Direction of arrival estimation method and device, medium and product

Abstract

The invention relates to the technical field of direction-of-arrival estimation, and discloses a direction-of-arrival estimation method and device, a medium and a product, which are applied to an analog beam forming system, and are used for carrying out grid-by-grid beam scanning in a preset multi-angle grid to obtain a beam power characteristic pattern of a target signal in the preset multi-angle grid; wherein the preset multi-angle grid comprises the arrival direction of the target signal; inputting the beam power characteristic pattern into a direction prediction model, and performing direction of arrival estimation on the target signal to obtain the direction of arrival of the target signal; wherein the direction prediction model is obtained by training according to a training power feature map of the training signal in a preset multi-angle grid. The method has the beneficial effects that after the target signal is received, the beam power characteristic pattern in the preset multi-angle grid is obtained through beam scanning, and the arrival direction of the target signal is predicted through the direction prediction model, so that high-precision direction-of-arrival estimation by using the analog beam forming system is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of direction of arrival estimation, and in particular to a direction of arrival estimation method, device, medium and product. BACKGROUND

[0002] The direction of arrival (DOA) is a core parameter of spatial signal positioning. The estimation technology for the direction of arrival mainly determines the azimuth and elevation of a spatial signal through an array antenna, and has important applications in the fields of radar, wireless communication, positioning and navigation, and intelligent sensing. Analog beamforming is a technology for changing the direction of an electromagnetic beam by controlling the signal phase of each antenna element in an antenna array. In a signal receiving system applying the analog beamforming technology, the received signals of all antenna elements are combined into a single output signal for subsequent processing to simplify the system architecture. However, when the direction of arrival is estimated through the above signal receiving system, the original signal of each antenna element cannot be obtained by the signal receiving system, which results in that the traditional algorithm relying on multi-channel signal processing cannot be used.

[0003] Therefore, it is urgent to propose a direction of arrival estimation method that can be applied to an analog beamforming system. SUMMARY

[0004] The present application provides a direction of arrival estimation method, device, medium and product. The beam power feature map in a preset multi-angle grid is obtained through beam scanning, and the direction of arrival of a target signal is predicted through a direction prediction model, so that high-precision direction of arrival estimation using an analog beamforming system is realized.

[0005] To achieve the above purpose, the main technical solution adopted by the present application includes: In a first aspect, the present application provides a direction of arrival estimation method applied to an analog beamforming system, which includes: performing grid-by-grid beam scanning in a preset multi-angle grid to obtain a beam power feature map of a target signal in the preset multi-angle grid; wherein the preset multi-angle grid contains the direction of arrival of the target signal; inputting the beam power feature map into a direction prediction model to estimate the direction of arrival of the target signal and obtain the direction of arrival of the target signal; wherein the direction prediction model is obtained by training a training power feature map of a training signal in the preset multi-angle grid.

[0006] The method for estimating the direction of arrival provided in the embodiments of the present application receives a target signal through a simulated beamforming system, performs beam scanning on a preset multi-angle grid based on the received signal, and obtains a beam power feature map containing the spatial characteristics of the target signal; a direction prediction model is trained in advance, and the direction of arrival of the target signal is estimated based on the beam power feature map, so as to determine the direction of arrival of the target signal. Compared with the related art, the beam power feature map in the preset multi-angle grid is obtained through beam scanning for the estimation of the direction of arrival of the target signal, thereby solving the problem that the traditional algorithm cannot be used due to the fact that the simulated beamforming system cannot obtain the original signal of each antenna element, and realizing the estimation of the direction of arrival using the simulated beamforming system. In addition, the direction prediction model is trained in advance by training the power feature map, so that the direction prediction model learns the mapping relationship between the power distribution in the preset multi-angle grid and the direction of arrival, thereby effectively utilizing the fine features in the beam power feature map and significantly improving the prediction accuracy and robustness of the direction of arrival.

[0007] Optionally, the performing beam scanning on a preset multi-angle grid based on the received signal to obtain a beam power feature map of the target signal in the preset multi-angle grid comprises: for any grid in the preset multi-angle grid, receiving the target signal at a receiving angle corresponding to the any grid to obtain a received signal of the target signal corresponding to the any grid; performing power calculation based on the received signal to obtain signal power of the target signal on the any grid; iterating through all grids in the preset multi-angle grid to obtain the beam power feature map based on the signal power of all grids.

[0008] Optionally, the received signal comprises signals received by the any grid in a plurality of sampling frames; and the performing power calculation based on the received signal to obtain the signal power of the target signal on the any grid comprises: performing power estimation on the received signal of the any grid in any sampling frame to obtain single-frame power of the any grid in the any sampling frame; performing time-domain averaging on the single-frame power of the any grid in the plurality of sampling frames to obtain the signal power of the target signal on the any grid.

[0009] Optionally, the inputting the beam power feature map into a direction prediction model to estimate the direction of arrival of the target signal comprises: performing forward propagation calculation based on the beam power feature map through the direction prediction model to output the direction of arrival of the target signal.

