CN102621535B - High-efficiency method for estimating covariance matrix structures - Google Patents

Abstract

The invention discloses a high-efficiency method for estimating covariance matrix structures, which belongs to the field of radar signal processing and mainly realizes estimating covariance matrix structures in adaptive detection of radar targets on the background with non-Gaussian clusters. The high-efficiency method aims to solve the problem that a traditional sampling covariance matrix and a traditional normalized sampling covariance matrix cannot lead an adaptive detector to a complete CFAR (constant false alarm rate) characteristic. A sampling covariance matrix is solved after pretreatment by means of dividing real-part data from virtual-part data, an obtained initiated matrix realizes the complete CFAR characteristic for clusters, a real part and a virtual part of auxiliary data are sufficiently utilized for iteration, accordingly, computation complexity in an iteration process is reduced, estimation precision is improved beneficially, and the high-efficiency method for estimating covariance matrix structures has a popularization and application value in adaptive CFAR detectors for radar targets.

Description

A kind of efficient covariance matrix structure method of estimation

One, technical field

The present invention is under the jurisdiction of radar target self-adapting detecting field, is specifically related to a kind of efficient covariance matrix structure method of estimation.

Two, background technology

It is an important content of Radar Targets'Detection that the self-adaption constant false alarm rate of radar target (constant false alarm ratio, CFAR) detects.For single array element Radar Targets'Detection, only need to be detected unit clutter power level by estimation, the self-adaptation CFAR that gets final product realize target detects.Yet, when radar is many array element radar, by the accumulation of the coherent to each array element echoed signal, can effectively improve target detection probability, but also need to estimate more clutter parameter simultaneously.

Under the Gaussian Clutter background, sample covariance matrix SCM (sample covariance matrix) is maximal possibility estimation (the maximum likelihood estimate of covariance matrix, MLE), directly utilize SCM to replace the actual value of covariance matrix can obtain the self-adaptation CFAR detecting device of radar target.Yet, in high resolving power or low incident angle situation, radar clutter more shows as non-Gauss's form, at this moment, Gauss model is the guinea pig clutter no longer accurately.

For non-gaussian clutter, usually adopt the constant random vector of ball (spherically invariant random vector, SIR V) to carry out modeling, as classical K Distribution Clutter, Weibull distribution clutter etc. can be used the SIRV modeling.SIRV is comprised of the texture component that means the clutter power level and the Gaussian random vector product that means the clutter speckle component, it has been generally acknowledged that, and the texture component value difference of different distance unit, but there is identical speckle component covariance matrix.The covariance square of speckle component is also referred to as the covariance matrix structure of clutter.Under this non-gaussian clutter background with the SIRV modeling, the people such as Conte E suppose that clutter covariance matrix is known, normalized matched filter (normalized matched Filter has been proposed, NMF), and further point out, only need to utilize the suitable estimated value of covariance matrix to replace the covariance matrix in NMF, can obtain corresponding adaptive detector ANMF (adaptive NMF).Because ANMF detecting device itself has unchangeability to the clutter power level, therefore, only need to be estimated the clutter covariance matrix structure.Under the non-gaussian clutter background in this SIRV modeling, the maximal possibility estimation of clutter covariance matrix structure is difficult to obtain, and therefore, the estimation problem of covariance matrix structure becomes the key that solves radar target self-adapting detecting under this clutter background.

A kind of method is to adopt the estimated value of SCM as the covariance matrix structure, and its adaptive detector obtained is called SCM-ANMF.SCM-ANMF has the CFAR characteristic to the clutter covariance matrix structure, but can't guarantee the CFAR characteristic to the clutter texture component, and therefore, SCM-ANMF is not complete CFAR detecting device.During another kind method, on each range unit clutter, adopt normalized method to eliminate the impact of texture component, adopt normalization SCM (normalized SCM, NSCM), as the estimation of covariance matrix structure, the adaptive detector obtained is called NSCM-ANMF.Although NSCM-ANMF has guaranteed the CFAR characteristic to the clutter texture component, but can't guarantee the CFAR characteristic to the clutter covariance matrix structure, therefore, SCM-ANMF is the CFAR detecting device fully.From the CFAR characteristic angle of adaptive detector, consider, SCM and NSCM are not effective methods of estimation.

