Distributed multi-curved-surface array full-space collaborative beamforming method and system

Abstract

The application discloses a kind of distributed multi-curved surface array full airspace cooperative beam forming method and system, belong to the field of spacecraft TT&C, method includes steps: S1, each curved surface array is respectively to the antenna element in each internal first weighting, and respectively synthesis primary beam;S2, processing center carries out secondary weighting to the signal of distributed curved surface array, to synthesize an equivalent large aperture phased array antenna, and form the expected adaptive beam.The application solves the problem that the antenna array technology of traditional TT&C system can only be applied to single-beam, single-target TT&C scene, and solves the problem that the antenna array technology of traditional TT&C system does not have the ability of adaptive suppression airspace interference.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aircraft TT&C, more particularly, to a distributed multi-curved-surface array full-space-domain cooperative beam forming method and system. BACKGROUND

[0002] The prior art related to the present application is introduced as follows:

[0003] 1) As for aircraft TT&C, TT&C is the abbreviation of measurement and control, measurement refers to the measurement and telemetry of the flight trajectory of the aircraft, control refers to the remote control of the aircraft, TT&C is abbreviated as TT&C (Tracking, Telemetering and Command), the first T refers to tracking and orbit determination, the second T refers to telemetry, and C refers to command and control. Aircraft includes aircraft (aircraft with an altitude of less than 20 km), near-space aircraft (aircraft with an altitude of 20-100 km), space aircraft (aircraft with an altitude of more than 100 km), and deep-space aircraft (aircraft with an altitude of more than 2x106 km). These aircrafts are the objects of TT&C, and specific aircrafts include satellites, spacecraft, space stations, deep-space probes, near-space aircrafts, unmanned aerial vehicles, airships, balloons, weapon platforms, etc. TT&C communication equipment includes radio, optical, and infrared TT&C communication equipment, and the main equipment in the current TT&C communication field is the radio continuous wave TT&C communication system.

[0004] 2) As for antenna array, with the development of deep-space exploration technology, the amount of data transmitted by the task is increasing, and the distance of deep-space exploration is increasing, resulting in a great loss of transmission path. Since the equivalent isotropically radiated power (EIRP) of the probe is very limited, it is necessary to use a large-diameter parabolic antenna at the ground station to increase the receiving area as much as possible to improve the signal-to-noise ratio (SNR). Taking the Mars exploration mission as an example, according to the parameters of the NASA mission, 1Mbps of data needs an antenna with a diameter of 80m. However, the larger the antenna aperture, the narrower the beam, and the more difficult it is to capture the downlink signal of the probe. At the same time, the large antenna structure and heavy weight make it very difficult to drive the servo. The 70m large antenna (weight 7260 tons, height 76m) of the NASA Deep Space Network (DSN) has reached the current engineering limit, therefore, it is necessary to use antenna array technology to realize a larger equivalent antenna aperture through multi-antenna signal synthesis.

[0005] 3) On the new generation of curved array TT&C system, in the new generation of TT&C system, super large scale curved array TT&C antenna is introduced. Due to its full airspace coverage array configuration and the use of multi-beam forming technology, the curved array antenna can realize single phased array antenna tracking and controlling multiple targets in full airspace at the same time. In the future, for the scenario of distributed deployment of multiple curved arrays, through cooperative array signal processing, an equivalent high gain array antenna with larger aperture can be formed, which will have the following potential advantages: 1) according to the specific requirements of TT&C tasks, the curved array resources are flexibly scheduled and managed, realizing higher resource utilization; 2) by adding adaptive beam forming algorithm, adaptive beam forming is supported, and the spatial interference suppression capability is possessed; 3) supporting mobile and flexible deployment of curved array, and updating the weight value in real time; 4) having partial invulnerability, and good system robustness.

[0006] 4) On adaptive array antenna, adaptive array antenna (Adaptive Array Antennas) forms a controllable antenna beam pattern by weighting the antenna elements, and uses adaptive filtering algorithm to dynamically optimize and adjust the weighting coefficient of each antenna element, so that the main lobe of the synthesized beam of the antenna array points to the desired direction, while producing beam nulls in the direction of the interference source, effectively resisting suppressive interference, and adaptively tracking the target user terminal, playing the role of spatial filtering. The anti-interference technology based on adaptive array antenna has been widely applied in navigation system anti-interference receiver, various aircrafts and other fields.

[0007] 5) On adaptive beam forming algorithm, adaptive beam forming, as one of the key technologies of array signal processing, has been widely applied in radar, communication, medical and sonar fields. Common adaptive beam forming algorithms mainly include least mean square (Least mean square, LMS) algorithm, sample matrix inversion (Sample matrix inversion, SMI) algorithm and minimum variance distortionless response (Minimum variance distortionless response, MVDR) algorithm. LMS algorithm cannot balance the contradiction between convergence speed and steady-state error; when the signal-to-noise ratio rises, the robustness of SMI algorithm will decrease; MVDR algorithm needs matrix inversion, which has large amount of calculation and is not suitable for real-time updating of optimal weight vector.

