FR2683913A1 - Method and device for processing and detecting with a high visibility factor, used for the spectral analysis of a sampled signal, and applications, particularly to Doppler radars - Google Patents

Abstract

The invention relates to a method and a device for processing and detecting with a high visibility factor, used for the spectral analysis of a sampled signal, and its applications, particularly to Doppler radars. According to the invention, from samples (Xn) of a burst of pulses in a remote gate, the signal is autoregressively modelled (1) in order to obtain the coefficients (a1 to a4) of the predictive filter, followed by a search for the poles (2) in order to determine the frequencies of the spectral components of the signal. The amplitude (A1 to A4) of these components is determined by adaptive filtering (4.1 to 4.4) in channels separately processing each component, after rejection of the other components by a blanking (whitening, bleaching) filter, and by carrying out the correction of the gain necessary to obtain a unit gain in the channels. The detection (5.1 ... 5.i) is achieved by using a criterion of contrast over the maximum in the forward and backward distance/frequency zones. Application particularly to maritime surveillance radars and to FFC systems.

Description

METHOD AND DEVICE FOR PROCESSING AND DETECTION
HIGH VISIBILITY FACTOR FOR SPECTRAL ANALYSIS OF A SIGNAL
SAMPLE AND APPLICATIONS, INCLUDING RADAR DOPPLER
The present invention relates to a high visibility factor processing and detection method for the spectral analysis of a sampled signal, to its applications to a frequency spectral analysis method for a radar
Doppler emitting in particular short bursts of coherent pulses and a spatial spectral analysis method for a beamforming system by calculation, and corresponding devices.

 It is known, in the field of Doppler radars, particularly for the observation of the battlefield, to emit short bursts of pulses so as to limit the risk of radar detection by the enemy. The use of a short burst amounts to creating a short time window of observation of a signal whose theoretical frequency spectrum extends to infinity.

 FIG. 1 represents such a spectrum Si, with for example a high peak of clutter F and a low power target C, as well as the spectrum S2 of the temporal observation window which, as it is known, comprises a lobe principal of width 2 / T if T is the duration of the window and secondary lobes of decreasing amplitude.

 The resulting spectrum S obtained by the radar receiver is constituted by the convolution S1 * S2. The sidelobes of a large echo can completely mask a weak target close to the loud echo.

 It is known to eliminate clutter echoes in a radar signal by various methods, in particular based on transverse filtering.

 The observation window weighting due to the aforementioned convolution on the one hand and the strong a priori weighting applied to the radar signal to reduce the amplitude of the sidelobes so as to better ensure the visibility of low SER targets. (Equivalent radar surface) on the other hand have the effect of widening the bandwidth of the filters resulting in poor visibility of the targets spectrally very close to the clutter.

 To increase the visibility factor in such cases, we have developed different methods based this time on the deconvolution of clutter echoes using autoregressive modeling. These methods produce very good results provided that the clutter has a minimum of stationarity in distance and / or in time. However, in certain circumstances and in particular in the case of low altitude maritime surveillance, this assumption of stationarity is not fulfilled. Indeed, it is known that the power of the sea echoes which is substantially proportional to the power -3 of the distance for the near areas passes rather abruptly to a proportionality to the power -7 of the distance when the distance increases. This passage takes place in a transition zone located at a distance from the radar essentially variable with the state of the sea. In addition, in the transition zone, the sea clutter presents variable behaviors with very strong peaks whose position in time and distance is random. Under these conditions, even the processes mentioned last lose much of their effectiveness.

 The subject of the invention is therefore a method of processing and detection which overcomes these drawbacks by avoiding the use of a so-called "noise-only" reference for the processing of echoes, that is to say calculated from the echoes clutter from distance resolution cells close to the one being processed, as is the case in known methods.

 It should be noted that up to now, only the case of the Doppler frequency domain and the visibility of a low SER target has been considered spectrally very close to the clutter, but there is a complete analogy with the problem. to detect in the spatial domain a target angularly very close to a jammer in the case of a system with beamforming by calculation (FFC). In both cases, the problem of signal processing is the same: detecting and locating a weak signal spectrally close to a very strong signal. For the FFC processing, the angular information is translated into a regular network and linear of sensors by a "spatial frequency" in all respects identical to the Doppler frequency; the sampled signal is only composed of a series of samples in the spatial domain at a given instant instead of a sequence of samples in the time domain.

