Ground & Reference Points
Also known as: earth, common, GND
11 min read
Quick Answer
Ground is the node in a circuit chosen as the zero-volt reference, against which every other voltage is quoted. The choice is a convention rather than a physical property. In a real circuit the ground conductor has resistance and inductance, so different points on it are not at exactly the same potential.
Intuition
Sea level, agreed by everyone
Heights on a map are given above sea level, which is not a physically privileged place. A hill is 300 m high because everyone agreed to measure from there. Choose a different datum and every number on the map changes, while every hill stays exactly the height it was.
Circuits do the same thing with voltage. A voltage only ever exists between two points, so quoting a single number for a node needs a reference that everybody has agreed on. That agreed node is called ground, and by definition it sits at zero volts. Everything else is quoted relative to it: a 5 V rail is 5 V above ground, a −12 V rail is 12 V below it.
Which node gets the job is genuinely arbitrary. Nominate the positive terminal of a battery-powered circuit as ground instead of the negative one and every voltage in it changes sign, while the circuit behaves in precisely the same way, because only differences drive current. Some older automotive and telecom systems are positive-earth for that reason, and they work perfectly well.
On a schematic, ground is drawn as a symbol instead of as wires: a short stack of shrinking horizontal lines, or a triangle, hanging below whatever connects to it. All of those symbols on a sheet refer to one node, and drawing them separately is what keeps a diagram readable.
Ground and earth get used interchangeably in conversation, and they are not the same thing. Ground in electronics usually means the circuit's zero-volt common. Earth means an actual connection to the planet, provided for safety. In battery-powered equipment the two are unrelated; in mains-connected equipment they may or may not be joined, and knowing which matters a great deal.
Practitioner
The copper is not ideal
A schematic shows ground as a single node holding a single value. On a board it is a piece of copper carrying every return current in the circuit, and copper has resistance. Current through that resistance produces a voltage, so two points the drawing calls zero volts do not sit at the same potential:
Worked example — Where the zero moves
A load draws 1.5 A and returns it through a ground path of 30 mΩ back to the supply.
Multiplying gives the voltage that appears along the return: 45 mV.
The load's local ground therefore sits that much above the supply's ground for as long as that current flows. Everything here is working exactly as designed. A schematic that ties both points to the same symbol is telling the truth about the connection and saying nothing at all about the copper that makes it.
How much that matters depends on where the return currents join. A small-signal circuit whose return meets the ground path downstream of a high-current load's return has its reference move every time the load current changes. Bring that return in upstream — at the source — and the reference holds still. Choosing between those two arrangements is what the name "star grounding" refers to, and it is a decision about topology rather than about copper area.
On a board the usual answer is a ground plane. A continuous copper layer gives every return current a low-resistance, low-inductance path directly beneath its outgoing signal, which keeps both the drop and the loop area small. Slots cut into that layer, or a signal routed across a gap in it, force the return to detour and undo the benefit; it is the most common self-inflicted layout problem there is.
The same care applies on the bench. Test equipment usually has its measurement ground bonded to mains earth, so clipping a scope probe's ground lead onto a node earths that node through the instrument. Where the circuit under test is isolated, that does no harm. Where the circuit is already referenced to the mains, it does considerable harm.
Real designs routinely carry several grounds. Analogue ground, digital ground, power ground and chassis ground are names for parts of one return system, joined at a single chosen point. The names are how a designer states where the currents are meant to flow, and they mean nothing unless that joining point has been picked on purpose and drawn on the schematic.
Engineer
Reference errors, and how measurements go wrong
Whatever offset the return currents put on the ground adds directly to any signal measured against it.
Worked example — A small signal on a moving reference
A sensor produces 100 mV and is read single-ended, meaning its output is measured against the local ground. That local ground sits 45 mV above the reference used by the measuring circuit.
The offset adds straight onto the signal, an error of 45 %.
Amplification is no help, since the offset is amplified along with the signal. Averaging does not remove it either, because this is a real voltage rather than noise, and it shifts whenever the unrelated load current shifts, which is what makes it so hard to pin down. A reading that jumps when a motor starts is usually this.
The structural fix is to measure differentially. Sense both ends of the signal source and respond only to the difference, and the shared offset stops mattering, because it arrives equally on both inputs. An instrumentation amplifier does exactly that, and how well it rejects a shared offset is quoted as its common-mode rejection. Three-wire and four-wire sensor connections exist for the same reason.
All of this is Kirchhoff's voltage law restated. Go round the loop containing the load, the return path and the measuring circuit, and the potentials sum to zero, which is exactly what requires the offset to be there. Draw ground as a real conductor instead of as a symbol and the arithmetic becomes entirely ordinary.
The sea-level picture has to be given up at this point. Ground is zero by definition at one point, and everywhere else it is whatever the return currents make it. An experienced designer asks where a current returns and what it passes through on the way, rather than whether a node is grounded.
