Multi-antenna radiation source direction finding method using nonlinear processing capability of neural network

Abstract

The invention discloses a multi-antenna radiation source direction finding method using neural network nonlinear processing capability, and relates to the technical field of electronic countermeasure, passive positioning and radar signal processing, and the method comprises the steps: training a preset neural network model through a constructed training sample until the neural network model reaches a preset training end condition; the method comprises the following steps: acquiring an unknown radiation source signal in a multi-channel acquisition mode, and extracting and acquiring a multi-dimensional feature vector from the unknown radiation source signal; inputting the feature vector into a neural network model meeting a training end condition for processing to obtain a radiation source azimuth angle of the unknown radiation source signal; the method does not depend on an ideal antenna array model, and can establish a complex mapping relation between the signal characteristics measured under the actual non-ideal hardware condition and the azimuth angle of the radiation source through a learning mode, thereby remarkably improving the robustness and precision of a direction finding system in a real environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electronic countermeasures, passive location and radar signal processing, and more particularly to a method for multi-antenna emitter direction finding using the nonlinear processing capability of a neural network. BACKGROUND

[0002] Emitter direction finding is a core technology in the fields of electronic reconnaissance, spectrum monitoring, passive location, etc. Traditional direction finding methods, such as the amplitude comparison method, interferometer method, beamforming method, MUSIC subspace algorithm, etc., are all based on an ideal antenna array model. These models strictly assume that:

[0003] 1. The antenna array elements have completely consistent amplitude and phase responses.

[0004] 2. The geometric position relationship between the antenna array elements is accurately known.

[0005] 3. The antenna array is in free space and is not affected by the surrounding environment (such as a radome).

[0006] However, in actual engineering practice, the above ideal assumptions are difficult to meet:

[0007] 1. Antenna and microwave component inconsistency: Each antenna array element and its subsequent microwave channel (including amplifiers, filters, mixers, etc.) will inevitably have manufacturing tolerances and temperature drifts, resulting in inconsistent amplitude gains and phase delays of each channel.

[0008] 2. Radome effect: To protect the antenna, a radome is usually installed. The radome will change the propagation path of electromagnetic waves, causing beam distortion, scattering, and inter-element coupling, which severely damages the original directional pattern of the antenna array and the phase relationship between the elements.

[0009] 3. Layout constraints: On some platforms (such as drones, missiles), the installation position of the antenna is strictly limited, and the theoretically optimal regular layout (such as a uniform linear array or circular array) cannot be achieved.

[0010] To overcome these non-ideal factors, traditional methods usually require a complex "calibration" process: measure the directional pattern of the array in a darkroom and construct a calibration table for compensation when actually used; but the calibration process is tedious, and when the environmental temperature and frequency change, the calibration data may be invalid, resulting in a sharp decrease in direction finding accuracy.

[0011] Therefore, there is an urgent need for a new direction finding method that can fundamentally "immunize" or "adapt" to these hardware non-idealities. SUMMARY

[0012] The application aims to provide a multi-antenna radiation source direction finding method using the nonlinear processing capability of a neural network, which is not dependent on an ideal antenna array model and can establish a complex mapping relationship between signal characteristics measured under actual non-ideal hardware conditions and the azimuth of the radiation source through learning, thereby significantly improving the robustness and accuracy of the direction finding system in a real environment.

[0013] The above technical objective of the application is achieved by the following technical solutions.

[0014] In a first aspect, the application provides a multi-antenna radiation source direction finding method using the nonlinear processing capability of a neural network, comprising the following specific steps:

[0015] The preset neural network model is trained by using the constructed training samples until the neural network model reaches the preset training end condition;

[0016] An unknown radiation source signal is acquired by a multi-channel acquisition method, and a multi-dimensional feature vector is extracted from the unknown radiation source signal;

[0017] The feature vector is input into the neural network model that has reached the training end condition for processing, and the azimuth of the unknown radiation source signal is obtained.

[0018] On the basis of the above technical solutions, the application can also be improved as follows.