 As will be seen later, the method according to the invention also applies to these two areas. Thus, the terms "spectral" or "spectral analysis" will be taken throughout the description that follows and the claims in a broad sense covering both the temporal and spatial domain, in the absence of other details.

 According to the invention, a high visibility factor processing and detection method is therefore provided for the spectral analysis of a sampled signal, characterized in that it comprises the steps of: - estimating by autoregressive signal analysis sampling the spectral positions of the components of said signal - separately estimating the power of each of these components by a matched filtering in real time; and - performing a detection on the components thus estimated to select those above a predetermined threshold.

 The method according to the invention takes advantage of the fact that the autoregressive estimation of the frequencies of the lines of the spectrum is very precise but that, on the other hand, the determination of the powers of these components being generally distorted, another method is then used consisting of to create p channels each dealing with a single component whose spectral position has been previously determined.

 According to another aspect of the invention, there is provided a high visibility factor processing and detection method for spectral frequency analysis of a time-sampled radar signal, said radar emitting short bursts of pulses. coherent, said method as described above being characterized in that said step of performing a detection on the estimated components consists, for each sequence of samples of a burst in a door at a given distance, to: - determine a criterion Contrast on maximum by calculating a threshold starting from the maximum of the components obtained in distance / frequency zones bordering in distance the treated cell containing the examined component; and comparing the power of said examined component with said threshold to determine whether said component corresponds to a target.

 According to yet another aspect of the invention, there is provided a high visibility factor processing and detection method for the spatial spectral analysis of a signal from at least one source and sampled in space by minus a linear array of N regularly spaced sensors providing at each given instant a set of N samples of the signal received by these sensors, said method being as described above first and being characterized in that it further consists, after elimination of the clutter by filtering, a: calculating the successive reflection coefficients of an autoregressive, recursive modeling with respect to the order of said sampled signal constituted by said N samples at the instant considered; and comparing the amplitude of said coefficients with a given threshold to determine the number of sources likely to be present and the corresponding order of said autoregressive modeling of the step of estimating the spectral positions, the latter using said p coefficients of reflection retained for the calculation of said coefficients ak.

 The invention also relates to devices for implementing the various aspects of the method according to the invention.

 The invention will be better understood and other features and advantages will become apparent from the description below and the accompanying drawings in which - FIG. 1, already described, illustrates spectra that explain one of the drawbacks to which the invention seeks to remedy - Figure 2, already described, shows a clutter diagram illustrating another problem solved by the invention - Figure 3 is a characteristic filtering diagram of the method according to the invention - Figure 4 is a diagram of principle of a device according to the invention applied to the frequency domain - FIG. 5 is a diagram of an element of the device of FIG. 4 - FIG. 6 is a diagram of a rejection filter used in the device according to the invention FIGS. 7 and 8 are diagrams illustrating the detection in the device according to the invention; FIG. 9 is a diagram illustrating the performances of the method according to the invention; FIG. of principle of a known system beamforming by calculation - the figure is a diagram of such a system implementing a variant of the method according to the invention - Figure 12 is a diagram of an embodiment known from a part of the system of the figure it; and FIG. 13 is a diagram of an alternative application of the method according to the invention to an FFC system.

 As explained above, some of the problems encountered in radar signal processing and clutter elimination are the widening of the filter bandwidth and the difficulty in eliminating clutter.

 The autoregressive methods for estimating the Doppler components of a radar signal offer a great advantage: they provide a very precise estimate of the frequency position of the spectral components in all circumstances. On the other hand, in the presence of more than one frequency mode and / or non-stationary clutter, the estimated power values are erroneous. The method according to the invention is therefore based on the idea, after having determined the position of the Doppler frequencies of the components, to use another method to determine the power of each of the components. This method consists of filtering each component to estimate its power.

 The method according to the invention furthermore consists, at least in the time domain, in performing a detection adapted by contrast criterion based on the maximum reached by the components in the areas surrounding the cell in terms of distance and frequency tested with respect to the spectral band likely to be occupied by the clutter, and based on a noise reference calculated in the same areas with respect to the other spectral bands.

 In a first part of the description, reference will be made more particularly to the problem in the time domain to then extend the solutions exposed to the spatial domain.

