CN104991236A - Monostatic MIMO radar non-circular signal coherent source DOA (Direction Of Arrival) estimation method - Google Patents

Abstract

The invention relates to the technical field of a radar, and specifically relates to a monostatic MIMO radar non-circular signal coherent source DOA (Direction Of Arrival) estimation method. Signal sampling data is obtained; dimension reduction processing is performed on a received signal; correctional spatial smoothing processing is performed on a data matrix after dimension reduction; characteristic decomposition is performed on a correlation matrix after smoothing; (5) a root solving polynomial is built, and angle estimation of a target is obtained. According to the method, the dimension reduction processing is performed on the received signal, so computational efficiency is greatly improved, and an influence on estimation performance can hardly be caused; the advantage of a non-circular signal is taken fully, so the utilization rate of the received data is improved, and performance of DOA estimation is improved; and, through correction for a spatial smoothing coherent solution method, the method is suitable for a non-circular signal direction measurement situation, and a DOA of a coherent signal source under the condition of the non-circular signal can be effectively estimated.

Description

A kind of MIMO radar not rounded signal coherence source, single base Wave arrival direction estimating method

Technical field

The invention belongs to Radar Technology field, be specifically related to MIMO radar not rounded signal coherence source, a kind of single base Wave arrival direction estimating method.

Background technology

In recent years, the MIMO technological thought of the communications field is incorporated into field of radar, MIMO (Multiple-InputMultiple-Output, the multiple-input and multiple-output) radar system of proposition causes and pays close attention to widely.MIMO radar makes full use of signal diversifying, spatial domain diversity gain, obtains the degree of freedom higher compared with conventional radar, and its target detection capabilities and parameter estimation capabilities etc. all obtain to be approved widely.Due to the plurality of advantages that MIMO radar is potential, the research and development about MIMO radar is quick, and wherein direction of arrival (Direction Of Arrival the is called for short DOA) estimation problem of MIMO radar is emphasis research topic wherein.DOA estimation method based on traditional phased-array radar utilizes round signal configuration to receive data matrix often, do not consider not rounded signal, but in practical application, the use of the not rounded signals such as BPSK, AM is more and more extensive, therefore, the characteristic making full use of not rounded signal carries out the important topic that DOA estimation is Estimation of Spatial Spectrum theory.In addition, not often independently between the signal source run in actual environment, but there is certain correlativity, or even relevant, this will cause the Subspace Decomposition class methods of widespread use, as MUSIC (multiple signal classification) method and ESPRIT (based on rotational invariance technology) method etc., and performance severe exacerbation even complete failure, so the DOA estimation in coherent signal source is also an important topic in Mutual coupling field.

DOA based on MIMO radar is estimated, general approach is transplanted at classic method, as the MUSIC method of subspace class being applied to the two-dimentional MUSIC method, dimensionality reduction MUSIC method etc. of MIMO radar, but these class methods often calculated amount are larger, and be not suitable for the method adopting not rounded signal direction-finding, when there is coherent, performance sharply worsens, and is difficult to realize effective DOA and estimates.

DOA for not rounded signal estimates, be generally by matrix-expand, then zygote spatial class method carries out DOA estimation.Utilize in the direction-finding method of not rounded signal numerous, NC (the not rounded)-MUSIC method that Pascal Charg é equals calendar year 2001 proposition has certain representativeness.The method, by expanding reception data matrix, obtains DOA in conjunction with Root-MUSIC method and estimates.But when there is coherent, the performance of NC-MUSIC method sharply worsens, and estimated accuracy and success ratio are starkly lower than the situation that only there is independent source.

DOA for coherent estimates, general employing space smoothing class methods and matrix reconstruction class methods wherein, take front-rear space smooth as representative.But when using not rounded signal to carry out DOA estimation, directly applying front-rear space smooth and will destroy original subspace structure, thus reliable DOA estimated result cannot be obtained.Therefore, how to make full use of the characteristic of not rounded signal, carry out estimating it is technical matters urgently to be resolved hurrily based on the coherent DOA of single base MIMO radar.

Summary of the invention

The object of the invention is to propose a kind of MIMO radar not rounded signal coherence source, single base Wave arrival direction estimating method adopting not rounded signal to carry out coherent Mutual coupling.

