EMC Measurement Uncertainty: How It Changes Pass or Fail
How EMC measurement uncertainty is built, which decision rule turns a reading into a verdict, how the CISPR 16-4-2 U_cispr rule works and what margin to keep.
Two accredited labs measure the same product. One reports a radiated emission peak 1.5 dB below the limit; the other reports it 1 dB above. Neither lab has necessarily made a mistake. Every EMC reading carries a measurement uncertainty of several decibels, and the verdict depends on how the decision rule handles that uncertainty. This post explains where the uncertainty comes from, the main ways it enters a pass/fail decision, including the CISPR 16-4-2 approach, and what it means for the margin you should design for.
What measurement uncertainty means in an EMC test
The expanded uncertainty U is a range around the reading within which the true value is expected to lie with a stated probability. EMC labs usually quote it with a coverage factor k = 2, which corresponds to about 95 % for a normal distribution. A reading of 38.5 dBµV/m with U = 4.8 dB therefore means that the true field strength, for this set-up, is expected to lie between 33.7 and 43.3 dBµV/m.
For radiated emission below 1 GHz, lab uncertainties of several decibels are normal, because the result depends on the antenna, the site and the cables as well as on the receiver. Conducted emission measured with a LISN usually has a smaller uncertainty, since there is no antenna and no site. The uncertainty treated in CISPR 16-4-2 is the measurement instrumentation uncertainty (MIU): it covers the test equipment, the facility and the method, not the product.
Where the decibels come from
| Contributor | Test | Why it varies | Usual treatment |
|---|---|---|---|
| Receiver or analyzer level accuracy | Conducted and radiated | Calibration tolerance, sine-wave accuracy, detector and pulse response | From calibration data, normal distribution |
| Antenna factor | Radiated | Calibration uncertainty, interpolation between calibration points, dependence on height above the ground plane | Normal distribution from the calibration certificate |
| Site or chamber imperfections | Radiated | Reflections from walls, absorbers and floor; the site is validated by normalized site attenuation (NSA) against a ±4 dB criterion | Rectangular distribution |
| Mismatch | Conducted and radiated | Reflections between antenna or LISN, cable, attenuator and receiver input, set by their VSWR | U-shaped distribution |
| LISN impedance and voltage division factor | Conducted | Tolerance of the 50 Ω / 50 µH impedance and of the calibrated division factor | Rectangular or normal |
| Antenna height, polarization and turntable steps | Radiated | The true maximum can fall between measured heights or angles | Rectangular distribution |
| Repeatability | Conducted and radiated | Scatter between repeated measurements of the same set-up | Type A evaluation, normal |
Each contributor is converted to a standard uncertainty: a value from a calibration certificate is divided by its stated coverage factor, a rectangular half-width by √3 and a U-shaped half-width by √2. The standard uncertainties, all in dB, are combined by root sum of squares, u_c = √(u₁² + u₂² + … + uₙ²), and multiplied by the coverage factor: U = 2 · u_c. Because the terms are squared, the two or three largest contributors decide the result; halving a small term changes almost nothing.
Three ways to turn a reading into a verdict
A reading and an uncertainty do not by themselves give a verdict. A decision rule is needed. ISO/IEC 17025:2017 requires a lab that issues a statement of conformity to define the decision rule, and to agree it with the customer unless it is already fixed by the specification or standard. ILAC-G8 describes the common options.
| Decision rule | Pass when | Who carries the risk | Where you meet it |
|---|---|---|---|
| Simple acceptance (shared risk) | Reading ≤ limit | Shared: a product exactly at the limit is about as likely to be truly above it as below it | Test reports where no other rule is specified |
| CISPR 16-4-2 rule | Reading ≤ limit if U_lab ≤ U_cispr; otherwise reading + (U_lab − U_cispr) ≤ limit | Shared up to U_cispr; a lab with a larger uncertainty is penalized by the excess | CISPR emission standards that refer to CISPR 16-4-2 |
| Guarded acceptance | Reading + U ≤ limit | Manufacturer: some compliant products are rejected | Customer specifications, internal release targets |
The CISPR 16-4-2 rule in practice
CISPR 16-4-2 gives a reference value U_cispr for each type of measurement, such as conducted emission with an artificial mains network or radiated emission in a given frequency range and test distance. It reflects what a lab with good equipment and practice should achieve. If the lab's own expanded uncertainty U_lab is equal to or smaller than U_cispr, the reading is compared directly with the limit. If U_lab is larger, the excess U_lab − U_cispr is added to the reading before the comparison.
An example with illustrative numbers: the limit is 40.0 dBµV/m, the reading 38.5 dBµV/m, and U_cispr for this measurement is taken as 5.2 dB. Lab A has U_lab = 4.8 dB, so it compares 38.5 dBµV/m with the limit: pass with 1.5 dB margin. Lab B has U_lab = 6.0 dB, so it adds 0.8 dB and compares 39.3 dBµV/m: still a pass, with 0.7 dB margin. Had the reading been 39.5 dBµV/m, Lab A would pass the product and Lab B would fail it at 40.3 dBµV/m. Under guarded acceptance with U = 4.8 dB, even the 38.5 dBµV/m reading would fail, because 38.5 + 4.8 = 43.3 dBµV/m.
What uncertainty means for your design margin
Lab uncertainty is only one reason a passing product can fail later. The same unit can move by several dB when cables are re-dressed, a different laptop is used as auxiliary equipment or the test is repeated in another chamber, and production units vary with component tolerances, ferrite batches and assembly. None of this is included in U_lab.
For series production, several CISPR product standards refer to the 80 %/80 % rule described in CISPR 16-4-3: with 80 % confidence, at least 80 % of the units produced must comply. A type test on one sample is the normal route to market, but an authority may later test units taken from the market, and a single sample measured close to the limit leaves little room for production spread.
Many teams therefore aim for a margin of several decibels in the final scan, often around 6 dB for radiated emission, and treat anything within 3 dB as a risk to understand rather than a pass to celebrate. On a pre-compliance bench the uncertainty is larger again, because the site is not validated and the probes and antennas may not be calibrated. Use the bench mainly to compare readings before and after a change, and keep a wider margin on absolute values.
Immunity tests: uncertainty of the applied level
In immunity testing the uncertainty concerns the disturbance applied to the EUT, not a reading compared with a limit. In a radiated immunity test to IEC 61000-4-3, for example, the field is calibrated over a uniform field area where it must lie within 0 to +6 dB of the nominal value at 75 % of the calibration points, so the EUT can see more than the nominal level at some positions. The test is run at the specified level, the lab states the uncertainty of that level, and the verdict comes from the performance criterion (A, B or C) of the product standard.
Common mistakes
- Adding U to every reading by habit when the applicable standard uses the CISPR 16-4-2 rule, or forgetting the correction when the lab's U_lab is above U_cispr.
- Blaming one lab when two labs differ by 3 or 4 dB at one frequency. Their combined uncertainties plus differences in cable layout and auxiliary equipment easily explain that.
- Assuming U covers the EUT set-up. Moving a cable by 10 cm can change a radiated peak by more than the lab's whole uncertainty.
- Taking the margin from a fast prescan with coarse frequency steps or few antenna heights. A peak missed between steps is not covered by the uncertainty budget; the final measurement must maximize each critical frequency.
- Not agreeing the decision rule before the test. ISO/IEC 17025 requires it to be defined, and a guarded-acceptance clause in a customer specification can turn a pass into a fail.
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.