EMC failure triage

How this works

Everything is a harmonic of something

f_n = n · f_clock

A square wave is a fundamental plus odd harmonics. Real digital signals are not symmetric, so even harmonics appear too. Given a failing frequency and a list of clocks, finding the source is division: divide the failure by each clock and see which gives something close to a whole number. This tool does that across every clock, every switcher, and every sum and difference of clock pairs, then ranks by how exact the fit is and how plausible the harmonic order.

The spectral envelope

corner 1:  f = 1 / (π · T)        20 dB/decade above this
corner 2:  f = 1 / (π · t_rise)   40 dB/decade above this

Amplitude falls with harmonic order, but the second corner is set by edge rate, not clock rate. A 25 MHz clock with a 1 ns edge has significant content past 300 MHz; the same clock with a 5 ns edge does not. This is why a series resistor at the driver — deliberately slowing an edge that never needed to be fast — is often the cheapest fix available, and why a slow bus can fail while a fast one passes.

Why the cable is usually the antenna

λ = 300 / f(MHz)   meters

efficient radiator ≈ λ/4, and λ/20 still radiates usefully

Below a gigahertz a board is electrically small and radiates poorly. Any attached cable is not small — and common-mode current on it is what the chamber measures. The failure frequency tells you which cable to suspect: at 143 MHz a quarter wave is about 520 mm, so a half-meter harness is close to ideal.

The diagnostic is simple and free. Remove cables one at a time and re-scan. When the peak drops, that cable is the antenna — and the fix is common-mode current, not shielding the board.

Spread spectrum, and why it hides things

Many switching regulators dither their frequency to spread energy across a band. Peak measurement drops several decibels, which is genuinely useful for passing a quasi-peak limit. But the energy is still there, spread into a hump rather than a spike — and it can desensitise a receiver on the same board. If your scan shows a broad raised region rather than a line, look for a spread-spectrum converter.

The fixes, roughly in order of what works

1. Fix the return path. A signal crossing a plane split makes a loop, and the loop is the antenna. This is the most common root cause and costs nothing to fix at layout.
2. Common-mode chokes on the offending cable. Cheap, effective, and addresses the actual radiator.
3. Slow the edge. A series resistor on a signal that does not need speed removes the high harmonics entirely.
4. Terminate shields properly. A pigtail is an inductor; a 360° termination at the connector shell is not.
5. Local decoupling at the source. Reduces the current loop feeding the harmonic in the first place.
6. Ferrite on the cable as a last-resort field fix. It works, but if it is load-bearing in production your design has a problem you have not found.

Where this stops being accurate

• Identifying a harmonic identifies a candidate, not a cause. Two clocks frequently produce coincident harmonics.
• Limit lines here are the common EN 55032 radiated values. Read the actual standard for your product family, detector type and distance — and note that quasi-peak and average limits differ.
• Below 30 MHz radiated emissions are generally not assessed; conducted emissions apply instead, with different limits and a different setup.
• Above 1 GHz the limits, detectors and measurement distance all change.
• Pre-compliance measurements routinely under-read a full chamber. Treat margin under 6 dB as a fail.

Related: EMC design support · design review · crosstalk estimator

Design guidance, not a manufacturing instruction. Verify every result against your fabricator's stack-up and the applicable standard before release. Closed-form models are approximations; the fab's field solver and process window are the authority.