GB2040036A - Signal analysis, processing and correlation - Google Patents

Abstract

Signals are received and correlated with stored representations. If a new signal is received it is stretched in time by a processor and then stored. The processor includes ultrasonic light modulators C1, C2, light source LS, L1, L2, S1 lenses L3 to L7, and bandpass filters S2, S3 to produce at photodetectors Dd-Da spatial distributions of light which are a function of an impulse fed to modulator C2 and the new signal fed to modulator C1. By virtue of the lenses L4 and L6, the distributions move relatively to one another at a predetermined speed giving rise to the time expansion. The photodetectors produce electrical signals, representing the correlation of the impulse and the new signal, the electrical signals being time expanded representations of the new signal. Once a signal has been correlated it is analysed. <IMAGE>

Description

SPECIFICATION Signal analysis, processing, and correlation The present invention relates to the analysis, processing and correlation of signals.

There is a known requirement to examine signals received within a given frequency band which arise from transmissions from an unspecified number of sources of unspecified strength and characteristics. A transmission may be with or without angle modulation, and may be either continuous or, more commonly, amplitude modulated (in particular pulsed on/off). Very often, the manner of modulation is characteristic of a particular source.

Economy of circuit hardware has usually required that the frequency band to be examined is made as wide as is consistent with the detection of significant transmissions in the presence of the general background noise. The bandwidth occupied by signals from any one source may be taken to be no greater, and normally appreciably less, than the acceptance band.

The aim is for the overall combination of incoming signals to be analysed to identify and separate signals originating from different sources, and from these to endeavour to categorise and assess the identity and implications of these sources. This activity is commonly described as electronic support measures (e.s.m.).

For pulsed signals, the natural unit or event to deal with is the time interval equal to the pulse duration itself. Once a pulse has ended, measurements made of relevant parameters, possibly in multiple channels with differing directional sensitivies, are put together to form a 'descriptor' package. On the presumption that signals transmitted from any one source all give rise to essentially the same descriptor, specific to the source, it can be made the basis for segregating one signal from another. Examples of immediately indicative parameters are pulse duration, mean frequency and bandwidth, while time of occurrence becomes relevant only in subsequent analysis.In respect of timing information, 'instantaneous' high-bandwidth measurements are appropriate, whereas to assess the carrier structure within a pulse implicates data from many time-resolution elements of duration approximately 1/(Acceptance Bandwidth).

The implementation of a scheme of this sort is straightforward, provided that pulses occur one at a time.

However, difficulties arise with a composite pulse-over-lap in pairing off each rise with its associated fall. In general, the narrower the acceptance bandwidth, the lower the probability of pulse overlap; thus it has been proposed to provide a receiver having many narrow-band channels.

Although an interrupted continuous wave (i.c.w.) transmission may, in the strict sense, be regarded as pulsed, the pulse duration tends to be inconveniently long for processing. Accordingly, analysis is conducted upon segments, with the descriptor including an extra item notifying whether the signal was still present at the time the segment ended. Obviously, this arrangement is applicable to true c.w. transmissions as well although it might then still be necessary forthe receiverto chop the input from time to time in order to sense the system noise level; from this, the detection threshold would be reset so as to maintain the required probability of false alarm.

With the exception of the above quasi-continuous detection, the interval between successive events attributable to a given source will usually have a minimum value which, for many surveillance applications, will be several tens of lis at least. The handling of data of this kind of repetition interval accords well with the capabilities of a general purpose digital computer. In contrast, the signal processing involved in the compilation of the descriptor, while not an exclusively analogue task, has to operate in real time, and with a bandwidth, (i.e. at a rate) rather too great for the input/output facilities of the general purpose computer.

The difficulties which arise in attempting to distinguish between signals which coexist within the working, or acceptance, frequency band encourage the consideration of pattern recognition techniques as an alternative method of extraction. Here, the signal itself (or some linear tranform of it) is compared with each of a number of stored patterns in order to determine which, if any, provides a close match. Such a system is comparatively insensitive to accompanying non-matching signals. In a practical system, this matching process would have to be performed in real time. Using electronic techniques, it can be a formidable task, particularly if the density of sources is such that there is not much time after one event before the next takes place. It does, however, fall within the reach of electro-acoustic-optical methods.