[0008] 6)About radial basis function neural network, radial basis function neural network (Radial Basis Function Neural Network, RBFNN) is a simple structure of three-layer feedforward neural network, has the characteristics of fast approximation, simple structure, fault tolerance and strong inductive ability for any nonlinear function, is widely used in signal processing and adaptive control and other fields. Radial basis function neural network is composed of input layer, hidden layer and output layer. Among them, the activation function of hidden layer neuron is called radial basis function. According to the analysis of functional theory, the function approximation ability of network can be improved when using Gaussian function as radial basis function.

[0009] In some prior art schemes, the traditional antenna array of TT&C system refers to the array composed of multiple parabolic antennas in a certain geographical range, which synthesizes the signals of the same signal source received by each antenna to improve the signal-to-noise ratio. The basic principle is to obtain and eliminate the time and phase difference between the signals of each antenna, and then coherently add to synthesize a narrow beam with high gain. The commonly used algorithms of traditional antenna array of TT&C system include Simple, Sumple, Eigen, etc. Under the condition that the noises of each antenna are not related, the SNR of the received signal of N antenna array is N times that of a single antenna in theory. If small and medium caliber antennas with small volume, light weight and wide beam are used for array, it is more beneficial for signal acquisition and tracking. In addition, in terms of capacity expansion, reliability, flexibility, multi-target communication and cost, the antenna array has more advantages than a single large antenna. The main shortcomings of the traditional antenna array of TT&C system are: (1) The scheme and algorithm of the traditional antenna array of TT&C system are designed for parabolic TT&C antennas, and can only be applied to the scene of single-beam and single-target TT&C. The curved array antenna in the new generation of TT&C system adopts multi-beam forming technology and supports full-space multi-target simultaneous TT&C. Therefore, the traditional antenna array technology of TT&C system cannot meet the needs of multi-curved array full-space collaborative beam forming of the new generation of TT&C system. (2) Since the traditional antenna array algorithm of TT&C system does not use adaptive beam forming algorithm, it only forms high gain beam for the target, and therefore does not have the ability to adaptively suppress spatial interference.

[0010] In other prior art solutions, the currently common adaptive beam forming algorithms include: Least Mean Square (LMS) algorithm, Sample Matrix Inversion (SMI) algorithm, Minimum Variance Distortionless Response (MVDR) algorithm, and the features of each algorithm are introduced as follows: (1) LMS adaptive beam forming algorithm, the LMS adaptive beam forming algorithm is a simple and relatively universal adaptive beam forming algorithm, and is also an algorithm using the steepest gradient descent method and the minimum mean square error criterion in the local Wiener filter theory. The LMS algorithm is favored by people due to its simple algorithm structure, fewer parameters, faster convergence and easy implementation. (2) SMI adaptive beam forming algorithm, the SMI adaptive beam forming algorithm is an algorithm based on the minimum mean square error criterion, and the principle of the algorithm is to directly obtain the inverse of the sample covariance matrix. (3) MVDR adaptive beam forming algorithm, the MVDR adaptive beam forming algorithm constrains the power of the expected signal direction to 1 and minimizes the array output power, and the MVDR adaptive beam can suppress interference and noise, because the interference and noise power of the array output is minimum at this time. The main shortcomings of the traditional adaptive beam forming technology are as follows: (1) the LMS algorithm cannot balance the contradiction between the convergence speed and the steady-state error; (2) the robustness performance of the SMI algorithm decreases when the signal-to-noise ratio increases; (3) the MVDR algorithm needs matrix inversion, and has high computational complexity and large amount of calculation, so it is difficult to update the antenna array weight vector in real time; (4) for the adaptive beam forming of large-scale antenna arrays, the above algorithms all have the problem of difficult implementation. SUMMARY

[0011] The purpose of the present application is to overcome the shortcomings of the prior art, and provide a distributed multi-curved surface array full-space cooperative beam forming method and system. In view of the shortcomings of the traditional antenna array technology of the measurement and control system, a distributed multi-curved surface array full-space beam forming scheme is proposed for the scene of the distributed deployment of the multi-curved surface array in the new generation of measurement and control system, which supports the requirements of multi-beam forming and flexible resource scheduling, solves the problem that the traditional antenna array technology of the measurement and control system can only be applied to the single-beam and single-target measurement and control scene, and at the same time, an adaptive beam forming algorithm is added to support real-time dynamic updating of beam weights and solve the problem that the traditional antenna array technology of the measurement and control system does not have the adaptive interference suppression capability in the space.

[0012] The purpose of the present application is achieved by the following scheme:

[0013] A distributed multi-curved array full-space collaborative beamforming method includes multiple distributed curved arrays and a processing center. The multiple distributed curved arrays are connected to the processing center to transmit parameters, control signals, received signals, and transmitted signal data. The processing center is used to control the multiple distributed curved arrays to complete their own beamforming and collaborative beamforming. The method specifically includes the following steps:

[0014] S1, each curved array performs first-level weighting on its internal antenna elements and synthesizes primary beams respectively;

[0015] In S2, the processing center performs secondary weighting on the signals of the distributed curved array to synthesize an equivalent large-aperture phased array antenna and form the desired adaptive beam.

