METHOD FOR THE MULTIPATH PASSIVE RADAR PROCESSING OF AN FM OPPORTUNITY SIGNAL FIELD OF THE INVENTION The present invention relates to the field of passive radars, and notably the field of passive radars operating on non-cooperating opportunity transmissions, such as FM transmissions intended for the public. It deals more particularly with the processing of the FM signals reflected by objects situated in the zone covered by such transmissions for their use for detection and location purposes. CONTEXT OF THE INVENTION - PRIOR ART When one wishes to achieve the radar coverage, temporary or permanent, of a geographical zone, the immediate solution generally consists in implanting autonomous, active radar systems, mobile or otherwise, in such numbers that the union of the zones covered by each of the systems corresponds to the zone that it is desired to cover. Thus to cover an extended zone it is possible to choose to use a given number of radars of short or medium range or a more restricted number of long-range radars. The merging of the data provided by each system makes It possible to achieve the desired coverage. Being autonomous active systems, each of them comprises a transmitter and a receiver whose price and complexity are dependent on the performance required by the application considered and in particular on the power transmitted by the transmitter, which power conditions the range and therefore the size of the zone covered by each system. Furthermore the deployment of such a structure even over a vast zone generally poses problems of closeness between systems covering neighboring zones within one and the same global geographical zone. These closeness problems can in part be solved by using systems operating in accordance with different frequency plans. Nevertheless, the deployment of such a set of autonomous radar systems is both complex and expensive. A known solution for decreasing the complexity and therefore the cost of such a structure making it possible to cover a wide geographical zone, consists in using multistatic active systems comprising a single common transmitter placed at a given point of the zone and delivering sufficient power to cover this zone, and a set of receivers distant from one another and distant from the source. In such a structure each receiver generally knows the position of the common transmitter. Furthermore, in such a structure, the transmitter and the receivers are in general synchronized. In this way the coherent processing of the signals received implemented by each receiver is advantageously a processing akin to a conventional bi-static radar processing. Nevertheless such a structure which requires in particular the installation of a transmitter and means of synchronization between the transmitter and the various receivers remains complex to implement especially in the case of a mobile structure. Another solution for decreasing the complexity of such a structure, consists in using a simply passive structure comprising receivers able to receive opportunity signals in a given frequency band, which frequency band preferably corresponds to that of transmitter equipment whose transmissions totally or partially cover the geographical zone that it is desired to cover. These transmitters being intended for a use other than the formation of a monitoring structure their transmissions are also called opportunity transmissions or non-cooperating transmissions. Among these opportunity transmissions may for example be cited the frequency modulation transmissions (FM transmission) intended for the general public which are generally transmitted by a local transmitter covering a given geographical zone each transmission being carried out on a frequency band, or FM channel, from a hundred kilohertz to two hundred kilohertz. This solution based on installing a passive structure, with no transmitter, presents the great advantage of limiting the complexity and therefore the cost of this structure. Its tactical deployment amounts to the dissemination over the zone to be monitored of one or more radar receivers performing the reception and the processing of the opportunity signals received. On the other hand the practical implementation of this purely passive solution comes up against a certain number of difficulties which explain why they are still little utilized. A first difficulty which appears when one wishes to implement such a structure resides in the complexity of the signal received. Indeed the signal received by each receiver corresponds at one and the same time to the direct reception of the signal transmitted by the opportunity source (FM transmitter), to the reception of the reflections of this opportunity signal off varied fixed obstacles, these signals possibly being regarded as what the person skilled in the art knows by the name clutter, and to the reception of the "useful" signals originating from the reflections of this same source by the mobile objects that it is sought to detect. The complexity of the signal received is moreover further exacerbated by the reception of spurious signals that may for example originate from other, distant, transmitters