Taking Time-Domain EMI Measurements According to International EMC Standards

Stephan Braun and Peter Russer

A system that measures in the time rather than the frequency domain can qualify products more quickly.

The characterization of the electromagnetic compatibility (EMC) of a device requires investigation of its electromagnetic interference (EMI) characteristics. International standards describe a variety of measurement procedures for various classes of devices. The device under test passes the test for interference if its EMI spectrum is below a specified threshold level.

Up to this point, EMI measurements have been made by EMI receivers operating in the frequency domain. EMI receivers that are used for full-compliance testing according to the global standards of the International Special Committee on Radio Interference (CISPR) of the International Electrotechnical Commission must fulfill the CISPR 16-1-1 specifications.¹ Measurements according to the U.S. Federal Communications Commission standards may be performed with equipment that meets either CISPR 16-1-1 or ANSI 63.2.²

Measurements in the frequency domain take a long time to acquire—up to several hours for a single frequency scan. For a device to be completely characterized, the measurements have to be carried out with the antenna at various elevations and polarizations. They have to be repeated with the device at various angular positions. A complete device characterization thus takes several days.

Time-domain EMI (TDEMI) measurements, theoretically, allow this time to be reduced by as much as five orders of magnitude. This technique involves digitizing the EMI signal with an analog-to-digital convertor. By applying digital signal processing techniques to the spectrum with a selectable frequency resolution, the compliance engineer selects the virtual intermediate-frequency (IF) filter and detector mode. The amplitude (magnitude) spectrum is determined by fast Fourier transform (FFT).

A TDEMI measurement system that uses a PC for digital signal processing has been able to reduce EMI measurement time by a factor of 50 relative to measurements in frequency domain.³ By performing the processing on digital signal processors or field-programmable gate arrays (FPGAs), engineers can shorten measurement time, in comparison with frequency-domain measurements, by a factor of about 2000.

During development of electric and electronic devices, precompliance measuring receivers are used to detect EMI problems at early project stages. Precompliance measuring receivers do not have to fulfill the international standards entirely. However, it is important to get precompliance measurement results that are comparable to full-compliance EMI measurements. Full-compliance EMI measurements have to be performed during the product qualification process. The measuring receivers used for full-compliance testing must fulfill CISPR 16-1-1 completely.

This article is intended to show how a TDEMI measurement system should be developed to meet the specifications of the current EMC standards CISPR 16-1-1 and ANSI 63.2.

The TDEMI Measurement System

Figure 1 is a block diagram of the TDEMI measurement system. The system uses an ultrabroadband antenna that is a combination of a logarithmic periodic and a biconical antenna to receive the signal. This HL562 Ultralog antenna from Rohde & Schwarz (Munich) covers the frequency range from 30 MHz to 3 GHz. For conducted-emission measurements, a line impedance stabilization network (LISN) in the frequency range of 9 kHz–30 MHz is used. The EMI signal is low-pass filtered to avoid aliasing effects.

Figure 1. Block diagram of the real-time multiresolution time-domain EMI measurement system.

A floating-point analog-to-digital conversion system that consists of three analog-to-digital convertors (ADCs) digitizes the EMI signal. This system subdivides the amplitude range into three regions, with each amplitude region being digitized by one ADC. The dynamic range specified in CISPR 16-1-1 thus is covered.^4,5 Microwave circuits are used to limit the signals for the ADCs and perform the floating-point analog-to-digital conversion.⁶ The digitized EMI signal may be stored in a high-speed digital memory or may be processed in real time by a digital signal processing unit.

In order to evaluate the spectrum in the peak, average, and root-mean-square (rms) detector modes, the system performs a recursive calculation of the spectrum⁷ And in order to evaluate the spectrum under the quasi-peak (QP) detector mode, a statistical model of the EMI signal is generated.³ From various stored samples of the signal, the system creates a statistical model that describes the absolute frequency of those samples for a given time interval. It processes the stored samples via the short-time fast Fourier transform (STFFT) and generates a statistical-equivalent spectrogram of the EMI signal. A digital QP detector model is used to obtain the spectrum in QP detector mode.