[0016] Furthermore, in step S1, each of the curved arrays performs primary weighting on the antenna elements within each of the arrays and synthesizes primary beams respectively, which specifically includes the following sub-steps: in the primary weighting stage, each curved array uses the CBF algorithm to generate its own internal element weight vector.

[0017] Furthermore, each of the curved arrays uses the CBF algorithm to generate its own internal element weight vector, which specifically includes the following sub-steps:

[0018] S1-1, the processing center will expect the beam angle Send to each surface array;

[0019] S1-2, each curved array performs a first-level weighting on the antenna elements within it and synthesizes the primary beam respectively; for the i-th curved array, P is activated according to the angle of the desired beam i antenna array elements, a three-dimensional rectangular coordinate system is established with the phase center of the curved array antenna as the origin, and the coordinates of the pth antenna array element are expressed as ξ p =[x p ,y p ,z p ] T , p represents the antenna element number, p=1,…,P i , x p represents the x-axis coordinate of the antenna array element p, y p represents the y-axis coordinate of the antenna array element p, z p represents the z-axis coordinate of antenna element p, [·] T Represents the transpose of a matrix or vector; the i-th surface matrix is ​​based on P i The three-dimensional rectangular coordinates of the active antenna elements and the angle of the desired beam The antenna array weight vector w is calculated according to the following formula i ;

[0020]

[0021] wherein, g(·) is a linear function of input ; the i-th curved array utilizes a weight vector w i to synthesize a primary beam and output a received composite signal x i (k).

[0022] Further, in step S2, the processing center performs secondary weighting on the signals of the distributed curved arrays, specifically including: in the secondary weighting stage, the processing center generates the weight vector between the curved arrays by neural network fitting, which can support real-time updating of the antenna array weight vector.

[0023] Further, in step S2, the processing center generates the weight vector between the curved arrays by neural network fitting, which can support real-time updating of the antenna array weight vector, specifically including the following sub-steps:

[0024] In the secondary weighting process, the i-th curved array sends its position coordinate information, delay τ i and received signal x i (k) to the processing center, wherein the delay τ i includes a fiber delay;

[0025] The processing center combines the received parameters and data of the multiple curved arrays and the expected beam direction into a real vector and inputs it into the input layer of the neural network, which outputs the antenna array weight vector w NN in real time after multiple layers of weighting and nonlinear processing.

[0026] The processing center then performs weighting processing on the signals of the multiple curved arrays using w NN .

[0027] Further, in step S2, the processing center performs secondary weighting on the signals of the distributed curved arrays, including an initialization stage, which specifically performs the following sub-steps:

[0028] S2-3, the processing center receives the position coordinate information, delay τ i , received signal x i (k) and node state information sent by the i-th curved array;

[0029] S2-4, the processing center determines the number N of curved arrays that need to perform cooperative beamforming according to the measurement and control link budget requirement or the antenna gain requirement;

[0030] S2-5, the processing center determines whether the current curved array is normal according to the received parameters of each curved array, the received signal and the node state information, and selects N normal working curved arrays as an effective curved array set;

[0031] S2-6, the processing center receives the parameters and the received signal of the N normal curved arrays, randomly specifies a spatial reference point as a coordinate origin, and establishes a self-defined three-dimensional rectangular coordinate system, and the processing center converts the position coordinates of each curved array into the coordinates of the self-defined three-dimensional rectangular coordinate system, and represents as Ω i i i i T , i represents the number of the curved array, x i represents the x-axis coordinate of the curved array i, y i represents the y-axis coordinate of the curved array i, and z i represents the z-axis coordinate of the curved array i, and the superscript T represents transposition.

[0032] Further, in step S2-5, the determination of whether the current curved array is normal and the selection of N normal working curved arrays as an effective curved array set specifically include the following sub-steps:

[0033] If any of the following three conditions is met, it means that the curved array h has been damaged or broken, and the curved array h is discarded without processing; otherwise, the curved array h is added to the effective curved array set.

[0034] Condition a), the processing center does not receive the data of the curved array h or receives incomplete data;

[0035] Condition b), the processing center receives abnormal parameter or signal data of the curved array h;

[0036] Condition c), the processing center receives abnormal node state of the curved array h.

[0037] Further, in step S2, the processing center performs two-level weighting on the signal of the distributed curved array, including a training phase, and the following sub-steps are specifically performed in the training phase:

[0038] S2-7, the processing center calculates the antenna array weight vector w opt according to the three-dimensional rectangular coordinates Ω, the delay τ, the received array signal vector x(k) and the angle of the expected beam of the N curved arrays received by the processing center according to the adaptive beam forming algorithm.

