FR2747792A1 - Multichannel adaptive beam forming for ground surveillance radar - Google Patents

Abstract

In each channel the samples of echoes received after transmission of a single burst undergo three-stage processing which includes selection (11) of a number of samples corresponding to one and the same range, frequency filtering (12) and storage of the resulting frequency samples in a memory (13). Inter-spectral matrices (Ri) for subarrays of successive sensors are estimated (2) and averaged (3) to obtain an overall inter-spectral matrix (R), from which spatial filtering for a pointing vector in the desired beam direction is calculated (4) and applied (5) to the frequency samples.

Description

BEAM FORMING PROCESS
ADAPTIVE FOR A RADAR OF
SOIL SURVEILLANCE, AND RADAR
IMPLEMENTING THE PROCESS
The present invention relates to an adaptive beam forming method for a ground surveillance radar, and to a radar implementing the method.

 The particular field of ground monitoring is currently experiencing recent developments on the battlefield, characterized by increasing integration between the different systems used. The direct consequence is a diversification of the threat to be taken into account for each weapon system. In particular, the operational requirements of a ground surveillance radar must take into account the use of vehicles and helicopters. The radar will therefore have to detect and track these two types of targets simultaneously, as well as infantrymen and, possibly, drones. This constraint of simultaneously tracking fast-speed targets (helicopters, drones) and low-speed targets (infantrymen, vehicles) immediately leads to incompatibility for single-beam antennas, with mechanical or electronic scanning, conventionally used in ground surveillance . Indeed, the tracking of fast targets requires a high rate of renewal, typically of the order of a second, while the detection of slow targets imposes a long duration of the pointing of the antenna in each direction.

On the other hand, another important constraint comes from the multiplication of electronic eavesdropping means which leads to search for radars as discreet as possible. This discretion can be obtained, for example, by emitting only a very small part of the time,
The observation of the entire area to be monitored being carried out in a single burst of pulses. In addition, the multiplication and the increase in the performances of the jammers impose on the ground surveillance radar to have anti-jamming capacities and to simultaneously eliminate the signals coming from the jammers.

 All of these new constraints have led the Applicant to study a ground surveillance radar using a fixed antenna at emission, illuminating a large angular sector in bearing, and at reception, a fixed network of elementary antennas covering the same sector, the beam formation being provided by calculation, adaptively.

 The calculation method for beam formation however needs to be adapted to the specific problems encountered in ground monitoring: It is for example necessary to determine the speed of the targets, which requires carrying out a Doppler analysis.

An object of the invention is to associate with the adaptive beamforming treatment of the antenna a Doppler filtering, in order to estimate the radial speed of the targets in addition to their angular location. Advantageously, the invention provides for carrying out this filtering
Doppler before the FFC, in order to eliminate the ground echoes as well as possible, before adapting the antenna diagram.

 In addition, an essential characteristic of the invention consists in carrying out the adaptive processing in deposit per cell, a cell being defined either by a box-distance, or by a casedistance by frequency band. This characteristic makes it possible to form an antenna diagram suitable for each cell, which was impracticable with the types of processing carried out until now. In addition, the processing by cell avoids passing on the problems linked to the saturation of the dynamics in a cell (very strong fixed echo for example) in the other cells.

 Finally, the method according to the invention takes into account the presence, very likely in ground monitoring, of coherent targets, that is to say targets having the same speed in the same cell, which disturbs the training treatments. adaptive beams in the absence of an appropriate remedy.

