Insights · Technical note

Two detectors called "average", and the limit only means one of them

The CISPR-average detector and a spectrum analyser's average detector agree on a steady signal and part company on everything else. How the responses differ, and which one a CISPR or FCC average limit is written for.

A power supply in burst mode is measured twice on the same bench. With the analyser's average detector it clears the conducted average limit by 6 dB. With the receiver's CISPR-average detector it fails by 5 dB. Nothing drifted between the two runs. Both detectors did exactly what they are defined to do, and the definitions are 11 dB apart for that signal. The limit was written for one of them.

In short
  • The CISPR-average detector is defined in CISPR 16-1-1 clause 7: a linear average of the signal envelope, indicated through a meter with a specified time constant, and the reading is the highest meter indication during the measurement time.
  • A spectrum analyser's average is a sample average over a dwell time, in linear, RMS or logarithmic units, with no meter time constant. Trace averaging and video filtering are different things again.
  • CISPR limits mean the CISPR-average. FCC limits define their own average, a 100 ms window under 47 CFR 15.35(b), and call up CISPR receivers through ANSI C63.4 and C63.10.

What the CISPR-average detector is

CISPR 16-1-1 defines the average detector for the measurement of the average value of the envelope of the signal through the reference bandwidth: 200 Hz in band A, 9 kHz in band B, 120 kHz in bands C and D, and a 1 MHz impulse bandwidth above 1 GHz. Two things make it more than an averager. First, the average is linear, of the envelope voltage, not of its logarithm. Second, the detector output is indicated through a meter with a defined time constant, 160 ms in bands A and B and 100 ms in bands C, D and E, and the value recorded is the maximum of the meter indication during the measurement time. The meter is emulated in software in any modern receiver, but the time constant is still there, and it is what the standard verifies.

The verification is in clause 7.3.3, intermittent, unsteady and drifting narrowband disturbances. A sine wave gated on for 160 ms every 1.6 s in bands A and B, or 100 ms in bands C, D and E, shall indicate 9 dB below the same sine wave left on, within 1 dB. That number is the whole difference in one figure. A plain average of the envelope over the 1.6 s period would read 20 dB down, because the signal is on for a tenth of the time. The CISPR meter, with its 160 ms time constant, charges most of the way up during the burst and the receiver records that maximum. The detector is built to make a short, repeating narrowband disturbance count for close to what it does to a receiver, not for its duty cycle.

Narrowband signal on for 160 ms every 1.6 s, band B. Reading relative to the same signal left on. on off 1.44 s on off Envelope against time meter indication peaks near 0.35 of the steady level Reading Peak: 0 dB CISPR-average: −9 dB Linear average over the period: −20 dB Log average: lower still (noise floor)
The 9 dB figure is the requirement of CISPR 16-1-1 clause 7.3.3 for bands A and B, with a tolerance of 1 dB. The 20 dB figure is arithmetic: a tenth of the time on, averaged linearly over the whole period. A logarithmic average spends nine tenths of the period averaging the noise floor in decibels and reads lower again.

What "average" means on a spectrum analyser

An analyser offers three things under the word, and none of them has a meter time constant.

  • The average detector. Each displayed point is the average of the samples taken during the dwell time for that point. The average can be of the voltage, of the power (RMS) or of the logarithm, depending on the setting, and the result depends on how long the dwell is compared with the signal's cycle.
  • Trace averaging. Successive sweeps are averaged point by point. This reduces noise on the display. It is not a detector, and a signal that is on in one sweep and off in the next is simply diluted.
  • Video filtering. A video bandwidth narrower than the resolution bandwidth smooths the detected envelope. In linear display mode this approximates a linear average; in logarithmic mode it averages decibels, which is a different quantity for any signal that is not steady.

For a steady sine wave all of these agree with the CISPR-average, because averaging a constant returns the constant. The differences appear as soon as the signal varies within the measurement time, which is exactly the case the CISPR detector was specified for.

SignalCISPR-average, CISPR 16-1-1 clause 7Analyser average, linear voltageAnalyser average, logarithmic
Steady narrowbandSame readingSame readingSame reading
Intermittent narrowband, 160 ms on every 1.6 s−9 dB, by requirement−20 dB if the dwell covers a whole period, otherwise anything between 0 and −20 dB depending on where the dwell fallsWell below −20 dB, dominated by the noise floor during the off time
Broadband pulses, repetition frequency well below the bandwidthIndication proportional to repetition frequency, verified by clause 7.3.2Similar, if linear and the dwell spans many pulsesMuch lower; the log of a mostly-zero envelope
Drifting or modulated narrowbandMeter follows the variation; the maximum is recordedVariation smoothed away over the dwellSmoothed away, biased low

Which limit means which

An average limit is only meaningful with the detector it was written for. The two families of standards answer differently.

