Ground is a place, not a number.
A wire has resistance, and current through resistance is a voltage. Run three amps back to the battery through a metre and a half of thin wire and three crimps, and the far end of that wire sits a quarter of a volt away from the near end — and the part of that quarter-volt which varies is what lands in your audio. Both ends are called ground. Both are drawn as the same symbol. They are not the same potential, and the difference is a signal source sitting in series with everything that references either of them.
That is common-impedance coupling, and it is the mechanism behind most of the cockpit noise that reaches our bench. The intercom is the usual victim because a microphone input is looking for a few millivolts of speech, so a few millivolts of somebody else's return current is not background — it is the same size as the signal. A pump, a landing light or an LED strip switching on and off writes its own current waveform straight into your audio.
The loop, and the star that removes it.
A ground loop is what you get when two devices are connected both by a signal wire and by a shared ground path: the two routes enclose an area, that area is a single-turn coil, and any changing magnetic field in the cockpit induces a current in it. The fix is topological rather than electrical. Give every instrument its own return to one point — the battery negative or a single grounding stud — and no two devices share a path, so none of them can write on another's reference. That is a star ground, and it is cheap to build and expensive to retrofit.
Metal aircraft can use the airframe as that single reference, which is why aviation electrics look the way they do. Composite aircraft have no such luxury: glass conducts nothing at all, and carbon conducts just enough to be a bad conductor, which is the worst of both worlds. In a modern glider every ground is a wire you chose to run, and the star has to be built deliberately because nothing else will do it for you.
Where the millivolts come from.
Start daisy-chained with three amps and the hum bar sits far above the audible line — that is one shared return doing all the damage. Switch to a star and it collapses by a factor of forty without a single component changing. The red rings are the other problem entirely: a switching regulator radiating into the satellite band, which no amount of grounding fixes. Only distance and filtering move that one.
Shields: ground one end, deliberately.
A shield is an electrostatic screen. Its job is to intercept capacitively coupled interference and give it a path away from the conductor inside, and for that it needs to be connected at one end. Connect it at both and you have handed the shield a second job it is bad at — carrying ground current — and built exactly the loop the star was meant to prevent. For audio and microphone lines the convention is to ground at the intercom, the end that defines the reference.
Coaxial antenna cable is the exception that confuses everybody, because it is grounded at both ends by construction. That is correct: at radio frequencies the shield is part of the transmission line and not a screen at all. Leave it alone. Where you can still get it wrong is the pigtail — a long thin wire from the shield to its termination is an inductor, and an inductor undoes the shield at exactly the frequencies you cared about. Short and fat wins.
Whine, buzz and silence: three faults, two of them audible.
A tone whose pitch follows the engine is alternator ripple arriving on a shared return. It appears the day something got tapped onto the wrong terminal, and it goes away when the offending device gets its own wire home. A fixed buzz at a hundred hertz or a few kilohertz is a switching load — a dimmer, a fuel pump, a strobe supply — and it usually stops when that load is switched off, which is also how you identify it.
The third one makes no sound at all, and it is the expensive one. A cheap switching regulator, and above all a USB charger, radiates harmonics across the bands the GPS and the traffic receiver live in. There is no hum to notice; what you get is a satellite count that used to be twelve and is now seven, and a traffic range that quietly halved. Since almost nobody links a phone charger to a FLARM range complaint, this fault survives for years. Unplug everything, note the satellite count, and add devices back one at a time.
In deinem Cockpit.
Everything here shares a root cause with the voltage-drop arithmetic in the cockpit power article — wire has resistance, and current turns it into a voltage that lands somewhere you did not intend. The difference is only which end you notice it at: the far instrument browns out, or the near one hums. While you have the panel open and a meter in your hand, measure the bus termination too: the CAN bus article explains why 60 Ω across the pair with the power off is the single most useful number in the aircraft.