A novel automatic digital quasi-peak detector for a time domain measurement system

A novel automatic Digital Quasi-Peak Detector for a Time Domain Measurement System Florian Krug

Stephan Braun Lehrstuhl fur Hochfrequenztechnik Technische Universitat Munchen Munchen, Germany 80333 Email: [email protected]

General Electric Glohal Research - Europe Freisinger Landstr. SO, Garching 8S748 Email: [email protected]

Absnocf-In this paper an advanced ultra-fast broadband time-domain EM1 (TDEMI) measurement system is presented. Measurements were performed in the 30 loo0 MHz range. Using digital signal processing lor spectral estimation and detection, the measurement time is reduced by a factor 10 in comparison to a conventional EM1 Receiver. A novel recording mutine for TDEMl measurement and digital signal pmcessing for proper QuasiPeak detection is described. Difterent amplitude resolutions are selected during the recording to enhance the dynamic range by about 50 dB. Measnremenl results are compared with l e results obtained with a conventional EM1 Receiver.

-

Peter Russer Lehrstuhl fur Hochfrequenztechnik Technische Universitlit Miinchen Miinchen, Germany 80333 Email: [email protected]

consists of single pulses or bursts that may have a broadband spectrum. In the general case we consider an EM1 signal that consists of the superposition of transient and stationary signals. In the following a novel signal recording routine for TDEMl measurement and Quasi-Peak detection is descrihed that operates for the general case. An example of an EM1 signal is shown

1. INTRODUCTION EMC and EM1 measurement equipment which allows to extract comprehensive and accurate information within shon measurement times will allow to reduce the costs and to improve the quality in circuit and system development. The drawback of todays EM1 receivers is the long measurement time up to several hours for a frequency band from 30 loo0 MHz. Spectral estimation via Fast Fourier Transformation (FFT) allows reducing the measurement time by a factor of more than 10.

11. TDEMI MEASUREMENT SYSTEM The TDEMl measurement system is depicted in Fig. I . It

Fig. I .

The lime-domain EMJ Measurement System

consists of a low noise amplifier, an anti-aliasing lowpass filter. an oscilloscope for data acquisition and a PC for digital signal processing. An algorithm for recording in time-domain and evaluation of EM1 spectra in the Peak, Average and RMS detector modes is presented in [I]. We distinguish between stationary and transient EM1 signals. An EM1 signal is named stationary, if the RMS Value calculated for a certain time interval d w s not dependent on the position of this interval in time. The transient EM1 signal 0-7803-8443-1/04/$20.000 IEEE.

919

in Fig. 2. The considered EM1 signal is a superposition of stationary and transient signals. A major drawhack of timedomain EM1 measurement systems compared with frequency scanning systems is their lower dynamic range. To Increase the dynamic range of the TDFMI system a novel strategy is used to record the stationary pm of the EM1 signal and the transient pans of the EM1 signal belonging to varous amplitude classes separately. In the following a routine is described, that records stationary and impulsive signal parts separately. An impulsive signal is detected if the signal amplitude exceeds a certain threshold. After every passing of this threshold the pulse is digitized, recorded over a preselected time interval and further processed by short-time fast Fourier transform (STFFT). The lengths of the time intervals between the pulses is recorded together with the STFFT spectra of the pulses. A time-dependent spectrum is generated that is sutistically equivalent to the time-dependent spectrum of the original signal. At each discrete frequency value the signal in time-

rn.g”,rudr

Fig. 3.

between pulses of the emission generated by a drill machine. A model of the stochastic pulse train is generated. In this

Dismibutian of sections an the

timescale

domain that is statistically equivalent to the demodulated IF-signal of a conventional EMI-Receiver is extracted. This signal is evaluated by the Quasi-Peak detector at each discrete frequency value. 111. TDEMI MEASUREMENT ALGORITHM In order to record stationary and impulsive signal parts separately different trigger levels are used. Setting the trigger level above the maximum level of the envelope of the stationary signal only signal parts that contain pulses are recorded. In order to record the stationary signal the trigger level is set below the minimum level of the envelope of the stationary signal. Thus mainly stationary signal sections are recorded. If during the recording a certain threshold is passed, the signal part is recognized as impulsive and is eliminated.

intervals between pulses [ms] Fig. 4.

Probability density estimation far

model the marks where pulses occur follow the probability distribution of the time intervals between pulses determined experimentally. Fig. S shows the marks to..t, on the time

c

.

-:

.;-

c*

0

‘*

1,

A. Starioiiar> sigrial recor-ding

1.

