阅读说明：本技术 一种基于功率特征值提取的改进块稀疏贝叶斯抗干扰方法 (Improved block sparse Bayesian anti-interference method based on power characteristic value extraction ) 是由 张海龙 张宁 李纪三 于 2020-06-12 设计创作，主要内容包括：本发明涉及一种基于功率特征值提取的改进块稀疏贝叶斯抗干扰方法,属于雷达信号处理技术领域,特别涉及解决电子干扰中的压制式窄带干扰以及主瓣干扰问题。首先采集雷达中频回波数据并缓存,接着初始化块稀疏贝叶斯数据块大小、设置学习速率和最大收敛次数；然后构造优化函数并建立块稀疏贝叶斯数学模型,添加扰动项；接着根据特征功率门限判断是否是干扰信号特征功率,如果是则构造块稀疏恢复子字典进行相应迭代；最后判断如果达到最大迭代次数,则利用稀疏恢复反射系数进行数据恢复,从而抑制干扰信号,有效提高目标检测概率。本发明可以分离雷达有用回波和干扰信号,有效改善目标信号的信干噪比、抑制干扰信号,降低运算量。(The invention relates to an improved block sparse Bayesian anti-interference method based on power characteristic value extraction, belongs to the technical field of radar signal processing, and particularly relates to a method for solving the problems of suppression type narrow-band interference and main lobe interference in electronic interference. Firstly, acquiring radar intermediate frequency echo data and caching, then initializing block sparse Bayesian data block size, and setting learning rate and maximum convergence times; then constructing an optimization function, establishing a block sparse Bayesian mathematical model, and adding a disturbance term; then judging whether the characteristic power is the interference signal characteristic power according to a characteristic power threshold, and if so, constructing a block sparse recovery sub-dictionary to perform corresponding iteration; and finally, if the maximum iteration times is judged, data recovery is carried out by using the sparse recovery reflection coefficient, so that interference signals are suppressed, and the target detection probability is effectively improved. The invention can separate radar useful echo and interference signal, effectively improve the signal-to-interference-noise ratio of target signal, inhibit interference signal and reduce the operation amount.)

1. An improved block sparse Bayesian anti-interference method based on power characteristic value extraction is characterized in that:

1) collecting radar intermediate frequency echo data, constructing a sparse recovery perception principle system model, wherein the radar intermediate frequency echo system mathematical model is as follows:

E_cho'＝Φ'[Re(σ_I ^T)Re(σ_U ^T)Im(σ_I ^T)Im(σ_U ^T)]^T+[_IU]^T；

wherein the perception matrix phi' is a 2N multiplied by 2M dimensional matrix, and the mathematical expression of the matrix expanded according to the rows can be expressed as [ Re (D) -im (D) Re (D)]D represents a plurality of dictionaries;_IUrepresents an additional superposition of the echo signals in addition to the interference; and assuming that a real part and a complex part are M-dimensional vectors respectively; the echo signal to be extracted is denoted as [ Re (σ) ]_I ^T)Re(σ_U ^T)Im(σ_I ^T)Im(σ_U ^T)]^TAnd defined as a 2N × 1 vector;

2) initializing perception matrix related parameters, adaptive learning type adaptive, super parameter pure _ gama and block matrix element size F_gWherein the blocking of the interference signal and the useful echo signal is carried out independently and the minimum blocking structure F is set according to the signal to interference plus noise ratio SINR_gIf the signal interference noise ratio SINR > T needs to be judged_ΔdB, set the minimum block structure to F_g10; otherwise, setting block minimum structure F_g＝6；T_ΔSelecting according to the actual radar background; carrying out corresponding division of complex number segmentation into real number on the acquired complex radar echo signals according to the set block structure, wherein the block structure of radar interfered signal elements is theta_I＝{[1:g_I],[g_I+g_U+1:2g_I+g_U]}; the block structure of the useful target echo signal elements is expressed as theta_U＝{[g_I+1:g_I+g_U],[2g_I+g_U+1:2g_I+2g_U]}; setting a learning rate learn _ lambda and a maximum iteration number max _ iter;

