CN115469265B - A method for azimuth estimation based on joint processing of acoustic vector array - Google Patents

Abstract

The present invention discloses a method for azimuth estimation of joint processing of acoustic vector arrays, establishes an acoustic vector array output signal model, constructs a covariance matrix _RV of joint processing of sound pressure and velocity according to the model, decomposes the covariance matrix _RV into an observation coefficient matrix and a residual covariance matrix R _uv , separates the observation azimuth from the cross-covariance matrix of sound pressure and velocity, and then reconstructs the Hermitian covariance matrix by singular value decomposition of the residual covariance matrix, and finally uses the reconstructed covariance matrix to implement spatial spectrum estimation to obtain an estimated azimuth. The present invention avoids the problem that certain azimuth signals are filtered out or weakened due to the fixed observation azimuth, and does not need to scan the observation azimuth, thereby solving the problem of high algorithm complexity, and enhances the anti-noise ability of acoustic vector array processing by singular value decomposition and reconstruction of the cross-covariance matrix of sound pressure and velocity, thereby improving the multi-target resolution and azimuth estimation performance under low signal-to-noise ratio conditions.

Description

A method for azimuth estimation based on joint processing of acoustic vector array

Technical Field

The invention belongs to the field of underwater acoustic array signal processing, and relates to an acoustic vector array joint processing azimuth estimation method, in particular to an acoustic vector array sound pressure and velocity joint processing azimuth estimation method based on covariance matrix decomposition.

Background technique

The acoustic vector sensor is usually composed of a sound pressure sensor and a particle velocity sensor, which can synchronously pick up the sound pressure and particle velocity information in the sound field at the same point in space. Since the perceived sound field information is more comprehensive, the signal processing methods of the acoustic vector sensor array are more diverse, which can basically be divided into two categories: based on the Nehorai processing framework and sound pressure and velocity joint processing.

The Nehorai processing framework for acoustic vector arrays was proposed by Arye Nehorai in 1994 (Arye Nehorai, Eytan Paldi. Acoustic vector-sensor array processing. [J]. IEEE Trans. Signal Processing, 1994, 42 (9)). The idea is to treat the velocity component output by the acoustic vector sensor as information independent of the sound pressure component and process it. Although this processing method enables the spatial spectrum estimation method based on the sound pressure array to be extended to the acoustic vector array, the performance improvement it brings is still not enough to meet the target direction estimation requirements in underwater low signal-to-noise ratio environments.

The core idea of the joint processing method of sound pressure and velocity is to make full use of the spatial correlation characteristics of sound pressure and velocity to suppress noise. Bai Xingyu et al. (Bai Xingyu, Jiang Yu, Zhao Chunhui. Detection of source number and azimuth estimation of acoustic vector array based on joint processing of sound pressure and velocity [J]. Acta Acoustics, 2008(01):56-61.) achieved high-resolution detection and orientation of long-range targets by constructing the cross-covariance matrix of sound pressure and velocity; Yao Zhixiang expanded the concept of the cross-covariance matrix of sound pressure and velocity (Yao Zhixiang, Hu Jinhua, Yao Dongming. An acoustic vector array azimuth estimation algorithm based on multiple signal classification method [J]. Acta Acoustics (Chinese Edition), 2008(04):305-309.), and used the combined directional gain of the vector sensor to further reduce the signal-to-noise ratio threshold of the azimuth estimation, thereby achieving better performance in terms of multi-target resolution and resolution probability. However, in the process of constructing the sound pressure-velocity cross-covariance matrix, the velocity components need to be projected onto a certain observation direction to obtain the combined velocity in that direction. The sound pressure-velocity joint processing method itself has a certain spatial filtering capability. Specifying the observation direction as a fixed value may cause some direction signals to be filtered out or weakened, thereby reducing the output signal-to-noise ratio or missing the source. If the observation direction is scanned with the spatial spectrum search direction, the cross-covariance matrix will change, which will dramatically increase the complexity of the algorithm.