transmitting in the same frequency band. Furthermore the useful signals (i.e. those backscattered by the mobile targets) generally exhibit a level very substantially lower than the direct reception, substantially lower than that of the reflections on fixed obstacles and substantially lower than that of the thermal noise. Another difficulty related to the implementation of such a structure resides in the fact that the opportunity transmission corresponds to an unknown signal whose properties are not controlled (bandwidth, level of the distance and Doppler sidelobes, etc.). So, the signal transmitted by the opportunity source must be Identified as such and isolated from the other signals received by the radar receiver in order to serve as reference during the correlation computations and to .allow the elimination of the spurious detections corresponding to Doppler and distance sidelobes. PRESENTATION OF THE INVENTION An aim of the invention is to propose a solution making it possible to implement a passive monitoring structure composed of one or more passive radar systems dispersed over the geographical zone to be covered, which solution thus makes it possible to solve the difficulties cited above. For this purpose, the subject of the invention is a method of processing the signal received by an FM passive radar comprising a plurality of reception pathways (Vi VN), the method comprising a phase of coherent processing carrying out a re-conditioning of the signals received and then a phase of non-coherent processing carrying out the construction of blips on the basis of the signals arising from the coherent processing. According to the invention the coherent processing comprises : - a step of forming cleansed target pathways, - a step of regenerating the reference signal, - a step of computing the cross-ambiguities between cleansed target pathways and regenerated reference signal, - a step of computing the auto-ambiguity of the regenerated reference signal. According to a preferred mode of implementation of the method according to the invention the non-coherent processing furthermore comprises : - a step of distance-Doppler extraction applied to the PEs created on completion of the step of normalization and of searching for the detections so as to construct blips, - a step of Doppler-distance purification applied to the blips constructed after distance-Doppler extraction and carried out by means of the signal resulting from the step of computing the auto-ambiguity of the regenerated reference signal, the determination of the attributes of the blips being finally carried out on the basis of the blips obtained after the Doppler- distance purification step. According to this preferred mode of implementation, the distance-Doppler extraction step uses the computation of the -3dB band of the regenerated reference signal. According to this preferred mode of implementation, the -3dB band of the regenerated reference signal is estimated on the basis of the signal resulting from the step of computing the autoambiguity of the regenerated reference signal. According to a preferred mode of implementation, the non-coherent processing furthermore comprises: - a step of distance ecartometry measurements, - a step of azimuthal ecartometry measurements, *hese two steps being applied to the blips constructed on completion of the Doppier-distance purification step. According to this mode of implementation the non-coherent processing also comprises a step of azimuthal purification and a step of limiting the lumber of blips sent. DESCRIPTION OF THE FIGURES The characteristics and advantages of the invention will be better appreciated by virtue of the description which follows, which description sets out the invention through a particular embodiment taken as nonlimiting example and which is supported by the appended figures, which figures represent: - Figure 1, a schematic diagram of all the steps of the processing according to the invention, - Figures 2 to 3, schematic illustrations of the operating principle of the step of forming cleansed reception pathways, - Figure 4, a schematic Illustration of the operating principle of the step of regenerating the reference signal, - Figure 5, a schematic illustration of the operating principle of the step of computing the reference signal/cleansed target pathways cross-ambiguities, - Figures 6 and 7, schematic illustrations of the operating principle of the step of extracting the blips, - Figure 8, a schematic illustration of the operating principle of the distance-Doppler purification step. DETAILED DESCRIPTION Attention is first turned to Figure 1, which presents the typical schematic overview of a signal processing method implementing the invention. The method according to the invention is here presented in its nonlimiting application to a passive radar comprising a plurality of independent reception pathways. To facilitate the understanding of the invention, the method is firstly presented in a general manner with all its processing steps. The steps enclosed in bold, specific to the invention, are detailed subsequently in the document. The