Fixing radiated emissions in an existing enclosure

A failed radiated-emissions test is a containment problem: the RF is already inside the box, and your job is to close the doors it uses to get out. Start from two numbers. Any seam, window or hole leaks hard once its longest side passes 1/20 of the failing wavelength — 24 in [600 mm] at 25 MHz, 6 in [150 mm] at 100 MHz. And 10 µA of common-mode current on a 1 m cable sits about 6 dB over the CISPR 32 Class B limit at 100 MHz. Work in this order: clamp a ferrite on every cable leaving the box, break the longest seams with conductive gasket or closer fasteners, and swap a plastic display window for a meshed or coated one. Most stubborn boxes pass with those three moves. The sections below show how to find which door is open and what each fix buys.

What do the emissions peaks tell you?

Each peak names its maker: a carrier at the clock frequency with harmonics (25, 50, 75, 125 MHz for a 25 MHz crystal) means a cable or aperture is carrying clock energy out, while a comb spaced at the switching-regulator frequency means the board itself is the antenna.

Read the scan against what the board contains, then against what leaves it. Frequency plus attachment points to the door to close first:

• Clock peak that grows when cables attach. A 25 MHz crystal shows at 25 MHz and its harmonics. Unplug the CAN or RS-232 lead; if the peak drops 10 dB, that cable is the antenna.
• Clock peak that stays with every cable off. The energy exits through an aperture: display window, lid seam, an unused gland hole.
• Broadband smear across 30-300 MHz. Peaks spaced at the switching frequency accuse the regulator. A comb at 45 MHz and its multiples accuses a memory-card clock the same way.
• A peak that wobbles when you press the lid. A joint is making and breaking contact; that is a seam fault, not a board fault.
• Level follows cable dress. Rerouting one lead moves a peak by 6 dB or more; common-mode current on that lead is carrying it.

Is it the cables or the enclosure leaking?

Measure before you guess: at 100 MHz, 10 µA of common-mode current on a 1 m cable radiates about 46 dBµV/m at 3 m, roughly 6 dB over the CISPR 32 Class B limit of 40 dBµV/m, so cables are guilty in most failed boxes.

The figure comes from the small-dipole formula, E = 60π·I·l/(λ·r), with λ = 3 m at 100 MHz, a 1 m cable and a 3 m measurement distance:

E = 60π × 0.00001 × 1 / (3 × 3) ≈ 2.1e-4 V/m ≈ 210 µV/m ≈ 46 dBµV/m

The model ignores the ground plane, so treat it as a prescan estimate good to a few dB. Halve the current and you buy 6 dB; the passing target works out near 5 µA per cable.

Work the exits in this order:

1. Common-mode current on the leads. Unshielded power and GPIO leads are the usual carriers, and a shielded cable with a pigtail shield behaves the same above 100 MHz.
2. Shield terminations. A 2 in [50 mm] pigtail is roughly 20 nH by the 10 nH/in wire rule, about 16 Ω at 125 MHz — enough to float the shield at the exact harmonics you fail.
3. The display window. A plastic window is a plain hole; the table below sets its limit.
4. Long seams and the lid rail. Painted contact surfaces, a missing gasket and wide screw pitch turn the flange into a slot antenna.
5. Spare openings. A gland hole behind a plastic plug leaks from the first megahertz; fit metal plugs or gland it properly.

The aperture arithmetic is the same one behind how much RF leaks through enclosure holes and seams. The same common-mode current shows up on the floor as inverter trips near a metal enclosure.

Size every opening against the frequency you fail at, keeping the longest dimension under 1/20 of a wavelength:

A slot leaks by its longest side, not its area: a 6 in × 0.02 in gap under the lid radiates almost like a 6 in round hole. Break any seam longer than the table allows with a conductive gasket or added fasteners.

How do you find the leak without an anechoic chamber?

A near-field probe kit and an RF current clamp find the guilty path in an afternoon: sniff the seams for the strongest field, then clamp each cable and flag anything over about 5 µA at the failing frequency as an antenna.

Rent an hour of bench time with the gear from most test houses; it is cheaper than a second chamber booking. Then run three checks in order:

1. Sniff the seams. Hold the loop probe against the window edge, the lid rail and the gland plate. The loudest point is the exit; the board under it is the source.
2. Clamp every cable. Measure at the failing frequency. Anything above about 5 µA there is an antenna and gets a ferrite first.
3. Close one door at a time. Tape foil over the window, bridge one seam with copper foil, unplug one cable. Whatever drops the peak by 10 dB was carrying it.

Lock out and confirm dead before opening a panel that is still wired to mains; the probes keep for five minutes. Current-clamp readings track chamber results within a few dB, which is plenty for ranking fixes.

How much do clamp-on ferrites cut the peaks?

Plan on 6-10 dB from one pass of a clamp-on core and 10-15 dB from two turns, with typical nickel-zinc cores showing 200-300 Ω resistive impedance between 50 and 300 MHz in their datasheets.