[0039]

[0040] wherein Ω = {Ω1, Ω2, …, Ω N ​​​​} represents the three-dimensional rectangular coordinates of N curved arrays, Ω i =[x i ,y i ,z i ] T represents the three-dimensional rectangular coordinates of the i-th surface matrix; τ=[τ1,τ2,…,τ N ] T Represents the delay vector from N curved arrays to the processing center, with a dimension of N×1, τ i represents the delay from the i-th curved array to the processing center, i=1,2,…,N, N represents the number of curved arrays; x(k)=[x1(k),x2(k),…,x N (k)] T represents the array signal vector obtained by the k-th sampling, with a dimension of N×1; represents the desired beam direction, θ0 is the horizontal azimuth angle, is the vertical pitch angle; f(·) is the angle with respect to the input Nonlinear function of

[0041] S2-8, the processing center will w opt As training samples to train the neural network so that the output of the neural network approaches w opt , and finally output the antenna array weight vector w NN ;

[0042] S2-9, the processing center calculates the training error vector e=w NN -w opt , and determine whether to end training; calculate the second-order norm of the error vector ‖e‖ 2 , Among them, e i represents the i-th element in the error vector, Σ(·) represents the summation function;

[0043] If the second-order norm of the error vector ||e|| 2 If it is greater than the preset threshold Threshold1, the training error vector e is fed back to the neural network to correct the parameters of the neural network, and the process jumps to step S2-7 to continue iterative training; otherwise, if the second-order norm of the error vector ||e|| 2 If the value is less than or equal to the preset threshold Threshold1, the training ends.

[0044] Furthermore, in step S2-8, the processing center w opt As training samples to train the neural network so that the output of the neural network approaches w opt , and finally output the antenna array weight vector w NN , specifically including the following sub-steps:

[0045] S2-8-1, the three-dimensional rectangular coordinates Ω, the time delay τ, the received array signal vector x(k), the angle of the expected beam of the N curved surface arrays are input into the trained neural network, and the output antenna array weight vector w is updated through multi-layer neural network weighting and nonlinear processing. inputting the neural network;

[0046] S2-8-2, the neural network is subjected to multi-layer weighting and nonlinear processing, and finally the antenna array weight vector w is output. NN , the dimension is N x 1.

[0047] Further, in step S2, the processing center performs secondary weighting on the signals of the distributed curved surface arrays, including a normal working phase, and the following sub-steps are specifically performed in the normal working phase:

[0048] S2-10, the processing center continuously inputs the three-dimensional rectangular coordinates Ω, the time delay τ, the received array signal vector x(k), and the angle of the expected beam of the N curved surface arrays into the trained neural network, and updates the output antenna array weight vector w' through multi-layer neural network weighting and nonlinear processing. NN , the dimension is N x 1.

[0049] S2-11, the processing center uses the updated antenna array weight vector w' NN to weight and sum the received array signal x(k) to obtain a synthesized signal y(k):

[0050]

[0051] to form an expected adaptive receiving beam.

[0052] S2-12, the processing center uses the updated antenna array weight vector w' NN to weight and output a single transmitting signal s(k) to obtain a transmitting array signal vector s(k):

[0053]

[0054] and distribute the N elements in s(k) to the N curved surface arrays for transmission, so as to form an expected adaptive transmitting beam.

[0055] S2-13, during the normal working process, if the processing center finds that there is an abnormal curved surface array h added, the curved surface array h is removed from the effective curved surface array set, and then the neural network is retrained by jumping to step S2-7; otherwise, jump to step S2-10 to continue updating the parameters and iteration.

[0056] ​A distributed multi-curved-surface array full-space collaborative beamforming system comprises a computer device, and a computer program is stored in a memory of the computer device, and when the computer program is loaded by a processor of the computer device, the method according to any one of the preceding items is executed.

[0057] The beneficial effects of the present application include:

[0058] (1) The present application supports unified and flexible scheduling and management of curved-surface array resources: a processing center selects a proper number of curved-surface arrays for collaborative beamforming according to the requirements of a measurement and control link budget or antenna gain, and the processing center can uniformly schedule and manage the curved-surface array resources according to the requirements of a measurement and control task, while generating multiple independent beams to respectively communicate with multiple spacecrafts in different regions of the sky, thereby achieving higher resource utilization.

[0059] (2) The present application supports adaptive beamforming capability: adaptive beamforming is supported due to the addition of an adaptive beamforming algorithm, and the adaptive beamforming has spatial interference suppression capability.

[0060] (3) The present application supports real-time signal processing capability of distributed multi-curved-surface array full-space collaborative beamforming: two-stage weighting is adopted to respectively realize beam synthesis of the curved-surface array itself and distributed collaborative beam synthesis between the curved-surface arrays, thereby reducing the implementation complexity of the signal processing of the collaborative beamforming.

[0061] (4) The present application supports real-time updating of antenna array weight vectors: a processing center generates antenna array weight vectors by using a neural network fitting, without complex operations such as matrix inversion, thereby reducing the operation complexity, reducing the operation amount, reducing the demand for computing resources, and supporting real-time updating of the antenna array weight vectors.

[0062] (5) The present application supports dynamic adaptation to real-time changes of array manifolds: a processing center takes the coordinate positions and fiber delays of the curved-surface arrays as inputs of a neural network, and real-time calculates and updates antenna array weight vectors, which can dynamically adapt to scenarios of real-time changes of array manifolds, such as scenarios of dynamic changes of the positions of the curved-surface arrays.

