Berkeley Nucleonics Corporation
ICX-FieldHawk technical paperBNC-AN-109 Rev AAugust 2026

EMC pre-compliance testing on your own bench

What CISPR specifies, which parts of it a general-purpose analyzer can reproduce, and how to make a bench scan that predicts a chamber result.

Abstract

A first radiated emissions test fails often enough that most teams budget for it. The expensive part is not the chamber day, it is the loop between the failure and the retest. This paper sets out what CISPR 16-1-1 specifies for a scan: five bands, four resolution bandwidths defined at 6 dB rather than 3 dB, detector time constants that govern dwell, and an overload factor that governs headroom. It derives the corrections that turn an indicated level into a field strength, works the noise floor and common-mode current arithmetic, and quantifies what a bench result may claim. This paper describes pre-compliance practice with a spectrum analyzer, not a certified CISPR receiver mode.

Who this paper is for

Hardware and compliance engineers who own a product's first emissions test and want the failures found before the chamber is booked. Test engineers building a repeatable pre-compliance scan. Engineering managers deciding whether a bench, a chamber day, or both is the cheaper route to a certificate.

Contents
  1. What a failed emissions test costs
  2. What the standard specifies
  3. Why a scan finds nothing and reports it as a pass
  4. Requirements for a pre-compliance bench
  5. Meeting the requirements, and what it costs
  6. Measurement practice
  7. What the bench cannot tell you
  8. Summary and selection
  9. Definitions, symbols and further reading

1What a failed emissions test costs

An accredited 3 m semi-anechoic chamber bills 2,000 to 4,000 US dollars a day, and a first radiated scan takes two or three of them [generic illustrative range]. The larger cost is the loop: the product fails at 240 MHz, somebody fits a ferrite, and the chamber is rebooked three weeks out. Two loops put a launch two months late.

The objection is that a bench is not a chamber. Correct, and not the end of it. A bench scan is not trying to produce a number a certification body will accept. It answers the three questions that set the loop count: which frequencies sit close to a limit, what is radiating them, and whether last night's change helped. None needs an accredited site. All three need the measurement made as the standard defines it, because a scan at the wrong bandwidth is not pessimistic, it is wrong.

2What the standard specifies

CISPR 16-1-1 describes a measuring receiver, not a spectrum analyzer, and it pins down what an analyzer leaves to the operator: band boundaries, resolution bandwidth, detector time constants and overload factor. Table 1 collects them.

Bandwidth breaks bench scans twice. CISPR specifies it at the −6 dB points; analyzers specify it at the −3 dB points, a narrower filter for the same number. And a narrowband carrier reads the same at any bandwidth wide enough to hold it, while a clock edge does not, because a broadband emission follows bandwidth in volts rather than power:

B3 = B6 / √2     ΔL = 20 log10(B2 / B1)
(1)
where
  • B3, B6 is filter width at 3 dB and 6 dB, in hertz
  • B1, B2 is the bandwidth used and the bandwidth the standard specifies, in hertz; 9 kHz and 120 kHz for the band B and band C settings, so ΔL is the correction to add and Figure 1 plots its negative, the error made
  • ΔL is change in indicated level, in dB
The 22.5 dB mistake

Scan band C at the band B bandwidth, which is what happens when the conducted setup is reused, and a broadband emission reads 20 log10(120 kHz / 9 kHz) = 22.5 dB low. A product 12 dB over the limit is displayed 10.5 dB under it. Not lost precision. A reversed verdict.

-60-40-2002040200 Hz1 kHz9 kHz120 kHz1 MHz10 MHz9 kHz in band C: 22.5 dBlowBand C, correct3 dB RBW at 120 kHz: 3.0dB highResolution bandwidth, at the 6 dB pointsError in indicated level (dB)Broadband emissionNarrowband carrierIllustrative. Curves are computed from the stated model, not measured data.
Figure 1. Bandwidth error is signed, and it lands only on broadband emissions: too narrow and a failing product passes, too wide and a passing one fails, while the narrowband carrier reads the same at every setting wide enough to hold it. Typing the CISPR 6 dB number into a 3 dB field over-reads by 3.0 dB.Derived

Figure 1 plots the error against bandwidth, and the sign is the thing to carry away. Table 1 then gives the settings the standard actually specifies, band by band.