other steps implementing known methods are simply mentioned so as to facilitate the description of the method which combines all of these steps. As illustrated by Figure 1, the method according to the invention implements two types of processing, a first type of processing 11, dubbed, in a manner known to the person skilled in the art, coherent processing which undertakes a re-conditioning of the signals received by the radar, followed by a second type of processing 12 dubbed non-coherent processing carrying out the construction of blips on the basis of the signals arising from the coherent processing. One speaks here of coherent processing for all the processing steps for which the signals are processed both In terms of amplitude and phase. Conversely one speaks of non-coherent processing when the-phase of the signal is no longer taken into account in the processing carried out on the signal. According to the invention, the coherent processing 11 principally comprises two steps, a first step 13 of so-called "cleansed target pathway formation", and a second step 14 of coherent integration by computing the cross-ambiguities between the signal received on each "cleansed" pathway and a reference signal corresponding to the signal transmitted by the opportunity source. The first step 13, of the coherent processing 11 is termed "cleansed reception pathway formation", since its objective is to minimize on the target pathways, pathways used in a known manner to search for detections or else "echo presences", the power of the nuisance sources which might limit the sensitivity of the radar. In the case of a passive radar, such nuisance sources may originate : - from the signal of relatively high level with respect to the useful signals and originating in a direct path from the reference transmitter, that is to say from the opportunity source. According to the invention, this direct signal, used moreover to carry out the step of coherent integration as reference signal, constitutes by its level a significant nuisance for the reception of the useful signals whose level is generally much lower, - from the multiple reflections of the signal transmitted by the source off fixed natural obstacles (multipaths of the reference transmitter), - from other FM transmitters working, at least partially, in the same band as the reference transmitter and whose signal reaches the receivers of the radar considered, - from the sources of radioelectric pollution of diverse nature. For a system comprising several acquisition pathways (multi-sensor system), as is the case for a passive radar- operating on FM opportunity transmissions, the elimination of the nuisance sources can be carried out in the spatial domain, by "antenna processing" of "Adaptive Computational Beam Forming" (Adaptive CBF) type or of "Opposition in the Side Lobes" (OSL) type which are types of processing well known to the person skilled in the art. These types of processing constitute the solutions customarily adopted for active radars working in the high frequency bands usually used for these radars. The elimination of the nuisance sources can also be carried out jointly in the temporal and spatial domains. One then speaks of spatio-temporal processing. It is this second alternative that is implemented in the method according to the invention. According to the invention, the algorithm used is an algorithm of spatio-temporal type with directional constraint, which processing is denoted by the term Spatio-Temporal Adaptive Formation (STAF, known by the term FAST in French standing for Formation Adaptative Spatio-Temporelle). This type of algorithm advantageously makes it possible to minimize the number (the multi-paths can be eliminated on the time axis) and the extent (by virtue of the use of the directional constraint) of blind angular zones appearing in the direction of the jammers. Here the term "jammer" is understood in the general sense of "disturbance". It should be noted that the FM waveforms not being separable in terms of distance and Doppler, the elimination of the multiple paths may not be done, as is customary for active radars, by applying a filter known to the person skilled in the art by the name MTI (Mobile Targets Indicating, known as VCM in French standing for Visualisation des Targets Mobiles). The second step 14 of the coherent processing 11 consists in characterizing in terms of distance and Doppler the signal samples received, by coherent integration of the signal received over a given duration. According to the invention, the signals reflected by the targets being delayed and Dopplerized versions of the reference signal, the coherent integration of the samples of the signal received can be carried out by determining the cross-ambiguity function defined for a time shift m, lying between 0 and D and a bistatic Doppler shift n lying between 0 and M-1 by : M-l , j252!l r,, amb^_r,)=ZVrefMvtergetcln[^<+n^]e M [1] k=0 for which relation :. - M represents the number of temporal samples integrated in coherence, - D represents the maximum number of time shifts tested, - Vref[k], 0 :S k < M - 1, the reference signal, - vtarget