In outline, in a previous proposal the incoming signal is caused to form a one-directional spatial distribution of light intensity (e.g. as a function of time or, perhaps frequency). The relative intensities are then weighted by means of a reference pattern of transmission (for instance, of variable opacity or refractivity), and the weighted means intensity derived: this constitutes a measure of the cross-correlation coefficient. By repeating this process for a number of relative displacements between distribution and reference (for instance by continuous longitudinal scanning) and comparing results, the point of maximum correlation can be determined.In the class of signal of which a particular reference pattern is representative, a detection is made whenever, after normalisation with respect to total signal energy, the peak value determined exceeds a pre-arranged threshoid of significance.

Generally, correlation would be performed for a number of distinct reference patterns, possibly sequentially but, because of the need for a quick verdict, more likely in parallel. Clearly, there would then be economic pressures in favour of minimising the number of correlation channels.

Systems have been proposed in which, over a long period of time, a 'library' is compiled of known signal classes, together with their respective 'signature patterns. Whenever the analysis system is put on the alert, an estimate is made of those known classes most likely to be encountered, and the correlators are primed with the corresponding reference patterns. There are some drawbacks to this scheme, such as (a) To ensure a good probability of classifying all signals received, reference patterns for many classes will have to be deployed, of which an appreciable proportion prove redundant.

(b) If a transmission should be received for which a match cannot be found, then a new array of patterns, with a matching pattern now included, has to be substituted. To elicit the pattern and compile a new array may impose an appreciable delay before a repetitive aequence of transmissions of this type can be resolved.

(c) In addition, if new pattern sets can be prepared only from entries already in the main library, it is not at all clear what search strategy would be needed in order to find the match with minimum delay (assuming that such a signature was already on file).

It is an object of the invention to provide an alternative method of analysing signals.

According to one aspect of the present invention, there is provided a method of analysing signals, comprising the steps of A) receiving a signal B) feeding it to a correlator for correlating a representation of it with a representation of a reference signal stored therein C) storing the representation of the received signal in the correlator as a further reference representation, if it does not correlate with any reference representation previously stored therein, for correlation with a subsequently received signal, and D) analysing both the reference signal and sequences of those received signals which correlate therewith, wherein steps A), B) and C) are carried out substantially at the time of reception of the said received signal, and step D) is carried out at least in part with a substantial, often random, delay with respect to the said time of reception.

Thus, in accordance with said one aspect, those activities, i.e. steps A), B) and C) which require a prompt real-time reaction are separated from these other, i.e. step D) which can tolerate a variable delay. Steps A) B) and C) requiring real-time reaction are equivalent to the assignment of a received transmission to a source (identified or not) and the formulation of a descriptor. Step D) involves the analysis of signal characteristics and of transmission sequences by now attributed to single sources, the deducing of p.r.f., p.r.f. stagger, scan rate etc. which, in turn, would form the basis for classifying the type of source and hence to deduce the likely class of conveyance involved. This more leisurely analysis task is well suited to data processing by general purpose digital data processors.

Thus, in accordance with said one aspect, for source/signal association, full characterisation is not treated as being of immediate importance and is left in abeyance. However, the question of whether a transmission is, on the one hand, a repeat of one received already (whatever its characteristics) since the beginning of operations, or/is on the other hand a stranger, is decided in real time.

In an embodiment of a system for implementing the method of said one aspect of the invention, the correlator has a store for storing representations of the reference signals which initially has zero content, with representations being entered into the store as and when a new (non-matching) signal is received.

Preferably, the delay between deciding that the signal has not been received before and a representation of it being stored should be low, e.g. about I ms. This would minimise the loss of time before resolving a repetitive sequence of such signals. Forming of representations must be sequential, but subsequent use of stored representations in parallel is possible.

