Calibration and measurement uncertainty (GUM, CISPR)

Author Hugues Orgitello Last updated 27 May 2026 ~14 min read

Guide, calibration and measurement uncertainty

Every EMC or radio test report ends with a number plus an uncertainty: 47.3 dBuV/m with U equal to 5.2 dB at k=2. Marginal pass/fail decisions, contractual disputes and post-launch surveillance findings often hinge on what that U value means, how it was built, and whether it was used correctly when the measurement was compared to a regulatory limit. This guide walks through the GUM (JCGM 100:2008) framework, the CISPR 16-4-2 budget references for EMC, the ISO/IEC 17025 traceability and decision rule requirements, and how an engineer or lab QA reviewer can read a budget annex critically, challenge a borderline conformity statement, or plan engineering margin under a regulatory limit.

Why uncertainty matters for certification

A test report without a stated uncertainty is, from an ISO/IEC 17025:2017 standpoint, incomplete. Three operational consequences flow from this.

• A measurement compared to a limit without uncertainty cannot support a defensible conformity statement under clause 7.8.6.
• A regulator confronted with a borderline level (within a few dB of the limit) will look at the budget annex first.
• An engineering team planning hardware margins needs to know how much room the test campaign actually leaves above and below the limit, not the limit minus the nominal measurement.

The framework that anchors all this is the GUM, jointly published by BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP and OIML as JCGM 100:2008. Its vocabulary counterpart is JCGM 200:2012, the VIM. CISPR 16-4-2 then applies GUM to EMC test methods; ETSI TR 100 028 does the same for radio testing.

GUM framework, the essentials

The GUM describes how to express, evaluate and combine uncertainty contributors around a measurand.

Type A and Type B evaluation

Type	Source	Standard uncertainty computation
Type A	Repeated observations on the measurand	Experimental standard deviation of the mean over n runs
Type B	Calibration certificate, datasheet, manufacturer tolerance, handbook value, judgement	Half-width divided by the coverage factor of the assumed distribution (1 for normal at 1 sigma, square root of 3 for rectangular, square root of 6 for triangular, square root of 2 for U-shaped)

The labels Type A and Type B describe how the contribution was evaluated, not whether it is random or systematic. A repeatability evaluated by twenty runs is Type A; the same physical effect characterised by a manufacturer datasheet is Type B. The GUM is explicit that both are treated identically in the budget.

Combined standard uncertainty

The combined standard uncertainty u_c is built by quadratic summation of the standard uncertainty contributors u_i, weighted by their sensitivity coefficients c_i (the partial derivative of the measurand with respect to the input quantity):

u_c squared equals the sum over i of (c_i times u_i) squared, plus 2 times the sum of covariance terms when contributors are correlated.

In EMC, contributors are largely treated as uncorrelated, so the cross terms collapse to zero. This is an explicit assumption documented in CISPR 16-4-2.

Expanded uncertainty and coverage factor

The expanded uncertainty U reported on the certificate is:

U equals k times u_c

with k=2 corresponding to a confidence level of approximately 95% under the assumption of an approximately normal output distribution (justified by central limit theorem when several contributors of similar magnitude combine). k=1 gives roughly 68%, k=3 gives roughly 99.7%. The vast majority of EMC and radio reports use k=2.

A report stating U equals 5.2 dB without a coverage factor is ambiguous, even though the convention is overwhelmingly k=2. ISO/IEC 17025 expects k to be explicit.

CISPR 16-4-2, the EMC budget reference

CISPR 16-4-2:2011, amended by A1:2014 and A2:2018, is the IEC standard that codifies how to apply GUM to canonical CISPR test methods. It defines a reference budget U_cispr per test method, computed by CISPR with a defined model and contributor list. Laboratories then compute their own U_lab and compare it to U_cispr.

Reference budgets

Test method	Frequency range	U_cispr (k=2), order of magnitude
Radiated emissions, OATS or SAC	30, 1000 MHz	About 5.2 dB
Radiated emissions above 1 GHz	1, 18 GHz	About 5.7 dB
Conducted emissions, LISN	0.15, 30 MHz	About 3.6 dB
Disturbance power, absorbing clamp	30, 300 MHz	Around 4.6 dB

These round figures come from CISPR 16-4-2 and are the reference values; the laboratory's actual U_lab can be smaller or larger depending on its real budget.

U_lab versus U_cispr, the decision mechanism

The standard sets up a comparison rule that resolves the asymmetry between labs with different capabilities.