cin[k], 0 to the thermal noise power on the pathway voppi. It should be noted furthermore that if R denotes the covariance matrix of the thermal noise taken on the N reception pathways, it Is possible to write : Rb=BR-BH [11] The thermal noise being assumed spatially white on the N reception pathways, we have : R = IN,N and hence: Rb =BB'^. The expression for the vector Wref is then : Wref = (BBH)-i.B-dref [12] Hence, the matrix O is therefore defined as the product: O = F • B. As has already been stated previously the principal role of step 13 consists in the formulation of the signal vopptot which is subtracted from the signal vtarget. According to the invention the signal vopptot Is produced by filtering on the basis of the signals voppi to voppN-i. The filtering implemented is of adaptive type and its object is to obtain a signal vopptot making it possible to best eliminate the undesirable signal components. In practice it is achieved by applying to each signal voppi to voppN-i a time filter whose coefficients are recomputed periodically with the sole aim of producing at each instant the desired signal vopptot. The principle for computing the various coefficients of the filter applied to each signal voppi (i varying from 1 to N-1) is presented in the subsequent description. To perform this computation the following two vectors are formed first of all: - vopp(k) vector of the values of the opposition signals, which is composed of temporal samples of the signals voppi (i varying from 1 to N-1) taken at various successive instants around the instant k, - h vector of the coefficients of the filters applied to the signals voppi. Moreover vtarget(k) denotes the value of the signal vtarget for the instant k (temporal sample k) considered. The expressions for these two vectors are respectively : ^ Voppl(k + Ri) - vopp(k) = Vopp,i(k-Ri) Vopp,2(k + R2) Vopp,2('^-f^2) Vopp,N-l(k + R2) Vopp,N-l(k-R2). - h = hi(-Ri) hi(Ri) h2{-R2) h2(R2) hN-l(-R2) hN-l(R2)j where Ri represents the delay of the opposition filter applied to the signal voppi, the length Li of the filter, that is to say the number of. coefficients, being equal to 2Ri+1, and where R2 represents the delay of the opposition filters ajDplied to the signals vopp2 to voppN-i, the length L2 of the filters being equal to 2R2+I. Hence, the expression for the signal vtargetcin corresponding to the cleansed target pathway is : vtargetcin(k) = vtarget(k)-h" vopp(k) [13] According to the invention the vector h is computed in such a way as to minimize the cost function J defined on K signal samples by the following relation : K J = Sktarget(k)-h^Vopp(k)| [14] The filter h minimizing the criterion J, is then defined by the following relation : ^ -^ vopp.vopp ■' vopp,vtarget [15] vopp.vopp where r, represents the matrix defined by the following relation K •"vopp.vopp = Xv°PP^'^)-v°PP(k)" [16] k=1 and where Tyopp^target represents the vector defined by the following relation: '"vopp.vtarget = EvoPP(k)- vtarget(k)" [17] k=1 According to the invention, the solution adopted for computing h is therefore of the type of the block version of the deterministic least squares algorithm. It should be noted therefrom that in order to improve the conditioning thereof, critical both on the space axis and on the time axis because of the high level of rejection of the receiver filters at the band limit, the matrix Fvopp.vopp is overloaded prior to the computation of the vector h according to a principle that is well known to the person skilled in the art. To optimize the performance of the formation of the cleansed pathways, the computation of the vector h can be updated on partially overlapping data blocks. The raw computation of the terms of the matrix Fvopp.vopp necessitates separate estimation of a number NT of terms equal to (Li + (N-2)L2) (Li + (N-2)L2+1)/2, which number corresponds to the terms situated in the upper triangular part of the Hermitian matrix Tvopp.vopp- The computational load necessary for such a computation is significant and this is why, in a preferred form of implementation, the method according to the invention carries out this computation by exploiting the particular structure of the matrix Tvopp.vopp-Indeed, the Hermitian matrix Tvopp.vopp can be written in the following form : ' voppi,voppi ' VOPP1,VOPP2 r r VOPP1,VOPPN_1 V0PP1,V0PP2 ' V0PP2,V0PP2 vopp.vopp ■• VOPPN_2,VOPPN_1 r» rH r V0PP1,V0PPN_1 vopPN_ .2.V0PPN_i V0PPN_1,V0PPN_I [18] each of the Hermitian sub-matrices Tvoppivoppi of the matrix Tvopp.vopp furthermore having the following structure : opPi.vopPi K K K ^vopP|(k + Ri)vopPi"(k + Ri) ^vopP|(k + Ri)vopP|'(k + Ri-1) ■•■ ^vopP|(k + R,)-vopPi'(k - R, + 1) k=1 k.1 K=t f. K-1 (5^vopP|(k + R|)-vopp'(k + Ri-1)r 2vopPi(k + Ri)vopP|'(k + Ri) :=1 , k=0 K ■ " ■ K-L|-H (5]wopPi(k + Ri)vopPi'(k-Ri + 1))" - - ]^vopP|(k+Ri)vopP|*(k + Ri) k.1 k=1-L|t1 [19] In this way, for any row m and for any column n of the matrix rvoppi.voppi it is possible to write, for m ^ n : K rvoppj,vopPi(n^.n)=I]vopPj(k + Rj+1-m)voppj (k + Rj+l-n) [20] k=1 or else : K-m+1 rvopPi.vopPi(fTi'n)= Z voppj(k + Ri)-voppj (k + Ri+(m-n)) [21] k=2-m or else: (1) rvopPi,vopPi("^>")= X^°PPi('^ + ^i)"^°PPi (k + Ri+(m-n)) + k=2-m (2) K-L + 1 Y, voppj(k + Ri)-voppj*(k + Ri+(m-n)) + k=i (3) K-m+1 2 voppj(k + Ri)voppi (k + Ri+(m-n)). [22] k=K-Li+2 By consulting relation [22], it is noted that the terms (1), (2) and (3) constitute, respectively, the sums of (m-1), (K-Li) and (Lj-m) products. It is furthermore noted that the term (2) depends only on the difference (m-n), and that it therefore need be computed only once for all the terms of Tvoppi.voppi that are situated on one and the same sub-diagonal. Finally it is noted that the computational load due to the computations of the terms (1) and (3) is negligible insofar as K- Lj» 1,-1. In this way, the estimation of the sub-matrices rvoppi.voppi is therefore akin, to first order, in terms of computational load, to the computation of L, values (the Lj values of the term (2)), although in a raw computation it corresponds to the computation of L, (Lj + 1 )/2 values, these values corresponding to the number of terms situated in the upper triangular part of the Hermitian matrix A voppj.voppi- By proceeding in an analogous manner, it is possible to compute at lesser cost the terms of the sub-matrices ryoppi.voppj (with \¥^]) which are situated on one and the same sub-diagonal. The computation of the term (2) for each sub-diagonal therefore constitutes the largest share of the computational load. It consists for a matrix of type rvoppi.voppi in evaluating the term C(m-n) defined for the parameters m (row number) and n (column number) by the following relation : K-Lj+1 C(m-n)= YJ voppj(k + R|)-voppj*(k + Rj+(m-n)) [23] with : - m s n, - 1 < m < Lj, - 1 < n ^ Lj. For large enough values of Lj it is advantageous, in terms of computational load, to perform the estimation of the terms C(m-n) in the spectral domain. To do this the following quantities are defined : -N = K-Li+1, -D = N+Li-1, - voppsynch(k) = voppi(k + Ri +1) for k = 0,..., N-1, - voppcomp(k) = voppi(k - Rj +1) for k = 0,..., D-1. Thereafter, on the basis of voppsynch(k) and voppcomp(k) the sequences voppsynchpad(k) and voppcomppaci(k) of length S are constructed by "0 padding", S being the smallest power of 2 greater than or equal to N+2(Li-1). Then VFoppsynch(l) and VFoppcomp(l) are computed, these being respectively the results of the fast Fourier transforms of voppcompad(k) and voppsynchpad(k). The temporal signal p(k) defined, for 0 ^ k < S-1, by : p(k) = FFT^ (VFoppsynch(l) VFoppcomp*(l)), is deduce from VFoppsynch and VFoppcomp. Hence, for each pair (m,n) such that: - 1 n) = t' C(m-n) = 2_,voppj(k + Rj)voppj (k + R-, +(m-n)) k=1 = p(S-Li+1-(m-n))for m-n ^ ^-U, = p(0) for m-n = 1-Li, In practice, to maximize the load reduction afforded by the computation in the spectral domain, the estimation of the values of the signals p(k) is performed on several contiguous temporal sub-blocks so as to use FFTs of reduced size. The values of C(m-n) are then obtained by summing the values of the signals p(k) computed on the various sub-blocks. This principle of passing to the spectral domain, described previously, can obviously also be used to reduce the computational load induced by the computation of the terms of the type C(m-n) of the matrices rvoppi.voppj with i ^ j, to the extent of course that the lengths Li of the filters justify this. It can also be used to perfomn the computation of the vector Tvopp.vtarget- Attention is now turned to Figure 4, which illustrates the principle implemented for carrying out step 15 of the method according to the invention, the step of regenerating a reference signal. In order to obtain the reference signal making it possible to perform the coherent integration and to facilitate the steps of extraction and Doppler/distance purification of the non-coherent processing, the method according to the invention implements, as was stated previously, a step termed "step of regenerating a reference signal" the aim of which is to obtain a reference signal which is as similar as possible to the signal transmitted by the source. This signal must be ridded, like the signal vtarget, of the signals arising from the multiple paths and other potential nuisance sources. One of the characteristics of the processing for regenerating a reference signal is that it must discriminate, in a blind manner, the reference signal from the other signals. The procedure used by the method according to the invention to carry out this blind discrimination uses the property of constant modulus of the FM signals to find a collection of purely spatial weightings which, applied to the signals received on the antenna, gives as output the estimation of the reference signal. The algorithms of the family known by the acronym CMA standing for "Constant Modulus Algorithm" make it possible to carry out the blind equalization of constant modulus signals. In a preferred form of implementation, the method according to the invention uses a variant CMA algorithm known by the acronym LS-CMA standing for "Least Square Constant Modulus algorithm". The latter in fact offers a good compromise between performance and computational load. Figure 4 illustrates the