Table 1. Columns three, seven and eight are from CISPR 16-1-1; the rest are derived from equation (1). The four bandwidths in column four are the whole of what makes a bench scan comparable to a laboratory scan. An instrument that cannot be set to 84.9 kHz cannot produce a CISPR-comparable band C or D measurement, and no post-processing recovers it, which is why a continuously settable resolution bandwidth matters more here than sensitivity does.
BandRange6 dB RBW3 dB equiv.StepPointsτc/τd/meter (ms)Overload
A9 to 150 kHz200 Hz141 Hz100 Hz1,41045 / 500 / 16024 dB
B0.15 to 30 MHz9 kHz6.4 kHz4.5 kHz6,6331 / 160 / 16030 dB
C, D30 MHz to 1 GHz120 kHz84.9 kHz60 kHz16,1671 / 550 / 10043.5 dB
E1 to 6 GHz1 MHz707 kHz500 kHz10,000nonen/a
Verify: bandwidth ranges permitted in bands A and E; post-detector overload factors.

From an indicated level to a field strength

An analyzer displays the level at its connector; a limit is field strength at a distance. Everything between is a signed correction:

E = Vind + AF + Lc − Gp
(2)
where
  • E is field strength at the measurement distance, in dBµV/m
  • Vind is level indicated at the analyzer input, in dBµV
  • AF is antenna factor, in dB(1/m); 17.5 at 240 MHz and 25 at 1 GHz [generic illustrative values]
  • Lc is cable and pad loss, in dB; 2.2 and 4.0 [generic illustrative values]
  • Gp is preamplifier gain, in dB; 30 [generic illustrative value]

The conducted case has the same shape, with the mains network voltage division factor replacing the antenna factor (Figure 2). Three terms vary with frequency, so a mid-band constant is decibels wrong at the edges.

fieldEquipmentmains portMains networkdivision factorPad and cablelossAnalyzerband B, 6.4 kHzEquipmentradiatedAntennaantenna factorCable and padlossPreamplifiergainAnalyzerband C, 84.9 kHzThe preamplifier buys sensitivity and spends headroom.
Figure 2. Four frequency-dependent corrections across the 16,167 points of bands C and D cannot be done by hand. They live in the instrument or nowhere.Schematic

3Why a scan finds nothing and reports it as a pass

The second failure has nothing to do with amplitude: the emission was not there when the analyzer was. A stepped scan visits each cell for the fraction of the sweep set by bandwidth over span:

td = Tsweep · B / S,   p = (td + te) / Te,   Pn = 1 − (1 − p)n
(3)
where
  • td, S, n is cell dwell; span, 970 MHz; sweeps in max hold
  • te, Te is event duration and period, 5 ms and 2 s

Across 30 MHz to 1 GHz at 120 kHz with a one second sweep [generic illustrative value], td = 1 s × 120 kHz / 970 MHz = 124 µs, so p = (124 µs + 5 ms) / 2 s = 2.562 × 10−3. One sweep finds a 5 ms commutation burst a quarter of one percent of the time (Figure 3).

020406080100Emitter5 ms burst, 2 s periodStepped scan124 us in the affected cell, once per sweepburst missedmissed againGap-free windowparked, 100 MHz at onceboth bursts measured wholeTime, 100 units = 4 sThe scan is in the right cell for 124 microseconds of every second, and the burst lasts 5 milliseconds once every two.
Figure 3. The geometry behind the arithmetic. The stepped scan is not slow, it is elsewhere: it visits the affected cell often but for 124 microseconds at a time, and the burst it is looking for arrives when it is somewhere else. Nothing about sweeping faster changes that, because a faster sweep shortens the dwell in the same proportion. Drawn for legibility rather than to scale: the true ratio is 124 microseconds of dwell inside a one second sweep against a 5 millisecond burst, which no drawing can show and the arithmetic below can.Schematic

Figure 3 is the same statement drawn rather than computed, and it is worth looking at before Figure 4, because that curve is a consequence and this is the cause.

0204060801001 s10 s1 min5 min15 min60 min14.2%53.7%99.0% at 30 minElapsed max hold timeProbability the event has been captured (%)Gap-free captureStepped, 5 msStepped, 200 us99 percentIllustrative. Curves are computed from the stated model, not measured data.
Figure 4. Thirty minutes of max hold to reach 99 percent confidence on a 5 ms emission that repeats every two seconds, or a single pass with gap-free capture. A one minute max hold finds it 14 percent of the time, and the faint trace, a 200 microsecond burst on the same period, has reached only 44 percent after a full hour and needs 7.9 hours for 99. Most pre-compliance scans that found nothing were not measurements of a clean product. They were measurements of too little time.Derived

4Requirements for a pre-compliance bench

None of the above names an instrument. Table 2 converts it into six requirements.