The embodiment has a device for forming the representations as well as for storing them. Preferably the device for forming and storing representations should have a fairly high writing speed, and a sufficient number of parallel channels to accommodate representations of signals from the maximum number of sources likely to be encountered together. However, it would be quite tolerable for erasure to occupy a much longer time interval (several minutes, say) pending the next call for use.

Whatever form the representation store may take, it is likely that the time for writing representations into it would be longer than the duration of the signal to be represented. Means for the temporary storage of the signal for subsequent read-out in stretched time are therefore needed.

It is another object of the invention to provide a signal processor capable of changing the scale in time of a signal, e.g. by stretching or expanding it in time, to make it conform to the writing speed characteristic of a device, such as a store, utilising the signal.

According to another aspect of the invention, there is provided a signal processor wherein means are arranged to produce spatial distributions of light which are functions of an impulse signal and a signal to be processed, the distributions moving relative to one another at a predetermined relative velocity, and means are arranged to respond to the said distributions in combination to produce an electrical signal representing the correlation of the impulse signal with the signal to be processed, the arrangement being such that the said electrical signal represents the signal to be processed scaled in time as a function of the said relative velocity.

It is a further object of the invention to provide a correlation arrangement into which a representation of a signal can be stored in real-time albeit in a total time interval longer than the duration of the signal to be represented.

According to a further aspect of the invention, there is provided a correlation arrangement including: (A) a signal processor according to said another aspect arranged such that the said electrical signal represents the said signal to be processed expanded or stretched in time; (B) a store for storing a first representation of the said electrical signal;; (C) a correlator comprising optical storage means on which an optical representation of the said electrical signal can be formed, means for forming the optical representation, means for producing a first spatial distribution of light which is a function of an optical representation of a signal formed on the optical storage means, means for forming a second spatial distribution of light which is a combination of the said first distribution and a further distribution of light, which is a function of the said signal to be processed, the first and further distributions moving relatively to another, and means responsive to the second distribution to produce a correlation signal representing the correlation of the signal to be processed and the signal represented on the optical storage means; and (D) means for causing the forming means to form on the optical storage means an optical representation of the signal stored in the said store if the signal to be processed does not have a predetermined correlation with any signal represented on the optical storage means.

The said optical storage means and forming means may be a cathodo-chromic tube as disclosed in British Patent 1 322 337 (EMI Ltd) for example. Alternatively writing by means of a scanned laser beam onto photochromic material or so as to locally remove a thin opaque film from a clear carrier could be used for forming and storing the optical representations.

For a better understanding of the present invention and to show how the same may be carried into effect, reference will now be made, by way of example, to the accompanying drawing, in which: Figure 1 is a block diagram of a singal analysing system in accordance with the said one aspect of the invention, Figure 2 shows a signal processor according to the said another aspect of the invention and a store as used in the system of Figure 1, Figure 3 Ai to iX and B to G and 4 are diagrams explaining the operation of the signal processor, and Figure 5 shows a correlator as used in the system of Figure 1.

Analysis system In the analysis system of Figure 1 a signal to be analysed, for example, a radar pulse from an unknown source, is received by a receiver 1 where it is mixed with a local oscillator signal to produce an l.F. signal of reduced carrier frequency e.g. 100 MHz. The l.F. signal is applied to a signal processor 2 (to be described hereinafter with reference to Figures 2,3 and 4) and to a correlator 3 (to be described hereinafter with reference to Figure 5). The correlator 3 is provided for correlating the l.F. signal with a stored reference representation. However, it is assumed that initially the correlator contains no stored reference representations. Thus if the l.F. signal is the "first" signal to be received its existence is recognised, but it is not identified within correlator 3 with a reference representation.This circumstance is notified to a controller 4 via its input 6. Controller 4 then causes a representation of the l.F. signal to be written into the correlator 3.

