Subarray adaptive spatial filtering method under non-uniform division

Abstract

The invention discloses a sub-array adaptive spatial filtering method under non-uniform division, and belongs to the technical field of array signal processing. After array elements are reasonably divided, conventional beam forming is carried out at an array element level by utilizing an FPGA, Capon adaptive beam forming is carried out at a sub-array level, and finally a weighting coefficient of the sub-array level is obtained. The FPGA implementation architecture mainly comprises: 1) an array element level steering vector calculation module; 2) calculating a dimensionality reduction transformation matrix and a steering vector dimensionality reduction module; 3) a data dimension reduction module; 4) a covariance matrix calculation module; 5) a snapshot accumulation module; 6) a matrix inversion module; and 7) a weighted vector calculation module. According to the sub-array adaptive spatial filtering method under non-uniform division provided by the invention, interference can be successfully suppressed after FPGA processing, and the consumption of hardware resources is reduced. And all operations including matrix inversion operation can be completed only by utilizing the FPGA development board while isomerism is not needed, so that the implementation difficulty of the system is reduced.

Description

Technical Field

[0001] The invention belongs to the technical field of array signal processing, in particular to a subarray adaptive spatial domain filtering method based on non-uniform division. Background Art

[0002] Phased array radar uses electrical control to adjust the direction of the radar beam in order to detect targets. This method is called electronic scanning. Broadly speaking, a phased array radar is a collection of many small traditional radars, each of which is called an array element. Similarly, the antenna array of a phased array radar is also composed of countless receiving units and radiating units. These small radars are combined to form an array antenna. Each unit is controlled by an independent phase shift switch. By controlling the phase of each antenna unit, different phase beams can be synthesized. Unlike traditional mechanical radars that rely on physical rotation for scanning, phased array radars can emit multiple beams at the same time, scanning and tracking at the same time, so they have many advantages that traditional mechanical radars cannot match.

[0003] The concept of spatial domain filtering was proposed in the digital beamforming (DBF) technology in the field of radar technology. It refers to forming a main lobe beam in a specific direction, receiving useful signals and suppressing interference signals from other directions. DBF technology is the basis for phased arrays to achieve adaptive adjustment. DBF technology is the main form of spatial domain filtering, which is called smart antenna in communications. When several signals that arrive at the same time or occupy the same frequency band are superimposed and incident, they can no longer be distinguished by traditional time domain filtering or frequency domain filtering. Since these signals are incident at different angles, it is possible to consider using the separability of the signal spatial domain to achieve spatial domain processing of the signal, which is essentially a multi-channel array signal processing system.

[0004] The electromagnetic environment in which modern radar systems are located is becoming more and more complex, and adaptive beamforming has become particularly important as an effective anti-interference means. In order to meet the performance indicators of engineering applications, such as main lobe width, signal-to-noise ratio, etc., a large array is generally required. If the echo data received by a large phased array radar containing hundreds or thousands of array elements is directly executed with a fully adaptive algorithm, a very large amount of calculation will usually be generated, which has extremely high requirements on the computing efficiency and hardware of the entire system, so it is difficult to implement in practice. Under the premise of not affecting the adaptive interference suppression effect, in order to facilitate the implementation of the real-time processing system, the array elements are usually reasonably divided according to the actual engineering situation, so that the array elements in the array can share a transceiver channel. While reducing the computing burden and saving hardware resources, the division method of the subarray can also be dynamically adjusted according to the actual situation, so that the system can optimize the beam shape and direction for different usage scenarios and target characteristics. Subarray-level beamforming is an efficient and flexible signal processing strategy that can provide better flexibility, reliability and economy while maintaining system performance. Summary of the invention

[0005] The object of the present invention is to provide a subarray adaptive spatial domain filtering method under non-uniform division.

[0006] The technical solution to achieve the purpose of the present invention is: a subarray adaptive spatial domain filtering method under non-uniform division, comprising the following steps:

[0007] Step 1: The element-level desired signal steering vector calculation module receives the angle information sent down, including azimuth angle information θ and pitch angle information The element-level desired signal steering vector is calculated using the planar array steering vector calculation method.

