EMC Pre-Compliance Testing: Bench Setup and Why Labs Differ
How to run EMC pre-compliance tests in-house: LISN and antenna setups, spectrum analyzer settings, noise floor checks, and why results differ from the lab.
A day in an accredited semi-anechoic chamber is expensive, and the worst time to discover a 10 dB overshoot is the week before a product launch. Pre-compliance testing moves the first look forward: you measure emissions in-house, with simpler equipment, early enough that a layout or filter change is still cheap. It tells you whether you are close to the limit, which frequencies and cables are the problem, and whether a fix moved things in the right direction.
What pre-compliance can and cannot tell you
The value of in-house testing is mostly relative. If a ferrite on the USB cable lowers a 150 MHz line by 8 dB on your bench, it will usually lower it in the chamber too, even if the absolute levels differ. A careful in-house setup often lands within a few dB of the lab for conducted emission, while radiated results can differ by much more, because the site, the antenna height search and the cable layout all change the reading. Treat the in-house number as an estimate with a wide error bar, and keep enough margin to cover it.
Conducted emission on the bench
Conducted emission is the easiest test to reproduce, because the LISN (line impedance stabilization network) defines the impedance the EUT sees and the measurement does not depend on the room. The setup needs a LISN that follows CISPR 16-1-2 (the 50 Ω / 50 µH V-network used from 150 kHz to 30 MHz), a spectrum analyzer or EMI receiver, and a metal reference ground plane. Bond the LISN to the ground plane with a short, wide strap; a long ground wire adds inductance and changes the result above a few MHz.
Arrange a table-top EUT the way CISPR 16-2-1 describes: on a non-conductive table, 0.4 m from a vertical reference ground plane, with the LISN 0.8 m from the EUT boundary. Put a transient limiter between the LISN output and the analyzer input, because switching the EUT on or off can produce spikes large enough to damage an unprotected input mixer. Measure both line and neutral, because the worst conductor is not always the same.
Radiated emission without a full chamber
Radiated emission is where in-house methods differ most. Each option trades cost against correlation, and many teams combine two of them: a current probe on the cables for fast comparisons, and an antenna measurement to estimate absolute levels.
| Method | What it shows | Main limitation | Best use |
|---|---|---|---|
| Antenna at 3 m in an office or open area | Approximate field strength in dBµV/m | Ambient signals (FM, TV, mobile) and reflections from walls and floor | Rough absolute estimate and ranking of failing frequencies |
| Antenna at 1 m | Higher signal, easier to see above the ambient | Near-field effects; correcting to 3 m or 10 m is unreliable | Comparing fixes on weak emitters |
| Current probe on each cable | Common-mode current in dBµA | Says nothing directly about enclosure or board radiation | Finding the radiating cable and checking ferrites |
| TEM or GTEM cell | Repeatable reading in a shielded volume | EUT and cables must be small compared with the cell; correlation depends on the model used | Small products and modules, before-and-after comparisons |
| Rented chamber time | Close to the formal test | Cost and scheduling | Final check before the formal test |
If you use an antenna, measure in both horizontal and vertical polarization, rotate the EUT through 360°, and vary the antenna height if you can. The formal test searches heights from 1 m to 4 m because reflections from the ground plane create peaks and nulls; an antenna fixed at one height can sit in a null and miss a failing line by many dB.
Spectrum analyzer settings that matter
A general-purpose spectrum analyzer can do useful pre-compliance work if it is set up like an EMI receiver. Use the CISPR resolution bandwidths: 9 kHz from 150 kHz to 30 MHz, 120 kHz from 30 MHz to 1 GHz, and 1 MHz above 1 GHz. CISPR specifies these at the −6 dB points, while many analyzers specify RBW at −3 dB, so use the EMI bandwidth setting if the analyzer has one. Start with the peak detector, because it is fast and never reads lower than quasi-peak or average; if the peak reading is below the lowest applicable limit (for conducted emission, the average limit), the other detectors will be too. Then remeasure the few frequencies with the smallest margin with quasi-peak and average detectors, with enough dwell time on each frequency.
Check the number of sweep points. The frequency step should be no more than half the RBW. From 30 MHz to 1 GHz with a 120 kHz RBW, that means a 60 kHz step, or about 16,200 points. Many analyzers default to 1001 points, which gives a step of 970 kHz across the same span, about eight times the RBW: narrow clock harmonics can fall between points and disappear from the scan. Either increase the number of points or split the range into narrower spans.
Finally, check for overload. A strong out-of-band signal or a large conducted transient can drive the input mixer into compression and create false lines. Add 10 dB of input attenuation: real emissions stay at the same corrected level, while products generated inside the analyzer drop or change by a different amount.
Is the noise floor low enough?
The analyzer must see well below the limit, or a passing result means nothing. The thermal noise floor is N = −174 dBm/Hz + 10·log₁₀(RBW) + NF. With a 120 kHz RBW and a noise figure of 25 dB, which is common for an analyzer without a preamplifier, N = −98.2 dBm, or 8.8 dBµV in 50 Ω. Adding an example antenna factor of 15 dB/m and 2 dB of cable loss gives a floor of 25.8 dBµV/m. That is only about 4 dB below the CISPR 32 Class B limit of 30 dBµV/m at 10 m (30 MHz to 230 MHz), and the peak detector displays noise even higher. With a noise figure of 10 dB, the same chain gives 10.8 dBµV/m. A low-noise preamplifier at the analyzer input is usually the cheapest way to get there; a common working rule is to keep the floor at least 6 dB, preferably 10 dB, below the limit.
Why lab results differ from yours
- Site: a chamber is a calibrated site with controlled reflections; an office has walls, furniture and ambient signals that add and cancel at unpredictable frequencies.
- Height and angle search: the lab maximizes each emission over antenna height, polarization and turntable angle, so a fixed-position bench measurement usually reads lower.
- Distance: correcting a 1 m reading to 3 m or 10 m with a 20·log rule fails in the near field, as our earlier article on 3 m to 10 m conversion explains.
- Cable layout: cable length, height above the ground plane and routing change common-mode resonances.
- Operating mode: the lab exercises the EUT in its worst-case mode. Use the same firmware, data traffic, display content and load.
Checklist before booking the lab
| Check | Target | Why |
|---|---|---|
| Conducted emission, line and neutral, peak scan | Below the average limit, or remeasured with quasi-peak and average detectors | Peak is a worst case; if it already meets the average limit, both limits pass |
| Radiated emission, both polarizations, 360° rotation | Several dB of margin at every frequency | Covers site differences and the lab height search |
| Ambient scan with the EUT off | All ambient lines identified | Avoids chasing broadcast signals |
| Noise floor | At least 6 dB, preferably 10 dB, below the limit | A floor near the limit hides failures |
| Worst-case operating mode defined | Same mode planned for the lab | Emissions can change by many dB between modes |
| Fixes documented | Each change with its measured effect in dB | Lets you remove parts that did nothing |
Common mistakes
- Declaring a pass from a single bench scan with 2 dB of margin. The uncertainty of an in-house radiated measurement is larger than that.
- Measuring with a laptop, USB hub or bench supply that is noisier than the EUT. Check the support equipment alone first.
- Running pre-compliance only once at the end. Its value is greatest when you repeat it at each hardware revision.
Calculators used in this post
More posts
Results and summaries are for reference. For certification and test reports use the latest official standard text and calibrated instrument data.