Real-Time Signal Processing

An EMI receiver processes the input signal at a single frequency by means of analog stages. During the measurement at one frequency, signal processing goes on continuously. Time-domain EMI measurement systems that use an ADC system and a digital memory can store and process the signal only for a limited time, the interval being determined by the size of the digital memory. Thus, for the peak, average, and rms detector modes, performance of recursive updates of the spectrum is necessary. Similarly, the statistical model of the EMI signal is needed for evaluating the spectrum under the QP detector mode.

Real-time digital signal processing solves this drawback—the limited record length of today’s available high-speed digital memory. A real-time STFFT processor can perform continuous calculation of the spectrogram and continuous detection of the obtained spectrogram at all calculated discrete spectral values. A further advantage is the shortening of the measurement time by a factor of about 2000. Because of the continuous processing and the possibility of providing a virtual IF signal, such a system can fulfill the entire CISPR 16-1-1 specification.

Currently available FFT cores for an FPGA processor make possible signal processing at a sampling rate of about 250 MHz in real time. In order to process data streams of approximately 2 gigasamples per second, it is mandatory to reduce the data rate. Digital down-conversion is the means for achieving that.8

Digital Down-Conversion. The frequency band 30 MHz– 1 GHz is subdivided into eight frequency bands, each of which is digitally down-converted to the baseband. The baseband signal is processed by a real-time-operating STFFT. The smaller frequency bands are processed sequentially.

The system filters the digitized signal with a digital band-pass filter. After the band-pass filtering, it down-samples the signal to a sampling rate of 250 megasamples per second. The down-conversion and the band-pass filtering may both be performed using a polyphase decimator.

The polyphase decimator consists of a down-sampling unit and a digital filter. Its advantage is that it reduces the data rate of all stages to the output sampling frequency. With the TDEMI system described here, the digital filter of the polyphase decimator needs only to run at 250 megasamples per second. The output signal of the decimator covers a bandwidth of 125 MHz. The polyphase decimator’s coefficients are set to predefined values for processing different frequency bands.

Spectral Estimation. Spectral estimation is performed with the discrete Fourier transform (DFT). The DFT is used to calculate the amplitude spectrum of a signal sample of length N. Because the DFT is based upon the assumption that this sample is continued periodically in time, the time record x[n] needs to be multiplied by a window w[n] of length N to avoid spectral leakage:

           (1)

A Gaussian window function w[n] is used. In order to obtain at every discrete frequency value the same filter response as with the IF filter of the conventional EMI receiver, the window has to be adapted to the IF filter with respect to impulse bandwidth and equivalent noise bandwidth. The DFT being applied to each of the data blocks is

           (2)

Figure 2. Spectrogram of a drill machine, as an example.

In order to obtain a spectrogram, the short-time fast Fourier transform is applied.⁹ A spectrogram depicts a spectrum over discrete frequencies and times (see Figure 2). Each spectrum is calculated by the DFT, using a time interval smaller than the total record length. This T_sBB corresponds with the increment used in the STFFT:⁹

           (3)

where ƒ_sBB is the baseband sampling frequency that is used to simulate the peak, quasi-peak, average, and rms detector modes. Details about implementation of the peak, average, and rms detector modes have been published elsewhere,¹⁰ as has a description of the implementation of the quasi-peak detector mode.

The selected bin width of the STFFT has to be smaller than half of the 6-dB bandwidth of the modeled IF filter in order to avoid spectral leakage. And to fulfill the Nyquist criterion, the baseband sampling frequency has to be at least twice the 6-dB bandwidth of the modeled IF filter. The Nyquist criterion says that the sample rate has to be twice the bandwidth of a signal to avoid aliasing. The relation between the bin width and the baseband sampling frequency is called the overlapping factor (Oƒ) and is given by

           (4)

In order both to avoid spectral leakage and to fulfill the Nyquist criterion, the overlapping factor of the STFFT must be at least 4.