Situation	Action on the measured value before limit comparison
U_lab <= U_cispr	None. The measurement is compared directly to the limit (shared risk)
U_lab > U_cispr	The excess (U_lab minus U_cispr) is added to the measured value before comparison

The mechanism guarantees that a lab less accurate than the CISPR reference cannot transfer that loss to the customer by issuing borderline pass verdicts. Conversely, a lab that does better than the reference is not penalised, because no margin is subtracted from the limit.

Implications for the customer

A customer reviewing a CISPR-style report should check three points.

• U_lab stated: the report or budget annex must give U_lab for the test concerned, not just the generic 5.2 dB.
• U_cispr reference cited: the comparison context must be explicit.
• Limit comparison method: whether the lab added (U_lab minus U_cispr) when applicable, or applied shared risk.

A pass statement at limit minus 1 dB in a configuration where U_lab equals 6 dB and U_cispr equals 5.2 dB means the limit comparison should have been done after adding 0.8 dB to the measured value. If the lab did not do that, the verdict is questionable.

ETSI TR 100 028 for radio

ETSI Technical Report TR 100 028 (in two parts) is the radio counterpart of CISPR 16-4-2. It applies GUM to mobile radio test methods (output power, modulation accuracy, spurious emissions, occupied bandwidth, adjacent channel power) and uses the same shared-risk logic. A radio test report under EN 300 328, EN 301 893 or EN 301 489 typically references TR 100 028 for its uncertainty budgets. The mechanism (laboratory budget compared to a reference, excess added before limit comparison) is identical.

Typical EMC uncertainty contributors

A CISPR-compliant radiated emission budget breaks down into a documented list of contributors, each evaluated Type A or Type B, each with an assumed distribution.

Contributor	Typical magnitude	Distribution	Notes
Receiver or spectrum analyser accuracy	1, 1.5 dB at k=2	Normal	From the calibration certificate
Antenna factor (calibration)	0.5, 1.5 dB at k=2	Normal	Datasheet plus calibration certificate
Antenna factor height variation	0.5, 2 dB	Rectangular	Variation 1, 4 m scan, biconical and log-periodic
Antenna directivity, cross-polarisation	0.5, 1 dB	Rectangular	Polarisation imperfection
Cable loss	0.2, 0.5 dB	Normal	Calibrated at install, drift on connectors
Mismatch antenna to cable	0.3, 1.5 dB	U-shape	Depends on VSWR at the interface
Mismatch cable to receiver	0.2, 0.5 dB	U-shape	Pre-selector or LISN inserted modifies this
Site attenuation (NSA)	1, 2 dB	Normal	Validated by NSA measurement on a reference site
Climatic conditions	0.1, 0.3 dB	Rectangular	If within calibration certificate range
EUT positioning repeatability	1, 3 dB	Normal	Largest variable contributor for complex EUTs
Cable layout repeatability	0.5, 2 dB	Normal	LISN and motherboard cables on a tabletop EUT

The quadratic sum of these terms, multiplied by k=2, yields U_lab. In practice EUT positioning and mismatch are the two terms most often dominating the budget for a small EUT in a SAC.

Mismatch, the term most often misread

The mismatch contribution between two elements with reflection coefficients Gamma_1 and Gamma_2 follows a U-shaped distribution centred on 0 dB, with half-width:

Half-width in dB approximately equals 20 log_10 (1 plus the absolute value of Gamma_1 times Gamma_2)

For an antenna with VSWR equal to 2 (Gamma equal to 0.33) connected directly to a cable with VSWR equal to 1.5 (Gamma equal to 0.2), the worst-case mismatch term is about 0.6 dB. With a U-shape, the standard uncertainty is half-width divided by square root of 2, so 0.42 dB Type B. Above a few hundred megahertz, antennas with VSWR up to 2.5 are common, and mismatch quickly becomes a leading contributor. Adding an attenuator at the antenna port (10 dB pad) drops Gamma seen at the cable by 20 dB and crushes the mismatch term, at the cost of dynamic range.

ISO/IEC 17025 traceability requirements

ISO/IEC 17025:2017 clause 6.5 requires every measurement underpinning a conformity decision to be traceable to the SI through an unbroken chain of calibrations.