 The method according to the invention therefore consists first of all in making an estimate by autoregressive spectral analysis of the p Doppler components of the signal, generated by the clutter and any targets.

 In a second step, the power of the Doppler components thus determined is evaluated by constructing p gain filters G j (with j varying between l and p) such that, as illustrated in FIG. G.

J is equal to l for the frequency fj for the rank filter j and to zero for each of the frequencies of the other components fi for all the values of i between l and p but different from j
The shape of the curve in Figure 3 is unimportant outside the frequencies of the estimated components specified above (with as an example p = 4) assuming pure frequency components. This way of proceeding makes it possible to avoid the widening created by the a priori weightings and by the window of observation.

FIG. 4 is a block diagram of a device according to the invention as part of the application to a radar
Doppler emitting short bursts of coherent pulses. The received signal is constituted for each door in distance from the radar by a sequence of N complex samples X corresponding to the N
n pulses of a burst.

 These samples are sent firstly to means of autoregressive modeling of order p. This order p corresponds to the maximum number of components that can be determined taking into account the operating conditions of the radar (number of pulses per burst, maximum number of spectral components likely to be present in a door in distance ... ).

For example and without being in any way limiting, with N = 9 pulses, we can choose p = 4.

The autoregressive modeling is carried out for example from the Marple algorithm which is described in more detail in the article "A new AR spectrum analysis algorithm" by
L. Marple, published in IEEE ASSP vol. 28, No. 4, 1980.

donc la transformée en Z du filtre prédictif dont il s'agit de trouver les racines, déterminant la position spectrale des composantes

This non-recursive modeling does not provide the poles but only the predictive model

therefore the Z-transform of the predictive filter whose roots it is to find, determining the spectral position of the components

A pole search circuit 2 calculates these roots Z1 to 24 of the polynomial P (Z) for example by the method of Bairstow described in detail in the book "Methods of numerical computation" of
JP. Nougier, 2nd edition, published by Masson.

The values Z1 to 24 are sent to a device 3 for filter synthesis and gain correction. The samples
X are on the other hand treated in p adapted filtering channels 4.l
n to 4.4 to estimate the power of the different components
Doppler. Each of these channels is intended to isolate a particular component to estimate its power independently of the other components by providing for each channel a gain equal to 1 for the component considered, the frequency of which has been defined by the means 1 to 3, and a gain equal to zero for all other components. The amplitude and frequency information of the processed components are then transmitted to a detection circuit 5.i for the distance gate of rank i.For each gate in distance, the same processing is performed and this has been schematized by representing a plurality of similar detection circuits 5.1, ... 5.i, ... respectively assigned to each door in distance l, ... i, ..., the assembly constituting the detection means 5.

Figure 5 shows the diagram of a suitable filtering channel. Such a channel, here for the frequency f1, comprises in series a "whitener" filter eliminating all the other components at the frequencies f2, f3, f4, a filter adapted to the frequency f1 and a gain correction. The whitening filter comprises in series (here 3) rejector filters 41 to 43 respectively rejecting the frequencies f2, f3 and f4. FIG. 6 represents an example of a response curve Fbl of such a rejection filter having a zero gain for fl. This curve is deduced by translation of a curve Fb. The shape of this curve Fb
oo can be any a priori outside the frequency O. We see that the gain of the rejection filter (Fbl) at the frequency of another component f2 is equal to the gain g21 measured on the initial curve Fb at the frequency f2 -fl . One can likewise obtain the
o gains at frequencies f3 and f4. The same remarks apply to the rejecting filters of frequencies f2, f3 or f4. We thus see that it suffices to keep in memory the points of the curve Fb
o to deduce the respective gains of the rejecting filters at the various frequencies.

sur les N échantillons de la rafale, moins trois en raison des filtrages réjecteurs antérieurs. Le circuit 45 extrait le module au carré de la quantité Z obtenue. Ce module au carré est alors soumis à une correction de gain de valeur Gl par le circuit 46. Cette correction de gain est obtenue à partir des différents gains des filtres de la voie.
Y samples of the signal cleared
n components at frequencies f2 to f4 are sent to the matched filter 44 which receives the value of the frequency f1 associated with the channel considered and performs the operation

on the N samples of the burst, minus three because of the previous rejecting filtering. The circuit 45 extracts the module squared from the quantity Z obtained. This squared module is then subjected to a gain gain correction G1 by the circuit 46. This gain correction is obtained from the different gains of the filters of the channel.