[0010] Optionally, the direction prediction model is trained by the following manner: obtaining the training power feature map, wherein the preset multi-angle grid contains the direction of arrival of the training signal, and the training power feature map corresponds to a training label representing the direction of arrival of the training signal; inputting the training power feature map into an initial prediction model, performing direction of arrival prediction on the training power feature map by the initial prediction model to obtain an initial prediction result, calculating a loss function value between the initial prediction result and the training label, adjusting parameters of the initial prediction model to obtain a parameter adjustment model, and repeating the process of direction of arrival prediction and parameter adjustment until a training stop condition is met, and obtaining the direction prediction model.

[0011] Optionally, the obtaining the training power feature map comprises: for any grid in the preset multi-angle grid, receiving the training signal under any channel condition in the preset channel conditions to obtain a received signal corresponding to the any grid under the training signal in the any channel condition; calculating the signal power of the training signal on the any grid according to the received signal, and obtaining a condition power feature map under the any channel condition according to the signal powers of all grids in the preset multi-angle grid; obtaining the training power feature map according to the condition power feature map under each channel condition in the preset channel conditions.

[0012] Optionally, before the training power feature map is input into the initial prediction model, the method further comprises: performing scale transformation on the training power feature map to obtain a gain simulation image; performing rotation transformation on the training power feature map to obtain an offset simulation image; obtaining a sample-enhanced training power feature map according to the gain simulation image and the offset simulation image, so as to be input into the initial prediction model for training.

[0013] In a second aspect, an embodiment of the present application provides a computer device, comprising a memory and a processor, which are communicatively connected with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method in any of the above embodiments.

[0014] In a third aspect, an embodiment of the present application provides a computer readable storage medium, having stored thereon computer instructions, the computer instructions being used to cause a computer to execute the method of any of the above embodiments.

[0015] In a fourth aspect, an embodiment of the present application provides a computer program product, comprising computer instructions, the computer instructions being used to cause a computer to execute the method of any of the above embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the specific embodiments of the present application or the prior art, the drawings needed in the description of the specific embodiments or the prior art will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0017] Figure 1 The schematic diagram of the architecture of the analog beamforming system; Figure 2 The step diagram of the direction of arrival estimation method provided by the embodiment of the present application; Figure 3 The step diagram of the beam scanning in the embodiment of the present application; Figure 4 The step diagram of the power calculation in the embodiment of the present application; Figure 5 The structure diagram of the direction prediction model in the embodiment of the present application; Figure 6 The step diagram of the training of the direction prediction model in the embodiment of the present application; Figure 7 The step diagram of the acquisition of the training power feature map in the embodiment of the present application; Figure 8 The beam power feature map in the embodiment of the present application; Figure 9 The step diagram of the data enhancement of the training power feature map in the embodiment of the present application; Figure 10 The stage flow chart of the direction of arrival estimation method provided by the embodiment of the present application; Figure 11 The prediction result comparison chart of different direction prediction models in the embodiment of the present application; Figure 12 The module chart of the direction of arrival estimation device provided by the embodiment of the present application; Figure 13 The structure diagram of a computer device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0019] Referring to Figure 1 As shown in the figure, the signal receiving system applying the analog beamforming technology includes an antenna array composed of multiple antenna units, each antenna unit is connected with an independent phase shifter, the phase of the signal received by the corresponding antenna unit is controlled through phase adjustment by the phase shifter. The output end of each phase shifter is connected with a combiner, the combiner collects all the phase-modulated signals into a single output signal and inputs the radio frequency chain for subsequent signal processing. The above signal receiving system has the advantages of simple structure, low cost and low power consumption, but when the direction of arrival is estimated by the above signal receiving system, the original signal of each antenna unit cannot be obtained by the signal receiving system, which leads to the fact that the traditional algorithm relying on multi-channel signal processing cannot be used.

[0020] Based on the above problem, the present application provides a direction of arrival estimation method, device, medium and product, applied to an analog beamforming system, performing grid-by-grid beam scanning in a preset multi-angle grid to obtain a beam power feature map of a target signal in the preset multi-angle grid; wherein the preset multi-angle grid contains the direction of arrival of the target signal; inputting the beam power feature map into a direction prediction model to estimate the direction of arrival of the target signal, and obtaining the direction of arrival of the target signal; wherein the direction prediction model is obtained by training according to a training power feature map of a training signal in the preset multi-angle grid.

[0021] The direction of arrival estimation method provided by the present application receives a target signal through an analog beamforming system, performs grid-by-grid beam scanning in a preset multi-angle grid according to the received signal to obtain a beam power feature map containing the spatial characteristics of the target signal; and estimates the direction of arrival of the target signal according to the beam power feature map by using a pre-trained direction prediction model, so as to determine the direction of arrival of the target signal.

[0022] Compared with the related art, the present application obtains the beam power feature map in the preset multi-angle grid by beam scanning for the direction of arrival estimation of the target signal, solves the problem that the traditional algorithm cannot be used due to the fact that the analog beamforming system cannot obtain the original signal of each antenna unit, and realizes the direction of arrival estimation by using the analog beamforming system.

[0023] In addition, the direction prediction model is pre-trained by training the power feature map, so that the direction prediction model learns the mapping relationship between the power distribution in the preset multi-angle grid and the direction of arrival, thereby effectively utilizing the fine features in the beam power feature map, and significantly improving the prediction accuracy and robustness of the direction of arrival.

