IEEE PEDS 2005

Time Domain Measurement System for ConductedEMI and CM/DM Noise Signal Separation

Yuang-Shung Lee Yu-Lin Liang Ming-Wang ChengDepartment of Electronic Engineering Department of Electronic Engineering Graduate Institute of A.S. E.

Fu Jen Catholic University Fu Jen Catholic University Fu Jen Catholic UniversityTaipei, Taiwan Taipei, Taiwan Taipei, Taiwan

lee@aee.f u.edu.tw sam emc.coM.tw

Abstract- This article discusses how to use a digital oscilloscope In recent years, with advancements in digital oscilloscopes,to match the time-domain measurement computer-aided sampling rates of an analogue-to-digital converter are now uptechnique and Fourier transformation to measure the conducted to several GB/s, which is sufficient for the Fourier Analyzer toEMI. Taking advantage of the multi-channel oscilloscope and the display information below the 30MHz range. For long,computer, it will deal with the Line and Neutral lines engineers have been measuring EMI with spectrum analyzers,synchronization. It can mitigate the phase coherency problem to where scanning of the frequency requires a long period of time,separate the common-mode (CM) and differential-mode (DM) making it impossible to catch the instantly vanishing noiseEMI noise at the same time without any hardware signals. This disadvantage is especially obvious whenimplementations. Due to the time-domain measuring method, measuring the radiated EMI (30MIHz-IGHz), normally takingwhich allows the noise amplitude and phase in parallel processing

u

for the whole spectrum range, measuring time is reduced. The up to 30 minutes [1,2]. Moreover, the conventional spectrumcomputerized time-domain measurement environment is based measuring system allows only one input signal, thereforeupon the LabVIEW test and measurement system. The requiring coordination of separators, such as a power combinerseparation results are compared with the frequency-domain or active noise separator, to separate CM and DM noise signals.measurements. The proposed time-domain measured system can On the contrary, while using the time domain measuringprovide a user friendly and convincible approach for fast EMI technique with a digital oscilloscope with multi-input features,measurement and countermeasures. direct calculations can be performed for signals retrieved from

CHI & CH2 of the oscilloscope for adding or subtracting noiseKeywords: Time domain measuremen; conducted EM; CM signals of the Line and Neutral lines. Not only can CM and

and DM separation; Lab VIEW; Fourier transformation DM noises be separated at the same time, but also a separator isno longer required. For the last part, Fourier Transformation is

I. INTRODUCTION used at the PC end for decomposing the measured signals intodifferent frequency components.Because EMI regulations in different countries vary, the

restrictions also differ. EMI emissions are mainly divided into Presently, some oscilloscopes already have built-in fasttwo categories: the radiated and the conducted. Conducted EMI Fourier transformation (FFT) functions; the economical benefitdenotes noise emission caused by conducting cables. In general, of this advancement is getting clearer, and the FFT applicationto lower the generation of conducted EMI, the R&D people is getting more and more popular. This study handles FFT.usually employ the noise separation technique, separating the within a PC, with the sampling rate set at 500MB/s, fornoise as common-mode and differential mode, for designing lowering the oscilloscope requirements and therefore savingEMI filters, respectively, to eliminate conducted EMI the cost of instrument procurement. Lastly, a calibratedgenerated from a switching power supply device. Take the spectrum analyzer is used to verify the accuracy of theCISPR regulations as an example, the measurement range is frequency amplitude and separation effects of CM and DM150 kHz - 30 MHz. The testing instrument, traditionally a noises.spectrum analyzer or an EMI receiver, works mainly byscanning the frequency. As for the testing environment, since II. SYSTEM DESCRIPTIONSthe switching power supply is small in size, a workbench setup Many studies have discussed the use of time-domainis normally sufficient; the ground and vertical waIls shall be Masurem measurnrdite eoffumished with grounded metal plates, to provide a shielding TDEMI measurement for measuring radiated EMI (30MHzs -effect that lowers EMI. An 80cm high workbench of non- 1GHz) [2]. Reference [3] has also proposed preliminary studiesconductive material is placed upon the metal plate, the tested on the conducted EMI In this section, we shall probe deeperobjectshallbe 40cm apart fromtheverticalwallsandpowinto the conducted EMI time-domain measurement and the

by LISN. The conducted noise generated by the tested object is ofealcommn-mode and diffegrentia-odfte nosofaes sopasatosent to the Spectrum Analyzer viaLISN.ofcmo-oeaddfrntlmde ois,oaso