Summary of the invention

In view of the above-mentioned prior art, the technical problem to be solved by the present invention is to provide a method for azimuth estimation of sound pressure and velocity joint processing of acoustic vector array based on covariance matrix decomposition, which separates the observation azimuth from the sound pressure and velocity cross-covariance matrix, avoids the problem of certain azimuth signals being filtered out or weakened due to the fixed observation azimuth, and eliminates the need to scan the observation azimuth, thereby solving the problem of excessive algorithm complexity.

To solve the above technical problems, the present invention provides an acoustic vector array joint processing azimuth estimation method, which establishes an acoustic vector array output signal model, constructs a covariance matrix _RV for sound pressure and velocity joint processing according to the model, decomposes the covariance matrix _RV into an observation coefficient matrix and a residual covariance matrix _Ruv , separates the observation azimuth from the sound pressure and velocity cross-covariance matrix, and then reconstructs the Hermitian covariance matrix by singular value decomposition of the residual covariance matrix, and finally uses the reconstructed covariance matrix to implement spatial spectrum estimation to obtain an estimated azimuth.

Furthermore, a method for azimuth estimation by joint processing of acoustic vector arrays of the present invention comprises the following steps:

Step 1: Establish an output signal model of an acoustic vector array with M array elements of arbitrary geometric shape, establish an x, y reference rectangular coordinate system with the leftmost array element as the origin, and the coordinates of the mth array element are ( _xm , _ym ), and obtain the acoustic vector array sound pressure channel output vector p(n) and vibration velocity x, y channel output vectors _vx (n), _vy (n):

where s(n)=[ _s1 (n),…, _sK (n)] ^T represents the source vector, K is the number of sources, the symbol T represents the transpose operation, A(θ) is the sound pressure array manifold matrix, A(θ)=[a( _θ1 ),a( _θ2 ),…,a( _θK )], _θk is _the azimuth of the kth source, the sound pressure array steering vector a( _θk )=[1,exp( _-j2πf0τ2 ),…,exp _{( -j2πf0τM )} ] ^T , _τm =( _xmcosθk + _ymsinθk )/C, C is the _speed of sound, _f0 is the center frequency of the source, _Φvx =diag[cos( _θ1 ),…,cos( _θK )], _Φvy =diag[sin( _θ1 ),…,sin( _θK) ] )] are the coefficient matrices of the velocity x and y channels respectively, n _p (n), n _vx (n), n _vy (n) are the background noise vectors of the sound pressure and velocity x and y channels respectively;

Step 2: Combine the velocity x and y channel output vectors v _x (n), v _y (n) to obtain the combined velocity at the observation orientation θ _r :

v _c (n) = cos (θ _r ) v _x (n) + sin (θ _r ) v _y (n),

Construct the covariance matrix R _V of the sound pressure and vibration velocity joint processing. R _V is obtained by the (p+v _c )v _c joint processing combination method. E{·} represents the expectation operation;

Step 3, decompose the covariance matrix _RV into the observation coefficient matrix _Tv (θ) of _vc (n), the observation coefficient matrix _Tu (θ) of p(n)+ _vc (n) and the residual covariance matrix _Ruv ;

Step 4, perform singular value decomposition on the remaining covariance matrix _Ruv to obtain a diagonal matrix Λ composed of non-zero singular values, and left singular vectors U and right singular vectors V composed of columns corresponding to Λ, _Ruv = ^UΛVH , and select U or V to construct a new covariance matrix _RC = ^UΛUH or _RC = ^VΛVH ;

Step 5: A new steering vector a _C (θ) is generated by the observation coefficient matrix T (θ) and the steering vector a (θ), a _C (θ) = ^TH (θ)a (θ), wherein, when _RC = UΛU ^H , T(θ) = T _u (θ), and when _RC = VΛV ^H , T(θ) = T _v (θ);

Step 6: Implement a spatial spectrum estimation method using the covariance matrix _RC and the steering vector _aC (θ) to obtain an estimated orientation.