[0063] (6) The present application supports dynamic tracking and suppression of spatial interference: a processing center takes a received array signal vector as an input of a neural network, and real-time calculates and updates antenna array weight vectors, which can dynamically adapt to changes in directions of interference sources in the received signal, and real-time adjust the null points of the beams to align the interference, thereby achieving dynamic tracking and suppression of spatial interference.

[0064] (7) The application has partial anti-destroying capability and good system robustness: each curved surface array sends its state information to the processing center, the processing center selects N normal working curved surface arrays as an effective curved surface array set, when a small number of curved surface arrays are destroyed, the processing center removes the invalid curved surface arrays from the effective curved surface array set, re-trains the neural network and updates the antenna array weight vector in time, so as to continue to maintain the expected beam pattern, only a weak performance loss is caused, and the partial anti-destroying capability is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0065] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description only constitute some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor under the premise of the drawings.

[0066] Figure 1 The figure is a schematic diagram of the system of the present application;

[0067] Figure 2 The figure is a schematic diagram of the curved surface array of the present application;

[0068] Figure 3 The figure is a schematic diagram of the two-stage weighting structure of the present application;

[0069] Figure 4 The figure is a system working process of the present application. DETAILED DESCRIPTION

[0070] All features disclosed in the embodiments of the present specification, or all steps in the methods or processes impliedly disclosed, can be combined and / or extended, replaced, in any manner, except for mutually exclusive features and / or steps.

[0071] The abbreviations in the present application are explained as follows:

[0072] RBFNN: Radial Basis Function Neural Network, radial basis function neural network;

[0073] LMS: Least Mean Square, minimum mean square error (beam forming algorithm);

[0074] SMI: Sample Matrix Inversion, sample matrix inversion (beam forming algorithm);

[0075] MVDR: Minimum Variance Distortionless Response, minimum variance distortionless response (beam forming algorithm);

[0076] CBF: Conventional Beamforming, conventional beamforming.

[0077] In the overall technical concept of the application, the preferred embodiment provides a distributed multi-curved array full-space cooperative beamforming method. The system of the method comprises N distributed curved arrays and a processing center, and the N distributed curved arrays and the processing center are connected through optical fibers to transmit parameters, control signals, received and transmitted signal data, and the processing center controls the N distributed curved arrays to complete cooperative beamforming. First, each curved array performs one-level weighting on the antenna elements in the respective internal curved array, and then synthesizes a primary beam, and then the processing center performs two-level weighting on the signals of the distributed curved arrays through a neural network algorithm. In the two-level weighting process, the i-th curved array sends its position coordinate information, optical fiber delay τ i , and received signal x i (k) to the processing center, the processing center combines the above parameters, data and expected beam direction of the N curved arrays into a real vector, and inputs the real vector into the input layer of the neural network. After multiple layers of weighting and nonlinear processing, the neural network outputs an antenna array weight vector w NN in real time, and the processing center performs weighting processing on the signals of the N curved arrays using w NN , thereby synthesizing an equivalent large-aperture phased array antenna and forming an expected adaptive beam.

[0078] The overall technical solution of the application uniformly schedules and manages the curved array resources through the processing center, and adopts a two-level weighting method to realize the beam synthesis of the curved array itself and the distributed cooperative beam synthesis between the curved arrays, thereby reducing the signal processing implementation complexity of the large-scale distributed array antenna. At the same time, the neural network is used to realize the distributed cooperative beam synthesis between the curved arrays, which reduces the demand for computing resources compared with the traditional algorithm, supports real-time updating of the antenna array weight vector, supports dynamic adaptation to the scene of real-time change of the array manifold, supports dynamic tracking and suppression of interference in different directions, and has partial anti-destroying capability.

[0079] Further, the system composition and working process of the application are described in detail as follows in combination with the drawings: as Figure 1As shown, the system of the present invention includes N distributed curved arrays (a first curved array 101, a second curved array 102, a third curved array 103, and an Nth curved array 104) and a processing center 105. The N distributed curved arrays are connected to the processing center 105 via optical fiber, enabling transmission of parameters, control signals, and receive and transmit signal data. The processing center 105 controls the N distributed curved arrays to implement collaborative beamforming. Each curved array can independently receive and transmit signals, acting as an independent antenna unit. The processing center 105 calculates weights for the N curved arrays and performs receive, combine, and transmit signal generation for the N curved arrays.

[0080] like Figure 2 As shown, the curved array i of the present invention is a traditional centralized array antenna, and its synthetic beam points to the target direction. Among them, θ0 is the horizontal azimuth angle, that is, the angle between the XY plane and the positive X axis, is the vertical pitch angle, that is, the angle with the XY plane.

[0081] like Figure 3 As shown, the present invention implements distributed collaborative beamforming between curved arrays through a two-stage weighting approach, achieving both internal curved array beamforming and distributed collaborative beamforming between arrays. In the first stage of weighting, each curved array uses the CBF algorithm to generate its own internal element weight vectors. In the second stage of weighting, the processing center uses neural network fitting to generate weight vectors between the curved arrays, supporting real-time updates of the antenna array weight vectors.

