Weak array signal phase enhancement direction finding method

Abstract

The invention discloses a weak array signal phase enhancement direction finding method. The method comprises the following steps: collecting ideal array data, taking a first array element as a reference, and calculating a phase difference between each array element and the first array element by adopting a phase discrimination method to construct a data label; collecting original receiving data, and calculating the phase difference between each array element and the first array element by adopting a phase discrimination method to construct data features; the data features serve as input, the data labels serve as labels, and training is conducted through an auto-encoder neural network to obtain a training model; extracting data features from original received data needing to be enhanced, and inputting the data features into the training model to obtain an enhanced phase difference; an array phase difference vector is constructed through the enhanced phase difference, a covariance matrix is solved, the covariance matrix and an amplitude value matrix of an original receiving data covariance matrix are used for reconstructing a new covariance matrix, and a spectrum estimation method is used for direction finding. According to the method, the direction finding precision of weak array signals with few array elements and few snapshots can be improved, and the system sensitivity is further improved.

Description

Technical Field

[0001] The present invention belongs to the field of radio direction finding, and relates to radio astronomy detection, deep sea detection, electromagnetic detection, passive positioning and radio monitoring management technology, and specifically relates to a direction finding method for phase enhancement of weak array signals. Background Art

[0002] Direction finding technology is widely used in key demand areas such as astronomy, radar, electronic reconnaissance, earthquake monitoring, and acoustic detection. Arrays can increase the array aperture, have good low signal-to-noise ratio (SNR) effects, and can use multiple signals of the same frequency for direction finding. Therefore, array signal direction finding performance is better. However, the impact of multipath and low signal-to-noise ratio has always been the two major difficulties in array signal direction finding. They can easily cause the received signal to weaken and the noise to increase, thereby affecting the direction finding accuracy, increasing the pseudo-peaks of the spatial spectrum, and causing the system detection performance to deteriorate. Therefore, the research on weak array signal direction finding methods is of great significance to improving system sensitivity and breaking through the detection limit.

[0003] There are two types of array signal direction finding. One is the model-driven algorithm based on classic theoretical models such as the multiple signal classification algorithm (MUSIC), maximum likelihood estimation algorithm (ML), rotationally invariant subspace algorithm (ESPRIT), sparse reconstruction algorithm and their improved algorithms. This type of direction finding algorithm is derived from established mathematical models and formulas, but there are many assumptions, such as: the signal is a narrowband signal, the noise is Gaussian white noise, and the signals are independent of each other. In actual situations, it is difficult to meet all these assumptions. Therefore, its direction finding performance is determined by the degree of matching between the actual model and the theoretical model, and is greatly affected by channel amplitude and phase disturbances, especially multipath. The second is the data-driven algorithm from the data end to the direction finding end, including: deep neural network (DNN), convolutional neural network (CNN) and other direction finding methods. This type of method does not have a definite mathematical formula, but relies on intelligent algorithms. It consists of a large amount of simulation or measured data to form a training set, which is input into the machine learning network for training and then the direction of arrival (DOA) is estimated. This type of method is greatly affected by the environment, which limits its engineering application. The above two methods only consider the direction finding algorithm itself, but ignore the data processing in the early stage of direction finding, and their direction finding effect is affected by amplitude and phase disturbance. Since the actual received signal may have complex environments such as multipath, strong noise, and weak signal, which leads to model mismatch, only studying the direction finding algorithm will have limited improvement on the direction finding performance.

[0004] Since the early noise reduction of array signals will affect the signal phase and cause the direction finding to deteriorate or even fail, in order to improve the direction finding performance of weak array signals, researchers often make improvements based on the above two types of algorithms, and there are few literatures that perform early preprocessing of array signal direction finding. Therefore, in order to address the problems of increased pseudo peaks and reduced direction finding accuracy caused by multipath, strong noise, and weak signals, it is necessary to study an early preprocessing method that does not affect the array signal phase to improve the direction finding performance of weak array signals. Summary of the Invention