Gini F has proposed the iterative estimate method of a kind of NSCM of utilization as the initialization matrix, referred to as NSCM-RE.NSCM-RE is under the limited number of time iteration, and corresponding ANMF does not have complete CFAR characteristic from theory, but NSCM-RE is through after iteration repeatedly, and corresponding ANMF has approximate CFAR characteristic.NSCM-RE is widely used in the self-adapting detecting of non-gaussian clutter at present.But NSCM-RE need to just can make corresponding adaptive detector have approximate CFAR characteristic through iteration repeatedly, the obvious like this computation complexity that increased.

Three, summary of the invention

1. the technical matters that will solve

The objective of the invention is provides a kind of efficient clutter covariance matrix structure method of estimation for many array element radar target self-adaptation CFAR under the non-gaussian clutter background detects, and the technical problem underlying that wherein will solve comprises:

(1) initialization of covariance matrix structure is estimated, the adaptive detector ANMF that makes estimated matrix be updated to obtain after the NMF detecting device has the CFAR characteristic to clutter covariance matrix structure and clutter texture component;

(2) efficient iterative estimate, make iteration have higher estimated accuracy later, and have lower computation complexity.

2. technical scheme

Efficient covariance matrix structure method of estimation of the present invention comprises following technical measures: at first, utilize the real part (I channel data) of auxiliary data to carry out the data pre-service divided by first yuan of number of imaginary part (Q channel data), again pretreated data are calculated to SCM, then, utilize the imaginary part (Q channel data) of auxiliary data to carry out the data pre-service divided by first yuan of number of real part (I channel data), again pretreated data are calculated to SCM, again to obtain the initialization estimated matrix of covariance matrix structure after two SCM additions that obtain divided by the auxiliary unit number, finally, utilize the initialization estimated matrix to carry out the estimated value that iteration obtains the covariance matrix structure.

In technique scheme, because auxiliary data real part and imaginary part comprise identical texture component value, utilize the real part of auxiliary data divided by first yuan of number of imaginary part, can effectively eliminate the impact that the texture texture is estimated the covariance matrix structure on the one hand; Can not introduce the correlativity between speckle component on the other hand, thereby guarantee that ANMF corresponding to initial estimation matrix obtained has the CFAR characteristic to clutter covariance matrix structure and clutter texture component.

In technique scheme, complex operation is equivalent to four real arithmetics due to one time, with NSCM-RE, compares, and the present invention only need to carry out real arithmetic, not only can reduce the calculated amount of half, also contributes to improve estimated accuracy.

3. beneficial effect

The present invention with NSCM-RE, compare have advantages of as follows:

(1) adaptive detector ANMF corresponding to the method has the CFAR characteristic to clutter covariance matrix structure and clutter texture component;

(2) the method has higher estimated accuracy than NSCM-RE under identical auxiliary data and identical iterations;

(3) the method computation complexity under identical auxiliary data and identical iterations is lower than NSCM-RE;

Four, accompanying drawing explanation

The auxiliary picture in picture 1 of instructions is adaptive detector ANMF block scheme corresponding to the present invention, and Fig. 2 is the process flow diagram that covariance matrix structure of the present invention is estimated.In Fig. 1, the function of device 1 is realized by Fig. 2, and install 2 calculating detection statistic, installing 3 is decision devices.

In Fig. 2, device 1 and device 2 are data pre-processor, and installing 3 is sample covariance matrix counters with installing 4, and installing 5 is iterators.

