EC-8.4 Grounding, Shielding and Where Noise Gets In

What this is and why it exists

The most frustrating hours in a laboratory are spent chasing something that is not in the circuit at all.

This topic is the map of where it comes from instead. It is placed early because everything after it inherits the same problems. A sensor reading, an embedded converter and a communication receiver are all measurements. All of them have interference arriving by the routes described here.

The useful division is not between good signal and bad noise. It is between noise that was always going to be there and noise that your own wiring invited in.

The vocabulary

• Thermal noise — the unavoidable noise produced by random motion of charge in any resistance.
• Noise floor — the level below which a measurement cannot see, set by the physics of the parts.
• Interference — unwanted signal coupled in from somewhere else.
• Ground loop — a closed path formed by two separate connections to ground.
• Single-point ground — a wiring scheme in which every ground connection meets at one physical place.
• Capacitive coupling — interference passing between conductors through the capacitance between them.
• Inductive coupling — interference passing through the magnetic field around a changing current.
• Common-mode interference — interference that appears equally on both wires of a pair.

The mental model

Separate the noise that comes with the physics from the noise that came with your wiring. They need entirely different responses.

Thermal noise exists in every resistance at every temperature above absolute zero. Shot noise exists wherever charge crosses a barrier one carrier at a time. Together they set a floor. You cannot go below it, and the only levers are lowering the resistance, cooling it, or narrowing the bandwidth you are looking through.

Interference from a nearby power supply, a motor, a switching converter or a mobile handset is a different animal. It came in through your wiring and it can be removed entirely. Almost all of the effort in a real laboratory goes here.

Ground loops are the most common single cause. If a circuit reaches ground by two different routes, those two routes enclose an area, and that area is a loop of wire. Any changing magnetic field passing through it induces a current, and that current produces a voltage along the ground conductor that adds to your signal.

The fix is to make sure there is only one route. Bring every ground connection to one physical point. When an instrument's mains earth creates a second route that you cannot remove, an isolated instrument or a differential measurement is the way out.

Shielding is misunderstood often enough to be worth stating carefully. A shield works by giving interference somewhere to go, which means it must be connected to something. A shield floating at both ends does very little. A shield connected at both ends is itself a loop. Connecting it at the source end only is the usual right answer.

The two coupling mechanisms behave differently and need different fixes. Capacitive coupling passes a changing voltage from one conductor to a neighbour through the capacitance between them. It grows with the voltage, with the frequency and with how close and how parallel the wires run. Separation helps, and so does a shield.

Inductive coupling passes a changing current from one loop to another through the magnetic field. It grows with the current, with the frequency and with the area of the loops. Separation helps less here, and reducing the loop area helps a great deal. Running the outgoing and returning conductors close together is the standard move.

That leads to the twisted pair. If interference reaches both wires equally, then a measurement of the difference between them removes it. Twisting makes the pickup on the two wires as nearly equal as possible, because each half twist swaps which wire is nearer the source. Every data cable in the building is twisted for this reason, and the same trick works for a sensor on a bench.

Averaging and filtering are the last two tools, and neither is free. Averaging a hundred readings improves the random part of the noise by a factor of ten and costs a hundred times the time. Filtering removes noise outside the band you care about and costs speed of response, because a narrow filter takes longer to settle.

Both fail against interference inside your signal band. A fifty-hertz hum cannot be filtered out of a measurement with real content at fifty hertz. Averaging over a whole number of mains cycles is the trick that removes it instead.

What you should now be able to explain or do

• Tell noise that was always present from interference that your wiring admitted.
• Find a ground loop in a bench setup and remove it with a single-point ground.
• Decide where a shield should be connected, and say why both ends is usually wrong.
• Distinguish capacitive from inductive coupling and choose the right fix for each.
• Explain why a twisted pair with a differential measurement rejects interference.
• Choose between averaging and filtering, knowing what each one costs.

Check yourself