[0082] like Figure 4 As shown, the system workflow of the present invention is divided into two stages:

[0083] Step 1: First-level weighting stage: Each curved array performs first-level weighting on its internal antenna elements and synthesizes primary beams respectively.

[0084] Step 2, secondary weighting stage: The processing center performs secondary weighting on the synthetic signal of the distributed curved array through a neural network algorithm, thereby synthesizing an equivalent large-aperture phased array antenna and forming the desired adaptive beam.

[0085] Among them, the secondary weighted stage can be further divided into three sub-stages:

[0086] (1) Initialization phase: complete the initial configuration of parameters, screening of valid surface array sets, and conversion of surface array coordinates;

[0087] (2) Training phase, completing the training of the neural network;

[0088] (3) During the normal operation phase, the trained neural network is used to generate antenna array weight vectors in real time, and the antenna array weight vectors are used to generate receiving beams and transmitting beams.

[0089] Further, the system workflow of the present application is described as follows:

[0090] 1. First-level weighting stage:

[0091] (1) The processing center sends the angle of the desired beam to each curved array.

[0092] (2) Each curved array respectively weights the antenna elements in itself and respectively synthesizes the primary beam.

[0093] Specifically, for the i-th curved array, according to the angle of the desired beam, P i antenna elements are activated, a 3-dimensional rectangular coordinate system is established with the phase center of the curved array antenna as the origin, and the coordinates of the p-th antenna element are represented as ξ p = [x p , y p , z p ] T , where p represents the antenna element number, p = 1, …, P i , x p represents the x-axis coordinate of the antenna element p, y p represents the y-axis coordinate of the antenna element p, z p represents the z-axis coordinate of the antenna element p, and [·] T represents the transpose of a matrix or a vector.

[0094] The i-th curved array calculates the 3-dimensional rectangular coordinates of the P i activated antenna elements according to the angle of the desired beam , and calculates the antenna array weight vector w i according to the conventional beam forming algorithm.

[0095]

[0096] wherein g(·) is a linear function of the input .

[0097] The i-th curved array synthesizes the primary beam using the weight vector w i , and outputs the received synthesized signal xi(k).

[0098] 2. Second-level weighting stage:

[0099] Initialization stage:

[0100] (3) The i-th curved array sends its position coordinate information, fiber delay τ i , and received signal x i to the processing center.​​(k) and node status information is sent to the processing center.

[0101] (4) The processing center determines the number of curved surface arrays N that need to be cooperatively beamformed according to the requirements of the telemetry link budget or the requirements of the antenna gain.

[0102] (5) The processing center determines whether the current curved surface array is normal according to the received parameters, received signals and node status information of each curved surface array, and selects N normal working curved surface arrays as the effective curved surface array set.

[0103] The processing center determines the criteria for the curved surface array h to be an abnormal node:

[0104] a) The processing center does not receive data from the curved surface array h or receives incomplete data;

[0105] b) The processing center receives abnormal parameters or signal data from the curved surface array h, such as parameters exceeding the normal range or signal data being all zeros;

[0106] c) The processing center receives abnormal node status from the curved surface array h, such as node status showing a radio frequency module fault code;

[0107] If any of the above three points is met, it means that the curved surface array h has been damaged or broken, so the curved surface array h is discarded and not processed; otherwise, the curved surface array h is added to the effective curved surface array set.

[0108] (6) After the processing center receives the parameters and received signals of all N normal curved surface arrays, it arbitrarily specifies a spatial reference point as the coordinate origin and establishes a self-defined 3D rectangular coordinate system. The processing center converts the position coordinates of each curved surface array into the coordinates of the self-defined 3D rectangular coordinate system, denoted as Ω i = [x i , y i , z i ] T , where i represents the curved surface array number, x i represents the x-axis coordinate of the curved surface array i, y i represents the y-axis coordinate of the curved surface array i, and z i represents the z-axis coordinate of the curved surface array i, and the superscript T represents transposition.

[0109] Training phase:

[0110] (7) The processing center calculates the antenna array weight vector w opt according to the received 3D rectangular coordinates Ω of the N curved surface arrays, the fiber delay τ, the received array signal vector x(k) and the angle of the desired beam according to the adaptive beamforming algorithm.

[0111]

[0112] Where, Ω={Ω1,Ω2,…,Ω N} represents the 3D rectangular coordinates of N curved arrays, Ω i =[x i ,y i ,z i ] T represents the 3D rectangular coordinates of the i-th surface matrix; τ=[τ1,τ2,…,τ N ] T represents the optical fiber delay vector from N curved arrays to the processing center, with a dimension of N×1, τ i represents the optical fiber delay from the i-th curved array to the processing center, i=1,2,…,N, N represents the number of curved arrays; x(k)=[x1(k),x2(k),…,x N (k)] T represents the array signal vector obtained by the k-th sampling, with a dimension of N×1; represents the desired beam direction, θ0 is the horizontal azimuth, that is, the angle between the X axis and the XY plane, is the vertical pitch angle, that is, the angle with the XY plane; f(·) is the angle with respect to the input nonlinear function.

