Bond the 5 V return to the enclosure and let the metal do the shielding. It works with no earth wire: at 50 MHz a 1 mm stainless wall is worth over 200 dB on its own. What an earth wire buys you is shock protection, not shielding. The 6 m wavelength at that frequency makes a 1.5 m earth lead a quarter-wave stub, a high-impedance connection that is useless as an RF reference. Treat the two questions separately. Bonding the return to the wall at one point near the SMPS output makes the box sit at one potential and become a shield instead of a radiator. Make the product safe the Class II way: double insulation between the mains and every touchable metal part, with touch current at or below 0.25 mA. The numbers for each decision follow.
Does a metal enclosure shield with no earth connection?
Yes — the metal wall gives you over 200 dB of shielding with no earth wire attached. Reflection and absorption happen in the metal itself; the earth wire only matters for safety and for low-frequency electric fields.
A shield works two ways. Most of the arriving wave reflects off the conductive surface. What gets in decays to nothing within a few hundredths of a millimetre of metal. Neither mechanism asks whether the box is earthed — the metal is doing the reflecting. Radiated-emission testing runs from 30 MHz upward (FCC Part 15 B, CISPR 32). Across that whole band your wall behaves the same way.
The earth wire is a different tool. It carries fault current so the breaker trips, and a two-wire product never had one, which is exactly why the Class II route exists. At radio frequencies the wire is nearly irrelevant anyway. A 1.5 m earth lead is a quarter wavelength of the 6 m wave, a high-impedance stub at precisely the frequency you care about. Panels that do have earth get their RF reference from bonding the 0 V to the wall, not from the green wire — grounding PCB ground to chassis when an earth point exists covers that case. On magnetic near fields, the one coupling a thin wall handles worst, whether enclosure metal attenuates magnetic fields has the numbers.
How much shielding does the wall itself give at 50 MHz?
About 220 dB for a 1 mm 304 stainless sheet: 75 dB of reflection at the surface plus about 145 dB of absorption inside it, computed for non-magnetic stainless at 2.5 % of copper's conductivity.
Work it for your own wall. Skin depth is the depth at which current falls to 37 %. In 304 stainless at that frequency it is 0.06 mm. A 1 mm wall is 1 / 0.06, about 16.7 skin depths, and absorption loss is 8.7 dB per skin depth, so 16.7 x 8.7 is about 145 dB. Reflection comes from the surface impedance of the metal, about 0.017 ohm for stainless, giving 20 x log(377 / (4 x 0.017)) = 75 dB. The reflection figure for any box metal sits within about 15 dB of this either way. The total is two orders of magnitude more than you need.
Put that against the test: the FCC Part 15 B limit at 3 m is 40 dBuV/m from 30 to 88 MHz. A product reading 48 dBuV/m needs 8 dB of shielding; the wall offers about 220. So the sheet is never what lets you down — the seam, the hole and the cable are. Inside the box your board also sits in the near field, which reaches 6 / 2pi, about 0.95 m. Electric-field reflection there is even stronger than the far-field figure.
Should you bond the 5 V return to the enclosure?
Yes, at one point, as close to the SMPS output or the noisy board as you can get. The bond gives radio-frequency currents a low-impedance way back to their source; without it the wall floats up as a plate that re-radiates.
Float the box and it capacitively couples to whatever is inside. A board of 300 x 200 mm sitting 25 mm from the wall couples through about 21 pF (8.85 pF/m x 0.06 m2 / 0.025 m). At 150 MHz — the third harmonic of the clock — 21 pF is 1 / (2pi x 150 MHz x 21 pF), about 50 ohms. That is an easy path onto 0.06 m2 of metal, and the box then radiates better than the board ever did. Bonded to the return, the box holds one potential and gives that current a short path home instead of an antenna.