More specifically, the invention relates to an adaptive beam forming method for a ground surveillance radar emitting pulses in bursts and comprising a reception network of N sensors followed by N reception channels each carrying out sampling and coding at the period T of the complex signals picked up, the method being characterized in that it consists in
- select, on each channel j, j being an integer varying from 1 to N, a number k of time samples Xj (iT) of said captured signals, i being an integer varying from 1 to k, said time samples corresponding to a single and even box-distance
- perform, on each channel j, a frequency filtering of the k time samples Xj (iT) in order to obtain a number k of frequency samples Yj (if);
- store said frequency samples Yj (if);
- estimate, for at most P + 1 subnetworks of (NP) successive sensors, The integer P may be zero, each subnetwork of index n deducing from the subnetwork of index (n-1) by a translation of at least one sensor, the interspectral matrices Rin relating to frequencies identified by the integer i, from the corresponding frequency samples Yj (if);
- deduce a global interspectral matrix R by averaging from the interspectral matrices Rin
- calculate, for at least one given pointing vector Soe, in a direction O where one wishes to form a beam, a spatial filtering according to the expression

where SOù is the transposed vector of said vector S and R-1, the inverse matrix of said matrix R;
- apply said spatial filtering to said frequency samples
Yj (if) corresponding respectively to said frequencies.

 The present invention also relates to a ground surveillance radar for implementing the method as described in claims 1 to 6.

The present invention, as well as its advantages, will be better understood in the light of the following description made with reference to the appended figures
- Figure 1 is a block diagram of the method according to the invention;
- Figure 2 is a possible embodiment of a reception network according to the invention for a ground surveillance radar.

 - Figure 3 is another embodiment of the reception network, using elementary processors.

FIG. 1 is a block diagram of the method according to the invention applied to a radar comprising a network of N sensors, followed by N channels identified by an index j. In this figure, it can be seen that the method generally consists in carrying out a first treatment 1 applying to each of the channels j, for j varying from 1 to
N, followed by a second treatment comprising the steps referenced 2 to 5, common to all the channels.

 Advantageously, the invention plans to work on the time samples received after the emission of a single burst of pulses.

 For each channel j, receiving complex signals conventionally sampled at a given period T, then coded, the first processing 1 consists first of all in selecting, in step 11, a number k of temporal samples Xj (iT) of the signals picked up corresponding to a single distance box, the index i varying from 1 to k. As it was said previously, the fact of working by casedistance makes it possible to form an optimal antenna diagram for each box-distance, and especially to avoid the repercussion of any problem (saturation) encountered in a box-distance on the others . The next step, referenced 12, is a frequency filtering of the k time samples Xj (iT) so as to obtain k frequency samples Yj (if).

A step 13 then stores these k frequency samples.

 At the end of these N first treatments 1, carried out respectively on each of the channels j, we are therefore able to recover k vectors Y (if) of dimension N whose N components are the frequency samples Yj (if), at a frequency identified by the integer i.

At the start of the second processing, each vector Y (if) will be used for a step 2 of estimation of an interspectral matrix
Dimension Ri (NxN) according to the formula
R i = Y (if) Y + (if) (1) where Y + is the vector transposed from the vector Y.

 The next step, referenced 3, consists in calculating a global interspectral matrix R, that is to say common to all the frequencies, by averaging the k matrices Ri estimated in step 2.

où
Soe est ie vecteur de pointage dans la direction O où l'on désire former un faisceau
S0 est le vecteur transposé du vecteur S+ e
R-1 est la matrice inverse de la matrice interspectrale globale R.The global interspectral matrix R is necessary for the chosen beamforming method, namely the so-called generalized maximum likelihood method, or CAPON method. This matrix R is in fact used in the next step, referenced 4, to calculate the spatial filtering according to the expression

or
Soe is the pointing vector in the direction O where we want to form a beam
S0 is the vector transposed from the vector S + e
R-1 is the inverse matrix of the global interspectral matrix R.

où d est la distance séparant deux capteurs et X, la longueur d'onde émise.In the most general case, the pointing vector is of dimension N, and one can write

where d is the distance between two sensors and X is the wavelength emitted.

 The method according to the invention provides for cutting the angular domain observed in several angular directions 0e indexed by the integer 1, so that there are several pointing vectors S, and as many spatial filters to be calculated according to expression (2 ).