[0019] Further, the training samples are obtained by the following method:

[0020] An actual direction finding system comprising at least two antenna elements is constructed;

[0021] One or more calibration radiation sources with known azimuths are placed at the measured azimuths;

[0022] The calibration radiation sources emit signals at each azimuth angle at a preset interval within a 0°-360° azimuth range;

[0023] The signals emitted at each azimuth angle are acquired by the actual direction finding system through a multi-channel acquisition method, and the amplitudes and phases of the signals emitted at each azimuth angle are calculated;

[0024] For the signals emitted at each azimuth angle, a multi-dimensional feature vector is synthesized by the corresponding amplitudes and phases;

[0025] The feature vectors of the signals at each azimuth angle and the corresponding azimuth angles are used to construct training samples.

[0026] Further, the actual direction finding system comprises at least the following components in a connection relationship: a radome, an antenna element, a microwave receiving channel, and a signal processing unit.

[0027] Further, the feature vector is obtained by the following way:

[0028] For the signals obtained by each channel, the signals are digitally down-converted to baseband to obtain complex signals, and the corresponding amplitudes and phases are calculated by the complex signals;

[0029] For each azimuth angle, a multi-dimensional feature vector is constructed by the amplitudes and phases of each channel.

[0030] Further, the multi-dimensional feature vector is specifically: F = [A1, A2,…, An, Δφ2, Δφ3, …,Δφn]; In the formula, F is the feature vector, An is the amplitude or amplitude difference of the signal collected by the nth channel, and Δφn is the relative phase of the signals collected by each channel and the signal collected by the first channel, wherein: Δφ2 = φ2 - φ1, Δφ3 = φ3- φ1, Δφn = φn - φ1.

[0031] Further, the amplitudes and phases are calculated by the following way: ; ; In the formula, denotes the amplitude of the signal collected by the nth channel, denotes the phase of the signal collected by the nth channel, denotes the quadrature component in the complex signal of the nth channel, is the in-phase component in the complex signal of the nth channel, is a square root function, is a four-quadrant arctangent function.

[0032] Further, the training end condition is that the number of iterations is reached or the loss function is not more than a threshold value, and the loss function is: ; In the formula, is the true value of the azimuth angle of the radiation source, denotes the predicted value of the azimuth angle of the radiation source.

[0033] In a second aspect, the application provides a multi-antenna radiation source direction finding system using the nonlinear processing capability of a neural network, which is applied to the multi-antenna radiation source direction finding method using the nonlinear processing capability of a neural network in any one of the first aspect.

[0034] The model training module is configured to train the preset neural network model by using the constructed training samples until the neural network model reaches the preset training end condition.

[0035] a feature vector construction module, configured to acquire an unknown radiation source signal through a multi-channel acquisition mode, and extract a multi-dimensional feature vector from the unknown radiation source signal;

[0036] an azimuth angle determination module, configured to input the feature vector into the neural network model reaching the training end condition for processing to obtain a radiation source azimuth angle of the unknown radiation source signal.

[0037] In a third aspect, the present application provides an electronic device, comprising: at least one processor, at least one memory and a data bus;

[0038] The processor and the memory complete mutual communication through the data bus; the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the method of any one of the first aspect.

[0039] In a fourth aspect, the present application provides a non-transitory computer readable storage medium, which stores computer instructions, and the computer instructions make the computer execute the method of any one of the first aspect.

[0040] Compared with the prior art, the present application has at least the following beneficial effects:

[0041] 1. High precision and strong robustness: the present application directly learns the input-output relationship of a real hardware system (including non-ideal factors), and the powerful nonlinear fitting capability of the neural network can accurately model and compensate for complex errors introduced by the antenna cover, channel inconsistency, etc., thereby obtaining higher direction finding precision and stability than traditional methods in a real environment.

[0042] 2. No strict requirement for antenna layout: the method does not depend on any specific array geometry model, so the antenna elements can be arranged arbitrarily, even using a sparse random array, greatly increasing the flexibility of system design, especially suitable for space-limited platforms.

[0043] 3. Free from complex real-time calibration: once the neural network model is trained, it does not need to be calibrated in real time during operation, reducing the requirement for system stability and simplifying the operation and maintenance process.