[0113] (8) The processing center will w opt As training samples to train the neural network so that the output of the neural network approaches w opt .

[0114] (8-1) The 3D rectangular coordinates Ω of the N curved arrays, the fiber delay τ, the receiving array signal vector x(k), and the angle of the desired beam are Enter the neural network.

[0115] (8-2) The neural network undergoes multi-layer weighting and nonlinear processing, and finally outputs the antenna array weight vector w NN , the dimension is N×1.

[0116] (9) The processing center calculates the training error vector e = w NN -w opt , and determine whether to end the training.

[0117] Calculate the 2nd order norm of the error vector ||e|| 2 , Among them, e i represents the i-th element in the error vector, and Σ(·) represents the summation function.

[0118] If the second-order norm of the error vector ||e|| 2If the 2-norm of the error vector is greater than a preset threshold Threshold1, the training error vector e is fed back to the neural network for correcting the parameters of the neural network, and the step (7) is jumped to continue the iterative training; otherwise, if the 2-norm of the error vector is less than or equal to the preset threshold Threshold1, the training is ended. 2 NN

[0119] Normal working phase:

[0120] (10) The processing center continuously inputs the 3-dimensional rectangular coordinates Omega of the N curved surface arrays, the fiber delay Tau, the received array signal vector x(k), the angle of the expected beam to the trained neural network, and the output antenna array weight vector w' is updated through multi-layer neural network weighting and non-linear processing. NN , and the dimension is N*1.

[0121] (11) The processing center uses the updated antenna array weight vector w' NN to perform weighted summation on the received array signal x(k) to obtain a synthesized signal y(k):

[0122]

[0123] Thus, the expected adaptive receiving beam is formed.

[0124] (12) The processing center uses the updated antenna array weight vector w' NN to perform weighted summation on the received array signal x(k) to obtain a synthesized signal y(k):

[0125]

[0126] Thus, the expected adaptive receiving beam is formed.

[0127] (13) In the normal working process, if the processing center finds that there is a new abnormal curved surface array h, the curved surface array h is removed from the effective curved surface array set, and then the step (7) is jumped to retrain the neural network; otherwise, the step (10) is jumped to continue updating the parameters and iterative work.

[0128] The technical scheme of the embodiment of the application has the following advantages:

[0129] 1) In the above technical scheme of the embodiment of the application, the beam synthesis of the curved surface array itself and the distributed cooperative beam synthesis between the curved surface arrays are respectively realized through two-stage weighting.

[0130] 2) In the above technical scheme of the embodiment of the application, in the first-stage weighting phase, each curved surface array respectively generates the internal element weight vector of itself by using the CBF algorithm.

[0131] 3) In the above technical solution of the embodiment of the present invention, in the secondary weighting stage, the processing center uses neural network fitting to generate weight vectors between curved arrays, supporting real-time updating of antenna array weight vectors.

[0132] 4) In the above technical solution of the embodiment of the present invention, each curved array sends its own position coordinate information, fiber delay τ i , receive signal x i (k) and node status information are sent to the processing center.

[0133] 5) In the above technical solution of the embodiment of the present invention, the processing center determines the number N of curved arrays required for collaborative beamforming based on the measurement and control link budget requirement or the antenna gain requirement.

[0134] 6) In the above technical solution of the embodiment of the present invention, the processing center selects N curved arrays that are working normally as the effective curved array set. When some curved arrays are destroyed, the processing center removes the invalid curved arrays from the effective curved array set, retrains the neural network and promptly updates the antenna array weight vector, thereby continuing to maintain the desired beam pattern and having partial anti-destruction capability.

[0135] 7) In the above technical solution of the embodiment of the present invention, the processing center determines the standard of the surface array h as an abnormal node:

[0136] a) In the above technical solution of the embodiment of the present invention, the processing center does not receive the data of the curved array h or the received data is incomplete;

[0137] b) In the above technical solution of the embodiment of the present invention, the processing center receives abnormal parameters or signal data of the curved array h, for example, the parameters are out of the normal range or the signal data are all zero;

[0138] c) In the above technical solution of the embodiment of the present invention, the processing center receives abnormal node status of the curved array h, for example, the node status displays a radio frequency module fault code;

[0139] If any of the above three points is met, it means that the surface array h has been destroyed or damaged.

[0140] 8) In the above technical solution of the embodiment of the present invention, the processing center uses neural network fitting to generate the antenna array weight vector, which does not require complex operations such as matrix inversion, reduces the computational complexity, reduces the amount of computation, and reduces the demand for computing resources, thereby supporting real-time updating of the antenna array weight vector.

[0141] 9), In the above technical solution of the embodiment of the application, the processing center takes the coordinate positions of the curved surface arrays and the fiber delays as inputs of the neural network, and calculates and updates the antenna array weight vector in real time, so that the array manifold can dynamically adapt to the scene of real-time change, such as the scene of dynamic change of the position of the curved surface array.