The object of the present invention is achieved like this:

A kind of MIMO radar not rounded signal coherence source, single base Wave arrival direction estimating method comprises:

(1) signal sample data is obtained:

X (l)=AS (l)+N (l), wherein, X (l)=[x ₁(l), x ₂(l) ..., x _mN(l)] ^ttie up for MN × 1 when fast umber of beats is l and receive data vector, M and N is respectively emission array and receiving array array number, A is array manifold matrix, and S (l) is for tieing up narrow band signal vector in K × 1, and signal type is not rounded signal, N (l) is for tieing up additive noise vector in MN × 1, noise type is white complex gaussian noise, l=1 ..., L $A=[a_r\left({\mathrm{\θ}}_1\right)\⊗a_t\left({\mathrm{\θ}}_1\right),a_r\left({\mathrm{\θ}}_2\right)\⊗a_t\left({\mathrm{\θ}}_2\right),...,a_r\left({\mathrm{\θ}}_K\right)\⊗a_t\left({\mathrm{\θ}}_K\right)],a_r\left({\mathrm{\θ}}_k\right)=[1,z_k,z_k^2,...,z_k^{N-1}{]}^T$ For receiving steering vector, for launching steering vector, symbol represent Kronecker product, θ _krepresent the incident angle of a kth information source, z _k=exp (-j π sin (θ _k)), k=1,2 ..., K, for the multiple initial phase of a kth signal, S ₀t () is for tieing up real vector in K × 1;

(2) dimension-reduction treatment is carried out to received signal:

X _rD(l)=W ^-1g ^hx (l)=BS (l)+W ^-1g ^hn (l), wherein, X _rDl () is (the M+N-1) × 1 dimension data vector after dimensionality reduction conversion, W and G is dimensionality reduction transition matrix, and B is that (the M+N-1) × K after dimensionality reduction conversion ties up array manifold matrix, B=[b (θ ₁), b (θ ₂) ..., b (θ _k)], for the steering vector after dimensionality reduction conversion,

$G=\left[\begin{array}{c}{G}_1\\ G_2\\ .\\ .\\ .\\ G_N\end{array}\right]\∈R^{\mathrm{MN}\×(M+N-1)},$ Wherein, G _n=[O ₁i _mo ₂] ∈ R ^{m × (M+N-1)}, n=1,2 ..., N, O ₁and O ₂be respectively the null matrix that M × (n-1) and M × (N-n) ties up, I _mfor M ties up unit matrix,

(3) to the space smoothing process that the data matrix after dimensionality reduction is revised:

r _fbfor to data correlation matrix R _sub2 × (M+N-P) that the space smoothing process carrying out revising obtains ties up square formation, and wherein P is the number of times of space smoothing, $Y_p=\left[\begin{array}{c}{X}_{\mathrm{RDp}}\\ X_{\mathrm{RDp}}^{*}\end{array}\right],$ X _rDpby X _rDthe p capable element capable to p+M+N-1-P composition, J is the switching matrix that 2 × (M+N-P) ties up, and the element on its counter-diagonal is 1, and the element on other positions is 0;

(4) feature decomposition is carried out to the correlation matrix after level and smooth:

To R _fbcarry out feature decomposition, obtain the noise subspace U of 2 × (M+N-P)-2 × K proper vector composition corresponding to the individual less eigenwert of 2 × (M+N-P)-2 × K;

(5) construct rooting polynomial expression, obtain the angle estimation of target:

Noise subspace U is divided into the submatrix that upper and lower two structures are identical, $U=\left[\begin{array}{c}{U}_1\\ U_2\end{array}\right],$ Utilize U ₁and U ₂structure rooting polynomial expression $P\left(z\right)=D_1^2-D_2{D}_3,$ Wherein $D_1=b^H\left(z\right)U_1{U}_1^{H}b\left(z\right),D_2=b^H\left(z\right)U_1{U}_2^H{b}^{*}\left(z\right),D_3=b^T\left(z\right)U_2{U}_1^{H}b\left(z\right),$ B (z)=[1, z, z ²..., z ^m+N-2] ^t, z=exp (-j π sin (θ)), angle () represents the phase place of getting plural number, obtains root of polynomial, and wherein K estimates the root of closest unit circle and the DOA of corresponding target.