The STFFT hardware exhibits a parallel-FFT core structure (see the block diagram in Figure 3). For digital implementation, the input signal is delayed by three N/4 shift registers (SR in the diagram). Before and after each shift register, the signal is extracted and multiplied by the Gaussian window function. Each signal subsequently is processed by an FFT core. The outputs of the four FFT cores are processed by the digital peak, average, rms, and quasi-peak detectors.

Figure 3. Digital implementation of short-time fast Fourier transform.

The output signal of the STFFT for a selectable frequency is multiplied by e^j2πƒIF, where ƒ_IF is a selectable virtual intermediate frequency. Through digital-to-analog conversion, the signal can be provided as an analog virtual IF signal.

Detector Modes. The spectrogram is evaluated in the peak, quasi-peak, average, and rms detector modes. The peak detector mode calculates the maximum magnitude at each discrete spectral value, thus:

           (5)

The average detector mode calculates the mean spectrum from the spectrogram. The formulation in this case is

           (6)

The rms detector mode calculates the rms value of the magnitude of the spectrogram as follows:

           (7)

Finally, the quasi-peak detector mode evaluates the emission according to a physiological disturbance against amplitude- modulation radio. A detailed description of a digital quasi-peak detector has been published.³

Full-Compliance EMI Measurements

CISPR 16-1-1 and ANSI 63.2 were originally written for EMI receivers that were, at that time, analog heterodyne or superheterodyne receivers. But in no standard is it said that the measuring apparatus has to be a heterodyne or superheterodyne receiver.

In CISPR 16-1-1, the measuring device is described as a black box that, for stationary and transient signals, must display a defined indication. Obviously, therefore, instead of an EMI receiver, any other measurement equipment—such as a TDEMI measurement system—that shows the correct indication as described in CISPR 16-1-1 may be used for full-compliance measurements.

The requirements given in CISPR 16-1-1 can be divided into three main categories. The first describes the relation of the indication of the measuring device to stationary and transient signals. The second group consists of requirements pertaining to the dynamic range. Required characteristics of all analog input and output connectors constitute the third category.

Response to Stationary and Transient Signals. The IF filter has to provide the critical masks specified in CISPR 16-1-1. The standard gives the critical masks for all selectable IF filters. Figure 4 supplies the example of a 120-kHz IF filter. It is evident that the requirements can be met by applying a Gaussian window function and the STFFT.

Figure 4. IF-filter response (120 kHz) for Band C/D (30 MHz– 1 GHz).

In order to ensure a fixed relation between transient and stationary response, CISPR 16-1-1 also specifies the impulse bandwidth of the IF filter. This bandwidth is defined by the following:

           (8)

In the standard, the relation between the window function and the impulse bandwidth is given by

           (9)

where W(f) is the Fourier transform of the window function w(t). When a Gaussian window function is used to model the IF filter, specifications regarding the critical masks of the IF filter and the required impulse bandwidth are fulfilled.

The average, rms, and quasi-peak detectors have to be implemented in a way that fulfills the weighting functions defined in CISPR 16-1-1. Obtaining weighting functions for a measurement system involves connecting a pulse generator to the system. The output voltage level is tuned for each selected pulse-repetition frequency to provide a constant indication of the measuring apparatus. The tuned voltage dependent on the pulse-repetition frequency describes the weighting curve of the detector.

An example of the measurement performed with a quasi-peak detector is shown in Figure 5. For the simulation of the quasi-peak detector, the original time constants presented in CISPR 16-1-1 were used. However, for some lower pulse-repetition frequencies, the deviation between CISPR 16-1-1 and the obtained result is about 2 dB. CISPR 16-1-1 does state that the time constants may be adapted. Experiments have shown that a decrease in the charging time constant along with an increase in the discharging time constant results in optimized weighting curves and thus minimized error.

Figure 5. Weighting curve of the QP detector for Band A (2–150 kHz), Band B (150 kHz–30 MHz), and Band C/D (30 MHz–1 GHz).

An investigation of the weighting curve of the rms detector has been published.⁵ For the peak, average, and rms detectors, the digital implementation will produce exactly the results given in CISPR 16-1-1.

