Staged beam forming method and system based on full-connection network

Abstract

The invention discloses a phased beam forming method and system based on a full-connection network, and mainly solves the problems of high calculation complexity, long calculation time delay, performance loss of heuristic beam forming and the like in the existing beam forming. According to the implementation scheme, the method comprises the steps that a full-connection network is learned in advance, and an amplitude weight predictor is formed; the amplitude weight predictor outputs an amplitude weight vector according to the angle difference between the interference direction and the beam pointing direction; calculating a corresponding phase weight vector according to the beam pointing angle; obtaining an array element complex weight vector by taking the Hadamard product of the amplitude weight vector and the phase weight vector; signals received or sent by the antennas are multiplied by the corresponding array element complex weight vectors and superposed, and beam forming is completed. According to the method, the fitting capability of the neural network and the segmentation forming strategy are fully utilized, the online calculation complexity is reduced, the beam forming steps and the network dimension are simplified, the beam forming efficiency and performance are improved, and the method can be used for signal enhancement and anti-interference equipment of communication, radar and satellites.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, and in particular relates to a phased beamforming method and system, which can be used in the fields of communication, radar, satellite, etc. Background Art

[0002] When the two communicating parties are sending and receiving information or the radar is receiving the target echo, the receiving end receives a mixed signal containing the target signal and other interference signals in space. While ensuring the quality of the received target signal, other interference signals need to be suppressed to achieve the reception of higher standard target signals. The appearance of interference signals will affect the reception and detection of target signals by the receiving end. Adaptive beamforming technology is a technology that suppresses interference in different directions and forms. The signal received by the array antenna can form a higher gain in its target direction through adaptive anti-interference technology, form a null in the interference and noise direction, suppress interference and noise, and increase the gain in the main lobe direction of the array antenna and reduce the gain in the interference plus noise direction.

[0003] The existing optimization solution methods of beamforming technology mainly adopt specific iterative algorithms and convex optimization algorithms. These iterative algorithms have the disadvantages of high computational complexity and long computational time. Therefore, beamforming technology cannot meet the high real-time requirements of the communication process in some scenarios. In order to solve this challenge, some researchers have proposed simple heuristic beamforming schemes. These heuristic beamforming schemes are directly calculated based on channel state information and do not require iteration, so they have low computational delay. However, due to the error in the arrival angle estimation, the amplitude and phase errors of the array element channel and the array element position error in the beamforming process, the MVDR algorithm cannot obtain accurate steering vectors and covariance matrices. Since then, a series of robust beamforming algorithms have been proposed, mainly including diagonal loading algorithm, worst performance optimal algorithm based on uncertain set constraints of steering vectors, projection subspace algorithm, interference plus noise covariance matrix reconstruction algorithm, etc. However, the reduction of delay in these algorithms comes at the cost of performance loss. The trade-off between delay and performance may limit the potential of beamforming technology and its application in practice. Due to recent advances in neural network technology, it becomes particularly important to predict the optimal beamforming weights in real time while considering performance and computational latency.

[0004] The use of neural network technology can train the neural network offline through a large amount of data, and retain the network parameters with better performance for online optimization. This method transfers the previous complex calculations of online optimization to the continuous iteration of offline training, and uses the trained neural network parameters to obtain the optimal beamforming solution, which greatly reduces the computational complexity and latency.

[0005] Patent document with application number CN202310899181.5 discloses a robust adaptive beamforming method based on a deep unfolding network, which uses a neural network to learn and predict interference signals and errors, then reconstructs the corresponding covariance matrix, and finally calculates the required weight vector based on other methods. This method does not fully utilize the learning ability and forward propagation of the neural network, but is only used to estimate the error, thereby increasing the robustness of the beamforming. In addition, some methods proposed in the literature use neural networks to predict the imaginary and real parts of the weight vector respectively, but this increases the output dimension of the neural network, resulting in a slowdown in the learning and prediction speed of the network. Summary of the invention

[0006] The purpose of the present invention is to address the defects of the above-mentioned prior art and propose a staged beamforming method and system based on a fully connected network to make full use of offline resources, improve beamforming efficiency, and make full use of the fitting ability of neural networks, simplify network output, reduce computational complexity and real-time running time, and improve beamforming efficiency.

