EMI & EMC Basics
15 min read
Quick Answer
Electromagnetic interference, EMI, is unwanted electrical energy that leaves one circuit and arrives in another. Electromagnetic compatibility, EMC, is the condition of emitting little enough and tolerating enough for equipment to coexist. Interference travels by conduction, by radiation, or by the electric and magnetic fields close to a conductor. The standard that applies in a market sets the limits.
Intuition
Nothing stays inside the box
A circuit does its work by moving charge about, and moving charge is what an electromagnetic field is made of. A current that changes puts a magnetic field into the space around its loop. A voltage that changes puts an electric field between whatever sits at one potential and whatever sits at another. Neither field is polite about the edge of the board. Some of it leaves, along the wires or straight through the air, and arrives somewhere it was not invited.
That arriving energy is electromagnetic interference, EMI. Designing so that a circuit emits little and tolerates what reaches it is electromagnetic compatibility, EMC. The two words describe one problem seen from either end, because a cable that radiates well also receives well and the physics does not know which way round the equipment is being used.
The surprising part is which signals cause the trouble. It comes from fast edges, which is not the same thing as a fast signal. An edge that takes 2.0 ns to switch needs a signal path good to roughly 175 MHz to arrive with its shape intact, and it will hand energy to anything nearby that responds up there, however slowly the signal itself repeats. A sluggish clock with sharp corners is a loud circuit.
Interference is the department's other kind of unwanted voltage, and it is worth keeping apart from the first. Noise is random and comes from inside the components. Interference is somebody else's signal, and it took a route you can trace and block.
Practitioner
What a designer can change
Interference reaches a victim circuit either along a conductor or through a field, and the field case splits again by which field is doing the work. Naming the route is most of a diagnosis.
A disturbance that travels on something the two circuits share, a power rail or a mains lead, is conducted. One that crosses open space as a wave and needs no connection at all is radiated. Between them sit the near-field mechanisms, which act over short distances through the fields themselves. Capacitive coupling works through the electric field between conductors at different potentials, so it grows with voltage swing. Inductive coupling works through the magnetic field around a current loop, so it grows with current swing and with the area that loop encloses.
Most of the terms in that description are fixed by the parts or by the physics. What a designer moves is the geometry of the current loop, which means the outward path and its return taken together, and the rate at which the edges change.
The relationship reads both ways. Given a bandwidth it returns the fastest edge that survives the path; given an edge, the same constant divided by the rise time returns the bandwidth the path has to have.
Worked example — A slow clock with sharp corners
A clock line repeats at 10 MHz, and its driver switches in 2.0 ns.
Carrying that edge intact takes a path good to 175 MHz, which is 17.5 times the rate at which the signal itself repeats. Whatever couples, couples up there.
Slowing an edge that has no reason to be fast is the cheapest fix in the subject. A series resistor at the driver will stretch it, and so will a slew-limited output. Take the same driver out to 10 ns and the requirement drops to 35 MHz. The emission spectrum comes down with it. The price is timing margin, and where the margin exists the fix is free.
Whatever survives the layout gets attacked physically, with a filter at the boundary or a shield around the source. Both are reported the same way: the ratio between what was measured without them and what is measured with them, on the decibel scale.
Worked example — Measuring what a shield is worth
An emission is measured at 500 µV with the enclosure open, and at 5.0 µV with it closed and properly bonded. Both are illustrative bench figures, not limits taken from any standard.
The shielding effectiveness is 40.0 dB.
The rest of the toolbox is short and it repeats across every design. Decoupling capacitors keep the fast return current in a small loop beside the chip that made it. A ferrite bead adds loss where a conductor leaves the board. Common-mode chokes work on the current that flows the same way along every conductor in a cable. A continuous return plane under a signal is worth more than any of them, because it lets each return current sit directly beneath its own outward path and shrinks the loop without anyone having to think about it.
Engineer
Reactance decides which path dominates
Both near-field mechanisms are ordinary circuit elements seen from an unusual angle. The capacitance between two conductors is a real capacitance, and the impedance it offers falls as the frequency rises:
Worked example — Two picofarads between neighbours
Conductors running alongside each other share a small mutual capacitance. Take 2.0 pF as an illustrative figure for a pair of adjacent board tracks, not a catalogue value.
At 1.0 MHz that capacitance presents 79.6 kΩ, high enough beside ordinary circuit impedances that almost nothing crosses.
At 175 MHz, the top of the band the earlier edge occupies, the same capacitance presents 455 Ω. The board is identical in both cases; only the frequency being asked about has moved.
