Short answer: give the module its own board with a continuous ground plane, shield it with at least 0.3 mm [0.012 in] of copper — about three skin depths at the switching frequency — hold it 1 in [25 mm] or more from the analog board, and bond the shield to chassis ground at one point away from the signal star ground. The copper figure assumes a 500 kHz converter of the dual-output buck-and-inverting-regulator type. Skin depth scales with the square root of frequency, so the same can covers the harmonics even better than the fundamental. A floating shield re-radiates what it collects. A shield wired into the audio ground turns your quietest node into the switcher's return path. The sections below carry the arithmetic, the build order and the checks that show the noise floor stayed put.
Does the converter really need a shield?
Run the steel-box test before you cut any copper. Close a steel lid over the powered converter and listen. If the preamp quiets, the switching pickup is radiated and a shield will work. If the noise survives unchanged, it is arriving on shared supply or ground leads.
At the fundamental the wavelength is 600 m, so everything inside a small box sits deep in the near field. A steel lid damps that near field by eddy action, which makes the test a clean radiated-versus-conducted decision. Noise that survives a closed steel box is riding in on the input lead, the output leads or a shared ground return. Copper inside the box cannot touch that path. Twist the lead pairs, fit split ferrites and separate the returns before you blame radiation.
Scope both rails into a dummy load and write the baseline down. Single-digit millivolts of switching ripple at rated load is a healthy small converter. Any 100 or 120 Hz component on the rails is a wiring or supply problem to clear first.
How thick does the copper shield have to be?
Make the copper shield at least 0.3 mm [0.012 in] thick — about three skin depths, absorbing roughly 28 dB — while 1 oz copper measures 0.035 mm and manages about 3 dB.
Worked through: skin depth is δ = √(ρ / πfμ), with copper resistivity ρ = 1.7 × 10⁻⁸ Ω·m, f = 5 × 10⁵ Hz and μ₀ = 1.26 × 10⁻⁶ H/m. The denominator is πfμ = 3.1416 × 5 × 10⁵ × 1.26 × 10⁻⁶ ≈ 1.98. Then δ = √(1.7 × 10⁻⁸ / 1.98) = √(8.6 × 10⁻⁹) ≈ 9.3 × 10⁻⁵ m, or 0.093 mm [0.0037 in].
Absorption in a sheet of thickness t is A = 8.69 × t/δ decibels. A 0.3 mm sheet gives 8.69 × (0.3 / 0.093) ≈ 28 dB; 1 oz foil gives 8.69 × (0.035 / 0.093) ≈ 3 dB. Three skin depths means the field that enters the sheet dies inside it. Surface reflection adds on top of that. These are absorption figures; the near-field geometry of a small loop source changes the reflection term, not the absorption term.
| Frequency | Skin depth in copper | Absorption of a 0.3 mm sheet |
|---|---|---|
| 500 kHz (fundamental) | 0.093 mm | about 28 dB |
| 1.5 MHz (3rd harmonic) | 0.054 mm | about 49 dB |
| 4.5 MHz (9th harmonic) | 0.031 mm | about 84 dB |
Skin depth shrinks with the square root of frequency, so the harmonics take care of themselves. Mains hum is the opposite case: at 60 Hz the skin depth in copper is about 8.5 mm. That is why copper ignores hum while steel shunts it. The 50 Hz field around signal filters is its own problem, covered in the guide to enclosure shielding for 50 Hz noise.
Count on the enclosure wall as a backstop rather than the shield. For 304 stainless — the grade our stainless wall-mount and HMI enclosures use — resistivity is about 7.2 × 10⁻⁷ Ω·m with μr near 1. The skin depth at the fundamental works out near 0.6 mm, so a 1.5 mm [0.059 in] wall is only 2.5 skin depths, about 22 dB of absorption. At 1–2 in of board spacing the wall is barely in the coupling path anyway, which is why the local can next to the module does the work. An aluminum wall of the same thickness, with a 0.12 mm skin depth, clears a dozen skin depths and backstops harder.
Does the converter board layout matter as much as the shield?
It matters more, and it comes first. Move the module off its two-layer carrier onto a board with a continuous ground plane. The field a loop radiates scales with loop area, and a board without a plane multiplies every loop.
