Space target direction finding method and system based on deep learning

Abstract

The invention discloses a space target direction finding method and system based on deep learning, and relates to the technical field of DOA (direction of arrival) estimation. Comprising the following steps: acquiring a covariance matrix of a wireless signal; converting the covariance matrix of the wireless signal from a complex number space to a real number space and carrying out serialization processing; a deep learning network model is constructed, the model comprises a DOA large interval classification module and a high-precision DOA estimation module, and the input of each module comprises two parts: one part is a real value sequence of a wireless signal, and the other part is an incident angle corresponding to each real value sequence; and realizing DOA estimation by using the trained deep learning network model. According to the invention, high-precision DOA estimation of the space target is realized through the deep learning network model, and the method is suitable for a multi-target-source scene.

Description

Technical Field

[0001] The present invention relates to the technical field of direction of arrival (DOA) estimation, and in particular to a space target direction finding method and system based on deep learning. Background Art

[0002] In today's signal processing and target positioning fields, DOA estimation occupies a pivotal position and plays a key role in accurately obtaining the azimuth information of the signal source. It is widely used in many important fields such as wireless communications, radar detection, and acoustic monitoring. Traditional DOA estimation methods, such as subspace-based algorithms (such as the MUSIC algorithm and the ESPRIT algorithm) and beamforming methods, can achieve certain estimation results under ideal conditions. However, with the increasing complexity of actual application scenarios, these traditional methods have gradually shown many limitations. On the one hand, in complex noise environments, such as impulse noise and Gaussian mixed noise, the traditional methods have weak anti-interference capabilities, and noise can easily mask signal characteristics, resulting in a significant decrease in the accuracy of DOA estimation. On the other hand, in the face of multipath propagation effects, that is, the signal reaches the receiving end after multiple reflections and refractions, it is difficult for traditional methods to accurately distinguish signals from different paths, and thus it is impossible to accurately estimate the true DOA information.

[0003] At the same time, deep learning technology has made rapid progress in recent years and has demonstrated excellent performance in many fields. With its powerful nonlinear mapping capabilities and the ability to automatically learn features from large amounts of data, deep learning provides new ideas and methods for solving complex signal processing problems.

[0004] Although there have been some related explorations in the research direction of using deep learning to achieve DOA estimation, there is still room for further optimization. For example, some existing methods often use a single neural network architecture for overall estimation when dealing with DOA estimation tasks, which makes it difficult to perform targeted processing at stages with different accuracy requirements. This results in either the inability to quickly lock the approximate DOA range in the rough estimation stage, which makes the subsequent precise estimation calculation too large and susceptible to error accumulation; or too much emphasis is placed on the initial rough estimation, while ignoring the key role of the precise estimation link in the accuracy of the final result. Summary of the invention

[0005] The purpose of the present invention is to propose a space target direction finding method and system based on deep learning, which realizes high-precision DOA estimation of space targets through a deep learning network model and is suitable for scenarios with multiple target sources.

[0006] According to a first aspect of an embodiment of the present disclosure, a space target direction finding method based on deep learning is provided, comprising the following steps:

[0007] Obtaining a covariance matrix of a wireless signal;

[0008] Convert the covariance matrix of the wireless signal from the complex space to the real space and perform serialization processing;

[0009] A deep learning network model is constructed, which includes a DOA large interval classification module and a high-precision DOA estimation module. The input of each module contains two parts: one is the real-valued sequence of the wireless signal, and the other is the incident angle corresponding to each real-valued sequence. The trained deep learning network model is used to realize DOA estimation.

[0010] In one embodiment, the DOA large interval classification module includes a connected sequence input layer, a first GRU layer, a second GRU layer, a fully connected layer, a softmax layer and a classification layer.

[0011] In one embodiment, the DOA large interval classification module divides the incident angle θ into a large interval [0°: 5°: 90°], and the corresponding training labels are:

[0012]

[0013] in, Indicates rounding down.

[0014] In one embodiment, the high-precision DOA estimation module includes multiple GRU sub-networks, referred to as subGRU i , i = 1, 2, ..., 17; each GRU sub-network includes: a connected sequence input layer, a first GRU layer, a second GRU layer, a third GRU layer, a fully connected layer, a softmax layer and a classification layer.