Table 2. Six requirements, derived rather than asserted. R1 and R4 are the two a general-purpose analyzer meets only if it was selected with them in mind.
RequirementThe numberFrom
R1Bandwidth at the CISPR 3 dB equivalents141 Hz, 6.4 kHz, 84.9 kHz, 707 kHzEq. (1)
R2Coverage of the applicable bands9 kHz to 1 GHz; 6 GHz above 108 MHz clocksBands A to E [verify]
R3Step at half bandwidth, dwell at the meter constant16,167 points, 100 msTable 1
R4Headroom for the overload factor43.5 dB, bands C and DTable 1
R5Corrections applied per pointFour termsEq. (2)
R6A capture that does not rely on luck99 percent in one passEq. (3)

5Meeting the requirements, and what it costs

The ICX-FieldHawk family is a general-purpose real-time analyzer, not a compliance receiver. R1 and R2. The datasheets publish a swept resolution bandwidth of 1 Hz to 10 MHz and coverage of 9 kHz to 9.5 GHz on the ICX-090 models or 40 GHz on the ICX-400 models, so all four CISPR equivalents fall inside. Whether that range is continuously settable, and 84.9 kHz can be set exactly, is not stated and is marked verify below.

R3 and R5. A CISPR-shaped scan is built, not selected. SCPI is standard, with C, Python, MATLAB, Qt and LabVIEW on Windows and Linux: a short script steps center frequency in 60 kHz increments across band D, 300 MHz to 1 GHz, holds 100 ms and records the peak. Antenna Factor, Amplitude Offset and amplitude correction are published functions on all three datasheets, so SpectraCore applies equation (2) per point from the antenna table and the measured cable loss. Note what that dwell costs: at 100 ms a point, a compliant pass over bands C and D takes 27 minutes. The one second sweeps of the intercept arithmetic above are the search, and the 100 ms dwell is the confirmation of what the search found. Running the confirmation scan as a search is the commonest way a bench turns into an afternoon.

Max hold against clear-write over a 100 MHz span, with spectrogram
Figure 5. Max hold against clear-write, 100 MHz span at 30 kHz resolution bandwidth, captured at 2.44 GHz rather than in band D because the traffic there is denser and the effect is identical. The upper trace is what accumulation buys, the lower one is a single instant, and the gap between them, 20 dB and more in places, is what a single sweep reports as absent.Measured

Figure 5 is the same effect on a screen, at a different frequency and bandwidth. A scan reported from a clear-write trace reports one instant per point, and everything intermittent is missing from it by construction.

Real-time density display with spectrogram
Figure 6. Real-time density with its spectrogram on an ICX-FieldHawk, 2.389 to 2.491 GHz in a 101.56 MHz span at 60.3 kHz resolution, outside every CISPR band this paper scans. Density accumulates every transform rather than the peak of a sweep, so an intermittent emitter resolves into a persistent trace with its duty cycle visible.Measured

R6. Figure 6 answers the intercept problem. The published engine is an FPGA transform with no missing samples across 100 MHz of analysis bandwidth, and its relation is published too: 100 percent probability of intercept equals 2 × N × D × 8 ns, or 32.768 µs at 2048 points. A 5 ms burst is 150 times that, so it is measured at true amplitude on the first pass. The window is 100 MHz wide, so band D, 300 MHz to 1 GHz, takes seven placements. Put equation (3) to that case and the comparison stops being rhetorical: seven placements parked for one event period each cover the band in 14 seconds at a probability of one, against the 1,795 sweeps and thirty minutes the stepped scan needs for 99 percent.

Worked example: what the bench can see

Referred to field strength, the noise floor is:

En = −174 + NF + 10 log10B + 107 + AF
(4)
where
  • −174 is thermal noise density at 290 K, in dBm/Hz; 107 converts dBm to dBµV in 50 Ω
  • NF is noise figure from antenna terminals to display, cable included; 15 dB for the analyzer alone [generic illustrative], consistent with the published −159.9 dBm/Hz at 1 GHz, RBW 1 kHz

In band D at 240 MHz, 10 log10(84,900) = 49.3 dB. With 2.2 dB of cable ahead of a 15 dB analyzer the system noise figure is 17.2 dB, so En = −174 + 17.2 + 49.3 + 107 + 17.5 = 17.0 dBµV/m against a 46 dBµV/m FCC Class B limit at 3 m: 29 dB of clearance. A 3 dB, 30 dB preamplifier [generic illustrative values] takes the system noise factor to 1.66 × (1.995 + 30.62/1000) = 3.362, or 5.3 dB, buying 11.9 dB on the 17.2 dB bare figure. Put the preamplifier at the antenna instead of at the analyzer and the cable stops costing anything at all: 3.1 dB, which buys 14.1 dB.