[0007] The technical solutions to achieve the above objectives include:

[0008] Technical Solution 1: A phased beamforming method based on a fully connected network, characterized by comprising the following:

[0009] Pre-learning the neural network to form an amplitude weight predictor Net;

[0010] The amplitude weight predictor outputs the amplitude weight vector w according to the interference angle difference norm ;

[0011] Get the beam pointing angle and calculate the corresponding phase weight vector w based on it phase ;

[0012] Take the Hadamard product of the amplitude weight vector and the phase weight vector to obtain the required weight vector w;

[0013] The signal received or sent by each antenna is multiplied by the corresponding weight vector w and superimposed to complete beamforming.

[0014] Further, the neural network is pre-learned to form an amplitude weight predictor Net, including:

[0015] 2a) constructing a linear array including N array elements and a spacing between adjacent array elements of d, where d is the array element spacing, wherein d=λ / 2, λ is the signal wavelength, and N is an integer greater than or equal to 1;

[0016] 2b) Based on the number of array elements N, a real number amplitude weight matrix A in the range of [0,1] is randomly generated as training data;

[0017] 2c) Use IFFT to calculate the antenna pattern f for each column of the real amplitude weight matrix A m , forming the pattern matrix F:

[0018] F=[f1,f2,…,f m ,…,f M ], where M is the dimension of the training data set;

[0019] 2d) Search for the local minimum values ​​of each column of the directional matrix F to form a set V m , where the set V m The minimum value of the zero sink position u m (u m ≠±1), according to u m Calculate the null angle vector Θ corresponding to the pattern matrix F null :

[0020] Θ null =[arcsin(u1),arcsin(u2),…,arcsin(u m )…,arcsin(u M )];

[0021] 2e) Using the ReLU function as the intermediate layer activation function, the Sigmoid function as the output layer function, and the mean square error MSE function as the loss function, a single-dimensional input and N-dimensional output amplitude weight predictor Net is constructed;

[0022] 2f) Set the zero sink angle Θ null As input values, each column of the real amplitude weight matrix A is used as the network output target value to fit the network amplitude weight predictor Net.

[0023] Further, the amplitude weight predictor Net outputs an amplitude weight vector w according to the interference angle difference norm ,include:

[0024] 3a) Based on the received mixed signal X(l), the interference angle θ is calculated using the MUSIC algorithm J ;

[0025] 3b) Calculate the interference angle θ J The angle θ with the beam pointing angle θ0 is used as the network input for forward propagation, and the output amplitude weight w norm :

[0026] w norm =Net(θ)

[0027] Among them, Net is the amplitude weight predictor.

[0028] Further, the corresponding phase weight vector w is calculated according to the beam pointing angle phase , the formula is as follows:

[0029]

[0030] Where λ is the signal wavelength, θ0 is the beam pointing angle, [·] T represents transpose, d is the array element spacing, and j is the imaginary unit.

[0031] The Hadamard product of the amplitude weight vector and the phase weight vector is taken to obtain the required weight vector w, and the formula is as follows:

[0032] w=w phase ⊙w norm , where ⊙ represents the Hadamard product.

[0033] Technical Solution 2: A phased beamforming system based on a fully connected network, characterized by comprising:

[0034] A training data generation module is used to construct the data required for the amplitude weight predictor fitting process;

[0035] The interference angle estimation module is used to estimate the direction of the interference signal according to the currently received signal and output the angle;

[0036] The amplitude weight prediction module first performs fitting based on the data generated by the training data generation module, and then predicts the amplitude weight vector of the array element based on the fitted network parameters and the angle output by the interference angle estimation module;

[0037] A phase weight generation module, used to construct a phase weight vector according to beam pointing;

[0038] The array element weight synthesis module is used to synthesize the amplitude weight vector of the array element predicted by the amplitude weight prediction module and the phase weight vector constructed by the phase weight generation module into a final array element weight vector.

[0039] Compared with the prior art, the present invention has the following advantages:

[0040] Firstly, the present invention adopts the method of pre-learning the neural network, which can transfer the complex calculation of online optimization to the continuous iteration of offline training, and obtain the optimal beamforming solution by using the trained neural network parameters, thereby reducing the complexity of online calculation.