[0044] 4. Potential super-resolution capability: the neural network may learn features beyond the resolution limit of traditional methods from data, and is expected to achieve super-resolution direction finding. BRIEF DESCRIPTION OF DRAWINGS

[0045] The drawings described herein are used to provide further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation of the embodiments of the present application. In the drawings:

[0046] Figure 1A method flow chart of the direction finding method in the embodiment of the present application;

[0047] Figure 2 A structure schematic diagram of the neural network model in the embodiment of the present application;

[0048] Figure 3 A connection schematic diagram of the lateral system in the embodiment of the present application;

[0049] Figure 4 A connection schematic diagram of the electronic device in the embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to make a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0051] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present application.

[0052] It should be noted that: similar reference numbers and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0053] In the description of the embodiments of the present application, “a plurality of” represents at least 2.

[0054] In the description of the embodiments of the present application, it should be further noted that, unless explicitly defined and limited, if the terms “set”, “install”, “connect”, “connect” appear, they should be understood in a broad sense, for example, can be fixedly connected, can be detachably connected, or integrally connected; can be mechanically connected, can be electrically connected; can be directly connected, can be indirectly connected through an intermediate medium, can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0055] Embodiment 1: In order to solve the problem that the current traditional direction finding method relies on an ideal antenna array model and needs to perform a complex calibration process, and the calibration process is cumbersome, and when the environmental temperature and frequency change, the calibration data may be invalid, resulting in a sharp decline in direction finding accuracy, the embodiment provides a multi-antenna radiation source direction finding method using the nonlinear processing capability of a neural network, and the method can be directly applied to two-dimensional antennas, such as Figure 1 As shown, comprising the following specific steps:

[0056] S1, training a preset neural network model by constructing a training sample until the neural network model reaches a preset training end condition.

[0057] Optionally, the training sample is obtained by the following method:

[0058] S11, constructing an actual direction finding system comprising at least two antenna elements.

[0059] Wherein, the actual direction finding system comprises at least the following components connected with each other: antenna cover, antenna element, microwave receiving channel and signal processing unit.

[0060] S12, placing one or more calibration radiation sources with known orientations at the measured orientations.

[0061] S13, emitting signals at each orientation angle by stepping at a preset interval within the 0°-360° orientation range by the calibration radiation source.

[0062] Specifically, an actual direction finding system comprising N (N≥2) antenna elements can be constructed, which generally comprises an antenna cover, an antenna element, a microwave receiving channel and a signal processing unit; in a microwave darkroom or an open field with known orientations, a calibration radiation source with a known orientation is placed at the measured orientation, and the radiation source emits signals at each orientation angle by stepping at a certain interval (such as 1°) within the 0°-360° orientation range, while the system synchronously collects and records the digital baseband or intermediate frequency signals received by all N channels.

[0063] Wherein, the environment for obtaining the training sample can be a microwave darkroom to exclude environmental interference such as multipath reflection; the radiation source uses a signal source and a standard horn antenna to emit a single-carrier continuous wave (CW) signal at a frequency of 3 GHz.

[0064] S14, acquiring the signals emitted at each orientation angle by the actual direction finding system through multi-channel acquisition, and calculating the amplitudes and phases of the signals emitted at each orientation angle.

[0065] Wherein, the radiation source is installed on a high-precision rotary table. The rotary table rotates in steps of 1° in a range of 0° to 359°. At each azimuth angle θ_i (i = 0, 1, 2,..., 359), the rotary table is stationary, and the system collects and records signals for a period of time.

[0066] S15, for each azimuth angle, a multi-dimensional feature vector is synthesized by corresponding amplitudes and phases.

[0067] Optionally, the above feature vector is obtained by the following method:

[0068] S151, for the signals obtained by each channel, the signals are digitally down-converted to baseband to obtain complex signals, and the corresponding amplitudes and phases are calculated by the complex signals.

[0069] Wherein, the above amplitudes and phases are calculated by the following method: ; ; In the formula, represents the amplitude of the signal collected by the nth channel, represents the phase of the signal collected by the nth channel, represents the quadrature component in the complex signal of the nth channel, is the in-phase component in the complex signal of the nth channel, is a square root function, is a four-quadrant arctangent function.

[0070] S152, for each azimuth angle, a multi-dimensional feature vector is constructed by the amplitudes and phases of each channel.

