Multi-target DOA estimation method and system based on GRU and biLSTM

Abstract

The invention discloses a multi-target DOA estimation method and system based on GRU and biLSTM, and relates to the technical field of wireless communication. Comprising the following steps: constructing a covariance matrix of a signal receiving array; performing spatial transformation and dimension reduction processing on the covariance matrix of the signal receiving array; a DOA estimation system is established, noise suppression is achieved through a biLSTM network, space target source number estimation is achieved through a GRU network, and then multi-target DOA estimation is achieved in combination with the subspace technology. According to the method, excellent performance can be shown under various signal-to-noise ratio conditions, and the limitation of a traditional DOA estimation technology is broken through. Especially in the aspect of DOA estimation accuracy, the method significantly reduces the root-mean-square error, greatly improves the resolution probability at the same time, and can achieve more accurate and more reliable DOA estimation even in a complex and changeable signal environment.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a multi-target DOA estimation method and system based on GRU and biLSTM. Background Art

[0002] In today's era of rapid technological development, DOA estimation plays an important role in many fields, and its technological development and practical applications show diversified characteristics. In the field of wireless communications, accurate DOA estimation is crucial for smart antenna systems to achieve beamforming. By accurately measuring the DOA of the signal source, the direction of the antenna beam can be flexibly adjusted, thereby enhancing the ability to receive or transmit signals in a specific direction in a targeted manner. This not only effectively improves the utilization efficiency of spectrum resources, but also significantly improves the communication quality and greatly reduces the problem of co-channel interference. In the field of radar detection, DOA estimation is a key means to determine the target direction. Whether it is a radar system used by the military to monitor air and sea targets, or a civilian field such as meteorological radar to detect the direction of precipitation areas, accurate DOA estimation is indispensable to provide accurate location information of the target.

[0003] Traditional DOA estimation methods mainly rely on array signal processing theory. Among them, classic algorithms such as MUSIC (Multiple Signal Classification) algorithm and ESPRIT (Estimated Signal Parameter Rotation Invariant Technology) algorithm can show good performance in an ideal noise-free environment. However, in actual application scenarios, it is often difficult to achieve this ideal state. Various factors seriously restrict the accuracy of traditional DOA estimation methods. Among them, noise interference is one of the most significant problems. In a wireless communication environment, various types of noise, such as background noise and noise caused by multipath propagation, are mixed in the received signal. These noises will change the characteristics of the covariance matrix, thereby weakening the accuracy of DOA estimation based on traditional algorithms using the orthogonal relationship between signal subspace and noise subspace.

[0004] In addition, multipath effect is also an important factor that cannot be ignored. During the propagation of wireless signals, the signal may reach the receiving antenna array via multiple different paths, which makes the received signal complex and difficult to accurately parse its original DOA information. When traditional algorithms process complex signals caused by multipath effect, it is often difficult to effectively extract accurate DOA information, which has an adverse effect on the final estimation accuracy. Summary of the invention

[0005] The purpose of the present invention is to propose a multi-target DOA estimation method and system based on GRU and biLSTM, and to realize DOA estimation based on deep learning by acquiring spatial signal data.

[0006] According to a first aspect of an embodiment of the present disclosure, a multi-target DOA estimation method based on GRU and biLSTM is provided, comprising the following steps:

[0007] Constructing the covariance matrix of the signal receiving array;

[0008] Performing spatial transformation and dimensionality reduction processing on the covariance matrix of the signal receiving array;

[0009] A DOA estimation system is established, which uses the biLSTM network to achieve noise suppression and the GRU network to estimate the number of spatial target sources, and then combines it with subspace technology to achieve spatial multi-target DOA estimation.