On a four-layer board the return current of every fast trace runs in the plane directly beneath it. The hot loop — input capacitor to switch to inductor to output capacitor — collapses to a few square millimetres. Halve the loop area and you halve the coupled field, so the plane is the cheapest decibels in the build. Switcher layout app notes say the same thing in net names: keep the sense, feedback, compensation and signal-ground nets short and on the quiet side of the board, away from the switch node.
The loudest single part on a typical dual-output module is the open drum-core inductor, whose field is uncontained by design. Fit a closed-core inductor, or turn the open one so its axis points away from the analog board. For module placement, DIN rail spacing and spare space around the board, the guide to backplate and DIN rail layout shows how to keep hot and quiet zones apart on one panel. The copper-pour ground-plane design guidelines show how to tie a pour into a plane without slitting it.
Where does the shield bond to ground?
Bond the shield to chassis ground at exactly one point, under a standoff with a short braid or solder lug. Place that bond away from the signal star ground, so the shield's milliamp-scale currents return through the chassis instead of through the audio ground.
A floating plate is worse than no plate. It picks up the switch-node field capacitively and re-radiates it from its far side, which is the antenna behaviour you were right to worry about. One bond gives the collected charge a return path and closes no loop, because a single-point connection has nothing to circulate in. The natural spot is the boss where the board's ground plane already bonds to the wall: the shield lug lands beside it, not on the audio star at the input jacks. Keep the braid short, because a pigtail adds roughly 1 nH per millimetre. At the 4.5 MHz ninth harmonic, 50 nH is about 1.4 Ω of series impedance — enough to undo the can at the frequencies that matter most.
The grounding and bonding guide covers how boards, shields and enclosure metal tie into one bonding system. Bolt the lug onto bare metal; NEC 250.12 requires paint, lacquer and enamel removed from contact surfaces at grounding connections.
How far should the converter sit from the analog board?
Keep 1 in [25 mm] between the converter board and the analog board as a working minimum, and 2 in [50 mm] on the side carrying the moving-coil input. Small-loop coupling falls with the cube of distance, so doubling the gap cuts pickup to about one eighth.
Two small loops a distance d apart couple through a mutual inductance that falls as 1/d³ once d is a few loop sizes. The arithmetic for the two spacings is (1/2)³ = 1/8. Orientation moves the number as much as distance does, because loops couple hardest when their axes line up. Mount the converter board edge-on to the cartridge input and put the copper shield between the two. A vertically mounted distribution board with the can facing the analog side does exactly that. The 25 mm figure is a working value for boards up to about 4 × 4 in [100 × 100 mm]; a board carrying a large unshielded input loop wants the 2 in figure on that side regardless.
Working depth, mounting height and the space around the board all come from the sizing routine in how to size an electrical enclosure.
What order do you build and wire the shield?
Work in this order so each change is measured on its own: record the baseline, quiet the wiring, rebuild the board, add the copper, make the one bond, set the spacing, then retest with the inputs shorted.
You need a sheet of 0.3 mm [0.012 in] copper or brass for a five-sided can (8 oz copper clad at about 0.28 mm is the single-board fallback), M3 brass standoffs with braid or solder lugs, split ferrite cores, an oscilloscope with a ×10 probe, a dummy load and shorted plugs for the preamp inputs. Unplug the supply and let the input and output capacitors bleed down before you lift the module off its sockets.
- Scope both rails into the dummy load and write down the baseline: ripple in millivolts at rated load, plus any 100 or 120 Hz component.
- Quiet the wiring first: twist each input and output pair (the guide to twisted pairs inside a grounded enclosure shows when it pays), fit the split ferrites, and keep the converter's return out of the audio return.
- Move the module onto its own four-layer board with a continuous plane, shrink the hot loop, and fit a closed-core inductor.
- Form the 0.3 mm copper into a five-sided can over the module; lap the seams and clamp or solder them along their length, opening the can only where the leads leave.
- Bond the can to the chassis at one standoff with a short braid, beside the plane-to-chassis bond and away from the audio star.
- Set the spacing at 1 in [25 mm] minimum and 2 in [50 mm] on the cartridge side, and keep connector cutouts off the wall that faces the converter; set them deliberately in the custom enclosure builder when you order, following the cutouts and glands guide.