[0071] Wherein, the above multi-dimensional feature vector is specifically: F = [A1, A2,..., An, Δφ2, Δφ3,..., Δφn]; In the formula, F is a feature vector, An is the amplitude of the signal collected by the nth channel or the amplitude difference, wherein the amplitude difference can select a certain antenna as an amplitude reference antenna, and the difference between the other antennas and the reference antenna is the amplitude difference; Δφn is the relative phase of the signals collected by each channel and the signal collected by the first channel, wherein: Δφ2 = φ2 - φ1, Δφ3 = φ3 - φ1, Δφn = φn - φ1; Similarly, Δφn can also be expressed as a phase difference, that is, a certain antenna is taken as a phase reference antenna, and the phase difference between other antennas and the antenna is obtained.

[0072] Wherein, a four-element antenna array working in S-band (2-4 GHz) can be adopted; four antenna elements are arranged in an asymmetric, non-uniform manner on a circuit board to simulate an arbitrary layout adopted in actual engineering due to space limitations. A radome of a specific material is installed outside the array. Each antenna channel is followed by an independent microwave receiving front end (including a low-noise amplifier, a mixer, and a filter), and due to component tolerances, there is inherent amplitude and phase inconsistency between channels. After sampling by an ADC, the signal is sent to an FPGA (such as Xilinx Zynq UltraScale+ RFSoC) for real-time processing.

[0073] Specifically, in the FPGA of the signal processing process, the signal of each channel is digitally down-converted (DDC) to baseband to obtain a complex signal I + jQ. Then the amplitude and phase of each channel signal are calculated; when constructing the feature vector, in order to reduce the drift of the absolute phase with frequency and initial time, we take the phase of channel 1 as the reference to calculate the relative phase. Finally, for each azimuth angle, a 7-dimensional feature vector F is generated, and the four channels are:

[0074] F = [A1, A2, A3, A4, Δφ2, Δφ3, Δφ4]; wherein A1 to A4 are the amplitude values of the four channels, Δφ2 = φ2 - φ1, Δφ3 = φ3 - φ1, Δφ4 = φ4 - φ1.

[0075] S16, a training sample is constructed by the feature vector of the signal at each azimuth angle and the corresponding azimuth angle.

[0076] Specifically, the multi-channel signal collected at each azimuth angle is processed; for the signal of each channel (element), the amplitude A_n (n = 1, 2,..., N) and the phase φ_n are calculated; the N amplitude values and N phase values are combined in order to form a 2N-dimensional feature vector F = [A1, A2,..., AN, φ1, φ2,..., φN]; the feature vector F is associated with the true azimuth angle label to form a training sample (F, θ_i); by traversing all azimuth angles, a complete training data set can be constructed.

[0077] Wherein, data of 360 azimuth angles can be collected, and 100 snapshots are collected at each azimuth angle to average random noise; the final data set contains 36,000 samples, 80% of the data (28,800 samples) is divided into a training set, and 20% of the data (7,200 samples) is divided into a test set.

[0078] Optionally, the constructed neural network model above can be a multi-layer perceptron (MLP); the number of nodes in the input layer of the network is 2N (corresponding to the dimension of the feature vector), and the number of nodes in the output layer is 1 (corresponding to the estimated value of the azimuth angle, and for omnidirectional direction finding, a softmax output can also be used for multiple angle classification); using the training data set, the network is trained through an optimization algorithm such as back propagation; the goal of training is to let the network learn to predict the correct azimuth angle from the input amplitude / phase features, and the training process is essentially to "remember" all hardware nonlinear effects such as radome, antenna inconsistency in the weights of the network model.

[0079] Specifically, the structure of the neural network model is as shown in Figure 2 The specific parameters of the network are as follows:

[0080] Input layer: 7 nodes corresponding to a 7-dimensional feature vector.

[0081] Hidden layer: 3 fully connected layers.

[0082] First hidden layer: 128 neurons, activation function is ReLU.

[0083] Second hidden layer: 64 neurons, activation function is ReLU.

[0084] Third hidden layer: 32 neurons, activation function is ReLU.

[0085] Output layer: 1 neuron, activation function is linear activation function, directly outputting the estimated value of the azimuth angle (unit: degree).