Beneficial effect of the present invention is:

Method of the present invention carries out dimension-reduction treatment to received signal, counting yield is improved greatly, and impacts estimated performance hardly; Make full use of the characteristic of not rounded signal, thus improve the utilization factor to receiving data, improve the performance that DOA estimates; By revising the decorrelation LMS method of space smoothing, making it be applicable to the situation of not rounded signal direction-finding, effectively can estimate the direction of arrival of coherent under not rounded RST.

Accompanying drawing explanation

Fig. 1 is method schematic diagram of the present invention.

Fig. 2 this method (NC-Root-MUSIC) and the change contrast simulation figure of Root-MUSIC method DOA estimated result square error with signal to noise ratio (S/N ratio).

Fig. 3 is that this method (NC-Root-MUSIC) is estimated the change contrast simulation figure of power with signal to noise ratio (S/N ratio) with Root-MUSIC method DOA.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described further.

The method comprises the following steps:

Step (10), dimension-reduction treatment is carried out to the signal data that described receiving array receives, obtain the signal data X after dimension-reduction treatment _rD(l);

Step (20), to data vector X _rDcarry out the space smoothing process revised, obtain the correlation matrix R smoothly _fb;

Step (30), to R _fbcarry out feature decomposition, obtain noise subspace, and construct rooting polynomial expression, the DOA obtaining information source estimates.

Wherein, step (10) comprising:

The data that step (110), all matched filters of acquisition receiving array export, structure dimensionality reduction transition matrix G, and G constructs W, wherein, $G=\left[\begin{array}{c}{G}_1\\ G_2\\ .\\ .\\ .\\ G_N\end{array}\right]\∈R^{\mathrm{MN}\×(M+N-1)},$ G _n=[O ₁i _mo ₂] ∈ R ^{m × (M+N-1)}, n=1,2 ..., N, O ₁and O ₂be respectively the null matrix that M × (n-1) and M × (N-n) ties up, I _mfor M ties up unit matrix,

Step (120), utilize above-mentioned two data transition matrixes to reception data process, obtain the data X after dimensionality reduction _rD(l).

Wherein, step (20) comprising:

Step (210), utilize X _rDthe p capable element capable to p+M+N-1-P composition X _rDp; Then X is utilized _rDpcomposition Data expansion matrix $Y_p=\left[\begin{array}{c}{X}_{\mathrm{RDp}}\\ X_{\mathrm{RDp}}^{*}\end{array}\right],$ Calculate the correlation matrix of extended matrix

Step (220), sum-average arithmetic is carried out to all data correlation matrix the switching matrix J that structure 2 × (M+N-P) ties up, the element on its counter-diagonal is 1, and the element on other positions is 0, then utilizes R _subthe space smoothing matrix revised is calculated with J $R_{\mathrm{fb}}=R_{\mathrm{sub}}+{\mathrm{JR}}_{\mathrm{sub}}^{H}J.$

Wherein, step (30) comprising:

Step (310), to R _fbcarry out feature decomposition, obtain the noise subspace U be made up of 2 × (M+N-P)-2 × K proper vector that the individual little eigenwert of 2 × (M+N-P)-2 × K is corresponding;

Step (320), utilize that the front M+N-P of U and rear M+N-P is capable forms two sub-matrix U ₁and U ₂, utilize U ₁and U ₂structure operator D ₁, D ₂, D ₃, $D_1=b^H\left(z\right)U_1{U}_1^{H}b\left(z\right),D_2=b^H\left(z\right)U_1{U}_2^H{b}^{*}\left(z\right),D_3=b^T\left(z\right)U_2{U}_1^{H}b\left(z\right),$ Utilize D ₁, D ₂and D ₃structure rooting polynomial expression wherein b (z)=[1, z, z ²..., z ^m+N-2] ^t, z=exp (-j π sin (θ)), angle () represents the phase place of getting plural number, obtains root of polynomial, and wherein K estimates the root of closest unit circle and the DOA of corresponding target.