3) constructing a system optimization cost function, wherein the cost optimization function is set as:

whereinRepresenting a likelihood function, gamma_iIs a non-negative hyperparameter, B_iRepresenting a positive definite matrix for characterizing the internal interconnection relationship between the blocks; estimating the related parameters of the expectation-maximization algorithm under the improved block sparse Bayesian framework by using the initial information and the radar echo matrix information;

4) judging whether the characteristic power of the interference signal is the characteristic power of the interference signal according to a power characteristic value threshold, wherein the definition of the characteristic power is defined as follows according to the physical meaning of the matrix characteristic value:

the threshold calculation formula is completed in an interference signal space, and a main covariance matrix corresponding to each vector divided into blocks is gamma_iB_iThe block number of the interference signal which is finally divided is g_I(ii) a l is set to 0.5 characterizing the weighting factor; if the interference signal is judged to exceed the calculation threshold, a target echo matrix sub-dictionary and an interference signal matrix sub-dictionary are respectively constructed as follows:

w of the above formula_NAnd e_mnShown as W_N＝exp(-j2π/N),Ε_mn＝exp(jπkt²) Wherein k represents the chirp rate of the transmitted signal; sub-dictionary D of interfering signals_IIs N multiplied by N dimension, and the useful target echo signal sub-dictionary is N multiplied by M dimension; if the power characteristic value exceeds the threshold, carrying out iterative operation of data recovery of the interference signal and the useful echo signal, otherwise traversing the next data;

5) after the iteration maximum value is judged, reconstructing an interference signal and reconstructing a useful echo signal by utilizing dictionary information; the specific reconstruction mode of the interference signal and the radar useful echo signal is as follows:

wherein the content of the first and second substances,andthe effective reflection coefficients of the interference signal and the useful echo signal are respectively recovered by the method.

Technical Field

The invention belongs to the technical field of radar signal processing, and is particularly suitable for a scene that electronic interference is narrow-band suppressed interference entering a radar receiver to influence radar detection probability.

Background

The electromagnetic environment of modern electronic battlefield is complex and changeable, and the electronic interference patterns designed aiming at the working mode of radar are different in form. Especially for a conventional search radar, narrow-band suppression type interference is an electronic attack means commonly used by enemies, frequency agility is adopted for a typical active anti-interference method of suppression type interference, a typical passive anti-interference method adopts a method of combining side lobe pair cancellation and side lobe suppression with signals and data processing points, track quality evaluation and the like to suppress interference signals, but the active anti-interference measure has adverse effects of overlarge burden on limited bandwidth resources of a phased array radar, overlarge loss of target detection probability brought by passive anti-interference measures and the like, and has no subordinate effect on interference from a radar main lobe. Therefore, it is one of the key technologies for radar target detection to reduce the signal-to-interference ratio loss, aiming at how to effectively suppress the interference signal by the narrowband pressure system interference, especially the interference from the main lobe direction.

The invention patent document of the university of Xian traffic in the application of the university of Sian 'a Gaussian mixture spectrum sensing method based on block sparsity characteristics' (publication number: CN108880713A, application number: CN201810490654.5) discloses a Gaussian mixture spectrum sensing method based on block sparsity characteristics. The method is mainly used for constructing a block sparse Bayesian mathematical model under Gaussian noise to finally reconstruct power spectrum information of a main user, but the method cannot be used for a non-Gaussian model and has poor recovery effect on a non-stationary noise environment and a narrow-band interference environment.

In the document, "frequency modulation communication comb interference suppression based on block sparse Bayesian learning" (war institute of war, 2018, Vol.9, No.1pp:9-12), a learning framework under a block sparse Bayesian mathematical model is established by using the sparse characteristic of interference signals expressed in the frequency domain, and comb interference is recovered by an expectation-maximization algorithm "is proposed. However, the algorithm cannot effectively recover the interference signal under the condition that the time domain of the interference signal is not sparse, and has a convergence problem when the data sample is large.