[0008] Step 2: After the calculation of the element-level expected signal steering vector is completed, it is sent to the expected signal steering vector dimensionality reduction module to obtain the dimensionality reduction transformation matrix T, which can divide the array into L sub-arrays. The element-level expected signal steering vector is reduced in dimension by the dimensionality reduction transformation matrix T to obtain the sub-array-level steering vector

[0009] Step 3: After the dimensionality reduction transformation matrix T is calculated, it is sent to the data dimensionality reduction module to reduce the dimension of the array input data X(t), that is, after conventional beamforming at the array element level, the sub-array level data X is obtained. sub (t). The incident signal on the front surface is composed of M unrelated interference signals and noise in space, and M<L;

[0010] Step 4: Subarray-level data X sub (t) After the calculation is completed, it is sent to the covariance matrix calculation module to obtain the sub-array level covariance matrix R X ;

[0011] Step 5: Subarray-level covariance matrix R X After the calculation is completed, it is sent to the snapshot accumulation module, and the elements of the matrix are accumulated and summed according to the number of K snapshots and the average is calculated to obtain

[0012] Step 6: Result after snapshot accumulation Send it to the matrix inversion module, perform Cholesky decomposition on it, and use the decomposed submatrix to calculate its inverse matrix

[0013] Step 7: After the matrix inversion is completed, The weighted vector calculation module is fed into the subarray-level steering vector calculated in step 2. At the subarray level, the Capon beamforming algorithm is used to calculate the subarray-level weight vector W Z And output.

[0014] Furthermore, in step 1, the planar array elements are located on the xz plane, and the number of elements is P×Q. The element-level steering vector is:

[0015]

[0016] Where d is the array element spacing, λ is the wavelength, (x i , z j ) are the coordinates of each array element relative to the reference array element, and i = 0, 1, ..., P-1, j = 0, 1, ..., Q-1.

[0017] Furthermore, the dimension reduction transformation matrix T in step 2 and step 3 satisfies:

[0018]

[0019] In the formula, each column of T0 represents a subarray, and is 1 if there is an array element at the corresponding position, otherwise it is 0. Subarray-level steering vector Element-level steering vector The relationship is: At time t, the relationship between the sub-array level sampling data and the array element level sampling data is: sub (t) = T H X(t).

[0020] Furthermore, step 4 uses the following method to calculate the subarray-level covariance matrix for a single snapshot.

[0021] R X =X sub (t)X sub (t)H

[0022] The samples received at the subarray level at time t are Xsub (t), 1≤t≤K, t is a positive integer, and K is the number of snapshots, that is, the number of sampling points.

[0023] Furthermore, step 5 uses the following method to calculate the estimate of the subarray-level covariance matrix under K snapshots:

[0024]

[0025] Furthermore, the matrix inversion method in step 6 assumes that the input matrix is ​​A and performs Cholesky decomposition (A = LDL H =CL H ), we get the lower triangular matrices C and L. By matrix operation AW = CL H W=E=CY, we can get the matrix Y. Finally, L H W = Y calculates the inverse matrix of the covariance matrix A The input matrix A in this module is the estimate of the sub-array level covariance matrix

[0026] The elements c in matrices C and L ij and l ij The calculation method is as follows:

[0027]

[0028] The inverse matrix Y of matrix C is calculated as follows:

[0029]

[0030] Finally, the inverse matrix W of A is calculated as follows:

[0031]

[0032] Furthermore, in step 7, the method of obtaining the weighted vector using the Capon beamforming algorithm at the subarray level is:

[0033] in is the subarray-level desired signal steering vector, and the solution to the optimization problem is the required subarray-level weighted vector:

[0034]

[0035] The beneficial effects of the present invention are as follows: compared with the conventional fully adaptive beamforming method, after the sub-array is reasonably divided, not only the interference suppression effect is good, but also the consumption of FPGA hardware resources is greatly reduced, the calculation speed is improved, and it is suitable for phased array radars with a large number of array elements; in addition, since there is no need to use an external DSP chip to perform heterogeneous processing on the processing board, all operations including matrix inversion operations can be completed by using only the FPGA development board, which reduces the difficulty of system implementation. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flowchart of FPGA design process.

[0037] Figure 2 It is the block diagram of the desired signal-oriented vector calculation module.

[0038] Figure 3 It is the block diagram of the expected signal-guided vector dimensionality reduction module.

[0039] Figure 4 This is the block diagram of the covariance matrix calculation module.

[0040] Figure 5 This is the block diagram of the snapshot accumulation module.

[0041] Figure 6 This is the block diagram of the matrix inversion module.