Five, embodiment

Fig. 1 is adaptive detector ANMF block scheme corresponding to the present invention.At first, utilize auxiliary unit data z _t(t=1,2 ..., K) by device 1, estimate clutter covariance matrix structure estimated value

The detected cell data z of recycling ₀Calculate by device 2 and calculate detection statistic λ,

$\mathrm{\λ}=\frac{{\left|p^H{\widehat{\mathrm{\Σ}}}^{-1}{z}_0\right|}^2}{\left(p^H{\widehat{\mathrm{\Σ}}}^{-1}p\right)\left(z_0^H{\widehat{\mathrm{\Σ}}}^{-1}{z}_0\right)}---\left(1\right)$

In formula, p is known direction vector.Finally, by device 3, to detection statistic λ and by the corresponding detection threshold T of given false-alarm probability, compare, differentiate target and whether exist, if λ >=T adjudicates as target exists, otherwise the judgement target does not exist.

Below in conjunction with Figure of description 2 (being the device 2 in Fig. 1), the present invention is described in further detail.With reference to Figure of description 2, the specific embodiment of the present invention is divided into following step:

Step 1: determine auxiliary range unit data z _t, t=1 ..., K, wherein, $z_t^{\left(1\right)}={[z_t^{\left(1\right)}\left(1\right),z_t^{\left(1\right)}\left(2\right),...,z_t^{\left(1\right)}\left(N\right)]}^T$ Expression is from the echo data of I passage, $z_t^{\left(2\right)}={[z_t^{\left(2\right)}\left(1\right),z_t^{\left(2\right)}\left(2\right),...,z_t^{\left(2\right)}\left(N\right)]}^T$ Expression is from the echo data of Q passage, and N means to receive signal radar array number.

Step 2: by installing 1 calculating z _tThe real part data are divided by z _tFirst component of imaginary data obtains

$y_t^{\left(1\right)}=z_t^{\left(1\right)}/z_t^{\left(2\right)}\left(1\right)---\left(2\right)$

Step 3: by installing 2 calculating z _tImaginary data is divided by z _tFirst component of real part data obtains

$y_t^{\left(2\right)}=z_t^{\left(2\right)}/z_t^{\left(1\right)}\left(1\right)---\left(3\right)$

Step 4: by installing 3 calculating

Sample covariance matrix,

${\widehat{\mathrm{\Σ}}}_0^{\left(1\right)}=\underset{t=1}{\overset{K}{\mathrm{\Σ}}}{y}_t^{\left(1\right)}{y}_t^{\left(1\right)T}---\left(4\right)$

Step 5: by installing 4 calculating

Sample covariance matrix,

${\widehat{\mathrm{\Σ}}}_0^{\left(2\right)}=\underset{t=1}{\overset{K}{\mathrm{\Σ}}}{y}_t^{\left(2\right)}{y}_t^{\left(2\right)T}---\left(5\right)$

Step 6: obtain the initial estimate of covariance matrix structure after the sample covariance matrix summation that step 3 and step 4 are obtained divided by the range unit number,

${\widehat{\mathrm{\Σ}}}_{\left(0\right)}=({\widehat{\mathrm{\Σ}}}_0^{\left(1\right)}+{\widehat{\mathrm{\Σ}}}_0^{\left(2\right)})/K---\left(6\right)$

Step 7: will

Utilize iterator to carry out iteration, for the initialization estimated matrix, m=0, m=m+1 after iteration, iterative process is

${\widehat{\mathrm{\Σ}}}_{(m+1)}=\frac{N}{K}\underset{t=1}{\overset{K}{\mathrm{\Σ}}}\left(\frac{z_t^{\left(1\right)}{z}_t^{\left(1\right)T}+z_t^{\left(2\right)}{z}_t^{\left(2\right)T}}{z_t^{\left(1\right)T}{\left({\widehat{\mathrm{\Σ}}}_{\left(m\right)}\right)}^{-1}{z}_t^{\left(1\right)}+z_t^{\left(2\right)T}{\left({\widehat{\mathrm{\Σ}}}_{\left(m\right)}\right)}^{-1}{z}_t^{\left(2\right)}}\right)---\left(7\right)$

Step 8: if after iteration m=M (M for set iterations),

The covariance matrix structure estimated value obtained for the present invention, otherwise, will

As input value, utilize step 7 to continue iteration.