Disclosure of Invention

The invention aims to provide an anti-interference method with high convergence rate and low target signal-to-noise ratio loss aiming at the problem that a receiver is mixed with narrow-band suppression interference from an antenna main lobe under the condition that the background is not stable white Gaussian noise so as to solve the problems that the target signal-to-noise ratio loss is serious, the narrow-band interference entering from a radar main lobe cannot be inhibited, the convergence rate is low and the like in the processing process of the prior art.

The specific technical scheme of the invention is as follows:

firstly, radar intermediate frequency echo data are collected, and a sparse recovery perception principle system model is constructed. The radar intermediate frequency echo system comprises:

E_cho'＝Φ'[Re(σ_I ^T)Re(σ_U ^T)Im(σ_I ^T)Im(σ_U ^T)]^T+[_IU]^T

wherein the perception matrix is a 2 Nx 2M dimensional matrix, and the matrix is expanded by rows and can be expressed as [ Re (D) -im (D) Re (D)]And D represents a complex dictionary._IUExpressed as an additional superposition of the echo signals in addition to the interference, a common superposition is embodied in the form of noise, and the real part and the complex part are assumed to be M-dimensional vectors respectively. The echo signal to be extracted is denoted as [ Re (σ) ]_I ^T)Re(σ_U ^T)Im(σ_I ^T)Im(σ_U ^T)]^TAnd is defined as a 2N × 1 vector.

Then initializing relevant parameters of a perception matrix, setting a block minimum structure F according to the size of elements of a block matrix and the signal to interference noise ratio (SINR)_gIf the signal to interference plus noise ratio SINR > T_ΔdB, set the minimum block structure to F_g10; otherwise, setting block minimum structure F_g＝6。T_ΔSelected according to the actual radar background. Performing corresponding division of complex numbers of acquired complex radar echo signals into real numbers of corresponding positions according to the set block structure, wherein the finally obtained radar interfered signal element sparse Bayesian block element structure is defined as theta_I＝{[1:g_I],[g_I+g_U+1:2g_I+g_U]}; the element structure of the sparse Bayesian block of the useful radar target echo signal is expressed as theta_U＝{[g_I+1:g_I+g_U],[2g_I+g_U+1:2g_I+2g_U]}. And setting the learning rate and the maximum iteration number.

Constructing a system optimization cost function:

the cost function is improved according to the radar echo characteristics by the expectation maximization algorithm under the block sparse Bayesian model, and a disturbance term is introduced_iAnd the unconvergence situation is prevented. Wherein

Representing a likelihood function, gamma_iIs a non-negative hyperparameter, B_iA positive definite matrix is represented for characterizing the inter-connection relationships between blocks.

And then, the calculation speed is accelerated by utilizing the defined threshold of the characteristic power, and the calculation efficiency is improved. The physical meaning according to the matrix eigenvalues is defined as:

the threshold calculation formula is completed in an interference signal space, and a main covariance matrix corresponding to each vector divided into blocks is gamma_iB_iThe block number of the interference signal which is finally divided is g_I。Set to 0.5 characterizing the weighting factor. And estimating the related parameters of the expectation-maximization algorithm under the improved block sparse Bayesian architecture by using the initial information and the radar echo matrix information. And judging whether the characteristic power of the interference signal is the characteristic power of the interference signal or not according to a set threshold, if the characteristic power exceeds the threshold, establishing a recovery sub dictionary of the interference signal and the radar useful echo signal, otherwise, traversing the next data to judge that the characteristic power reaches the maximum value, and then utilizing dictionary information to reconstruct the interference signal and the useful echo signal. If the interference signal is judged to exceed the calculation threshold, respectively constructing a radar useful echo matrix sub-dictionary and an interference signal matrix sub-dictionary as follows:

w of the above formula_NAnd e_mnShown as W_N＝exp(-j2π/N),Ε_mn＝exp(jπkt²) Where k represents the chirp rate of the transmitted signal. Sub-dictionary D of interfering signals_IIs of dimension N × N, useful radar echo D_UThe signal sub-dictionary is N × M dimensional.

