Advances in Radio Science (24) 2: 27 32 Copernicus GmbH 24 Advances in Radio Science A novel quasi-peak-deecor for ime-domain EMI-measuremens F. Krug, S. Braun, and P. Russer Insiue for High-Frequency Engineering, Technische Universiä München, Arcissrasse 21, 8333 Munich, Germany Absrac. In his paper, an advanced ulra-fas, broadband ime domain EMI measuremen sysem is described. Measuremens were performed in he 3 1 MHz range. The digial signal processing of EMI measuremens allows o emulae in real-ime he modes of convenional analogous equipmen, e.g. Peak-, Average-, RMS- and Quasi-Peak- Deecor. Wih he presened ime domain measuremen sysem he measuremen ime can be reduced by a facor of 1. A novel signal recording rouine for ime-domain EMI (TDEMI) measuremens and Quasi-Peak-Deecion is described. Measuremen resuls obained from he invesigaion of a drill machine, monior and lapop obained wih he imedomain elecromagneic inerference (TDEMI) measuremen sysem are discussed. The resuls obained wih he described sysem have been compared wih measuremens performed wih a convenional EMI receiver. 1 Inroducion Due o he rapid developmen of new elecronic producs and due o emerging new echnologies he abiliy o achieve and o improve elecromagneic compaibiliy is a major challenge in developmen of elecronic producs. EMC and EMI measuremen equipmen which allows o exrac comprehensive and accurae informaion wihin shor measuremen imes will allow o reduce he coss and o improve he qualiy in circui and sysem developmen. In he pas and currenly, radio noise and elecromagneic inerference (EMI) are measured and characerized by superheerodyne radio receivers. The disadvanage of his mehod is he quie long measuremen ime of ypically 3 min for a frequency band from 3 MHz o 1 GHz. Since such a long measuremen ime resuls in high es coss, i is imporan o look up for possibiliies o reduce he measuremen ime wihou loss of qualiy. The digial processing of EMI measuremens using Correspondence o: F. Krug (fkrug@ieee.org) Fourier Transform allows he decomposiion of a signal measured in ime-domain ino is specral componens. In general he digial processing of EMI measuremens allows o emulae in real-ime he various modes of convenional analogous equipmen and i is also possible o inroduce new conceps of analysis (Krug, 22a). In he following a new signal recording rouine for ime-domain EMI (TDEMI) measuremen and Quasi-Peak- Deecion is described. The measuremen rouine inhibis inelligen riggering o record he EMI signal. This makes i possible o reduce daa enormously. A mehod o reconsruc an equivalen of he original signal is described. An accurae ime-domain measuremen wih Average-, RMS- and Quasi- Peak-Deecion on all signal ypes are possible. 2 Advanced TDEMI measuremen concep A deailed hardware descripion of he used TDEMI measuremen sysem has already been presened in Krug (22b, 23). In Fig. 1 he advanced TDEMI measuremen concep is shown. The daa acquisiion process for he imedomain measuremen sars wih he sample process of he oscilloscope. Then he specra via he Shor-Time-Fourier- Transform (STFT) is digially compued. Afer he STFT he errors due o he frequency characerisics of anenna, ransmission line, amplifier and ani-aliasing filer are correced by signal processing. Afer his he analysis of Peak-, RMS-, Average- and Quasi-Peak-values of he EMI signal is performed. 3 Auomaed TDEMI measuremen algorihm for class C signals When measuring signal wih ransien envelope, he pulses can record separaely, which allows o sample pulses wih differen ampliudes wih an opimum verical resoluion of he Analog-o-Digial-Converer (ADC). Firs he measuremen rouine wih one verical resoluion is described.