operating principle of this algorithm. The LS-CMA algorithm is a block iterative algorithm for which the weightings applied to the input signals Vi to VN are computed by successive iterations on one and the same block of data. Its object is to determine the vector Vref and the vector WCMA which are solutions of the least squares problem formulated by the following relation: H argmin wcma.vref = II ^CMA • V - Vref II [24] in which the expressions for v, Vref and WCMA are respectively vi(1) - Vi(M) VN(1) - VN(M) Vref = l^ref('<) ••• Vref(M)| w CMA WiCMA WNCMA V is the nriatrix of the signals received V|(k) on the N reception pathways over the duration of the coherent integration. WCMA is the vector of the CMA weightings making it possible to reconstruct the reference signal having a constant modulus, The signal Vref is for its part the constant modulus signal which represents the estimation of the reference signal. This is the so-called "regenerated" reference signal. A feature of the processing method according to the invention is that it must be able to take account of the large dynamic swing of the processed signals, the reference signal possibly being received with a level 70dB higher than that of the noise. This significant dynamic swing can cause diificufties in implementing the corresponding digital computations. To take account of this feature of the processed signals, step 15 of the method begins, as illustrated by Figure 4, the operation 41 of actually evaluating the weightings WCMA, such as is provided for by the CMA algorithm, with an operation 42 of so-called "whitening of the signal space" and dimension reduction. The signal space whitening operation begins with the computation of the sensor signals covariance matrix Rv defined by the following relation: Rv=oVv^ [25] M The (positive defined) matrix Rv is thereafter diagonalized in an orthonormal basis according to : Rv = Uv Dv Uv", where the eigenvalues of Rv are ranked in decreasing order along the diagonal of Dv- The matrices Dv and Uv, are thereafter respectively used to construct the matrices Dvred and Uvred, which matrices are extracted from Dv and Uv by retaining only the eigenvalues of these matrices, and the associated eigenvectors, whose dynamic swing with respect to the largest eigenvalue of Dv is less than about 50dB (so-called dimension reduction operation). The whitening filter Fb applied to the input data Vi(k) is then deduced from Dvred and Uvred through the relation : Ffa - Uvred ■ Dvred [26] Hence the whitened input data Vbi(k) are obtained on the basis of the input data Vj(k) by using the following relation : Vb=Fb^.V [27] These whitened data constitute according to the invention the input data of the LS-CMA iterative algorithm. Initially, the vector of the weightings WcwiAb in the whitened signals space is initialized to a value w^^Ab init- '^ ^^^ reference signal is the most powerful, the value given hereinafter can be used : WcMAb=^CMAb init = Then, at each iteration of the algorithm, a new value of WcwAb is computed in accordance with the algorithm described below : - The vector Y of the signals obtained as output from the pathway formation is formed with the value of WcMAb obtained at the previous iteration. The expression for Y is : Y=WcMAb''Vb [28] - The vector Ymod_cst> associated with a constant modulus signal, is deduced from Y. Ymod_cst is a vector whose components are obtained by dividing each component of Y by its modulus : Y™d_cst=fY/|Yil]i [29] - The components of the weighting vector WcMAb are updated on the basis of Ymod_cst by performing the following operation : ^CMAb = ( Ymod est " pinv(Vb))H [30] where pinv denotes the Moore-Penrose inverse. Hence, after a few iterations (of the order of 5 to 10), the value of the components of the sought-after weighting vector WCMA is deduced from those of the vector WcMAb by applying the relation : WcMA=FbWcMAb [31] The vector representing the reference signal 43 is then obtained by applying the weighting coefficients to the input signal 44. Attention is now turned to Figure 5 which illustrates in a schematic manner the operating principle of the operation performed during step 14 of the method according to the invention, step of so-called "computing the reference signal/cleansed target pathways cross-ambiguities" which carries out the actual coherent integration of the signals back-scattered by the observed objects, or targets. The object of this step is to carry out an analysis of the signal received on the distance axis and on the Doppler axis, through distance-Doppler cells (m,n) of defined size and on a likewise defined distance-Doppler domain. If we initially consider the number M of temporal samples integrated in coherence, the maximum number D of distance shifts tested, the reference signal Vref[k] obtained by step 15 of the method for 0 ^ k ^ M-1 and the cleansed target signal Vtargetcin[k] obtained by step 13 for 0 < k < M-1+D, the signals reflected by the targets being moreover considered to be delayed and Dopplerized versions, that is to say suffering a frequency shift due to the movements of the target, of the reference signal, the