 The circuit 46 thus provides the exact value of the power | A1 | Ai 12 of the component considered-since the gain of the adapted filtering channel has been reduced to l by this circuit 46.

 The values of the roots Z2, Z3, Z4, of the frequency f1 and of the gain correction G1 are provided by the device 3 (FIG 4) for the synthesis of the filters, which calculates G 1, in particular from the gains of the rejector filters obtained, such as It has been explained, for example, by addressing a read-only memory in which the gains of the filter lZ 1 are stored successively by the values f1-f2, f1-f3 and f1-f4 for the path to f1. The device 3 also provides the values Za 'Zb and Z as well as f adequate for each channel.

c
Such a treatment has the main effect of suppressing the window and weighting effect.

 The detection means 5 (FIG. 4) which are used have the particularity of being based, at least in the Doppler frequency band likely to be occupied by clutter, on a criterion of contrast both in distance and frequency.

For this, as indicated in FIG. 7, two zones are considered which spatially close the treated cell Ci for the distance gate of rank i, namely a so-called advanced zone ZA and a so-called backward zone ZR. The processed cell C i is defined by a distance gate i and a frequency band AF centered on the frequency fj of the processed component.

The forward zone ZA extends to the next P / 2 gates from gate i + 1 and to the frequency band AF around f. Likewise, the retracted zone ZR extends over P / 2 gates in distance preceding the gate i, from the gate in distance il, and on the same frequency band AF around fj. As it is symbolized in FIG. 8, if the average level of the noise power is Sb which defines a frequency / distance plane in the diagram, we will retain in the zones ZA and ZR the maximum reached by the components included in the diagram. these two areas. Here this maximum MAX is symbolized by the segment HE.On deduces a detection threshold F (MAX) according to the value MAX which is used to calculate the contrast between the cell being processed C. and the maximum of the two zones ZA and ZR. If the power of the component in cell C is symbolized by the segment AB less than AD, there is no detection of a target; if this power is symbolized by a segment AC greater than AD, there is detection of a target. This target presence information and frequency information
Doppler and distance are provided at the output of the corresponding 5.i circuit. The determination of the threshold is made from the power and frequency information provided by the adapted filters corresponding to the distance / frequency cells of the zones ZA and ZR.

 It is clear that this statistic performed on the maximum is penalizing in the case of Gaussian white noise, its purpose being to ensure the probability of false alarm required on nonstationary clutter essentially in distance. Also, to increase the visibility outside the Doppler frequency band likely to be occupied by clutter, [-o, x to + o, x in reduced frequency], we replace the previous threshold F (MAX) with a calculated threshold in a classical way from the average of the powers determined on the zones advanced and retreated.

The implementation of the method according to the invention thus described can be carried out very simply using a programmable signal processor of the 320 C type from Texas Instruments to perform the autoregressive modeling and the search for pale a PROM memory (or a set of such memories) providing the gain correction of p filter stages made from programmable circuits as previously or from specialized circuits of the ASIC type or IMS A100 of the society
INMOS, the choice being essentially dictated by the necessary calculation rate.

 The implementation can be remote door gate distance, independently, to the detection stage which takes into account simultaneously several gates in distance.

For the latter, the processing rate is that of bursts of samples (and no longer that of samples) and the detection function can be performed by a programmable circuit.

 FIG. 9 is an example of curves showing the visibility factor on real sea clutter echoes with probability of detection of 0.3 and probability of false alarm of 10 6 and bursts of 9 pulses. The curve FVr corresponds to a conventional treatment by bank of filters, the curve FVt corresponds to the theoretical value for a treatment by bank of filters on a clutter supposed homogeneous in distance (in practice the probability of false alarm of l0 6 is no longer reached because peaks in the reduced frequency band [0,1; OJ cause false alarms) and the FVi curve corresponds to the treatment according to the invention. It can thus be seen that, for low SER targets corresponding for example to a signal-to-noise ratio of the order of 10 dB, the method according to the invention makes it possible to increase the band "useful", that is to say outside the bandedou clutter, about 40%.

