Ground Noise: Loop, Common Impedance and Return Path Problems
A ground conductor is treated as a reference, and in a schematic it is. On a board it is a piece of copper with resistance and inductance, and at the frequencies a digital circuit uses, the inductance is what matters. Current flowing through that impedance produces a voltage, and a voltage on the ground a circuit measures against is indistinguishable from noise in the signal. Most of the interference problems that appear in a working board trace back to this one fact.
Why a Ground Conductor Is Not a Zero Volt Node
The distinction between resistance and impedance is the key. Resistance is what a conductor presents to direct current. Impedance is what it presents to alternating current, and for copper it is dominated by inductance rather than resistance. Measuring a ground trace with a multimeter returns a value in milliohms, which is the resistance, and that reading says nothing about the impedance the trace presents to a switching signal.
Two numbers make the scale of the effect concrete. A piece of copper 10 centimetres long, 1.5 millimetres wide and 50 micrometres thick has a resistance of roughly 0.026 ohm, which is negligible in most circuits. Its inductance, at about 0.8 microhenries per metre along a length much greater than its width, is around 0.08 microhenries. Multiplying that inductance by the angular frequency of the signal gives the reactance, and at the frequencies present in a digital edge the reactance is far larger than the resistance. That is why a ground connection which measures perfectly becomes a voltage source once the circuit starts switching.
Ground Loop Interference
Ground loop interference appears where two pieces of equipment are connected by a cable and both are grounded. Their ground potentials differ, and that difference drives a current around the loop formed by the two grounds and the cable between them. Because the circuit is not perfectly balanced, the current in each conductor of the cable differs, and the difference appears as a differential voltage across the signal pair.
The diagnostic signature of this failure is memorable: disconnecting one ground makes the problem disappear. That is not a solution, but it identifies the mechanism, and it explains why the effect is most pronounced at low frequency. At high frequency the loop impedance rises and other mechanisms take over.
Measuring the Noise
Ground noise is measurable, and measuring it resolves arguments that inspection cannot. A wide bandwidth probe placed between two points on the plane, with a short ground lead or a coaxial connection, shows the actual voltage difference during switching. Probing between the analogue reference and the digital reference at the same moment shows exactly how much noise the sensitive circuit sees.
The measurement should be made under the worst case load rather than at idle, because the noise depends on the current being switched. Working at the device pins rather than at the supply entry also matters, since the noise that affects a device is the noise at its own reference pin, not the average across the board. Once the source and the path are identified, the fix is usually a change in geometry rather than in components.
Practical Design Rules
Keep the reference continuous under every high speed net, and treat any plane split as a boundary that no trace may cross. Give every connector a defined return, place decoupling against the pin it serves, and route sensitive analogue traces over a quiet region rather than under a noisy package. These rules cost nothing but attention and eliminate most ground noise problems before they exist.
Where a design does need a boundary, make it explicit and control the joint. Write down where the two reference regions meet, why the boundary is there and which nets may cross it, and verify the arrangement with a probe during bring up. A deliberate split that is understood is safe; an accidental one created by a slot, a connector or a via fence is what turns a working prototype into a design that fails intermittently in the field. Our notes on power plane design cover the related supply side of the same picture.
FAQ
What causes ground noise on a PCB? Current flowing through the resistance and inductance of the reference plane develops a voltage across it. The noise is worse when the return path is long, when it is shared between noisy and sensitive circuits, or when a plane split forces the current to detour.
Does a thicker plane remove ground noise? It reduces the resistance but does little for the inductance, which is what dominates at high frequency. Geometry, and particularly the continuity of the path, matters far more than copper thickness.
Should I use separate analogue and digital grounds? Usually a single continuous plane with careful placement is better. A split is justified only where a defined boundary prevents a specific noise current, and it must have one controlled joining point.
Common Impedance Coupling
Common impedance coupling is the mechanism that matters most inside a digital board. Where several circuits share a section of ground, the current from one circuit flows through the same impedance as the current from another. The voltage that develops across that shared impedance modulates the ground potential of every circuit connected to it, so a switching current in one place appears as a disturbance somewhere else.
The size of the shared impedance determines the coupling. It can be reduced by widening the conductor, by using a plane instead of a trace, or by changing the topology so that the two circuits no longer share the path. Single point grounding removes the shared conductor entirely by giving each circuit its own connection to a common reference, and it is effective at low frequency, where the impedance of the individual connections is small.
Above a few megahertz, single point grounding stops working, because the individual connections have enough inductance of their own to make each one a different potential. At those frequencies the answer is the opposite: a continuous plane, so that every circuit connects to the same low impedance surface over a wide area.
Return Path Problems
A signal returns to its source through the path of least impedance, and at high frequency that path lies directly beneath the trace, in the adjacent plane. Any discontinuity in that plane forces the return current to detour, and the detour has two consequences: it enlarges the loop, which increases both radiation and susceptibility, and it changes the impedance of the line over that section.
This is why a signal that crosses a split in a reference plane is a design error rather than a matter of degree. It is also why the position of a connector and the arrangement of planes around it matter, because a return current that has to leave the board through a cable must find its way to the connector. Where the routing plan has to allocate ground and power conductors, the reasoning is set out in ground routing and power trace planning.
Layout Measures That Work
Four measures cover most ground related interference in a board layout. Use a ground plane rather than a network of traces, so that the impedance between any two points is low and predictable. Keep the return path continuous under every high speed signal, so that no current is forced to detour. Separate the noisy part of the ground from the quiet part, and join the two at a single controlled point rather than letting them share an uncontrolled path. And keep the loops small, because the field a loop radiates is proportional to its area.
The order matters. A plane is the foundation, and the other measures work on top of it. A design that separates grounds carefully but returns them through a trace network has solved nothing, because the trace network puts back the impedance that the separation was meant to control.
Grounding and Power Distribution Together
Ground noise and power distribution are the same problem seen from two sides. Current leaves a source, travels through the power network, passes through the load and returns through the ground network. If either half of that path is narrow, the voltage the device actually receives differs from the voltage the design assumes, and the difference appears as noise.
Designing the two together is therefore more effective than treating either in isolation. The layout that gives a device a wide, short return path also gives it a low impedance supply, and the same plane structure serves both. Where a plane has to be divided to keep two supplies apart, the division should be arranged so that no signal crosses it, because a split lying under a routed signal creates exactly the return path discontinuity the plane was there to prevent. The rules for splitting a plane without creating that problem are described in power plane splitting rules, and the effect of the harmonic content of the return current is discussed in ground current and harmonic distortion.
FAQ
Why does a ground connection cause interference when it measures zero ohms? Because the measurement gives resistance, while a switching signal sees impedance, which for copper is dominated by inductance. At high frequency the inductive term is far larger than the resistive one.
What is common impedance coupling? It happens when two circuits share a length of ground. The current of one produces a voltage across the shared impedance, and that voltage modulates the reference of the other circuit.
Should a board use single point grounding? It works at low frequency, where the individual connections have little impedance of their own. Above a few megahertz a continuous plane is the better answer, because it presents one low impedance surface to every circuit.
Why is crossing a plane split a design error? Because the return current cannot follow directly beneath the trace. It has to detour, which enlarges the loop, increases radiation and changes the impedance of the line over that section.