The speed of writing a representation into the correlator is insufficient for the writing to be performed wholly within the duration of the l.F. signal to be analysed. Temporary storage is therefore necessitated and is provided by the signal processor 2 and an intermediate store 5. It is assumed that the signal bandwidth is too high for direct acceptance into store 5. The signal processor, as described hereinafter stretches the I.F.

signal in time to make it match the write-in capacity of the store 5. The stretched signal is fed to the store 5, supplanting previous contents, and is then held until called for by controller 4 to be applied to the correlator.

A set of reference representations may thus be written into the correlator. If the l.F. signal is identified with a reference representation already in the correlator it is, of course, unnecessary to write a representation of it into the correlator.

By means of the processore 2, store 5, correlator 3, and controller 4, a signal can be correlated to identify whether or not it is new and, if it is, a representation of it is written into the correlator substantially in real-time.

The controller 4 is also arranged to feed signals to a general purpose computer so that a more thorough and long term analysis of them in combination can take place. This analysis is of signals which correlate with any one stored representation (and are thus "known") and involves the derivation of, for example, pulse repetition frequency (p.r.f.), p.r.f. stagger, scan rate etc. from long term examination of sequences of the signals. Analysis of pulse content, for one pulse within the sequence may also be relevant. The analysis is not carried out in real time.

Signal processor and store Referring to Figure 3A i consider an input signal, p(t), which is a continuous function of time. Let us divide it into a number, M, of interleaved trains 31, q (m,t), (m = 1,2..., M), each train comprising segmentas 32 of it, separated by intervals of zero amplitude (see Figures 3A ii to v). We can define the then th segment of them thtrainby q (m, t) = p(t) .. (1) for n+(m-1)/M < t/T < n+mlM where n is an integer. It follows that segments 32 appear in each train 31 at time intervals of T.

For example, if M=4 then, in the first train (i.e. m= 1 ) the nth segment would exist from t=nT to t= (n+ 1 /4)T.

From each train, it is required to generate an output r(m,t) (see Figures 3A vi to ix) such that r(m,t) = p(t/M + (n+(m-1)/M)(M-1)T/M .. .. (2) for n+(m-1)/M < t/T < n+1+(m-1)/M Thus, continuing with the same example, in the first channel the nth segment, T/4 long, occurring from nT to (n+1/4)Twould, in T(1,t), be uniformly slowed down so asto fill the interval from t = nTto (n+1)T.

It will be appreciated that, neglecting discontinuities at the ends of segments, the output r(m,t) occupies a frequency band of the width I/M forp(t).

For the purpose of explaining the basis of the operation of the processor, let us assume the signal, p(t), whose waveform with respect to time is shown in Figure 3A and assume also that M=4. The segments of interest will be those 32, corresponding to the train q (I,t), where t goes from 0 to T/4, from Tto 5T/4 and so on.

Referring to Figures 3 B to G, this signal p(t) is caused to travel linearly, and with low atenuation in a propagating medium across an aperture of width X. The speed of propagation is chosen equal to 4X/3T. At the same time, a single impulse 33 is caused to traverse the same aperture but, at speed X/T. Their instantaneous positions at tiT = 0, 1/4, 1/2,3/4 & 1 are shown in Figure 3B to F respectively.

It is arranged for the impulse 33 to serve as a windowing function which, by multiplication, results in spatial domain samples being taken of the continuous propagating signal, p(t). The values of the resulting continuous time function output at the instants t=0, T/4, T/2, 3T/4 will therefore be seen to be equal to the respective values of the input at a proportion 0, 1/4, 1/2,3/4 of the way along the first segment. These are p(0), p(T/16), p(T/8), p(3T/1 6), and their plot in Figure 3G clearly exhibits the required time expansion; (3G shows r(l,t) = p(t/4) for n = 0).

An example of a processor implementing the above described principle is shown in Figure 2, for M=4.

Here light passes through two ultrasonic column light modulators C1 and C2 in succession, the speed of propagation being the same for both, and the required 4/3 ratio is obtained by optical magnification.