[0041] Secondly, the present invention directly constructs the angle θ between the interference and beam pointing and the amplitude weight vector w norm The mapping relationship gives full play to the learning ability of the fully connected network and simplifies the beamforming steps.

[0042] Thirdly, the present invention adopts a strategy of separately forming an amplitude weight vector and a phase weight vector, thereby simplifying the network output dimension of the amplitude weight predictor and shortening the training time and prediction time, thereby improving the beamforming efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of the implementation of the phased beamforming method based on a fully connected network of the present invention;

[0044] Figure 2 yes Figure 1 Schematic diagram of the predictor forward propagation and synthesis of the final weights.

[0045] Figure 3 It is a block diagram of the staged beamforming system based on a fully connected network of the present invention. DETAILED DESCRIPTION

[0046] The invention is further described in detail below with reference to the accompanying drawings.

[0047] Embodiment 1: A phased beamforming method based on a fully connected network.

[0048] Reference Figure 1 , the implementation scheme of this example includes:

[0049] (i) Pre-learning the neural network to form an amplitude weight predictor Net.

[0050] The neural networks can be divided into three main types: feedforward neural networks, feedback neural networks, and graph neural networks, where:

[0051] Feedforward neural network is a simple neural network, also known as multi-layer perceptron MLP. The signal is transmitted from the input layer to the output layer in one direction without feedback in between. Commonly used model structures include: convolutional neural network, BP neural network, RBF neural network, etc.

[0052] Feedback neural network is a directed cyclic graph or undirected graph with strong associative memory and optimization computing capabilities. Its output is not only related to the current input and network weights, but also to the previous input of the network. Commonly used model structures include: RNN, Hopfield network, Boltzmann machine, LSTM, etc.

[0053] Graph neural network GNN is a kind of network based on graph, which is a set of functions organized by graph structure in topological space for relational reasoning. Several network models of graph neural network include: graph convolutional network, graph autoencoder, graph generation network, graph recurrent network, graph attention network, etc.

[0054] Feedforward neural networks have strong function fitting capabilities, feedback neural networks have strong associative memory capabilities and can effectively associate contextual information, while graph neural networks are more suitable for modeling and analyzing complex and structured data. norm The mapping relationship is similar to the functional relationship, so the present invention uses a feedforward neural network. Moreover, since the network input is only one-dimensional data and the network output is multi-dimensional data, it does not require operations such as convolution to extract features from the input data. Therefore, the present invention preferably uses a BP neural network, that is, a fully connected network, but other types of network models can also be used.

[0055] Since most neural networks need to be trained to learn the characteristic information in the data, the present invention needs to pre-learn the preferred fully connected network to construct the angle θ between the interference and beam pointing and the amplitude weight vector w norm The mapping relationship is implemented as follows:

[0056] Step 1: Build a fully connected network.

[0057] The ReLU function is used as the intermediate layer activation function, the Sigmoid function is used as the output layer function, and the mean square error MSE function is used as the loss function. The network input is a single-value angle θ, and the output is the array element amplitude weight w norm , that is, the number of neurons in the input layer of the amplitude weight predictor is 1, the number of neurons in the output layer is the same as the number of array elements N, the hidden layer dimension and the corresponding number of neurons are adjusted as needed, and the total number of layers is L, forming a fully connected network, such as Figure 2 shown.

[0058] The above functions are expressed as follows:

[0059] The ReLU function is expressed as: The ReLU function can improve the computational efficiency and gradient descent of the neural network, reduce the problems of overfitting and gradient disappearance, and can converge quickly in SGD due to the linear and non-saturated properties of ReLU. In addition, the computational complexity is low and no exponential operations are required, which is why this function is used as the activation function of the intermediate layer in the present invention.

[0060] The activation function includes Sigmoid function, Tanh function, ReLU function, Softmax function, etc. This example adopts but is not limited to Sigmoid function, which is expressed as follows: The Sigmoid function is used as the final function of the network output layer to compress the output of the network to [0,1] to meet the requirements for the magnitude of the amplitude weight. In addition, its gradient is smooth, which is easy to derive and prevents sudden gradient changes during model training.

[0061] The Tanh function is expressed as: In fact, the Tanh function is a deformation of the Sigmoid function: Unlike Sigmoid, Tanh is "zero-centered", so in some applications, Tanh is better than Sigmoid. However, in the case of saturated neurons, Tanh still does not solve the gradient vanishing problem, and the present invention requires that the output result is in [0,1].