[0042] Figure 7 It is the block diagram of the weighted vector calculation module. DETAILED DESCRIPTION

[0043] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0044] Step 1: Figure 2 The block diagram of the desired signal steering vector calculation module at the array element level is shown in Figure 1. The FPGA receives the azimuth and elevation angle information θ and Then, the Cordic ip core calculates the corresponding sine and cosine values. The first level multiplication is completed by sending it to the multiplier IP core. In the second level multiplication operation, 2πd / λ is set as a parameter and multiplied by the result of the first level multiplication operation. In the third level multiplication operation, x i and z j Multiply it with the result of the second-level operation. The present invention selects a planar array and both P and Q are 24, then the result output by the adder is a column vector with a dimension of 576*1. Finally, it is serially input into the Cordic IP core to calculate the array element-level steering vector The real and imaginary parts of .

[0045] Step 2: Figure 3 The block diagram of the dimension reduction module of the desired signal steering vector. The present invention chooses to divide the array plane into 12 sub-arrays non-uniformly. Since in each column of the dimension reduction transformation matrix T (dimension 576*12), when the array element exists during the sub-array division, the position element is replaced by the steering vector, and when the array element does not exist, the position element is 0. Each column of the matrix T0 (dimension 576*12) is pre-stored serially in ROM. The matrix T0 only contains elements 0 and 1, so no multiplier is required to calculate the matrix T. When the array element-level steering vector When inputting, each column of T0 in ROM is read to determine whether the corresponding column positions of matrix T retain the steering vector. When the steering vector is reduced in dimension, 12 complex multipliers are used to multiply the corresponding elements of the matrix, and finally the accumulation can be obtained to output row by row. The dimension of the guidance vector is reduced from 576*1 to 12*1.

[0046] Step 3: The function of the data dimension reduction module is similar to that of the expected signal-guided vector dimension reduction module and is universal. Similarly, this module outputs x by row sub (t). Its dimension is reduced from 576*1 to 12*1.

[0047] Step 4: Figure 4 This is the block diagram of the covariance matrix calculation module. sub (t) After inputting row by row, the data is stored, and the conjugate is taken (the real part remains unchanged, and the sign of the imaginary part is reversed) and then stored in registers R0'~R11'. The conjugated data is moved in one by one, and complex multiplication is performed with the original data, and the covariance matrix can be output row by row. The dimension of the covariance matrix is ​​12*12.

[0048] Step 5: Figure 5 This is the block diagram of the snapshot accumulation module. Snapshot accumulation requires summing the covariance matrix. After accumulating 128 snapshots, the average is calculated and then output. Since the data structure of the covariance matrix is ​​parallel input per row, in order to increase code reusability, the accumulation of one row is abstracted into a submodule, which can be instantiated 12 times when used. The dimension of the accumulated covariance matrix is ​​12*12.

[0049] Step 6: Figure 6 The block diagram of the matrix inversion state machine is shown in Figure 1. The multiplication and addition operation module and the division operation module are both completed by the IP core. State machine 1 completes the matrix decomposition, and i and j are the indexes of the corresponding positions of the matrix. The i-th row of matrix C and the j-th column of matrix L are taken out from RAM according to the current position index (i, j) (i, j = 0, 1, ..., 11) and input into the multiplication and addition module to calculate c. ij and store it in RAM, and c ij and c jj Input the division operation module to calculate l ijand store them in RAM. After all elements are calculated, jump to the idle state. State machine 2 calculates the inverse matrix of matrix C. According to the current position index (i, j) (i, j = 0, 1, ..., 11), the i-th row of matrix C and the j-th column of matrix Y are taken out from RAM and input into the multiplication and addition module, and the result is added to C. ii Input the division operation module to get y ij And store it in RAM. After all elements are calculated, jump to the idle state. State machine 3 calculates the inverse matrix of matrix A. According to the current position index (i, j) (i, j = 0, 1, ..., 11), the i-th row of matrix L and the j-th column of matrix W are taken out from RAM and input into the multiplication and addition module. After the result is output, w is calculated. ij And store it in RAM. After all elements are calculated, jump to the idle state. At this point, the matrix inversion is completed, and the inverse matrix dimension is 12*12.

[0050] Step 7: The weighted vector calculation module can be completed in three steps. First, the inverse matrix Row-wise input and subarray-level steering vectors Multiply, the result is a column vector The dimension is 12*1; then the sub-array level steering vector is conjugate transposed and multiplied by the column vector to obtain Finally, the weighted coefficient can be obtained by using the divider IP core. Therefore, it is necessary to design a universal row-column multiplication module.