[0024] According to an embodiment of the present application, a direction of arrival estimation method is provided. It should be noted that the steps shown in the flowchart can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0025] Referring to Figure 2 In this embodiment, a direction of arrival estimation method is provided, which is applied to an analog beamforming system. The method comprises: S100. Perform grid-by-grid beam scanning in a preset multi-angle grid to obtain a beam power feature map of the target signal in the preset multi-angle grid; wherein the preset multi-angle grid contains the direction of arrival of the target signal.

[0026] S200. Input the beam power feature map into a direction prediction model to estimate the direction of arrival of the target signal, and obtain the direction of arrival of the target signal; wherein the direction prediction model is obtained by training a training power feature map of a training signal in the preset multi-angle grid.

[0027] The preset multi-angle grid can be a two-dimensional grid obtained by grid division of an angle space relative to an antenna array. The dimensions of the preset multi-angle grid can correspond to the azimuth angle and the elevation angle, respectively, and each grid corresponds to a pair of spatial angles. It can be understood that the direction of arrival of the target signal is included in the preset multi-angle grid, i.e. the azimuth angle corresponding to the direction of arrival is within the azimuth angle range of the preset multi-angle grid, and the elevation angle corresponding to the direction of arrival is within the elevation angle range of the preset multi-angle grid. For example, for an angle space covering an azimuth angle range of [-60°, 60°] and an elevation angle range of [-60°, 60°], the angle space is divided into 50x50 grids, obtaining 2500 grids, and the angle resolution of each grid is about 2.4°.

[0028] The direction prediction model can be a prediction model using a deep learning method, including a deep neural network and machine learning, etc. Before performing direction of arrival estimation on the target signal, the direction prediction model is trained in advance according to a training power feature map of the training signal in a preset multi-angle grid, to learn the mapping relationship between the power distribution and the direction of arrival in the preset multi-angle grid. The training signal can be a real signal, and an actual training power feature map is generated as training data by performing signal reception and beam scanning on the analog beamforming system. The training signal can also be a simulation signal, and a training power feature map is generated as training data based on simulation results. In some embodiments, the training power feature map can also be obtained based on real signals and simulation signals respectively as training data, to improve the breadth of the training data and enhance the training effect.

[0029] Specifically, the phase shifters in the analog beamforming system are controlled to adjust the signal reception phase of the antenna elements in the analog beamforming system, thereby changing the reception angle of the analog beamforming system, receiving the target signal in different signal reception directions, and obtaining the received signals corresponding to different signal reception directions. It can be understood that the target signal arrives at the antenna elements at different times, and there is a phase difference between the received signals of each antenna element. The phase difference between different received signals represents the spatial phase information of the target signal. Exemplarily, the antenna elements can be arranged in the form of a uniform planar array (UPA), and the size of the array is 8x8, and the distance between each antenna element is half a wavelength.

[0030] Further, each signal reception direction corresponds to a grid position in the preset multi-angle grid, and after obtaining the received signal of each grid position, the signal power in each grid position is calculated by a beam scanning algorithm, and then a beam power feature map of the target signal in the preset multi-angle grid is obtained, representing the signal power received by the analog beamforming system in different signal reception directions. It can be understood that the size of the beam power feature map is the same as that of the preset multi-angle grid, and the power data in the beam power feature map can be normalized data.

[0031] It should be noted that the analog beamforming system has a directional gain characteristic. When the signal receiving direction of the analog beamforming system is aligned with the direction of arrival of the target signal, the maximum main lobe gain can be obtained, the signal receiving capability of the analog beamforming system is improved, and the corresponding grid in the preset multi-angle grid shows a power peak value in the beam power characteristic map. Similarly, when the signal receiving direction of the analog beamforming system deviates from the direction of arrival of the target signal, the main lobe gain decreases significantly as the direction deviates, resulting in a decrease in the signal power of the corresponding grid in the preset multi-angle grid, which in turn affects the signal-to-noise ratio of the corresponding grid, and the corresponding grid cannot form an effective peak value in the beam power characteristic map. It can be understood that the power distribution in the beam power characteristic map represents the matching relationship between the signal receiving direction of the analog beamforming system and the direction of arrival of the target signal. According to the power distribution, the direction of arrival of the target signal can be estimated to determine the direction of arrival of the target signal.

[0032] Further, the beam power characteristic map is input into a direction prediction model, the direction of arrival of the target signal is estimated according to the beam power characteristic map, and the power distribution in the beam power characteristic map is mapped to the direction of arrival of the target signal by the direction prediction model. It should be noted that the direction prediction model can be obtained by training a training power characteristic map of a training signal in a preset multi-angle grid. In the case of a deep neural network as the direction prediction model, the network weights and network biases of the deep neural network are iteratively adjusted based on the training power characteristic map. After training, the obtained network weights and network biases are saved to obtain the direction prediction model.

[0033] The direction of arrival estimation method provided in this embodiment receives a target signal through an analog beamforming system, performs grid-by-grid beam scanning on the target signal in a preset multi-angle grid, and obtains a beam power characteristic map containing the spatial characteristics of the target signal. A pre-trained direction prediction model is used to estimate the direction of arrival of the target signal according to the beam power characteristic map, thereby determining the direction of arrival of the target signal.