It also spends headroom, which is R4 arriving as a bill. At +10 dBm output-referred 1 dB compression [generic illustrative value] it compresses at −20 dBm in, or 87 dBµV. Band C wants 43.5 dB of overload factor, so the highest uncompressed level is 43.5 dBµV, or 63.2 dBµV/m at the antenna, 17 dB above the limit. Measure a badly failing product with the preamplifier out. That ceiling assumes the preamplifier at the analyzer; move it to the antenna and it sees the extra 2.2 dB, so the ceiling falls to 61.0 dBµV/m. The antenna position buys 2.2 dB of noise figure and spends 2.2 dB of headroom, which is the same trade run in both directions.

The same arithmetic in band E at 1 GHz gives 35.5 dBµV/m bare against a 54 dBµV/m limit, and the conducted scan sits nearly 60 dB below its own. Sensitivity rarely fails a pre-compliance bench. R4 and R5 decide whether a scan is trustworthy.

6Measurement practice

  1. Start with the conducted scan. A mains network, a 10 dB pad and 150 kHz to 30 MHz find much of what fails a mains-powered product. If yours is a mains box with no radio and one short cable, that scan and a limit line may be all the pre-compliance you need.
  2. Use a current probe before an antenna. Cable common mode dominates below 1 GHz. Rearranging the short-radiator relation, ICM = 2rE / (μ0fL): at 200 MHz on a 1 m cable against the Class B limit of 150 µV/m at 3 m, that is (2 × 3 × 150 × 10−6) / (1.2566 × 10−6 × 2 × 108) = 3.6 µA, or 11.1 dBµA. Through a 15 dBΩ transfer impedance [generic illustrative value] that is 26.1 dBµV, well clear of the noise floor. Four microamps of common mode on a one meter cable fails Class B, and a clamp finds it.
  3. Hold the near-field probe still. Loop field falls as the cube of distance, so 5 mm to 10 mm changes the reading by 60 log10(2) = 18.1 dB, more than any margin you are chasing. Near-field probing ranks sources, it does not measure them.
  4. Take the dwell from Table 1 and the max hold from Figure 4, with the peak detector for both. Thirty minutes if anything intermittent is suspected. Quasi-peak never exceeds peak, so a peak scan that clears the limit clears it under any detector. Keep one scan with the equipment off, and give the analyzer its published 10 minutes at 25 °C first.

7What the bench cannot tell you

Uncertainty separates a bench result from a certificate; Table 3 compares two benches.

Table 3. Combined in quadrature and expanded at k = 2, discipline is worth 6.2 dB, more than most products carry in margin. The site term dominates both columns, so a ground plane and a fixed geometry beat an instrument upgrade.
ContributionCareful ±dBuiCasual ±dBui
Receiver accuracy, antenna factor2.0, 1.31.382.0, 3.52.32
Cable loss and mismatch0.6, 1.20.922.0, 2.52.12
Site, height and polarization3.01.737.04.04
Preamplifier and repeatability1.0, 1.61.702.0, 3.03.21
Expanded, k = 2U = 5.9U = 12.1
Illustrative. Verify the CISPR 16-4-2 figure against the current edition.
The honest limitation

No EMC measurement function appears in any published Berkeley Nucleonics specification table, and nothing here is claimed for one, though a pre-compliance software option was released in August 2026 (verify) and has not reached those tables yet; the quasi-peak detector of Table 1 is not published on this family. Everything above uses general-purpose datasheet functions, and a peak scan bounds a quasi-peak result rather than replacing it. That is what makes peak scanning safe for pre-compliance and wrong for compliance.

Three further limits. A bench gives no legally defensible number, because the site is not validated. It does not reproduce a 10 m open area result by arithmetic: extrapolating by 20 log10(10/3) = 10.46 dB assumes a far-field single-source geometry a real product lacks below 100 MHz. And 5.9 dB of expanded uncertainty means a reading 4 dB under the limit is not a pass. Class A equipment gets 10 dB more room, so this bench suits industrial products best.

