DE3639500C2 - - Google Patents

Description

The invention relates to a method for the two-dimensional, ie effective in space and time suppression of clutter echoes in echo signals for detecting moving targets with a radar receiver of a moving radar, with a phased array antenna, sent over the M transmit pulses and M echo signals over N antenna elements of the array antenna and receive from the radar receiver the echo signals received from each of the N antenna elements in receive channels to N * M input signals consisting of N signal sequences of length M , amplified, down-converted and digitized, and then from the N signal trains given predetermined weighting factors (b ₁ ... b _N ) a lobe formation takes place, by which a single signal sequence of length M is generated from the N signal sequences, which signal is fed to a Doppler filter bank which delivers a search signal at its outputs. The invention further relates to the embodiment of a radar receiver for carrying out the method according to the preamble of claim 1.

Locating moving targets near the ground, such as Tiefflieger, Vehicles, by an elevated position, for. Legs flying platform with which the radar device is moved, z. B. the on-board radar of an aircraft can be significantly increased there influences such as shadowing in undulating terrain and on jams of the radar lobe near the ground can be reduced. The desired target signal of the moving target, however, must from the entire, including all interfering signals (clutter) containing echo field by suppression of clutter of all kinds by means of a Clutterfilters be won.

Regarding the clutter echoes arise at z. B. be flying radars have some disadvantages, of which the following two of particular importance for the development of a Clutterfilters are:

• a) The various clutter elements are under different azimuth and elevation angles relative to the flight axis of the radar. The radial component of the relative velocity between the clutter element and the axis of flight of the radar is thus proportional to the cosine of the angle between the locus vector of the clutter element (seen from the radar) and the axis of flight. Since the relative radial velocity between Clutterelement and radar determines the Doppler frequency of the radar echoes are those Bewegtclutterechos, in particular of a homogeneous background, doppler colored, ie the received Cluttersignale have all possible Doppler frequencies between v _max (airspeed), and - v _max on. In horizontal flight, curves of constant Doppler frequency form a flock of hyperbolas that open in the direction of flight (or against it). If you walk on a range ring ent long, so you can step through the entire Clutterdopplerspektrum of v _max to - v _max.
• b) When flying over different terrain formations be the radar device encounters zones with different returns scattering behavior, d. H. the clutter background is temporal and spatially variable. It follows that the clutter suppression system should be stoopaptic, d. H. it must have the ability to adapt to changing disturbances adapt.

From a) it follows that clutter signals of several parameters, namely azimuth, elevation, distance and Doppler frequency hang. However, there is a fixed relationship between the location coordinates of each clutter element and its Doppler frequency.

FIG. 1 shows such a distance-azimuth Doppler network. Here, φ is the azimuth angle, v is the clutter radial velocity, r / h is the range / flight altitude ratio, and HNr is the number of the respective hyperbola.

If one limits oneself to a distance ring, then one recognizes that the Clutterleistung is a function of two parameters: Azimut φ , Doppler frequency f _D and Clutterradialgeschwindigkeit v , and can be represented for a Radarzielbereich by a two-dimensional power spectrum. In FIG. 2, this is an example of a Bewegtclutterspektrums for the line of sight of the transmitting lobe φ _B = 0 °, ie in the flight direction, is provided. There is a maximum - main lobe clutter at d = 0 ° and the associated clutter frequency v = v _max . Any destinations to be detected are located in the transmission lobe direction φ = 0 ° and at all possible speeds - v _max . , , + v _max , that is along the left upper edge in Fig. 2. Along the diagonal one sees the side lobe clutter distributed over all Doppler velocities. CNR means the clutter noise ratio, M the number of echo pulses, N the number of sensors.

Considering a certain distance ring, a dependence of the clutter power of two parameters remains: azimuth φ and Doppler frequency f _D (or relative velocity v) . A filter for suppressing such spurious signals must therefore be effective in two dimensions, ie in φ and v (or in space and time).

It follows that the backscattered echo field in two dimensions tions, and must be recorded spatially and temporally. These assumptions are made by a coherent pulse Doppler Radar met with phased array antenna, the Sensors as spatial sampling and the echo pulses (coherent Echofolgen) are regarded as temporal sampling. When Clutter model becomes a cluster of the clutter echo samples. Time covariance matrix formed.  

Most of the hitherto known methods of suppression in moving radars are based on a modification of the known 2 pulse clutter canceller.