[0034] Compared with related technologies, the present application obtains a beam power characteristic map in a preset multi-angle grid through beam scanning for direction of arrival estimation of a target signal, solves the problem that traditional algorithms cannot be used due to the inability of an analog beamforming system to obtain the original signals of each antenna element, and realizes direction of arrival estimation using an analog beamforming system.

[0035] In addition, the present application also pre-trains a direction prediction model through a training power characteristic map, so that the direction prediction model learns the mapping relationship between the power distribution in the preset multi-angle grid and the direction of arrival, thereby effectively utilizing the fine features in the beam power characteristic map and significantly improving the prediction accuracy and robustness of the direction of arrival.

[0036] ReferenceFigure 3 As shown, as an embodiment of the present application, a grid-by-grid beam scanning is performed in a preset multi-angle grid to obtain a beam power feature map of the target signal in the preset multi-angle grid, including: S110. For any grid in the preset multi-angle grid, receiving the target signal at a receiving angle corresponding to the any grid to obtain a receiving signal of the target signal corresponding to the any grid.

[0037] S120. Performing power calculation according to the receiving signal to obtain a signal power of the target signal on the any grid.

[0038] S130. Iterating through all grids in the preset multi-angle grid, obtaining a beam power feature map according to the signal powers of all grids.

[0039] Specifically, the dimensions of the preset multi-angle grid can correspond to the azimuth angle and the elevation angle respectively, and each grid corresponds to a spatial angle pair. For any grid in the preset multi-angle grid, the phase shifter in the analog beamforming system is controlled to adjust the receiving angle of the antenna unit to align with the spatial angle of the any grid to receive the target signal, and the receiving signal of each antenna unit is obtained. The receiving signal is phase-adjusted by the phase shifter connected to the antenna unit, and the receiving signals of all antenna units are added after phase adjustment to obtain the receiving signal of the target signal corresponding to the any grid.

[0040] Further, in the any grid, the receiving signal of the target signal is preprocessed, and the preprocessed receiving signal is analog-to-digital converted to obtain a digital signal sequence of the any grid, which includes an in-phase signal component and a quadrature signal component. The power calculation is performed according to the digital signal sequence to obtain the signal power of the target signal on the any grid. It can be understood that the digital signal sequence can be signal data corresponding to any sampling frame, and the instantaneous power calculation is performed on any sampling point in the any sampling frame to obtain the instantaneous power of the any sampling point. The instantaneous powers of all sampling points in the any sampling frame are averaged to obtain the power data corresponding to the any sampling frame as the signal power of the target signal on the any grid. In other embodiments, power calculation can also be performed in other ways, which are not limited.

[0041] Further, all grids in the preset multi-angle grid are iterated to perform power calculation on each grid to obtain the signal power of the target signal on each grid. According to the position of each grid in the preset multi-angle grid, the signal power of the target signal on each grid is mapped to the corresponding position in the preset multi-angle grid, and the signal power of each grid is normalized to generate a beam power feature map.

[0042] Reference Figure 4As shown, as an embodiment of the present application, the received signal includes any grid received signal in a plurality of sampling frames; power calculation is performed according to the received signal to obtain the signal power of the target signal on any grid, including: S122. Power estimation is performed on the received signal of any grid in any sampling frame to obtain the single-frame power of any grid in any sampling frame.

[0043] S124. Time domain averaging is performed on the single-frame power of any grid in a plurality of sampling frames to obtain the signal power of the target signal on any grid.

[0044] Specifically, for any grid in a preset multi-angle grid, power estimation is performed according to the received signal of the grid in any sampling frame to obtain the single-frame power of the grid in the sampling frame. It should be noted that the power data obtained from the single-frame received signal has reliability problems. On the one hand, the received signal in any sampling frame usually has a large variance, and the signal data fluctuates greatly around the true data of the target signal; on the other hand, in the case of low signal-to-noise ratio, the signal data is easily buried by noise data, resulting in that the signal peak value cannot be effectively detected.

[0045] Considering the above reasons, the embodiment respectively obtains the single-frame power of the grid in a plurality of sampling frames, and performs time domain averaging on the single-frame power in the plurality of sampling frames to obtain the signal power of the target signal on any grid. It can be understood that the plurality of sampling frames used for time domain averaging can be consecutive adjacent sampling frames, can be a plurality of sampling frames arranged based on the same period interval, or can be a plurality of sampling frames arranged at irregular intervals. By performing time domain averaging in a plurality of sampling frames, the variance of the signal power is reduced, the power data accuracy of the grid is improved, and the reliability of the beam power feature map is improved, thereby providing a more stable data basis for the direction of arrival estimation. Exemplarily, the number of sampling frames used for time domain averaging can be 10 frames, each frame contains 1000 shots, and the time length of each shot is related to the hardware sampling rate.

[0046] As an embodiment of the present application, the beam power feature map is input into a direction prediction model to estimate the direction of arrival of the target signal, including: S210. According to the beam power feature map, forward propagation calculation is performed by the direction prediction model to output the direction of arrival of the target signal.