But, as already indicated in the introduction, the invention is not limited to the time domain. In the space domain, there is known systems, including radar, beamforming by calculation (FFC) where a receiving antenna is formed of a linear array of N sensors regularly spaced A / 2, where x is the length of d wave of the transmitted frequency. The formation of a beam in a given angular direction (defined with respect to the normal to the network) consists of re-phasing all the samples received by the various sensors and emanating from this direction. This operation is completely similar to the time domain phasing performed during a Fourier transform for a given Doppler frequency. In the case of FFC radar, the visibility of a low SAR target angularly very close of a strong jammer poses the same problems as the visibility of a low SER target at the same distance and frequency
Doppler near a target or clutter of high power.

In order to be able to detect the weak target, "high resolution" algorithms such as the Capon algorithm described for example in the article "High-resolution frequency-wavenumber spectrum analysis" have been developed by
J. Capon in the review Proc. IEEE, vol 57, No. 8, 08/1969 - the MUSIC algorithm described for example in the article "multiple emitter location and signal parameter estimation" by
RO Schmidt in the journal IEEE Trans. on Antennas and
Propagation, vol. AP 34, 03/1986; or - the ESPRIT algorithm described for example in the article "ESPRIT-Estimation of Signal Parameters via Rotational Invariance"
Techniques "by R. Roy and T. Kailath in the journal IEEE ASSP, vol 37, No. 7, 07/1989.

All these algorithms allow, in the presence of N sensors, to theoretically detect Nl frequency targets
Doppler different.

 However, all require to perform the possible detection of a target (or more) per distance resolution cell to implement the algorithm globally, that is to say with all its potentiality to separate Nl targets, even in the absence of a target.

 To understand the extremely high calculation load that this implies, we can take the example of a ground surveillance radar having for example 128 Doppler filters and 400 gates in distance. The implementation of a FFC per filter and per gate distance leads to 51200 implementations of the FFC algorithm. However, even in ground surveillance where the target density is very high, it leads to a maximum of 400 targets (taking 90 "opening in the field and 20 km range). 8% of the implementations of the algorithm Figure 10 schematizes this situation of the classical systems: N C1 to CN sensors whose received signals processed by a conventional algorithm are used to form Nl beams. Thus allows to detect between 0 and Nl targets depending on the case, the algorithm being used anyway to form all the Nl beams.

 It is clear that, in place of conventional algorithms, it would be possible to implement the method of the invention already described in the time domain. However, according to another characteristic of the invention, it is proposed to carry out a sequential implementation in the form of N-1 stages, each stage making it possible to locate in the field and to detect an additional target.

 Figure 11 symbolizes this implementation. The signal is, this time, constituted by the N complex samples (phase and quadrature) obtained at the output of N C1 to CN sensors at a given moment, after passing through the receivers 60.1 to 60.N and the clutter filters 61.1 to 61 .N associated.

At each stage And at EtN 2 of beam formation
o FFC5, processing circuits 70.0 to 70.N-2 respectively calculate the successive reflection coefficients of an autoregressive modeling of the signal, recursive with respect to the order. For example, it is possible to use the algorithm of
Burg (with Levinson recursion) described in J. Burg's Maximum Entropy Spectral Analysis, PhD-Stanford University, 1975.

At each successive stage, corresponding to an increasing order of the modeling, the amplitude of the modulus of the calculated reflection coefficient is tested. If this amplitude is greater than a predetermined threshold, it is that there is an additional target; the number M of targets detected in the corresponding processing circuits is then incremented by 1. In each stage, the number M obtained is tested (71.0 to 71.N-2). If M is greater than or equal to the order of the reflection coefficient of the stage considered, the next stage is passed by transmitting to it the direct and delayed channels, the value of M and the control signal coming from the test 71.1. If M is smaller than the order j of the reflection coefficient of the stage, calculation circuits (73.1 to 73.N-1) associated with this stage calculate the angular directions ("spatial frequency") Fi and the respective power
Pi of the targets (1 <i <j-1) using a device 73.j-1 similar to that of Figure 4. The only differences reside in the fact that the means 1 autoregressive modeling, d order jl, can be very simplified by sending them the values of all the reflection coefficients already calculated by the Eto stages Etj-1 (Fig.11) which allows to deduce directly coefficients a1 to a; 1 necessary for the calculation of the poles by the search circuit 2 as will be seen below - in that it is not essential to provide detection means 5 since the target detection has already been made on the amplitude of the reflection coefficients; however, if a more precise detection than that carried out on the reflection coefficients is desired, detecting means 5 can be maintained.