The magnetic side behaves in the mirror image. Any conductor has inductance, and the impedance that inductance offers climbs with frequency:
Worked example — Ten nanohenries called a ground
A short piece of wire, or a via down to a plane, has inductance. Roughly a centimetre of it comes to about 10 nH, again as an illustrative figure.
At 175 MHz its reactance is 11.0 Ω, against a DC resistance of a few milliohms. Return current at that frequency develops a voltage across it, and everything referenced to the far end sees its own reference moving.
Taken together those two results explain why interference is a high-frequency subject even in slow equipment. A coupling capacitance that is an open circuit at audio becomes a usable path at a hundred megahertz; a piece of wire that is a dead short at DC becomes an impedance in the return. Frequency alone converts both. Ground loops is the low-frequency version of the second effect, where the culprit is resistance instead of reactance.
Every number above came out of a lumped model, and each of its assumptions gives way somewhere.
The rise-time constant belongs to a single-pole response. A repeating pulse train has no such spectrum: it has a line spectrum at multiples of its repetition rate, sitting under an envelope with more than one breakpoint, and harmonics is where that structure comes from. The frequency the constant returns marks where the energy is still substantial, not a ceiling above which there is none.
Treating a conductor as a lumped impedance holds while it is short against a wavelength. At 175 MHz a free-space wavelength is 1.71 m, and a twentieth of that is 0.0857 m. Texts put the threshold for efficient radiation somewhere between a twentieth and a tenth of a wavelength, and either way a cable of that order has stopped being a connection and started being an antenna. On a board the wave travels more slowly than in free space, so the physical length that matters is shorter still.
Close to the source, the electric and magnetic parts of the field are not yet in the fixed ratio a propagating wave carries, and the boundary conventionally sits about a sixth of a wavelength out. Inside it the two couple by different mechanisms and answer to different cures. A thin conductive wall reflects an electric field well and does very little against a low-frequency magnetic one, which yields to physical distance or to a high-permeability enclosure, and best of all to a smaller loop at the source.
Shielding effectiveness carries the same caveat. It is a measured figure under stated conditions rather than a property of the metal, belonging to one frequency, one field orientation and one enclosure, and the worst opening in that enclosure sets it.
Professional
What a conformity test asks, and when to ask it
Emission limits come from standards, and which standard applies depends on the market and on what kind of equipment is being sold. CISPR, a committee of the International Electrotechnical Commission, publishes measurement methods and limits used across much of the world. In the United States the Federal Communications Commission's Part 15 rules cover radio-frequency devices, including the unintentional radiators most digital equipment falls under. Both families draw a line between equipment meant for residential environments and equipment meant for industrial ones, and the residential class is the harder of the two to satisfy. The applicable standard states the frequency ranges, the measurement geometry, the detector and the limit lines. Nothing written here substitutes for reading the one that governs.
The emission side splits in two. Below a crossover frequency the standard names, the measurement is conducted: the disturbance is read on the mains lead through a line impedance stabilisation network, a fixture that presents a defined impedance so results repeat between laboratories. Above the crossover the measurement is radiated, taken with an antenna at a specified distance in a chamber or on an open test site.
Immunity is the other half of the certificate, and its methods run the same coupling paths backwards: electrostatic discharge into the enclosure and its controls, fast transient bursts on the cables, surges from switching and lightning, radiated fields from nearby transmitters, and disturbances injected directly into the leads. Severity levels are graded, and the standard that applies names them. Electrostatic discharge is the one a product meets long before it meets a laboratory, and a TVS diode at each exposed pin is the usual answer.
Most of the real work happens in pre-compliance, because chamber time is expensive and a failure discovered there has already cost the booking. A near-field probe and a spectrum analyser will rank the sources on a board inside an hour, and the ranking is the useful output even though the absolute numbers will not match a chamber. The oscilloscope is the other instrument, and its own bandwidth is part of every reading it gives: one specified at 500 MHz has a rise time of 0.70 ns, quick enough to show an edge of 2.0 ns without flattering it.
Shields fail at their openings and their penetrations long before they fail as metal.
Worked example — The seam that undoes the shield
The same enclosure is measured again with one seam left unbonded. The emission comes back at 50 µV, against the 500 µV it started from.
Shielding effectiveness is now 20.0 dB, where the properly closed box gave 40.0 dB.
