Grounding Done Right: Loops and Star Points

Ground is copper, not a magic zero

Guides 1 and 2 left you with two ideas: real circuits are bathed in noise, and the first habit that fights it is a local decoupling capacitor sitting right beside every hungry chip. This guide tackles the other half of that habit — the return path. On a schematic every ground symbol is one tidy node, a perfect 0 V that all voltages are measured against. On the real board that node is a stretch of copper, and copper has resistance and inductance. The instant current flows through it, the two ends are no longer at the same voltage, and 'ground' quietly stops being zero.

Put real numbers on it with nothing fancier than Ohm's law. Suppose a return trace has 0.1 ohm of resistance and the circuit's output stage pulls 1 A back through it. The voltage from one end of that copper to the other is V = I times R = 1 A times 0.1 ohm = 100 mV. Now imagine a delicate 10 mV sensor signal that happens to use the same copper as its reference: it is sitting on top of a 100 mV bump that wags around with the load current. The signal you cared about is buried ten times over, and no amplifier downstream can rescue it — because the bump is inside the reference it measures against.

Shared impedance: why signal ground is not power ground

That 100 mV bump is the villain of grounding, and it has a name: shared-impedance coupling. Whenever a noisy, high-current return and a quiet, sensitive return flow through the same piece of copper, the big current's IR drop appears in series with the small signal. The fix is structural, not a component: give the loud currents their own return path and the quiet ones theirs, so they meet only where you choose. Engineers name these signal ground (the quiet reference for inputs, sensors, the front of an amplifier) and power ground (the muscular return for output stages, motors, the ground pin of a switching regulator).

SHARED-IMPEDANCE GROUND  (bad)         SINGLE-POINT 'STAR' GROUND  (good)

 quiet stage o----+                     quiet stage o-----------+
                  |                                             |
 loud  stage o----+---[ Z_g ]---o GND    loud  stage o------+    |
                  ^                                         |    |
           both returns share                               +----+---o GND  (one node)
           the SAME copper Z_g ;
           I_loud x Z_g lands in         each return has its OWN wire ;
           series with the quiet          they meet ONLY at the star
           stage's reference              point  ->  no shared drop

The rule of thumb falls straight out of the picture: never let a high-current return wander through copper that a sensitive node relies on. Route the output-stage current, the LED current, the regulator's return back to the supply by their own path, and bring the quiet signal returns back separately. They are all 'ground' and they will all join eventually — the art is choosing the single spot where they join, so no big current ever crosses a small one's reference on the way.

The ground loop and the 50/60 Hz hum

Connect a piece of equipment to ground in two places and you have built a closed loop of wire — a ground loop. Two things then go wrong. First, that loop is a single-turn coil, and the world around you is awash in the magnetic field of mains wiring; by inductive coupling (guide 4's subject) the loop catches that field and a 50 or 60 Hz current is induced to circulate. Second, the two 'ground' points are rarely at exactly the same potential — a few millivolts to volts of difference across a building's wiring is normal — so a current flows around the loop just to equalize them. Either way a 50/60 Hz voltage now rides on your reference, and in audio you literally hear it: the classic hum.

The same return path bites at high speed too, through inductance rather than resistance. A short wire is roughly 1 nH per millimetre, so a 2 cm lead is about 20 nH. When a digital chip switches and yanks a current step of, say, 0.1 A in 10 ns, the di/dt is 0.1 A / 10 ns = 10^7 A/s, and the voltage kicked up across that 20 nH is V = L times di/dt = 20 nH times 10^7 = 200 mV. That spike is ground bounce, and it is exactly why guide 2 begged you to keep the decoupling capacitor's loop tiny: the cap can only hold the rail steady if the current it dumps does not have to fight a long, inductive return. Decoupling and grounding are the same problem seen from two ends.

Star grounding versus the ground plane

If shared impedance and loops are the disease, the cleanest cure at low frequencies is star grounding, also called single-point grounding. Picture the spokes of a wheel: every stage runs its own return wire back to one central node — the star point, usually right at the supply's negative terminal or the main ground connection. Because no two returns share copper, no current's IR drop lands in another's reference, and because everything meets at a single point there is no loop to enclose area. For audio, precision DC, and instrumentation this is close to ideal.

But star grounding quietly fails as frequency climbs, because a wire's enemy is no longer its resistance but its inductance — and a long thin return spoke is all inductance. At high speed the return current does not want to follow a defined wire at all; given a solid sheet of copper it flows back along the path of least inductance, which is *directly underneath the signal trace*. So fast and digital boards do the opposite of star: they lay down an unbroken ground plane, a continuous copper sheet that hands every signal its own tight, low-inductance return hugging the trace above it. There is no single 'right' scheme — single-point star for low-frequency precision, an uninterrupted plane for high-frequency and digital, and mixed-signal boards carefully partition the two.

One warning follows straight from 'return current hugs the trace': do not cut a slot in the plane under a fast signal. If a gap forces the return current to detour around it, you have just created a big loop exactly where you were trying to avoid one — radiating, picking up, and adding inductance. A ground plane works because it is continuous; the moment you carve it up, the neat return path you were relying on goes looking for the long way round.

Kelvin sensing: measure where it matters

There is one more grounding idea that turns a precision measurement from hopeless into easy: Kelvin, or four-wire, sensing. The problem it solves is the very first one in this guide — the IR drop in a wire that carries current. If you both deliver current to a load and measure its voltage through the same two wires, the wires' own resistance adds straight into what you read. The trick is to split the jobs: one pair of wires forces the current, and a separate pair senses the voltage right at the load's terminals. Because the sense pair feeds a high-impedance input (an op-amp draws almost no current), almost no current flows in it, so it has almost no IR drop, and you read the load voltage cleanly.

1. Pick the point you actually care about — the two terminals of the load itself, or the two pads of a current-sense resistor. That, not a convenient spot on a nearby trace, is where the true voltage lives.
2. Run two more conductors, the sense lines, from exactly those terminals. Land them as close to the component body as the footprint allows, so no shared copper sits between the part and the tap.
3. Feed the sense pair into a high-input-impedance difference or instrumentation amplifier, so the lines carry negligible current and develop negligible drop of their own.
4. Keep the force (current-carrying) pair separate the whole way; let it have all the IR drop it likes — the sense pair never sees it. This is exactly how a Wheatstone bridge, an RTD, and a 4-20 mA loop stay accurate over long, lossy wiring.

Notice what every idea in this guide had in common: not one of them was a new component you could bolt on at the end. Where the return current flows, where two grounds meet, how big a loop encloses, where you tap a voltage — these are decided by the layout, the same theme guides 1 and 2 kept circling. Grounding done right is not a fix you apply to a finished circuit; it is part of designing the circuit, and the engineers who internalize that are the ones whose boards work the first time. Guide 4 picks up the coupling mechanisms — capacitive, inductive, conducted — that this loop-and-plane thinking was already quietly defending against.