[0005] In view of this, the present invention proposes a direction finding method for weak array signal phase enhancement, aiming to solve the problems of low direction finding accuracy and many false peaks of weak signals.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows: As a first aspect, the present invention provides a direction finding method for weak array signal phase enhancement, including the following steps: Collect ideal array data, taking the first array element as a reference, using the phase discrimination method to find the phase difference between the m-th array element and the first array element under different DOAs, and constructing data labels; Collect original received data, taking the first array element as a reference, using the phase discrimination method to find the phase difference between the m-th array element and the first array element under different DOAs, and constructing data features; Taking the data features as input and the data labels as labels, use the Auto encoders neural network for training to form a training model; Extract the data features of the original received data to be enhanced, input them into the training model, and obtain the enhanced phase difference; Construct an array phase difference vector from the enhanced phase difference, and find the covariance matrix. Reconstruct a new covariance matrix from the amplitude value matrix of the covariance matrix of the enhanced phase difference and the covariance matrix of the original received data, and use the spectral estimation method for direction finding; Where m = 2, 3, 4... M, and M is the number of array elements.

[0007] Further, the collection of ideal array data is specifically as follows: If K far-field narrowband signals are incident on a uniform array with M array elements, K < M, and the array element spacing is half a wavelength, then at a certain moment t, the following mathematical model for receiving ideal array signals is established: ; Where ; is an M×K array manifold matrix of the desired signal, θ k is the incident angle of the desired signal, is a desired signal direction vector of M×1, d is the spacing between adjacent array elements, is the wavelength of the signal; is a desired signal vector of K×1, s k (t) is the signal envelope.

[0008] Further, taking the first array element as a reference, use the phase discrimination method to find the phase difference between the m-th array element and the first array element under different DOAs , where m = 2, 3, 4... M, and M is the number of array elements, to form data labels ; Taking the first array element as the reference, the phase difference between the mth array element and the first array element under different DOAs is calculated by using the phase detection method , where m=2,3,4…M, M is the number of array elements, constituting the data feature .

[0009] Furthermore, the phase detection method is specifically as follows: taking the first array element as a reference, using frequency domain phase detection technology, assuming that the first array element receives the signal The spectrum function is , then the mth array element receives the signal The spectrum function is: ; in, is the time delay between the two channels, f is the signal frequency; the cross spectrum of the two channels is multiplied by becomes: ; At this time, the phase difference between the two channels is: .

[0010] Furthermore, the acquisition of original received data is specifically: acquiring measured array data at different angles under different noise environments or simulating array data at different angles with different signal-to-noise ratios; the simulation of array data at different angles with different signal-to-noise ratios is obtained by adding Gaussian white noise with different signal-to-noise ratios to the mathematical model of the received ideal array signal.

[0011] Preferably, the autoencoder neural network structure is: input one-dimensional M-1 phase feature data, the encoding part passes through the first convolution layer, the first activation function layer, the second convolution layer, and the second activation function layer in sequence; the first convolution layer input is 1, the output is 16, and the convolution kernel is 3; the second convolution layer input is 16, the output is 32, and the convolution kernel is 3; the decoding part uses deconvolution size recovery, and passes through the first one-dimensional transposed convolution layer, the third activation function layer, the second one-dimensional transposed convolution layer, and the flattening layer (Flatten); the first one-dimensional transposed convolution layer input is 32, the output is 16, and the convolution kernel is 3; the second one-dimensional transposed convolution layer input is 16, the output is 1, and the convolution kernel is 3; the first, second, and third activation function layers all use ReLU activation functions.

[0012] Furthermore, the array phase difference vector is constructed by the enhanced phase difference, and the covariance matrix is ​​obtained. The covariance matrix and the amplitude value matrix of the covariance matrix of the original received data are reconstructed into a new covariance matrix, and the spectral estimation method is used for direction finding. Specifically, the phase difference enhanced by the autoencoder neural network is assumed to be , then the array phase difference vector constructed by the enhanced phase difference is: ; From the constructed array phase difference vector, the covariance matrix is ​​obtained : ; Assume that the amplitude value matrix of the covariance matrix of the original received data is , then the reconstructed new covariance matrix is: ; in, for Hadamard; Direction finding is performed based on spectral estimation methods.

