PCB shield cans for RF amplifier stages: layout and fit

Short answer: give each gain block its own can once chain gain passes about 60 dB, and stitch the can wall to ground every 10-15 mm. Keep the interstage filter inside the can with the stage that feeds it. At 110 dB of forward gain, a four-stage chain amplifies its own leakage as readily as the signal. The isolation therefore belongs in the layout, not in a lid added after the first spin oscillates. The numbers below are worked for a 10-400 MHz, 50 Ω chain. They cover where the walls go, the fence pitch they need at 400 MHz, what crosses each wall, and the bench checks that prove the isolation.

Why a 110 dB chain needs walls

Every feedback path adds to the forward gain when you sum it in decibels. A loop of 110 dB of forward gain plus a -115 dB leakage path sits at -5 dB, which stays quiet. A -105 dB path closes the loop past 0 dB, and the chain sings. You want every reverse path below about -120 dB: the 110 dB of gain plus a 10 dB margin for phase and part tolerance.

This chain amplifies broadband noise across a 220 MHz passband, so there is always a frequency where the phase comes around. The layout has to beat the gain everywhere in the band, not at one spot. That is what the walls buy you.

Published figures for board-level shields run 40-80 dB per well-grounded can. Split the chain into gain blocks and give each boundary its own wall, and the boundary isolations add. Three interstage walls at the bottom of that range already sum to 3 x 40 = 120 dB, which meets the -120 dB target before routing losses help. Walls that land near the top of the range give you margin to burn. This is also why the working threshold sits near 60 dB of chain gain. Below it, a ground fence with clean routing holds. Above it, every gain block gets a can of its own.

One more property matters: a can blocks electric fields and passes magnetic ones. Shrink every signal and return loop so magnetic coupling has almost no area to work on, and keep large currents off the outside of any can.

What you need

• Premade SMD shield cans with solderable walls and removable lids, or a hand-folded can from 0.2-0.5 mm tinned copper or brass sheet. Removable lids make bring-up far easier than a one-piece cover.
• Ground fence pads and fence vias for the pitch below, plus a pad for every wall corner.
• A pi decoupling network (capacitor, ferrite bead or small inductor, capacitor) for every supply that crosses a wall.
• 0 Ω jumpers in each interstage, so you can bypass a stage during bring-up.
• Footprints for resistive pi pads between stages, with 3 dB and 6 dB values on the bill of materials.
• 50 Ω routing over an unbroken ground plane, and thermal vias under each amplifier's exposed pad.

Working the fence pitch for a 110 dB chain at 400 MHz

Start with the inputs. The top carried frequency is 400 MHz, the upper edge of the example chain's band of interest. The board is FR-4, with a relative permittivity of about 4.4 at VHF. The fence rule RF layout guides give is a via pitch of one-twentieth of a wavelength, λ/20, at the highest frequency the wall must contain. This article works to 10-15 mm and takes whichever limit is tighter.

The arithmetic is short. In air, λ = 300 / 400 = 0.75 m = 750 mm. FR-4 slows the wave by a factor of √4.4 ≈ 2.10, so the wavelength on the board is 750 / 2.10 ≈ 358 mm. The λ/20 limit works out at 358 / 20 ≈ 17.9 mm. The 10-15 mm fence is tighter than that, so it governs, and the margin covers both the wave in the substrate and the cavity of air above it.

The decision: fence every wall at 10-12 mm, which leaves margin against the 17.9 mm limit, and treat 10-15 mm as the loosest pitch you accept. Budget about 8 fence vias per 100 mm of wall run (100 / 12 ≈ 8.3). Put a grounded pad in every corner, because corners end up sparse more often than any other spot. The low end of the band never sets the pitch: at 10 MHz the wavelength is 30 m.

