Ground Loops & Grounding Practice
14 min read
Quick Answer
A ground loop is a closed path formed when two pieces of equipment share more than one ground connection. Current circulating in that path crosses the impedance the two share, and the resulting voltage adds directly to any signal referenced to it. It appears as mains hum or as a drifting offset.
Intuition
The hum that arrives with the second cable
A mixing desk on its own is quiet. A power amplifier on its own is quiet. Join the two with a signal lead and a low buzz appears in the speakers, pitched at the frequency of the mains supply, and it sits there whatever the volume control is doing.
Nothing inside either box has changed. What changed is the number of paths joining them. Each unit was already connected to the earth pin of its own mains socket, so the two metal cases were already linked through the building's wiring. The signal lead has now added a second link between them. Those two paths, taken together, close a ring, and that ring is a ground loop.
A ring of conductor is not harmless. Any current that finds its way round it, and any disagreement between the two earth points it joins, shows up as a small voltage between the two cases. Neither case is at fault, and both are connected as intended. What is missing is the assumption the equipment was built on, that both ends of the cable agree on where zero volts is.
The receiving equipment cannot separate that voltage from the wanted signal, since both arrive on the same pair of wires. It amplifies the pair together, and out of the speaker comes the music with the mains supply's own rhythm underneath it.
Turning the gain down is the first thing anyone tries, and it does nothing to the balance between the two, since gain acts on the interference as well. Disconnecting one of the two earths does silence it, and takes a safety conductor out of the installation on the way.
Practitioner
What the shared copper adds to the signal
Whatever joins the two chassis, a cable screen or a run of mains earth wire, has resistance of its own. Push current along it and a voltage appears between its ends, and those ends are the two points the drawing calls ground.
Worked example — The offset along a shared return
Two chassis are joined by a ground conductor of 50 mΩ, and the return current of something else in the system, a motor drive or a heater supply, comes to 2.0 A along the same path.
Between the two ends sits 100 mV.
Nothing here has failed. The conductor is doing the job it was installed to do, and the two chassis simply stop agreeing about zero for as long as that current flows.
The conductor resistance used there is an illustrative example parameter, not a catalogue value, though it is an easy one to reach: a painted joint in a rack, or a length of screen braid crimped into a pigtail, will land somewhere in that region.
How much the offset matters depends on what it lands on top of. Line-level audio and a thermocouple output are both smaller than it, and the comparison between a wanted voltage and an unwanted one is normally reported as a decibel ratio.
Worked example — The signal set against the offset
A sensor on one chassis produces 10 mV, and an amplifier on the other chassis reads it against its own local ground.
The offset of 100 mV arrives added to the signal. It is 10 times the size of what was wanted, and the ratio between the two comes to -20.0 dB.
The negative sign says the wanted part is the smaller of the two. Gain will not recover it, and neither will averaging, since this is a real voltage that tracks somebody else's load current.
Finding one on the bench is a process of elimination. A loop needs both of its connections, so pulling the signal lead should silence it and so should unplugging the second unit's mains lead; whichever one silences it names the pair of earth points the ring runs between. Battery power on one unit is the quickest diagnostic of all, and a diagnostic is all it is. A multimeter across the two chassis with the signal lead unplugged reads the difference directly, on AC volts, but only where both units are safe to touch and the meter is rated for the installation.
Engineer
The loop drawn as a circuit
Draw the ground conductor as a resistance instead of as a symbol and the arrangement resolves into two meshes sharing one branch. The load's heavy current circulates in one mesh, the signal current in the other, and the shared branch carries both. Apply Kirchhoff's voltage law round the signal mesh and the voltage across that branch enters the sum regardless of which current put it there. The mechanism has its own name, common-impedance coupling, and that name is the more useful of the two: a ring with nothing shared inside it does no harm, while a shared conductor with no ring around it still ruins a measurement.
The remedy follows from the name. Nothing has to be cut; the two returns have to stop overlapping.
Worked example — Same conductor, returns separated
The load's return is rerouted so that it reaches the supply along a path of its own. The conductor between the two chassis is left carrying nothing but the signal's own return, which comes to 1.0 mA.
The offset across it drops to 50 µV, and the signal-to-noise ratio moves from -20.0 dB to 46.0 dB.
The copper is the same copper and the ring is still closed. What changed is which currents share a conductor.
Single-point grounding, sometimes called star grounding, is that result written as a rule: bring each return separately to one agreed node, and no circuit's current can appear in another circuit's reference. Stating it as a rule hides the reasoning, and the rule then gets applied where it does not belong.
The resistance model holds over a limited range. It describes the low-frequency case, where a short conductor's impedance is essentially its resistance. Let the frequency climb and the inductance of the same piece of wire takes over, so the identical current produces a far larger voltage and length starts to matter more than cross-section. Single-point grounding fails at the same boundary: a long return to a star point is a long inductor, and above the frequency where that dominates, many short connections to a plane beat one tidy central one.
A second mechanism produces the same symptom and needs separating from the first. The ring encloses an area, and a changing magnetic field through that area induces a voltage round it with no shared conductor involved at all. A mains transformer or a nearby cable run will supply the field. Common-impedance coupling is treated by rearranging returns; induced-loop coupling is treated by shrinking the enclosed area, which in practice means running a signal and its return as a twisted pair along the same route. EMI and EMC covers both coupling paths, and neither of them is noise in the sense that lesson defines: each has a source, and each stops when its source does.
One more limit sits in the arithmetic above, which treated the interfering current as fixed. In a real installation it is not. It divides between every parallel path back to the supply, so bonding two chassis together more heavily diverts current into the new bond instead of removing it. Thickening the shared conductor lowers the voltage it develops and raises the share of current it takes, and the two effects work against each other. The improvement is real, and it is smaller than the change in resistance alone would suggest.