The core adds series impedance to the common-mode loop, so the gain depends on the loop it fights. A loop at 150-400 Ω with a 250 Ω core in it loses current in proportion — realistically that 6-10 dB. Two turns around a wide core quadruple the impedance below self-resonance and add 3-6 dB, with the useful peak sliding down by roughly an octave. Fit the core at the wall exit; either side of the wall works, because the ferrite only sees the loop. Pick material from the datasheet curve at the failing frequency: broadband EMC cores hold 200-300 Ω from tens of megahertz to 1 GHz, while a 25-45 MHz peak needs a curve still steep down there.

Where the board is still open to change, a 22-33 Ω series resistor in each memory-card data line slows the edges driving the 45 MHz comb. A capacitor alone on those lines often makes the peak worse; the resistor damps instead of tuning.

How do you seal a display window without losing the view?

Swap the plastic window for a conductive-mesh or coated-film window gasketed into a bare-metal frame: a mesh with 1 mm openings stays under the λ/20 limit up to 15 GHz, and the perimeter bond, not the mesh, decides the shielding.

Three builds exist, in rising order of shielding: a transparent conductive film laminated to the window, a wire mesh laminated between glass or acrylic, and a mesh framed for bolt-over mounting on the cutout. Film shields less, with tens of dB set by its surface resistance; mesh openings never set the limit below a few gigahertz. The frame land is the weak point. Run a continuous conductive gasket around it, and treat one painted or anodized spot in that land as a hole the size of the whole window.

Do the arithmetic before spending. A scan failing at 125 MHz, the fifth harmonic of a 25 MHz crystal, gives λ = 300/125 = 2.4 m, so the limit is 2.4/20 = 120 mm [4.7 in]. An 8 in [203 mm] wide plastic window is nearly double that and radiates even in a steel box. The same cutout with 1 mm mesh sits 120 times under the limit, and the frame gasket now decides the result.

The geometry matches the water side of window sealing; the compression and fastener checks behind a leak-free frame are the ones in display window gasket and fastener checks.

What makes a seam leak after the lid is back on?

Two things: coatings and fastener pitch. A lid rail under powder coat measures kilohms instead of milliohms at RF, and bare screws 6 in apart leave a 6 in slot, over the λ/20 limit for every frequency above 25 MHz.

Mask or scrape a continuous contact rail about 0.25 in [6 mm] wide on both mating faces, bright metal on bright metal. Getting that contact through a coating is the same job as in grounding a powder-coated aluminum enclosure. With bare rails alone, the fasteners set the seam length: against a 125 MHz failure, screws every 4.7 in [120 mm] keep each gap under λ/20.

A conductive gasket replaces that pitch arithmetic with a continuous seam. Fabric-over-foam, conductive elastomer and knitted mesh all work on stainless. Compress fabric-over-foam to 20-30% of free height, and give the flange 0.5 in [12 mm] of land to carry a 0.09 in [2.3 mm] profile. Flatness and the fastener pattern hold that compression evenly, the subject of enclosure cover flatness. Do not count on spring fingers spaced inches apart unless their own pitch passes the same λ/20 test.

How do you keep the fix passing at the retest?

Book the chamber only when your own numbers say pass: common-mode under 5 µA on every cable, loop-probe readings along every seam at the noise floor, and 6 dB of margin in the prescan, because labs vary 3-4 dB between sites and setups.

Re-run both bench numbers after each fix and log them: current clamp per cable, loop probe per seam. When both are quiet, add headroom. Walk into the retest with 6 dB of margin, not 1 dB, and the result survives a different antenna position or ground-plane height.

Write the fixes into the build record so the next unit repeats them: ferrite part and turns count, gasket profile, masked rail positions, mesh window part number. Torque-check the lid fasteners after the first service cycle, because a gasket squeezed past 40% takes a set and will not recover. Leave lossy absorber sheets for one stubborn resonance: they add 5-10 dB there, but they cost per panel and they insulate, trading an emissions margin for a heat problem.

When is a new enclosure the cheaper fix?

Re-spec the box when the band-aids cost more assembly time than a purpose-built body, or when the walls have no flat gasket land: a custom body with mesh-window lands, 360° shield entries and a gasketed flange is delivered in 10 business days from your approval of the drawing.

Add the per-unit labor first. A ferrite on each of four cables, foil on a window and gasket strips on a lid can pass today and haunt production for years. A body built for the job ends that: a flange wide enough for the gasket, a mesh window with its own gasket land, bulkhead gland positions where every shield lands 360°, and drawn metal plugs for the spare holes.

Lay the window land and gland positions out in the cutout designer, check opening options against the cutouts, glands and HMI windows guide, and send the layout with the quote for custom stainless enclosures or a fully custom body.

Catalog dimensions to compare

Compare these catalog dimensions with your component layout. Open each enclosure for its rating, material and usable space, then check those specifications against your application.

Questions