Low-complexity direction of arrival estimation method and system based on biLSTM network
Through the low-complexity wave-to-direction estimation method based on biLSTM network, the accuracy and robustness of traditional DOA estimation in multipath propagation and strong noise environments are solved, and high-precision and low-complexity DOA estimation in impulse noise environments are achieved.
Patent Information
- Application Number
- CN202510056097.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-14
- Publication Date
- 2025-05-13
AI Technical Summary
Traditional DOA estimation methods are difficult to effectively distinguish the signal source direction from the multipath signal interference in multipath propagation and strong noise environments, resulting in reduced accuracy and insufficient robustness.
Using a low-complexity wave-to-reach direction estimation method based on biLSTM network, by obtaining the covariance matrix R of the wireless signal, expanding and converting it to the real space, the biLSTM network is trained to suppress impulse noise, reconstructing the covariance matrix of the noise-free array signal, and combining the least squares rotation invariance technology for DOA estimation.
Fast and real-time signal source positioning and monitoring are achieved in impulse noise environment, improving the accuracy and robustness of DOA estimation and reducing the computational complexity.
Smart Images
Figure CN119986525A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of direction of arrival (DOA) estimation, and in particular to a low-complexity direction of arrival estimation method and system based on a biLSTM network. Background Art
[0002] The core goal of DOA estimation is to determine the direction information of the signal source relative to the receiving array, which is crucial for target positioning, tracking, and signal processing. Traditional DOA estimation methods are mainly rooted in array signal processing theory, covering classic algorithms such as conventional beamforming (CBF) and minimum variance distortionless response (MVDR) beamforming. In an ideal signal environment, these methods can show certain effects.
[0003] However, as the actual application scenarios become increasingly complex, these methods face many severe challenges. In particular, in a multipath propagation environment, the signal will reach the receiving array through multiple paths, causing the signal to suffer severe fading and distortion. This makes it difficult for traditional methods to effectively distinguish the direction of the real signal source from the interference caused by the multipath signal, which greatly reduces the accuracy of DOA estimation.
[0004] In addition, in scenarios with strong noise interference, such as radar working environments with complex electromagnetic environments or noisy underwater sonar detection environments, noise will mask signal characteristics, thereby seriously inhibiting the performance of traditional algorithms and reducing their robustness. Traditional signal processing methods are unable to cope with complex and changeable signal environments, such as multipath propagation, pulse noise interference, and signal frequency overlap. In particular, pulse noise, which has non-Gaussian characteristics, is sudden and high-intensity, and can seriously interfere with the stability and accuracy of the array receiving signal. In DOA estimation, this will cause the traditional algorithm to fail, the signal characteristics to be severely distorted, and the estimation results to have large deviations, making it impossible to accurately determine the DOA of the signal source. Summary of the invention
[0005] The purpose of the present invention is to propose a low-complexity direction of arrival estimation method and system based on biLSTM network, so as to realize fast real-time positioning and monitoring of spatial wireless target sources in an impulse noise environment.
[0006] According to a first aspect of an embodiment of the present disclosure, a low-complexity direction of arrival estimation method based on a biLSTM network is provided, comprising the following steps:
[0007] Obtain the covariance matrix R of the wireless signal;
[0008] The covariance matrix R of the wireless signal is expanded and converted to the real number space to form a data pair p = {r in ,r non};
[0009] Using data pair p = {r in ,r non} Train the biLSTM network to suppress impulse noise; in practical applications, the output of the trained biLSTM network is the denoised vector r n ' on ;
[0010] The vector r n ' on Convert to complex space and reconstruct the covariance matrix R′ of the noise-free array signal using the inverse operation non , the covariance matrix R′ non It is combined with the total least squares rotation invariant technique to achieve DOA estimation.
[0011] In one embodiment, the covariance matrix R of the wireless signal is obtained as follows: Assuming that a uniform linear array ULA is composed of N equally spaced antennas, the spacing between adjacent antennas is d, and K signal sources are incident on the uniform linear array, the signal received by the array is expressed as:
[0012] x(t)= A(θ) s(t)+ v(t) (1)
[0013] Where A(θ) is the steering matrix, each column of which is the steering vector corresponding to each signal source direction, s(t) is the signal source signal vector, and v(t) is the noise vector;
[0014] The covariance matrix R of the wireless signal is:
[0015] R = E[x(t)x H (t)] (2)
[0016] Where E[·] represents the expected operation, (·) H Represents the conjugate transpose operation.
