A clean CISPR scan is not a clean design

A multimedia product passes CISPR 32 conducted emissions with margin, ships, and field-fails intermittent radiated emissions at 1 GHz. The conducted scan never showed it — because CISPR 32 conducted measurement stops at 30 MHz, and the resonance lived at 870 MHz. The energy was always there; the commercial test simply does not look that high on the conducted port.

The MIL-STD conducted-emissions setup looks further, and that makes it a powerful diagnostic even for commercial products.

The frequency-window mismatch

CISPR 32 (the harmonized successor to CISPR 22 for multimedia equipment) tests conducted emissions 150 kHz–30 MHz and radiated emissions to 6 GHz in current editions. MIL-STD-461H CE102 (released April 17, 2026) tests conducted emissions on power leads 10 kHz–10 MHz, and the broader RE102 radiated requirement under 461H now extends coverage to 40 GHz for select equipment classes — a notable expansion over Rev G's 18 GHz ceiling.

The gap that matters: there is a 30 MHz–1 GHz band where energy is conducted onto cables (and re-radiates) but no commercial conducted test observes it. A switching regulator can be quiet at its fundamental and low harmonics yet exhibit a sharp resonance at 800 MHz–1 GHz — typically the interaction of the switch-node capacitance with package and layout inductance, or a power-loop resonance.

💡Use the MIL setup as a diagnostic

Even for a commercial product, run a conducted-emissions sweep with the MIL-STD-461H LISN/current-probe setup across the full 10 kHz–1 GHz range during development. It will surface the 800 MHz–1 GHz resonances that CISPR 32's 30 MHz conducted ceiling hides — before they become a field return.

Reading the delta

Compare your measured conducted spectrum against the commercial limit extrapolated past 30 MHz. A measured level that exceeds the extrapolated commercial limit by more than 6 dB at any frequency 30 MHz–1 GHz is a flag. Here is why 6 dB is the threshold:

A switching converter's fundamental drifts with temperature — typically ±5 to ±15% over −40 to +85 °C for an RC-timed oscillator. A resonance has finite Q; if the fundamental shifts so a harmonic walks into a resonant peak, the emission at that frequency can rise sharply. A 6 dB margin at room temperature can evaporate at temperature extremes. Treat > 6 dB over the extrapolated limit as a probable failure at corner conditions, and 0–6 dB as "marginal — fix it."

The physics of the 800 MHz–1 GHz resonance

The switch node sees a loop: high-side switch output capacitance C_oss, the power-loop inductance L_loop (the physical area enclosed by input cap → switches → output), and parasitic trace inductance. This forms a resonant tank:

f_ring = 1 / (2π·√(L_loop·C_oss))

For L_loop = 2 nH and C_oss = 15 pF, f_ring = 919 MHz. The switching edge rings this tank every cycle; the ring decays with the loop's Q. Minimize L_loop (tight power loop, vias close, smallest enclosed area) to push f_ring up and out, or add a small RC snubber across the switch node to damp Q below 2.

CS115 and the medical crossover

CS115 bulk-cable injection (10 kHz–200 MHz) provides immunity data directly relevant to IEC 60601-1-2 Edition 4.1 medical-device requirements, which mandate that clinical function be maintained — not merely that the device not be destroyed — during EMI exposure. A device qualified to the MIL setup gives you immunity headroom that maps cleanly onto the medical environment's 3–10 V/m field requirements.

POCONS connection

When the 800 MHz–1 GHz switch-node resonance is real and the power-loop is already minimized, the residual radiator is the switch node itself. A POCONS custom board-level shield over the regulator caps the near-field coupling so the resonance — even if it still rings — cannot reach a cable or an aperture. It converts a "respin the power loop" problem into a stamped-part fix.

One thing

A clean CISPR 32 conducted scan only proves you are clean to 30 MHz. Run the MIL-STD-461H setup to 1 GHz in development; treat any point > 6 dB over the extrapolated commercial limit as a temperature-corner failure waiting to happen.