The specific reconstruction mode of the interference signal and the radar useful echo signal is as follows:

wherein the content of the first and second substances,andthe effective reflection coefficients of the interference signal and the radar useful echo signal recovered by the method are respectively. And performing subsequent radar signal processing and interference identification of radar echo by using the reconstructed information.

The invention is described in further detail below with reference to fig. 1.

Drawings

FIG. 1 is a processing flow chart of the improved block sparse Bayesian anti-interference method based on power eigenvalue extraction of the present invention.

Fig. 2 is a schematic diagram of the pulse pressure of the collected radar echo after adding the narrowband radio frequency interference in the embodiment of the present invention.

Fig. 3 is a waveform diagram of an interference signal recovered after being processed by a power eigenvalue-based improved block sparse bayesian anti-interference method in an embodiment of the present invention.

Fig. 4 is a waveform diagram of a radar useful echo signal recovered after processing by using a power eigenvalue-based improved block sparse bayesian anti-interference method in the embodiment of the present invention.

FIG. 5 is a comparison graph of a radar useful echo signal which is recovered after being processed by a power characteristic value-based improved block sparse Bayesian anti-interference method and subjected to pulse pressure modulo calculation with a direct pulse pressure modulo calculation without an interference signal.

Fig. 6 shows the improvement of the signal-to-interference-and-noise power ratio after the power eigenvalue-based improved block sparse bayesian interference rejection method is adopted in the embodiment of the present invention.

FIG. 7 is a distortion degree statistic of a radar useful echo signal which is recovered after being processed by a power characteristic value-based improved block sparse Bayesian anti-interference method and subjected to pulse pressure modulo calculation and a signal which is not subjected to interference and subjected to direct pulse pressure modulo calculation.

Detailed Description

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments, but the present invention is not limited thereto. The schematic diagram of the power characteristic value-based improved block sparse Bayesian anti-interference method is shown in FIG. 1, and the specific process is as follows:

the method comprises the following steps: and acquiring radar intermediate frequency echo data to construct a sparse recovery perception principle system model.

Considering the structural characteristics of the radar intermediate frequency echo, the radar intermediate frequency echo system is as follows:

E_cho'＝Φ'[Re(σ_I ^T)Re(σ_U ^T)Im(σ_I ^T)Im(σ_U ^T)]^T+[_IU]^T

wherein the perception matrix is a 2 Nx 2M dimensional matrix, and the matrix is expanded by rows and can be expressed as [ Re (D) -im (D) Re (D)]，_IUExpressed as an additional superposition of the echo signals in addition to the interference, a common superposition is embodied in the form of noise, and the real part and the complex part are assumed to be M-dimensional vectors respectively. The echo signal to be extracted is denoted as [ Re (σ) ]_I ^T)Re(σ_U ^T)Im(σ_I ^T)Im(σ_U ^T)]^TAnd is defined as a 2N × 1 vector.

Step two: initializing relevant parameters of a perception matrix, the size of elements of a block matrix, and setting a learning rate and the maximum iteration number;

1) the blocking of the interference signal and the useful echo signal is carried out independently and in accordance with the signal interferenceNoise ratio SINR setting block minimum structure F_gWhen the signal to interference plus noise ratio SINR > T_ΔdB sets the minimum structure of the block as F_gOtherwise, set the minimum blocking structure F to 10_g6. Where T is set_ΔThe radar has an element block structure of a narrow-band interference signal subjected to electronic suppression, which is expressed as theta, and 5_I＝{[1:g_I],[g_I+g_U+1:2g_I+g_U]Equally, a useful radar echo signal element block structure can be defined as Θ_U＝{[g_I+1:g_I+g_U],[2g_I+g_U+1:2g_I+2g_U]}。

2) The adaptive learning type adaptive is initialized to 2, the maximum number of iterations max _ iter is set to 20, the learning rate learn _ lambda is set to 0.001, and the super parameter pure _ gama is set to 0.01.

