EMC Test Failures: 10 Common Root Causes and First Fixes
Ten root causes behind most EMC test failures, from converter hot loops and cable currents to ESD resets and RF demodulation, with symptoms and first fixes.
Most EMC test failures are not exotic. The same few design and preparation mistakes appear in lab report after lab report, whatever the product. This article collects ten of them, grouped into emission, immunity and process failures, with how each shows up in the test data and what to try first.
Emission failures
1. A switching converter with a large hot loop
The fast current loop of a buck or boost converter, made of the input capacitor and the two switches, carries edges of a few nanoseconds. If that loop is large, the converter fails twice: in conducted emission, as narrow lines at the switching frequency and its harmonics from 150 kHz upward, and in radiated emission, as a broad hump where the switch node rings. The ringing frequency follows f₀ = 1 / (2π·√(LC)); 1 nH of loop inductance with 1 nF of switch capacitance rings at about 159 MHz. First fix: put the input capacitor next to the switches, keep the switch-node copper small and separate the filter input from its output.
2. Common-mode current on cables
A cable radiates when common-mode current flows on it, and less than 10 µA on a cable about 1 m long can exceed CISPR 32 Class B. The current is driven by small voltages between the ground points where cables connect, for example return current crossing the plane between two connectors, or a connector shell that reaches the chassis only through a thin trace. First fix: group connectors on one edge, bond connector shells directly to the chassis and filter every line where it leaves the board.
3. Clock edges faster than the function needs
The harmonic envelope of a trapezoidal clock is flat up to 1/(πτ), falls at 20 dB per decade up to 1/(πt_r) and at 40 dB per decade above, where τ is the pulse width and t_r the rise time. A 1 ns edge puts the second corner at 318 MHz; slowing it to 3 ns moves the corner to 106 MHz. The envelope is unchanged below 106 MHz, 5.5 dB lower at 200 MHz and 9.5 dB lower above 318 MHz. First fix: a series resistor at the driver, lower drive strength, and spread spectrum where the interface allows it.
4. Interrupted return paths
High-frequency return current flows in the plane directly under its trace. When the trace crosses a split between power areas or a slot in the ground plane, or changes layers between two reference planes with no stitching via nearby, the return current detours, the loop grows and the gap is driven like a slot antenna. First fix: continuous reference planes under fast signals and a stitching via next to every layer change; on the current board, a capacitor across the split near the crossing can help.
5. Internal cables and floating metal
Flat flexible cables to displays and cameras, and heatsinks that are not bonded to ground, act as antennas inside the enclosure. A heatsink is coupled capacitively to the fast signals of the chip it cools, and at 1 GHz a quarter wavelength is only 75 mm, so a typical heatsink can resonate within the measured range. First fix: bond heatsinks to ground at several points, use flex cables with a ground layer or a shield bonded at both ends, and keep internal cables short and close to metal.
Immunity failures
6. ESD current that flows through the circuitry
An ESD pulse rises in under 1 ns, and the result depends less on where the spark lands than on where the current flows next. Typical failures are resets caused by a long, unfiltered reset or enable line, a connector shell bonded to logic ground through a thin trace instead of to the chassis, or a seam in a plastic enclosure that lets an air discharge reach a board edge. First fix: give the current a short path to the chassis before it reaches the circuit, place TVS diodes at the connector with a short ground connection, and add an RC filter on reset inputs.
7. RF demodulated by analog circuits
Radiated immunity (IEC 61000-4-3) and conducted immunity (IEC 61000-4-6, from 150 kHz to 80 MHz) use a carrier with 80 % amplitude modulation at 1 kHz. Junctions at op-amp inputs, sensor interfaces and audio stages rectify the RF, so the circuit sees a 1 kHz tone or a DC offset. The symptom is a reading that drifts only at certain frequencies, often where an attached cable resonates. First fix: filter at the input pins before the first active device, such as a series resistor or ferrite with a small capacitor to ground, and bond cable shields at the entry point.
8. Surge and burst protection that returns to the wrong place
In EFT/burst (IEC 61000-4-4) and surge (IEC 61000-4-5) tests the protection parts usually work; the path fails. A TVS diode or Y-capacitor connected through a long trace adds inductance, so the clamped voltage rises during the fast edge, and protection current returning through logic ground disturbs the circuit. First fix: return protection directly to chassis or protective earth at the entry point and keep the unfiltered side of every filter away from the filtered side.
9. Firmware that cannot recover
Immunity results are judged against performance criteria. In the generic standards, criterion A means normal operation within specification during the test, criterion B allows temporary degradation that recovers by itself, and criterion C allows temporary loss of function that recovers by itself or is restored by operator action. Common gaps are a watchdog serviced from a timer interrupt that keeps running while the main loop hangs, buses without timeouts or retries, and peripheral settings that are never refreshed. First fix: service the watchdog from the main loop, add timeouts and retries, and rewrite key registers periodically.
Process failures
10. A setup or build that differs from the design
Some failures have nothing to do with the design: the product runs in a mode that is not its worst case, the auxiliary laptop or power adapter produces the failing emission, or the cables are routed differently from the pre-compliance setup. Others appear after the design passed: an alternative regulator with faster edges, a second-source ferrite bead with a different impedance curve, a cable with a poorer shield termination, or a board built with a different stackup. First fix: write operating modes and cable layout into the test plan, measure auxiliary equipment on its own, and treat EMC-critical parts as controlled items.
Quick reference
| # | Root cause | Signature in the data | First check |
|---|---|---|---|
| 1 | Converter hot loop | Lines at the switching frequency and harmonics; hump at the ringing frequency | Input capacitor placement, switch-node area, filter routing |
| 2 | Cable common-mode current | Level changes with cable position or a clamp ferrite | Connector bonding to chassis, filtering at the board edge |
| 3 | Fast clock edges | Narrow lines at clock harmonics far above the fundamental | Rise time, drive strength, series termination |
| 4 | Interrupted return path | Near-field maximum at a plane gap, not at the source | Plane splits and layer changes under fast signals |
| 5 | Internal cables, floating metal | Peaks that move when a heatsink or flex cable is touched | Heatsink bonding, flex cable shielding |
| 6 | ESD current path | Reset or lockup at specific discharge points | Reset line filtering, connector shell bonding |
| 7 | RF demodulation | Drift or 1 kHz tone at certain frequencies | Filtering at analog input pins |
| 8 | Protection return path | Upset or damage although protection is fitted | Ground connection length of TVS and Y-capacitor |
| 9 | Firmware recovery | Product stops and needs a manual reset | Watchdog, bus timeouts, register refresh |
| 10 | Setup or build differences | Failure not seen in pre-compliance or earlier tests | Operating mode, auxiliary equipment, part substitutions |
Common mistakes
- Fixing the symptom at the test site. A clamp ferrite that passes the retest hides a cable current that can return with the production cable.
- Assuming one cause per failure. A failing line often combines two items, such as a clock harmonic carried out by a poorly bonded cable.
- Passing with no margin. A 1 dB pass can fail on the next unit or at another lab because of unit-to-unit spread and measurement uncertainty.
Calculators used in this post
More posts
Results and summaries are for reference. For certification and test reports use the latest official standard text and calibrated instrument data.