Lay the walls in before you route

1. Draw the can walls before any trace routing. Split the example chain FILTER - LNA - FILTER - A2 - A3 - A3 - FILTER into three blocks: input filter with the LNA, the bypassable mid stages, and the output pair with its final filter. A wall crosses between blocks, never between a filter and the amplifier it feeds.
2. Drop the ground fence along every wall: vias at 10-12 mm down to the unbroken plane below, with a pad in each corner. Run ground strips around the LNA's neighbourhood even where no can goes, because they cost nothing and help everywhere.
3. Cross each wall only where you planned it. A supply crosses through its pi network, landed on the wall line with the shortest decoupling loop you can draw. A signal crosses on a short 50 Ω span with ground vias on both sides of the wall. Keep several dielectric heights of clearance between that span and the wall, because the wall reads as coplanar ground and pulls the impedance down.
4. Fit a 0 Ω jumper in every interstage. Bring the chain up one stage at a time, because a 110 dB chain switched on all at once tells you only that something oscillates somewhere.
5. Leave a pi pad footprint between every pair of stages and fit it only where a stage misbehaves. A 3 dB pad placed after the LNA costs almost nothing in noise figure, and it buys isolation in both directions.
6. Give the heat a way out. A sealed can stops the convection a bare part enjoys, so drop thermal vias from each amplifier's exposed pad to the plane. Check junction temperature with the lid on, because the can changes the answer. The same logic scales up, and the sealed-enclosure component-layout guide works it at instrument size.
7. Set the stack-up before routing: signal on top over an unbroken ground plane, power on an inner layer. A PWR-GND-SIG-GND order also spreads part heat into the outer copper while keeping a solid reference under the signal traces.

Checks on the bench

Calculate the floor before you measure it. Thermal noise at 290 K is -174 dBm/Hz, and the example filters pass 20-240 MHz, so the noise bandwidth is 220 MHz and adds 10 log10(220,000,000) = 83.4 dB. The input shelf therefore sits at -174 + 83.4 = -90.6 dBm. After 110 dB of gain the output shelf lands near -90.6 + 110 = +19.4 dBm, plus a few decibels of noise figure. Compare that level with the output stage's maximum in its datasheet. A last stage that cannot deliver a continuous +19 dBm will clip the shelf, and that is what the pad footprints are for.

Then measure with the lids off and the input terminated in 50 Ω. Bring stages in one at a time with the 0 Ω jumpers and fit the cans last. A healthy chain shows the calculated shelf and nothing else. A narrow spur that stays put while you tune the input is a loop that has closed. The stage below the last quiet one is where to look.

Press gently on a lid while you watch the output floor. A spur that shifts under your thumb points to a resonance in that cavity. The usual culprits are a sparse corner or a supply that crossed the wall unfiltered. Lids on should drop the floor or leave it level; a floor that rises with lids fitted means a wall is coupling where it should be isolating.

Mistakes that cost the isolation

• One can across the whole chain. It puts input and output in the same cavity, and the 110 dB between them finds a path.
• Sparse corners and a floating lid seam. Corners take a pad, and the seam takes the same 10-15 mm pitch as the wall.
• A supply crossing the wall raw. The decoupling path is the first coupling route into a stage, so every supply lands through its pi network at the wall line.
• Expecting the can to fix magnetic coupling. Shields block electric fields, not magnetic ones; small loops fix that.
• Stray current riding the outside of a can. Route returns so the can exterior carries none, or the can becomes a radiator.
• A wall hugging the 50 Ω trace. Keep several dielectric heights of clearance so the coplanar capacitance leaves the match alone.

From the board to the host box

Settle the host box before you freeze the board outline, the way the PCB enclosure fit guide argues at board scale. Tie the can ground to the case at one planned point. The chassis-ground connection point article covers where that point belongs, and the ground-to-chassis guide covers when a board ground should tie to metal at all. Treat every opening in the case wall exactly like a can wall, bonded or filtered; the RF feedthrough planning article walks that path.

When the board heads for the field rather than the lab bench, order the host box with the openings you actually need. A custom stainless enclosure builds to your size and every cutout from one piece and arrives in 10 business days, counted from drawing approval for custom work. It is rated NEMA 4X / IP66 by the manufacturer; not UL listed. Lay the openings out in the cutout designer, which checks the spacing and exports the layout to send with the quote request.

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