0-7803-9296-5/05/$20.00 ©2005 IEEE 1640

provide a low-cost solution for generating automatically a Panel and varying according to the time scale of thepreliminary certification for the conducted EMI. oscilloscope) shall be at least two times that of the highest

Same as of general languages, the main parts of LabVIEW signal frequency. To a maximum conducted EMI of 30 MIHz, a

consist of the user-machine interface (Front Panel) and the minimum sampling rate of60Mz is required. The upper limitBlock Diagram [1]. In this study, the Front Panel including the of the testing frequency Fn = Fs / 2 is therefore called asetting options of the oscilloscope and the display of the results. Nyquist frequency. If the signal exceeds the upper limit,The block diagram where procedures are arranged in the Aliasing Frequency will occur to avoid this, the sampling ratesequence for carrying out calculations and treatments in the is set to 500MB/s, much higher than that required for theflowchart format. measurement [4].

III. TIME DOMAIN MEASUREMENT PRINCIPLES R = f / L (3)esolton s engthSignals in EMI measurement are often random. There are

transient and abrupt components in the signal besides Length is the depth of memory, and Resolution is the resolutionharmonics and noises. However, with a long enough observing of each point in the frequency domaintime ATIME, the random EMIsample signal x(t) (to< t < + From the above formula, we can see that memory lengthATIME) can contain all information included in the signal. and sampling rate of the oscilloscope are related to theTherefore the signal is considered as stable [2]. When definition of the frequency axis. In the example, for a samplingmeasuring a signal with an oscilloscope, we normally measure rate of 500MB/s and memory length of 50,000 (bytes), thewith a time-domain, i.e. Y-axis stands for voltage and X-axis resolution will be 10 kHz, as derived from (3). This is veryfor time. This intuitive method is not applicable to EMI close to the 9 kHz bandwidth of the RBW (resolutionmeasurement. According to regulations, we need to observe in bandwidth) filter.a frequency domain. In a frequency domain, the Y-axis is for Spectral leakage is a product of the Fourier Transformationvoltage but the X-axis is for frequency; the frequency domain hypothesis, by which we assume that discrete time extends itsshows the signal intensity in each frequency. For complex period in the entire time-domain precisely; i.e. all signalshybrid noise signals, it is difficult to recognize the frequency . . .tiproperty in the time domain; but in the frequency domain, the cprising the discret esqnc are afncti of the.significant components are obvious. Since the oscilloscope e n Heerod is relevant to the length of the timesamples time-domain signals in a discrete way, it is a usual t lpractice in oscilloscope frequency domain measurement to use a non-integer number of cycles, meaning that the hypothesistief .t. f . the condition does not stand, spectral leakage will occur. In reality,discrete Fourier transformation (DFT). it is difficult to control the EMIl noise measurement to a time-record that corresponds to complete waveforms, i.e. non-

First, sampling rate and sampling period are defined as j continuous waveforms do exist. This will cause the energy ofand

the frequency component to leak to the neighboring frequency,and Ts , respectively so T, =1/f .Continuous-time input thus introducing an error to the measuring result.signal is x(t), to n = 0,1 ...(N -1), the sampled signal is To solve the spectral leakage problem, we may use the

window function to increase the record time.x[n] = x(nT)(1x[n] =x(nTs) (1) x. [n] = x[n]w[n] (4)

where n is the length of the oscilloscope's memory capacity.DFT is defined as where x[n] is a window function.

N-i The principle of this method is to change the waveform toX(fk) = x[n]ej (2) zero at the ending point of the record time, and then when we

n=O repeat the record time there will be no transient phenomenon.Naturally the window will modify the time record and affect

k = 0919..(K 1) the frequency domain; yet selecting a properly designed= 0,1,l .(K-1) window basically has greater improvements in the frequencydomain than a design that has no window at all. Therefore, toand fk = kfs 1K: where k is the number of points in the solve the spectral leakage problem, this study incorporated the

frequency domain, K = N flat-top window to the Fourier transformation formula for time-domain measurement [4].