It may be noted that, in FIG. 11, the initial stage Et and the last stage EtN, 1 have a schematic diagram
o different from the other floors.

 In the Eto stage, if the reflection coefficient, of order 1, tested is lower than the threshold, it means that there is no target (M = 0) and the processing stops there (block 72) , this information being delivered by block 72.

 In the stage Et N-i 'it is useless to provide processing circuits for calculating a reflection coefficient of order N since, in principle, only N-1 beams can be formed. In this stage, it is therefore simply expected to perform the calculation of the N-1 components (calculation circuits 73.N-1) and is also provided alert means 74 to signal the saturation of the system.

 From the description above, one of the essential advantages of this application can be deduced from the FFC: it is that the calculations performed are only the calculations that are useful for the effective detection of the targets present. Thus in the distance gates without targets, only the reflection coefficient of order 1 is calculated and its module tested.

 In a remote gate where there are targets - the order of the reflection coefficient K and its modulus are calculated at each order up to p + 1 - the signal filtered by a trellis filter is calculated at each order up to p the Levinson recursion calculation is carried out in order to pass reflection coefficients to the prediction coefficients a1 to a; the poles are calculated by the Bairstow method; the appropriate filters are calculated; and - the corrected modules are extracted with respect to the gain.

 It can be shown that the design load for the method according to the invention is several hundred times lower than that which would be obtained using a conventional algorithm such as that of Capon in the case of surveillance radar.

 Another very great advantage of the method according to the invention is that it can be validly implemented with a very small number of echoes.

 FIG. 12 represents the known diagram of a possible embodiment of a part of the processing circuits 70.0 to 70.N-2 of FIG. 11 for the calculation of the reflection coefficients as well as the conserved part of the modeling means 1 autoregressive of Figure 4 used for the FFC.

à partir des échantillons de la voie directe et de la voie retardée.
X samples coming from N sensors at a given moment are
n rearranged in a time sequence of these N samples and sent on a first stage 101 trellis filter, the samples X being applied to both inputs of
n the stage, say "direct input" to the memory 111 and "delayed entry" to the memory 112 through a register 110 ensuring the necessary delay equal to a sampling period. The memories 111 and 112 store the signals in memory the time required for the calculation of the reflection coefficient K1 by the circuit 131. The reflection coefficient K1 is calculated according to a relationship such that

from the samples of the direct and the delayed channels.

The basic structure of the trellis filter further comprises two complex multipliers 113 and 114 and two subtractors 115 and 116, the multiplier 114 receiving the reflection coefficient K1 and the multiplier 113 receiving the conjugate value K1 *. In general, the n-order stage of a trellis filter receiving on a direct channel a signal and (ni) and on a delayed channel a signal ft (n-1), provides on its two outputs
etc. ") = and (ni) - K ft (nl) z
ft (n) ftn-ltZ - Kn * and (ni) where z is a delay operator and K the reflection coefficient of
n the order n.The stages 101, 102 ... 10p up to the maximum order p considered are connected in series and at each stage is associated a calculation circuit 131, 132 ... 13p for calculating the coefficient reflection of the corresponding order. It is these reflection coefficients that are used in the various stages Et to EtN 1 of FIG. 11.

o
Moreover, as explained above, the means 1 of autoregressive modeling of Figure 4 can be reduced to a calculation circuit of the predictive coefficients a1 to a from these reflection coefficients K1 to Kp already calculated p. This circuit is indicated by the reference 140 in FIG. 12. As for calculating a stage of the trellis filter, it is possible to compute the predictive filtering element of order n + 1, by using the coefficients a1 ... a year calculated previously, as well as the new coefficient of reflection Kn + 1 provided by the trellis filter.On uses for this a device 140 whose operating principle is given by the following recurrence formula corresponding to a filtering limited to the order n + 1

<tb> -i <SEP> 0 <SEP> -1
<tb> (n) <SEP> (n) * <SEP> a <SEP> (n + 1)
<tb> a1 <SEP> n <SEP> i <SEP>
<tb> (n) <SEP> (n + 1) <SEP>
<tb> a2 <SEP> a2 <SEP>
<tb>: <SEP> -Kn-1 <SEP>. <SEP> = <SEP> ~ <SEP> Kn-l <SEP>
<tb> year (n) <SEP> a1 (n) *
<tb> 0 <SEP> -1
<tb> where aj) constitutes the filtering coefficient of the filter of order n + 1 (1 # js n + 1), where Kn + 1 constitutes the (n + i) th th reflection coefficient calculated by the (n +1) cell of the trellis filter and where the coefficient aj (n) constitutes the jth filtering coefficient of the filter of order n (1 # j # n).