[0051] The above-mentioned embodiments only express the preferred implementation modes of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for adaptive spatial filtering of subarrays under non-uniform partitioning, characterized in that: It includes the following seven steps: Step 1: The element-level desired signal steering vector calculation module receives the angle information sent down, including azimuth angle information θ and pitch angle information The element-level desired signal steering vector is calculated using the planar array steering vector calculation method. Step 2: After the calculation of the element-level expected signal steering vector is completed, it is sent to the expected signal steering vector dimensionality reduction module to obtain the dimensionality reduction transformation matrix T, which can divide the array into L sub-arrays. The element-level expected signal steering vector is reduced in dimension by the dimensionality reduction transformation matrix T to obtain the sub-array-level steering vector Step 3: After the dimensionality reduction transformation matrix T is calculated, it is sent to the data dimensionality reduction module to reduce the dimension of the array input data X(t), that is, after conventional beamforming at the array element level, the sub-array level data X is obtained. sub (t). The incident signal on the front surface is composed of M unrelated interference signals and noise in space, and M<L; Step 4: Subarray-level data X sub (t) After the calculation is completed, it is sent to the covariance matrix calculation module to obtain the sub-array level covariance matrix R X ; Step 5: Subarray-level covariance matrix R X After the calculation is completed, it is sent to the snapshot accumulation module, and the elements of the matrix are accumulated and summed according to the number of K snapshots and the average is calculated to obtain Step 6: Result after snapshot accumulation Send it to the matrix inversion module, perform Cholesky decomposition on it, and use the decomposed submatrix to calculate its inverse matrix Step 7: After the matrix inversion is completed, The weighted vector calculation module is fed into the subarray-level steering vector calculated in step 2. At the subarray level, the Capon beamforming algorithm is used to calculate the subarray-level weight vector W Z And output.

2. The method for subarray adaptive spatial domain filtering under non-uniform division according to claim 1, characterized in that: In step 1, the planar array elements are located on the xz plane, and the number of elements is P×Q. The element-level steering vector is: Where d is the array element spacing, λ is the wavelength, (x i , z j ) are the coordinates of each array element relative to the reference array element, and i=0, 1, ..., P-1, j=0, 1, ..., Q-1.

3. The method for subarray adaptive spatial domain filtering under non-uniform division according to claim 1, characterized in that: The dimension reduction transformation matrix T in step 2 and step 3 satisfies: In the formula, each column of T0 represents a subarray, and is 1 if there is an array element at the corresponding position, otherwise it is 0. Subarray-level steering vector Element-level steering vector The relationship is: At time t, the relationship between the sub-array level sampling data and the array element level sampling data is: sub (t) = T H X(t).

4. The method for subarray adaptive spatial domain filtering under non-uniform division according to claim 1, characterized in that: The step 4 uses the following method to calculate the sub-array level covariance matrix for a single snapshot. R X =X sub (t)X sub (t) H The samples received at the subarray level at time t are X sub (t), 1≤t≤K, t is a positive integer, and K is the number of snapshots, that is, the number of sampling points.

5. The method for subarray adaptive spatial domain filtering under non-uniform division according to claim 1, characterized in that: The step 5 uses the following method to calculate the estimate of the subarray-level covariance matrix under K snapshots:

6. The method for subarray adaptive spatial domain filtering under non-uniform division according to claim 1, characterized in that: The matrix inversion method in step 6 assumes that the input matrix is ​​A, and performs Cholesky decomposition (A = LDL H =CL H ), we get the lower triangular matrices C and L. By matrix operation AW = CL H W=E=CY, we can get the matrix Y. Finally, L H W = Y calculates the inverse matrix of the covariance matrix A The input matrix A in this module is the estimate of the sub-array level covariance matrix The elements c in matrices C and L ij and l ij The calculation method is as follows: The inverse matrix Y of matrix C is calculated as follows: Finally, the inverse matrix W of A is calculated as follows:

7. The method for subarray adaptive spatial domain filtering under non-uniform division according to claim 1, characterized in that: The method of obtaining the weighted vector using the Capon beamforming algorithm at the subarray level in step 7 is: in is the subarray-level desired signal steering vector, and the solution to the optimization problem is the required subarray-level weighted vector:

Images