[0017] In one embodiment, the covariance matrix R of the wireless signal is expanded, specifically:
[0018]
[0019] Among them, σ 2 I is the covariance matrix of spatial white noise, and σ k 2 is the power of the kth signal source, a(θ k ) is the steering vector of the kth signal source.
[0020] So we get:
[0021]
[0022] R non Expanded into matrix form:
[0023]
[0024] This matrix satisfies the following conditions:
[0025]
[0026] Take the matrix R non The first row and diagonal elements of form a new vector:
[0027] r non =[r 11 ,r 12 ,…,r 1N ,r 22 ,…,r NN ] (7)
[0028] Convert a vector to real space:
[0029]
[0030] Among them, Imag(·) and Real(·) represent the real part and imaginary part respectively, and max(‖·‖) represents the maximum value of the modulus.
[0031] In one embodiment, the first row and diagonal elements of the noise-containing covariance matrix R in equation (2) are r i ' n ,Right now:
[0032] r i ' n =[R 11 ,R 12 ,…,R 1N ,R 22 ,…,R NN ] (9)
[0033] Converting it to real number space is:
[0034]
[0035] In one embodiment, the biLSTM network includes a sequence input layer, multiple biLSTM layers, a fully connected layer and a regression layer.
[0036] In one embodiment, the vector r n ' on The conversion to complex space is:
[0037] r out =r n ' on (1:2N-1)+jr n ' on (2N:4N-2) (11).
[0038] In one embodiment, the covariance matrix R′ of the noise-free array signal is reconstructed by using the inverse operation non :
[0039]
[0040] According to a second aspect of an embodiment of the present disclosure, a low-complexity direction of arrival estimation system based on a biLSTM network is provided, comprising:
[0041] A signal acquisition module, which acquires the covariance matrix R of the wireless signal;
[0042] The input vector building module expands the covariance matrix R of the wireless signal and converts it into the real number space to form a data pair p = {r in ,r non};
[0043] The impulse noise suppression module uses the data pair p = {r in ,r non} Train the biLSTM network to suppress impulse noise; in practical applications, the output of the trained biLSTM network is the denoised vector r n ' on ;
[0044] The DOA estimation module transforms the vector r n ' on Convert to complex space and reconstruct the covariance matrix R′ of the noise-free array signal using the inverse operation non , the covariance matrix R′ non It is combined with the total least squares rotation invariant technique to achieve DOA estimation.
[0045] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored and running on the memory, wherein when the processor executes the program, the low-complexity direction of arrival estimation method based on a biLSTM network is implemented.
[0046] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the low-complexity direction of arrival estimation method based on a biLSTM network is implemented.
[0047] Compared with the prior art, the above technical solution adopted by the present invention has the following advantages: the present invention is based on a bidirectional long short-term memory (biLSTM) network, which effectively suppresses impulse noise. The network uses the deformation of the covariance matrix of the noisy array output signal as input data, and its output is the processed covariance matrix of the noise-free Toeplitz structure, thereby obtaining a reliable DOA estimation.
[0048] The least squares rotation invariance technique is used to process the covariance matrix of the noise-free Toeplitz structure to achieve high-precision DOA estimation under low computational complexity. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] The drawings in the specification, which constitute a part of the present application, are used to provide further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.
[0050] Figure 1 This is a schematic diagram of the biLSTM network structure;
[0051] Figure 2 Schematic diagram of DOA estimation performance analysis. DETAILED DESCRIPTION
[0052] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.
[0053] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.
[0054] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0055] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and systems according to various embodiments of the present disclosure. It should be noted that each box in the flowchart or block diagram can represent a module, a program segment, or a part of a code, and the module, program segment, or a part of a code may include one or more executable instructions for implementing the logical functions specified in each embodiment. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the flowchart and / or block diagram, and the combination of boxes in the flowchart and / or block diagram can be implemented using a dedicated hardware-based system that performs a specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.