[0010] In one embodiment, the covariance matrix of the signal receiving array is constructed as follows: the signal receiving array adopts a uniform linear array, the array includes M omnidirectional antennas, and the signal transmitted by the space wireless target is received through the array; assuming that the number of snapshots of each received signal is N, and the complex amplitude of the signal is s(t), the signal received by the mth antenna is expressed as:

[0011] x m (t) = s(t-τ m )+n m (t) (1)

[0012] Among them, τ m is the delay, n m (t) is the noise vector;

[0013] Assuming s(t) is a narrowband signal, it is approximated by a complex exponential form:

[0014] s(t)=Ae j2πft +w(t) (2)

[0015] where A is the amplitude of the signal, w(t) is any noise that may be present, and f is the frequency of the incident signal; therefore, the signal received by the mth omnidirectional antenna is reformulated as:

[0016] x m (t) = Ae j2π(m-1)dsin(θ) / λ e j2πft +n m (t) (3)

[0017] Where θ is the incident angle of the signal, λ is the wavelength of the incident signal, and d is the distance between adjacent antennas;

[0018] If x(t) is defined as a vector containing the outputs of all antennas, then the vector is expressed as:

[0019] x(t)=a(θ)s(t)+n(t) (4)

[0020] Where a(θ) is the array manifold, s(t) is the signal matrix, and n(t) is the noise matrix;

[0021] The covariance matrix corresponding to formula (4) is expressed as:

[0022] R = E[x(t)x H (t)] (5)

[0023] Among them, E[·] is the mathematical expectation operation, (·) H is the conjugate transpose operation.

[0024] A plurality of groups of signals with a snapshot number of N are sampled from the signals received by the array, and the corresponding covariance matrix is ​​obtained using equation (5) as the output of the array.

[0025] In one embodiment, the spatial transformation and dimensionality reduction processing of the covariance matrix of the signal receiving array is specifically performed as follows:

[0026]

[0027] Where R(:,i) represents the i-th column of R, ‖·‖1 represents the 1-norm, |·| represents the modulo operation, (·) T is the transpose operation, where real(·) and imag(·) represent the real part and imaginary part, respectively.

[0028] Assume that R in The corresponding covariance matrix of the noise-free signal is obtained by processing formula (6) as the vector R non , then R in and R non A data unit is denoted as {R in ,R non}; Multiple groups of {R in ,R non The data units form a data set which serves as the input of the DOA estimation system.

[0029] In one embodiment, R in The corresponding number of space target sources L n A data unit is denoted as {R in ,L n}, multiple groups of {R in ,L n The data units form a data set which serves as the input of the DOA estimation system.

[0030] In one embodiment, the biLSTM network includes a sequence input layer, two consecutive biLSTM layers, a fully connected layer and a regression layer, and its output is a noise suppressed sequence R′.

[0031] In one embodiment, the GRU network includes a sequence input layer, two consecutive GRU layers, a fully connected layer, a softmax layer and a classification layer, and R′ is used to replace the corresponding {R in ,L n}R in the data unit in Form {R′,L n} is used as the input of the GRU network, and the output is the estimated number of spatial target sources L nst .

[0032] In one embodiment, for the denoised sequence R′, to perform DOA estimation, it is first necessary to reconstruct R′ into an M×M dimensional complex valued matrix E, as follows:

[0033]

[0034]

[0035] Then, R′ re and R′ im Transform from real space to complex space:

[0036] E=R′ re +j·R′ im (9)

[0037] Finally, E and the output L of the GRU network are combined nst Combined with subspace technology, spatial multi-target DOA estimation can be achieved.

[0038] According to a second aspect of an embodiment of the present disclosure, a multi-target DOA estimation system based on GRU and biLSTM is provided, including:

[0039] A signal acquisition module constructs the covariance matrix of the signal receiving array;

[0040] A preprocessing module performs spatial transformation and dimensionality reduction processing on the covariance matrix of the signal receiving array;

[0041] The DOA estimation module uses the biLSTM network to achieve noise suppression and the GRU network to estimate the number of spatial target sources, and then combines it with subspace technology to achieve spatial multi-target DOA estimation.

[0042] 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 multi-target DOA estimation method based on GRU and biLSTM is implemented.

[0043] 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 multi-target DOA estimation method based on GRU and biLSTM is implemented.

[0044] Compared with the prior art, the above technical solution adopted by the present invention has the following advantages: the present invention integrates deep learning with array antenna technology, and provides a wave direction estimation method based on deep learning by making full use of the spatial signal data captured by the array antenna. This method can show excellent performance under various signal-to-noise ratio conditions, breaking through the limitations of traditional DOA estimation technology. In particular, in terms of the accuracy of DOA estimation, this method significantly reduces the root mean square error and greatly improves the probability of resolution. Even in complex and changeable signal environments, it can achieve more accurate and reliable DOA estimation. This technological breakthrough not only optimizes the performance indicators of DOA estimation, but also brings improvement potential to multiple fields such as wireless communications, radar detection, and sonar positioning, laying a solid foundation for achieving high-precision positioning and signal analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] 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.