28 F. Krug e al.: A novel quasi-peak-deecor for ime-domain EMI-measuremens Oszilloscope Inelligen Recording pulses ime samps STFT Deecors Peak 6 5 Probabiliy Densiy Esimae Quasi-Peak Time-Domain signal Reconsrucion a each Specral poin Fig. 1. Advanced TDEMI measuremen concep. Average RMS Esimae Value 4 3 2 Realime 1.5 1 1.5 2 2.5 3 3.5 ime [s] x 1 3 Fas Frame Fig. 3. Probabiliy densiy esimaion for he emission of a drill machine. Fig. 2. Principle of fas frame mode. random disribuion of dela-imes 3.1 Trigger ime measuremen 1 2 3 n The used oscilloscope needs o offer a fas frame mode. In his mode he oscilloscope capures he absolue imes of rigger evens which occur in sequence. In case he oscilloscope riggers on pulses, he imes beween he pulses can calculaed. The principle of he fas frame mode is shown in Fig. 2. A measuremen of a cerain amoun of rigger imes, a calculaion of he reliable saisics abou which imes beween pulses occur mosly. Figure 3 shows an example for a probabiliy densiy esimae calculaed over imes beween pulses radiaed by a hand-held drill machine. The imes which correlae o he local maxima of he graph will be used for signal reconsrucion. They are randomly bu according o heir relaive frequency of occurrence disribued on he imescale. An example for a reconsruced imescale is shown in Fig. 4. 3.2 Signal measuremen The sampling rae f s mus mee he Nyquis requiremens. Recording occurs in a manner of several single sho measuremens. Time per division on he oscilloscope mus be se ha single pulses fi in single records. The recorded pulse is used for signal reconsrucion only if i mees he following requiremens: Firsly he pulse does no clip and secondly i is unique which means, he pulse is in a specral comparison o already saved pulses no correlaed or similar o one of hose. This leads o a enormous daa reducion. From he record of a high amoun of pulses, a reliable esimaion is calculaed. The esimaion gives an informaion in wha relaion he frequency of occurrence of he pulses sand o each oher. Weighed wih ha esimaion, he pulses are randomly Fig. 4. Reconsruced imescale from he probabiliy densiy esimaion. disribue on he marks of he imescale as shown in Fig. 4. The resul of he pulses disribuion on he reconsruced imescale is shown in Fig. 5. As he signal conains pulses of varying ampliudes, he rouine which is described above needs o be done for differen oscilloscope verical resoluions. The oscilloscope rigger-level mus be adaped o each verical oscilloscope seing (e.g. fix rigger level on he oscilloscope screen). This ensures improvemen of he measuremen dynamic range. From his resuls as many imescales as many verical scales have measured. Las sep is o merge hese imescales ogeher o one signal represenaion. 3.3 Merger o one signal represenaion Tha procedure of merge o one signal represenaion is shown for he example of wo imescales in Fig. 6. In he fas frame mode he oscilloscope also riggers on pulses which clip. Tha means he imes beween rigger evens measured consider he clipping pulses. So randomly disribued pulses on he imescale of he measuremen wih higher resoluion are replaced by he randomly disribued pulses of he measuremen a lower oscilloscope resoluion. This needs o be done a places where hey overlap. In case here exis measuremens for furher oscilloscope verical seings, he procedure described above would be done ieraively.
F. Krug e al.: A novel quasi-peak-deecor for ime-domain EMI-measuremens 29 u () u 2 ( ) R 1 C R 2 Criically damped meer 1 2 3 n Fig. 5. Pulses disribuion on he reconsruced imescale. Fig. 7. Analog Quasi-Peak-Deekor. higher resoluion x[k] b 1 b lower resoluion a_1 a_2 a_3 n b_1 b_2 n -a 1 T y[k] resul Fig. 8. Signal flow of a IIR1-filer. a_1 a_2 a_3 b_1 b_2 n Fig. 6. Example of a signal reconsrucion. 4 Auomaed TDEMI measuremen algorihm for class A signals In case he signal shows no pulsar envelope, or if imes beween pulses are shorer han fas frame mode is able o follow updaing rigger imes (e.g. limi of oscilloscope Tekronix TDS 7154 is 6 µs), doing saisical analysis brings no meri. A deailed measuremen descripion of his class of signal have already been presened in Krug (22c). 