Traceability chain

Level	Actor	Document
SI definitions	BIPM	Mise en pratique, KCDB
National primary standards	NMI (NIST, NPL, PTB, LNE, INMETRO, KRISS, NIM)	Primary standard certificate
Secondary standards (lab calibration body)	Accredited calibration laboratory under ILAC MRA	Calibration certificate with U and CMC
Working instruments (EMC or radio lab)	Accredited testing laboratory	Internal calibration record, recall date

Every link in the chain must show:

• the measurand (what is calibrated),
• the reference standard used (model, serial, previous calibration date),
• the method (calibration procedure or reference document),
• the environmental conditions (temperature, humidity, pressure),
• the result with the expanded uncertainty U at k=2,
• the due date for the next calibration.

A certificate that omits any of these is not traceable for the purposes of clause 6.5.

CMC versus MU

Term	Definition	Where it lives
CMC (Calibration and Measurement Capability)	The smallest uncertainty the lab can routinely achieve for a given measurand in its scope	ILAC MRA accreditation scope, BIPM KCDB for NMIs
MU (Measurement Uncertainty, on a specific report)	The actual uncertainty for the specific test configuration measured	Test or calibration report

The MU is typically equal to or larger than the CMC for the same measurand, because the report reflects contributors absent from the CMC evaluation (real EUT geometry, cable routing, climatic excursion). A report whose stated MU equals the CMC across all tests is statistically suspect; a report whose MU is many dB worse than the CMC, without explanation, indicates a configuration or methodology problem.

ILAC G8 decision rules

ISO/IEC 17025:2017 clause 7.8.6 requires the lab to record the decision rule applied when a conformity statement is issued. ILAC G8:09/2019 codifies four canonical rules.

Decision rule	Logic	Effect on pass criterion
Simple acceptance, shared risk	The measured value is compared directly to the limit (assuming U_lab <= U_cispr in the EMC case)	Pass if measured <= limit
Simple acceptance with a guard band of U_lab	A guard band equal to U_lab is subtracted from the limit	Pass only if measured <= (limit minus U_lab)
Non-binary statement	Four-way decision: pass, fail, conditional pass, conditional fail	Conditional pass when measured + U_lab > limit but measured < limit
Bayesian decision	Probabilistic decision with prior	Used in high-stakes calibration, rarely in EMC

The customer contracts the decision rule with the lab before the campaign. A test plan that does not specify the decision rule defaults to shared risk under CISPR 16-4-2 for EMC tests, or to the rule embedded in the relevant ETSI standard for radio tests.

Engineering margin

For an internal engineering pass/fail decision (not a regulatory statement), a design team typically targets the measured level to sit at the limit minus U_lab, providing about 95% probability that the true value is below the limit. For safety-critical or surveillance-sensitive devices, the margin is often increased to limit minus 2 times U_lab (about 99% protection on the assumption of a normal output distribution).

Choosing the right distribution for Type B contributors

The choice of the assumed distribution for a Type B contributor is rarely random; it follows the physical nature of the source.

Source	Recommended distribution	Why
Calibration certificate giving U at k=2	Normal, with the half-width equal to U/2 used as the standard uncertainty	The certificate already implies a normal output by stating k=2
Datasheet tolerance with no specific assumption	Rectangular, half-width divided by square root of 3	The conservative default when the manufacturer states only a min and max
Resolution of a digital display	Rectangular, half-width equal to half the last digit	The true value is anywhere within the last digit window with uniform probability
Triangular tolerance (manufacturer states a central value with falling probability)	Triangular, half-width divided by square root of 6	Used when datasheets state a typical value with extreme bounds
Mismatch term, RF interfaces	U-shape (arcsine), half-width divided by square root of 2	The standing wave nature of the mismatch produces a bimodal distribution at the worst-case bounds
Effects derived from a calibration curve fit	Normal, half-width from the residual fit error	The curve fit itself is a statistical operation

Picking the wrong distribution silently moves the standard uncertainty by a factor of square root of 3 over square root of 2 (about 1.22) at most, but it can be enough to flip a borderline pass/fail when the budget is dominated by a single Type B term.

Inter-laboratory comparisons and proficiency testing

ISO/IEC 17025:2017 clause 7.7 requires the laboratory to monitor the validity of its results, and proficiency testing (PT) or inter-laboratory comparisons (ILC) are the canonical means. The accreditation body (UKAS, COFRAC, DAkkS, A2LA, JAB) typically requires participation at least every two to four years for each major measurand.