[0046] Figure 1 This is a schematic diagram of the biLSTM network structure;

[0047] Figure 2 This is a schematic diagram of the GRU network structure;

[0048] Figure 3 This is a diagram of simulation experiment results. DETAILED DESCRIPTION

[0049] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0050] 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.

[0051] 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.

[0052] 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.

[0053] Embodiment 1:

[0054] The signal processing field has always been committed to extracting valuable information from complex signal environments. DOA estimation is an important branch of it, and its accuracy directly affects many subsequent application links. This embodiment provides a multi-target DOA estimation method based on GRU and biLSTM, including the following steps:

[0055] S1. Construct the covariance matrix of the signal receiving array;

[0056] Specifically, the signal receiving array used in the present invention is a uniform linear array, which includes M omnidirectional antennas, and receives the signal transmitted by the space wireless target through the array. Assuming that the number of snapshots of each received signal is N, and the complex amplitude of the signal is s(t), the signal received by the mth omnidirectional antenna can be expressed as:

[0057] x m (t) = s(t-τ m )+n m (t) (1)

[0058] Among them, τ m is the delay, n m (t) is the noise vector.

[0059] Assuming s(t) is a narrowband signal, it can be approximated by a complex exponential form:

[0060] s(t)=Ae j2πft +w(t) (2)

[0061] Where A is the amplitude of the signal, w(t) is any noise that may be present, and f is the frequency of the incident signal. Therefore, the signal received by the mth antenna can be re-expressed as:

[0062] x m (t) = Ae j2π(m-1)dsin(θ) / λ e j2πft +n m (t) (3)

[0063] Where θ is the incident angle of the signal, λ is the wavelength of the incident signal, and d is the distance between adjacent antennas.

[0064] If x(t) is defined as a vector containing all antenna outputs, then the vector can be expressed as:

[0065] x(t)=a(θ)s(t)+n(t) (4)

[0066] Where a(θ) is the array manifold, s(t) is the signal matrix, and n(t) is the noise matrix.

[0067] The covariance matrix corresponding to formula (4) is expressed as:

[0068] R = E[x(t)x H (t)] (5)

[0069] Among them, E[·] is the mathematical expectation operation, (·) H is the conjugate transpose operation.

[0070] A plurality of groups of signals with a snapshot number of N are sampled from the signals received by the array, and the corresponding covariance matrix is ​​obtained using equation (5) as the output of the array.

[0071] S2. Performing spatial transformation and dimensionality reduction on the covariance matrix of the signal receiving array;

[0072] As the input of the deep learning network, the covariance matrix R needs to be spatially transformed and dimensionally reduced as follows:

[0073]

[0074] Where R(:,i) represents the i-th column of R, (·) T is the transpose operation, where real(·) and imag(·) represent the real part and imaginary part, respectively.

[0075] Assume that R in The corresponding covariance matrix of the noise-free signal is obtained by processing formula (6) as the vector R non , then R in and R non A data unit is denoted as {R in ,R non}. Further, multiple groups of {R in ,R non} data units form a data set as the input of the deep learning network. In addition, R in The corresponding number of space target sources L n A data unit is denoted as {R in ,L n}, further, by sampling the wireless signal received in the array multiple times, multiple groups of {R in ,L n}The data units form a dataset, which serves as the input of the deep learning network.

[0076] S3. A DOA estimation system is established, which uses a biLSTM network to achieve noise suppression and a GRU network to estimate the number of spatial target sources.

[0077] like Figure 1 As shown in the figure, the biLSTM network consists of a sequence input layer, two consecutive biLSTM layers, a fully connected layer and a regression layer, and its output is the noise suppressed sequence R′. The number of hidden units in the two consecutive biLSTM layers is 64 and 128 respectively.

[0078] In order to obtain accurate DOA estimation, the GRU network is used to estimate the number of spatial target sources, such as Figure 2 As shown. Replace the corresponding {R in ,L n}R in the data unit in Form {R′,L n} is used as the input of the GRU network, and the output is the estimated number of spatial target sources L nst The network consists of a sequence input layer, two consecutive GRU layers, a fully connected layer, a softmax layer and a classification layer. The number of hidden units in the two GRU layers is 120 and 60 respectively.