For one thing, there are methods that use the method of "clutter In this procedure, the maximum of the Clutterspektrums in the restricted area of the 2-Puls-Canceller ge pushed, so that some extinction of the main lobe Clutters done. Application has this principle in the TACCAR System found (M.I. Skolnik "Radar Handbook" McGraw Hill, 1970) p. 17-32). A disadvantage of TACCAR and related methods is that no motion compensation takes place, so that the from the main lobe detected clutter spectrum a relatively high Doppler bandwidth has only part of the 2-pulse canceller can be suppressed. Sidelobe Clutter has from Main lobe clutter different Doppler frequencies and is over not suppressed at all.

On the other hand, there is the so-called DPCA technique (displaced phase center antenna). Here is the movement of the platform of the Radar compensated by that between every two pulses of a receiving array on a second, in flight direction behind it mounted array is switched such that the platform speed between two pulses is just canceled. The method has the disadvantage that at comparable Auf wand around 6 dB of transmission energy for clutter suppression get lost.

A precursor to adaptive 2-dimensional clutter suppression for moving radars is in Andrews' Radar Antenna Pattern Design for Platform Motion Compensation "IEEE Trans, AP, Vol. Ap-26, No. 4, July 1978, pages 566-571. Here are the Ge weights of two evaluation sets of a sensor array opti that the difference of their clutter responses is minimal. An adaptive realization is not treated. All of these known method is common that only two consecutive the following echoes are used for clutter suppression.

In the publication by Brennan, LE, Mallett, ID, Reed, IS: "Adaptive Array in Airborne MTI Radar", IEEE AP-24, No 5, Sept. 1976, pages 607-615, it is noted that the moving clutter problem (such as other types of interference suppression) can in principle be solved by the optimal detector (likelihood ratio test = LR test).

The optimal detector is known to be given by the Likelihood Ratio Test ( LR Test Probability Test):

With

x Vector of the received signals
R clutter covariance matrix of the spatio-temporal samples
s Vector of the target signals
n vector of clutter and noise values

If the expression on the left exceeds a threshold, then the decision "target exists", otherwise not. Equation (1) is a very general rule which can be applied to arbitrary decision problems. In case of detection of flying targets in moving clutter, R has the dimension ( MN) × (MN) , the vectors x , n and s have the dimension MN .

The optimal receiver is known to consist of a two-stage filter, a noise suppression component and a target matched filter (pre-whiten and match). The first part consists of the inverse covariance matrix of the clutter signals. In the case of 2-dimensional signal sampling, the order of such a matrix may take on dimensions that are no longer suitable for real-time use. An example: A small antenna has N = 32 sensors, the coherent signal processing works with M = 32 coherent pulses. The order of the matrix is N × M = 2¹⁰ = 1024. The matrix contains (N · M) ² = 1,048,576 elements, for their inversion one needs (N · M) ³ = 2³⁰≈10⁹ arithmetic operations. These dimensions can not be mastered by today's computing capacity and will hardly be so in the future.

Basically, resulting from the 2-dimensional Signalver work no particular problems; the limitation of profit is due to the receiver noise (as well as in the 1-dimensional Case). The to the realization of the optimal detector too However, driving effort is so high that he is currently with the Available devices neither in terms of cost nor computing time to cope with.

In the unpublished DE-OS 35 43 577 a suboptimal adaptive radar receiver with an active array antenna for two-dimensional suppression of clutter echoes in received echo signals of a moving radar is already described, is used in which instead of the LR test with a suitable Vortransformation to reduce the computational effort and subjecting all received signals to transformation. For pre-transformation, a space search channel (search lobe and Doppler filter) in the azimuth and Doppler frequency is assigned to the clutter adapted space Doppler channels as Clutterhilfskanäle and at the output of this channel multiple, the spatio-temporal correlation matrix is estimated, inverted and used as an adapted filter.

The invention is also based on the adaptive radar receiver of DE-OS 35 43 577 underlying consideration of creating a sub-optimal method to the LR test by suitable pre-transformation of all received signals, the order of interference suppression and matched filtering - pre-whiten and match - is reversed.

The object of the invention is to provide a method and a suitable radar receiver for adaptive, d. H. two-dimensional - in space and time - effective oppression of moving clutter to create a working method that the favorable characteristics of the optimal detector, however is reduced so far in the computing effort to be operated that a realization is possible.

The invention solves the task with the by the Features of claim 1 characterized method and with a radar receiver according to the characterizing features of Claim 5.

Advantageous developments are specified in the subclaims.