A proficiency testing scheme circulates a stable artefact through participating laboratories and compares the results against a reference value (assigned by a high-tier NMI or by the robust mean of the participant set). The result is expressed as a z-score:

z equals (lab result minus reference) divided by the standard deviation of the participant set

A z-score with absolute value below 2 is considered satisfactory, between 2 and 3 questionable, above 3 unsatisfactory. A lab that comes out unsatisfactory must open a root-cause investigation and may have its accreditation scope suspended for the measurand concerned. When choosing a laboratory for a marginal pass/fail dossier, asking for the latest PT z-score on the measurand of interest is a fair and informative question.

Calibration intervals and what they actually mean

A calibration certificate is valid for a stated period, typically expressed as a recommended recall interval and a due date.

Instrument	Typical interval	Drivers
EMC receiver, spectrum analyser	12 months	Internal reference oscillator drift, attenuator wear
Biconical antenna	12 to 24 months	Balun drift, mechanical wear of element joints
Log-periodic antenna	12 to 24 months	Same as biconical, plus dipole element fatigue
Horn antenna (above 1 GHz)	24 to 36 months	Lower mechanical stress, drift is slower
LISN	12 months	Internal resistor drift, isolation degradation
RF cable set	6 to 12 months	Connector wear, the most common source of out-of-tolerance drift
EMI filter and reference load	24 months	Passive elements, slow drift
EUT support, tripod, turntable	Mechanical check yearly, dimensional cal at install	Geometry and reproducibility, no electrical cal

The interval is recommended, not regulatory: the laboratory can shorten it after a known shock or excursion (transport drop, climatic event), or extend it on stable instruments with documented drift trends. ISO/IEC 17025 clause 6.4 requires a documented policy on intervals, and traceability of any deviation from that policy.

A recurring audit finding is the antenna calibration past due by a few weeks, used to issue a borderline conformity report. The lab usually argues that the drift is small; the regulator will discount the test anyway, because the certificate validity is binary.

National metrology institutes and the ILAC MRA

The traceability chain depends on the national metrology institute (NMI) chosen by the lab calibration body. The choice has practical consequences.

NMI	Country	Strength in EMC and radio metrology
NIST	United States	Reference for FCC-aligned testing, very strong RF metrology
NPL	United Kingdom	Reference for ETSI and CISPR work, antenna calibration leader
PTB	Germany	Dominant in EU laboratories, complete CISPR coverage
LNE	France	Reference for French laboratories, ANFR alignment
INMETRO	Brazil	Primary reference for Anatel-aligned testing
KRISS	South Korea	KCS and KC certification chains
NIM	China	CCC and SRRC certification chains
NMIA	Australia	RCM and ACMA-aligned testing

All listed NMIs are signatories of the BIPM CIPM MRA, which makes their calibration certificates mutually recognised. At the laboratory level, the ILAC MRA recognises ISO/IEC 17025 accreditations issued by signatory accreditation bodies (UKAS in the UK, COFRAC in France, DAkkS in Germany, A2LA in the US, JAB in Japan, NABL in India). A calibration certificate issued by a lab accredited by a non-MRA body is technically traceable but administratively non-recognised outside the issuing region.

Worked example, radiated emission at 100 MHz

A concrete example illustrates how the budget is built and how the comparison rule applies.

EUT measured at 100 MHz: 37 dBuV/m at 3 m, vertical polarisation, on SAC. EN 55032 Class B limit at 3 m, 30 to 230 MHz: 40 dBuV/m quasi-peak. Margin to limit: 3 dB.

The laboratory budget annex states:

Contributor	u (dB)	Distribution
Receiver accuracy	0.5	Normal
Antenna factor calibration	0.6	Normal
AF height variation	0.7	Rectangular
Cable loss	0.2	Normal
Mismatch antenna-cable	0.5	U-shape
Site attenuation NSA	0.8	Normal
EUT positioning	1.2	Normal
Climatic	0.1	Rectangular

Combined standard uncertainty u_c equals the square root of (0.5 squared plus 0.6 squared plus 0.7 squared plus 0.2 squared plus 0.5 squared plus 0.8 squared plus 1.2 squared plus 0.1 squared) = square root of 3.32 = 1.82 dB. Expanded U at k=2 = 3.64 dB.

U_lab = 3.64 dB. U_cispr at this frequency = 5.2 dB. U_lab is below U_cispr, so shared risk applies: the measured 37 dBuV/m is compared directly to the 40 dBuV/m limit. Pass.

If the same lab measured 39 dBuV/m instead, the verdict would still be pass under shared risk, even though the true value at k=2 lies between 35.4 and 42.6 dBuV/m. A producer wanting tighter protection would contract for simple acceptance with guard band: pass only if measured <= 40 minus 3.64 = 36.36 dBuV/m. Under that rule, 39 dBuV/m fails.