 Due to the high potential density of the targets, the sampling step of the angular domain by the beams formed must be sufficiently fine, that is to say finer than the theoretical angular resolution x of the antenna, namely D rad . if D is the dimension of the antenna.

 It guarantees that each target is correctly angularly sampled and that discrimination between two angularly adjacent targets is possible.

 In step 5 of FIG. 1, each spatial filtering for a direction A1 is applied to the vectors Y (if), that is to say to the frequency samples Yj (if) for each frequency.

The method according to the invention therefore makes it possible to obtain a plurality of beams adapted to each distance box, and after exploitation, the position of the targets in the plane (angle, frequency
Doppler), i.e. in the plane (bye, if). Noise jammers, spread in frequency by definition, can then be easily detected and located angularly because they occupy the entire frequency domain observed in a direction A1. This detection can be done, for example, by post-processing at TFAC in frequency, TFAC being the French abbreviation for Rate of Constant False Alarm.

 Preferably, the method further provides for the elimination of the diffuse clutter of soil by not taking account of the annoying frequency samples Yj (if), for example for i = 1, 2 and k.

 The global interspectral matrix R is then estimated from the matrices Ri relating to the frequencies taken into account, to the number of (k-3) in the example taken; similarly, spatial filtering in a direction 81 then preferably applies to the vectors Y (if) for i corresponding to the frequencies taken into account.

 The method according to the invention advantageously provides for the possibility of performing low-pass filtering before frequency filtering 12, in order to limit the speed range observed. This additional precaution allows the ground surveillance radar not to make unnecessary calculations on signals which do not concern it, such as for example signals from jet planes for which the frequency is high.

 The method according to the invention, as it has just been described in its most general form for a better understanding, does not however solve the problem of coherent targets which, as we said before, have a high probability of presence. In the case of these coherent targets, the global interspectral matrix R obtained in step 3 of FIG. 1 is singular, and can therefore no longer be inverted. The calculation of a spatial filtering according to expression (2) is then impossible. This is why the method of the invention advantageously provides for the use of a technique known as spatial averaging which has the effect of decorrelating the coherent signals, and of rendering the matrix R its reversibility properties.

 This technique consists of cutting the network of N sensors into at most P + 1 subnetworks of index n of (NP) successive sensors, each subnetwork of index n deducing from the subnetwork of index (n- 1) by a translation of at least one sensor. For each subnetwork of index n, the method of FIG. 1 is applied up to step 2, so as to estimate the interspectral matrices Rin according to formula (1) in which the vectors Y (if) and Y + (if) then have the dimension of the subnetwork n, namely (NP). The global interspectral matrix R is then estimated by averaging from the Rin matrices, starting for example by carrying out a first average on all the sub-networks, that is to say on the index n, followed by a second average over all frequencies, i.e. over the index i. One can, in this way, treat at most P coherent targets, to the detriment however of the number of degrees of freedom of the radar which is reduced to the dimension of the sub-arrays.

Finally, the invention also provides another version of the method which performs not only the box-distance processing, but also the frequency band. This advantageously makes it possible to best analyze targets located in a given speed domain, without taking into account the targets of another speed domain which would however be located in the same distance box. To do this,
The architecture of the processing is identical to the method presented in FIG. 1 up to step 2 of estimation of the matrices Ri (or of the matrices
Rin in the case where the spatial average technique on subnets is used). A global interspectral matrix Rbm by frequency bands bm is then calculated by averaging on the index i the matrices Ri (or on the indices n and i the matrices Rin) corresponding to the frequencies belonging to the frequency band bm. The different spatial filtering along directions 8 are then calculated according to expression (2) in which R-1 is replaced by Rbm. There is therefore a spatial filtering by angle 8 targeted and by frequency band bm. Each filtering then applies to the samples Yj (if) corresponding to the frequency band considered.

