GB1496837A - Method and apparatus for measuring the frequency and/or period of a signal - Google Patents

Abstract

1496837 Measuring frequency and period YOKOGAWA-HEWLETT-PACKARD Ltd 19 Nov 1975 [20 Nov 1974] 47695/75 Heading G1U A method of measuring the frequency or period of a signal (such as a fetal heart beat) which although substantially periodic is contaminated by various randomly occurring other signals, so that its shape varies in successive cycles thereby making the identification of a particular point within each cycle of the signal, and hence its period, difficult, the method being of the type in which in order to produce a train of pulses having the same frequency as that of the sensed signal, the sensed signal is cross-correlated with a reference signal and a pulse is produced upon the occurrence of each successive peak in the cross-correlation signal, is characterized in that the reference signal is periodically modified in accordance with the actual shape of the sensed signal, in such a way that the shape of the reference signal tends to follow the average shape of the sensed signal. Although the system described is entirely electrical the principle of the crosscorrelation method is understood most easily from the optical analogy. Firstly the method assumes that the shape that the sensed signal would have in the absence of contamination is known, and this ideal shape is taken as the reference signal. Both the sensed signal and the reference signal have to be thought of as being drawn on glass slides and made into masks by leaving the areas below the signals transparent and making the areas above the signals opaque. The reference signal mask need only be of one cycle in length, but the sensed signal mask must be thought of as extending indefinitely. The reference mask is then held stationary, the sensed mask is slid uniformly along it, and light is directed through the two masks. As the signal mask is moved the amount of light emerging (which corresponds to the crosscorrelation function of the two signals) varies from a low level, when the reference signal and sensed signal register least well with each other, to a maximum level when the two signals register best with each other. Clearly perfect registration is only possible if the sensed signal is not contaminated, but in practice quite a lot of contamination can be present in the sensed signal without changing the position in time of the successive peaks in the total amount of light passing through the two masks. The occurrence of these peaks is detected, and an output pulse train is produced from which the frequency or period of the sensed signal can be obtained. Hitherto, however, the requirement that the ideal form of the sensed signal be known has limited the application of the method to systems such as radar, where the echo should be a copy of the transmitted pulse. However, the invention shows how by starting with an estimate of the ideal form of the sensed signal - this estimate being taken as the initial shape of the reference signal and being referred to as a 'seed' - it is possible to modify the reference progressively, using the sensed signal itself, so that it approaches the unknown ideal form. As described the sensed signal 10, Fig. 1, is repeatedly sampled and digitized 11, and a block of 257 of the samples are stored in a recirculating sensed signal register 18, circulating at 256 times (=2À5 ms/cycle) the sampling rate. This results in each sample appearing on a lead S and being followed by the immediately proceeding 255 samples before the next sample occurs. (The 256th preceeding sample is written over by the next sample). The reference signal is stored in a 256 word recirculating register 30, which is so phased with the sensed signal store that the occurrence of a sample on lead S for the first time coincides with the beginning of the reference signal on a lead D. The two signals S and D are multiplied 19 and their products accumulated, 33, 35, to give the cross-correlation signal C defined by:- Thus immediately following each sample of the sensed signal, and before the next sample occurs, the corresponding value of the cross-correlation signal is computed using the current sample and the preceeding 255 samples. The successive values of the cross-correlation signal are converted to analogue form 39, appropriately smoothed, and passed to a peak detector 40, which produces one pulse, on a terminal 41, after a fixed delay of 250 ms, (this delay being less than the minimum expected period of the sensed signal) for each peak detected, unless a further larger peak occurs before the delay of 250 ms expires in which case the delay recommences from the further larger peak and the earlier peak is disregarded. The system described so far merely implements the general prior art method and would operate satisfactorily if the reference signal were optimal. However, this is not so and therefore means are provided to modify the reference signal progressively. The principle employed to modify progressively the reference signal is that those sample values S which result in an increasing cross-correlation signal C are averaged with the corresponding existing values of the reference signal, thereby producing an updated reference signal. Thus, whenever the crosscorrelation signal C is increasing in value, this being detected by the unit 40, the samples S then occurring are fed into an intermediate recirculating register 24, overwriting those samples already there. Once the cross-correlation signal C stops increasing, indicating a peak, firstly the delay period of 250 ms is initiated and secondly further input of samples into the intermediate register is blocked. Eventually (providing no further peak occurs) the 250 ms delay will expire and an output pulse 41 of 2.5 ms duration will occur. Simultaneously with the output pulse, a control unit 25 forming part of the reference signal store causes the reference signal to be modified by replacing each of the 256 values thereof by the arithmetic average of the existing value and the corresponding value in the intermediate store 24. This operation takes one complete cycle of 2À5 ms and therefore occurs between two consecutive data sampling instants. The output pulse 41, in addition to being passed to a frequency determining unit, not shown, also resets the peak detector unit 40, so that once the crosscorrelation function C again starts to increase, the samples then occurring can be fed into the intermediate store. Thus, progressively, the reference signal is modified in such a way as to maximize the largest peak in the cross-correlation signal. Peak occurrence detector Fig. 2 The successive values of the cross-correlation signal C are converted to analogue form 39, Fig. 1, and smoothed 403, Fig. 2, and the resulting waveform is applied to a peak detector 405, 409. Whilst the input waveform C is rising, the input to an amplifier 411 is positive and a transistor 417 is held conductive, thereby providing a positive signal 429. Once the input waveform begins to decrease, the output of amplifier 411 goes positive, thereby turning a transistor 413 off so that a capacitor C 1 commences to charge via a resistor R 1 , and eventually after 250 ms, unless a further larger peak occurs, an amplifier 419 produces a pulse which (a) discharges the peak storage capacitor 409, and (b) is shaped to provide a final output pulse that can be passed to an appropriate frequency measuring circuit.