[0056] Embodiment 1:
[0057] In many key areas of modern science and technology, such as wireless communications, radar detection, sonar systems, earthquake monitoring, etc., signal processing plays an extremely important role. Among them, the estimation of the direction of arrival (DOA) is the core technical requirement of many application scenarios. For example, in wireless communications, accurate DOA estimation helps base stations to intelligently adjust signal transmission and reception strategies to achieve efficient signal transmission and interference avoidance; in the field of radar detection, accurate judgment of the DOA of the target can greatly improve the accuracy and speed of target positioning, tracking and identification, and enhance national defense security capabilities; sonar systems rely on DOA estimation to accurately detect the position of underwater targets, assisting in marine resource exploration and naval combat operations; in earthquake monitoring, through the DOA analysis of seismic wave signals, the source location can be more accurately determined, providing a key basis for earthquake early warning and disaster assessment. Therefore, this embodiment provides a low-complexity direction of arrival estimation method based on a biLSTM network, comprising the following steps:
[0058] S1. Obtain the covariance matrix R of the wireless signal;
[0059] Specifically, assuming that a uniform linear array ULA consists of N equally spaced antennas, the spacing between adjacent antennas is d, and there are K signal sources incident on the uniform linear array, the signal received by the array can be expressed as:
[0060] x(t)= A(θ) s(t)+ v(t) (1)
[0061] Where A(θ) is the steering matrix, each column of which is the steering vector corresponding to the direction of each signal source, s(t) is the signal source signal vector, and v(t) is the noise vector.
[0062] The covariance matrix R of the wireless signal is:
[0063] R = E[x(t)x H (t)] (2)
[0064] Where E[·] represents the expected operation, (·) H Represents the conjugate transpose operation.
[0065] In the actual DOA estimation process, this embodiment uses a uniform linear array including 6 antennas to receive signals, and uses formula (2) to obtain the covariance matrix R of the array output signal to achieve signal acquisition.
[0066] S2. Expand the covariance matrix R of the wireless signal and convert it into real number space to form a data pair p = {r in ,r non};
[0067] In order to further reduce the dimension of the biLSTM network input data, for a uniform linear array, if the covariance matrix of the noise-free output signal is defined as R non , it can be seen that it usually has a Toeplitz structure. This structure reflects the statistical characteristics of the signal, that is, its temporal stationarity. Therefore, formula (2) can be further expanded into the following form:
[0068]
[0069] Among them, σ 2 I is the covariance matrix of spatial white noise, and is the power of the kth signal source, a(θ k ) is the steering vector of the kth signal source.
[0070] So we can get:
[0071]
[0072] R non Expanded into matrix form:
[0073]
[0074] This matrix satisfies the following conditions:
[0075]
[0076] Take the matrix R non The first row and diagonal elements of form a new vector:
[0077] r non =[r 11,r 12 ,…,r 1N ,r 22 ,…,r NN ] (7)
[0078] Convert a vector to real space:
[0079]
[0080] Among them, Imag(·) and Real(·) represent the real part and imaginary part respectively, and max(‖·‖) represents the maximum value of the modulus.
[0081] Similarly, the first row and diagonal elements of the covariance matrix R containing noise in equation (2) are defined as r i ' n ,Right now:
[0082] r i ' n =[R 11 ,R 12 ,…,R 1N ,R 22 ,…,R NN ] (9)
[0083] Convert it to real space:
[0084]
[0085] The implementation of biLSTM network for suppressing impulse noise can be regarded as a data mapping mechanism, that is, mapping the covariance matrix of the array output signal containing impulse noise to the covariance matrix of the noise-free array output signal. in ,r non} is defined as the input of the biLSTM network.
[0086] S3. Using data pair p = {r in ,r non} Train the biLSTM network to suppress impulse noise; in practical applications, the output of the trained biLSTM network is the denoised vector r n ' on ;
[0087] In the DOA estimation algorithm based on statistical models, the non-Gaussian nature of impulse noise often makes it difficult for these algorithms to obtain accurate and reliable DOA estimation. In view of this, a biLSTM network is proposed. Figure 1As shown, the network includes a sequence input layer, multiple biLSTM layers, a fully connected layer, and a regression layer. To achieve the best impulse noise suppression performance, the number of hidden units in the biLSTM layer of this embodiment is set to 120.
[0088] The input of biLSTM network training process is p = {r in ,r non} data pair. In actual application, the input is r in , the output is the denoised vector r n ' on .
[0089] S4. Transform the vector r n ' on Convert to complex space and reconstruct the covariance matrix R′ of the noise-free array signal using the inverse operation non , the covariance matrix R′ non It is combined with the total least squares rotation invariant technique to achieve DOA estimation.
[0090] Specifically, for the denoised vector r n ' on It cannot be directly applied to DOA estimation, so it needs to be reconstructed into the covariance matrix corresponding to the array output signal.