[0013] Preferably, the spectrum estimation method includes a MUSIC algorithm, and the spatial spectrum function using the MUSIC algorithm is: ; in, is the noise subspace, It is the conjugate transpose, and the direction can be determined according to the spatial spectrum function.

[0014] As a second aspect, the present invention further provides a direction finding device for phase enhancement of weak array signals, comprising a plurality of sensors, a multi-channel receiver and a processor, for implementing the above-mentioned direction finding method for phase enhancement of weak array signals.

[0015] Furthermore, the direction-finding device for phase enhancement of weak array signals described in the present invention can be divided into a hardware part and a software part; the hardware part is used for signal reception, and the software part is used for controlling hardware, transmitting data and processing data signals; the hardware part mainly includes a sensor array, a front-end amplification system and a receiver, wherein the sensor array can be a linear array, a surface array, a volume array, etc., and the sensor can be an acoustic sensor, an electromagnetic signal sensor, etc.; the front-end amplification system includes: a filter and a low-noise amplifier, which are used for RF front-end amplification of weak signals; the receiver can be a zero intermediate frequency receiver, a frequency conversion receiver, a superheterodyne receiver, an acquisition card, etc.; the software part is mainly a data processing part, including a processor, a memory, a network interface and a non-volatile memory, etc.

[0016] As a third aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-mentioned direction finding method for phase enhancement of weak array signals.

[0017] The beneficial effects of the present invention are: (1) The present invention proposes a weak array signal phase enhancement method, which innovatively proposes to extract the phase difference for constructing the array phase difference vector and the new covariance matrix. This method is used as the signal preprocessing in the early stage of the direction finding algorithm. Under the premise of not affecting the array signal phase, it can effectively remove the noise of the array signal phase and improve the equivalent signal-to-noise ratio, which is beneficial to the subsequent direction finding. (2) After a lot of exploration and experiments, the present invention selects the autoencoder neural network structure as a deep learning method for phase enhancement and designs the corresponding network structure accordingly, achieving good results; (3) The direction finding method of the present invention can improve the problem of many pseudo peaks and low precision in the spectrum caused by multipath and low signal-to-noise ratio in traditional methods. It requires fewer snapshots and array elements, has strong robustness, and is conducive to improving system sensitivity. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the method of the present invention; Figure 2 This is the phase enhancement comparison diagram when SNR=-15dB; Figure 3 Normalized spectra of the two methods when SNR=-15dB; Figure 4 This is the phase enhancement comparison diagram when SNR=10dB; Figure 5 Normalized spectra of the two methods when SNR=10dB; Figure 6 It is the relationship diagram between SNR and RMSE; Figure 7 It is a comparison diagram of phase enhancement in multipath; Figure 8 It is the normalized spectra of the two methods when there are multiple paths; Fig. 9 Diagram of direction-finding equipment for phase enhancement of weak array signals. DETAILED DESCRIPTION

[0019] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0021] Embodiment 1: To solve the problems of increased false peaks and reduced direction-finding accuracy caused by multipath, strong noise, and weak signals, the present invention proposes a direction-finding method for enhancing the phase of weak array signals. As Figure 1 shown, it is a schematic flowchart of the above-mentioned direction-finding method for enhancing the phase of weak array signals. Specifically, the method includes the following steps: Step S101: Collect ideal array data. Taking the first array element as a reference, use the phase discrimination method to find the phase differences between the second to the Mth array elements and the first array element under different DOAs, and construct data tags; The ideal array data is clean array data without noise. If K far-field narrowband signals are incident on a uniform array with M array elements, K < M, and the element spacing is half a wavelength, then the mathematical model for receiving the ideal array signal at a certain moment t is: ; Among them, ; is an M×K array manifold matrix of the desired signal, θ k is the incident angle of the desired signal, is a direction vector of the desired signal with dimensions M×1, d is the spacing between adjacent array elements, is the wavelength of the signal. is a desired signal vector with dimensions K×1, s k (t) is the signal envelope. According to the above array model, ideal array data at different angles can be constructed.

