What a ground loop is, what causes it, and how to break it

“Ground” on a schematic is one node. On a board, in a rack or across a room it is copper with resistance and inductance, and two points on it are at the same voltage only while no current flows between them. A ground loop is what happens when two circuits share a return path and one of them pushes current through it, or when two grounds are joined by more than one route and something drives current round the ring they form. In both cases the ground moves, and every circuit referenced to it sees the movement as a signal.

The three places it shows up are the three sources this article draws from. ADI’s MT-031 does the arithmetic for a shared ground wire on a board: one inch of #22 wire carrying a logic edge drops 200 mV, “an error of about 10 %” on a 2 V signal. Tektronix’s probe primer describes the bench version, a scope and a circuit on different mains outlets with the probe’s shield closing the loop, and its symptom — “line frequency noise (50 or 60 Hz)” on every trace. TI’s grounding articles describe the board-level version, a trace across a split plane whose return current has to go round the split, “a really big loop” that is also “a very effective dipole antenna”. The physics is the same in each; the remedy depends on which one it is.

What causes it: a shared impedance and a current through it

MT-031’s opening example is the whole mechanism. Buss wire is “totally unacceptable as a ‘ground’”, because of its inductance:

For instance, #22 gauge wire has about 20 nH/inch inductance. A transient current having a slew rate of 10 mA/ns created by a logic signal would develop an unwanted voltage drop of 200 mV at this frequency flowing through 1 inch of this wire … For a signal having a 2-V peak-to-peak range, this translates into an error of about 10 % (approximately 3.5-bit accuracy). Even in all-digital circuits, this error would result in considerable degradation of logic noise margins.

— ADI, MT-031

The 200 mV is V = L·di/dt: 20 nH × 10 mA/ns. Its resistance never enters — at logic edge rates the inductance is the whole problem, and it scales with length. MT-031’s figure 1 is the classic picture: “the digital return current modulates the analog return current … The ground return wire inductance and resistance is shared between the analog and digital circuits, and this is what causes the interaction and resulting error.” An ADC whose analog ground reference is that wire reads the digital circuit’s edges as signal.

MT-031’s own fix is the star: “make the digital return current path flow directly to the GND REF … the fundamental concept of a ‘star,’ or single-point ground system.” And its own caveat is why the star is rarely what gets built: “implementing the true single-point ground in a system which contains multiple high frequency return paths is difficult because the physical length of the individual return current wires will introduce parasitic resistance and inductance … In practice, the current returns must consist of large area ground planes for low impedance to high frequency currents.” A plane is a star with a thousand points. The mixed-signal grounding article is the argument for one plane; the loop here is the reason.

The bench version: the oscilloscope

Tek’s primer names the loop most engineers meet first:

Noise injection into the grounding system can be caused by unwanted current flow in the ground loop existing between the oscilloscope common and test circuit power line grounds and the probe ground lead and cable. Normally, all of these points are, or should be, at zero volts, and no ground current will flow. However, if the oscilloscope and test circuit are on different building system grounds, there could be small voltage differences or noise on one of the building ground systems … The resulting current flow will develop a voltage drop across the probe’s outer cable shield. This noise voltage will be injected into the oscilloscope in series with the signal from the probe tip.

— Tektronix, ABCs of Probes

The loop is the two safety earths, the building ground between the outlets, and the probe shield. It is a large loop — metres of it — so it is also an antenna, and the primer adds that the noise “is often line frequency noise (50 or 60 Hz). Just as often, though, the noise may be in the form of spikes or bursts resulting from building equipment, such as air conditioners, switching on and off.” Its diagnostic is that the noise does not depend on the signal: “poor grounding can result in noise appearing on any signal of any speed”, which distinguishes it from the ground-lead ringing that only appears on fast edges.