[0047] Reference Figure 5As shown, in the present embodiment, the direction prediction model can be a deep neural network of a regression model, including three convolution layers, a flatten layer and two fully connected layers connected in sequence, wherein the convolution layers can be a Convolutional Neural Network (CNN), the first convolution layer can use 32 convolution kernels with a size of 3x3 to perform convolution operation with a step of 1, and perform nonlinear change operation through a ReLU activation function, to preliminarily extract spatial features of the input beam power feature map, to obtain a first feature map. Similarly, the second convolution layer can use 64 convolution kernels with a size of 3x3 to perform convolution operation with a step of 1, and perform nonlinear change operation through a ReLU activation function, to extract spatial features of the first feature map output by the first convolution layer, to obtain a second feature map. The third convolution layer can use 128 convolution kernels with a size of 3x3 to perform convolution operation with a step of 1, and perform nonlinear change operation through a ReLU activation function, to extract spatial features of the second feature map output by the second convolution layer, to obtain a third feature map as a spatial feature map of the beam power feature map. It can be understood that the second and third convolution layers can further extract spatial features of the beam power feature map on the basis of the output of the first convolution layer, to enhance the extracted spatial features.

[0048] The flatten layer can be used to flatten the spatial feature map output by the convolution layer, to obtain a one-dimensional feature vector and input the fully connected layer. The first fully connected layer can include 512 neurons, to reduce and fuse the feature vector, to obtain a low-dimensional feature representation of the beam power feature map. The second fully connected layer can include 2 neurons, to estimate the direction of arrival of the grid according to the low-dimensional feature representation of the beam power feature map, to output the direction of arrival of the target signal.

[0049] Specifically, the beam power feature map is input into the direction prediction model, wherein the first convolutional layer uses 32 convolutional kernels with a size of 3x3 to perform convolution operation on the beam power feature map with a step of 1, and the spatial size of the output result is kept unchanged by setting padding. After obtaining the convolution result, batch normalization processing is performed on the convolution result, and nonlinearity is introduced through the ReLU activation function to obtain the first feature map, which has the same spatial size as the beam power feature map and a channel number of 32, containing the preliminary spatial features of the beam power feature map. Similarly, the first feature map is input into the second convolutional layer, which uses 64 convolutional kernels with a size of 3x3 to perform convolution operation on the first feature map with a step of 1, and the spatial size of the output result is kept unchanged by setting padding. After obtaining the convolution result, batch normalization processing is performed on the convolution result, and nonlinearity is introduced through the ReLU activation function to obtain the second feature map, which has the same spatial size as the first feature map and a channel number of 64, containing more complex intermediate spatial features. The second feature map is input into the third convolutional layer, which uses 128 convolutional kernels with a size of 3x3 to perform convolution operation on the second feature map with a step of 1, and the spatial size of the output result is kept unchanged by setting padding. After obtaining the convolution result, batch normalization processing is performed on the convolution result, and nonlinearity is introduced through the ReLU activation function to obtain the third feature map, which has the same spatial size as the second feature map and a channel number of 128, containing high-level semantic features related to the direction of arrival of the target signal.

[0050] Further, the third feature map is input into the flattening layer, which expands and connects all the channel feature maps in the third feature map in the spatial dimension to obtain a one-dimensional feature vector with a length equal to the product of the spatial size and the channel number of the third feature map.

[0051] Further, the feature vector is input into the first fully connected layer, in which each neuron performs weighted summation on the feature vector and adds a bias, and then performs nonlinear transformation on the processing result through the ReLU activation function to obtain the feature output of each neuron. According to the feature output of all neurons, a high-dimensional feature vector is obtained, which has a dimension of 512 and represents the low-dimensional feature representation of the beam power feature map. The high-dimensional feature vector is input into the second fully connected layer, in which 2 neurons are used to perform linear transformation on the high-dimensional feature vector respectively to obtain the feature data corresponding to the neurons, which represent the azimuth angle and the elevation angle of the target signal respectively, as the output of the direction of arrival of the target signal.

[0052] Referring to FIG. 1, Figure 6 As shown in FIG. 1, as an embodiment of the present application, the direction prediction model is trained in the following manner: S310. Obtain a training power feature map; wherein the preset multi-angle grid contains a direction of arrival of a training signal, and the training power feature map corresponds to a training label representing the direction of arrival of the training signal.

[0053] S320. Input the training power feature map into an initial prediction model, perform direction of arrival prediction on the training power feature map by the initial prediction model to obtain an initial prediction result, and calculate a loss function value between the initial prediction result and the training label to adjust parameters of the initial prediction model and obtain a parameter adjustment model.

[0054] S330. Take the parameter adjustment model as the initial prediction model, repeat the above direction of arrival prediction and parameter adjustment process until a training stop condition is met, and obtain a direction prediction model.

[0055] Specifically, the training signal can be a real signal or a simulation signal with a preset direction of arrival, and the direction of arrival of the training signal is in an angle space corresponding to the preset multi-angle grid. The training power feature map is generated according to the training signal as training data, and each training power feature map corresponds to a training label to represent the direction of arrival of the training signal.