Professional
Fixing it without removing the earth
The structural answer is to stop referencing the signal to either chassis. A differential receiver responds only to the difference between two wires and ignores whatever both of them carry in common, so an offset that arrives equally on the pair is rejected rather than amplified. How completely it is rejected is quoted as common-mode rejection, and the figure a design needs comes out of an error budget, not out of a preference.
Allow the offset computed earlier to contribute no more than 100 µV of error. Suppressing 100 mV to that level asks for at least 60.0 dB of common-mode rejection: comfortable for an instrumentation amplifier, and beyond a difference stage thrown together from ordinary resistors, whose rejection is set by how well four resistors match. Balanced audio interconnects apply the same reasoning under a different name. The three-pin connectors on professional equipment carry a signal as a difference between two conductors, with the screen kept out of the signal path entirely and left to do nothing but intercept fields.
Isolation goes further and opens the ring. An isolating transformer passes a signal magnetically, an optocoupler passes it optically, and an isolated converter does the same for a supply rail. Each removes the conductive path completely, so no current can circulate at all, and each is paid for in bandwidth, in linearity, in power or in money. Fibre is the limiting case, with no residual coupling capacitance worth arguing about. Where the connection has to stay conductive, a common-mode choke offers impedance to the circulating current while leaving the differential signal alone, and ferrite beads do a coarser version of the same job.
Cable screens force a decision with no universal answer. Bond a screen at one end only and it cannot carry loop current, but the free end behaves as an antenna at radio frequencies and the shielding suffers. Bond both ends and the shielding is good, at the price of closing the very ring this lesson is about. Bonding one end directly and the other through a capacitor is the usual compromise, since a capacitor is close to a short circuit at radio frequencies and close to an open circuit at the mains frequency.
Instruments deserve the same scrutiny as the equipment under test. Clip two oscilloscope probes to different points on one board and their ground clips are joined through the instrument's chassis, which is a ground loop the operator has just built, and on a board with any current flowing in its ground plane the two traces will disagree because of it. A differential probe or an isolated input measures between two points without tying either to earth, and scope probes covers what that costs.
Inside a product a designer chooses all of this freely: which returns share copper, and where an isolation barrier goes. The building is different. How equipment is bonded and earthed in an installation is set by the wiring regulations in force in that country, the work is licensed in most places, and no hum is worth a modification to it.
Safety
Ground lift adapters sold as a hum cure work by disconnecting the protective earth conductor, which is the one repair named here only so it can be ruled out. That conductor exists to carry fault current so that a fuse or a breaker operates quickly and exposed metalwork never becomes live. Take it away and the metalwork is held safe by insulation alone, with nothing left to trip when the insulation fails. Cutting or taping over an earth pin does the same damage permanently, and leaves the next person no way of knowing.
An unearthed chassis does not sit at zero either. Leakage through a supply's own input filtering can hold it well above earth potential, and that is what the faint tingle from a metal case is. Treat it as a warning, not a quirk of the equipment.
Defeating an oscilloscope's earth so that it stops closing a loop is the same mistake with an instrument attached. The whole chassis and every exposed connector on it then float at whatever the circuit under test is doing. Use a differential probe or an isolated input instead, check the CAT rating of the leads and the instrument first, and take electrical safety fundamentals as the governing practice. Each cure described above works with the earth connection left as the installer made it.
Common mistakes
- Cutting or lifting the mains earth to stop hum — the protective conductor goes with it, and the first insulation fault then leaves a live chassis with nothing to trip. Fix the signal connection instead.
- Blaming the ring instead of the shared conductor — a loop with nothing shared inside it does no harm. Ask which currents pass through the same piece of copper.
- Adding gain to overcome the offset — interference and signal are amplified together, so the ratio between them will not move.
- Assuming a heavier bond solves it — lowering the shared conductor's resistance also increases the current it attracts, so the offset falls by less than the resistance change predicts.
- Bonding cable screens at both ends by habit — that is right for radio-frequency shielding and wrong for a low-frequency loop. Decide per cable, or bond the far end through a capacitor.
- Treating hum as noise — it is deterministic interference with a source that can be found, so filtering and averaging are the wrong instruments for it.
Frequently asked questions
What causes a ground loop?
More than one ground connection between the same two pieces of equipment. Typically each unit is earthed through its own mains lead and the two are also joined by the screen of a signal cable, which closes a ring. Current circulating in that ring, or a difference in potential between the two earth points, ends up added to the signal.
Why does a ground loop sound like a hum?
The circulating current comes from the mains supply and from the loads connected to it, so it carries the supply frequency along with harmonics of it. The fundamental on its own would be a smooth hum; the harmonics are what turn it into a buzz. The pitch is a useful clue, since it identifies the mains as the source before any measurement is taken.
Is a ground lift adapter a safe fix?
No. It works by breaking the protective earth connection, so the equipment loses the conductor that carries fault current and operates its protective device. It removes a hum and creates the conditions for a live chassis. The remedies in this lesson all work with the earth left intact.
Should a cable screen be grounded at one end or both?
One end where the trouble is a low-frequency loop, both ends where the cable has to shield against radio frequencies. Bonding one end directly and the other through a capacitor covers both cases, since the capacitor closes the screen at high frequencies and leaves it open at the mains frequency.
Do ground loops affect digital links too?
Yes, though the symptom differs. A digital receiver has voltage margin, so a slow offset is often absorbed instead of heard, and the failure shows up as occasional corrupted data or as emissions radiated by the circulating current. Ethernet is transformer-coupled at each end, and many industrial buses are isolated too, which keeps the question from arising on those links.