Furthermore, step 3 decomposes the covariance matrix _RV into the observation coefficient matrix _Tv (θ), _Tu (θ) and the residual covariance matrix _Ruv as follows:

Step 3-1, set the observation direction θ _r as the spatial spectrum search direction θ, and extract the observation coefficient matrix T _v (θ) in v _c (n):

In the formula, υ(θ)＝[cos(θ),sin(θ)] ^T , υ(θ) is the array manifold of the single velocity sensor;

Step 3-2, extract the observation coefficient matrix T _u (θ) of the p(n)+v _c (n) item in the covariance matrix:

In the formula, u(θ)＝[1，cos(θ),sin(θ)] ^T , u(θ) is the array manifold of a single vector sensor;

Step 3-3, decompose the sound pressure and vibration velocity joint processing covariance matrix _RV into the form of multiplying the observation coefficient matrix and the residual covariance matrix _Ruv :

Where R _uv is the residual covariance matrix, specifically:

Furthermore, the spatial spectrum estimation method is a CBF method, an MVDR method or a MUSIC method.

Furthermore, the spatial spectrum estimation method is the MVDR method, and the expression of the spatial spectrum _PC (θ) is:

Furthermore, the covariance matrix _RV is obtained by the pv _c joint processing combination method, and the covariance matrix _RV is decomposed into the observation coefficient matrix _Tv (θ) of _vc (n) and the residual covariance matrix _Ruv .

Furthermore, the covariance matrix _RV is obtained by the p(p+ _vc ) joint processing combination method, and the covariance matrix _RV is decomposed into the observation coefficient matrix T _u (θ) of p(n)+v _c (n) and the residual covariance matrix R _uv .

Furthermore, the covariance matrix _RV is obtained by the (p+ _vc ) ² joint processing combination method, and the covariance matrix _RV is decomposed into the observation coefficient matrix T _u (θ) of p(n)+ _vc (n), the observation coefficient matrix T _u (θ) of p(n)+ _vc (n) and the residual covariance matrix R _uv .

Beneficial effects of the present invention: The present invention proposes a new method for joint processing of sound pressure and velocity by acoustic vector array: In view of the contradiction between the azimuth estimation performance and the algorithm calculation amount caused by the selection of the observation azimuth in the traditional joint processing of sound pressure and velocity, the present invention separates the observation azimuth from the sound pressure and velocity mutual covariance matrix, reconstructs the Hermitian covariance matrix by singular value decomposition of the remaining covariance matrix, and finally uses the covariance matrix to implement the spatial spectrum estimation algorithm, thereby avoiding the problem of certain azimuth signals being filtered out or weakened due to the fixed observation azimuth, and there is no need to scan the observation azimuth, thus solving the problem of excessive algorithm complexity. The present invention does not need to select the observation azimuth, and by decomposing and reconstructing the sound pressure and velocity mutual covariance matrix, the noise resistance of the acoustic vector array processing is enhanced, thereby improving the multi-target resolution and azimuth estimation performance under low signal-to-noise ratio conditions.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 Algorithm flow chart;

Figure 2 Acoustic vector uniform linear array model;

Figure 3(a) is a spatial spectrum comparison diagram for SNR = 0dB;

Figure 3(b) is a spatial spectrum comparison diagram for SNR = -5dB;

Fig. 4 Curve of target resolution probability changing with signal-to-noise ratio;

Figure 5. Root mean square error versus signal-to-noise ratio curve.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

The present invention aims at the contradiction in the traditional sound pressure and vibration velocity joint processing method that the fixed observation orientation may cause some orientation signals to be weakened or filtered out due to spatial filtering, while the observation orientation scanning greatly increases the complexity of the algorithm. The present invention separates the observation orientation from the sound pressure and vibration velocity cross-covariance matrix, reconstructs the Hermitian covariance matrix by singular value decomposition of the remaining covariance matrix, and finally uses the covariance matrix to implement the spatial spectrum estimation algorithm.