The invention operates on the principle of match and post whiten for the two-dimensional clutter suppression, wherein one or more search channels are formed by lobe formation and Doppler filtering of the received signals, which are still affected by all noise components in direction and magnitude, then the clutter suppression occurs. Instead of performing the known probability ratio test according to equation (1) at full rank N × M , according to the invention, the LR test is applied in the transformed range. For this purpose, the received signals of a linear transformation using the eigenvectors of the clutter covariance are subjected, so that the resulting vector space (or: the number of signal values to be evaluated per unit time) is significantly reduced. As a result, the on wall of the subsequent signal processing, interference suppression, spatially-temporally matched filter is reduced to an economically feasible level.

Thus, the invention enables the creation and operation a clutter-adaptive filter to suppress clutter of all kinds, in particular ground clutters, in radars, which are on relatively moving to the earth platforms be Find. Such a clutter filter allows the discovery of moving targets such as aircraft and vehicles aboard one flying radar platform, provided the goals in their Doppler frequency of that of background disturbing differ.

Theoretical Preliminary Examinations in the Publications of Klemm, R .: "Suboptimum Clutter Suppression for Airborne Phased Array Radars", PROC RADAR 82, London, UK, October 1982 and Klemm, R .: "Adaptive Clutter Suppression for Airborne Phased Array Radars", IEE Proc. Pts. F and H, Vol. 1, February 1983, pages 125-132 have revealed an important property of the covariance matrix of the spatial and temporal samples of the clutter echoes. If the number of sensors N and those of the echo values M , then one obtains an (N × M) (N × M) -Kovian matrix; however, the number of parent values is a maximum of N + M. It follows, on the one hand, that the number of eigenvalues, in particular for large values of N and M, is small compared with the number of samples or the order of the covariance matrix. It also follows that for a given total number of samples for N = M , the minimum number of clutch eigenvalues is reached, ie when the number of echo pulses equals the number of sensors. According to the invention, this finding is reversed by using the eigenvectors of the space-time covariance matrix of the clutter sampling values associated with the cluster eigenvalues as auxiliary channels for interference suppression. This makes it possible to derive on the basis of the optimal detector a processor for locating moving targets against moving clutter background in such a way that the favorable properties of the optimal processor are retained, but due to the appropriate transformation of the received signals, the amount of arithmetic operations and processing time of subsequent signal processing is reduced so far that a practical realization is possible.

For carrying out the invention, the covariance matrix R of the spatio-temporal samples must be determined in principle in this invention, however, a linear transformation of the received data is performed in simplification according to the invention using the eigenvectors of the covariance matrix, with the then the evaluation of the received signals, ie Clutterunterdrückung , is made.

The invention is carried out in such a manner that instead of carrying out the LR test according to (1) at full rank N × M , only one transformed reduced test according to equation (2) is carried out:

The transformed variables are defined as follows:

x _T = T * x (2a)
s _T = T * s
n _T = T * n
R _T = T * R T

Since the target signal s and thus also s _{T is} initially unknown, it is replaced by the control vector b ( φ , v ):

b _r ( φ, ν ) = T * b ( φ, ν ) (2b)

Since the matrix T anticipates the signal integration (lobe formation, Doppler filtering), and the sidelobe contributions to the target signal can be neglected over the main lobe contribution, is:

with e as the unit vector, where R _T ^-1 e represents the first column of R _T ^-1 and (2) simplifies

T is a NM × L matrix (L N + M) . It is chosen according to the invention so that

• a) no signal energy is lost  
• b) receive the clause echoes with high CNR
• c) L « MN , so that the desired savings in computation time, arithmetic operations and hardware costs are achieved
• d) L is not less than the number of parent values (N + M) .

If these conditions are fulfilled, the same behavior can occur of (3) as expected from the optimal detector in equation (1) become.

The structure of the transformation matrix forms the core of the invention described herein. In order to satisfy the established conditions, the matrix T , which is a rectangular matrix with N × M columns and L rows, must be composed of two parts:

• 1) 1 main channel, consisting of lobe formation (spatial) and Doppler filters (temporally, a Doppler to be suppressed channel) for the signal adaptation
• 2) L - 1 auxiliary channels of spatial + temporal samples, which serve to measure the disturbances and span the clutter subspace of R as well as possible so that the clutter echoes are received with the highest possible CNR (clutter / noise ratio). For this purpose, according to the invention, the eigenvectors belonging to a clutter covariance matrix that are intrinsic to the clutter are used.