Keep the bond short and wide. A straight wire carries about 1 nH per mm — the figure EMC design handbooks use for round wire — so 30 mm of wire is 30 nH, which is 28 ohms at 150 MHz. Use a bracket or a stud with direct metal contact, and keep the path under about 25 mm. If a cable leaves the box, bond the return again at the connector, where the cable meets the wall.
One thing to settle before you commit: many two-wire SMPS modules fit a capacitor from the DC output to the rectified mains side for EMC, and the module datasheet will show it. If it is there and you bond the return to the case, its leakage lands on your metal — the arithmetic is in the touch-safety section below. For tracing which peak comes from the supply in the first place, troubleshooting RF emissions from switch-mode power supplies walks the usual suspects.
Can you connect the enclosure to neutral for EMI?
No — neutral is a live conductor, so bonding the case to it puts mains potential on every touchable surface and feeds your switching noise straight back into the supply. The EMI argument does not survive the safety rules. This is the one wiring move in this design that is outright dangerous, so write it plainly on your build documents: the enclosure connects to the return and to nothing else.
Neutral sits near earth at the transformer, not at your socket. Reverse the plug, open the neutral upstream, or share the circuit with an unbalanced load, and the case rides at line potential through exactly the impedance you added for EMI reasons. The screening job that neutral bond was meant to do is done better by the return bond, which drains the noise at its source, and by sealing the seams, which stops the rest escaping.
How does a product with a metal case and no earth stay touch-safe?
By Class II double insulation: every mains-to-metal path carries two independent insulation layers, the SMPS is a certified double-insulated module, and touch current on the case stays at or below 0.25 mA, the Class II limit from the IT-equipment standard line.
Class II is a whole-product property. The module must be certified for double-insulated service — encapsulated plastic modules usually carry the marking — and everything you add around it has to keep the two layers: shrouded mains terminals, insulated boots on the crimps, strain relief that cannot wear through to the case. Between the line conductors and any metal a person can touch you need basic insulation plus supplementary insulation, or one layer rated as reinforced. Mark the rating plate 'Class II' or with the double-square symbol, the recognised marking for a double-insulated appliance. Safe design for a metal enclosure without protective earth works through the insulation system in detail.
Then do the touch-current arithmetic for the bond you just made. Say the module datasheet shows 2.2 nF from the 5 V return to the mains side. At 230 V, 50 Hz, that is 1 / (2pi x 50 x 2.2 nF) = 1.45 Mohm, so 230 / 1.45 Mohm = 0.16 mA flows onto the case through the hand of anyone touching it. That clears the 0.25 mA Class II limit, the cap the IT-equipment standard line (IEC 60950-1) set, with 0.09 mA of margin — so measure the finished product, because line voltage, a second capacitor and supply harmonics all push the number up. If it measures over the limit, swap to a module without the output capacitor, or leave the case floating and put a small shield can over the noisy board instead.
Where does an ungrounded shield actually leak?
At the seams, the holes and the cable pass-throughs, never in the metal. A slot starts passing energy once it is longer than about one twentieth of a wavelength, which at 150 MHz is 100 mm — shorter than most door seams.
| Where it leaks | The number that decides | What closes it |
|---|---|---|
| Door seam | one twentieth of the wavelength: 100 mm at 150 MHz, 33 mm at 450 MHz | Continuous conductive gasket, or bare-metal contact points well inside half those lengths |
| Ventilation holes | each opening under the same limit — 33 mm at 450 MHz | A pattern of small round holes rather than slots |
| Display window | one hole the size of the window | Keep it under the limit at the highest failing harmonic, or use mesh-laminated glass |
| Cable through a plain hole | the cable carries the noise outside and radiates it | Shielded cable with a 360 degree gland bond, or a ferrite at the wall plus the return bonded at the connector |
Paint, powder coat and anodising are insulators, so a painted seam is an open seam electrically — does paint affect EMI shielding in metal enclosures covers what to do about it. Where a cable leaves the box, the shield termination carries the whole benefit, and cable shield termination at the enclosure entry shows which points to bond. Before blaming the board alone, clamp a ferrite on each internal cable where it leaves the noisy board: harmonics reach the outside world through wiring more often than through the air. Fixing radiated emissions in an existing enclosure takes the next step when the box alone does not clear the margin.