When there is correlativity between each information source, the order of signal covariance matrix has loss in various degree, thus reduces the performance of DOA estimation.Space smoothing process is a kind of effective ways signal being carried out to decorrelation LMS.Its basic thought is that the array manifold structure of each submatrix is identical by even linear array being divided into several overlapped submatrix, and carry out sum-average arithmetic computing to the covariance matrix of each submatrix, the order of signal covariance matrix is restored, thus realizes decorrelation LMS.Front-rear space smooth is except forward direction space smoothing, conjugation is carried out to each submatrix and oppositely reconstitutes a subarray, then respectively sum-average arithmetic process is carried out to the covariance matrix of the submatrix through front-rear space smooth process, form total covariance matrix, realize decorrelation LMS.

Traditional space smoothing algorithm is smoothing to whole signal covariance matrix, but when utilizing not rounded signal to carry out direction finding, the structure of covariance matrix changes, and therefore, this method is applied directly in the covariance matrix process of not rounded signal data and can not obtains correct DOA estimation.Array partition is multiple submatrix by the present invention, the data of each submatrix are utilized to expand respectively, calculate data covariance matrix, carry out sum-average arithmetic, then the space smoothing process of correction is adopted, while make use of not rounded characteristics of signals, ensure that the structure that new covariance matrix is identical with former covariance matrix, thus realize decorrelation LMS.

Describe the present invention below in conjunction with accompanying drawing.

As shown in Figure 1, first method of the present invention carries out dimension-reduction treatment to the reception data vector of single base MIMO radar receiving array, then the space smoothing process that the characteristic of not rounded signal is revised the data matrix after dimensionality reduction is utilized, obtain data correlation matrix, feature decomposition is carried out to it, thus obtains noise subspace, finally utilize noise subspace to construct rooting polynomial expression, obtain root of polynomial, the DOA obtaining coherent estimates.

The present invention is based on single base MIMO radar, its array structure adopts even linear array, and emission array array number is M, and receiving array array number is N, and array element distance is half-wavelength.Suppose in same investigative range, have K arrowband far field objects, its incident angle is respectively θ ₁, θ ₂..., θ _k, then can construct array manifold matrix:

$A=[a_r\left({\mathrm{\θ}}_1\right)\⊗a_t\left({\mathrm{\θ}}_1\right),a_r\left({\mathrm{\θ}}_2\right)\⊗a_t\left({\mathrm{\θ}}_2\right),...,a_r\left({\mathrm{\θ}}_K\right)\⊗a_t\left({\mathrm{\θ}}_K\right)]=[a\left({\mathrm{\θ}}_1\right),a\left({\mathrm{\θ}}_2\right),...,a\left({\mathrm{\θ}}_K\right)]$

Wherein, $a_r\left({\mathrm{\θ}}_k\right)=[1,z_k,z_k^2,...,z_k^{N-1}{]}^T$ For receiving steering vector, $a_t\left({\mathrm{\θ}}_k\right)=[1,z_k,z_k^2,...,z_k^{M-1}{]}^T$ For launching steering vector, symbol represent that Kronecker (Kronecker) amasss, z _k=exp (-j π sin (θ _k)), k=1,2 ..., K, θ _krepresent the incident angle of a kth information source.

Arrowband not rounded signal phasor is tieed up in K × 1

S(l)＝[s ₁(l),s ₂(l),...,s _K(l)] ^T＝ΦS ₀(l)

Wherein, for the multiple initial phase of a kth signal, S ₀l () is for tieing up real vector in K × 1.

When then fast umber of beats is l, the output data vector accepting all matched filters of array is

X(l)＝AS(l)+N(l)

Wherein, X (l)=[x ₁(l), x ₂(l) ..., x _mN(l)] ^tfor MN × 1 dimension data vector, N (l)=[n ₁(l), n ₂(l) ..., n _mN(l)] ^tfor additive noise vector is tieed up in MN × 1, noise type is white complex gaussian noise, l=1 ..., L.

Fig. 2 is completely relevant at signal, and emission array array number is 8, and receiving array array number is 6, and array element distance is half-wavelength, and fast umber of beats is 1280, and information source number is 2.

Fig. 3 is completely relevant at signal, and emission array array number is 8, and receiving array array number is 6, and array element distance is half-wavelength, and fast umber of beats is 1280, and information source number is 2.