Claims (9)

1. an efficient covariance matrix structure method of estimation is characterized in that comprising the following steps:

Step 1: be identified for receiving signal radar array number, be detected unit and auxiliary data unit, record the echo data of each range unit;

Step 2: the data to each range unit of I passage are carried out pre-service;

Step 3: the data to each range unit of Q passage are carried out pre-service;

Step 4: the pretreated data of I passage are asked to sample covariance matrix;

Step 5: the pretreated data of Q passage are asked to sample covariance matrix;

Step 6: after the sample covariance matrix summation that step 4 and step 5 are obtained, divided by auxiliary range unit number, obtain the initialization estimated value of covariance matrix structure;

Step 7: utilize initialization estimated value and the auxiliary data of the covariance matrix structure obtained to carry out iteration, its iterative process is

In formula, N means the radar array number, and K means auxiliary range unit number, and t means the range unit numbering,

Expression is from the echo data of I passage,

Expression is from the echo data of Q passage,

Mean the covariance matrix structure estimated value after iteration m time, wherein, when m=0,

The initialization estimated value that means the covariance matrix structure;

Step 8: judge that whether iterations meets and should require, if meet, stop iteration, the estimated value of output covariance matrix structure, if do not meet, utilize estimated value that step 7 obtains to turn back to step 7 and continue iteration.

2. a kind of efficient covariance matrix structure method of estimation according to claim 1, it is characterized in that, described step 1 is specially: first be identified for receiving the radar array number N of echoed signal, adopt plural form to record the echo complex data z of each range unit corresponding to each array element of radar _t, z _t=[z _t(1), z _t(2) ..., z _t(N)] ^T, t is the range unit numbering, t=0 means detected unit, and t=1,2 ..., K means auxiliary data range unit, z _tReal part mean the data of I passage, imaginary part means the data of Q passage, z _t(n), n=1,2 ..., N means that range unit that n array element receives is numbered the echoed signal of t.

3. a kind of efficient covariance matrix structure method of estimation according to claim 1, is characterized in that, described step 2 is specially: for each auxiliary range unit t, to vectorial z _tThe real part data

All divided by

Imaginary data

(1) obtain

4. a kind of efficient covariance matrix structure method of estimation according to claim 1, is characterized in that, described step 3 is specially: for each auxiliary range unit t, to vectorial z _tImaginary data All divided by z _t(1) real part data

(1) obtain

5. a kind of efficient covariance matrix structure method of estimation according to claim 1, is characterized in that, described step 4 is specially: ask

The sample covariance matrix ∑ ⁽¹⁾.

6. a kind of efficient covariance matrix structure method of estimation according to claim 1, is characterized in that, described step 5 is specially: ask

The sample covariance matrix ∑ ⁽²⁾.

7. a kind of efficient covariance matrix structure method of estimation according to claim 1, is characterized in that, described step 6 is specially: ask ∑ ⁽¹⁾With ∑ ⁽²⁾And ∑, then ∑ is obtained to the initialization estimated matrix divided by auxiliary unit number K

8. a kind of efficient covariance matrix structure method of estimation according to claim 1, is characterized in that, described step 7 is specially: will

As initialized estimated matrix, utilize z _tReal part data and imaginary data carry out iteration, obtain the iterative estimate value

The initialization value of m is 0, and once, the value of m adds 1 to every iteration.

9. a kind of efficient covariance matrix structure method of estimation according to claim 1, is characterized in that, described step 8 is specially: the value to m is judged, if m reaches the iterations M of setting, be m=M, iteration stops, the estimated value of output covariance matrix structure

If m<M, continue step 8.

Images