FIG. 2 represents a possible embodiment of a reception network according to the invention for a ground surveillance radar. In the nonlimiting example taken, the network comprises 16 sensors 14 followed by 16 reception channels with current index j. A reception channel j comprises first of all a receiver 15 which performs, in a conventional manner, the demodulation and the sampling-coding at a period T of the complex signals picked up, coming from the targets in response to a burst of pulses emitted by the radar. The samples are then transmitted to selection means 16, for example a memory, which select the time samples Xj (iT) corresponding to the same distance box. Preferably, each channel j of the reception network comprises, at the output of memory 16, a low-pass filter 17 which limits the speed range observed. To give a numerical example, let's set the number of pulses of a burst to 512, the radar recurrence frequency being of the order of 7 KHz. The signals received are preferably processed by block of 256 samples, the results then being averaged over all the blocks. The low-pass filter 17 covers approximately half of the band in order to process the useful Doppler band, of 3.5 KHz in the digital example taken. At the end of the low-pass filter 17, there are 128 time samples Xj (iT), i varying from 1 to 128. The 128 time samples Xj (iT) are then filtered frequently by first means 18 preferably carrying out a transform of
Fast Fourier and providing 128 frequency samples Yj (if).

In accordance with the method of FIG. 1, the reception network comprises several modules 19 for estimating and storing the interspectral matrices Ri for each frequency indexed by the integer i, a module 20 for calculating the global matrix R by averaging the matrices Ri, a module 21 for calculating the transfer function of at least one spatial filter to be applied in an aiming direction 81 according to formula (2), and second means 22 for applying this spatial filter to the samples
Yj (iT). A module 23 then calculates the standard squared of the result from the means 22 in order to give the value of the power received as a function of the viewing angle And and of the frequency considered. Of course, the reception network can be adapted to carry out the processing of coherent targets by the technique of spatial averages, and / or the processing by frequency band.

 All the calculation steps being carried out on digital samples, it is advantageous to use elementary processors according to an architecture such as that represented in FIG. 3.

 The elementary processors can be constituted from signal processing processors of the ADSP 2100 type marketed by Analog Devices or TMS 320 from Texas Instrument.

 Recall that an elementary processor essentially consists of a program memory in which the microcode is stored, corresponding in our case to the steps of the method in FIG. 1, of a first data memory in which the input signal is stored , a calculation unit, and a data memory for storing the results of intermediate and final calculations.

 FIG. 3 represents a parallel architecture of elementary processors 25 each processing Q 'distance boxes, The integer Q' being determined as a function of the computing capacity of an elementary processor. If Q is the total number of distance boxes to be processed, then a number q of elementary processors 25 will be required equal to the integer part of (Q / Q '+ 1).

An input interface 24 makes it possible to distribute sequentially to the various elementary processors 25 the N-tuples of complex samples originating from the N sensors, after demodulation in phase and quadrature and digital coding of the latter. The input interface 24 consists for example of (q + 11 memories making it possible to store all the q packets of Q 'N-tuples in response to a pulse transmitted. The writing to these (q + 1) memories is carried out by cyclically so as to be able to simultaneously write to one of them and read to another. The data read is then transmitted to the first data memories of the elementary processors which can then carry out the beamforming calculations according to the method of the The results found are then transmitted to other elementary processors 26 which carry out the actual detection of the targets by any known technique, for example by processing with
TFAC in frequency. An output interface 27 then makes it possible to transmit the results of the preceding analyzes, for example to any display means.