[0015] In one embodiment, each GRU subnetwork corresponds to three small intervals divided by the DOA large interval classification module, covering a range of 15°; two adjacent GRU subnetworks overlap by 5°, and each GRU subnetwork divides the range of 15° into intervals of 0.1°.

[0016] In one embodiment, the DOA estimated by the DOA large interval classification module is θ in , then the GRU sub-network selection principle is:

[0017]

[0018] In one embodiment, for the output of the high-precision DOA estimation module, it is assumed that the estimated DOA is θ e , the corresponding predicted probability is p; assuming the estimated DOA is θ e -0.1°, the corresponding predicted probability is p 1, the estimated DOA is θ e +0.1°, the corresponding predicted probability is p 2 , then the final result of DOA estimation is:

[0019] θ DOA =θ e ·p+(θ e -0.1°)·p 1 +(θ e +0.1°)·p 2 .

[0020] According to a second aspect of an embodiment of the present disclosure, a space target direction finding system based on deep learning is provided, comprising:

[0021] A wireless signal acquisition component, which acquires the covariance matrix of the wireless signal;

[0022] A signal preprocessing component converts the covariance matrix of the wireless signal from the complex space to the real space and performs serialization processing;

[0023] The data processing component builds a deep learning network model, which includes a DOA large-range classification module and a high-precision DOA estimation module. The input of each module contains two parts: one is the real-valued sequence of the wireless signal, and the other is the incident angle corresponding to each real-valued sequence; the trained deep learning network model is used to realize DOA estimation.

[0024] 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 method for direction finding of a space target based on deep learning is implemented.

[0025] 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 method for direction finding of a space target based on deep learning is implemented.

[0026] Compared with the prior art, the above technical scheme adopted by the present invention has the following advantages: the present application realizes high-precision DOA estimation of space targets through a deep learning network model, and is suitable for scenarios with multiple target sources; in the field of wireless communications, it helps to accurately locate the signal transmission source, optimize the pointing direction of the base station antenna, and improve the communication quality; in the field of radar detection, it can more accurately determine the direction of the target and improve the detection efficiency of the radar; in the field of acoustic monitoring, such as the positioning of underwater sound sources or aerial noise sources, it can also play an important role, providing new technical support and solutions for the development of related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] 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.

[0028] Figure 1 This is a schematic diagram of the overall structure of the deep learning network model;

[0029] Figure 2 This is a schematic diagram of the DOA large-interval classification module structure;

[0030] Figure 3 Schematic diagram of the high-precision DOA estimation module structure. DETAILED DESCRIPTION

[0031] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0032] 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.

[0033] 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.

[0034] 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.

[0035] Embodiment 1:

[0036] The essence of DOA estimation is to accurately determine the direction information of the signal source from the received signal. Traditional DOA estimation methods often rely on complex mathematical models and manual feature extraction. In the face of complex and changeable practical application scenarios, such as strong interference and multipath effects, the estimation accuracy and stability will be greatly affected. Therefore, this embodiment provides a space target direction finding method based on deep learning, including the following steps:

[0037] S1. Obtain the covariance matrix of the wireless signal;

[0038] Specifically, assume that the signal receiving array includes N antennas, which are evenly placed on the circumference. M signal sources are incident on the array at the same time. The signal received by the array is:

[0039]

[0040] Where x(t) is the received signal vector, a(θ m ,φ m ) is the steering vector of the mth source, θ m and φ m They represent the azimuth and elevation angles of the mth source respectively, and n(t) is the noise vector.

[0041] For a sufficient number of snapshots T, the covariance matrix R of the wireless signal is x It can be estimated as:

[0042]

[0043] in,(·) H represents the conjugate transpose.

[0044] S2. Convert the covariance matrix of the wireless signal from the complex space to the real space and perform serialization processing;

[0045] Using a deep learning network for direction finding is essentially a mapping: mapping the array output signal to the DOA of the target source. For the uniform circular array containing N antennas, the covariance matrix R obtained according to formula (2) is x is complex-valued. Since the input of the deep learning network is a real-valued sequence, R x It is necessary to convert from complex space to real space and serialize.