A pinhole aperture in a stop plate, S1, is illuminated by a near-monochromatic light source, LS, by way of a condenser lens pair, LI, and 2. This aperture behaves as a secondary uniform source for collimating lens L3, which is associated with ultrasonic light modulator column, Cl.

Collimated light leaving Cl is focused by 'transforming' lens, L4, onto the plane of stop S2 where it is band pass filtered in the well-known manner (see Felstead Trans IEEE AES -3 No.6 Nov.1967 pp 907-914) in respect of spatial frequencies; suitably proportioned apertures permit only selected frequency components to pass through to the next stage. Light diverging from a given spatial element in CI, and passing through the filter aperture(s) in S2, is rendered convergent by lens, L5, to form a reduced image of Cl in the plane of a second ultrasonic light modulator, C2. Immediately prior to its arrival at C2, light is passed through collimating lens, L6. The distances, f4 and f6, shown in the figure will be seen to be the focal lengths of L4 and L6 respectively as well as the respective object and image distances of the imaging lens, L5.In the present example, f6/f4 = 3/4.

If the full aperture of C2 were to be employed, it would be followed by a single transforming lens which gives rise to a frequency distribution in its transverse focal plane. In the present case, however, there are four such lenses, L7 a to d, of width one quarter of the column length, each cooperating with a corresponding zone (a) to (d), of the column. Note that these lenses are each a central section cut from a spherical lens, and consequently have the same focal length for all planes of incidence. Taking the situation when neither C1 or C2 is energised, four datum images of S1 aperture would be formed in the plane of the output stop S3.

Considering in isolation the effect of lens L17a, it will be seen that it would form a frequency-dependent distribution in S3 in a similar way to L4 with S2. Offset from the datum image is a spatial frequency band-pass filter in the form of an extended slit aperture in S3, parallel to the propagation axis of C2. Light passing this aperture is collected and caused to fall on a square-law photo-detector, Da, to give an output whose dominant alternating components are selected by band-pass electric filter, Fa. Similar arrangements apply for each of the other three outputs.

The processor operates as follows: The sampling pulse waveform is generated in Cl, while the signal waveform is carried by C2. The speeds of propagation in each are nominally equal but, in the plane of C2, the image of Cl's events travels only 3/4 as fast. It the required sampling is to be carried out, it is necessary to form the product of instantaneous values for corresponding points along C2, an operation performed in the photo-detector as a consequence of its square law characteristic.

In order that both amplitude and phase information should be conveyed by the output, the sampling pulses have to take the form of a pulse of an alternating carrier. This is most easily accomplished by exciting the transducer, X1, of C1 with a short duration oulse, so producing a broad spectral distribution in the plane of S2. An offset aperture formed in S2 extracts a band of spatial frequencies whose width and centre value determine respectively the envelope and carrier of the sampling pulse in the plane of C2. Additionally, there is a low-pass filter in the form of a small central aperture which accepts the zero order (zero frequency) component.

A suitable transmission characteristic for the complete S2 filter is shown in Figure 4(a), with spatial frequency in units of 5/v where B Hz denotes the i.f. bandwidth and v m/s denotes the speed of propagation of acoustic signals along C2, corresponding to the value 4X/3T of Figures 3B to G.

The resulting modulation function at C2 instantaneously is of the form of a carrier, of spatial frequency 3 B/v cycles/metre, modulated by a pulse envelope whose main lobe is some 0.85 v/B m wide. This envelope corresponds to the impulse, 33, in Figure 3. Furthermore, if the propagation speed isv m/s in Cl as well as C2 then, because of the optical reduction, the resulting wave packet travels along C2 at a speed of 3 v/4 (i.e. X/T).

Let us denote the envelope by a (x) where a (0) is the peak value. Then the modulation function, including the zero frequency term, go, may be expressed as go + g (x,t), where g (x,t) = a (x-3vt/4) exp U2(3BN).(x-3vt/4)i .. .. (a) If transducer X1 is left unenergised, then z (x) is zero and, in C2, g (x,t) is a uniform go. With a representative signal fed to X2, the resulting modulation function of light leaving C2 is of the form go s'(x,t) where s'(x,t) = so + 2b(x-vt). cos[2x(fslv).(x-vt)l .. .. (p) Here 2b.(-vt) and fs are respectively the envelope and the carrier frequency (i.f. band) of the signal, applied at the point = 0.