Embodiment 1:

The technical solution adopted by the present invention to solve the technical problem comprises the following steps:

Step 1: Establish an output signal model of an acoustic vector array with M array elements of arbitrary geometric shape, establish an x,y reference rectangular coordinate system with the leftmost array element as the origin, and the coordinates of the mth array element are ( _xm , _ym ), and obtain the acoustic vector array sound pressure channel output vector p(n) and vibration velocity x, y channel output vectors _vx (n), _vy (n):

where s(n)=[ _s1 (n),…, _sK (n)] ^T represents the source vector, K is the number of sources, the symbol T represents the transpose operation, A(θ) is the sound pressure array manifold matrix, A(θ)=[a( _θ1 ),a( _θ2 ),…,a( _θK )], _θk is _the azimuth of the kth source, the sound pressure array steering vector a( _θk )=[1,exp( _-j2πf0τ2 ),…,exp ₍ -j2πf0τM ₎ ] ^T , _τm =( _xmcosθk + _{ymsinθk )} /C, C is the _speed of sound, _f0 is the center frequency of the source, _Φvx =diag[cos( _θ1 ),…,cos( _θK )], _Φvy =diag[sin( _θ1 ),…,sin( _θK) ] )] are the coefficient matrices of the velocity x and y channels respectively, n _p (n), n _vx (n), n _vy (n) are the background noise vectors of the sound pressure and velocity x and y channels respectively;

Step 2: Combine the velocity x and y channel output vectors v _x (n), v _y (n) to obtain the combined velocity v _c (n) at the observation orientation θ _r .

v _c (n) = cos (θ _r ) v _x (n) + sin (θ _r ) v _y (n),

And construct the covariance matrix _RV of the sound pressure and vibration velocity joint processing. _RV can be obtained by the joint processing combinations including but not limited to _pvc , p(p+ _vc ), (p+ _vc ) _vc and (p+ _vc ) ² . Taking the (p+ _vc ) _vc combination as an example, E{·} represents the expectation operation;

Step 3, decompose the covariance matrix _RV into the observation coefficient matrix and the residual covariance matrix _Ruv ;

Step 4, perform singular value decomposition on the remaining covariance matrix _Ruv to obtain a diagonal matrix Λ composed of non-zero singular values, and a left singular vector U and a right singular vector V composed of columns corresponding to Λ, _Ruv = ^UΛVH , and select U or V to construct a new covariance matrix _RC = ^UΛUH or _RC = ^VΛVH ;

Step 5: The new steering vector a _C (θ) is generated by the observation coefficient matrix T (θ) and the steering vector a (θ), and a _C (θ) = ^TH (θ) a (θ), where when _RC = UΛU ^H , T (θ) = T _u (θ), and when _RC = VΛV ^H , T (θ) = T _v (θ);

Step 6: Implement a spatial spectrum estimation method using the covariance matrix R _C and the steering vector a _C (θ), wherein the spatial spectrum estimation method includes but is not limited to CBF, MVDR and MUSIC methods, etc., wherein the spatial spectrum P _C (θ) of the MVDR method is expressed as

Step 3 of the present invention, the covariance matrix _RV decomposition specifically comprises the following steps:

Step 3-1, set the observation direction _θr as the spatial spectrum search direction θ, and extract the observation coefficient matrix _Tv (θ) in _vc (n)

In the formula, υ(θ)＝[cos(θ),sin(θ)] ^T , υ(θ) is the array manifold of the single velocity sensor;

Step 3-2, extract the observation coefficient matrix T _u (θ) of the p(n)+v _c (n) item in the covariance matrix

In the formula, u(θ)＝[1，cos(θ),sin(θ)] ^T , u(θ) is the array manifold of a single vector sensor;

Step 3-3, decompose the sound pressure and vibration velocity joint processing covariance matrix _RV into the form of multiplying the observation coefficient matrix and the residual covariance matrix _Ruv , where the (p+ _vc ) _vc covariance matrix _RV can be decomposed into

Where _Ruv is the residual covariance matrix,

Embodiment 2:

In conjunction with Figure 1, the present invention includes the following steps:

Step 1, taking the acoustic vector uniform linear array as an example, establish the array output signal model, as shown in Figure 2, consider the two-dimensional isotropic noise field, the array takes the first array element position as the reference point, M array elements are evenly spaced along the positive direction of the y axis, the spacing between adjacent array elements is d, there are K independent far-field equal-power narrow-band sources, the incident angle is θ _k , k = 1, 2, ..., K, defined as the angle between the source and the positive direction of the x axis, then the sound pressure channel output vector p (n) and the vibration velocity x, y channel output vectors v _x (n), v _y (n) are:

Where s(n) = [s ₁ (n), …, s _K (n)] ^T represents the source vector; A(θ) is the sound pressure array manifold matrix,

A(θ)=[a(θ ₁ ), a(θ ₂ ),…, a(θ _K )] (2)

Where a(θ _k ) is the steering vector of the kth signal source on the sound pressure sensor array.

a(θ _k )=[1,exp(-j2πd sinθ _k /λ),…, exp(-j(M-1)2πd sinθ _k /λ)| ^T (3)

Wherein, λ represents the incident wavelength of the information source, the symbol T represents the transposition operation, Φ _vx =diag[cos(θ ₁ ), …, cos(θ _K )], Φ _vy =diag[sin(θ ₁ ), …, sin(θ _K )] are the coefficient matrices of the vibration velocity x and y channels respectively, n _p (n), n _vx (n), n _vy (n) are the background noise vectors of the sound pressure and vibration velocity x and y channels respectively;

Step 2: Combine the velocity x and y channel output vectors v _x (n), v _y (n) to obtain the combined velocity v _c (n) at the observation orientation θ _r .

v _c (n) = cos (θ _r ) v _x (n) + sin (θ _r ) v _y (n) (4)

Construct a covariance matrix _RV of the sound pressure and vibration velocity joint processing. _RV can be obtained by a combination of joint processing including but not limited to _pvc , p(p+ _vc ), (p+ _vc ) _vc and (p+ _vc ) ^2. Taking the combination of (p+ _vc ) _vc as an example,

In the formula, E{·} represents the expected operation;

Step 3, decompose the covariance matrix _RV into the observation coefficient matrix and the residual covariance matrix _Ruv . The decomposition steps of the covariance matrix _RV of the (p+ _vc ) _vc combination form are:

Step 3-1, set the observation direction _θr as the spatial spectrum search direction θ, and extract the observation coefficient matrix _Tv (θ) in _vc (n)

In the formula, υ(θ)＝[cos(θ),sin(θ)] ^T , υ(θ) is the array manifold of the single velocity sensor;

Step 3-2, extract the observation coefficient matrix T _u (θ) of the p(n)+v _c (n) item in the covariance matrix

In the formula, u(θ)＝[1，cos(θ),sin(θ)] ^T , u(θ) is the array manifold of a single vector sensor;

Step 3-3, decompose the sound pressure and vibration velocity joint processing covariance matrix _RV into the form of multiplying the observation coefficient matrix and the residual covariance matrix _Ruv ,

Where _Ruv is the residual covariance matrix,

Step 4: Perform singular value decomposition on the residual covariance matrix _Ruv .

R _uv ＝U _u Λ _uv V ^H (10)

Wherein Λ _uv is the singular value matrix of R _uv , which can be written as Λ _uv =[Λ, 0 _2M×M ] ^T , assuming λ _m is the mth non-zero singular value, Λ = diag(λ ₁ , ..., λ _2M ), and λ ₁ ≥λ ₂ ≥ … ≥λ _2M ; V is the right singular vector of R _uv , U _u is the left singular vector of R _uv , and the left singular vector U of the column corresponding to Λ is,

U＝U _u [I _2M×2M , 0 _2M×M ] ^T (11)

Where 0 _2M×M is a 2M×M dimensional zero matrix, I _2M×2M is a 2M×2M dimensional unit matrix, and Λ and singular vectors U or V are selected to construct a new covariance matrix

R _C =UΛU ^H or R _C =VΛV ^H (12)

Step 5: The new steering vector a _C (θ) is generated by the corresponding observation coefficient matrix T (θ) and the steering vector a (θ)

a _C (θ) = ^TH (θ) a (θ) (13)

Wherein, when _RC = UΛU ^H , T(θ) = T _u (θ), when _RC = VΛV ^H , T(θ) = T _v (θ);

Step 6: Implement a spatial spectrum estimation method using the covariance matrix _RC and the steering vector _aC (θ). The spatial spectrum estimation method includes but is not limited to CBF, MVDR and MUSIC methods. Taking the MVDR method as an example, the spatial spectrum _PC (θ) is expressed as

The first K maximum values of the spatial spectrum _PC (θ) are taken as the estimated azimuth of the far-field target.