[0062] The Softmax function is expressed as: The Softmax function often acts as an activation function in the output layer of a neural network. The value of the output layer is mapped to the range of 0-1 through the activation function, and the neuron output is constructed into a probability distribution for multi-classification problems. The larger the mapping value of the Softmax activation function, the greater the possibility of the true category.

[0063] The MAE function is expressed as: Among them, x represents the independent variable of the function, e represents the natural constant, Net(x i ) is the predicted value output by the predictor, y i is the target value, i.e., the element of each column in the real magnitude weight matrix A. MAE is more robust to outliers because it is calculated with the same weight regardless of the error size (absolute errors do not amplify differences). It is often used in scenarios where outliers may represent important information or corrupt data.

[0064] The MSE function is expressed as: This function is the most commonly used loss function in regression problems. It is a function used to measure the difference between the model prediction result and the actual result. This example uses but is not limited to the absolute error function MAE as the loss function of the fully connected network. Compared with MAE, MSE is more suitable for accurate prediction scenarios.

[0065] Step 2: Construct data for network training and obtain the network amplitude weight predictor Net:

[0066] 2.1) constructing a linear array comprising N array elements with a spacing of d between adjacent array elements, where d is the array element spacing, wherein d=λ / 2, λ is the signal wavelength, and N is an integer greater than or equal to 1;

[0067] 2.2) According to the number of array elements N, a real number amplitude weight matrix A in the range of [0,1] is randomly generated as training data, which is expressed as follows:

[0068]

[0069] Among them, a n,mis the element in the nth row and mth column, n=1,2,3…,N, m=1,2,3…,M, N represents the number of array elements, M represents the size of the data set, and the elements in A satisfy: a n,m ∈[0,1];

[0070] 2.3) Use IFFT to calculate the antenna pattern f of each column in the real amplitude weight matrix A m :

[0071]

[0072] Among them, f m (θ) is the antenna pattern vector with an angle of θ calculated from the weight coefficients in the mth column of A, g n (θ) is the gain of each element antenna at angle θ. Here, it is assumed that each element is an omnidirectional antenna, that is, g n (θ)=1;a n,m is the real amplitude weight of each array element, θ0 is the angle between the array beam pointing and the normal;

[0073] 2.4) By f m The pattern matrix F is:

[0074] F=[f1,f2,…,f m ,…,f M ]

[0075] 2.5) Search for the local minimum values ​​of each column of the directional matrix F to form a set V m , where the set V m The minimum value of the zero sink position u m , its u m ≠±1; according to u m Calculate the null angle vector Θ corresponding to the pattern matrix F null :

[0076] Θ null =[arcsin(u1),arcsin(u2),…,arcsin(u m )…,arcsin(u M )];

[0077] 2.6) Set the zero sink angle θ null As input values, each column of the real amplitude weight matrix A is used as the network output target value;

[0078] 2.7) Use back propagation and Adam algorithm to adjust the network weight and bias of the amplitude weight predictor Net, and fit the network amplitude weight predictor Net:

[0079] 2.7.1) Back propagate the loss function MSE and calculate the current gradient value gt , and then by g t Calculate the first-order moment estimate m of the current gradient t :

[0080] m t =β1m t-1 +(1-β1)g t

[0081] Among them, β1 is the adjustable parameter of the first-order moment estimate;

[0082] 2.7.2) According to the current gradient value g t Calculate the second-order moment estimate v of the current gradient t :

[0083] v t =β2v t-1 +(1-β2)g t 2

[0084] Among them, β2 is the adjustable parameter of the second-order moment estimate;

[0085] 2.7.3) Estimate m based on the first-order moment of the current gradient t and the second moment estimate v t Update the fully connected network parameters to obtain the weight parameter W of the current updated network i and the bias parameter B i :

[0086]

[0087] Among them, W′ i and B′ i is the parameter before update, i=1,2,3,…,L, α is the learning rate; ε is v t Compensation parameters to prevent v t Approaching zero;

[0088] 2.7.4) According to the updated amplitude weight predictor parameter W i and B i , we get x1,x2,…,x i ,…,x L-1 Data transferred between layers:

[0089] x1=ReLU(W1θ+B1)

