Grounding and Shielding: Why Your Signal Is Noisy

The symptom arrives as a measurement problem. A 4-20 mA loop reads two per cent high whenever the line runs. An encoder gains counts under acceleration. A Modbus link retries. Somebody replaces the transmitter, the fault stays, and the transmitter gets a reputation it does not deserve. Almost none of this is instrument failure. It is grounding and shielding, and it is one of the few areas of plant engineering where the correct practice is well established, widely documented and routinely ignored.

Start by separating two jobs that share a word. Protective earth exists to carry fault current safely and operate a protective device; it is sized for hundreds of amps and its performance is judged by impedance under fault. A signal reference exists to give a low-level measurement a stable zero. These are different requirements, and the habit of treating any green-and-yellow conductor as interchangeable with a signal common is where most noise problems begin.

The classic failure is the ground loop. Bond a cable shield to earth at both ends and you have connected two points that are not at the same potential — across a large plant the difference can be volts, and during a motor start it is not stable. The shield, being a low-resistance conductor between them, carries the resulting current, and that current couples into the signal it was supposed to protect. The shield has become an antenna's opposite: a conductor injecting noise rather than draining it.

The remedy depends on frequency, which is why the advice sounds contradictory. For low-frequency analogue signals — thermocouples, RTDs, 4-20 mA — bond the shield at one end only, conventionally at the control system end, and leave it insulated at the field end. There is no loop, and at those frequencies a single-ended shield still does its job. For high-frequency and digital signals — industrial Ethernet, high-speed encoders, drive cables — bond both ends, with a full 360-degree termination, because at those frequencies the shield needs to be a continuous coaxial enclosure and a single-ended connection is not one. Where both ends must be bonded across a potential difference, run a parallel equipotential bonding conductor alongside the cable so the fault and circulating currents have a path that is not the shield.

That 360-degree detail matters more than it sounds. A shield terminated by twisting the drain wire into a pigtail and landing it on a terminal is fine at 50 Hz and useless in the megahertz, because a few centimetres of thin wire is an inductor and an inductor is an open circuit to fast transients. Use a proper EMC gland or shield clamp that contacts the braid around its whole circumference. Every high-frequency installation guide says this and most field terminations ignore it.

Twisting is separate from shielding and often more important. Magnetic coupling induces a voltage in the loop area between two conductors; twisting them reverses the loop every twist so the induced voltages cancel. A twisted pair with no shield rejects magnetic interference better than a shielded untwisted pair, which is why the pair geometry is worth protecting — do not untwist more than you must at the termination.

Then there is segregation, which is free at design time and expensive afterwards. Power and signal cables belong in separate trays or conduits with a stated separation distance, and where they must cross, they cross at ninety degrees to minimise coupling. Variable frequency drives deserve individual attention because they are the dominant noise source in most plants: the fast switching edges that make them efficient also radiate, so use shielded motor cable, terminate it 360 degrees at both the drive and the motor, and keep the earth path back to the drive short and direct. A drive installed with an unshielded motor cable will find every sensitive circuit within ten metres.

Where the potential difference cannot be engineered away, buy isolation. Isolated analogue inputs, signal isolators and galvanically isolated communication interfaces break the loop by design and are cheaper than a week of intermittent-fault investigation. This is the honest answer for long runs between buildings, for anything crossing between separately earthed systems, and for the retrofit where you cannot control what the existing installation did.

Before bonding anything, measure. Put a voltmeter between the two ends of the shield before you connect them and see what is actually there. Put a clamp meter around a shield that is already bonded at both ends and see whether current is flowing. Those two measurements take ten minutes and settle arguments that otherwise run for weeks, because they replace a theory about what the installation should do with a number describing what it does.

Write six things into the design and the noise problems mostly do not happen: which conductor is protective earth and which is signal reference and where they meet, the shield termination policy per signal type, the separation distances, the drive cabling standard, where isolation is used, and who verifies it at installation. The last one is the one that gets dropped, and it is the reason correct drawings so often produce noisy plants.