In above-mentioned data vector, there is more redundant information, and do not make full use of the feature of not rounded signal, therefore, if directly utilize it to carry out direction finding not only can cause unnecessary computation burden, and have lost a part of useful information.First the present invention carries out dimension-reduction treatment to above-mentioned data vector, obtains the data vector after dimensionality reduction, then utilizes the characteristic of not rounded signal to expand and space smoothing data, while guarantee counting yield, takes full advantage of the characteristic of not rounded signal.

Structure dimensionality reduction transition matrix:

$G=\left[\begin{array}{c}{G}_1\\ G_2\\ .\\ .\\ .\\ G_N\end{array}\right]\∈R^{\mathrm{MN}\×(M+N-1)}$

Wherein, G _n=[O ₁i _mo ₂] ∈ R ^{m × (M+N-1)}, n=1,2 ..., N, O ₁and O ₂for the null matrix that dimension is M × (n-1) and M × (N-n), I _mfor M ties up unit matrix.

Data vector then after dimensionality reduction is

X _RD(l)＝W ^-1G ^HX(l)＝BS(l)+W ^-1G ^HN(l)

Wherein, B=[b (θ ₁), b (θ ₂) ..., b (θ _k)]=W ^-1g ^ha, for the steering vector after dimensionality reduction conversion.

Through dimension-reduction treatment, former data vector is tieed up by MN × 1 and is converted to (M+N-1) × 1 dimension, eliminates a large amount of redundant informations.Utilize the space smoothing process that not rounded characteristics of signals is revised the data vector after dimensionality reduction:

$Y_p=\left[\begin{array}{c}{X}_{\mathrm{RDp}}\\ X_{\mathrm{RDp}}^{*}\end{array}\right]$

Wherein, X _rDpfor by X _rDp capable to p+M+N-1-P row element composition virtual submatrix data, calculate Y _pcorrelation matrix and sum-average arithmetic is carried out to all P correlation matrix obtain the level and smooth covariance matrix of forward direction then obtain the covariance matrix after smoothing processing:

$R_{\mathrm{fb}}=R_{\mathrm{sub}}+{\mathrm{JR}}_{\mathrm{sub}}^{H}J$

Wherein, J is the switching matrix that 2 × (M+N-P) ties up, and the element on its counter-diagonal is 1, and the element on other positions is 0.

To the signal covariance matrix R after smoothing processing _fbcarry out Eigenvalues Decomposition, then can obtain the noise subspace U be made up of 2 × (M+N-P)-2 × K proper vector that 2 × (M+N-P)-2 × K less eigenwert is corresponding, utilize that the front M+N-P of U and rear M+N-P is capable forms two sub-matrix U ₁and U ₂:

$U=\left[\begin{array}{c}{U}_1\\ U_2\end{array}\right],$

Utilize U ₁and U ₂structure operator D ₁, D ₂, D ₃:

$D_1=b^H\left(z\right)U_1{U}_1^{H}b\left(z\right)$

$D_2=b^H\left(z\right)U_1{U}_2^H{b}^{*}\left(z\right)$

$D_3=b^T\left(z\right)U_2{U}_1^{H}b\left(z\right)$

Wherein, b (z)=[1, z, z ²..., z ^m+N-2] ^t, z=exp (-j π sin (θ)).

Utilize D ₁, D ₂and D ₃rooting polynomial expression:

$P\left(z\right)=D_1^2-D_2{D}_3$

Obtain root of polynomial, choose wherein K and, to the root z of closest unit circle, z is calculated, angle () represents the phase place of getting plural number, and the DOA namely obtaining corresponding target estimates.

Compared with existing DOA estimation method, adopt MIMO radar not rounded signal coherence source, list base of the present invention Wave arrival direction estimating method, the calculated amount of single base MIMO radar direction finding can be reduced.The present invention is adopted the structure of front-rear space smooth method according to the covariance matrix of not rounded signal data to be revised, combine front-rear space smooth and not rounded signal direction-finding, effectively reduce the correlativity between Received signal strength, realize the effective estimation to the coherent angle of arrival.In sum, the present invention is based on single base MIMO radar, not rounded signal can be utilized effectively to estimate the direction of arrival of coherent.