The corresponding spatial spectrum is shown in Figure 4(b) together with the acceptance band, shown dotted. When this spectrum is realised in the plane of S3, it is filtered according to the transmission characteristic of Figure 4(c) provided by the aperture. The transmitted component, s(x,t), of s'(x,t) is therefore s (x,t) = b (x-vt) exp [j2z (fs/v) (x-vt)] .. .. (y) When both X1 and X2 are energised, the light transmitted by the S3 aperture corresponds at C2 to a function proportional to u (x,t) = so g (x,t) + gos (x,t) .. .. (6) Then, at a given time, t, the component of instantaneous power, incident upon detector D, and originating from near x is proportional to I u (x,t) 2 2, Of this, the alternating component of interest is due to the cross-product of the constituents and is so go a (x-3vt/4).b(x-vt).cos [2a (3B-fs)x/v - 2as (2.25B-fs)t] .. ..

Integrating over the total range of x yields the total instantaneous power. In practice the value of a and hence of the integrand, departs significantly from zero only over a range of about + 0.3 v/B in x. The magnitude of (3B-fs) is low enough for only a small error to be incurred if, when integrating, the central value, xo(t), of x is used in the argument of the cosine. It is then possible to move the cosine to outside the integral. The reduced integrand involves the product of a and b and, since b changes slowly compared with a, is determined by the value of bat and about xo(t). Thus, in effect, b is being sampled by a just as the signal envelope 32' is to be sampled by the impulse envelope 33.

From (e), we see that xo(t) is equal to 3vt.4 and, by substitution, we find that the argument of the cosine becomes 2 (fs/4)t. Clearly, the whole input signal, and not just its envelope has been stretched four times in time. Figure 4(b) shows that the pass-band is well above zero, so it is possible to separate the alternating signal from variations due to drift by means of a band-pass filter.

The operation of the system, as described above and in respect of zone (a) of C2, realises the principles put forward above and shown in Figure 3A to F except that it includes the limitation wherein, after the signal segment and sampling oulse have both traversed the observation aperture, width X, the three following segments would already be within the aperture and in transit: they could not therefore be sampled by any subsequent pulse 33. However, the addition of zone (b) associated with lens L7b enables the original pulse to continue its slow scanning of the past signal while, at the same time, a new pulse embarks on a new cycle with a new segment from the next value of n. With all four zones (a) to (d) equipped, a pulse can thus be fed into X1 at intervals of one quarter of the transit time down C1.It will be seen that, except when entering and leaving, only one sampling pulse influences each photodetector at any one time.

The outputs of filters Fa to Fd form respectively the four inputs to quarters of a store 5. Clock pulses, at B/4 Hz are applied via OR gate, OG, causing the inputs to be sampled simultaneously, and their values to be stored sequentially. After a delay of 1/3B s, that is one quarter of a period of a period of the stretched centre frequency, provided by delay, another set of (quadrature) samples is stored. Thus a variable is stored whose envelope is a signal 32" expanded in time relative to the signal 32' from which it is derived as shown in Figure 3G. Carrier phase is conveyed by the relation between the in-phase and quadrature samples. It will be appreciated that the sampled values in these sequences are equal to those which would be obtained from frequency translation to two band-limited quadrature base band channels of + B/8 Hz, followed by sampling of each of B/4 Hz.

For this purpose, during the time T of propagation of the impulse 33 past all four detectors Da to Dd, detector Da is sampled for a time period a during which impulse 33 moves past that detector and so on for time periods b, c and dassociated with the detectors Db, Dc and Dd. The store 5 may comprise charge-coupled devices as described for example by Hobson in a review starting on page 925 of Proc.l.E.E., 124 No.1 lR, Nov.1977.