The signal adaptation according to point 1) is determined by a target signal adjusted weighting, d. H. the elements of a Columns of the matrix must match the spatio-temporal samples correspond to an expected target signal. The signal adaptation is a vector with those of a specific target signal expected partial signals

s _{n, m} ( φ , n ) ~ exp j β _n exp j ω _D τ _m (4)

in which

n = sensor index
m = impulse index
τ _m = m th delay of the echo
ω _D = Doppler frequency due to the relative speed between radar and target
β _n = phase center related phase at the n th sensor due to the direction of incidence

For a regular rectangular sensor array parallel to Earth becomes

where x _n , y _{n represent} the sensor coordinates and R represent the elevation.

The first column of the transformation T looks like this:

in which

is a club formation and the coefficients

c _m ( v ₀) = exp j ω _D (v ₀) τ _m (8)

represent the coefficients of a Doppler filter. Thus presents the scalar product

y = x * t ₁

Such a "search channel" is adapted to a destination in a particular direction φ ₀ and at a certain radial velocity which depends on the target speed, the platform speed and the direction of flight of the target.

The remaining L -1 columns of T are used to estimate the clutter noise and should be designed to receive them with the highest possible clutter / noise ratio.

For the invention, the occupancy of the L -1 clutter vectors in T plays a central role. First, a preview is made.

Received clutter signals are statistically described by a space-time covariance matrix R , as previously introduced. As you know, this matrix can also be represented in the following way:

where λ _{i is} the eigenvalues, e _i the eigenvectors of R. By ordering the eigenvalues by size, one obtains a series of multiple eigenvalues that represent the channel noise and, as a rule, N + M eigenvalues that belong to the clutter. An example of this is shown in Fig. 3 for a linear antenna with N _x sensors and M transmit pulses. It can be seen as a law that, neglecting clutter fluctuations, the number of cluster eigenvalues is less than or equal to the sum of the number of sensors and echo pulses: L N + M. The minimum number results for N = M , ie L 2 N. The prerequisite here is that the pulse repetition frequency is not smaller than the Nyquist frequency of the clutter echoes.

Since the Clackereigenwerte usually stand out clearly from the noise, you can divide R into a clutter and a noise component:

The two parts of the matrix are unitary to one another, so that the left part represents the vector part space of the clutter, the right one the one of the noise. The invention is based on the fact that the remaining L -1 columns of T are filled with the L -1 eigenvectors e _i ^c which belong to the cluster eigenvalues, ie the L -1 largest eigenvalues. The auxiliary channels thus formed just stretch the clutter vector space, so that in principle all clutter is detected, the matrix T, however, causes a significant reduction in the order of the subsequent signal processing. With this transformation matrix T and the equations (2a) and (3), the principle of the adaptive processor for suppressing moving clutter is given. However, it is still necessary to perform the inversion of the matrices R _T (m) , which cause the clutter suppression and are not known. It should be constantly adapted to the constantly changing clutter conditions.

Usually you ask all possible target Doppler frequencies at once set direction simultaneously, so that for the calculation rule (3)

where m denotes a specific target Doppler frequency. This also means that M matrices of the form

T * _m R T _m (m = 1 ... M)

must be inverted. T _m is the one transformation, containing the m th Doppler filter.

The inversion of the matrices R _T (m) can be performed very effectively taking into account that all the L x L matrices R _T (m) have a common (L -1) x (L -1) sub-matrix, viz the covariance matrix of the auxiliary signals. So the matrices have the shape

With

a cluttering power in the search signal
b Vector of the cross-correlation values between search and auxiliary signals
D covariance matrix of the auxiliary signals

Only a and b also depend on the frequency m .

We assume that D ^{-1 is} known. Then you can specify the inverse matrix as follows:

in which

Δ = D - 1 / a b b *. If D is known, Δ ^{-1 is obtained} as follows

The algorithm thus comprises the following steps:

a) Treasures D , furthermore a and b for all target Doppler frequencies m = 1. , , M
b) Calculate D ^-1
c) Compute Δ _m ^-1 for all m = 1. , , M (E.13).
d) Calculate R ^-1 for all m = 1. , , M (Eq.12).

Step b requires (L -1) ³ multiplications
Step c requires approx. L ² M multiplications
Step d requires approximately L ² M multiplications.

Overall, there is something about (L -1) ³ + 2 L ² M multiplications.