[0022] Taking the first array element as a reference, assuming the signal received by the first array element has a spectral function of , then the spectral function of the signal received by the mth array element is: ; Among them, is the time delay between two channels, and f is the signal frequency. Multiply the cross-spectrum of the two channels on both sides, that is: to become: ; At this time, the phase difference between the two channels is: ; Therefore, use the phase discrimination method to calculate the phase difference between the second array element and the first array element, the phase difference between the third array element and the first array element... the phase difference between the Mth array element and the first array element, and form the data tag

[0023] Step S102, collecting the original received data, taking the first array element as the reference, using the phase detection method to calculate the phase difference between the second array element to the Mth array element and the first array element under different DOAs to construct data features; The original received data is noisy array data or array data received in the actual environment. Gaussian white noise with different signal-to-noise ratios is added to the mathematical model of the received array signal to simulate array data at different angles with different signal-to-noise ratios. Each type is simulated at least N times (N>10); or measured array data at different angles under different noise environments are collected.

[0024] Taking the first array element as a reference, the phase difference between the second array element and the first array element is calculated by using the phase detection method in step S101. , the phase difference between the third array element and the first array element ...the phase difference between the Mth array element and the 1st array element , which constitutes the data features .

[0025] Step S103: According to the data features and data labels, the autoencoder neural network is used for training and forming a training model; The data features are used as input, the data labels are used as labels, and the phase data features are denoised using an autoencoder neural network to form a training model. The structure of the autoencoder neural network is as follows: input one-dimensional M-1 phase feature data, the encoding part passes through the first convolution layer, the first activation function layer, the second convolution layer, and the second activation function layer in sequence; the first convolution layer input is 1, the output is 16, and the convolution kernel is 3; the second convolution layer input is 16, the output is 32, and the convolution kernel is 3; the decoding part uses deconvolution size recovery, and passes through the first one-dimensional transposed convolution layer, the third activation function layer, the second one-dimensional transposed convolution layer, and the flattening layer in sequence; the first one-dimensional transposed convolution layer input is 32, the output is 16, and the convolution kernel is 3; the second one-dimensional transposed convolution layer input is 16, the output is 1, and the convolution kernel is 3; the first, second, and third activation function layers all use ReLU activation functions.

[0026] Step S104, extracting data features from the original received data to be enhanced and inputting the data features into the training model to obtain an enhanced phase difference; Extract data features from the original received data to be enhanced by the method in step S102, and input the extracted data features into the training The model is trained to form an enhanced phase difference .

[0027] Step S105: construct an array phase difference vector from the enhanced phase difference, and obtain a covariance matrix, reconstruct a new covariance matrix by combining the covariance matrix with the amplitude value matrix of the original received data covariance matrix, and perform direction finding using a spectrum estimation method such as a MUSIC algorithm.

[0028] The array phase difference vector constructed by the enhanced phase difference is: ; From the constructed array phase difference vector, the covariance matrix is ​​obtained : ; Preferably, the constructed covariance matrix and the amplitude value of the covariance matrix of the original received data can be reconstructed into a new covariance matrix; Assume that the amplitude value matrix of the covariance matrix of the original received data is , then the reconstructed new covariance matrix is: ; in, It is Hadamard.

[0029] The eigenvalue decomposition of R is: ; in, is the eigenvalue, sort the eigenvalues ​​as: ; in, These K eigenvalues ​​are related to the signal, and their corresponding eigenvectors constitute the signal subspace , the remaining smallest MK eigenvalues ​​are equal to the noise power, and their corresponding eigenvectors constitute the noise subspace .

[0030] At this time, the spatial spectrum function is: ; in, is the noise subspace, It is the conjugate transpose, and the direction can be determined according to the spatial spectrum function.

[0031] Embodiment 2: Corresponding to the embodiment 1 of the aforementioned direction finding method for phase enhancement of weak array signals, the present invention further provides an embodiment of a direction finding device for phase enhancement of weak array signals, such as Fig. 9 As shown, it includes multiple sensors, a front-end amplification system, a multi-channel receiver and a processor, which are used to implement a direction finding method for phase enhancement of a weak array signal in the above embodiment.