Reading a test report budget annex

A 17025-compliant test report typically includes an annex with the uncertainty budget. A reviewer should look at six points.

1. U_lab stated, with explicit k (k=2 is the convention), per test method.
2. U_cispr reference cited (CISPR 16-4-2 for EMC, ETSI TR 100 028 for radio).
3. Comparison rule applied (shared risk if U_lab <= U_cispr; excess added if U_lab > U_cispr).
4. Contributor list with values and distributions, allowing the reviewer to spot a missing term (mismatch, height variation).
5. Calibration due dates for instruments used (receiver, antenna, LISN, cable set), all in-date at the test date.
6. Climatic conditions during the test, within the certificate validity range.

A report failing any of these points has a defensible gap that can be raised in a customer audit or a regulator surveillance.

For the radiated emission counterpart that produces the levels whose uncertainty this guide quantifies, see radiated emissions. For the immunity counterpart whose field uniformity has its own GUM-style budget, see IEC 61000-4-3. For the chamber categories whose NSA contribution sits in the budget, see EMC chambers, SAC, FAR, OATS, reverb. For pre-compliance setups where uncertainty budgets are deliberately relaxed, see pre-compliance, TEM cell, near-field.

Frequent pitfalls

Pitfall	Consequence
Comparing measured value to limit without adding (U_lab minus U_cispr) when U_lab > U_cispr	Borderline pass declared on a non-compliant device, exposes producer to recall
Ignoring the mismatch term in non-50-ohm setups	Budget understated by 1, 2 dB in upper frequencies, undeclared risk on the report
Re-using an antenna calibration past its due date	Test invalid for traceability, regulator can reject the dossier
RBW or VBW used in measurement differs from calibration certificate configuration	Equivalence broken, undeclared term, audit finding
Climatic conditions during test outside certificate validity range	Calibration certificate not applicable, traceability chain broken
Confusing CMC and MU	Customer assumes lab capability applies to specific report, underestimates real budget
No decision rule referenced on the conformity statement	Report incomplete per ISO/IEC 17025 clause 7.8.6, audit finding
Coverage factor k not stated	U value ambiguous, defensibility weakened in dispute

Going further

• Radiated emissions, EMC test: the level measurement whose U budget this guide quantifies
• IEC 61000-4-3, radiated RF immunity: the immunity counterpart, with its own field uniformity budget
• EMC chambers, SAC, FAR, OATS, reverb: the test sites whose NSA contributes to U_lab
• Pre-compliance, TEM cell, near-field: low-budget setups where uncertainty is deliberately wider
• Glossary: definitions of GUM, VIM, U_lab, U_cispr, CMC, MU, shared risk, decision rule

See also

• PCB design for EMC: return paths, decoupling, stackup
• Antenna design and impedance matching for IoT
• Radio: RX blocking, selectivity and intermodulation tests
• Radiated emissions EMC test: pre-scan and final scan

Working on reviewing the uncertainty budget annex of an EMC or radio test report and supporting marginal pass/fail decisions for a real product? AESTECHNO can audit your design and certification plan. Get in touch →

Need help?

A 30-minute audit with AESTECHNO engineers.

AESTECHNO is the electronics design house that edits spilma. We help product teams scope their certification effort and avoid the common mistakes.

Book a free audit

Sources & references

1. JCGM 100:2008, Evaluation of measurement data, Guide to the expression of uncertainty in measurement (GUM) , BIPM JCGM www.bipm.org/en/committees/jc/jcgm/publications
2. JCGM 200:2012, International Vocabulary of Metrology (VIM) , BIPM JCGM www.bipm.org/en/committees/jc/jcgm/publications
3. CISPR 16-4-2:2011 + A1:2014 + A2:2018, Uncertainty in EMC measurements , IEC CISPR webstore.iec.ch/publication/65
4. ISO/IEC 17025:2017, General requirements for the competence of testing and calibration laboratories , ISO IEC www.iso.org/standard/66912.html
5. ILAC G8:09/2019, Guidelines on Decision Rules and Statements of Conformity , ILAC ilac.org/publications-and-resources/ilac-guidance-series/
6. ETSI TR 100 028, Uncertainties in the measurement of mobile radio equipment characteristics , ETSI www.etsi.org/deliver/etsi_tr/100000_100099/10002801/
7. BIPM Key Comparison Database (KCDB) , BIPM www.bipm.org/kcdb/