Fis. 5. Marks on the time %ale

First of all samples of the stationary signal part are recorded. The amplitude resolution of the oscilloscope is adjusted above the maximum level of the envelope of the stationary signal, to provide a maximum signal-to-noise ratio (SNR). Recording is done hy several single shot measurements. Thus the signal is sampled in a numher of intervals of equal length. In order to reduce the computation time during the STFFT an algorithm to reduce thr required number of stored niznal sections is used. I ) A/gorirlir?i fo reduce die required rrroirber of sigrial serrioris: The complete set of recorded sections is divided into groups of similar sections. Two sections are similar if the ahsolute value of the cross-correlation coefficient of the power spectra is higher than a certain preselected threshold. To reduce the computation time during STFFT and to reduce also the memory requirements each group of similar sections is replaced by one representative section and an ahsolute frequency, which corresponds to the number of sections of this group. The absolute frequency is the number of nccurrences of a certain event during a random experiment. Weighted with the absolute frequency the representative sections are placed in a stochastic distribution over the time scale. The result of the distribution of representative sections over the timescale is shown in Fig. 3. B. Inrpiilsiw sigrral rerordirig I ) Meastrrenierrt af time irrren,als benweir trarisierir EM/ pulses: The time intervals between pulses are measured hy an oscilloscope, which is set in fast-frame mode [ 3 ] . To ohtain a probability density estimation a statistical kernel smoothing method is applied to the measured time intervals [4]. Fig. 4 shows the probability density estimation for the time intervals 0-7803-8443-1/04/$20.000 IEEE.

he emission of a drill machine

scale. 2 ) Algorirltnt to reduce the required riimi6er af prilses: Recording of pulses is done by several single shot measurements. Triggered data acquisition has already been presented in [I]. In the following a routine is described to reduce the number of required pulses. The complete .set of recorded pulses is divided into groups of similar pulses. Two sections are similar if the absolute value of the cross-correlation coefficient of the power spectra is higher than a certain preselected threshold. Weighted with the absolute frequency the representative pulses are placed on the marks on the time scale. The result of the distribution of pulses on the timescale is shown in Fig. 6. l%@ilUdC

I,,

1,

1:

1:

Fig. 6. Dislribulion of pulses on h e

I

c

timescale

3 ) Errlrortcenierir of the djrianlic range: In the general case the pulses show different maximum amplitudes. In order to enhance the dynamic range during the acquisition of impulsive signal various amplitude resolutions are selected. The used oscilloscope provides stepped amplitude resolutions i.e. 20 mV, SO mV. 100 mV, 200 mV, 500 mV. 1 V 2 V S V and I O V.

920

~

mayityde

Measurements are performed at every amplitude setting of the oscilloscope. For each amplitude setting only these pulses are recorded that have maximum amplitude that is smaller than the amplitude resolution of the chosen setting and larger than the amplitude resolution of the previous setting. e.g.: Pulses that are recorded in the amplitude resolution 100 mV have a maximum amplitude that is higher than SO mV but lower than 100 mV. In order to record in this case only pulses with an amplitude higher than 50 mV, the trigger level is set above 50 mV. Recording of various amplitude resolutions enhances the dynamic range for impulsive signal evaluation according to

maTitude,

I

,

/

/

'

!

I

In the discussed example the dynamic range of the measurement is enhanced by 54 dB. C. Merging to one sratisrically equivdenr signal By recording of the time intervals between the pulses we obtain the probability discribution of this time intervals as depicted in Fig. 4. Assuming that the Occurrences of the pulses are statistically independent we can reconstruct for every amplitude interval the pulse train by appending pulses separated by intervals chosen of random length. The length of these intervals follows the above mentioned probability distribution. This reconsuuction is performed for every amplitude interval and for every set of similar impulsive signals. By this procedure we have constructed a pulse train statistically equivalent to the pan of the original signal with pulse amplitudes within a certain amplitude interval. During the reconstruction of the stationary signal a train of consecutive sections is generated. The reconstruction of the statistically equivalent signal to the complere original signal is perfohed by adding these reconstructions for all amplitude intervals. In Fig. I the process of adding the reconstructed signal of pulses from a single amplitude resolution with the reconstructed stationary signal is presented. A section of the stationary signal is replaced by a pulse of the next amplitude resolution, if the pulse is within the time interval of this section. In Fig. 8 the process of adding the reconstructed signal of pulses of the next amplitude resolution with the already merged signals is depicted. A section of the stationary signal or impulsive signal is replaced by a pulse of the next amplitude resolution, if the pulse is within the time interval of this section. The described algorithm is iteratively performed for all amplitude resolutions. The result is a reconstructed signal that is statistically equivalent to the original EM1 signal. ESTIMATION IV. SPECTRAL

A. Wi'ridowfurtcriori Spectral estimation is performed using the Discrete Fourier Transform (DIT). The D I T is used to calculate the amplitude spectrum of a signal sample o f length A'. In addition DFT 0-7803-8443-1/04/$20.00 0 IEEE.

92 I

Fig. 7.

Merging impulsiye d p n d seclions with walionary signal sections

magnitude

magnitude

Fig. 8.

Adding the timescale of next higher amplitude

is based upon the assumption that this sample is continued periodically in time. So the time record z[n]needs to he multiplied with a window w [ i i ] of length N to avoid spectral leakage: z(n1 = z [ n ]w[n]

(2)

As the window function w[n]a Gaussian window is used. In order to get at each discrete frequency value the same filter response as the intermediate-frequency-filter (IF-filter) of the conventional EMI-Receiver the window needs to be adapted to the IF-filter regarding impulse-bandwidth and equivalent noise-bandwidth, The DFI being applied to each of the data

For e[ir]