A cable screen terminated by a short wire to the connector shell, the arrangement everyone calls a pigtail, does the same damage for the same reason: the wire puts an inductance in the one path that had to stay low-impedance. Bonding the screen to the shell around its whole circumference removes it. The argument carries over to slots in a panel, gaskets along a lid seam, and any connector that brings an unfiltered wire through the wall.
Filter parts that bridge a mains conductor to earth or across the line are safety-rated components in their own classes, and fitting an ordinary capacitor in either position creates a hazard, whatever its voltage rating says. Electrical safety covers why.
Leave room for the fixes before you need them. Footprints for a bead, a common-mode choke and a pair of capacitors cost a few square millimetres at layout and nothing at all if they go unpopulated, while adding them after a failed test costs a board revision and a second booking. Unpopulated filter footprints are the cheapest insurance on a board, and they are the first thing sacrificed when a layout gets tight.
Common mistakes
- Blaming the clock rate for a radiated emission — the spectrum reaches to roughly the reciprocal of the rise time, so a slow signal with sharp corners is the usual source and the driver is where to fix it.
- Fitting a shield before fixing the layout — an enclosure adds cost to every unit ever built, while a shortened return path costs nothing after the first board revision.
- Terminating a cable screen with a pigtail — that short wire is an inductance in the only path that mattered. Bond the screen to the connector shell around its full circumference.
- Treating emission and immunity as separate projects — they run the same coupling paths in opposite directions, so a change that helps one nearly always helps the other.
- Reading a pre-compliance scan as a pass — a bench setup is not the standard's setup, and the numbers move when cable positions, ground plane and distance change. Use it to rank problems.
- Expecting a conductive box to stop a magnetic field — thin conductive walls do very little against low-frequency magnetic coupling. Physical separation helps, high-permeability material helps more, and shrinking the loop at the source beats both.
Frequently asked questions
What is the difference between EMI and EMC?
EMI is the interference itself, the unwanted energy travelling from one circuit to another. EMC is the property of equipment that neither emits an unacceptable amount of it nor malfunctions when it arrives. One names the phenomenon; the other names the requirement a product has to meet.
Why do fast edges matter more than a high clock rate?
A signal's spectrum extends to roughly the reciprocal of its rise time, not of its repetition rate. A slow clock switched by a very fast driver has energy far above its own frequency, and that energy is what couples. Slowing edges that do not need to be fast is often the entire fix.
Will a metal enclosure fix a radiated emission problem?
Sometimes, and only if it is continuous. A shield is worth what its worst opening allows, so seams, ventilation slots and cable entries govern the result. A conductive wall is also weak against low-frequency magnetic fields, which pass through it almost unchanged.
Can EMC testing be done without a laboratory?
Pre-compliance work on a bench, with a near-field probe and a spectrum analyser, will find and rank the sources on a board and is worth doing early. It will not produce a conformity result, because the absolute numbers depend on a setup the standard specifies and a bench cannot reproduce.
Which standard applies to a given product?
That depends on the market it is sold into and on the category the equipment falls in, and the same product can face different limits in different regions. Look up the standard that governs the actual case; emission limits also differ between the residential and industrial classes, and the residential one is stricter.
Knowledge check
A signal path has to carry an edge of 1.0 ns. What bandwidth does it need? (Show answer)
Two adjacent tracks share 2.0 pF of mutual capacitance. Why does the crosstalk between them worsen as edges get faster? (Show answer)
A short lead of 10 nH is drawn on the schematic as a ground connection. When does that stop being true? (Show answer)
A series resistor slows an edge from 2.0 ns to 10 ns. What has that bought, and what did it cost? (Show answer)
A product passes its emission tests but locks up whenever a two-way radio is keyed beside it. Which half of EMC has failed, and why is the layout still the place to look? (Show answer)
References
- International Electrotechnical Commission, CISPR — International Special Committee on Radio Interference — the committee whose publications define measurement methods and disturbance limits used in much of the world.
- United States Federal Communications Commission, Title 47 of the Code of Federal Regulations, Part 15 — the rules covering radio-frequency devices, including unintentional radiators, in the United States.
- Bureau International des Poids et Mesures, The International System of Units (SI), 9th edition — the definition of the metre, which fixes the speed of light used for the wavelength in Layer 3.
- Henry W. Ott, Electromagnetic Compatibility Engineering, Wiley — coupling mechanisms, shielding practice and cable-screen termination.
- Clayton R. Paul, Introduction to Electromagnetic Compatibility, Wiley — the field theory behind the near-field and far-field distinction.