C
Editorial
Hello from CosmosForum Editors —
Instrument and industrial boards that put Wi-Fi, SDR, telemetry, or short-range radios next to low-noise analog front-ends often add a via fence around the RFIC and still fail isolation or desense. Isolation is a **return-current** problem before it is a “more vias” problem. Radio and science-instrument teams keep rediscovering the same failure modes that II / lab-instrumentation bring-up already knows by another name: a pretty gerber fence that does not stitch the grounds the part actually references, or a connector / USB cut that opens the return plane and defeats the wall.
We opened a technical companion on CosmosForum with rfforum editorial on exactly this checklist. Cross-link: https://www.cosmosforum.org/t/43
### 1) Pitch is a start, not a guarantee
Fence pitch starting near λ/20 at the highest aggressor frequency is a useful rule of thumb — not a certificate. Keep the fence close to the paddle / package edge. A sparse ring far from the RFIC is decoration.
### 2) Stitch the grounds the part actually references
If the RFIC references multiple ground layers, the fence must stitch those layers — not only the top copper that looks dense in the 2D view. Instrument parallel: do not declare shield integrity from a silkscreen ring alone.
### 3) Return-plane cuts kill fences
Connector, USB, slot, or cable-exit gaps in the reference plane are classic self-inflicted desense. Dense bridging or a redesign beats “add vias and hope.” Same habit as documenting known-bad IO channels instead of averaging them into a soft-sensor.
### 4) Same-fixture A/B before declaring victory
Measure isolation or desense **before vs after** fence changes on the same fixture. Changing the harness, lid, or cable dress in the same step invents ghost wins. Freeze a written acceptance idea: max pitch at aggressor band, max paddle-edge gap, mandatory stitch list of reference layers, plus a same-fixture desense delta gate.
### Acceptance question we are collecting war stories on
When a Wi-Fi / SDR / telemetry module must coexist with a low-noise instrument front-end on one paddle, do you freeze that **fence acceptance triple + same-fixture desense delta**, or is “vias look dense on the gerber” still enough before bring-up?
Companion technical thread on CosmosForum (rfforum editorial note + admin engineering reply):
https://www.cosmosforum.org/t/43
Methods and failure modes only — no product pitches, no sock accounts. If iiunion has a production note on instrument-board RF coexistence, shield/fence acceptance, or desense bring-up gates that science / RF teams should steal, a technical reply on either side is welcome.
Clear returns, honest fences —
CosmosForum Editors
https://www.cosmosforum.org/
Instrument and industrial boards that put Wi-Fi, SDR, telemetry, or short-range radios next to low-noise analog front-ends often add a via fence around the RFIC and still fail isolation or desense. Isolation is a **return-current** problem before it is a “more vias” problem. Radio and science-instrument teams keep rediscovering the same failure modes that II / lab-instrumentation bring-up already knows by another name: a pretty gerber fence that does not stitch the grounds the part actually references, or a connector / USB cut that opens the return plane and defeats the wall.
We opened a technical companion on CosmosForum with rfforum editorial on exactly this checklist. Cross-link: https://www.cosmosforum.org/t/43
### 1) Pitch is a start, not a guarantee
Fence pitch starting near λ/20 at the highest aggressor frequency is a useful rule of thumb — not a certificate. Keep the fence close to the paddle / package edge. A sparse ring far from the RFIC is decoration.
### 2) Stitch the grounds the part actually references
If the RFIC references multiple ground layers, the fence must stitch those layers — not only the top copper that looks dense in the 2D view. Instrument parallel: do not declare shield integrity from a silkscreen ring alone.
### 3) Return-plane cuts kill fences
Connector, USB, slot, or cable-exit gaps in the reference plane are classic self-inflicted desense. Dense bridging or a redesign beats “add vias and hope.” Same habit as documenting known-bad IO channels instead of averaging them into a soft-sensor.
### 4) Same-fixture A/B before declaring victory
Measure isolation or desense **before vs after** fence changes on the same fixture. Changing the harness, lid, or cable dress in the same step invents ghost wins. Freeze a written acceptance idea: max pitch at aggressor band, max paddle-edge gap, mandatory stitch list of reference layers, plus a same-fixture desense delta gate.
### Acceptance question we are collecting war stories on
When a Wi-Fi / SDR / telemetry module must coexist with a low-noise instrument front-end on one paddle, do you freeze that **fence acceptance triple + same-fixture desense delta**, or is “vias look dense on the gerber” still enough before bring-up?
Companion technical thread on CosmosForum (rfforum editorial note + admin engineering reply):
https://www.cosmosforum.org/t/43
Methods and failure modes only — no product pitches, no sock accounts. If iiunion has a production note on instrument-board RF coexistence, shield/fence acceptance, or desense bring-up gates that science / RF teams should steal, a technical reply on either side is welcome.
Clear returns, honest fences —
CosmosForum Editors
https://www.cosmosforum.org/