Although the processor has been described with reference to signal expansion, it is to be appreciated that it could be used for signal compression by suitable variation of the relative velocity of movement of the impulse 33 and signal 32.

Time domain correlator Figure 5 shows an optical arrangement for correlating, in the time domain, an incoming signal with a fixed reference pattern. For the purposes of illustration, it is assumed that the pattern is stored on a cathodo-chromic target 5T, which can be opacity-modulated by a current-modulated electron beam, EB, from an electron gun, G. Before passing the focus coil, FC, the beam makes its way through a small hole in an inclined plane mirror, M, later being deflected as required via deflexion coils DC. G, M and 5T are all mounted within an evacuated envelope, E, and form the cathodo-chromic tube mentioned herein. A drive circuit 50 is provided to receive the incoming signal read sequentially from the store 5 of Figures 1 or 2.

Turning now to the optical layout, a near-monochromatic light source 5LS, illuminates a pinhole aperture in stop 5S1, by way of a condenser lens pair 5L1,2. This aperture behaves as a secondary uniform source for lens 5LS which converges the light onto a small area of M whence it diverges through 5T then to be concentrated by lens 5L4 onto lens 5L5. The light leaving 5L5 falls upon collimating lens 5L6 : the powers and relative distances of 5L4, 5L5 and 5L6 are so chosen that an image of 5T is formed in the exit plane of 5L6.

Collimated light leaving 5L6 passes normally through an uitrasonic column, 5C, and is then focused by the 'transforming' lens 5L7, onto the plane of a stop 5S2. An ultrasonic transducer, 5X, receives the l.F. signal to be correlated with a reference pattern and launches into 5C bulk compression waves whose local wavefronts contain the direction of collimation; the instantaneous pressure (and hence refractive index) at any transverse plane of the column forms a delayed linear replica of the input i.f. signal. In the absence of an electrical input, and for a uniform target 5T, 5L7 forms a real spot image of 5S1 upon the opaque part of 5S2, as shown in the figure.

Now, the effect of diffraction introduced at 5T and/or 5C is to spread the point image along the line where the plane of 5S2 intersects that of the diagram. It can be shown (for example, see Felstead ibid) that 5S2 is in a 'frequency transform plane'. In other words, the mean power of the light falling on 5S2 is distributed according to the spatial frequencies obtaining in the plane of 5C, so providing a spatial power spectrum. A slit aperture, 5A, in 532 is of such dimensions as to allow components with frequencies within the observation bandwidth to pass through it onto a square-law photo detector 5D.Provided that its sensitivity is substantially uniform over the area illuminated, 5D supplies an output which gives a measure of incident light power.

As a wavetrain passes up 5C, the part of the output of 5D selected by a band-pass filter, 5F, gives a continuous succession of instantaneous cross-correlation coefficients between it and the fixed spatial pattern on 5T; that is, their cross-correlation function expressed with real time as the variable. Generally, it alternates at the mean frequency of the wavetrain, while its envelope gives a measure of its agreement with the phase structure of the reference.

The use of power rather than amplitude permits the use of a non-coherent light source and an aperture in 5S1 of appreciable area.

It has been tacitly assumed in the above description of Figure 5 that just as the planar wave front propagated from 5X extends to some distance in the direction, N, normal to the plane of the diagram, so too does the image of the pattern in 5T, the local opacity being uniform along the direction normal to the plane of the Figure. However, it is quite practical for 5T to carry a number of differing 'line' patterns, all parallel to the propagation axis of 5C, and disposed side by side down the normal direction. In this way, a number of correlations could be performed simultaneously, provided that a sensor 5D could be assigned to each.