[0046] R x Transformed into real number space, it can be expressed as:

[0047] R′ x =[real(R x ),imag(R x )] (3)

[0048] Here, real(·) and imag(·) represent the real part and imaginary part, respectively.

[0049] R′ x The serialization can be done as follows:

[0050] First, by x (i,:) is defined as R′ x The i-th row of (where i = 1, 2, ..., N) is obtained as follows:

[0051] R′=[R′ x (1,:),R′ x (2,:),…,R′ x (N,:)] (4)

[0052] Then, by normalizing R′, we can get:

[0053]

[0054] Where max(·) represents the maximum value of the modulus.

[0055] In order to handle the case where the uniform circular array contains different numbers of antennas, it is necessary to expand equation (5). Assume that the maximum number of antennas contained in the uniform circular array is N max , that is, N max ≥N. Then, R″ will be rewritten as:

[0056]

[0057] Define the DOA corresponding to R as P, construct The number pairs are used as a dataset for training and validating deep learning networks, where U is the total number of number pairs.

[0058] S3. Figure 1 As shown, a deep learning network model is constructed, which includes a DOA large interval classification module and a high-precision DOA estimation module, and the trained deep learning network model is used to realize DOA estimation.

[0059] Deep learning based DOA estimation is regarded as a classification problem; therefore, the minimum margin of angle classification determines the accuracy of DOA estimation.

[0060] like Figure 2As shown in FIG. 1 , the DOA large interval classification module includes a sequence input layer, two gated recurrent unit (GRU) layers, a fully connected layer, a softmax layer, and a classification layer. The input of this module consists of two parts: one is the real-value sequence of the wireless signal, and the other is the incident angle corresponding to each real-value sequence; in this module, the incident angle θ is divided into a large interval [0°: 5°: 90°], and its corresponding training labels are:

[0061]

[0062] in, Indicates rounding down.

[0063] like Figure 3 As shown in Figure 2, the high-precision DOA estimation module contains multiple GRU sub-networks, called subGRU i (i=1,2,…,17); Each GRU subnetwork consists of seven layers: a sequence input layer, three GRU layers, a fully connected layer, a softmax layer, and a classification layer. Similar to the DOA large interval classification module, during the training of the GRU subnetwork, the input data contains two parts: one is the real-valued sequence of the wireless signal, and the other is the incident angle corresponding to each real-valued sequence. In order to obtain high-precision DOA estimation, each GRU subnetwork corresponds to three small intervals divided by the DOA large interval classification module, covering a range of 15°. Two adjacent GRU subnetworks overlap by 5°, and each GRU subnetwork divides the range of 15° into intervals of 0.1°.

[0064] Assume that the DOA estimated by the DOA large interval classification module is θ in , then the GRU sub-network selection principle (i.e., selector control principle) is:

[0065]

[0066] For the output of the high-precision DOA estimation module, in order to further improve the accuracy of DOA estimation, it is assumed that the estimated DOA is θ e , the corresponding predicted probability is p; the estimated DOA is θ e -0.1°, the corresponding predicted probability is p 1 , the estimated DOA is θ e +0.1°, the corresponding predicted probability is p 2 , then the final result of DOA estimation is:

[0067] θ DOA =θ e ·p+(θ e -0.1°)·p 1 +(θ e+0.1°)·p 2 (9)

[0068] After the training and verification of the deep learning network model is completed, the preprocessed data of the signal of the space target received by the uniform circular array is used as the input of the deep learning network model. The output is processed by formula (9) to obtain high-precision DOA estimation, thereby completing the direction finding task of the space target source.

[0069] Embodiment 2:

[0070] This embodiment provides a space target direction finding system based on deep learning, including:

[0071] A wireless signal acquisition component, which acquires the covariance matrix of the wireless signal;

[0072] A signal preprocessing component converts the covariance matrix of the wireless signal from the complex space to the real space and performs serialization processing;

[0073] The data processing component builds a deep learning network model, which includes a DOA large-range classification module and a high-precision DOA estimation module. The input of each module contains two parts: one is the real-valued sequence of the wireless signal, and the other is the incident angle corresponding to each real-valued sequence; the trained deep learning network model is used to realize DOA estimation.