[0086] Training configuration:

[0087] Loss function: mean squared error (MSE), i.e. ; in the formula, is the true value of the azimuth angle of the radiation source, represents the predicted value of the azimuth angle of the radiation source.

[0088] Optimizer: Adam, initial learning rate set to 0.001.

[0089] Batch size: 32.

[0090] Training period (Epochs): 500. Early stopping strategy is set, and when the validation set loss does not decrease for 10 consecutive periods, the training is automatically terminated to prevent overfitting.

[0091] Data preprocessing: Before training, the input feature vector is standardized, that is, the mean of each feature dimension is subtracted and divided by its standard deviation, so that the data mean is 0 and the variance is 1, to speed up network convergence.

[0092] Training environment: The model is trained on a desktop computer using the Matlab deep learning framework. After training, the final model weight and structure are fixed.

[0093] S2, the unknown radiation source signal is obtained by a multi-channel acquisition method, and a multi-dimensional feature vector is extracted from the unknown radiation source signal.

[0094] S3, the feature vector is input into the neural network model that reaches the training end condition for processing to obtain the radiation source azimuth of the unknown radiation source signal.

[0095] Among them, the trained neural network model can be deployed to an actual signal processing unit (such as FPGA, GPU or DSP); when an unknown radiation source signal arrives, the receiving system collects multi-channel signals in the same way, and extracts a 2N-dimensional feature vector F_real-time in real time; F_real-time is input into the trained neural network model, and the output value θ_estimated of the model is the final calculated radiation source azimuth.

[0096] In order to verify the effect of the application, the MLP direction finding method and the traditional interferometer direction finding method can be compared on the same hardware platform. The interferometer method assumes that the channels are ideal and uses an ideal array manifold to solve the angle, and the comparison results are shown in Table 1:

[0097] Table 1 Direction finding method Average direction finding error (test set) Maximum direction finding error (test set) Notes Conventional interferometer method 5.2° >20° Due to the antenna cover and channel inconsistency, the error is large, especially in a certain direction, there are jumps. The invention (MLP method) 0.8° 3.5° The accuracy is significantly improved, the performance is stable, and there is no jump.

[0098] As can be seen from Table 1, the method based on artificial neural network proposed in the embodiment can effectively learn and compensate the nonlinear amplitude and phase errors introduced by the arbitrary layout of the antenna, the antenna cover and the microwave components, and realize a direction finding precision and stability much higher than the traditional method under actual non-ideal hardware conditions.

[0099] Embodiment 2: The application provides a multi-antenna radiation source direction finding system using the nonlinear processing capability of a neural network, as shown in Figure 3 The multi-antenna radiation source direction finding method using the nonlinear processing capability of a neural network applied in embodiment 1 comprises:

[0100] The model training module is configured to train the preset neural network model by using the constructed training sample until the neural network model reaches the preset training end condition.

[0101] a feature vector construction module, configured to acquire an unknown radiation source signal through a multi-channel acquisition mode, and extract a multi-dimensional feature vector from the unknown radiation source signal;

[0102] an azimuth angle determination module, configured to input the feature vector into the neural network model reaching the training end condition for processing, to obtain a radiation source azimuth angle of the unknown radiation source signal.

[0103] Embodiment 3: The embodiment of the present application provides an electronic device, as shown in the figure, comprising at least one processor, at least one memory and a data bus. Figure 4

[0104] The processor and the memory complete mutual communication through the data bus; the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the method in Embodiment 1.

[0105] Embodiment 4: The embodiment of the present application provides a non-transitory computer readable storage medium, which stores computer instructions, and the computer instructions make the computer execute the method in Embodiment 1.

[0106] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can adopt a computer program product in the form of one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0107] The present application is described with reference to flowcharts and / or block diagrams according to the method, device (system) and computer program product of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one flow or multiple flows and / or blocks Figure 1 The device for implementing the functions specified in one flow or multiple flows and / or blocks.

[0108] ​These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0109] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0110] Those of ordinary skill in the art can understand that all or part of the steps of the above-mentioned facts and methods can be completed by programs instructing relevant hardware, and the programs involved or the programs mentioned can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: at this time, the corresponding method steps are derived, and the storage medium can be ROM / RAM, a magnetic disc, an optical disc, etc.