However, this effort can be significantly reduced again by applying the invention - eigenvectors of a clutter covariance matrix as auxiliary channels for the interference liberation. The use of cluster eigenvectors in the matrix R now has the advantage that each one of the eigenvectors enters with the weight given by the associated eigenvalue, in other words, there are "important" and "less important" cluster eigenvectors. In turn, the cluster's own spectrum depends on the nature of the reflective background. With increasing directivity, the number of large own values decreases. One can therefore come to a less expensive (calculating time, hardware effort) method in which one saves some of the eigenvalues belonging to smaller eigenvalues. Inverting the matrices R _T (m) costs several N ³ operations, so that e.g. As a halving of the dimension (L / 2 × L / 2 instead of L × L) brings a computational savings by a factor of 8. He according to the invention this is made use of. In a radar receiver operating with auxiliary channels with eigenvectors, a reduced covariance matrix based on the L -1 first, ie the eigenvectors belonging to the largest eigenvalues, is used. It has been found that a reduction of the auxiliary channels in the occupation with eigenvectors ie the reduction of the number of degrees of freedom L compared to N + M up to half or up to a third still lead to very good results, ie not noticeable loss of profit , This is due to the fact that in the Jacobian algorithm used for the eigenvalue decomposition, the eigenvalues are sorted by size and, when trimming the vector space stretched by the cluster eigenvectors, the small eigenvalues are automatically affected. FIG. 4 shows the dependence of the gain on SCR (signal / clutter ratio) on the number of clutter filter degrees of freedom L at constant N and M.

Plotted is the gain in signal / clutter ratio over the target speed normalized to the unique speed range. At the place of the clutter adjustment, the corresponding clutter element is interpreted as a target (here at v / v _max = 0.4). The clutter spectrum represents a projection of a spatial clutter distribution into the Doppler frequency range. This distribution was measured with the ELRA system and used here to obtain the most realistic possible ratios. As can be seen, with a reduction in the number of degrees of freedom from L = N + M = 24 to 12, there is no loss of profit.

In the following Tables 1 to 3 is a cost comparison of the required arithmetic operations between optimal detector method, auxiliary channel method with L = N + M degrees of freedom and half-dimension L / 2 reduced auxiliary channel method with eigenvectors is provided. The auxiliary channel methods are referred to as sub-optimal methods.

The cost of lobe formation and Doppler filtering is about the same in all cases. In suboptimal cases, these operations are performed before clutter suppression, in the best case afterwards. The cost of estimating R _t is much lower than for the large matrix R ; this point is disregarded. The effort for filter calculation, ie the required complex multiplies for the auxiliary channel method and the optimal method are compared. It is based on that N = M and thus L = 2 N. Then the numbers of multiplications listed in Table 1 result for different values of N for the optimal method. For the method according to the invention with L -1 auxiliary channels from the eigenvectors, the considerably smaller number of arithmetic operations listed in Table 2 already result. This reduced number of arithmetic operations permits the realization of the two-dimensional clutter suppression in the first place. A further significant reduction of the arithmetic operations by the factor is fiction, according to possible by the auxiliary channel method is operated with a reduction in the degrees of freedom in half, as is apparent from the analogous calculation example according to Table 3. If one needs the matrix inversion for L = 32 62 559 arithmetic operations in the simple auxiliary channel method, the effort for reducing the degrees of freedom is reduced by half to 7 471 arithmetic operations.

Table 1

optimal detector

Table 2

suboptimal detector with L -1 auxiliary channels at L = M + N

Table 3

suboptimal detector with L -1 auxiliary channels for eigenvectors and reduction of degrees of freedom

The eigenvector method requires, in principle, that in order to determine the auxiliary channels in the matrix T, the eigenvector decomposition of R would have to be performed real-time. This, however, makes the process more complicated again. According to the invention it is therefore proposed that the eigenvector decomposition of a matrix is made off-line and so he held eigenvectors in the signal processor as clutter Hilfskanäle be introduced. This model matrix can, for. B. obtained by prolonged averaging in a particular azimuth and elevation expanded area and brought up to date from time to time. Another possibility proposed according to the invention consists in calculating such a model matrix by calculation. According to the invention, the case of an omni-directional background is provided here, this contains all possible azimuth and Doppler frequency components in the same way and thus covers all possible inhomogeneous cases.

The reduced covariance matrices R _T (m) of equation (3) can be estimated by averaging over an azimuth and distance extended clutter area. A processor according to (3) is therefore stoop-adaptive. Corresponding embodiments of the invention are the features of claims 3 and 4 can be removed.

An even simpler receiver adapted only once can be obtained by fixing the matrix for omnidirectional clutter transformed according to (2) for R _T (m) in equation (3). The method is slightly less powerful than the previously described during the radar operation adaptive method, since even with highly directive clutter situation always the entire clutter range is hidden in azimuth and Doppler frequency. Furthermore, unbalanced influences such as sensor coupling and multipath propagation are not taken into account as in an adaptive system. Nonetheless, the non-adaptive method during radar operation is of interest in practice because of its simplicity.