- Retest with the inputs shorted and the gain at maximum: converter off, then on, rails against the baseline.
How do you confirm the shield worked?
The proof is comparative. With the inputs shorted and the preamp at full gain, the output noise floor holds when the converter switches on, and rail ripple stays at your millivolt baseline. Any change traces to wiring, not to missing shielding.
- Rail scope at rated load: baseline ripple unchanged and no new 100 or 120 Hz component. A hum that appears with the converter points at a shared return; the control signal ground loop checks for a cabinet article walks you through finding it.
- Shorted-input test at full gain, converter off and then on: the floor should sit still on the scope, with the screen at 1–5 mV per division to start.
- Slide test: move the converter board from 2 in [50 mm] to 1 in [25 mm] and watch the floor; an unchanged trace says the plane and the can are carrying the load.
- Touch test: press a cloth-covered finger on the can while listening. No change means the bond is solid; new hum means the bond or the return path needs a second look.
- Radio sweep: walk an AM pocket radio past the box across 530–1700 kHz, which covers the fundamental and its low harmonics, and listen for warble over the can seams.
What mistakes ruin the shielding?
Two failures account for nearly every bad shield: a can left floating, which collects the field and re-radiates it like an antenna, and a can bonded into the signal star ground, which turns the preamp's quietest node into the switcher's return path.
- Counting the pour as the shield. One ounce of copper is worth about 3 dB; the shield is the 0.3 mm can, and the pour underneath exists for the return currents.
- Slotting the plane under the module. A return current that detours around a slot grows the loop by the detour's area, so keep the plane solid beneath the inductor and the switch.
- Bonding through a long pigtail. At roughly 1 nH per millimetre, a long lead makes the bond inductive right where the harmonics live; bolt the braid down over a short path.
- Leaving the drum-core inductor facing the preamp. Its field is uncontained by design; fit a closed core or turn the axis away.
- Putting connector cutouts on the wall facing the converter. Every aperture is a leak at the harmonics, so place the inlet and the signal pass-throughs where the spacing plan wants them.
- Trusting the wall instead of the can. A 1.5 mm stainless wall gives about 22 dB of absorption and sits out of the coupling path at 1–2 in of spacing; the local can does the work, and the wall backstops it.
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.
| Enclosure | Material | Outside H x W x D / display | Rating |
|---|---|---|---|
| 6 × 6 × 4 in | 304 stainless | 6 x 6 x 4 in | Rated NEMA 4X / IP66 by the manufacturer; not UL listed |
| 8 × 6 × 4 in | 304 stainless | 8 x 6 x 4 in | Rated NEMA 4X / IP66 by the manufacturer; not UL listed |
| 10 × 8 × 4 in | 304 stainless | 10 x 8 x 4 in | Rated NEMA 4X / IP66 by the manufacturer; not UL listed |
Questions
Is a 1 oz copper pour enough to shield a 500 kHz converter?
One ounce is 0.035 mm, about a third of the 0.093 mm skin depth, so it absorbs roughly 3 dB — fine as a ground plane, thin as a shield. The can wants 0.3 mm of copper or brass, three skin depths, for about 28 dB before surface reflection adds on top.
Where does the shield bond when the board carries the audio star ground?
At the standoff where the board's ground plane already bonds to the wall — nowhere else. A single bond closes no loop for the shield's milliamp-scale currents, so they return through the box metal instead of the audio ground. Bolt the lug onto bare metal: a lug through paint or oxide reads high and drifts with humidity. NEC 250.12 requires paint removed from contact surfaces at grounding connections for exactly this reason.
How much clearance does a converter need from a moving-coil input?
Hold 1 in [25 mm] as the working minimum and 2 in [50 mm] on the cartridge side. Coupling between small loops falls with the cube of distance, so 25 mm to 50 mm is an eightfold drop; edge-on mounting and a closed-core inductor move it further still.
Does the stainless enclosure wall do the shielding on its own?
Only as a backstop. The skin depth in 304 stainless at 500 kHz is about 0.6 mm, so a 1.5 mm wall gives roughly 22 dB — but at 1–2 in of board spacing the wall is barely in the coupling path, which is why the local copper can does the work.