[0074] Embodiment three:

[0075] 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 method for space target direction finding based on deep learning when executing the program, including:

[0076] Obtaining a covariance matrix of a wireless signal;

[0077] Convert the covariance matrix of the wireless signal from the complex space to the real space and perform serialization processing;

[0078] A deep learning network model is constructed, which includes a DOA large interval classification module and a high-precision DOA estimation module. The input of each module contains two parts: one is the real-valued sequence of the wireless signal, and the other is the incident angle corresponding to each real-valued sequence. The trained deep learning network model is used to realize DOA estimation.

[0079] Embodiment 4:

[0080] A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the method for space target direction finding based on deep learning is implemented, including:

[0081] Obtaining a covariance matrix of a wireless signal;

[0082] Convert the covariance matrix of the wireless signal from the complex space to the real space and perform serialization processing;

[0083] A deep learning network model is constructed, which includes a DOA large interval classification module and a high-precision DOA estimation module. The input of each module contains two parts: one is the real-valued sequence of the wireless signal, and the other is the incident angle corresponding to each real-valued sequence. The trained deep learning network model is used to realize DOA estimation.

[0084] 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.

[0085] 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.

[0086] 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 space target direction finding method based on deep learning, characterized in that: The following steps are involved: Obtaining a covariance matrix of a wireless signal; Convert the covariance matrix of the wireless signal from the complex space to the real space and perform serialization processing; A deep learning network model is constructed, which includes a DOA large interval classification module and a high-precision DOA estimation module. The input of each module contains two parts: one is the real-valued sequence of the wireless signal, and the other is the incident angle corresponding to each real-valued sequence. The trained deep learning network model is used to realize DOA estimation.

2. According to claim 1, a space target direction finding method based on deep learning is characterized in that: The DOA large interval classification module includes a connected sequence input layer, a first GRU layer, a second GRU layer, a fully connected layer, a softmax layer and a classification layer.

3. According to claim 1, a space target direction finding method based on deep learning is characterized in that: The DOA large interval classification module divides the incident angle θ into intervals [0°:5°:90°], and the corresponding training labels are: in, Indicates rounding down.

4. According to claim 1, a space target direction finding method based on deep learning is characterized in that: The high-precision DOA estimation module includes multiple GRU sub-networks, called subGRU i , i = 1, 2, ..., 17; each GRU sub-network includes: a connected sequence input layer, a first GRU layer, a second GRU layer, a third GRU layer, a fully connected layer, a softmax layer and a classification layer.

5. The method for space target direction finding based on deep learning according to claim 4, characterized in that: Each GRU subnetwork corresponds to three small intervals divided by the DOA large interval classification module, covering a range of 15°; two adjacent GRU subnetworks overlap by 5°, and each GRU subnetwork divides the 15° range into intervals of 0.1°.

6. According to claim 4, a space target direction finding method based on deep learning is characterized in that: Assume that the DOA estimated by the DOA large interval classification module is θ in , then the GRU sub-network selection principle is:

7. The method for space target direction finding based on deep learning according to claim 1, characterized in that: For the output of the high-precision DOA estimation module, assume that the estimated DOA is θ e , the corresponding predicted probability is p; the estimated DOA is θ e -0.1°, the corresponding predicted probability is p1, and the estimated DOA is θ e +0.1°, the corresponding predicted probability is p2, then the final result of DOA estimation is: i DOA =θ e ·p+(θ e -0.1°)·p1+(θ e +0.1°)·p2.

8. A space target direction finding system based on deep learning, characterized in that: include: A wireless signal acquisition component, which acquires the covariance matrix of the wireless signal; A signal preprocessing component converts the covariance matrix of the wireless signal from the complex space to the real space and performs serialization processing; The data processing component builds a deep learning network model, which includes a DOA large-range classification module and a high-precision DOA estimation module. The input of each module contains two parts: one is the real-valued sequence of the wireless signal, and the other is the incident angle corresponding to each real-valued sequence; the trained deep learning network model is used to realize 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 space target direction finding method based on deep learning 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 space target direction finding method based on deep learning is implemented.

Images