The above describes the specific implementation methods of each part of the invention content, and the following analyzes the simulation example.

Assuming that the noise field is isotropic, consider an equally spaced linear array of acoustic vectors with 8 array elements, the array element spacing is half a wavelength, there are two unrelated far-field equal-power narrowband signals, the incident angles are -4° and 5° respectively, assuming that the noise is stationary Gaussian white noise, the number of snapshots is set to 1000, and the number of Monte Carlo tests is 100. The (p+v _c )v _c combination is selected to construct the cross-covariance matrix, and the performance of the conventional method based on the Nehorai processing framework, the traditional sound pressure and vibration velocity joint processing method in which the observation azimuth is set to 0° and 50° respectively, and the three cases of scanning the azimuth with spectrum search, and the method of the present invention are compared respectively. Among them, the method of the present invention selects the left singular value vector to construct a new cross-covariance matrix.

Figure 3 (a) and Figure 3 (b) are spatial spectrum comparison diagrams of various methods under the conditions of signal-to-noise ratio SNR = 0dB and SNR = -5dB, respectively. It can be found that the present invention not only has a lower spatial spectrum background level and a sharper target spectrum peak than other methods, but also under the condition of a signal-to-noise ratio of -5dB, when the two target spectrum peaks of other methods are overlapped and difficult to distinguish, the method of the present invention can still effectively estimate the directions of the two targets.

Figure 4 is a curve showing the target resolution probability of each method as a function of the signal-to-noise ratio. It can be found that under low signal-to-noise ratio conditions, the target resolution success probability of the method of the present invention is higher than that of other methods, and the target resolution success probability can reach 1 at a lower signal-to-noise ratio threshold.

Figure 5 shows the root mean square error (RMSE) curves of each method as the signal-to-noise ratio changes. It can be found that when the signal-to-noise ratio is above -6dB, as the target resolution probability increases, the RMSE of each method gradually decreases, among which the performance of the method of the present invention is the best. When the signal-to-noise ratio is greater than -4dB, the RMSE of the method of the present invention is less than 0.5°. To achieve a RMSE similar to that of the method of this paper when SNR = -4dB, the signal-to-noise ratio required by other methods is greater than 0dB.

Claims (5)