Step three: and constructing a system optimization cost function, and estimating the related parameters of the expectation maximization algorithm under the improved block sparse Bayesian architecture by using the initial information and the radar echo matrix information.

The cost optimization function is set as:

the cost function is improved according to the radar echo characteristics by the expectation maximization algorithm under the block sparse Bayesian model, and a disturbance term is introduced_iAnd the unconvergence situation is prevented. The occurrence of a misconvergence situation is prevented. WhereinRepresenting a likelihood function, gamma_iIs a non-negative hyperparameter, B_iA positive definite matrix is represented for characterizing the inter-connection relationships between blocks.

Step four: and judging whether the characteristic power of the interference signal is the characteristic power of the interference signal according to a set threshold, and if the characteristic power of the interference signal exceeds the threshold, establishing a recovery sub-dictionary of the interference signal and the radar useful echo signal.

According to the invention, the power characteristic value threshold is defined according to the third step, and the calculation amount is reduced by utilizing the threshold, so that the calculation speed is increased inevitably, and the calculation efficiency is improved. The physical meaning according to the matrix eigenvalues is defined as:

the threshold calculation formula is completed in an interference signal space, and a main covariance matrix corresponding to each vector divided into blocks is gamma_iB_iThe block number of the interference signal which is finally divided is g_I。Set to 0.5 characterizing the weighting factor. If the interference signal is judged to exceed the calculation threshold, respectively constructing a radar useful echo matrix sub-dictionary and an interference signal matrix sub-dictionary as follows:

w of the above formula_NAnd e_mnShown as W_N＝exp(-j2π/N),Ε_mn＝exp(jπkt²) Where k represents the chirp rate of the transmitted signal. Sub-dictionary D of interfering signals_IIs of dimension N × N, and the radar useful echo signal sub-dictionary is of dimension N × M.

Step five: and reconstructing the interference signal and the echo signal by using the sub-dictionary information to recover the original radar useful echo signal.

The concrete reconstruction formula is expressed as:

wherein the content of the first and second substances,andrespectively interfering signals andand respectively obtaining radar effective reflection coefficients of radar useful echo signals after sparse recovery.

Step six: and performing pulse compression processing and interference signal identification by using the reconstructed information.

The invention is further illustrated by data simulation.

Suppose there are two point target signals in the simulation area, the target speed is 0m/s and 300m/s, the target distance is set to 3.3km and 5.1km, the target signal-to-interference-and-noise ratio is set to 0dB and-12 dB, the radar carrier frequency is set to 3GHz, the radar emission signal is set to linear positive frequency modulation signal, the time width of the signal is set to 2us, and the bandwidth is set to 1 MHz. The light speed is set to 3 × 10⁸m/s。

Fig. 2 shows a direct pulse pressure model diagram after radar echo signals and noise and frequency modulation narrowband interference are mixed without any anti-interference measures. It can be seen from the figure that although the signal-to-noise ratio is improved through pulse compression, the interference signal is still far larger than the target signal, the modulo main lobe after pulse pressure is also submerged in the interference, target detection under subsequent given indexes cannot be performed, and the detection threshold is calculated according to the false alarm rate to perform conventional detection, which inevitably causes a target missing detection phenomenon. Fig. 3 and 4 show interference signals and radar useful echo signals recovered by the improved block sparse bayesian anti-interference method based on power characteristic value extraction, and it can be seen from the figures that radar target echo signals and interference signals are basically and completely separated. FIG. 5 is a diagram illustrating the comparison of the echo pulse pressure modulus value recovered by the present invention with the original radar echo pulse pressure modulus value without interference. The distortion degree before and after comparison is shown in fig. 7, the maximum loss of the pulse pressure modulus after interference separation is 1.5%, the requirement of engineering design index is completely met, and the average signal-to-interference-plus-noise ratio brought by the block sparse Bayes anti-interference method based on power characteristic value extraction through the method is 12.48dB as shown in fig. 6.