FIG. 5 shows an example of the same clutter distribution as in FIG. 4. As can be seen, for L = 12 and even L = 18 there are again optimal profit curves, which are even more favorable than for L = 24. This can be explained by the fact that a too large number of clutter channels already the target signal vector space is clipped by the clutter filter. As a model for calculating the eigenvectors, a covariance matrix for omnidirectional clutter was used.

Fig. 6 shows the comparison with conventional methods, ie antenna lobe formation and temporal clutter filter in series. One recognizes large losses over almost half the Doppler frequency range.

Insofar as curves are shown in the illustrative FIGS. 1-6, these are based on the following data and parameters:

The invention will be described below with reference to exemplary embodiments explained in more detail.

Fig. 7 shows an overview of the two methods for moving-clutter suppression, which were explained above.

Fig. 7a shows the structure as it is derived directly from the LR test (pre-whiten and match) and is known as the optimal detector. The received data are freed from the clutter in the space-time domain (x, y, t) . This is followed by a search channel formation (lobe in the direction φ _s , Doppler filter for speed v _s ).

Fig. 7b shows a block diagram of the auxiliary channel method (match and post whitening). The receive data is a search channel ( φ _s , v _s ) and several clutter channels (possibly several sets φ _c , v _c ) supplied. This is followed by a release of the search signal from the clutter.

The principle of the new radar receiver is shown in FIG. 8. The output signals of the N sensors pass through the usual coherent receiving channels (amplifiers, mixers, AD converters) and are then stored in the echo memory according to sensor, distance and delay. By forming a weighted sum over the complex input signals (b ₁, b ₂, ..., b _N ,), a search lobe is formed with a lobe-forming network for all echo delays (and range elements). The echo values of the search lobe are transformed by means of a Doppler filter bank DFB into the Doppler frequency range. The Doppler filterbank (weighted sums on the echo delays) performs a spectral analysis of the received signals, ie, the M values of the search lobe are each examined together for the individual Doppler frequencies, ie velocities. Together with the search lobe, the outputs of the Doppler filter bank M are search channels. Lobe formation and Doppler filter bank are well-known in radar equipment and Vor gangs.

The output signals of the receive channels are also supplied to the sub-channels from a matrix of L -1 to eigenvectors belonging to the largest eigenvalues. These each have the dimension NM and therefore have an effect both in the sensor and in the echo pulse dimension. The L -1 outputs of the auxiliary channels are each linked to one output of the Doppler filter bank and together form the reduced signal vector space in which the subsequent signal processing and evaluation takes place in an L × M matrix. The signal vector x is multiplied by the first column of R _T ^-1 (m) (clutter suppression). This happens separately for each of the M frequencies. Of the M signals of the remaining frequency channels, the largest in terms of size is selected and brought to threshold decision and display.

In particular, covariance matrices R _T (m) of dimension L × L between each output of DFB and all auxiliary channels are estimated (covariance estimate KS) . In this case, the (L -1) × (L -1) covariance sub-matrix of the auxiliary channels need only be estimated once; further M vectors of dimension L contain the cross-correlation values between the M search channels and the L -1 sub-channels and the services of the search channels. The estimation has to be done in azimuth and distance extended clutter areas, in order to avoid influences of individual target echoes.

In a special processor (FB filter calculation), the M matrices are inverted. The first columns R _t ^-1 (m) e represent the searched M ratings, ie the clutter filter.

The maximum of the squares of the filtered data becomes compared with a threshold and when exceeded to off evaluation (display, fine location, tracklogic etc.).

The filtering section BN operates such that each search channel 1. , , M (ie, the M outputs of the Doppler filter bank DFB in Fig. 8) is guided via filter coefficients to a summing element. The summers are also connected at other inputs to all L -1 subchannels via further filter coefficients. The effect of the filter coefficients is that the auxiliary channels are used to estimate the clutter signal in the respective search channel and to subtract it from the current search channel signal. At the outputs, sorted by Doppler frequencies, the echo signals released by the clutter appear. The evaluation network implements the calculation rule x _T R _T ^-1 (m) e in equation (3a).

Of the output signals of the evaluation network BN , as shown in FIG. 8, the squares of squares are formed by means of squaring units. With the aid of a comparison circuit MAX , the maximum of the M signals is searched for, fed to a threshold detector and finally to the display.