In practice, a digital oscilloscope converts analogue signals,via an A/D Converter, into digital ones. From the previous IV. CM ANDDM NOISE TIME-DOMAIN SEPARATIONsection, the oscilloscope samples and quantifies thecontinuous-time input signals. The sampling frequency (j ) The oscilloscope converts waveforms into digitalinformation by way of samplingad A/D conversion, andcorresponds to the interval of 1 / fs . According to the Nyquist registers the data into memory and displays the same throughtheorem, the sampling rate (normally displayed in the Front the monitor screen [5].

1641

With LabVIEW as the development environment, thewaveform retrieved by the oscilloscope is transmitted to a PC OscilIdosopewith the parallel 1EEE488 GPIB interface; data are retrieved bylooping for N-times and calculated according to differentselections of the detector. One has to confirm first, before SetupOscilloscopeproceeding the time-domain test, that the oscilloscope is setpn scope

available with DC offset Voltage when there is no input signal;it means that when no signal is being input to the oscilloscope, v e il1neoise adjustthe displayed value at its smallest vertical scale shall be a amplitde v vstraight line near the zero line. Since conducted EMI is oftendenoted in dBjiV, a small drift of the oscilloscope can cause anerror as large as 10 dBjiV or even more. In case of existence of wdrifts, it can either be calibrated automatically back to normalby the built-in function, or the machine has to be sent to themaker for calibration. Simulatemean-value m pea-l

The oscilloscope input impedance can be set to 50Q, which detacor ditectoris equal to the LISN output impedance, system coupling __-impedance is therefore not a problem; but the output value is inVrms, which requires conversion, by (5), into dBAV, which is Dispia N lOnesthe unit used by EMI measurement. noise signal

dB1V = 20log( Vo) (5)I#V |DM,nise signa

where V. is the sampling voltage retrieved by theoscilloscope.

Because of the need of introducing EMI design in thepreliminary stage of product development, it is of great help to rpt andd eiconotthe R&D people, for the reference of the filter design, to exceeded vlue ePortseparate CM and DM noise signals in the pre-compliancemeasurement. From [6], we have

Figure 1. Program flowchart

Vu^ = VCU + VDU (6)LIne VCM + VDM (6)V. MEASUREMENT SYSTEM VALIDATION TEST

Ne-tral -CM - DM To ensure accuracy of the measurement, this study uses aVN,tl.l V VDM (7) spectrum analyzer, including a power combiner, to perform

validation tests against the results on L and N lines noise, andTherefore, CM and DM noise signals can be calculated by CM and DM noise separation. The test environment

-specifications are shown in Fig. 2; the list of instruments andVCM = Vine + VNut,al) / 2 (8) validation test equipments are shown in table I.

VDM = (YLine VNet,al) /2 (9)Inserting waveforms retrieved from CHI and CH2 of the Vertical coAdtve ----Line

oscilloscope into formulas (8) and (9), respectively, we can NsurueNeutralthen separate CM and DM noise signals in the time-domain atthe same time and present the result of the Fourier analysis via 40 e DUTsoftware calculation. One particular point worth mentioning is GPIBthat, in order to match the phases while measuring noisesignals from the Line and neutral lines, the measuring action of Refeawce goundingthe oscilloscope must be paused, via software program, before scestarting the retrievals. Furthermore, to minimize measurementerror and instability, the length and property of the measuring Ls -.cables must be kept as identical as possible [6]. The flowchartofthe program is shown in Fig. 1. Figure 2. Measuring environment using the Oscilloscope

1642

TABLE 1. LIST OF INSTRUMENTS -iI TS

Make Model = = = - =Digital Oscilloscope Tektronic CSA7404B > _ _lSignal Generator Agilent E4438C iSpectrum analyzer Agilent E4404B ') ..V .Power combiner Mini-circuits ZFSC-2-75;ZFSCJ-2-1 - fr.ILISN 1Seaward INone -.-~ -

Actual products are tested for comparison, they are, c _ - - - - _ _respectively, a 250 W (forward converter) switching power

Z _ - __supply of a desk-top PC and a 60W (fly-back converter) ISO du

switching power supply of a Notebook PC with the following Frequency(MHz)specifications, as shown in table II. 3 Suctrum analyr Line (L) noise signal

Figs. 3 and 4 show a comparison of the results of Line andNeutral measurement of the 250W and 60W power supplies,measured, respectively, with the digital oscilloscope and withthe Spectrum Analyzer. Figs. 5 and 6 show the comparisons of 7V-the results for CM and DM. >