[0090] x2=ReLU(W2x1+B2)

[0091] …

[0092] x i =RelU(W i xi-1 +B i )

[0093] …

[0094] x L-1 =ReLU(W L-1 x L-2 +B L-1 )

[0095] Among them, W1, W2, …, W i ,…,W L-1 are the weight parameters of each layer, B1, B2,…, B i ,…,B L-1 is the bias parameter of each layer, L is the number of network layers, and ReLU is the activation function;

[0096] 2.7.5) According to the output data x of the L-1 layer L-1 and the weight parameter W of the Lth layer L and the bias parameter B L , we get the following expression of the amplitude weight predictor Net:

[0097] Net(θ)=Sigmoid(W L x L-1 +B L )

[0098] Among them, Sigmoid is the activation function, and Net(θ) is the output of the amplitude weight predictor when the input is θ.

[0099] (II) Forming the amplitude weight vector w in stages norm and the phase weight vector w phase , and the array element weight vector w is synthesized to complete the beamforming.

[0100] Step 3: Obtain the interference angle θ based on the mixed signal X(l) J , the amplitude weight predictor Net outputs the amplitude weight vector w according to the interference angle difference θ norm .

[0101] 3.1) The array antenna receives the mixed signal X(l) consisting of the communication signal and the interference signal sent by the transmitter:

[0102]

[0103] in, is the signal received by the nth array element at snapshot time l;

[0104] is the angle θ along the direction of arrival kThe direction vector element of the change, j is the imaginary unit, λ is the signal wavelength, d n is the array element spacing;

[0105] s k (l) is the kth complex signal s k The value at snapshot time l;

[0106] 3.2) Use the multi-signal decomposition algorithm MUSIC to calculate the interference angle θ of the received mixed signal X(l) J ;

[0107] 3.3) Calculate the interference angle θ J The angle θ with the beam pointing angle θ0 is used as the network input for forward propagation, and the output amplitude weight w norm :

[0108] w norm =Net(θ)

[0109] Among them, Net is the amplitude weight predictor.

[0110] Step 4: Calculate the phase weight vector w phase , and use it with the amplitude weight vector w norm The array element weight vector w is synthesized to finally complete the beamforming.

[0111] 4.1) According to the beam pointing angle θ0, the corresponding phase weight vector w is calculated phase :

[0112]

[0113] Where λ is the signal wavelength, θ0 is the beam pointing angle, [·] T represents transpose, d is the array element spacing, and j is the imaginary unit.

[0114] 4.2) Take the Hadamard product of the amplitude weight vector and the phase weight vector to obtain the required array element weight vector w:

[0115] w=w phase ⊙w norm

[0116] Wherein, ⊙ represents the Hadamard product.

[0117] 4.3) The mixed signal X(l) is weightedly superimposed by the array element weight vector w to obtain the processed signal Y:

[0118] Y=w T X(l)

[0119] in,[·] T Indicates transpose.

[0120] 4.4) Using the array element weight vector w, the antenna pattern f is calculated using IFFT:

[0121]

[0122] Where f(θ) is the antenna array gain in the direction of angle θ, and w n is the complex amplitude weight of each array element, and θ0 is the angle between the array beam pointing and the normal.

[0123] This completes the phased beamforming based on the fully connected network.

[0124] Embodiment 2: A phased beamforming system based on a fully connected network.

[0125] Reference Figure 3 The system of this example includes: a training data generation module 1, an interference angle estimation module 2, an amplitude weight prediction module 3, a phase weight generation module 4, and an array element weight synthesis module 5. Among them:

[0126] The training data generation module 1 is used to construct the data required for the amplitude weight predictor fitting process. The data construction method is to randomly generate array element amplitude weights, and then search for the null angle of the antenna radiation pattern under the amplitude weight, thereby forming the null angle and amplitude weight data. The generated training data is transmitted to the amplitude weight prediction module 3.

[0127] The interference angle estimation module 2 is used to estimate the direction of the interference signal according to the signal received by the current array and output the angle. The core of this module is the DOA estimation algorithm, which estimates the direction of the interference signal from the mixed signal received by the array, and then outputs the required angle by subtracting it from the beam pointing angle, and transmits the angle to the amplitude weight prediction module 3.