coherent integration carried out by the method according to the invention consists in computing the quantity amb(m,n), called the cross-ambiguity. This quantity is given by the following relation : M-1 , i2TTnl< amb(m.n) = Vv^f|k]Vtargetcin ^^+^'\^ ^ 132] k=0 where Vtargetcin*[k] corresponds to the conjugate of Vtargetcin[k], and where the parameter m, lying between 0 and D, characterizes the delay of the target signal with respect to the reference signal and the parameter n, lying between 0 and M-1, characterizes the bistatic Doppler frequency of the target. The computation of the function amb(m,n) can naturally be carried out in a direct manner by computing, for each value of k and of m, the products of the type Vref[k]vtargetcin*[k+m], and then by evaluating the FFTs of the temporal signals thus obtained for each value of m. However the computational load resulting from such a computation is generally excessive. This is why in the method according to the invention an indirect fast computation procedure, which advantageously induces a lower computational load, is used. The principle of this procedure is described subsequently in the description. The principle of the procedure described here relies on the fact that the speed of the targets to be processed not exceeding 2 to 3 times the speed of sound, the maximum Doppler frequency to be instrumented, fdmax, is much lower than the ambiguous Doppler frequency for a sampling frequency ofthe order of 150kHz. Thus, for example, for a passive radar operating in the general-public FM band, that is to say at a frequency close to f = 100 MHz, the value of fdmax is: F,_=^X-.2J000»667Hz. d"i« X - ,3 Hence it is possible to neglect the variation in the Doppler phase of the signal received over sub-blocks of a few tens of successive samples. Thus, as the Doppler phase variation of the signals received can be neglected over a few tens of consecutive samples, it is possible to estimate the cross-ambiguity by performing correlation computations solely on the distance axis on sub-blocks of L samples, and then by computing the FFT of these correlations in distance. The computation of the distance correlations can furthermore be perfomned in the spectral domain, the effect of this being to substantially decrease the computational load. It is on these principles that the processing carried out is based. According to this preferred form of implementation, the computation of the cross-ambiguities is performed in two steps. The first step consists in computing, on sub-blocks 51 and 52 of L samples of the regenerated reference signal and of the cleansed target signal, the elementary distance correlations cdlst(m,s) 53 defined by the following relation : (s+1)L-1 CdiSt(m,S)= Y, Vref [k]Vtargetcln*[k + m] k=sL (S+1)L-1 _ [33] k=sL with : 0^mndisf), satisfies the following relation : logmod(ndop , n^ist) > logmod(ndop + i, n^jst + J) [43] for each distance-Doppler bin (ndop+i,ndist+j) belonging, as illustrated by Figure 7, to a neighborhood E, formed of the union of the groups of distance-Doppler bins 71 to 74 encompassing the bin 75 considered and defined by : E = il^dop " "I ■ "dop + U^ hdist ' "ssechdist - f^dist ' ul ^ {"dop -1. "dist } ^ {"dop + 1. "dist } ^"^^^ ^ {["dop - "< - "dop + ""l^^ hdist + ^' "dist + nssechdist 1) where ngggchdist represents the nearest integer (EPP) to the ratio —55Q& BsdB representing the estimation of the 3dB bandwidth of the FM signal received for the current burst. As may be noted in the foregoing, in order to be carried out, the extraction step 18 according to the invention has to know the value Bade of the band of the signal at 3dB. The method according to the invention determines this band Bade by implementing a computation step 111 on the regenerated reference signal. According to the invention, the 3dB band of the reference signal is deduced from the auto-ambiguity function of the regenerated reference signal, by applying the procedure described hereinafter. The section along the distance axis of the auto-ambiguity function can be approximated on a linear scale and around the zero distance, by a Gaussian. As a result, on a logarithmic scale, the distance cut may be regarded as a parabola : Auto-amb-dist-log(d) = ad^ + bd + c [45] where : Auto-amb-dist-log(d) is the logarithmic modulus of the distance section of the auto-ambiguity function, d characterizing the distance variable, a, b and c denote the coefficients of the parabola modeling the distance cut on a logarithmic scale. The distance section of the auto-ambiguity function being a maximum at zero distance, the coefficient b is zero. The width of the lobe of the distance section of the auto-ambiguity function on a logarithmic scale does not depend on the coefficient c (which fixes the value of the auto-ambiguity function at zero distance), but solely on the coefficient a. The value of the coefficient a can be estimated, on the basis of the reference signal auto-ambiguity function computed in step (19) by the procedure described hereinafter, or any other parabolic approximation algorithm. Let auto_amb_log(m,n) = 20*log10 |auto_amb(m,n)| be the logarithmic modulus of the auto-ambiguity function computed in step (19). ^i2(auto_amb_log(i,0)-auto_amb_log(0,0)) a = t^ [46] 14 i=-2 The auto-correlation of the reference signal on a linear scale being assumed Gaussian, the DSP of the reference signal is also Gaussian. Its 3dB band can be deduced from a, by : B.B- ,-/^ [47, TT.