[0032] An embodiment of a direction-finding device for phase enhancement of weak array signals of the present invention can be divided into a hardware part and a software part. The hardware part is used for signal reception, and the software part is used for controlling the hardware, transmitting data and processing data signals. The hardware part mainly includes a sensor array, a front-end amplification system and a receiver, wherein the sensor array can be a linear array, a surface array, a volume array, etc., and the sensor can be an acoustic sensor, an electromagnetic signal sensor, etc.; the front-end amplification system includes: a filter and a low-noise amplifier, which are used for RF front-end amplification of weak signals; the receiver can be a zero intermediate frequency receiver, a frequency conversion receiver, a superheterodyne receiver, an acquisition card, etc. The software part is mainly a data processing part, including a processor, a memory, a network interface, and a non-volatile memory, etc.

[0033] The software part can be applied to any device with data processing function, including but not limited to computers, industrial computers and other entities or devices. Fig. 9 In the figure, a structural diagram of any device with data processing capability in which a direction-finding device with phase enhancement of weak array signals described in the present invention is located is shown. In addition to the processor, memory, network interface and non-volatile memory, any device with data processing capability in which the device in the embodiment is located can also include other necessary hardware according to the actual function of the device with data processing capability. The above-described embodiment of the direction-finding device with phase enhancement of weak array signals is only schematic, and some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of the present invention. The division of the units is only a logical function division, and there can be other division methods in actual implementation, for example, multiple parts can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0034] Embodiment 3: The embodiment of the present invention further provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the direction finding method for phase enhancement of a weak array signal in the above-mentioned embodiment 1 is implemented.

[0035] The present invention may be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted via the computer-readable storage medium. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which is implemented in the process Figure 1 The computer-readable storage medium may be a memory, a magnetic disk, an optical disk, etc.

[0036] Embodiment 4: This embodiment verifies the effectiveness of the method of the present invention by comparing the weak array signal direction finding performance, direction finding accuracy under different SNRs, and direction finding performance under multipath conditions with the method of the present invention and the existing MUSIC method through simulation experiments. Among them, the existing MUSIC method is the most classic direction finding algorithm in this field, which can obtain the direction of arrival of the signal without any prior information support, and introduces the concept of "vector space" into the field of spatial spectrum estimation. Specifically, the data covariance matrix is ​​obtained from the received data of the array, and the covariance matrix is ​​subjected to eigenvalue decomposition. The eigenvectors corresponding to different eigenvalues ​​constitute mutually orthogonal signal subspaces and noise subspaces. The eigenvectors corresponding to large eigenvalues ​​constitute the signal subspace, and the eigenvectors corresponding to small eigenvalues ​​constitute the noise subspace. The orthogonality between the two subspaces is used to estimate the DOA of the signal.

[0037] First, the weak array signal direction finding performance of the two methods is compared. During simulation, the incoming wave direction is 40°, an 8-element uniform linear array is used, the distance between adjacent antennas is half a wavelength, and the snapshot is 200 during data processing. During training, the signal-to-noise ratio of the original received data varies from [-15,10] dB, with an interval of 1 dB. 100 groups of original received data and ideal array data are simulated at each SNR, for a total of 2600 groups of data. The phase difference of these data is extracted by the phase detection method to form data features and data labels. An autoencoder neural network is used during training. The network is a very useful neural network architecture, which is widely used in data denoising, data dimensionality reduction, feature extraction, etc. In an embodiment of the present invention, the network structure is as follows: input one-dimensional M-1 phase feature data, the encoding part passes through the first convolution layer, the first activation function layer, the second convolution layer, and the second activation function layer; the first convolution layer input is 1, the output is 16, and the convolution kernel is 3; the second convolution layer input is 16, the output is 32, and the convolution kernel is 3. The decoding part uses deconvolution size recovery, which passes through the first one-dimensional transposed convolution layer, the third activation function layer, the second one-dimensional transposed convolution layer, and the flattening layer; the first one-dimensional transposed convolution layer has an input of 32, an output of 16, and a convolution kernel of 3; the second one-dimensional transposed convolution layer has an input of 16, an output of 1, and a convolution kernel of 3. The activation function layers all use the ReLU activation function. During the test, the SNR of the original received data array signal is -15 dB. Figure 2 This is the phase enhancement comparison diagram when SNR=-15 dB. Figure 4 This is a phase enhancement comparison diagram when SNR=10 dB. It can be seen from the figure that when the signal-to-noise ratio is high, the phase of the original received data is not much different from the phase of the ideal array data; when the signal-to-noise ratio is low, the phase of the original received data deviates greatly from the phase of the ideal array data, and the enhanced phase after the training model is almost the same as the phase of the ideal array data. Figure 3 Normalized spectra of the two methods when SNR=-15 dB, Figure 5 The normalized spectra of the two methods when SNR=10 dB. It can be seen from the figure that the direction finding spectrum of the method of the present invention (MUSIC spectrum after phase enhancement) is sharper, the average noise spectrum value is lower, and the direction finding is more accurate, especially when the signal-to-noise ratio is low. At this time, the MUSIC spectrum of the original received data cannot be used for direction finding, but the method of the present invention can accurately find the direction.