The primer’s remedies, in order: “minimize ground loops by using the same power circuits for the oscilloscope and circuit under test”, keep probe cables away from power cables, and if that is not enough, “open the ground loop” with “a ground isolation monitor”, “a power line isolation transformer on either the test circuit or on the oscilloscope”, “an isolation amplifier”, or “differential probes to make the measurement (rejects common-mode noise)”. And the line it sets in bold: “in no case should you attempt to isolate the oscilloscope or test circuit by defeating the safety three-wire ground system.” The ground-lift adapter that removes the hum removes the thing that trips the breaker when a chassis goes live.

The board version: the split plane

TI’s SLYT499 gives the rule that decides where a return current goes: “the high-frequency signals have a characteristic of following the path of least impedance (inductance). The path’s inductance is determined by the loop area that the path encloses … the return current flows on the plane adjacent to the trace.” A trace with a plane under it has a return path a few tenths of a millimetre away, and the loop it encloses is a sliver. Cut the plane under the trace and the return has to find the nearest copper that connects the two sides:

If the two planes are connected only at the power supply, the return current is forced to flow all the way back to the power-supply ground, which is a really big loop! Also, the analog and digital ground planes, which are at different RF potentials and connected with long wires, unfortunately form a very effective dipole antenna.

— TI, SLYT512

That loop does three things at once. It has inductance in proportion to its area, so the return current develops a voltage across it — the two halves of the plane are no longer one ground. It radiates, which is why SLYT499 reports that “during EMC tests, most problems are observed when traces are routed across a slot or a split in a ground or power plane.” And it picks up: “low-level analog currents flowing in large loops are susceptible to interference.” SLYT512’s fix, when a split cannot be avoided, is a bridge “at one location”, with all the crossing traces routed over it so each has “a return path directly underneath … producing a very small loop area”; its alternatives are an optoisolator, a transformer or “a true differential signal (where the signal flows down one trace and returns on the other trace with no ground needed for the return current)”. Its preference is no split at all.

MT-031 gives the same warning for systems with more than one board: where planes “become connected at several points”, that is “creating the possibility of ground loops and making a single-point ‘star’ ground system impossible” — and it puts a number on how much ground difference a system tolerates before the loop is more than a noise problem: “DC or ac voltages of more than 300 mV between the two ground planes can not only damage ICs but cause false triggering of logic gates and possible latchup.”

How to break one

A ground loop needs two things: more than one connection between two grounds, and a current that flows round the ring they make. Every fix removes one of them.

• One connection. The star of MT-031, the single bridge of SLYT512, the scope and the circuit on one outlet (Tek): if the two grounds meet once, there is no ring. On a board the one connection is a plane, which is every point at once with almost no impedance between them.
• No shared impedance. Where two circuits must share a return, make it wide and short — a plane, not a wire — so that the current one of them pushes through it drops nothing the other can see. MT-031’s 20 nH per inch is the number to beat.
• No ring at all. Isolation: an optocoupler, a transformer, an isolated supply, a differential pair that carries its own return. Tek’s isolation transformer and differential probe are the bench versions.
• Reject what the loop produces. A differential input rejects the common-mode voltage the loop puts on both lines; a common-mode choke raises the loop’s impedance at the frequencies that matter without breaking the DC path that safety requires.

What does not break a loop is a thicker ground wire. The problem is inductance, and doubling the diameter of a wire barely changes it; only shortening it, or replacing it with a plane, does.

Worked numbers

MT-031's wire        #22, 20 nH/inch
logic edge           10 mA/ns
1 inch shared        20 nH × 10 mA/ns                    = 200 mV     10 % of a 2 V signal
3 inches shared                                          = 600 mV
30 mA/ns edge, 1 in  20 nH × 30 mA/ns                    = 600 mV

MT-031's limit       > 300 mV between two ground planes: false triggering, latchup, damage
Tek's symptom        50/60 Hz and switching spikes on every trace, independent of the signal

The quantity to remember is the 200 mV per inch. A ground that is a wire is a resistor at DC and a signal source at every edge, and the current that makes it one is somebody else’s.