5 Specral esimaion 5.1 Windowing Specral esimaion is done wih use of Discree Fourier Transform (DFT). DFT compues he ampliude specrum on a ime record of lengh N. In addiion DFT is based upon he assumpion ha his ime record is repeaed hroughou ime. So he ime record x[n] needs o be muliplied wih a window w[n] of lengh N o avoid specral leakage: z[n] = x[n] w[n]. (1) The window funcion is equivalen o he bandwidh of he Inermediae-Frequency-filer (IF-filer) of he convenional EMI-Receiver. The window funcion needs o be adaped o he IF-filer regarding impulse-bandwidh and equivalen noise-bandwidh. Appropriaely he IF-filer is modelled wih a gaussian window. The DFT being applied o each of he daa blocks and is defined as follows (Marple, 1987): Z[k] = N 1 n= 2πkn j z[n] e N n = 1... N, k = 1... N (2) 5.2 Shor-ime-Fourier-ransform In order o receive he filer response in form of a ime dependen ampliude specrum, a DFT calculaion using record samples less han he lengh of he ime record N is necessary. The process of ieraively shifing he window along he ime record followed by compuing a DFT is he Shor-Time- Discree-Fourier-Transform (STFT) (Cohen, 1989). The window shif is done wih an appropriae incremen. The ime domain resoluion T sbb of he calculaed ampliude specra is imporan for an accurae filer response. T sbb corresponds wih he incremen used in he STFT as follows (Cohen, 1989): T sbb = 1 f sbb (3) f sbb is he baseband sampling frequency. The signal represenaion in Fig. 6 and ime-response for each pulse calculaed via STFT is given o he model of he Average-, Peak-, RMS,- and Quasi-Peak-Deecor. 6 Deecor model 6.1 Analog Quasi-Peak-Deecor In Fig. 7 he circui of a analog Quasi-Peak-Deecor is shown. The demodulaed signal u() of he convenional
3 F. Krug e al.: A novel quasi-peak-deecor for ime-domain EMI-measuremens Table 1. CISPR specificaions for Quasi-Peak-Deecor. Frequency range Band A Band B Band C/D 9 15 khz,15 3 MHz 3 1 MHz IF 6dB Bandwidh,2 khz 9 khz 12 khz τ c 45 ms 1 ms 1 ms τ d 5 ms 16 ms 55 ms τ m 16 ms 16 ms 1 ms / dbµv 25 2 15 1 5 QP ESCS 3 QP TDEMI Accurae releaive inpulevel for consan oupu [db] 3 25 2 15 1 5 5 1 Band A Band B Band C,D CISPR 16 1 TDEMI 15 pulse [Hz] 1 1 1 1 2 repeiion frequency 1 3 1 4 Fig. 9. CISPR 16-1 and TDEMI-Sysem pulse response curve. EMI-Receiver charges he capacior C by he resisor R 1 as long as u() is above u 2 (). This is a ypical RC-charging wih he ime consan τ c. If he he inpu signal u() is lower han u 2 (), he volage u 2 () is discharged by resisor R 2. A criically damped meer wih he ime consan τ m is used o display he ampliude. A buffer is necessary o avoid any repercussions. The maximum of he shown value is aken as Quasi-Peak-Value. The values of ime consans and IF bandwidhs for CISPR 16-1 (CISPR16-1, 1999) are shown in Table 1. For proper Quasi-Peak-Deecion he inpu signal mus be provided up o 2 s, o have a ready seady sae. The problem of odays Quasi-Peak-Deecors is ha provide he Quasi-Peak-value of a single frequency wihin 2 s. For abou 15 frequency poins, in a normal CISPR C, D measuremen abou 9 h are necessary. 6.2 Digial Quasi-Peak-Deecor model For he inegraion of a Quasi-Peak-Deecor ino a TDEMI- Sysem i is necessary o provide a digial equivalen realizaion, ha can be run a leas 1 imes faser han he analog equivalen. u() is provided by inelligen recording echnique and a specral esimaion via STFT. For modelling he charging and discharging process digial IIR1-filers are 5 1 2 3 4 5 6 7 8 9 1 Frequency / MHz Fig. 1. Drill machine: Comparison beween TDEMI-Sysem and conv. EMI-Receiver. used. Figure 8 shows he signal flow of a IIR1-filer. For calculaing he filer coefficiens following ransfer funcion is used: H (s) = 1 1 + τs. (4) Also he bilinear ransformaion for characerizing he RCsysem is used. The bilinear ransformaion is defined as follows: s = 2 z 1 z + 1. (5) s is he frequency-domain variable, z is he discree imedomain variable and is he ime-domain discreisaion. In order o ge beer numeric precision no he normal opology of a IIR2-sysem, bu a cascading of wo idenical IIR1-filers for modelling he criically damped meer were used. The maximum of he oupu signal is aken as Quasi-Peak-Value. Charging and discharging is decided by a comparaor, while in discharging mode he coefficien b is se o zero. The model of he digial Quasi-Peak-Deecor was wrien for a convenional Inel Penium III 1 GHz processor, wih he use of inrinsics. Therefore simulaion ime of a single Quasi- Peak-value is abou.4 s. 6.3 Signal represenaion for Quasi-Peak-Deecion A main problem is providing a suiable signal for he Quasi- Peak-Deecor. By recording up o 2 s wih a TDEMI-Sysem a 5 GS/s a leas 1 Gbye RAM necessary. Therefore he signal is reconsruced from represenaive ime-domain signal pars and heir corresponding ime samps. The reconsrucion is done afer he STFT and he demodulaion. A ime-domain signal is creaed for 2 s and is esimaed by he Quasi-Peak-Deecor. The sampling frequency of he imedomain signal mus be chosen properly o minimize he error beween he digial and he analog implemenaion. For