[0038] Secondly, the direction finding accuracy of the two methods under different SNRs is compared. During simulation, the direction of the incoming wave is 40°, an 8-element uniform linear array is used, the distance between adjacent antennas is half a wavelength, and the snapshot is 200 during data processing. During training, the signal-to-noise ratio of the original received data varies in the range of [-15,10] dB, with an interval of 1 dB. 100 sets of ideal array data are simulated at each SNR, for a total of 2600 sets of data. The autoencoder neural network is also used during training. During testing, the SNR range of the original received data signal is [-10,10] dB, with an interval of 5dB. 100 Monte Carlo simulations are used at each SNR, and the direction finding accuracy is expressed by the root mean square error (RMSE): ; in is the number of Monte Carlo simulations, and the true angle of the signal is , is the signal source n c Monte Carlo DOA estimation. Figure 6 : is a relationship diagram between SNR and RMSE. It can be seen from the figure that the RMSE of the two methods decreases with the increase of SNR. When the SNR is low, the MUSIC spectrum direction finding accuracy deteriorates. The direction finding accuracy of the method of the present invention is better than that of the MUSIC method, especially when the SNR is low, the accuracy is much better than the MUSIC spectrum estimation result.

[0039] Finally, the direction finding performance under multipath conditions is compared. During simulation, the incoming wave direction is 40°, an 8-element uniform linear array is used, the distance between adjacent antennas is half a wavelength, and the snapshot is 200 during data processing. During training, the signal-to-noise ratio of the original received data varies from [-15,10] dB, with an interval of 1 dB. 100 sets of original received data are simulated under each SNR, for a total of 2600 sets of data, and an autoencoder neural network is used during training. During testing, 0.19+0.91j times the multipath and SNR=-13 dB noise are added to the data array. Figure 7 is the phase enhancement comparison diagram in multipath, Figure 8 It is the normalized spectrogram of the two methods when there are multiple paths. As can be seen from the figure, the phase of the array data after adding multipath and noise is obviously offset, and the direction finding will fail when the spectrum estimation is directly used. The phase after the phase enhancement algorithm proposed by the present invention almost coincides with the phase of the ideal array data, and the direction finding is accurate when the spectrum estimation is used. Figure 8 It can also be seen that the spectrum estimation effect of array data after multipath is significantly worse.

[0040] Based on the present embodiment, it can be concluded that the method of the present invention, as a signal preprocessing method in the early stage of the direction finding algorithm, can effectively remove the noise of the array signal phase and improve the equivalent signal-to-noise ratio, which is beneficial to the subsequent direction finding; and can improve the problem of the traditional method being affected by multipath and low signal-to-noise ratio, resulting in many pseudo peaks in the spectrum and low precision, requires fewer snapshots, has strong robustness, and is beneficial to improving the sensitivity of the system.

[0041] It should be noted that, in the present invention, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method or article including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, method or article including the element.

[0042] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A direction finding method for enhancing the phase of weak array signals, characterized in that, it includes the following steps: Collect ideal array data. Taking the first array element as a reference, use the phase discrimination method to find the phase difference between the m-th array element and the first array element under different DOAs, and construct data labels; Collect original received data. Taking the first array element as a reference, use the phase discrimination method to find the phase difference between the m-th array element and the first array element under different DOAs, and construct data features; Use the data features as inputs and the data labels as labels, and train using an autoencoder neural network to form a training model; Extract the data features from the original received data to be enhanced, input them into the training model, and obtain enhanced phase differences; Construct an array phase difference vector from the enhanced phase differences, and find the covariance matrix. Reconstruct a new covariance matrix from the magnitude value matrix of the covariance matrix of the original received data and the covariance matrix, and use the spectral estimation method for direction finding; where m = 2, 3, 4…M, and M is the number of array elements.