1. A method for azimuth estimation of joint processing of acoustic vector arrays, characterized in that: a model of acoustic vector array output signals is established, a covariance matrix R _V of joint processing of sound pressure and velocity is constructed according to the model, the covariance matrix R _V is decomposed into an observation coefficient matrix and a residual covariance matrix R _uv , the observation azimuth is separated from the cross-covariance matrix of sound pressure and velocity, and then a Hermitian covariance matrix is reconstructed by singular value decomposition of the residual covariance matrix, and finally the spatial spectrum estimation is performed using the reconstructed covariance matrix to obtain an estimated azimuth, and the specific steps are as follows: Step 1: Establish an output signal model of an acoustic vector array with M array elements of arbitrary geometric shape, establish an x,y reference rectangular coordinate system with the leftmost array element as the origin, and the coordinates of the mth array element are ( _xm , _ym ), and obtain the acoustic vector array sound pressure channel output vector p(n) and vibration velocity x, y channel output vectors _vx (n), _vy (n): where s(n)=[ _s1 (n),…, _sK (n)] ^T represents the source vector, K is the number of sources, the symbol T represents the transpose operation, A(θ) is the sound pressure array manifold matrix, A(θ)=[a( _θ1 ),a( _θ2 ),…,a( _θK )], _θk is _the azimuth of the kth source, the sound pressure array steering vector a( _θk )=[1,exp( _-j2πf0τ2 ),…,exp _{( -j2πf0τM )} ] ^T , _τm =( _xmcosθk + _ymsinθk )/C, C is the _speed of sound, _f0 is the center frequency of the source, _Φvx =diag[cos( _θ1 ),…,cos( _θK )], _Φvy =diag[sin( _θ1 ),…,sin( _θK) ] )] are the coefficient matrices of the velocity x and y channels respectively, n _p (n), n _vx (n), n _vy (n) are the background noise vectors of the sound pressure and velocity x and y channels respectively; Step 2: Combine the velocity x and y channel output vectors v _x (n) and v _y (n) to obtain the combined velocity at the observation orientation θ _r : v _c (n) = cos (θ _r ) v _x (n) + sin (θ _r ) v _y (n), Construct the covariance matrix R _V of the sound pressure and vibration velocity joint processing. R _V is obtained by the (p+v _c )v _c joint processing combination method. E{·} represents the expectation operation; Step 3, decompose the covariance matrix _RV into the observation coefficient matrix _Tv (θ) of _vc (n), the observation coefficient matrix _Tu (θ) of p(n)+ _vc (n) and the residual covariance matrix _Ruv ; Step 3-1, set the observation direction θ _r as the spatial spectrum search direction θ, and extract the observation coefficient matrix T _v (θ) in v _c (n): In the formula, υ(θ)＝[cos(θ),sin(θ)] ^T , υ(θ) is the array manifold of the single velocity sensor; Step 3-2, extract the observation coefficient matrix T _u (θ) of the p(n)+v _c (n) item in the covariance matrix: In the formula, u(θ)＝[1,cos(θ),sin(θ)] ^T , u(θ) is the array manifold of a single vector sensor; Step 3-3, decompose the sound pressure and vibration velocity joint processing covariance matrix _RV into the form of multiplying the observation coefficient matrix and the residual covariance matrix _Ruv : Where R _uv is the residual covariance matrix, specifically: Step 4, perform singular value decomposition on the remaining covariance matrix _Ruv to obtain a diagonal matrix Λ composed of non-zero singular values, and left singular vectors U and right singular vectors V composed of columns corresponding to Λ, _Ruv = ^UΛVH , and select U or V to construct a new covariance matrix _RC = ^UΛUH or _RC = ^VΛVH ; Step 5: A new steering vector a _C (θ) is generated by the observation coefficient matrix T (θ) and the steering vector a (θ), a _C (θ) = ^TH (θ)a (θ), wherein, when _RC = UΛU ^H , T(θ) = T _u (θ), and when _RC = VΛV ^H , T(θ) = T _v (θ); Step 6: Implement a spatial spectrum estimation method using the covariance matrix _RC and the steering vector _aC (θ) to obtain an estimated orientation. 2. The method for azimuth estimation using joint processing of acoustic vector arrays according to claim 1, wherein the spatial spectrum estimation method is a CBF method, a MVDR method or a MUSIC method. 3. The method for azimuth estimation by joint processing of acoustic vector array according to claim 1, characterized in that: the covariance matrix _RV is obtained by a pv _c joint processing combination method, and the covariance matrix _RV is decomposed into an observation coefficient matrix _Tv (θ) of _vc (n) and a residual covariance matrix _Ruv . 4. The method for azimuth estimation by joint processing of acoustic vector array according to claim 1, characterized in that: the covariance matrix _RV is obtained by a p(p+ _vc ) joint processing combination mode, and the covariance matrix _RV is decomposed into an observation coefficient matrix T _u (θ) of p(n)+v _c (n) and a residual covariance matrix R _uv . 5. The method for joint processing azimuth estimation of an acoustic vector array according to claim 1, characterized in that the covariance matrix _RV is obtained by a (p+ _vc ) ² joint processing combination mode, and the covariance matrix _RV is decomposed into an observation coefficient matrix T _u (θ) of p(n)+ _vc (n), an observation coefficient matrix T _u (θ) of p(n)+ _vc (n) and a residual covariance matrix R _uv .