The filtered design that still fails

A power supply with a textbook input filter — measured 60 dB differential insertion loss at 100 MHz — fails CS114 at 90 MHz. The filter is not broken. The failure is happening on a path the filter does not see: the harness as a whole, carrying common-mode current that the injection probe deposits across all conductors simultaneously.

CS114 under MIL-STD-461H (released April 17, 2026) drives bulk current onto the interconnecting cable from 10 kHz to 200 MHz. The methodology was updated in 461H: the bulk-cable-injection calibration and limit/leveling procedure were revised to improve repeatability between labs. The forced current is common-mode, referenced to the ground plane, and your differential π-filter is essentially transparent to it.

The physics: the harness is a resonant antenna

A cable harness over a ground plane is a transmission line with a characteristic CM impedance of roughly 100–300 Ω. Drive it with a bulk-injection probe and it behaves as a resonant structure. Standing waves form at:

f_n = n·c / (2·ℓ·√ε_eff) (n = 1, 2, 3 …)

For a 1.5 m harness with ε_eff ≈ 1.2 (cable jacket loading), the first resonance is near 91 MHz — and that is exactly where CS114 failures concentrate. At resonance, the CM current builds a standing wave with current antinodes where the harness presents low impedance. At an antinode, even a small shield transfer impedance Z_T delivers a large coupled voltage:

V_coupled = Z_T · I_CM(antinode) · ℓ_eff

The cure is to spoil the resonance — add distributed loss along the line so the standing wave cannot build amplitude.

⚠️One ferrite at one end does nothing

A single ferrite at the connector sits at whatever point of the standing wave geometry dictates. If that point is a current node, the ferrite — which dissipates I²R — sees almost no current and does almost nothing. Distributed placement guarantees at least one core lands near an antinode regardless of frequency.

The solution: distributed sleeves at λ/8 spacing

Place ferrite sleeves along the harness at intervals of λ/8 at your highest frequency of concern. At 100 MHz, λ = 3 m, so λ/8 = 37.5 cm. A single core lumped at one end under-populates the harness and leaves resonances un-damped; λ/8 is the physics-based interval.

Each sleeve must present real impedance — resistive, not just reactive — at the problem frequency, because the goal is dissipation. Specify ≥ 300 Ω of impedance at the target frequency:

Fair-Rite 31 material (Mn-Zn/Ni-Zn hybrid) peaks at 200–300 MHz, delivering 250–350 Ω per pass through 25–300 MHz. Ideal for CS114's upper band.
Fair-Rite 43 material peaks lower and suits the 10–50 MHz range.
Two passes (N = 2) quadruple impedance: Z ∝ N².

Worked example: a 1.5 m harness for 10–200 MHz

Highest concern: 200 MHz → λ = 1.5 m → λ/8 = 18.75 cm. A 1.5 m harness needs 8 sleeves.
Material: 31 material for the upper band; one 43-material core near the source for 10–50 MHz content.
Verify per-core impedance: a snap-on 31-material core, single pass, ≈ 280 Ω at 150 MHz. Two passes on the two cores nearest the connector → 1.1 kΩ each, anchoring the most exposed length.
Total distributed CM impedance across the harness: > 2 kΩ — enough to drop the standing-wave Q below the point where any antinode threatens Z_T coupling.

POCONS connection

Ferrite sleeves manage current along the cable. They cannot fix what happens where the cable enters the box. If the harness shield is pigtailed at the enclosure, that 15–25 nH of inductance reintroduces the coupling the sleeves just removed. A POCONS custom shield with an integrated 360° cable-entry gland terminates the braid directly to the can wall — zero pigtail — so the sleeve network and the shield work as one continuous low-impedance system.

One thing

300 Ω of resistive ferrite impedance at your problem frequency, distributed every λ/8 along the harness — anything less, or anything lumped at one end, is expensive decoration. CS114 is a standing-wave problem; solve it with distributed loss.