby. What coupling path surprised you most - radiated, conducted through the harness, or through shared grounds?

EV inverter noise rarely stays in one coupling bucket; the path that surprises people is often the one they did not model.

Common field patterns:
1. Conducted on HV / LV harnesses: broadband edges couple into sensor and CAN / Ethernet returns if filters and common-mode chokes are optimistic. A "radiated" complaint often starts as a harness antenna.
2. Shared grounds / chassis: inverter return currents modulate the reference for adjacent ECUs; star vs mesh and bond inductance matter as much as shielding cans.
3. Near-field magnetic / electric from busbars and cables: thin shields help E-field; low-impedance loops and twist help H-field. Mixing them up wastes a revision.

A useful triage: unplug / reroute one harness at a time while watching the victim spectrum, then check bond resistance and filter insertion under real current. Share band of the victim (AM / FM / GNSS / cellular) and whether the noise tracked RPM or switching frequency - that usually names the first fix.

— rfforum 编辑部 / Editorial Team

The path that surprised me most on EV work was not the radiated field from the inverter itself, but the common-mode current on the high-voltage cable acting as an antenna. The inverter switching edge is fast, the HV cable is long, and if the shield is terminated at only one end, or through a pigtail with significant inductance, the shield current has nowhere clean to return — so it flows on the outside of the shield and radiates, or couples into the low-voltage harness running alongside it. A shield that measures well at DC can be almost useless at the switching harmonics if its termination is poor.

That leads to a practical triage order. First, check the HV cable shield termination: a 360-degree termination at both ends, with low-impedance bonds to the enclosure, is worth more than a thicker shield can. Second, use a current probe on the HV cable, the LV harness and the chassis bond while the inverter is running — the cable carrying the largest common-mode current at the victim frequency is usually the radiating element, and that measurement is much more decisive than a near-field probe alone. Third, look at the busbar and DC-link loop area inside the inverter: a large loop is a magnetic source that a thin enclosure will not stop, so minimising loop area and adding a low-impedance return path often beats adding shielding material.

Also check whether the noise tracks switching frequency or RPM, and which victim band is affected. If the victim is a GNSS or cellular receiver, even a small conducted path through a shared ground can dominate, because the receiver is looking for a signal far below the inverter noise. In that case, filtering the LV harness at the enclosure boundary and separating the HV and LV return paths usually gives more improvement than another layer of shielding on the inverter can.

Thanks for sharing, very thorough. We learned this lesson the hard way too.