-10 'ogio(e) af2 'smp where fsmp denotes the sampling frequency. The value Bade of the 3dB band thus obtained on completion of step 111 is used during step 18 for the computation of ngggg^disf Attention is now turned to Figure 8 which illustrates the operating principle of the Doppler-distance purification step 112 according to the Invention. The object of this step is to eliminate the spurious blips corresponding to the distance/Doppler sidelobes of targets of high level. Indeed the FM-like reference signal exhibits, distance-wise and Doppler-wise, sidelobes of high level and which are relatively far from the principal lobe. These sidelobes can create, during the Normalization and Detections Search step 17, echo presences (EP) which are not eliminated by the Doppler-distance extraction step 18 which analyses each EP through its near environment. To carry out this Doppler-distance purification operation, the method according to the invention uses the distance-Doppler auto-ambiguity signal of the regenerated signal. According to the invention the extracted blips are ranked in order of decreasing level. The processing then consists in translating the autoambiguity function of the reference signal distance-wise and Doppler-wise in such a way that its origin 81 coincides with the distance-Doppler bin of the blip 82 considered. The level of the autoambiguity function is then adjusted on the basis of the level of the processed blip so as. to exceed this level by a value called the overshoot threshold. The function thus realized constitutes the purification template 83 used. Hence all the blips 84 - with the exception of the current blip 82 and of the blips situated in its near neighborhood - whose amplitude is below the purification template are considered to originate from Doppler or distance sidelobes and are deleted. The blips such as the blip 85, whose amplitude is above the purification template are, for their part retained. It should be noted that the illustration of Figure 8 constitutes a simplified two-dimensional representation in which all the processed blips are situated in the same distance bin. This representation simplified for the sake of clarity nevertheless makes it possible to clearly illustrate the operating principle of the purification step. On completion of this purification step the remaining blips are used for the subsequent processing steps. CLAIMS 1. A method of processing the signal received by an FM passive radar comprising a plurality of reception pathways (V1, ..., VN), the method comprising a phase of coherent processing (11) carrying out a re-conditioning of the signals received and then a phase of non¬ coherent processing (12) carrying out the construction of blips on the basis of the re-conditioned signals and then the determination of the attributes associated with the blips constructed, characterized in that the coherent processing comprises : - a step (13) of forming cleansed reception pathways, - a step (15) of regenerating a reference signal, - a step (14) of computing the cross-ambiguities between cleansed target pathways and regenerated reference signal, - a step (19) of computing the auto-ambiguity of the regenerated reference signal. 2. The method as claimed in claim 1, characterized in that the non¬ coherent processing furthermore comprises : - a step (18) of distance-Doppler extraction applied to the blips constructed, before determination of the attributes associated with these blips, - a step (112) of Doppler-distance purification applied to the blips constructed after distance-Doppler extraction and carried out by means of the signal resulting from step (19) of computing the auto-ambiguity of the regenerated reference signal, the determination of the attributes of the blips being carried out on the basis of the blips obtained after the Doppler-distance purification step (112). 3. The method as claimed in claim 2, characterized in that step (18) of distance-Doppler extraction uses the computation of the -3dB band of the regenerated reference signal. 4. The method as claimed in claim 3, characterized in that the -3dB band of the regenerated reference signal is estimated on the basis of the signal resulting from step (19) of computing the autoambiguity of the regenerated reference signal. 5. The method as claimed in any one of the preceding claims, characterized in that the non-coherent processing furthermore comprises: - a step (113) of distance ecartometry measurements, - a step (114) of azimuthal ecartometry measurements, these two steps being applied to the blips constructed on completion of the Doppler-distance purification step (112). 6. The method as claimed in claim 5, characterized in that it furthemnore comprises, a step (115) of azimuthal purification and a,:., step (116) of limiting the number of blips sent.