[0091] First, vector r n ' on Convert to complex space:
[0092] r out =r n ' on (1:2N-1)+jr n ' on (2N:4N-2) (11)
[0093] The inverse operation is then used to reconstruct the covariance matrix of the noise-free array output signal as follows:
[0094]
[0095] In order to further reduce the computational complexity, the noise-free covariance matrix R′ non It is combined with the total least squares rotation invariant technique to achieve DOA estimation.
[0096] To verify the performance advantages of the present invention, simulation experiments were conducted under the following conditions: Two independent orthogonal phase-shift keying signal sources are incident on a uniform linear array, which includes N=6 antennas. The incident angles of the two signal sources are θ1=20° and θ2=60°, respectively. It is assumed that the characteristic index of the pulse noise contained in the array output signal satisfies 1≤α≤2. This means that the range of noise is from strong pulse Cauchy noise to pulseless Gaussian noise. Since the second-order statistics of the alpha stable distribution random variable are unbounded except for the characteristic index α=2, the generalized signal-to-noise ratio (GSNR) is used to evaluate the signal power.
[0097] The characteristic index is set to α = 1.4, from Figure 2 (a) It can be seen that, ignoring the statistical error, for two signal sources, the root mean square error (RMSE) of each algorithm gradually decreases with the generalized signal-to-noise ratio, and the performance of the present invention is always the best. Figure 2 It can be found in (b) that the change of the generalized signal-to-noise ratio has no effect on the distinguishable probability of the FLOCR-MUSIC algorithm and the present invention, that is, within the range of change of the generalized signal-to-noise ratio, both algorithms can completely and effectively separate the two incident signal sources. However, for the TLS-ESPRIT, FLOM-MUSIC and CRCO-MUSIC algorithms, the improvement of the generalized signal-to-noise ratio can enhance their ability to distinguish the two signal sources. It is particularly important to point out that when the generalized signal-to-noise ratio is 10dB, the TLS-ESPRIT algorithm still cannot completely distinguish the two signal sources, which is related to its inability to offset the influence of impulse noise. From Figure 2 It can be seen from the experimental results that the present invention is superior to the existing algorithms in terms of both DOA estimation accuracy and spatial signal source separation capability.
[0098] Embodiment 2:
[0099] This embodiment provides a low-complexity direction of arrival estimation system based on a biLSTM network, including:
[0100] A signal acquisition module, which acquires the covariance matrix R of the wireless signal;
[0101] The input vector building module expands the covariance matrix R of the wireless signal and converts it into the real number space to form a data pair p = {r in ,r non};
[0102] The impulse noise suppression module uses the data pair p = {r in ,r non} Train the biLSTM network to suppress impulse noise; in practical applications, the output of the trained biLSTM network is the denoised vector r n 'on ;
[0103] The DOA estimation module transforms the vector r n ' on Convert to complex space and reconstruct the covariance matrix R′ of the noise-free array signal using the inverse operation non , the covariance matrix R′ non It is combined with the total least squares rotation invariant technique to achieve DOA estimation.
[0104] Embodiment three:
[0105] An electronic device comprises a memory, a processor and a computer program stored and running on the memory, wherein the processor implements the above-mentioned low-complexity direction of arrival estimation method based on biLSTM network when executing the program, comprising:
[0106] Obtain the covariance matrix R of the wireless signal;
[0107] The covariance matrix R of the wireless signal is expanded and converted to the real number space to form a data pair p = {r in ,r non};
[0108] Using data pair p = {r in ,r non} Train the biLSTM network to suppress impulse noise; in practical applications, the output of the trained biLSTM network is the denoised vector r n ' on ;
[0109] The vector r n ' on Convert to complex space and reconstruct the covariance matrix R′ of the noise-free array signal using the inverse operation non , the covariance matrix R′ non It is combined with the total least squares rotation invariant technique to achieve DOA estimation.
[0110] Embodiment 4:
[0111] A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the low-complexity direction of arrival estimation method based on a biLSTM network is implemented, comprising:
[0112] Obtain the covariance matrix R of the wireless signal;
[0113] The covariance matrix R of the wireless signal is expanded and converted to the real number space to form a data pair p = {r in ,r non};
[0114] Using data pair p = {r in ,r non} Train the biLSTM network to suppress impulse noise; in practical applications, the output of the trained biLSTM network is the denoised vector r n ' on ;
[0115] The vector r n ' on Convert to complex space and reconstruct the covariance matrix R′ of the noise-free array signal using the inverse operation non , the covariance matrix R′ non It is combined with the total least squares rotation invariant technique to achieve DOA estimation.