In a covariance estimator, the output signals 1. , , M of the Doppler filter bank DFB and the L -1 auxiliary channels calculates the covariance matrix D of the auxiliary channels according to the equation (11) and the cross-correlation vectors b and the a for all M target Doppler frequencies (m = 1 ... M) . The correlation estimates can be calculated by temporal as well as by spatial averaging over several distance elements.

The correlation values determined in the covariance estimator are fed to a circuit for filter calculation FB . In the circuit, according to the algorithms given by Equations (12) and (13), the filter coefficients, ie the first columns of R _T ^-1 (m) , are calculated. These are fed to the rating network. The adaptation speed, ie the speed with which the filter coefficients are recalculated, can be independent of the actual filtering process, ie the operation performed by the evaluation network. It depends on the speed with which the clutter characteristics change as a result of platform movement.

In the case of the non-adaptive method, the blocks "covariance estimation" and "matrix inversion" are saved. On the other hand, in the block "Filtering", the coefficients of the first columns of the precomputed inverse matrices R _T ^-1 (m) for omnidirectional clutter are stored.

From the estimated covariance matrix R _T (m) , an eigenvector decomposition can be performed parallel to the adaptation and filtering process, which updates the auxiliary channels in longer time periods. Alternatively, completely off-line computed eigenvectors, eg. As such a matrix for omnidirectional clutter, are kept in a memory and used as auxiliary channels.

The receiver thus contains three parts that are in different Run off speed:

• - The evaluation of the data with search and auxiliary channel coefficients and the filtering must be strictly in real time.
• - The adaptation of the filter (covariance estimation, matrixinver sion) must be able to follow changes in the clutter environment, but can generally be slower than the data filtering.
• - The auxiliary channels are determined either by external data adjusted or also to the changing environment customized. This adjustment can be even slower than that To proceed with adaptation.

A realized according to the invention radar receiver discovered moving radar targets (aircraft, driving witness) from clutter background from a moving platform out. The receiver exercises the following functions in principle out:

• - spatial and temporal target signal integration (clubs education, Doppler filter bank)
• - estimate of the clumping properties
• - Calculation of a clutter filter
• - Clutter suppression
• - Discovery and advertisement.

The receiver is adaptive, d. H. he can differ Liche Clutterverhältnisse, z. B. when flying over various Automatically adjust landscape shapes.

As a result of the adaptive method, interference suppression also works when the radar is stationary, against moving (clouds) and stationary mischievous (ground) clutter. The receiver is equal to the optimal LR detector (gain in signal-to-clutter ratio). In comparison with the optimal detector, a receiver according to the invention simplifies and reduces the complexity by orders of magnitude. As a result, a realization becomes possible at all.

Since the receiver is a modification of the optimal detector he concludes the principle of motion compensation implicitly with one.

For side viewing with omnidirectional transmitter, linear Receiving group and before homogeneous Clutterverteilung be the Clutterechos right-left-symmetrical. This will be the Clutter signals, filter calculation and filtering are real. This results in a computer saving by a factor of 4. This also applies to pure Querabbetrieb with direktiver transmitting antenna.

The radar receiver according to the invention for moving radars has the following characteristics:

• - Adaptivity
• - optimal motion compensation
• - optimal Clutterunterdrückung with reduced effort  
• Any number M of echo pulses can be used for clutter filtering.

With the invention described here, the computational effort be significantly reduced for adaptation and filtering, so that a realization for not too big antennas with today's Technology and using modern algorithms (systolic arrays) is possible.

Abbreviations and terms

Abbreviations and terms β _n spatial phase position e unit vector φ azimuth angle φ _B line of sight L Order of the transformed system λ , η discovery thresholds m temporal index M Number of echo pulses n Vector of the clutter echoes n _T transformed clutter vector n sensor Index N Number of sensors R Clutterkovarianzmatrix R _T T * R T R _T Estimated value of R _T S Vector of complex target signal components S _T transformed target signal vector T Transformation matrix to reduce the order of the system R elevation m Delaying the m-th echo pulse v relative radial velocity between clutter and radar v _Max Plate shape speed x Vector of the complex received signals x _T transformed received signal vector x _n sensor coordinate y _n sensor coordinate CNR Clutter / noise ratio SNR Signal / noise ratio a Clutter performance in the search channel b Vector of cross-correlation values between search and auxiliary channels D Covariance matrix of the auxiliary channels f _D Doppler frequency L Number of eigenvectors used as auxiliary channels

Claims (5)