[0079] For the denoised sequence R′, to perform DOA estimation, we first need to restore R′ to an M×M dimensional real-valued matrix R′ non , then we first need to reconstruct R′ into an M×M dimensional complex valued matrix E, as follows:

[0080]

[0081]

[0082] Then, R′ re and R′ im Transform from real space to complex space:

[0083] E=R′ re +j·R′ im (9)

[0084] Finally, E and the output L of the GRU network are combined nst Combined with subspace technology, spatial multi-target DOA estimation can be achieved.

[0085] Simulation experiment and result analysis: Assume that two far-field narrowband quadrature phase shift keying (QPSK) signals are incident on a uniform linear array consisting of 8 antennas. The spacing between adjacent antennas meets the half-wavelength condition, and the incident angles are θ1 = 10° and θ2 = 30° respectively.

[0086] Figure 3 is a comparison of the DOA estimation performance of different algorithms under different signal-to-noise ratios (SNRs). The number of snapshots is set to 500. Figure 3 As can be seen in (a), as the SNR increases, the root mean square error (RMSE) of various algorithms generally shows a downward trend. This phenomenon can be explained as follows: as the SNR increases, the noise is relatively weakened, which improves the DOA estimation accuracy of various algorithms, thereby reducing the root mean square error. However, the present invention has the best performance under different SNRs.

[0087] Figure 3 (b) is the variation of the resolvable probability of different algorithms with SNR. The present invention always presents a higher resolvable probability under different SNRs. At medium and high SNR levels, the present invention is superior to ESPRIT, MUSIC and root-MUSIC in the ability to accurately distinguish multiple signal sources. It can be seen that the present invention can handle complex signal environments more effectively. The experimental results prove the superiority of the present invention in DOA estimation, and its lower root mean square error and higher resolvable probability can achieve higher DOA estimation performance.

[0088] Embodiment 2:

[0089] This embodiment provides a multi-target DOA estimation system based on GRU and biLSTM, including:

[0090] A signal acquisition module constructs the covariance matrix of the signal receiving array;

[0091] A preprocessing module performs spatial transformation and dimensionality reduction processing on the covariance matrix of the signal receiving array;

[0092] The DOA estimation module uses the biLSTM network to achieve noise suppression and the GRU network to estimate the number of spatial target sources, and then combines it with subspace technology to achieve spatial multi-target DOA estimation.

[0093] Embodiment three:

[0094] An electronic device includes a memory, a processor, and a computer program stored and running on the memory, wherein the processor implements the above-mentioned multi-target DOA estimation method based on GRU and biLSTM when executing the program, including:

[0095] Constructing the covariance matrix of the signal receiving array;

[0096] Performing spatial transformation and dimensionality reduction processing on the covariance matrix of the signal receiving array;

[0097] A DOA estimation system is established, which uses the biLSTM network to achieve noise suppression, the GRU network to estimate the number of spatial target sources, and then combines the subspace technology to achieve DOA estimation of spatial multi-target sources.

[0098] Embodiment 4:

[0099] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned multi-target DOA estimation method based on GRU and biLSTM, including:

[0100] Constructing the covariance matrix of the signal receiving array;

[0101] Performing spatial transformation and dimensionality reduction processing on the covariance matrix of the signal receiving array;

[0102] A DOA estimation system is established, which uses the biLSTM network to achieve noise suppression, the GRU network to estimate the number of spatial target sources, and then combines the subspace technology to achieve DOA estimation of spatial multi-target sources.

[0103] 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.

[0104] 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.

[0105] 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 multi-target DOA estimation method based on GRU and biLSTM, characterized in that: The following steps are involved: Constructing the covariance matrix of the signal receiving array; Performing spatial transformation and dimensionality reduction processing on the covariance matrix of the signal receiving array; A DOA estimation system is established, which uses the biLSTM network to achieve noise suppression and the GRU network to estimate the number of spatial target sources, and then combines it with subspace technology to achieve spatial multi-target DOA estimation.