[0056] Further, an initial prediction model is constructed, parameters of the initial prediction model are initialized to obtain an initial prediction model with initial parameters, and the initial prediction model is trained by using the training power feature map. In some embodiments, the training process of the initial prediction model can use a random sample shuffling and batch training strategy. First, all the training power feature maps are randomly shuffled to obtain a shuffled training data set, and the training data set is divided into batches to obtain multiple training batches. For any training batch, the initial prediction model is input into the initial prediction model, the direction of arrival of the training signal is predicted according to the training power feature map, and the initial prediction result of the training signal is obtained. Based on a preset loss function, a loss function value between the initial prediction result and the training label corresponding to the training power feature map in the any training batch is calculated, and the initial prediction model is adjusted according to the loss function value to obtain a parameter adjustment model. The parameter adjustment model is taken as the initial prediction model, the next training batch is input into the initial prediction model, and the above direction of arrival prediction and parameter adjustment process is repeated until a training stop condition is met, the training is stopped, and a direction prediction model is obtained.

[0057] Exemplarily, the number of training power feature maps can be 10,000 groups, and the size of each training batch can be 32 groups. The loss function used for training the initial prediction model can be Mean-Square Error (MSE), the optimizer used for parameter adjustment can be Adam optimizer, and the learning rate can be 0.001. The training stopping condition can be that the loss function value no longer decreases in a preset number of training rounds, or the number of training rounds reaches a preset threshold, where the preset threshold of the number of training rounds can be 100.

[0058] Referring to Figure 7 As shown, as an embodiment of the present application, the training power feature map is obtained, including: S312. For any grid in the preset multi-angle grid, receive the training signal under any channel condition in the preset channel condition to obtain the received signal of the training signal corresponding to any grid under any channel condition.

[0059] S314. Calculate the signal power of the training signal on any grid according to the received signal, and obtain the conditional power feature map under any channel condition according to the signal power of all grids in the preset multi-angle grid.

[0060] S316. Obtain the training power feature map according to the conditional power feature map under each channel condition in the preset channel condition.

[0061] Specifically, in actual application scenarios, when receiving the target signal at the receiving angle corresponding to any grid, due to the directional gain characteristics of the analog beamforming system, when there is a deviation between the receiving angle and the incident angle of the target signal, the target signal will appear gain drop or waveform distortion. In addition, the channel condition of the analog beamforming system when receiving the target signal can be dynamically changing, and is not a single fixed value, such as Signal-to-Noise Ratio (SNR) or the number of signal sources, etc. When the channel condition changes, the received signal obtained by the analog beamforming system will be affected accordingly. Under the superposition of various error factors, the fidelity of the beam power feature map is deteriorated, which cannot effectively extract the power peak, and further affects the accuracy of the direction of arrival estimation. Therefore, if the training power feature map is obtained only under a single channel condition for model training, the obtained direction prediction model may have insufficient generalization ability and be easily affected by environmental factors.

[0062] In view of the above reasons, the embodiment pre-selects multiple preset channel conditions, receives the training signal under any channel condition of the multiple preset channel conditions, and obtains the received signal corresponding to any grid of the training signal under the any channel condition. Under the any channel condition, power calculation is performed according to the received signal to obtain a conditional power feature map. Similarly, the conditional power feature map is obtained under each preset channel condition, and the training power feature map is obtained according to the conditional power feature maps under all preset channel conditions, and is used for training the direction prediction model. Exemplarily, when the channel condition is signal-to-noise ratio, the preset channel conditions can include cases where the signal-to-noise ratio is -10 dB, 0 dB, 10 dB, and 20 dB, etc. In the case of a real signal as the training signal, the signal-to-noise ratio can be changed by adjusting the environmental noise or the transmission power of the training signal; in the case of a simulation signal as the training signal, the signal-to-noise ratio can be changed by changing the variance of the simulation signal.

[0063] The direction of arrival of the training signal can be a direction angle of 10° and a pitch angle of 30°. The training power feature map obtained based on the above signal-to-noise ratio can refer to FIG. 6, which shows that in the case of a low signal-to-noise ratio, the power distribution in the training power feature map is affected by the background noise, and the correct peak position cannot be detected. With the improvement of the signal-to-noise ratio, the power peak in the training power feature map can be focused on the actual direction of arrival of the training signal. Figure 8

[0064] It should be noted that by pre-selecting multiple preset channel conditions and generating training power feature maps under different preset channel conditions, the training data used to train the direction prediction model is enhanced, so that the direction prediction model can learn the mapping relationship between the power distribution and the direction of arrival under different conditions, thereby being able to accurately predict within a certain range of channel conditions, and improving the reliability and robustness of the direction prediction model.

[0065] Referring to FIG. 6, Figure 9 As an embodiment of the present application, before the training power feature map is input into the initial prediction model, it further includes: S342. Scale transformation is performed on the training power feature map to obtain a gain simulation image.

[0066] S344. Rotation transformation is performed on the training power feature map to obtain an offset simulation image.

[0067] S346. The training power feature map after sample enhancement is obtained according to the gain simulation image and the offset simulation image, and is used for inputting the initial prediction model for training.