At speed the offending quantity is inductance rather than resistance. A rapidly changing current produces a voltage across the inductance of the return path as well as across its resistance, and that voltage can be far larger. In digital circuits it appears as ground bounce: the local ground of a chip lifts during a fast output transition, and any input judged against that ground is momentarily misread. The remedies are shorter return paths, more of them and local decoupling. All of that is geometry, and a schematic shows none of it.
Where grounds meet in more than one place, they form loops. Two pieces of equipment each connected to earth and also joined by a signal cable close a path that can enclose changing magnetic flux, and a current circulates in it. In audio that current is mains hum; in instrumentation it is a wandering offset. Ground loops covers the problem and its remedies.
Professional
Grounding as a design discipline
Grounding is decided at system level, and deciding it late is expensive. Where the returns join, which subsystems share copper, where the chassis connects and where isolation sits are all architectural choices. Revisiting them after layout means moving connectors, adding isolators, or accepting a specification that cannot be met.
An error budget is what makes a ground offset visible, because converting it into the units of the measurement stops a millivolt-scale figure looking negligible. Take a converter with a 3.3 V full-scale range and 1024 steps, giving 3.22 mV per step. The same 45 mV the return path produced earlier is worth 14 codes of error: an offset a voltmeter would barely notice, consuming a meaningful part of a ten-bit reading. Adding resolution makes that offset worth more codes rather than fewer, which is why extra bits bought without attention to grounding buy nothing. See analog vs digital.
Protective earth is there for safety, and its job is to carry fault current so that a protective device operates. It is sized for that fault current and may carry leakage current continuously, which makes it a poor place to reference a measurement. Treating it as a signal ground imports every other appliance's leakage into your circuit.
Isolation removes the shared reference altogether. Optocouplers, isolated converters, transformers and isolated instrument inputs let two subsystems each keep a ground of their own with no galvanic connection between them. That answers both a ground loop and a differential voltage larger than any amplifier can reject, and it is paid for in power, bandwidth and money.
Splitting a ground plane is usually the wrong instinct. The intuitive fix — cut the plane so digital currents cannot reach the analogue section — forces the returns into long detours around the cut, increasing loop area and often making both interference and emissions worse. The modern default is one continuous plane with the parts and their return paths placed deliberately, and a split is reserved for a genuine isolation barrier.
Star grounding stops working as the frequency rises. Single-point returns are exactly right at low frequencies. Once the frequency climbs, the inductance of a long single-point return dominates, and the answer becomes many short connections to a plane instead. Most real designs end up a hybrid, chosen per subsystem according to the frequencies that subsystem works at.
Safety
An oscilloscope's ground clip is usually bonded to mains earth through the instrument's own power lead. Clipping it to a point in a mains-referenced circuit — the primary side of a non-isolated supply, for instance — earths that point through the scope. The result is a short circuit through the probe lead, or the entire chassis of the instrument sitting at a hazardous potential.
Never attach a ground clip to a circuit that is not isolated from the mains. Use a differential probe or an isolated input for such measurements, and check the CAT rating of the equipment first. Defeating the instrument's own earth connection to "float" it is not an acceptable substitute: it removes a safety conductor and leaves the whole chassis live. The general practice is set out in electrical safety fundamentals, and the probe side in scope probes.
Common mistakes
- Treating ground as zero volts everywhere — it is zero at one chosen point. Everywhere else it is whatever the return currents put there.
- Joining a sensitive circuit's return downstream of a high-current return — the sensitive circuit's reference then moves with the load current, and the symptom looks like a sensor fault.
- Measuring a small signal single-ended over a long return — the ground offset adds directly to the signal and cannot be amplified away. Use a differential connection.
- Using protective earth as a signal reference — it carries fault and leakage current from everything else on the installation and is not quiet.
- Splitting a ground plane to keep sections apart — the returns detour around the cut, enlarging loop areas and usually making interference worse.
- Clipping a scope ground onto a mains-referenced circuit — the clip earths that node through the instrument. Use a differential or isolated probe.
Frequently asked questions
What does ground mean in a circuit?
The node chosen as the zero-volt reference, against which all other voltages are quoted. The choice is a convention; only voltage differences have physical effect.
Is ground the same as earth?
Not necessarily. Ground is the circuit's zero-volt common; earth is a physical connection to the planet, provided for safety. In battery equipment they are unrelated; in mains equipment they may be bonded.
Why do two ground points measure differently?
Because the conductor between them carries current and has resistance and inductance. The resulting voltage is real, and it changes whenever the current does.
What is a ground loop?
A closed path formed when two pieces of equipment are connected to earth at different points and also joined by a signal cable. Circulating current in that loop appears as hum or as a wandering offset.
Should I split my ground plane into analogue and digital sections?
Usually not. A split forces return currents to detour, increasing loop area and often worsening interference. Prefer one continuous plane with deliberate placement of parts and returns.