[0116] Those skilled in the art should understand that the modules or steps of the present disclosure can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present disclosure is not limited to any specific combination of hardware and software.
[0117] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
[0118] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.
Claims
1. A low-complexity direction of arrival estimation method based on biLSTM network, characterized in that: The following steps are involved: Obtain the covariance matrix R of the wireless signal; The covariance matrix R of the wireless signal is expanded and converted to the real number space to form a data pair p = {r in ,r non }; Using data pair p = {r in ,r non } Train the biLSTM network to suppress impulse noise; in practical applications, the output of the trained biLSTM network is the denoised vector r′ non ; The vector r′ non Convert to complex space and reconstruct the covariance matrix R′ of the noise-free array signal using the inverse operation non , the covariance matrix R′ non It is combined with the total least squares rotation invariant technique to achieve DOA estimation.
2. According to claim 1, a low-complexity direction of arrival estimation method based on a biLSTM network is characterized in that: The covariance matrix R of the wireless signal is obtained as follows: Assuming that a uniform linear array ULA consists of N equally spaced antennas, the spacing between adjacent antennas is d, and there are K signal sources incident on the uniform linear array, the signal received by the array is expressed as: x(t)= A(θ) s(t)+ v(t) (1) Where A(θ) is the steering matrix, each column of which is the steering vector corresponding to each signal source direction, s(t) is the signal source signal vector, and v(t) is the noise vector; The covariance matrix R of the wireless signal is: R=E[x(t)x H (t)] (2) Where E[·] represents the expected operation, (·) H Represents the conjugate transpose operation.
3. According to claim 2, a low-complexity direction of arrival estimation method based on a biLSTM network is characterized in that: The covariance matrix R of the wireless signal is expanded as follows: Among them, σ 2 I is the covariance matrix of spatial white noise, and is the power of the kth signal source, a(θ k ) is the steering vector of the kth signal source; So we get: R non Expanded into matrix form: This matrix satisfies the following conditions: Take the matrix R non The first row and diagonal elements of form a new vector: Convert a vector to real space: Among them, Imag(·) and Real(·) represent the real part and imaginary part respectively, and max(‖·‖) represents the maximum value of the modulus.
4. According to claim 2, a low-complexity direction of arrival estimation method based on a biLSTM network is characterized in that: The first row and diagonal element of the covariance matrix R containing noise in formula (2) is r′ in ,Right now: r′ in =[R 11 ,R 12 ,…,R 1N ,R 22 ,…,R NN ] (9) Converting it to real number space is:
5. According to claim 1, a low-complexity direction of arrival estimation method based on biLSTM network is characterized in that: The biLSTM network includes a sequence input layer, multiple biLSTM layers, a fully connected layer, and a regression layer.
6. According to claim 1, a low-complexity direction of arrival estimation method based on biLSTM network is characterized in that: The vector r′ non The conversion to complex space is: <h2 style=";text-align:left;direction:ltr">r<h2 style=";text-align:left;direction:ltr"> out <h2 style=";text-align:left;direction:ltr"> 5r′<h2 style=";text-align:left;direction:ltr"> non <h2 style=";text-align:left;direction:ltr"> (1:2N-1)+jr′<h2 style=";text-align:left;direction:ltr"> non <h2 style=";text-align:left;direction:ltr"> (2N:4N-2) (11)。 7. A low-complexity direction of arrival estimation method based on biLSTM network according to claim 6, characterized in that: Reconstruct the covariance matrix R′ of the noise-free array signal using the inverse operation non :
8. A low-complexity direction-of-arrival estimation system based on a biLSTM network, characterized in that: include: A signal acquisition module, which acquires the covariance matrix R of the wireless signal; The input vector building module expands the covariance matrix R of the wireless signal and converts it into the real number space to form a data pair p = {r in ,r non }; The impulse noise suppression module uses the data pair p = {r in ,r non } Train the biLSTM network to suppress impulse noise; in practical applications, the output of the trained biLSTM network is the denoised vector r′ non ; The DOA estimation module converts the vector r′ non Convert to complex space and reconstruct the covariance matrix R′ of the noise-free array signal using the inverse operation non , the covariance matrix R′ non It is combined with the total least squares rotation invariant technique to achieve DOA estimation.
9. An electronic device comprising a memory, a processor and a computer program stored and running on the memory, characterized in that: When the processor executes the program, the low-complexity direction of arrival estimation method based on the biLSTM network is implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, a low-complexity direction of arrival estimation method based on a biLSTM network is implemented.