Radiated Emissions Failure? A Step-by-Step Debugging Order

A practical order for debugging a radiated emissions failure: read the harmonic pattern, find the radiating cable, locate the source with near-field probes.

Published: 2026-11-07 · Last updated: 2026-11-20

The lab report says the product failed radiated emission by 6 dB at 125 MHz, and the retest is booked in two weeks. The tempting response is to add every ferrite and gasket in the drawer and hope. A fixed order works better: read the chamber data, reproduce the problem on the bench, find the structure that radiates, find the source that drives it, and then fix and verify with relative measurements.

Step 1: Read the chamber data before touching the product

For each failing frequency, write down the level, the margin, the detector (peak or quasi-peak), the antenna polarization and height, and the turntable angle. Ask the lab for the full peak scan, not only the final quasi-peak table, and for photos of the setup showing how every cable was routed. The angle and polarization tell you which side of the product and which cable orientation produced the maximum, and the photos let you rebuild the same layout on the bench.

Then check the detector. Quasi-peak is never higher than peak, and for a steady clock harmonic the two are almost equal, so a narrowband line that fails on peak will usually fail on quasi-peak too. Only impulsive, low-repetition noise reads much lower on quasi-peak. Also check whether a line is narrowband or broadband by changing the resolution bandwidth: a single clock harmonic keeps the same level, while broadband noise rises or falls with the bandwidth.

Reading the harmonic pattern

The spacing between failing lines is often the fastest clue. Suppose the scan shows lines at 125, 175 and 225 MHz. They are 50 MHz apart but are not multiples of 50 MHz; they are the 5th, 7th and 9th harmonics of 25 MHz. Missing even harmonics point to a square wave with a duty cycle close to 50 %, so the likely sources are a 25 MHz crystal, oscillator or clock output and the traces it drives. List every clock, PLL output and switching frequency on the board and divide each failing frequency by them; the candidate that divides all of them is usually the source.

Spectral signatureLikely sourceFirst check
Narrow lines at odd multiples of one frequencyClock or oscillator with a duty cycle of about 50 %Find the clock; check its edge rate, series termination and return path
Narrow lines at every multiple, even and oddClock with an unequal duty cycle, or a periodic data busCompare idle and active modes; look at bus and memory lines
Lines that widen in proportion to frequencySpread-spectrum clock; the deviation scales with the harmonic numberConfirm the spread setting; a 0.5 % spread is 0.625 MHz wide at 125 MHz
Broad hump over tens of MHz that changes with loadSwitching converter edges and switch-node ringingProbe the switch node; check the hot loop and snubber; compare light and full load
Peak that moves when a cable is lengthened or reroutedCable resonance; the cable is the antennaTry a clamp ferrite at the EUT end and a different cable length
Lines that appear only with data trafficHigh-speed interface such as USB, HDMI or EthernetTest with the link idle, then active; inspect the connector shield bonding

Step 2: Reproduce the failure on the bench

You do not need a chamber to make progress, only a repeatable relative measurement. Set the product up on a table with its cables in the same layout as the test, and clamp a current probe around each cable in turn, a few centimeters from the connector. Use a spectrum analyzer in max hold over the same frequency span as the failure. Convert each reading to common-mode current with the probe's transfer impedance: I (dBµA) = V (dBµV) − Z_T (dBΩ). The cable that carries the most current at the failing frequency is the first suspect. Cable radiation is very sensitive: as our earlier article on cable current worked out, a few microamps of common-mode current on a cable around 1 m long can exceed the CISPR 32 Class B limit.

Mark the probe positions with tape and keep the cable layout fixed, because moving a cable by a few centimeters can change the reading by several dB. Before trusting any line, switch the product off and look again: FM broadcast stations around 88 to 108 MHz, television and mobile signals all appear on a bench spectrum.

Step 3: Find the antenna

Below roughly 300 MHz, small products rarely radiate efficiently from the board itself; the cables do. At 100 MHz the wavelength is 3.0 m, so a 0.75 m cable is a quarter wavelength and a 1.5 m cable is a half wavelength, both efficient antennas. The test is simple: remove cables one at a time where the product can still operate, replace them with battery power or a filtered connection, or put a clamp ferrite on each cable at the product end. The cable whose removal or ferrite drops the failing line by several dB is the antenna.

Be careful how you read the result. A cable that carries no signal at the failing frequency can still radiate if it is connected to a ground point whose potential differs from another cable's ground point. Return current flowing through the inductance of the ground plane between two connectors creates a small voltage, and a few millivolts between two cables on opposite sides of a board can drive enough common-mode current into them. In that case the cure is on the board, not on the cable. If the emission does not change when every cable is removed or choked, look at the enclosure: slots, seams, unbonded heatsinks and display cables inside the product.

Step 4: Locate the source with near-field probes

With the antenna identified, trace back to the source. A loop (H-field) probe responds to current and finds traces, loops and switch nodes that carry high-frequency current; hold its loop perpendicular to the board and parallel to the trace so the magnetic field passes through it. An E-field probe responds to voltage and finds high-dV/dt nodes such as switch nodes, heatsinks and floating metal. Work from the suspect connector inward: its pins, its ground connection, the filter in front of it, and the circuit that drives it.

Near-field readings are relative. A probe close to a compact, well-returned loop can show a strong signal that radiates very little, while a weak signal on a ground pin next to a cable connector can be the real problem. Compare levels only at the same frequency, with the same probe, at the same position and height.

Step 5: Fix one thing at a time and verify

Change one thing at a time and record the change in dB at the same bench positions. Try the cheap measures first: in firmware, turn off unused clocks and outputs, reduce GPIO drive strength, or enable spread spectrum where the interface allows it. On the board, add or increase a series resistor on the clock line, add a ferrite bead or filter at the connector, and bond the connector shield or ground directly to the chassis. Layout changes and new enclosure parts come last. Aim for a bench reduction larger than the chamber overshoot, because bench and chamber do not correlate exactly. After the fix, check the neighboring harmonics as well; when the worst line comes down, the next one often becomes the new worst case.

Common mistakes

  • Changing several things at once. If the line drops, you do not know which change helped, and you cannot remove the parts that did nothing.
  • Comparing absolute near-field levels with the chamber limit. A probe reading in dBµV has no fixed relationship to dBµV/m at 3 m or 10 m.
  • Debugging in a different operating mode. Use the same firmware, data traffic and load as in the chamber, or the failing line may not even be present.
  • Treating a ferrite that works on the bench as the solution without checking the production cable and its routing, which can shift the resonance.
The harmonic and wavelength figures in this article are worked examples. Near-field and current-probe measurements guide debugging but do not replace a compliance measurement at an accredited test site.

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.