2. The method according to claim 1, characterized in that, the collection of ideal array data is specifically as follows: If K far-field narrowband signals are incident on a uniform array with M array elements, K < M, and the array element spacing is half a wavelength, then at a certain moment t, the following mathematical model for receiving ideal array signals is established: ； in, ; is an M×K array manifold matrix of the desired signal, θ k is the expected signal incident angle, is an M×1 desired signal direction vector, d is the spacing between adjacent array elements, is the wavelength of the signal; is a K×1 expected signal vector, s k (t) is the signal envelope.

3. The method according to claim 1, characterized in that, The first array element is used as a reference, and the phase difference between the mth array element and the first array element under different DOAs is calculated by using the phase detection method. , where m=2,3,4…M, M is the number of array elements, constituting the data label The first array element is used as a reference, and the phase difference between the mth array element and the first array element under different DOAs is obtained by using the phase detection method. , where m=2,3,4…M, M is the number of array elements, constituting the data feature .

4. The method according to claim 3, characterized in that, The phase detection method is specifically as follows: Assume that the first array element receives the signal The spectrum function is , then the mth array element receives the signal The spectrum function is: ； in, is the time delay between the two channels, f is the signal frequency; the cross spectrum of the two channels is: ； At this time, the phase difference between the two channels is: 。 5. The method according to claim 1 or 2, characterized in that, the collection of the original received data is specifically to collect the measured array data at different angles in different noise environments or simulate the array data at different signal-to-noise ratios and different angles; the simulation of the array data at different signal-to-noise ratios and different angles is obtained by adding Gaussian white noise with different signal-to-noise ratios to the mathematical model of the received ideal array signal.

6. The method according to claim 1, characterized in that, the structure of the autoencoder neural network is as follows: Input one-dimensional M - 1 phase feature data, and the encoding part successively passes through the first convolutional layer, the first activation function layer, the second convolutional layer, and the second activation function layer; the input of the first convolutional layer is 1, the output is 16, and the convolutional kernel is 3; the input of the second convolutional layer is 16, the output is 32, and the convolutional kernel is 3; the decoding part uses transposed convolution for size restoration and successively passes through the first one-dimensional transposed convolutional layer, the third activation function layer, the second one-dimensional transposed convolutional layer, and the flattening layer; the input of the first one-dimensional transposed convolutional layer is 32, the output is 16, and the convolutional kernel is 3; the input of the second one-dimensional transposed convolutional layer is 16, the output is 1, and the convolutional kernel is 3; the first, second, and third activation function layers all use the ReLU activation function.

7. The method according to claim 1, characterized in that, the construction of the array phase difference vector from the enhanced phase differences, finding the covariance matrix, reconstructing a new covariance matrix from the magnitude value matrix of the covariance matrix of the original received data and the covariance matrix, and using the spectral estimation method for direction finding is specifically as follows: Assume that the phase difference enhanced by the autoencoder neural network is , then the array phase difference vector constructed by the enhanced phase difference is: ; From the constructed array phase difference vector, the covariance matrix is ​​obtained : ； Assume that the amplitude value matrix of the covariance matrix of the original received data is , then the reconstructed new covariance matrix is: ； in, For Hadamard; Perform direction finding according to the spectral estimation method.

8. The method according to claim 7, characterized in that, The spectrum estimation method includes a MUSIC algorithm, wherein the spatial spectrum function of the MUSIC algorithm is: ； in, is the noise subspace, is the conjugate transpose.

9. A direction finding device for phase enhancement of weak array signals, comprising a plurality of sensors, a multi-channel receiver and a processor, It is characterized in that A direction finding method for implementing phase enhancement of weak array signals as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the steps in the direction finding method for phase enhancement of weak array signals as described in any one of claims 1 to 8 are implemented.

Images