The circuit 140 comprises only a simple logic calculation cell performing the calculations of the type
an + i) = aj (n) - Kn + 1 a ##### as well as a loop between its output and its input allowing to use as many times as necessary the same material following the order of the filter that we want to achieve.

 For the device of Figure 11, it has been considered only the case of a remote door gate processing distance. But one can also consider a treatment refined by gate distance and Doppler frequency.

In this case, as shown in FIG. 13, after each clutter filter 61.1 to 61.N a treatment is added.
Doppler 62.1 to 62.N by fast Fourier transform (FFT) for example. All the samples of the same row of Doppler filtering (corresponding to a particular Doppler frequency) resulting from the N Doppler treatments 62.1 to 62.N are sent to a beam forming device by the respective calculation FFCS1 to FFCSQ, identical to the device FFC5 of Figure 11.

 This system could be further improved by replacing each Doppler filter with autoregressive detection which provides both the indication of the presence of a target and its Doppler frequency. In this case, it would be possible to effect the formation of beams only when a target is detected without having to recalculate the reflection coefficients.

 Of course, the described embodiments are in no way limitative of the invention.

Claims (17)

 A high visibility factor processing and detection method for the spectral analysis of a sampled signal, characterized in that it comprises the steps of: estimating, by autoregressive analysis of the sampled signal, the spectral positions of the components of said signal - estimate separately the power of each of these components by a filtering adapted in real time; and - performing a detection on the components thus estimated to select those above a predetermined threshold.  2. Method according to claim 1, characterized in that said step of estimating the spectral positions of the components consists of an autoregressive modeling of predetermined order p, from a sequence of N samples constituting said signal to provide the coefficients ak. (1 sksp) of the predictive model, and in a search for the poles of the corresponding predictive filter.  3. Method according to claim 2, characterized in that said autoregressive modeling uses the algorithm of Marple.  4. Method according to one of claims 2 or 3, characterized in that said search for the poles consists of calculating the roots of the polynomial resulting from the Z-transform of said predictive filter.  5. Method according to any one of claims 1 to 4, characterized in that said step of separately estimating the power of each component comprises, for each component, a: - eliminating from said sampled signal all the other components by centering refocusing filtering respectively on the values provided by said poles corresponding to these other components - performing a matched filtering of said sampled and filtered signal, performed on said component; and - performing a gain correction at the output of said filtering adapted to take account of the gains of the different filtering and restore the power of said component  A high visibility factor processing and detection method for spectral frequency analysis of a time sampled radar signal, said radar transmitting short bursts of coherent pulses, said method according to any of the claims. 1 to 5 being characterized in that said step of performing a detection on the estimated components consists, for each sequence of samples of a burst in a door at a given distance, in - determining a maximum contrast criterion by calculating a threshold from the maximum of the components obtained in distance / frequency zones enclosing in distance the treated cell containing the component examined; and comparing the power of said examined component with said threshold to determine whether said component corresponds to a target.  7. Method according to claim 6, characterized in that, outside the spectral band capable of containing clutter, said contrast criterion is replaced by a criterion consisting in calculating a threshold based on the average of the components obtained in said zones. distance / frequency enclosing the treated cell in distance.  A high visibility factor processing and detection method for spatial spectral analysis of a signal from at least one source and sampled in space, by at least one linear array of N regularly spaced sensors providing each instant given a set of N samples of the signal received by these sensors, said method according to any one of claims 1 to 5 being characterized in that it further consists, after filtering the clutter, a: - calculate the successive reflection coefficients of an autoregressive, recursive modeling with respect to the order of said sampled signal constituted by said N samples at the instant considered; and comparing the amplitude of said coefficients with a given threshold to determine the number of sources likely to be present and the corresponding order of said autoregressive modeling of the step of estimating the spectral positions, the latter using said p coefficients of reflection retained for the calculation of said coefficients ak.  9. The method of claim 8, characterized in that said detecting step is constituted by said step of comparing the amplitude of said reflection coefficients to a given threshold.  