[0111] The above detailed description of the specific embodiments of the present application further illustrates the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for multi-antenna direction finding of a radiation source using the nonlinear processing capability of a neural network, characterized in that, The method comprises the following specific steps: training a preset neural network model through the constructed training sample until the neural network model reaches a preset training end condition; acquiring an unknown radiation source signal through a multi-channel acquisition mode, and extracting a multi-dimensional feature vector from the unknown radiation source signal; inputting the feature vector into the neural network model reaching the training end condition for processing to obtain a radiation source azimuth of the unknown radiation source signal.

2. The method of claim 1, wherein the neural network is configured to perform nonlinear processing of the received signals. The training sample is obtained through the following method: constructing an actual direction finding system comprising at least two antenna elements; placing one or more calibration radiation sources with known azimuths at measured azimuths; emitting signals at each azimuth angle at preset intervals through the calibration radiation sources within a 0°-360° azimuth range; acquiring the signals emitted at each azimuth angle through the actual direction finding system in a multi-channel acquisition mode, and calculating the amplitudes and phases of the signals emitted at each azimuth angle; for the signals emitted at each azimuth angle, synthesizing a multi-dimensional feature vector through the corresponding amplitudes and phases; constructing the training sample through the feature vectors of the signals at each azimuth angle and the corresponding azimuth angles.

3. The method of claim 2, wherein the neural network is configured to perform nonlinear processing of the received signals. The actual direction finding system at least comprises an antenna cover, antenna elements, microwave receiving channels and a signal processing unit in a connection relationship.

4. The method of claim 2, wherein the neural network is trained by using a plurality of training data sets, each of which is obtained by using a different antenna array. The feature vector is obtained through the following method: for the signals acquired through each channel, performing digital down-conversion to baseband to obtain complex signals, and calculating the corresponding amplitudes and phases through the complex signals; for each azimuth angle, constructing a multi-dimensional feature vector through the amplitudes and phases of each channel.

5. The method of claim 4, wherein the neural network is trained by using a plurality of training data sets, each of which is obtained by using a different antenna array. The multi-dimensional feature vector is specifically: F = [A1, A2,…, An, Δφ2, Δφ3, …,Δφn]; wherein, F is the feature vector, An is the amplitude or amplitude difference of the signal acquired by the nth channel, and Δφn is the relative phase of the signals acquired by each channel and the signal acquired by the first channel, wherein: Δφ2 = φ2 - φ1, Δφ3 = φ3 -φ1, Δφn = φn - φ1.

6. The method of claim 4, wherein the neural network is trained by using a plurality of training data sets, each of which is obtained by using a different antenna array. The amplitudes and phases are calculated through the following method: ； ； wherein denotes the amplitude of the signal acquired by the n-th channel, denotes the phase of the signal acquired by the n-th channel, denotes the quadrature component in the complex signal of the n-th channel, is the in-phase component in the complex signal of the n-th channel, is the square root function, is the four-quadrant arctangent function.

7. The method of claim 1, wherein the neural network is trained using a plurality of training data sets, each of the training data sets including a plurality of input data and a corresponding output data. The training end condition is that the number of iterations or the loss function does not exceed a threshold value, and the loss function is: ； wherein is the true value of the azimuth of the radiation source, denotes the predicted value of the azimuth of the radiation source.

8. A multi-antenna radiation source direction finding system that utilizes the nonlinear processing capability of a neural network, characterized by, It comprises: a model training module configured to train a preset neural network model through a constructed training sample until the neural network model reaches a preset training end condition; a feature vector construction module configured to acquire an unknown radiation source signal through a multi-channel acquisition mode, and extract a multi-dimensional feature vector from the unknown radiation source signal; an azimuth angle determination module configured to input the feature vector into the neural network model reaching the training end condition for processing to obtain a radiation source azimuth of the unknown radiation source signal.

9. An electronic device, comprising: It comprises: at least one processor, at least one memory and a data bus; wherein the processor and the memory complete mutual communication through the data bus; The memory stores program instructions executable by the processor, and the processor invokes the program instructions to execute the method in any one of claims 1-7.

10. A non-transitory computer-readable storage medium, comprising: The non-transitory computer-readable storage medium stores computer instructions that cause a computer to perform the method of any one of claims 1-7.

Images