Claims (9)

 1. Adaptive beamforming method for a ground surveillance radar emitting pulses in bursts and comprising a reception network of N sensors followed by N reception channels each performing sampling and coding at the period T of the complex signals picked up, the process being characterized in that it consists in  - select (11), on each channel j, j being an integer varying from 1 to N, a number k of time samples Xj (iT) of said captured signals, i being an integer varying from 1 to k, said time samples corresponding at one and the same distance box  - perform, on each channel j, a frequency filtering (12) of the k time samples Xj (iT) in order to obtain a number k of frequency samples Yj (if);  - memorize (13) said frequency samples Yj (if>  - estimate (2), for at most P + 1 subnetworks of (NP) successive sensors, The integer P may be zero, each subnetwork of index n deducing from the subnetwork of index (zero) by a translation of at least one sensor, the interspectral matrices Rin relating to frequencies identified by the integer i, from the corresponding frequency samples Yj (if);  - deduce (3) a global interspectral matrix R by means of swimming from the interspectral matrices Rin  - calculate (4), for at least one given pointing vector, in a direction A1 where it is desired to form a beam, a spatial filtering according to the expression

 - Apply (5) said spatial filtering to said frequency samples Yj (if) corresponding respectively to said frequencies.  where S + is the transposed vector of said vector S and R-1, the inverse matrix of said matrix R;  2. A method of forming adaptive beams according to claim 1, characterized in that said global interspectral matrix R is deduced from the interspectral matrices Rin by carrying out a first average on the index n of the sub-networks, then a second average on the index i of said frequencies.  3. A method of forming adaptive beams according to any one of claims 1 to 3, characterized in that a global interspectral matrix Rbm is deduced by frequency band t by averaging on the index i the matrices Ri or Rin corresponding at frequencies belonging to the frequency band l + z, and in that a spatial filtering by direction At and by frequency band s is calculated according to the expression

 4. A method of forming adaptive beams according to any one of claims 1 to 3, characterized in that, from the estimation step (2) of the interspectral matrices Rin, the soil clutter is eliminated by taking only the frequency samples Yj (if) for i varying from 3 to (kl).  5. A method of forming adaptive beams according to any one of the preceding claims, characterized in that, before the frequency filtering step (12), a low-pass filtering is carried out to limit the speed range observed.  6. A method of forming adaptive beams according to any one of the preceding claims, characterized in that, The integer P being zero, one estimates interspectral matrices Ri relating to frequencies indicated by the index i, and in that said interspectral matrix R is deduced by the average on the index i of the interspectral matrices Ri.  7. Ground surveillance radar for implementing the method according to claims 1 to 6, said radar transmitting pulses in bursts and comprising a reception network of N sensors followed by N reception channels each carrying out a sampling-coding at the period T of the complex signals received, the radar being characterized in that the reception network comprises:  means for selecting (16) a number k of temporal samples Xj (iT) of said captured signals, i being an integer varying from 1 to k, and j, an integer varying from 1 to N, said corresponding temporal samples at a single distance box  - first means (18) which effect a transform of Fourier of k time samples Xj (iT) in order to obtain a number k of frequency samples Yj (if);  - several modules (19) which estimate and store, for at most P + 1 subnetworks of (NP) successive sensors, The integer P can be zero, each subnetwork of index n deducing from the subnetwork of index (nl) by a translation of at least one sensor , the interspectral matrices Rin relating to frequencies identified by the integer i, from the corresponding frequency samples Yj (if)  - a module (20) which calculates a global interspectral matrix R by means of swimming from the interspectral matrices Rin;  - a module (21) which calculates, for at least one given pointing vector S, in a direction BL where it is desired to form a beam, the transfer function of a spatial filter according to the expression

 - a module (23) which calculates the norm squared of the result from the second means (22).  - second means (22) which apply said spatial filter to said frequency samples Yj (if) corresponding respectively to said frequencies  where SRt is the transposed vector of said vector Sd9! and R-1, the inverse matrix of said matrix R;  8. Ground surveillance radar according to claim 7, characterized in that the reception network further comprises, on each of the N reception channels, a low-pass filter (17) which limits the speed range observed.  9. Ground surveillance radar according to any one of the preceding claims, characterized in that said selection means (16) constitute an input interface (24), and in that said first and second means (18, 22 ) and said modules (19, 20, 21, 23) are produced in the form of elementary processors (25) to which the input interface (24) supplies the N-tuples it has stored.