A

TABLE II. SPECIFICATIONS OF TESTED DEVICES Z 191Desk top PC Notebook PC

Input Specifications I I0V-230V IOOV-240V Frequency(MHz)Output Specifications 5V - 25A ;12V - 10A 19V ' 3.42A 3 (c) Oscilloscope Neutral (N) noise signalPower Rating 250W 60W Rf I dB4 R, 1dB

Circuit Configumtion Forward Flyback

Separation is measured, respectively, by the digital D=loscilloscope time-domain and by the spectrum analyzerincluding the power combiner. To facilitate comparison of the

WS3measured results, the control-limit line is set to 5OdBjiV; the * S3Rfrequency range is in CISPR specifications: 150 kHz - 30MHz; ZX-axis is in linear format; and amplitude range is 0-100 dBIjV St kHZS= = T1 T(each scale unit represents lOdB,uV). The resolution bandwidth .R, BW 1$ kH= UBH1419 384S _ (401 .t.)(RBW) filter of the spectrum analyzer is 9 kHz; and the Frequency(MHz)oscilloscope simulation bandwidth resolution (3) is 10 kHz. 3 (d) Spectrum analyzer Neutral (N) noise signal

Figure 3. Comparison of Line and Neutral noise signal measurement, 250WComparing Figs. 3 and 4 shows that the maximum errors in

the Line and Neutral noise signals occur in the low-frequencyarea. The main reason for this is the uneven frequency responseof the power combiner in the low-frequency range, and part ofthe reasons is the uncertainty between the instruments. =L

Successively we compare the CM and DM separation results inFigs. 5 and 6; we leam from the references that the resultmeasured by the power combiner shall deduct 3dB to get theactual output value [6]. Results of the two measurements, apartfrom the low-frequency, are normally having an error less than Z6dB.

Frequency(MHz)4 (a) Oscilloscope Line (L) noise signal

R.f I8 dB4 RI,n 5Sde=L Z1,l

YG O |.~~~~18 - - -

1643Z Id S

zFrequcncy(MHz)

3 (a) Oscilloscope Line (L) noise signal S --15 H --- 1 - ~Frequency(MH1z)

4 (b) Spectrum analyzer Line (L) noise signal

1643

x~~~~~~~~~~~'v x-B _ __ ______ _ __

S~~~~~~~~~~~~~~~o>) >- F

z ITzFrequency(MHz)

Frequency(MHz) 5 (d) Spectrum analyzer DM noise signal4(c) Oscilloscope Neutral (N) noise signal Figure 5. Comparison ofCM and DM noise signal separation results, 250W

5.40z1W ASs.n S 45-

=L518~~~~~~~~~~~~~~~~~~~~~~~~=

>8Ul 2=* = = = =m ;11_S3FCF 5) __ > _ _ _ _

F)~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~5

S9.I 158 kp Sa. 38 W.zIS LR8.. B8 9 W,4 VOW 38 W.5 S---. 881.2 -4 (401 ...II--)Frequency(MHfz)

4 (d) Spectrum analyzer Neutral (N) noise signal

Figure 4. Comparison ofLine and Neutral noise signal measurement, 60W Frequency(MHz)6 (a) Oscilloscope CM noise signal

L"Is/

5z=L~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~5so1"Wt sd3 ~~~~~~~~~~~~~~~~~0

. Om88 kw8 44 38 4,44 S... 881.2. (Wi ow,Frequency(MHz) Frequency(MHz)

5 (a) Oscilloscope CM noise signal 6 (b) Spectrum analyzer CM noise signal

Ml $2 IVd>V FI '