F. Krug e al.: A novel quasi-peak-deecor for ime-domain EMI-measuremens 31 5 4 QP ESCS3 QP TDEMI Accurae 4 35 3 QP ESCS3 QP TDEMI / dbµv 3 2 1 / dbµv 25 2 15 1 5 1 1 2 3 4 5 6 7 8 9 1 Frequency / MHz 5 5 1 15 2 25 3 Frequency / MHz Fig. 11. Lapop: Comparison beween TDEMI-Sysem and conv. EMI-Receiver. Fig. 12. SVGA Monior: Comparison beween TDEMI-Sysem and conv. EMI-Receiver. Band CD (CISPR16-1, 1999) abou 5 khz is sufficien. The reconsrucion is equivalen o reconsrucing he signal in he ime-domain and applying he STFT and Quasi-Peak- Deecion, as long as he pulse response of he IF-filer do no overlap. The error is minimized by selecing he appropriae mode of he inelligen recording algorihm. 7 Measuremen resuls 7.1 Simulaion CISPR 16-1 pulse For approving he Quasi-Peak-Deecion in ime-domain he CISPR 16-1 pulse response curves simulaed a a single frequency is shown in Fig. 9. The CISPR 16-1 and he TDEMI- Sysem pulse response curve show a mah for all pulse repeiion frequencies. 7.2 Emission measuremen Figure 1 shows he resul of a measuremen wih he TDEMI-Sysem on a drill-machine in band C,D for Quasi- Peak-Deecion. The average deviaion over he whole frequency range is below 3 db. The drill-machine emi pulses wih irregular repeiion frequency. Pulses of he drillmachine exremely differ in ampliude and specrum. The auomaed TDEMI measuremen algorihm for class C signals was used. Figure 11 shows he resul of a measuremen wih he TDEMI-Sysem on a Penium 2 MHz Lapop in band C, D. The average deviaion over he whole frequency range is below 3 db. A lapop emi pulses wih a high, no regular repeiion frequency. The auomaed TDEMI measuremen algorihm for class A signals was used. In Fig. 12 a measuremen wih he TDEMI-Sysem of a SVGA-Monior in band C is shown. A good mach of he emission measured wih he convenional EMI-Receiver and he TDEMI-Sysem is shown. Even he low specral lines of he monior power supply show a maximum difference below 1.5 db. The monior emi pulses wih a high bu consan repeiion frequency. The auomaed TDEMI measuremen algorihm for class A signals was used. 8 Conclusion The presened broad-band ime-domain elecromagneic inerference measuremen sysem allows o emulae in realime he various modes of convenional deecor modes. Wih he presened ime domain measuremen sysem he measuremen ime can be reduced by a facor of 1. A novel signal recording rouine for ime-domain EMI (TDEMI) measuremens and Quasi-Peak-Deecion is described. Measuremen resuls obained from he invesigaion of a drill machine, monior and lapop obained wih he ime-domain elecromagneic inerference (TDEMI) measuremen sysem are discussed. The resuls obained wih he described sysem have been compared wih measuremens performed wih a convenional EMI receiver. References Krug, F. and Russer, P.: Time-domain Broad-band EMI Measuremen Techniques, 32h European Microwave Conference, Milan, Ialy, 23 27 Sepember 22, 641 644, 22a. Krug, F. and Russer, P.: Ulra-fas broadband EMI measuremen in ime domain using classical specral esimaion, 22 IEEE MTT-S Inernaional Microwave Symposium Diges, 2 6 June, Seale, USA, 2237 224, 22b. Krug, F. and Russer, P.: Ulra-fas broadband EMI ime-domain measuremen sysem, 22 Inernaional Symposium On Elecromagneic Compaibiliy Diges, 9 13 Sepember, Sorreno, Ialy, 379 384, 22c.
32 F. Krug e al.: A novel quasi-peak-deecor for ime-domain EMI-measuremens Krug, F. and Russer, P.: The Time-Domain Elecromagneic Inerference Measuremen Sysem, IEEE Transacions on Elecromagneic Compaibiliy, 45, 2, 33 338, 23. Marple, S.-L.: Digial Specral Analysis wih Applicaions, ISBN - 8493-7892-3, Prenice-Hall, 1987. Cohen, L.: Time-Frequency Disribuions A Review, Proceeding of he IEEE, 77, 7, 941 981, 1989. CISPR16-1: Specificaion for radio disurbance and immuniy measuring apparaus and mehods Par 1: Radio disurbance and immuniy measuring apparaus, Inernaional Elecroechnical Commission, 1999.