Field evidence

An honest note on maturity. An EMC pre-compliance software option was released on this platform in August 2026, and it appears in no published specification table yet, so nothing in this paper rests on it: every scan above is built from general-purpose analyzer functions that are published. Nor is there deployment history to report, because no bench of ours has yet run a product through a chamber afterwards to score the prediction. Every requirement is derived from CISPR 16-1-1 and evaluated against the base instrument's published figures, and the arithmetic in sections 2 to 4 is the part that does not change whatever SpectraCore becomes.

8Summary and selection

Nothing above is a compliance measurement. The scan is shaped like CISPR's, the instrument is not a CISPR receiver, and the certificate still comes from an accredited site. Table 4 collects the published values with their conditions.

Table 4. Published values with their conditions.
QuantityRelation or valueCondition
CISPR 3 dB equivalents141 Hz, 6.4 kHz, 84.9 kHz, 707 kHzGaussian filter
Swept bandwidth, coverage1 Hz to 10 MHz; 9 kHz to 9.5 or 40 GHzsettability marked verify
Amplitude accuracy, max CW input±2.0 dB to 9.5 GHz, ±3.0 above; +23 dBm10 min warm-up, 25 °C, spur reject; preamp off
100 percent intercept2 × N × D × 8 ns32.768 µs at N = 2048, D = 1
Source: the ICX-FieldHawk datasheets.

Bands A to D cover a mains-powered product with clocks below 108 MHz, and the module form sits beside the host running the script. Above 108 MHz, band E to 6 GHz is mandatory [verify the cap]; the handheld and rugged forms take that script outdoors.

9Definitions, symbols and further reading

Overload factor
The margin in dB by which an impulse peak may exceed the indicated level without any stage compressing.
Quasi-peak detector
A detector with defined charge, discharge and meter time constants, so its reading depends on repetition rate as well as amplitude. It never exceeds a peak detector.
Symbols used in this paper.
SymbolMeaningUnits
EField strength at the measurement distancedBµV/m
EnNoise floor referred to field strengthdBµV/m
VindLevel indicated at the analyzer inputdBµV
B1, B2Bandwidth used; bandwidth the standard specifiesHz
AF, Lc, Gp, NF, FAntenna factor; cable loss; preamplifier gain; noise figure and factordB(1/m), dB
B3, B6, S, τc, τd, td, te, Te, r, LBandwidth at 3 dB and 6 dB; span; quasi-peak constants; dwell; event duration and period; distance and radiator lengthHz, s, m
ICM, ZT, μ0Common-mode current; transfer impedance; permeability of free spacedBµA, dBΩ, H/m
p, Pn, n, N, D, uc, U, kIntercept probability and cumulative form; sweeps; transform size; decimation; uncertainty; coverage factordB, dimensionless
AF  antenna factor
AMN  artificial mains network
CISPR  Comité International Spécial des Perturbations Radioélectriques
EMC  electromagnetic compatibility
RBW  resolution bandwidth
SCPI  standard commands for programmable instruments

Further reading

If you are building a pre-compliance bench and want the bandwidth, dwell and headroom arithmetic worked against your own product, our application engineers would be glad to help.

Verification note

The following values in this paper are not yet confirmed against a published Berkeley Nucleonics datasheet and are marked verify in the text. They must be confirmed before this paper is released.

  • Whether the ICX-FieldHawk resolution bandwidth is continuously settable across its published 1 Hz to 10 MHz range, so 84.9 kHz and 141 Hz can be entered exactly. R1 turns on the answer.
  • Standard editions: CISPR 16-1-1, 16-2-3, 16-4-2, CISPR 32; ANSI C63.4 as incorporated in 47 CFR Part 15; the 6 GHz cap; the CISPR 16-4-2 figures; §15.109(g); bandwidth ranges in bands A and E; post-detector overload factors.
  • An EMC pre-compliance option was released in August 2026 but appears in no published specification table, so no EMC function, quasi-peak detector or limit-line set is claimed here and nothing above depends on one. Every instrument parameter in the worked examples is generic and illustrative, labeled at each use.
  • This paper reports no deployment history and no measured pre-compliance result. The two instrument captures are general-purpose measurements rather than band C or D scans, and their conditions are stated in their captions.