What should you specify on the enclosure order?
A one-piece welded body in 304 or 316L stainless, a door with a conductive gasket, bond studs beside the power entry and the noisy board, and a gland for every cable — all on your cutout drawing, delivered in 10 business days counted from your approval of the drawing.
- Material for the environment: 304 stainless for indoor and general duty, 316L where salt air, brine or chlorides are present. The RF numbers never pick the grade, because every box metal clears the bar by two orders of magnitude.
- Body made from one piece with welded seams, so the door is the only long slot you have to seal.
- Door with a continuous conductive gasket and bare contact surfaces; specify the gasket if the box will be opened in service.
- Two bond studs inside: one within 25 mm of the power entry, one at the corner of the noisy board. You set every cutout, so put the SMPS and the radiating board on the same wall and keep the bond short.
- Every cable entry cut and listed on the drawing — the cutouts and glands guide gives the sizes.
- Environmental rating picked for the site: our Type 4X / IP66 enclosures are rated NEMA 4X / IP66 by the manufacturer; not UL listed. That rating covers water and corrosion, and it is separate from every EMC decision above.
Custom enclosures take your material, dimensions and cutout list from one piece; the enclosure selector narrows the choice by environment and inside size. Bond the return, seal the door, keep the mains off the case, and the two-wire metal box passes on both counts.
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 |
|---|---|---|---|
| 12 × 10 × 6 in | 304 stainless | 12 x 10 x 6 in | Rated NEMA 4X / IP66 by the manufacturer; not UL listed |
| 20 × 16 × 8 in | 304 stainless | 20 x 16 x 8 in | Rated NEMA 4X / IP66 by the manufacturer; not UL listed |
| 24 × 20 × 8 in | 304 stainless | 24 x 20 x 8 in | Rated NEMA 4X / IP66 by the manufacturer; not UL listed |
Questions
Would a bigger earth wire help if you added one?
Length matters, gauge barely does. A straight conductor carries about 1 nH per mm whatever its cross-section, so 300 mm of 10 mm2 earth wire is around 300 nH, which is about 280 ohms of reactance at 150 MHz. Keep any earth lead under a tenth of a wavelength at your highest failing frequency, and bond the return to the wall regardless; the wire's real job is carrying fault current.
Is aluminum a better shield than stainless steel?
At 50 MHz a 1 mm aluminum wall reflects about 89 dB where 304 stainless reflects about 75 dB, and aluminum absorbs more as well. Both numbers sit two orders of magnitude above what a real door seam delivers, so radio performance never picks the material. Pick 304 stainless for indoor and general duty, and 316L where salt air or chlorides are present.
Does the box also fix conducted emissions on the mains lead?
Only a little. Conducted emissions run from 150 kHz to 30 MHz, and that energy leaves down the line and neutral leads, so the wall is barely in the path. The module's input filter and, if needed, a ferrite on the input pair do that job; bonding the case to the return rail helps the common-mode share at the top of the band.
Does a hinge or a latch count as an RF bond?
Treat it as a coincidence, never a bond. Paint, anodizing and oxide films sit in the pivot, and the path through a hinge is tens of millimetres of uncertain contact. Cross the door with a conductive gasket or a bond strap shorter than 25 mm; at 150 MHz even a strap that long is about 23 ohms, so shorter and wider wins.
How many decibels over the limit can a metal box rescue?
Bonded and seam-sealed, a metal box usually clears a product sitting 3 to 10 dB over, because the capacitive coupling it removes is normally the dominant path. A board 15 to 20 dB above the 40 dBuV/m Class B limit at 3 m needs fixes at the source — series resistance on clocks, a small shield can over the board — with the box supplying the last 10 dB.