10. Method according to one of claims 8 or 9, characterized in that said recursive autoregressive modeling with respect to the order uses the Burg algorithm.  11. A high visibility factor processing and detection device for spectral frequency analysis of a radar signal sampled in time, said radar emitting short bursts of coherent pulses, said device implementing the method according to FIG. one of claims 6 or 7 and being characterized in that it comprises - means (1) of autoregressive modeling of order p, from each sequence of samples of a burst in a door at a given distance, for providing said coefficients ak of the predictive model - means (2) for finding the poles (Z1 to Z4) of the corresponding predictive filter - p adapted filtering channels (4.1 to 4.4) for separately estimating the power of each component whose spectral position has determined by said pale search means (2) - filter synthesis and gain correction means (3) for controlling each of said channels; and detection means (5.1, 5.i) for comparing the power of each estimated component in said channels with a threshold calculated from the components obtained in distance / frequency zones enclosing the treated cell in distance.  12. Device according to claim 11, characterized in that each of said adapted filtering channels comprises -pl rejector filters (41 to 43) in series to respectively reject each of the pi components other than that associated with said channel considered - a suitable filter ( 44) at the frequency of said component under consideration; and gain correction means (46) for compensating the gains of the different filters of the channel.  A high visibility factor processing and detection device for spatial spectral analysis of a signal from at least one source and spatially sampled by at least one linear array of N regularly spaced sensors providing each given instant a set of N samples of the signal received by said sensors, said device implementing the method according to any one of claims 8 to 10 and being characterized in that it comprises, after receiving and filtering the clutter for each sensor processing means (70.0 to 70.N-2) for calculating the successive reflection coefficients of an autoregressive, recursive modeling with respect to the order of said sampled signal constituted by said N samples at the instant considered and for compare the amplitude of these coefficients with a given threshold, so as to determine the number of sources likely to be present and the order p cor respondent of said autoregressive modeling of the step of estimating the spectral positions; and means (73.1 to 73.N-1) for determining the respective spatial direction of said sources with respect to said sensors and their respective power from said samples and said reflection coefficients provided by said processing means.  14. Device according to claim 13, characterized in that said determining means (73.1 to 73.N-1) comprise - means (1) of autoregressive modeling of order p, recursive with respect to the order, to provide said coefficients ak of the predictive model from said reflection coefficients - means (2) for finding the poles (Z1 to Z4) of the corresponding predictive filter - p adapted filtering channels (4.1 to 4.4) for separately estimating the power of each component whose spatial direction has been determined by said pole finding means (2); and - filter synthesis and gain correction means (3) for controlling each of said channels.  15. Device according to claim 14, characterized in that each of said matched filtering channels comprises - pl rejector filters (41 to 43) in series to respectively reject each of the p-l1 components other than that associated with said channel considered - a filter adapted (44) to the spatial direction of said considered component; and gain correction means (46) for compensating the gains of the different filters of the channel.  16. Device according to one of claims 14 or 15, characterized in that said means (1) for autoregressive modeling comprise a calculation circuit (140) of the coefficients a k from the reflection coefficients K1 to Kp according to the recurrence relation.

<tb> where a constitutes the jth coefficient of the filter of order n + 1 jth (with 1 - j <n + i) and where Kn + 1 constitutes the (n + 1) th reflection coefficient. <tb> 0 <SEP> -1 <SEP> year + 1 <tb> (n + 1) <tb> year (n) <SEP> a1 (n) * <SEP> <tb> = <SEP> Kn-1 <SEP> <tb> a2 (n) <SEP> a2 (n + i) <SEP> <tb> a1 <SEP> n <SEP> i <SEP> <tb> (n) <SEP> a <SEP> (n) * <SEP> a <SEP> (n + i) <SEP> <tb> -i <SEP> O <SEP> -i <SEP>  17. Device according to any one of claims 14 to 16, characterized in that said processing means (70.0 to 70.N-2) comprise means for calculating said reflection coefficients according to Burg's algorithm.