The margin you think you have

Engineers internalize FCC Part 15 Subpart B as "Class B, 3 meters, you'll be fine." Then a 2.1 MHz buck converter's 480th harmonic lands at 1.008 GHz and the EUT exceeds the limit by 4 dB. The problem is not noise — it is that the limit and your shield both work against you above 1 GHz.

47 CFR §15.109(a) sets Class B radiated field-strength limits. Above 960 MHz the average-detector limit is 54 dBµV/m at 3 m, with a quasi-peak component, and CISPR 32-aligned measurement now extends the certification band to 6 GHz. There is no relief above 1 GHz — the ceiling stays flat while your emission profile and your enclosure both deteriorate.

ℹ️Why above 960 MHz is the trap

Below 960 MHz the Class B limit steps with frequency. Above it, the limit is constant — but radiated coupling efficiency from a PCB rises with frequency (a 1 cm trace is an increasingly efficient antenna as λ shrinks), and aperture leakage worsens. You are climbing a down escalator.

The physics: 6 dB per octave, twice

Two independent rolloffs erode high-frequency margin:

Aperture leakage. A slot of length L radiates with shielding effectiveness:

SE_aperture = 20·log₁₀(λ / 2L)

Every octave of frequency halves λ, costing 6 dB of SE per octave. A 30 mm seam giving 26 dB SE at 500 MHz gives only 20 dB at 1 GHz and 14 dB at 2 GHz.

Common-mode cable radiation. A cable carrying CM current I_CM radiates a field that scales linearly with frequency:

E ≈ (1.26×10⁻⁶ · f · I_CM · L) / r (V/m, free space, short-cable approximation)

Double the frequency, double the field — another effective 6 dB/octave.

Now layer on the harmonic content. A trapezoidal switching waveform with rise time t_r has a spectral envelope that is flat to f₁ = 1/(π·t_d), rolls at −20 dB/decade to f₂ = 1/(π·t_r), then −40 dB/decade. Modern GaN switchers with t_r ≈ 2 ns put the second knee at ≈ 159 MHz — meaning harmonic energy is still falling at only 20 dB/decade as it passes through 1 GHz. There is real spectral content exactly where the limit is flat and the shield is leaking.

The solution: budget the suppression before layout

Two strategies, in priority order:

Keep the switching fundamental low enough that the harmonic of concern is attenuated by waveform shape. If practical, hold f_switch ≤ 200 kHz so that by 1 GHz you are 3.5 decades past the second knee — 70+ dB of natural rolloff.
Where high f_switch is mandatory (GaN power density, fast transient response), budget 20–30 dB of explicit suppression at 1–6 GHz: ferrite bead loading on every cable penetration, an unbroken copper pour with via stitching at ≤ λ/10, and a board-level shield over the switching node.

Field case: the 1.008 GHz exceedance

A consumer SSD controller exceeded Class B by 4 dB at 1.008 GHz — the 480th harmonic of a 2.1 MHz buck. Lowering f_switch was off the table (transient-response spec). The fix: a POCONS custom board-level shield, a single-piece can with a soldered fence, placed over the switching node and inductor. Measured SE at 1 GHz: 38 dB. Result moved from +4 dB to −12 dB versus limit — 16 dB of headroom from one stamped part.

The economics matter. Above 960 MHz, every dB of margin is bought with copper, ferrite, or a redesign spin. A board-level shield converts a layout-and-stackup problem — weeks of spins — into a bill-of-materials line item.

One thing

Above 960 MHz the limit stops helping you and your enclosure starts hurting you. Design switcher harmonics, cable filtering, and shielding for 6 GHz from day one — retrofitting margin at 1 GHz is the most expensive dB in the BOM.