Claims (1)

1. MIMO radar not rounded signal coherence source, a single base Wave arrival direction estimating method, is characterized in that:

(1) signal sample data is obtained:

X (l)=AS (l)+N (l), wherein, X (l)=[x ₁(l), x ₂(l) ..., x _mN(l)] ^ttie up for MN × 1 when fast umber of beats is l and receive data vector, M and N is respectively emission array and receiving array array number, A is array manifold matrix, and S (l) is for tieing up narrow band signal vector in K × 1, and signal type is not rounded signal, N (l) is for tieing up additive noise vector in MN × 1, noise type is white complex gaussian noise, l=1 ..., L $A=[a_r\left({\mathrm{\θ}}_1\right)\⊗a_t\left({\mathrm{\θ}}_1\right),a_r\left({\mathrm{\θ}}_2\right)\⊗a_t\left({\mathrm{\θ}}_2\right),...,a_r\left({\mathrm{\θ}}_K\right)\⊗a_t\left({\mathrm{\θ}}_K\right)],$ $a_r\left({\mathrm{\θ}}_k\right)={[1,z_k,z_k^2,...,z_k^{N-1}]}^T$ For receiving steering vector, for launching steering vector, symbol represent Kronecker product, θ _krepresent the incident angle of a kth information source, z _k=exp (-j π sin (θ _k)), k=1,2 ..., K, for the multiple initial phase of a kth signal, S ₀t () is for tieing up real vector in K × 1;

(2) dimension-reduction treatment is carried out to received signal:

X _rD(l)=W ^-1g ^hx (l)=BS (l)+W ^-1g ^hn (l), wherein, X _rDl () is (the M+N-1) × 1 dimension data vector after dimensionality reduction conversion, W and G is dimensionality reduction transition matrix, and B is that (the M+N-1) × K after dimensionality reduction conversion ties up array manifold matrix, B=[b (θ ₁), b (θ ₂) ..., b (θ _k)], for the steering vector after dimensionality reduction conversion,

$G=\left[\begin{array}{c}{G}_1\\ G_2\\ \·\\ \·\\ \·\\ G_N\end{array}\right]\∈R^{\mathrm{MN}\×(M+N-1)},$ Wherein, G _n=[O ₁i _mo ₂] ∈ R ^{m × (M+N-1)}, n=1,2 ..., N, O ₁and O ₂be respectively the null matrix that M × (n-1) and M × (N-n) ties up, I _mfor M ties up unit matrix,

(3) to the space smoothing process that the data matrix after dimensionality reduction is revised:

r _fbfor to data correlation matrix R _sub2 × (M+N-P) that the space smoothing process carrying out revising obtains ties up square formation, and wherein P is the number of times of space smoothing, x _rDpby X _rDthe p capable element capable to p+M+N-1-P composition, J is the switching matrix that 2 × (M+N-P) ties up, and the element on its counter-diagonal is 1, and the element on other positions is 0;

(4) feature decomposition is carried out to the correlation matrix after level and smooth:

To R _fbcarry out feature decomposition, obtain the noise subspace U of 2 × (M+N-P)-2 × K proper vector composition corresponding to the individual less eigenwert of 2 × (M+N-P)-2 × K;

(5) construct rooting polynomial expression, obtain the angle estimation of target:

Noise subspace U is divided into the submatrix that upper and lower two structures are identical, $U=\left[\begin{array}{c}{U}_1\\ U_2\end{array}\right],$ Utilize U ₁and U ₂structure rooting polynomial expression $P\left(z\right)=D_1^2-D_2{D}_3,$ Wherein $D_1=b^H\left(z\right)U_1{U}_1^{H}b\left(z\right),D_2=b^H\left(z\right)U_1{U}_2^H{b}^{*}\left(z\right),D_3=b^T\left(z\right)U_2{U}_1^{H}b\left(z\right),$ $b\left(z\right)={[1,z,z^2,...,z^{M+N-2}]}^T,$ z＝exp(-jπsin(θ))， $\mathrm{\θ}=-\arcsin \left(\frac{\mathrm{angle}\left(z\right)}{\mathrm{\π}}\right),$ Angle () represents the phase place of getting plural number, obtains root of polynomial, and wherein K estimates the root of closest unit circle and the DOA of corresponding target.