This can be arranged by disposing a cylindricai lens 5L8 in the position shown dotted, with its surface generators parallel to the axis of propagation in 5C. (See, for example, Stroke, G.W.: 'Introduction of Coherent Optics. ..etc...', Academic Press, 1966, Chap V Figs 19 and 20). It is so proportioned as to form a one-dimensional image of 5C in the normal direction on 5S2. A set of slits 5A1, 5A2 (not shown) would be opened in 5S2, each in effective alignment with a different line on 5T. Associated with each slit would be respective sensors, 5D1, 5D2 (not shown) and filters 5F1, 5F2 (not shown) and subsequent signal detection circuits comprising a full-wave rectifier and a comparator. A more flexible mechanical layout results if apertures and detectors 5A1, 5D1; 5A2; 5D2; could be linked by means of a light guide each : this would allow the intervening path to be curved where necessary, while confining the cross-section.

Generally (see Felstead, ibid), each line pattern would take the form of a modulated spatial carrier sharing substantially the same bandwidth as is served by 5C. However, it would be advisable to make one line to be of uniform transmission so that the mean power in the incoming signal could be sensed. With this known, it becomes possible to normalise the responses, and hence the decision behaviour of the signal detectors.

Claims (11)

1. A signal processor wherein means are arranged to produce spatial distributions of light which are functions of an impulse signal and a signal to be processed, the distributions moving relative to one another at a predetermined relative velocity, and means are arranged to respond to said distributions in combination to produce an electrical signal representing the correlation of the impulse signal with the signal to be processed, the arrangement being such that the said electrical signal represents the signal to be processed scaled in time as a function of the said relative velocity.

2. A processor according to Claim 1, wherein the means to produce spatial distributions of light include ultrasonic light modulators arranged to receive the impulse signal and the signal to be processed respectively.

3. A processor according to Claim 2, wherein the modulators are arranged so that acoustic representations of the said signals propagate therethrough at the same rate, the processor further including lens means arranged to project an image of the spatial distribution of light derived from one of the modulators onto the other modulator, the magnification factor of the lens means defining the said relative velocity.

4. A signal processor substantially as hereinbefore described with reference to Figures 2, 3A to G and 4 of the accompanying drawings.

5. A correlation arrangement including: (A) a signal processor according any one of claims 1 to 4 arranged such that the said electrical signal represents the said signal to be processed expanded in time; (B) a store for storing a first representation of the said electrical signal; (C) a correlator comprising optical storage means on which an optical representation of the said electrical signal can be formed, means for forming the optical representation, means for producing a first spatial distribution of light which is a function of an optical representation of a signal formed on the optical storage means, means for forming a second spatial distribution of light which is a combination of the said first distribution and a further distribution of light, which is a function of the said signal to be processed, the first and further distributions moving relatively to another, and means responsive to the second distribution to produce a correlation signal representing the correlation of the signal to be processed and the signal represented on the optical storage means; and (D) means for causing the forming means to form on the optical storage means an optical representation of the signal stored in the said store if the signal to be processed does not have a predetermined correlation with any signal represented on the optical storage means.

6. A correlation arrangement according to Claim 5, wherein the storage.means and forming means comprise cathodo-chromic material and means for scanning the material with an electron beam to form on the material the said optical representation.

7. A correlation arrangement according to Claim 5, wherein the storage means and forming means comprises photochromic material or a clear carrier carrying an opaque film, and means for scanning a laser beam across the material or the carrier to form the said optical representation on the storage means.

8. A correlation arrangement according to Claim 5,6 or 7, wherein the means for forming the second distribution includes an ultrasonic light modulator arranged to receive the signal to be processed, and means for projecting onto the modulator an optical image of the said optical representation formed on the storage means.

9. A correlation arrangement, substantially as hereinbefore described with reference to Figures 2, 3A to G, 4 and 5 of the accompanying drawings.

10. Asignal analysis system, comprising A) means for receiving a signal, B) a correlation arrangement as claimed in any one of claims 5 to 9 arranged to receive that signal as the said signal to be processed; and C) means for analysing the characteristics of a signal which correlates with any signal represented on the optical storage means.

11. A signal analysis system substantially as herein before described with reference to Figures 1, 2, 3A to G, 4 and 5 of the accompanying drawings.