[0068] ​Specifically, to further improve the robustness of the orientation prediction model, an optimization strategy is employed to perform sample augmentation on the training power feature map, thereby expanding the coverage of the training data and improving the model training effect. Sample augmentation can include scaling and rotation transformations. Scaling transformation involves adjusting the image scale of the training power feature map to change its image size, simulating gain changes between antenna elements in real-world application scenarios, resulting in a gain simulation image. Rotation transformation involves rotating the training power feature map to change its image orientation, simulating angular offsets of antenna elements in real-world application scenarios, resulting in an offset simulation image. Sample augmentation is performed on the training power feature map based on the gain simulation image and the offset simulation image, resulting in an augmented training power feature map, which is then used as input to the initial prediction model for training. For example, sample augmentation can be performed on 50% of the data in all training power feature maps, the scaling factor can be between 0.9 and 1.1, and the rotation angle range can be within ±5°, resulting in a total of 15,000 training power feature maps.

[0069] Reference Figure 10 As shown, the method provided in this application can be divided into an offline stage and an online stage. The offline stage can include acquiring a training power feature map and training a direction prediction model. The process of acquiring the training power feature map can include signal reception and beam scanning to obtain the training power feature map of the training signal in a preset multi-angle grid. The process of training the direction prediction model can include neural network training and optimization to obtain the direction prediction model. The online stage can include acquiring a beam power feature map and estimating the direction of arrival. The process of acquiring the beam power feature map can include real-time signal acquisition and feature map generation. By receiving the target signal and using a beam scanning algorithm, the beam power feature map of the target signal in a preset multi-angle grid is obtained. The process of estimating the direction of arrival can include neural network prediction and outputting angle results. The direction prediction model trained in the offline stage estimates the direction of arrival based on the beam power feature map and outputs the direction of arrival of the target signal.

[0070] Based on the method provided in this application, the performance of different types of direction prediction models is validated in preset multi-angle grids of various grid sizes. The grid sizes used can include 31×61, 61×31, and 121×61. Direction prediction models are generated using classification and regression models, respectively. Direction of arrival estimation is performed on the same test dataset, and the results can be referenced. Figure 11 As shown. In this embodiment, the classification model can be any common model structure, and no specific limitation is made here; the model structure of the regression model can be referred to Figure 5 As shown, other model structures are also possible.

[0071] according to Figure 11 It can be seen that the estimation errors of each model gradually decrease with the increase of grid size when estimating azimuth and elevation angles respectively. With a grid size of 31×61, the estimation error of the regression model is within 1.5°, while the estimation error of the classification model is around 2°. With a grid size of 61×31, the estimation errors of both the regression and classification models are within 1.5°, and the estimation errors of the classification model in azimuth estimation and the regression model in elevation angle estimation are within 1°. With a grid size of 121×61, the estimation error of the regression model is around 0.5°, the estimation error of the classification model in azimuth estimation is 0.72°, but its estimation error in elevation angle estimation is around 1.5°. Comparing the estimation errors of the classification and regression models in different estimation tasks shows that the estimation error of the regression model is smaller than that of the classification model in the same estimation task; therefore, the direction prediction model generated using the regression model has higher accuracy.

[0072] Accordingly, please refer to Figure 12 This application provides a direction-of-arrival estimation device for use in an analog beamforming system. The device includes: The beam scanning calculation module 1210 is used to perform grid-by-grid beam scanning in a preset multi-angle grid to obtain the beam power characteristic map of the target signal in the preset multi-angle grid; wherein the preset multi-angle grid contains the arrival direction of the target signal.

[0073] The direction-of-arrival estimation module 1220 is used to input the beam power feature map into the direction prediction model to estimate the direction of arrival of the target signal and obtain the direction of arrival of the target signal; wherein, the direction prediction model is trained based on the training power feature map of the training signal in a preset multi-angle grid.

[0074] In some alternative implementations, the beam scanning calculation module 1210 includes: The signal receiving unit is used to receive the target signal at the receiving angle corresponding to any grid in the preset multi-angle grid, and obtain the received signal of the target signal corresponding to any grid.

[0075] The power calculation unit is used to perform power calculations based on the received signal to obtain the signal power of the target signal on any grid.

[0076] The grid traversal unit is used to traverse all grids in the preset multi-angle grid and obtain the beam power characteristic map based on the signal power of all grids.

[0077] In some alternative implementations, the power calculation unit includes: The single-frame power estimation subunit is used to estimate the power of the received signal of any grid in any sampling frame, so as to obtain the single-frame power of any grid in any sampling frame.

[0078] The multi-frame time-domain averaging sub-unit is used to perform time-domain averaging of the single-frame power of any grid within multiple sampling frames, so as to obtain the signal power of the target signal on any grid.

[0079] In some alternative implementations, the direction-of-arrival estimation module 1220 includes: The forward propagation calculation unit is used to perform forward propagation calculations based on the beam power characteristic map using the direction prediction model, and outputs the arrival direction of the target signal.

[0080] In some alternative implementations, the apparatus further includes a prediction model training module, comprising: The training data acquisition unit is used to acquire the training power feature map; wherein, the preset multi-angle grid contains the arrival direction of the training signal, and the training power feature map has a corresponding training label representing the arrival direction of the training signal.

[0081] The model training and adjustment unit is used to input the training power feature map into the initial prediction model, and then use the initial prediction model to predict the direction of arrival based on the training power feature map to obtain the initial prediction result. The loss function value between the initial prediction result and the training label is calculated to adjust the parameters of the initial prediction model to obtain the parameter-adjusted model.

