The Design Brief #010 — CISPR 32 to 6 GHz: Redesigning Emissions Strategy When the Test Band Triples

When the test band triples, the strategy breaks

For two decades, multimedia-equipment emissions strategy was built for a 1 GHz radiated ceiling. CISPR 32 now specifies radiated-disturbance measurement to 6 GHz for equipment with internal sources above 108 MHz — and every mitigation technique sized for 1 GHz has to be re-derived, because shielding effectiveness, aperture leakage, and cable coupling all behave differently when λ shrinks from 30 cm to 5 cm.

The physics at 6 GHz

Aperture leakage. Shielding effectiveness of a slot is SE = 20·log₁₀(λ/2L). At 1 GHz a 15 mm seam gives 20 dB. At 6 GHz the same seam gives 20·log₁₀(50 mm / 30 mm) = 4.4 dB — effectively no shielding. The seam length that was acceptable at 1 GHz is a wide-open window at 6 GHz. The maximum tolerable slot for 20 dB SE at 6 GHz is just 2.5 mm.

Waveguide-below-cutoff behavior. Any opening behaves as a waveguide; below its cutoff frequency it attenuates strongly. Cutoff for a rectangular aperture of width w is f_c = c/(2w). For f_c = 6 GHz, w must be ≤ 25 mm — but real attenuation requires operating well below cutoff, so honeycomb vents and small-diameter perforations become mandatory, not optional, on 6 GHz-rated enclosures.

Cable resonances. A cable becomes an efficient antenna near λ/2. At 6 GHz, λ/2 = 25 mm — almost any unmanaged pigtail, connector tail, or internal flex resonates. Common-mode cable radiation dominates the 1–6 GHz emissions profile.

The conducted side: the 9–30 MHz squeeze

While the radiated band tripled, the conducted band did not move — but the failures did. As switching frequencies rise, more harmonic energy lands in 9–30 MHz, and that is where Class B headroom is now thinnest. Recall FCC Part 15 / CISPR 32 Class B is ~10 dB stricter than Class A. The practical guidance: where possible, hold the switching fundamental below 150 kHz so its harmonics in the conducted band are few and well-attenuated, and never assume that 10 dB of margin at 30 MHz implies anything about 1–6 GHz behavior — the radiated band is governed by entirely different coupling.

FCC enhanced review: AI accelerator cards

A current regulatory development for 2026: FCC Part 15 Subpart B now places AI-accelerator GPU cards under enhanced review. These boards combine multi-hundred-amp core rails, sub-nanosecond switching edges, and dense high-speed SerDes — a worst-case emissions source — and they frequently install into host chassis whose apertures were never characterized to 6 GHz. Expect closer scrutiny of both the card's own emissions and the card-plus-host system.

Worked example: re-qualifying a 1 GHz design to 6 GHz

A media-gateway box passing CISPR 32 at 1 GHz fails at 4.8 GHz:

1. Diagnose: near-field scan localizes a 12 mm lid seam radiating at 4.8 GHz (SE there ≈ 6 dB).
2. Reject the obvious: a continuous gasket on the lid is the textbook answer but adds assembly cost and a service-access penalty.
3. Source-level fix instead: a POCONS board-level shield over the SerDes and clock domains contains the 4.8 GHz energy at its origin, so the leaky lid seam no longer has anything to leak. Measured: 31 dB SE at the shield, system passes at 4.8 GHz with 14 dB margin — no enclosure rework.

POCONS connection

At 6 GHz the cheapest decibel of shielding is the one applied closest to the source, because the source occupies a few square centimeters while the enclosure has square decimeters of seam to police. A stamped board-level shield — engineered for the specific component footprint, with a fence soldered to the ground pour — contains the emission before it ever reaches an aperture. For 6 GHz CISPR 32 compliance, source-level shielding is no longer optional.

One thing

CISPR 32 to 6 GHz is a new test, not a tighter old one. A 15 mm seam that gave 20 dB at 1 GHz gives 4 dB at 6 GHz. Shield at the source, characterize every aperture below cutoff, and never read conducted margin as radiated margin.