[0128] The amplitude weight prediction module 3 first performs fitting based on the data generated by the training data generation module 1, and then predicts the amplitude weight vector of the array element based on the fitted network parameters and the angle output by the interference angle estimation module 2. The core of this module is a fully connected network, which forms an amplitude weight predictor after learning the data transmitted by the training data generation module 1. The amplitude weight predictor performs forward propagation based on the angle input by the interference angle estimation module 2 to output the amplitude weight vector, and transmits the vector to the array element weight synthesis module 5.

[0129] The phase weight generating module 4 is used to construct a phase weight vector according to a known beam pointing angle, and output the phase weight vector to the array element weight synthesis module 5 .

[0130] The array element weight synthesis module 5 is used to synthesize the amplitude weight vector of the array element predicted by the amplitude weight prediction module 3 and the phase weight vector constructed by the phase weight generation module 4 into a final array element weight vector, that is, to obtain the Hadamard product of the amplitude weight vector and the phase weight vector.

[0131] The above description is only a specific example of the present invention and does not constitute any limitation to the present invention. Obviously, after understanding the content and principle of the present invention, professionals in this field may make various modifications and changes in form and details without departing from the principle and structure of the present invention. For example, the fully connected activation function and loss function may use other types in addition to the types used in this example; in the interference angle estimation, in addition to the MUSIC algorithm used in this example, other DOA estimation algorithms may also be used. However, these modifications and changes based on the idea of ​​the present invention are still within the scope of protection of the claims of the present invention.

[0132] It should be noted that the step numbers in the specification and patent claims of the present invention are only for a clear description of the implementation scheme of the present invention and for ease of understanding, and the order of the step numbers is not limited.

Claims

1. A phased beamforming method based on a fully connected network, characterized in that: The steps include: Pre-learning the neural network to form an amplitude weight predictor Net; The amplitude weight predictor outputs the amplitude weight vector w according to the interference angle difference norm ; According to the beam pointing angle, the corresponding phase weight vector w is calculated phase ; Take the Hadamard product of the amplitude weight vector and the phase weight vector to obtain the required weight vector w; The signal received or sent by each antenna is multiplied by the corresponding weight and superimposed to complete beamforming.

2. The method according to claim 1, characterized in that The neural network is pre-learned to form an amplitude weight predictor Net, including the following: 2a) constructing a linear array including N array elements and a spacing between adjacent array elements of d, where d is the array element spacing, wherein d=λ / 2, λ is the signal wavelength, and N is an integer greater than or equal to 1; 2b) Based on the number of array elements N, a real number amplitude weight matrix A in the range of [0,1] is randomly generated as training data; 2c) Use IFFT to calculate the antenna pattern f for each column of the real amplitude weight matrix A m , forming the pattern matrix F: F=[f1,f2,…,f m ,…,f M ] Among them, M is the dimension of the training data set; 2d) Search for the local minimum values ​​of each column of the directional matrix F to form a set V m , where the set V m The minimum value of the zero sink position u m (u m ≠±1), according to u m Calculate the null angle vector Θ corresponding to the pattern matrix F null : Θ null =[arcsin(u1),arcsin(u2),…,arcsin(u m )…,arcsin(u M )]; 2e) Using the ReLU function as the intermediate layer activation function, the Sigmoid function as the output layer function, and the mean square error MSE function as the loss function, a single-dimensional input and N-dimensional output amplitude weight predictor Net is constructed; 2f) Set the zero sink angle Θ null As input values, each column of the real amplitude weight matrix A is used as the network output target value to fit the network amplitude weight predictor Net.

3. The method according to claim 1, characterized in that The amplitude weight predictor Net outputs an amplitude weight vector w according to the interference angle difference. norm , including the following: 3a) Based on the received mixed signal X(l), the interference angle θ is calculated using the MUSIC algorithm J ; 3b) Calculate the interference angle θ J The angle θ with the beam pointing angle θ0 is used as the network input for forward propagation, and the output amplitude weight w norm : In norm =Net(θ) Among them, Net is the amplitude weight predictor.

4. The method according to claim 1, characterized in that: The corresponding phase weight vector w is calculated according to the beam pointing angle phase , the formula is as follows: Where λ is the signal wavelength, θ0 is the beam pointing angle, [·] T represents transpose, d is the array element spacing, and j is the imaginary unit. The Hadamard product of the amplitude weight vector and the phase weight vector is taken to obtain the required weight vector w, and the formula is as follows: in=in phase ⊙in norm Here, ⊙ represents the Hadamard product.