The standard’s requirements pertaining to the response to stationary and transient signals can be fulfilled by correct implementation of the window function and the detector mode. Because this system will usually show the same results, it can be used for precompliance measurements. Signal processing can be performed on a conventional PC.

Dynamic Range. An overview of the results of a published derivation of the dynamic range of a multiresolution time-domain EMI (MRTDEMI) system and a TDEMI system is presented here.⁴

ANSI and CISPR have different requirements according to the dynamic range of a measurement system (see Table I). While CISPR requires a dynamic range of 40 dB over the complete frequency range, the spurious free dynamic range (SFDR) is mandatory for a single sinusoidal signal. Other spurious signals such as the harmonics or mixing products with internal signals must be below 40 dBc. ANSI requires a rejection of spurious responses that are more than 20% removed from the sinusoidal input signal. The rejection must be more than 60 dB.

EMI receivers employ a preselective input filter to achieve sufficient dynamic range. A preselective filter can increase the dynamic range for transient broadband signals and strong narrowband signals. However, owing to the limited slope of analog filters, strong narrowband signals are suppressed only when they lie a certain frequency range apart from the measured signal.

TDEMI measurement systems are broadband systems and have no preselective filter. Their only input filter is the aliasing filter that ensures that the signal is sampled according to the Nyquist theorem.

The following discussion distinguishes between the TDEMI measurement system that uses one ADC to sample the input signal and the MRTDEMI system that uses several parallel ADCs to obtain an improved dynamic range. ANSI and CISPR requirements that apply to the analog-to-digital conversion systems are compared in Table II.

A TDEMI measurement system with 8-bit resolution can satisfy the ANSI standard if the ADC provides sufficient SFDR. The SFDR is limited by the nonlinear behavior of the ADC, by spurious signals that are coupled between the analog and digital sections of an ADC, and by the linearity of any preamplifiers. In order to fulfill the specified range of the input signal, an additional switchable attenuator would be necessary.

The MRTDEMI system with three 8-bit ADCs does not need any selectable attenuator. In order to fulfill the requirements in CISPR 16-1-1, a single-ADC system must have at least 16-bit resolution. But only high-speed ADCs with 10-bit resolution are available now. Thus, a TDEMI system with 16-bit resolution is currently not feasible. Use of an MRTDEMI system with three 8-bit ADCs, on the other hand, does fulfill the requirements of CISPR 16-1-1.

Analog Inputs and Outputs. An EMI measuring apparatus requires a radio-frequency input that must show an impedance of 50 Ω and a maximum voltage standing-wave ratio (VSWR) of 2.0. In addition, the antialiasing filter of a TDEMI measurement system has to have a VSWR better than 2.0 in the passband.

For disturbance analysis, analog outputs of the IF signal and the response of the quasi-peak detector are mandatory. These requirements cannot be fulfilled by a system that stores the signal in time domain and performs a recursive update. An algorithm for the quasi-peak detector that uses a statistical model of the EMI signal can generate a statistically equivalent IF signal. However, because a statistically equivalent IF signal is not suitable for disturbance analysis, the quasi-peak detector algorithm cannot fulfill these requirements.

Real-time signal processing, though, does allow an IF signal that shows the same response as the IF signal of an EMI receiver to be generated for a selected frequency. A simulation of the quasi-peak detector could be used to generate the response of that detector. By this means, a TDEMI measurement system can fully satisfy CISPR 16-1-1 requirements in this area. Digital signal processing hardware is necessary for real-time operation. With such a system and method, measurement time can be reduced by a factor of about 2000.

EMI Measurements in Time Domain

The emissions from a conventional desktop computer were measured in the average detector mode (see Figure 6). The maximum difference between the result of the measurement in time domain and the measurement performed with a conventional EMI receiver has been below 2 dB.

Figure 6. Emissions measurement of a desktop computer.

The nature of the desktop computer’s emissions was investigated via an azimuth scan performed with the TDEMI measurement system (see Figure 7). The computer’s various emission sources can be seen to have their maxima at different angles. Because of the TDEMI system’s short measurement period, it can perform a complete characterization of the device under test in the time required to perform a single scan in frequency domain.