2. According to the multi-target DOA estimation method based on GRU and biLSTM in claim 1, it is characterized in that: The covariance matrix of the signal receiving array is constructed as follows: the signal receiving array adopts a uniform linear array, which includes M omnidirectional antennas, and receives the signal transmitted by the space wireless target through the array; Assume that the number of snapshots of the received signal each time is N, and the complex amplitude of the signal is s(t), then the signal received by the mth antenna is expressed as: x m (t)=s(t-τ m )+n m (t) (1) Among them, τ m is the delay, n m (t) is the noise vector; Assuming s(t) is a narrowband signal, it is approximated by a complex exponential form: s(t)=Ae j2πft +w(t) (2) where A is the amplitude of the signal, w(t) is any noise that may be present, and f is the frequency of the incident signal; therefore, the signal received by the mth omnidirectional antenna is reformulated as: x m (t)=Ae j2π(m-1)dsin(θ) / λ And j2πft +n m (t) (3) Where θ is the incident angle of the signal, λ is the wavelength of the incident signal, and d is the distance between adjacent antennas; If x(t) is defined as a vector containing the outputs of all antennas, then the vector is expressed as: x(t)=a(θ)s(t)+n(t) (4) Where a(θ) is the array manifold, s(t) is the signal matrix, and n(t) is the noise matrix; The covariance matrix corresponding to formula (4) is expressed as: R=E[x(t)x H (t)] (5) Among them, E[·] is the mathematical expectation operation, (·) H is the conjugate transpose operation; A plurality of groups of signals with a snapshot number of N are sampled from the signals received by the array, and the corresponding covariance matrix is ​​obtained using equation (5) as the output of the array.

3. According to the multi-target DOA estimation method based on GRU and biLSTM as claimed in claim 1, it is characterized in that: The spatial transformation and dimensionality reduction processing of the covariance matrix of the signal receiving array is specifically performed as follows: Where R(:,i) represents the i-th column of R, (·) T is the transpose operation, where real(·) and imag(·) represent the real part and imaginary part, respectively. Assume that R in The corresponding covariance matrix of the noise-free signal is processed by equation (6) to obtain the vector Rnon. Then Rin and Rnon form a data unit and are recorded as {R in ,R non }; Multiple groups of {R in ,R non The data units form a data set which serves as the input of the DOA estimation system.

4. According to claim 3, a multi-target DOA estimation method based on GRU and biLSTM is characterized in that: R in The corresponding number of space target sources L n A data unit is denoted as {R in ,L n }, multiple groups of {R in ,L n The data units form a data set which serves as the input of the DOA estimation system.

5. According to the multi-target DOA estimation method based on GRU and biLSTM as claimed in claim 1, it is characterized in that: The biLSTM network includes a sequence input layer, two consecutive biLSTM layers, a fully connected layer and a regression layer, and its output is a noise-suppressed sequence R′.

6. According to the multi-target DOA estimation method based on GRU and biLSTM as claimed in claim 1, it is characterized in that: The GRU network includes a sequence input layer, two consecutive GRU layers, a fully connected layer, a softmax layer and a classification layer, and replaces the corresponding {R in ,L n }Rin in the data unit forms {R′,L n } is used as the input of the GRU network and the output is the estimated number of spatial target sources Lnst.

7. According to claim 5, a multi-target DOA estimation method based on GRU and biLSTM is characterized in that: For the denoised sequence R′, to perform DOA estimation, it is first necessary to reconstruct R′ into an M×M dimensional complex valued matrix E, as follows: Then, R r ' e and R i ' m Transform from real space to complex space: E=R r ′ e +j·R i ′ m (9) Finally, E and the output L of the GRU network are combined nst Combined with subspace technology, spatial multi-target DOA estimation can be achieved.

8. A multi-target DOA estimation system based on GRU and biLSTM, characterized in that: include: A signal acquisition module constructs the covariance matrix of the signal receiving array; A preprocessing module performs spatial transformation and dimensionality reduction processing on the covariance matrix of the signal receiving array; The DOA estimation module uses the biLSTM network to achieve noise suppression and the GRU network to estimate the number of spatial target sources, and then combines it with subspace technology to achieve spatial multi-target 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 multi-target DOA estimation method based on GRU and biLSTM 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 multi-target DOA estimation method based on GRU and biLSTM is implemented.

Images