EMI, SNR and communications: when the cable is a poor antenna
Every copper pair is an antenna. Category ratings assume a quiet test. Plants put variable-frequency drives, welders, and poorly bonded shields in the same tray as Ethernet. Signal-to-noise ratio (SNR) is the margin between the wanted symbol and everything else. When SNR collapses, you see CRC, FCS, unexplained retrains, and fieldbuses that “need a new card.” Noise / SNR / EMI is the name of that service; it sits between active diagnostics and grounding.
How the investigation is actually run
Stand in the tray with operations during a drive start, a weld, or a compressor across-the-line start. Eyes on the error counter, clamp on the shield, meter between cabinet earth and the far-end earth. That fifteen minutes is more scientific than a week of protocol traces taken in a quiet office. Photograph gland plates and whether the shield is landed on a painted surface.
If the media is fibre and the endpoints still reset, stop talking about antennas and start talking about power quality at the cabinet. If the media is copper and the errors die when a drive is in bypass, you have a working hypothesis even before a spectrum analyser arrives. Segregation, bonding, specified motor cable, and a media change are four different purchases; pick after the path is named.
Write the coupling path in one sentence a millwright can test. If you cannot, you are not done. Ferrites scattered on day one are not a method.
Symptoms
Errors that track motor speed or welder duty. Analog 4–20 mA that hops when a drive starts. Serial links that work on a laptop on the desk and fail in the cabinet. Wireless blamed; the AP is fine; the backhaul copper is not. See also intermittent network faults.
Causes
Common-impedance coupling through a shared “ground” that is actually a noisy steel structure. Shield drained at the wrong end, or at both ends when the building has large 60 Hz potential difference — circulating current on the shield. Unshielded cable in a tray with unfiltered motor leads. Gaps in bonding at cabinet entries. High dv/dt from modern drives without a specified cable and gland.
Investigation
Correlate error counters with electrical events (current, not folklore). Measure voltage between “grounds” at both ends of the run. Inspect shield termination. A clamp on the shield during a drive start is often more honest than another protocol analyser. If the path is optical and still fails, the problem is not this EMI story — look at power of the endpoints and at software. Cross-read when the network problem is electrical and grounding investigation.
Conclusions name the coupling path. Mitigation is bonding, segregation, specified cable, filtering, or a media change — after the path is named, not as a kit of ferrites on day one.