1_- E 10 kft "isff k%t S 14 (,M ptX_ 1

Frequency(MHz)S(b)SpectumanlezerCM nise signal Frequency(MHz)

- ~~~~~~~6(c) Oscilloscope DM noise signal-.4i- - - - -L- FreunyMz

LW -~~~~~~~~~~~~~~~~~~~~~~~=

Frequency(MHz) 6 d) Spectrum analyerFrequencysMHz aFrqecSMz6cOsilosop SetuanlzrDMnoise signal5 (c) Osciloscope DM nise signalFigure 6. Comparison ofCMand DM noisesignal separation results, 60W

1644

Usually, an oscilloscope has two input channels, so we can frequency noise signals that we do mind are "smuggled" withinperform noise signal measurements for both Line and Neutral these 60Hz low-frequency signals. This phenomenon causeslines at the same time. Tedious cable replacement can be the oscilloscope to have excessive signal variation whileavoided and the total measurement time can be reduced to conducting time-domain measurements, which in turn yieldsabout one third. More importantly, the price of the oscilloscope inaccurate triggering of the oscilloscope. This phenomenon canis much lower than that of the Spectrum Analyzer, and the be minimized by using the AC voltage source combined withrange of voltage tolerance is also wider than that of spectrum the statistical method of taking the mean value of multipleanalyzers; CM and DM noise signals can be separated at the samples to expose the noise characteristics.same time without the need of jigs. We can therefore save thecost for the surge limiter and power combiner as well.Moreover, displaying the oscilloscope measurements by way of VI. CONCLUSIONcomputer, versatile applications can be derived. For example, This study proposes a time-domain measurement system ofreport generation, remote monitoring and 3D effect of the a conducted EMI and noise signal separation technique, andmeasurement results add value to it. This is also one of the provides an alternative solution to EMI pre-compliance in themajor advantages of virtual instruments. The following Fig. 7 R&D stage. In the past, the spectrum analyzer is the mostis a 3D rendering of the measured results, which facilitates standard instrument for EMI measurements; nowadays,observation of the continuous distribution of the conducted computer-aided measurement, configured with time-domainEMl. conducted EMI measurement technique, is providing the

following three advantages: shorter validation time, added-value and lower equipment costs. Using an oscilloscope tomeasure Line and Neutral noise signals at the same time notonly simplifies operation procedures and shortens thevalidation time, but also separates CM and DM noise signals inthe same measurement. This lowers significantly cost becausethe separator is no longer required when using an oscilloscopeinstead of the spectrum analyzer. Added-value is increased bydisplaying calculation results ofthe retrieved signals via the PC

z i 3 _ 1l - iw ffi ill! p monitor thus allowing versatile additional functions, such as3D curve observation and generation of reports. Proved byFmlueztyMR, actual validations, the total system measurement error is less

than 6dBpV. Using the time-domain system for conductedFigure 7. Line (L) noise signal 3D-graphic-curve,25oW EMI measurement is advancing the EMI issue to the product-

development stage, and therefore results in a shortened R&DHowever, since the core components of an oscilloscope are cycle as well as improved competitiveness ofthe product

A/D converters, different resolutions of ADCs cause differentdynamic range limitations. These limitations may result in REFERENCESpoorer distinction of abruptly changing signals compared with [1] Y. S. Lee and Y. Shu, "Line Filter Design of Switching Mode Powerto Spectrum Analyzers. Market available oscilloscopes are Supply Using Software Approximation for Conducted Emissionusually 8bit A/D converters. Therefore, the dynamic range Separation" Inter-national Conference on Power Electronics and Driveprovided is about 54dB, while that of the Spectrum Analyzer is Systems, PEDS2003, pp. 1339-1344, November 2003.about 75dB. The calculation formula is as follows: [2] F. Krug and P. Russer, "The Time-Domain Electromagnetic Interference

Dynamic Range (dB) = SNR (dB) Measurement System," IEEE Transactions on Electromagnetic= 20 loglO(RMSFull - scale / RMENoise) (10) Compatibility, Vol. 45, pp. 330-338,MAY 2003.

[3] H. L. Su & K. H. Lin, "Computer-aided design ofpower line filters with aHowever, for a preliminary verification system, what we low cost common and differential-mode noise diagnostic circuit," IEEE

mind most is the value that exceeds the control limit. Moreover, International Symposium on Electromagnetic Compatibility, Vol. 1, pp.a maximum voltage test will be performed prior to starting the 511-516, Aug. 2001.program, for adjusting the oscilloscope into its optimal vertical [4] Agilent, "The Fundamentals Of Signal Analysis," AN. 243, Jun. 2000.display range and for lowering the influence caused by the [5] Tektronix,,"CSA 7404 User Manual," Mar. 2003.dynamic range limitation. Furthermore, when conducting [6] T. Guo, D. Chen, and F. C. Lee, "Separation of The Common-Mode andoscilloscope time-domain measurements, it is essential to Differential-Mode Conducted EMI Noise," IEEE Transactions on Powerconfirm that no excessive 60Hz low-frequency signal is Electronics, Vol. 11, No.3, pp. 480-488, May 1996.combined with the input, because the LISN cannot totallyisolate low-frequency signals. The 150 kHz - 30 MHz high-

1645