[0082] The model training iteration unit is used to take the parameter-adjusted model as the initial prediction model and repeat the above process of direction of arrival prediction and parameter adjustment until the training stopping condition is met, thus obtaining the direction prediction model.

[0083] In some optional implementations, the training data acquisition unit includes: The conditional signal receiving subunit is used to receive the training signal for any grid in the preset multi-angle grid under any channel condition in the preset channel conditions, so as to obtain the received signal of any grid corresponding to the training signal under any channel condition.

[0084] The conditional power calculation subunit is used to calculate the signal power of the training signal on any grid based on the received signal, and to obtain the conditional power feature map under any channel condition based on the signal power of all grids in the preset multi-angle grid.

[0085] The training data acquisition subunit is used to obtain the training power feature map based on the conditional power feature map of each channel condition in the preset channel conditions.

[0086] In some optional implementations, the prediction model training module further includes a training data augmentation unit, including... The scaling subunit is used to scale the training power feature map to obtain a gain simulation image.

[0087] The rotation transformation subunit is used to perform rotation transformation on the training power feature map to obtain an offset simulation image.

[0088] The data augmentation subunit is used to obtain the training power feature map of the sample augmentation based on the gain simulation image and the offset simulation image, which is then used as input to the initial prediction model for training.

[0089] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0090] In this embodiment, the direction-of-arrival estimation device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0091] Please see Figure 13 , Figure 13 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application, such as... Figure 13 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 13 Take a processor 10 as an example.

[0092] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0093] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0094] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0095] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0096] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0097] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0098] This application provides a computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method of any embodiment of this application.

[0099] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

[0100] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0101] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0102] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0103] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0104] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0105] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0106] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0107] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0108] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0109] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A direction-of-arrival estimation method, characterized in that, The method, applied to an analog beamforming system, includes: A grid-by-grid beam scan is performed in a preset multi-angle grid to obtain a beam power characteristic map of the target signal in the preset multi-angle grid; wherein, the preset multi-angle grid includes the arrival direction of the target signal; The beam power feature map is input into the direction prediction model to estimate the direction of arrival of the target signal, thereby obtaining the direction of arrival of the target signal; wherein, the direction prediction model is trained based on the training power feature map of the training signal in the preset multi-angle grid.

2. The method according to claim 1, characterized in that, The step of performing a grid-by-grid beam scan in a preset multi-angle grid to obtain a beam power characteristic map of the target signal in the preset multi-angle grid includes: For any grid in the preset multi-angle grid, the target signal is received at the receiving angle corresponding to any grid, and the received signal of the target signal corresponding to any grid is obtained; The power of the target signal on any grid is obtained by calculating the power based on the received signal. Traverse all grids in the preset multi-angle grid and obtain the beam power characteristic map based on the signal power of all grids.

3. The method according to claim 2, characterized in that, The received signal includes the signal received by any grid within a multi-sampling frame; the step of calculating the power of the target signal on any grid based on the received signal includes: Power estimation is performed on the received signal of any grid in any sampling frame to obtain the single-frame power of any grid in any sampling frame. The signal power of the target signal on any grid is obtained by performing a time-domain average of the single-frame power of any grid within the multi-sampling frames.

4. The method according to claim 1, characterized in that, The step of inputting the beam power feature map into the direction prediction model to estimate the direction of arrival of the target signal and obtain the direction of arrival of the target signal includes: The direction prediction model is used to perform forward propagation calculations based on the beam power characteristic map, and the arrival direction of the target signal is output.

5. The method according to claim 1, characterized in that, The direction prediction model is trained using the following method: Obtain the training power feature map; wherein, the preset multi-angle grid contains the arrival direction of the training signal, and the training power feature map corresponds to a training label representing the arrival direction of the training signal; The training power feature map is input into the initial prediction model, and the initial prediction model performs direction of arrival prediction based on the training power feature map to obtain the initial prediction result; the loss function value between the initial prediction result and the training label is calculated to adjust the parameters of the initial prediction model to obtain the parameter-adjusted model; Using the parameter adjustment model as the initial prediction model, the above process of direction of arrival prediction and parameter adjustment is repeated until the training stopping condition is met, thus obtaining the direction prediction model.

6. The method according to claim 5, characterized in that, The step of obtaining the training power feature map includes: For any grid in the preset multi-angle grid, the training signal is received under any channel condition in the preset channel conditions to obtain the received signal of the training signal corresponding to any grid under the any channel condition; The signal power of the training signal on any grid is calculated based on the received signal, and the conditional power feature map under any channel condition is obtained based on the signal power of all grids in the preset multi-angle grid. The training power feature map is obtained based on the conditional power feature map under each channel condition in the preset channel conditions.

7. The method according to claim 5, characterized in that, Before inputting the trained power feature map into the initial prediction model, the method further includes: The training power feature map is scaled to obtain a gain simulation image; The training power feature map is rotated to obtain an offset simulated image; The enhanced training power feature map is obtained from the gain simulation image and the offset simulation image, and is used as input to the initial prediction model for training.

8. A computer device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the method of any one of claims 1 to 7.

10. A computer program product, characterized in that, Includes computer instructions for causing a computer to perform the method of any one of claims 1 to 7.

Images