Figure 7. Azimuth scan of a desktop computer.

Emissions measurement of a handheld mixer also has been carried out. The maximum difference between the measurement performed in time domain and results obtained by an EMI receiver in this case also has been below 2 dB (see Figure 8).

Figure 8. Emissions measurement for a handheld mixer.

Conclusion

Time-domain EMI measurement systems can fulfill various parts of the international EMC standards. A single-resolution system will show the correct response to transient and stationary signals for any IF bandwidth and peak, quasi-peak, average, and rms detector mode. Its limited dynamic range does not meet the CISPR 16-1-1 requirement for transient signals. But a multiresolution TDEMI measurement system can fulfill entirely the requirements of the dynamic range defined in CISPR 16-1-1.

With both the TDEMI measurement system and the MRTDEMI system, a conventional PC can be used for signal processing. Both setups will produce results that are comparable to results obtained by an EMI receiver, while offering the advantage of a measurement time reduced by a factor of 50.

In order to satisfy CISPR 16-1-1 fully, the data need to be processed continuously in order to provide the IF signal and the response of the quasi-peak detector. A real-time time-domain EMI measurement system that uses digital hardware is able to process the EMI signal in a continuous manner. Such a system can fulfill CISPR 16-1-1 in all respects while also reducing the measurement time by a factor of 2000.

References

1. CISPR 16-1-1, “Specification for Radio Disturbance and Immunity Measuring Apparatus and Methods—Part 1-1: Radio Disturbance and Immunity Measuring Apparatus—Measuring Apparatus” (Brussels: International Electrotechnical Commission, 2003).
2. ANSI C63.2:1996, “American National Standard for Electromagnetic Noise and Field Strength Instrumentation, 10 Hz to 40 GHz—Specifications” (Piscataway, NJ: Institute of Electrical and Electronics Engineers (IEEE), 1996).
3. S Braun, F Krug, and P Russer, “A Novel Automatic Digital Quasi-Peak Detector for a Time Domain Measurement System,” in 2004 IEEE International Symposium on Electromagnetic Compatibility Digest, vol. 3 (Piscataway, NJ: IEEE, 2004), 919–924.
4. S Braun, M Aidam, and P Russer, “Development of a Multiresolution Time-Domain EMI Measurement System That Fulfills CISPR 16-1,” in proceedings of 2005 IEEE International Symposium on Electromagnetic Compatibility (Piscataway, NJ: IEEE, 2005): 388–393.
5. S Braun and P Russer, “Time-Domain EMI Measurements Performed with a Multi-Resolution System for Product Development and Compliance,” The International Journal of Electromagnetic Compatibility, EMC Test and Design Guide, November 2005: 18–26.
6. S Braun and P Russer, “A Low-Noise Multiresolution High-Dynamic Ultra-Broad-Band Time-Domain EMI Measurement System,” IEEE Transactions on Microwave Theory and Techniques 53 (2005): 3354–3363.
7. F Krug and P Russer, “The Time-Domain Electromagnetic Interference Measurement System,” IEEE Transactions on Electromagnetic Compatibility 45, no. 2 (2003): 330–338.
8. AV Oppenheim and RW Schafer, Discrete-Time Signal Processing (Englewood Cliffs, NJ: Prentice-Hall, 1999).
9. L Cohen, “Time-Frequency Distributions—A Review,” Proceedings of the IEEE 77, no. 7 (1989): 941–981.
10. F Krug, T Hermann, and P Russer, “Signal Processing Strategies with the TDEMI Measurement System,” in 2003 IEEE Instrumentation and Measurement Technology Conference Proceedings (Piscataway, NJ: IEEE, 2003): 832–837.

Stephan Braun is a research scientist with the Institute for High-Frequency Engineering at the Technische Universität München in Germany. He can be reached at stephan.braun@tum.de. Peter Russer is a professor at the institute. He can be contacted via e-mail at russer@tum.de.