CISPR limits

Every CISPR product standard measures with a receiver conforming to CISPR 16-1-1, through the methods of CISPR 16-2-1 and 16-2-3. When CISPR 32, CISPR 11, CISPR 14-1, CISPR 15 or CISPR 25 print an average limit, the detector is the CISPR-average of clause 7, with the reference bandwidth of the band and the meter time constant. Below 30 MHz that is the 9 kHz bandwidth and the 160 ms meter. Above 1 GHz it is the 1 MHz impulse bandwidth, the 100 ms meter and a linear average; CISPR 16-1-1 states the logarithmic-average requirement above 1 GHz as under consideration, which means a log-average reading is not the specified quantity. CISPR 16-1-1 also defines an RMS-average detector, with its own weighting, but the product standards in common use do not yet set limits against it.

FCC limits

The FCC defines its own average in 47 CFR 15.35(b). When an emission limit is expressed as an average and the emission is pulsed, the field strength is determined by averaging over one complete pulse train, including the blanking intervals, provided the train does not exceed 100 ms. Where it does, or as an alternative, the measurement is the average absolute voltage over the 100 ms interval in which the field strength is at its maximum. The same clause sets a peak limit 20 dB above the average limit. ANSI C63.4 and C63.10, which the FCC references for the methods, call up a CISPR 16-1-1 receiver, and C63.10 additionally allows the average of some transmitters to be derived from a peak measurement with a duty-cycle factor. So an FCC average is a 100 ms window, in the worst 100 ms, made with a CISPR receiver. For most signals it lands close to the CISPR-average with its 100 ms meter, but it is a separate definition, and the certificate and the report should say which one was applied.

The other thing to know about Part 15 is where the average applies at all. Section 15.209 states that its radiated limits are based on a CISPR quasi-peak detector except in 9 kHz to 90 kHz, 110 kHz to 490 kHz and above 1 GHz, where the limits are based on an average detector. Below 1 GHz outside those two bands, an average reading against a 15.209 limit is the wrong detector, however it was averaged.

LimitRangeDetector the limit is written for
CISPR 32, 11, 14-1, 15 conducted150 kHz to 30 MHzQuasi-peak and CISPR-average, 9 kHz bandwidth, 160 ms meter
CISPR 14-1, CISPR 15 conducted9 kHz to 150 kHzQuasi-peak and CISPR-average, 200 Hz bandwidth, 160 ms meter
CISPR 32 radiated1 GHz to 6 GHzPeak and CISPR-average, 1 MHz impulse bandwidth, 100 ms meter, linear
CISPR 25 conducted and radiated150 kHz to 2.5 GHzPeak or quasi-peak, and CISPR-average, per the class tables
FCC 15.209 radiatedBelow 1 GHzCISPR quasi-peak, except 9 to 90 kHz and 110 to 490 kHz, which are average
FCC 15.209 radiatedAbove 1 GHzAverage per 15.35(b), 100 ms window, with peak 20 dB above
FCC 15.107 conducted150 kHz to 30 MHzQuasi-peak and average, measured with a CISPR receiver per ANSI C63.4; 15.35(b) governs pulsed emissions
FCC 15.247, 15.407 intentional radiatorsPer rule partAverage per 15.35(b) and ANSI C63.10, with the duty-cycle provisions of C63.10

Set-up decisions that change the answer

Choose the detector by the limit, not by the menuIf the limit is from a CISPR standard, the detector is the CISPR-average, which the instrument may label CISPR-AV, EMI average or similar. The description should mention the meter time constant. A detector labelled simply "average" is a sample average and is not the specified quantity for anything that varies.
Make the measurement time longer than the signal's cycleThe CISPR-average reading is the maximum of the meter indication during the measurement time. A burst that repeats every 2 s is missed by a 500 ms dwell. CISPR 16-2-1 and 16-2-3 expect the measurement time to cover the slowest variation of the emission, and a receiver that reports a low CISPR-average with a short dwell has not measured the disturbance.
Never average in decibelsLog averaging is a display convenience for noise. For an intermittent or pulsed emission it reads low by an amount that depends on the noise floor, which is not a property of the product. If the analyser's average is the only option, set it to linear voltage and accept that the meter time constant is still missing.
Above 1 GHz, check three settingsThe 1 MHz impulse bandwidth, a linear average and a measurement time long enough for the meter to settle. A 1 MHz resolution bandwidth with a log average and a fast sweep gives a number that looks like a CISPR-average reading and is not one.

Why this is a calibration question too

The property that separates the CISPR-average from a plain average is the meter time constant, and it is verified by the intermittent test in clause 7.3.3 of CISPR 16-1-1: the 160 ms or 100 ms gated signal that shall read 9 dB down. A receiver whose average detector has never been tested that way has an average detector of unknown type. The sine-wave amplitude test cannot tell them apart, for the same reason the two detectors agree on a steady signal. That is why a receiver calibration to CISPR 16-1-1 includes the pulse and intermittent tests, and why a spectrum analyser calibration does not settle the question. The previous note covers what such a certificate has to contain.

Where we stand

Under the accredited scope we calibrate to, CISPR detector response, pulse response and bandwidths are verified from 9 kHz to 1 GHz, including the average detector's intermittent response, and the certificate shows the measured deviation for each detector and band. The EMI receiver calibration guide lists the parameters. If a product reads differently on two detectors that are both called average, send us the two traces and the standard you are testing to, and we will tell you which reading the limit means.

Tell us what you need to test

We reply within one working day with a quote or a plan.

Contact us