5. The method according to claim 2, characterized in that: The real amplitude weight matrix A generated in step 2b) is expressed as follows: Among them, a n,m is the element in the nth row and mth column, n=1,2,3…,N, m=1,2,3…,M, N represents the number of array elements, M represents the size of the data set, and the elements in A satisfy: a n,m ∈[0,1].

6. The method according to claim 2, characterized in that In step 2c), the antenna pattern vectors of each column of matrix A are calculated using the following formula: Among them, f m (θ) is the antenna pattern vector calculated from the mth column weight coefficients in A, g n (θ) is the antenna gain of each array element. Here, it is assumed that each array element is an omnidirectional antenna, that is, g n (θ) = 1. n,m is the real amplitude weight of each array element, and θ0 is the angle between the maximum beam pointing of the array and the normal.

7. The method according to claim 2, characterized in that: The ReLU function, Sigmoid function, and mean square error MSE function in step 2e) are expressed as follows: Among them, x represents the independent variable of the function, e represents the natural constant, Net(x i ) is the predicted value output by the predictor, y i is the target value, that is, the element of each column in the real amplitude weight matrix A.

8. The method according to claim 2, characterized in that: In step 2f), the network amplitude weight predictor Net is fitted, and the network weight and network bias are adjusted by back propagation and Adam algorithm, which includes: 2f1) According to the current gradient value g t Calculate the first-order moment estimate m of the current gradient t : m t =β1m t-1 +(1-β1)g t Among them, β1 is the adjustable parameter of the first-order moment estimate; 2f2) According to the current gradient value g t Calculate the second-order moment estimate v of the current gradient t : v t =β2v t-1 +(1-β2)g t 2 Among them, β2 is the adjustable parameter of the second-order moment estimate; 2f3) Estimate m based on the first-order moment of the current gradient t and the second-order moment estimate v t Update the network parameters to obtain the parameters W of the current updated network i and B i : Among them, W′ i and B′ i is the parameter before update, i=1,2,3,…,L, α is the learning rate; ε is v t Compensation parameters to prevent v t Approaching zero. 2f4) According to the fitted amplitude weight predictor parameter W i and b i , we get x1,x2,…,x i ,…,x L-1 Data transferred between layers: x1=ReLU(W1θ+B1) x2=ReLU(W2x1+B2) … x i =ReLU(W i x i-1 +B i ) … x L-1 =ReLU(W L-1 x L-2 +B L-1 ) Among them, W1, W2, …, W i ,…,W L-1 are the weight parameters of each layer, B1, B2,…, B i ,…,B L-1 is the bias parameter of each layer, L is the number of network layers, and ReLU is the activation function; 2f5) Using the output data x of the L-1 layer L-1 and the weight parameter W of the Lth layer L and the bias parameter B L , the amplitude weight predictor Net is expressed as follows: Net(θ)=Sigmoid(W L x L-1 +B L ) Among them, Sigmoid is the activation function, and net(θ) is the output of the amplitude weight predictor when the input is θ.

9. The method according to claim 3, characterized in that: The mixed signal X(l) received by the array antenna in step 3a) can be expressed as: in, is the signal received by the nth array element at snapshot time l; is the angle θ along the direction of arrival k The direction vector element of the change, j is the imaginary unit, λ is the signal wavelength, d n is the array element spacing; s k (l) is the kth complex signal s k The value at snapshot time l.

10. A phased beamforming system based on a fully connected network, characterized in that: include: A training data generation module is used to construct the data required for the amplitude weight predictor fitting process; The interference angle estimation module is used to estimate the direction of the interference signal according to the currently received signal and output the angle; The amplitude weight prediction module first performs fitting based on the data generated by the training data generation module, and then predicts the amplitude weight vector of the array element based on the fitted network parameters and the angle output by the interference angle estimation module; A phase weight generation module, used to construct a phase weight vector according to beam pointing; The array element weight synthesis module is used to synthesize the amplitude weight vector of the array element predicted by the amplitude weight prediction module and the phase weight vector constructed by the phase weight generation module into a final array element weight vector.

Images