EMI Bench Testing Notes: What a Spectrum Analyzer Under $500 Actually Shows You

Pre-compliance EMI testing with a $400 SDR and a near-field probe set tells you more than you'd expect — and less than you need for a formal chamber test. Notes from bench sessions on a wearable hub before we paid for a semi-anechoic slot.

Setup

• SDR: RTL-SDR Blog V3 (~$40) + Ham It Up upconverter optional for sub-100 MHz — we used direct sampling hack for HF on V3
• Spectrum software: SDR++ 1.1 on Linux; later borrowed Rigol DSA815 for comparison
• Probes: BeeHive 101A near-field H-field set (~$180 used) + homemade loop from coax
• DUT: nRF52840 wearable hub, 2-layer flex, BLE advertising + 200 Hz IMU SPI

Goal: find "obvious stupid" emissions before FCC unintentional radiator filing — not certify.

What a sub-$500 bench shows

You can rank fixes

Before/after comparisons reliable relative to your bench setup:

Change	~Peak at 2440 MHz (BLE)	Harmonic 4880 MHz
Baseline layout	-42 dBm ref	-50 dBm
Add π-filter on antenna feed	-48 dBm	-58 dBm
Fix ground stitch under nRF	-52 dBm	-62 dBm
Shield can (prototype tin)	-58 dBm	-70 dBm

Absolute dBm meaningless vs. chamber; delta guides where to spend layout time.

You find clock harmonics fast

Sweep 30–3000 MHz with H-probe over MCU crystal, SPI SCK, DC-DC SW node. Our ADXL355 SPI at 5 MHz showed strong 15 MHz, 25 MHz spurs — series R on SCK cut 8 dB at third harmonic (correlated with SPI debug session).

Use narrow RBW (3 kHz) for spur hunting; wide RBW (100 kHz) for BLE band shape.

Near-field localization works

H-probe peaks over physical source — DC-DC inductor vs. antenna mismatch vs. cable acting as radiator. E-probe (when we borrowed one) showed flex cable edge as slot antenna — reroute ground return dropped cable radiation.

Pair with scope triggers on DC-DC switching node to time-correlate spur frequency with Fsw.

What it doesn't show

• Conducted emissions on USB/power lines — need LISN ($$$ or rental)
• Far-field levels — near-field peak ≠ pass/fail at 3 m distance
• Absolute compliance margin — your -58 dBm isn't FCC Part 15.247 limit translated
• Spurious under load modulation — BLE frequency hopping smears in cheap SDR; use dedicated receiver or chamber for final

Formal chamber quote: $3.2k/day regional — still needed; bench saved two revisits (~$6k + 3 weeks).

Method (repeatable)

1. Golden photo of probe position per test — reproducibility matters more than calibration.
2. Max hold spectrum 60 s normal operation mode.
3. Mark suspected frequencies — correlate to clocks in firmware (grep MHz schematic).
4. One change at a time — or you'll argue with yourself.
5. Log environmental: laptop PSU 2 m away added 30 MHz garbage until we switched to battery laptop.

SDR++ settings that worked

Start: 2400 MHz, Stop: 2483 MHz, RBW: 10 kHz
Antenna: near-field probe via 30 dB attenuator (protect SDR front-end)

Without attenuator, overload falsifies peaks when probe near hot spot.

GPS timing side experiment

Same SDR bench used for GPS timing notes — 1 PPS reference helped correlate EMI events to radio TX schedule; BLE connection event every 30 ms visible as comb.

Real findings that shipped fixes

1. Missing ground under nRF ANT pin — layout tool poured wrong; bench caught, chamber would have failed radiated.
2. Heater PWM on same return as RF — moved return path; conducted noise on battery line dropped (measured with scope + current probe, not SDR alone).
3. USB-C cable shield termination — PD board flex transition; SDR showed 480 MHz spike when cable attached.

When to stop bench and book chamber

• Product intentional radiator (BLE, Wi-Fi) — yes, schedule chamber early for baseline
• Delta fixes plateau — last 3 dB may need chamber to verify far-field
• Customer contract requires third-party lab report — bench isn't legal evidence

Equipment upgrade path

If budget allows one step up: Rigol DSA815-TG (~$1.5k used) — tracking generator for filter validation. Still not chamber.

Near-field probes: buy once, use on every rev.

What I'd do next

Build emi-snapshots/ repo folder — PNG + probe position JSON per board rev. CI can't run EMI yet; human diff catches regressions when mechanical team rotates antenna 2 mm.

Train every HW engineer on 2-hour SDR bench orientation — cheaper than one chamber redo.

Pre-compliance EMI with cheap gear is triage, not verdict. Use it to make chamber visits boring.

SDR limitations we hit

• Overload near BLE antenna — false harmonics without 30 dB attenuator
• Image responses below ~24 MHz on V3 — know your dongle's aliasing
• No conducted emissions — LISN still rental

Borrowed DSA815 correlated one near-field peak to far-field — within 6 dB on one spur; not repeatable on all — still useful for trend.

Shield can prototype economics

Tin can + copper tape: $3, 2 dB improvement on 2.4 GHz spur. Production shield: $0.41 BOM + pick-place. Bench proves worth before mechanical commits.

What I'd do next