1. A method for two-dimensional, ie in space and time, effective seed suppression of clutter echoes in echo signals to detect moving targets with a radar receiver of a moving radar, with a phased array antenna, sent over the M transmit pulses and M echo signals over N antenna elements of the array antenna and from the radar receiver amplifying, down-mixing and digitizing the echo signals received from each of the N antenna elements in receive channels to N * M input signals consisting of N signal sequences of length M , and subsequently from the N signal sequences under specification of certain weighting factors (b ₁. ... b _N ) a lobe formation takes place, by which a single signal sequence of length M is generated from the N signal sequences, which signal is fed to a double filter bank, which supplies a search signal at its outputs, characterized

• a) that L -1 auxiliary signals by scalar multiplying the vector representing the N · M input signals with those L -1 eigenvectors of the (N · M) × (N · M) covariance matrix, which are the largest L -1 eigenvalues associated formed using as a (N × M) × (N × M) covariance matrix either a precomputed or one of the N × M input signals,
• b) from an m-th search signal and from the L -1 auxiliary signals a reduced covariance matrix R _T (m) of dimension L is formed and its inverse _T ^-1 R (m) is calculated,
• c) that the vector x _T representing the m th search signal and the L -1 auxiliary signals is scalar multiplied by the first column of the inverse reduced covariance matrix R _T ^-1 (m) ,
• d) that the method steps b and c are performed for each search signal, so that a set of output signals whose number is equal to the number of search signals,
• e) that of the set of output signals, the largest absolute square is selected and compared with a detection threshold and displayed when they are exceeded.

2. The method according to claim 1, characterized in that in the determination of the reduced covariance matrices R _T (m) from the L -1 auxiliary signals and one of the m- th search signals the all reduced L × L covariance matrices R _T (m) common (L -1) × (L -1) - Subcovariance matrix of the auxiliary signals determined only once, z. B. estimated, and is used for all M search signals. 3. The method according to claim 1 or 2, characterized in that for the purpose of adaptation to changing Clutterverhältnisse the reduced Kovarizmatrizen R _T (m) from the L -1 auxiliary signals and one of the m- th search signals during the radar operation of received echo signals are estimated. 4. The method according to any one of claims 1 to 3, characterized in that for the purpose of adaptation to changing Clutterverhältnisse the reduced covariance matrices R _T (m) from the L -1 auxiliary signals and one of the m- th search signals are estimated from echo signals from a be received in azimuth and distance extensive area. 5. Radar receiver with adaptive two-dimensional suppression of clutter echoes in echo signals of a moving coherent radar to detect moving targets with a phased array with N antenna elements to receive the M echo signals of transmitted M pulses, with N receiving channels connected to the N antenna elements coherent receivers with echo latches, in which the received echo signals are amplified, down-converted and digitized into N × M input signals consisting of N signal sequences of length M such that the N receive channels are fed to a lobe-forming network connected in series with a Doppler filter bank, in which the N Signal sequences under specification of certain weighting factors (b 1 .. b _N ) a lobe formation takes place, is generated by the N signal sequences a single signal sequence of length M, which is fed to the Doppler filter bank, which at their M outputs each one Search signal, characterized in that L -1 are provided at L N + M auxiliary channels whose inputs are the N × M input signals and L -1 eigenvectors of the (N × M) × (N × M) covariance matrix having the largest L -1 eigenvalues, are supplied, and on the output side L -1 auxiliary signals are obtained by scalar multiplying the vectors of the N × M input signals and the L -1 largest eigenvalues; that the M outputs of the Doppler filter bank and the L -1 outputs of the auxiliary channels are fed to a rating network which executes the computation instruction x * _T R _T ^-1 (m) e , where x _{T is} the transformed vectors of the input signals R _T (m) the m th reduced covariance matrix formed from the L -1 auxiliary signals and from a respective m-th search signal, m is a specific target Doppler frequency, and e represents a unit vector; in that a device for estimating the reduced covariance matrix R _T (m) is provided, to which the L -1 auxiliary signals and the m th search signals are supplied on the input side and whose outputs are fed to a filter calculation device for determining the inverse reduced covariance matrix R _T ^-1 (m) previously obtained by estimation R _T (m) are fed and the outputs of the filter calculation means are supplied to the evaluation network, in accordance with the calculation rule of the m -th search signal and the L -1 auxiliary signals representing vector x _T with the first column of the inverse reduced covariance matrix R _T ^-1 (m) is scalar multiplied, and the evaluation network is followed by a threshold detector for evaluating the filtered data supplied by the evaluation network or the maximum whose squares are squares.