SIGLENT: Home / Support / Application
How to correctly use spectrum analyzers for EMC pre-compliance tests
1   Introduction. 
2   Relevant standards. 
3   Standard related requirements. 
        3.1     Amplitude units. 
        3.2     Resolution bandwidth. 
        3.3     Frequency resolution. 
        3.4     Sweep time. 
        3.5     Detectors. 
4   Internal attenuator, pre-amplifier. 
5   Distortion considerations. 
6   General sensitivity considerations. 
        6.1     Conducted noise testing with LISN. 
        6.2     Conducted noise testing with RF current probes. 
        6.3     Radiated noise testing with TEM-cells. 
        6.4     Radiated noise testing with antennas. 
7   Input protection. 
8      History. 


1 Introduction

Spectrum analyzers with EMC pre-compliance testing capability have become very affordable in recent years. EMC pre-compliance testing capability is usually sold as “EMI-option” and offers CISPR filters and Quasi-Peak detectors in addition to the standard features of spectrum analyzers.
Spectrum analyzers offer a wide range of parameter settings and need to be set up correctly in order to make measurements as close as possible to the requirements of EMC standards. EMC standard related requirements affect correct setting of RBW filter, video bandwidth, detector type, frequency span and sweep time. Radiation limits and transducer characteristics affect settings, which are necessary to achieve a good compromise between high sensitivity and low distortion.
Measurement plots documented in this application note are created using a Siglent SSA3021X Plus, an entry level EMI- spectrum analyzer with excellent price performance ratio. 

2 Relevant standards

Several standards specify EMC test set ups and requirements for measurement equipment. Most prominent are the CISPR 16 and EN 61000-4 series. There are additional relevant standards, such as CISPR 25, Mil-461, DO 160 and more. This document is mainly focused on the CISPR 16 standard to keep this application note as compact as possible.

3 Standard related requirements

3.1 Amplitude units

 In RF applications, [dBm] is the predominant amplitude unit. [dBm] is a logarithmic power unit, which makes sense, as input and output impedance of RF building blocks are typically designed to 50 Ohm.
In EMC pre-compliance applications, impedance of EUTs and power supply sources is hardly predictable. Consequently, emission limits are predominantly specified in [dBµV] and [dBµA] amplitude units. Standardized transducers such as LISN, CDN, RF current probes and others are used to establish interfaces with defined impedance, in order to connect 50 Ohm measurement equipment. 

3.2 Resolution bandwidth

Typically, spectrum analyzers use Gaussian shaped IF filters in a 1 -3 -10 sequence, e.g. 100 Hz, 300Hz, 1 kHz, 3kHz, 10 kHz, 30 kHz…
In order to be compliant with CISPR standards, the spectrum analyzer must additionally provide so called CISPR-filters:

Besides specifying filter shape, impulse response and side lobe suppression, CISPR specifies frequency bands and the corresponding filter bandwidths that have to be used:
Frequency range CISPR filter bandwidth
9 kHz – 150 kHz 200 Hz
150 kHz – 30 MHz 9 kHz
30 MHz – 1 GHz 120 kHz
Above 1 GHz 1 MHz
The smaller the bandwidth, the lower the base noise level. You may already have observed steps in test house plots, which are caused by switching filter bandwidth:
Transition from 9 kHz to 120 kHz RBW at 30 MHz – step in base noise level

Spectrum analyzer base noise level versus resolution bandwidth

3.3 Frequency resolution

Spectrum analyzers sweep the frequency range in discrete steps. Typically, the number of frequency steps per sweep is identical with the number of display pixels in X-direction. The Siglent SSA3021X, as an example, has a resolution of 751 equidistant frequency points per sweep. Other common spectrum analyzers have 601 measurement points per sweep.
Spectrum analyzer usually power up with the sweep set to full span and RBW set to 1 MHz.
When feeding the analyzer with a signal, it may be observed that frequency and amplitude is not displayed correctly. A brief calculation and looking at the filter curves and spacing between adjacent frequency points and the reason becomes obvious. Dividing the span of 2.1 GHz by 751 frequency points results in adjacent frequency points being spaced by approximately 2.8 MHz:
Input signals may fall in between two adjacent filter curves or into the shoulder of a filter curve. Consequently, the signal is attenuated and the analyzer display shows a lower amplitude value – the measurement value is incorrect. The displayed frequency will be corresponding with the center frequency of the closest measurement frequency point and offset / incorrect as well.
Let´s take another example and look at a typical conducted emission measurement. In most cases, this measurement covers the frequency range up to 30 MHz and requires a CISPR RBW of 9 kHz. Attempting to make a full sweep across the entire 30 MHz results in a spacing of 30 MHz / 751 = 39.9 kHz. A significant part of the spectrum will not be measured at all:

In order to cover the entire spectrum within the span of a frequency sweep, CISPR 16 specifies that adjacent frequency points shall not be spaced more than half of the resolution bandwidth. In case of the example above, the spacing shall not be more than 9 kHz / 2 = 4.5 kHz.
With this information in mind, the frequency span settings have to be chosen in order to fulfill the frequency spacing and RBW specifications of CISPR 16:
Number of measurement points per sweep: 751  (Siglent SSA3021X)
Frequency range CISPR filter bandwidth Maximum frequency span
9 kHz – 150 kHz 200 Hz 75 kHz
150 kHz – 30 MHz 9 kHz 3.38 MHz
30 MHz – 1 GHz 120 kHz 45 MHz
Above 1 GHz 1 MHz 375 MHz
Consequently, a conductive emission measurement for the frequency range 150 kHz to 30 MHz has to be split into at least 29.85 / 3.38 = 9 segments with a span of 3.38 MHz.
Doing such a measurement manually, would be a tedious process. Various analyzers let you increase the default number of measurement points to a higher value. Newer analyzers also offer the capability to select standard conformant EMI measurement routines, which also ensure that adjacent measurement points have the correct frequency spacing. The disadvantage is that the resulting graph is still limited to the number of available display pixels.
Tekbox offers EMCview, an EMI measurement software, which splits the measurement into consecutive sweep segments. The measurement values of all sweeps are then stitched together to a single graph. EMCview also simplifies EMI measurements by providing a vast list of preconfigured measurements.

Conducted noise measurement with EMCview.
The 30 MHz sweep is split into 12 segments with 2.5 MHz span each

3.4 Sweep time

CISPR 16 differs between wide band and narrow band noise. Narrow band noise is typically caused by clock signals. Wideband noise is caused by data signals. As the spectrum of data signals is caused by a more or less arbitrary bit sequence, it is dynamic and wide band. Furthermore, signals may be present or not, depending on tasks running on the controller. Sweeping too fast would miss pulses and not correctly measure the wide band noise spectrum.
Consequently, CISPR 16 specifies minimum sweep times, depending on frequency range and detector:
Frequency range Peak detector Quasi-peak detector
9 kHz – 150 kHz 100 ms / kHz 20 s / kHz
150 kHz – 30 MHz 100 ms / MHz 200 s / MHz
30 MHz – 1 GHz 1 ms / MHz 20 s / MHz
CISPR 25 specifies minimum sweep times below:
Frequency range Peak detector Peak detector Quasi-peak detector
150 kHz – 30 MHz 10 s / MHz 10 s / MHz 200 s / MHz
30 MHz – 1 GHz 100 ms / MHz 100 ms / MHz 20 s / MHz
above 1 GHz 100 ms / MHz 100 ms / MHz n.a.
Longer sweep times have an averaging effect, reducing the noise level:

Base noise with 500 ms sweep time versus 10 s sweep time

3.5 Detectors

Most conducted and radiated emission tests have limits specified for average detector and quasi - peak detector.
Whereas measurement scans with average and peak detector can be carried out reasonably fast, quasi-peak detectors require a measurement time of 1 second per measurement point for measurement receivers and similar long time for spectrum analyzers. A single, complete measurement scan may take several hours, when carried out with quasi peak detectors.
However, there is a workaround, which reduces measurement time significantly:
The measurement result of the peak detector is always higher than the measurement result of the average detector.
The measurement result of the quasi-peak detector will always be somewhere in between the results of the average and peak detector. The measurement result of the quasi-peak detector will never be higher than the measurement result of the peak detector.

Consequently, a complete scan will be carried out, using the peak detector and the result will be compared against the quasi-peak limits. If the peak detector measurement is within QP-limits, the EUT has passed the test. If the peak detector result has a few spurious, which scratch or cross the limit line, there is still the chance that the quasi-peak result is within the limits. However, if the spurious is 10 dB or more above limits, the chance is pretty slim.
To verify, a selective re-measurement, using quasi-peak detector, will be carried out at only the frequency points, where the peak detector measurement crosses the limit line.
When selectively re-measuring spurious with critical amplitude, it also needs to be considered, that the spurious may have drifted in frequency in the time that passed between peak detector measurement and selective re-measurement with quasi peak detector. Especially spurious originating from switched mode regulators may drift considerably over time and temperature. Doing a selective re-measurement may completely miss the spurious at a later time or have the spurious frequency offset far enough to obtain a wrong measurement result. EMCview offers a selective measurement option considering frequency drift. Instead of just measuring at a single frequency, the quasi-peak measurement can be carried out across several adjacent frequency points. EMCview will then make a peak search through this frequency points to ensure capturing the correct quasi peak amplitude.

Example of spurious drifting over time. Both measurements were taken with the same settings, but with a time difference of 15 minutes.

The screenshot below shows a plot from a test house showing the concept of selective quasi-peak measurement. The orange graph shows the peak detector measurement with the blue markers at the frequencies, where the quasi-peak limits are violated. The red markers show the results of selectively re-measuring with quasi peak detector.