[0142] 10), In the above technical solution of the embodiment of the application, the processing center takes the received array signal vector as an input of the neural network, and calculates and updates the antenna array weight vector in real time, so that the direction of the interference source in the received signal can dynamically adapt to the change, and the null point of the beam can be adjusted in real time to align the interference, thereby realizing dynamic tracking and suppression of the spatial domain interference.

[0143] In other preferable embodiments of the application, the following embodiment one is provided:

[0144] The application system architecture or scene of the embodiment one is that multiple curved surface arrays jointly form a desired beam, and involves a network element being a curved surface array and a processing center. In a first weighting stage, a CBF algorithm is used to generate an antenna array weight vector, and in a second weighting stage, a radial basis function neural network is used to approximate an MVDR algorithm to generate an antenna array weight vector. The system working process of the embodiment is specifically described as follows:

[0145] 1, First weighting stage:

[0146] (1) The processing center sends the angle of the desired beam to each curved surface array.

[0147] (2) Each curved surface array respectively performs first weighting on the internal antenna elements thereof, and respectively synthesizes a primary beam.

[0148] For the i-th curved surface array, P i antenna elements are activated according to the angle of the desired beam, a 3-dimensional rectangular coordinate system is established with the phase center of the curved surface array as the origin, and the coordinates of the p-th antenna element are represented as ξ p = [x p , y p , z p ] T , wherein p represents the antenna element number, p = 1, …, P i , x p represents the x-axis coordinate of the antenna element p, y p represents the y-axis coordinate of the antenna element p, z p represents the z-axis coordinate of the antenna element p, and [·] T represents the transpose of a matrix or a vector.

[0149] The i-th curved surface array calculates the 3-dimensional rectangular coordinates of the P i activated antenna elements according to the angle of the desired beam , and sends the calculated 3-dimensional rectangular coordinates to the processing center.​ Compute antenna array weight vector w according to CBF beamforming algorithm i :

[0150]

[0151] where, represents the array manifold vector of the i-th curved array in the direction of with dimensionality P i ×1.

[0152]

[0153] represents the desired beam direction, θ0is the horizontal azimuth angle, i.e., the angle with the X positive axis in the XY plane, is the vertical elevation angle, i.e., the angle with the XY plane; represents the unit vector of the desired beam direction:

[0154]

[0155] exp(·) represents the exponential function with base of the natural constant e, j represents the imaginary part of a complex number, π represents the circular constant, f represents the signal frequency, c represents the speed of light, ξ p = [x p , y p , z p ] T represents the 3D Cartesian coordinates of the p-th antenna element, p = 1, …, P i , P i represents the number of active antenna elements of the i-th curved array, sin represents the sine function, cos represents the cosine function, [·] T represents the transpose of a matrix or vector.

[0156] The i-th curved array synthesizes a primary beam using the weight vector w i . For example, the output signal of a synthesized receive beam is:

[0157]

[0158] where x i (k) represents the synthesized receive signal obtained by the i-th curved array in the k-th sampling, represents the array signal vector obtained by the i-th curved array in the k-th sampling, with dimensionality P i ×1; [·] H represents the conjugate transpose of a matrix or vector.

[0159] 2. Secondary weighting stage:

[0160] Initialization phase:

[0161] (3) The i-th curved array sends its position coordinate information, fiber delay τ i , received signal x i (k) and node state information to the processing center.

[0162] (4) The processing center determines the number of curved arrays N that need to be cooperatively beamformed according to the measurement and control link budget requirements or antenna gain requirements.

[0163] (5) The processing center determines whether the current curved array is normal according to the received parameters, received signals and node state information of each curved array, and selects N normal working curved arrays as the effective curved array set.

[0164] The processing center determines the criteria for determining that the curved array h is an abnormal node:

[0165] a) The processing center has not received the data of the curved array h or has received incomplete data;

[0166] b) The processing center receives abnormal parameters or signal data of the curved array h, such as parameters exceeding the normal range or signal data being all zeros;

[0167] c) The processing center receives abnormal node state of the curved array h, such as node state showing radio frequency module fault code;

[0168] If any of the above three points is met, it means that the curved array h has been damaged or broken, so the curved array h is discarded and not processed; otherwise, the curved array h is added to the effective curved array set.

[0169] (6) After the processing center receives the parameters and received signals of all N normal curved arrays, the processing center is designated as the coordinate origin and a custom 3D rectangular coordinate system is established. The processing center converts the position coordinates of each curved array into the custom 3D rectangular coordinate system coordinates, denoted as Ω i = [x i , y i , z i ] T , where i represents the curved array number, x i represents the x-axis coordinate of the curved array i, y i represents the y-axis coordinate of the curved array i, and z i represents the z-axis coordinate of the curved array i, and the superscript T represents transposition.

[0170] Training phase:

[0171] (7) The processing center determines the angle of the desired beam Calculate the antenna array weight vector w according to the traditional MVDR adaptive beamforming algorithm MVDR .

[0172] The weight vector of the MVDR beamformer is:

[0173]

[0174] Among them, R x Represents the covariance matrix of the received array signal vector, R x =E(x(k)x H (k)), with a dimension of N×N; x(k) represents the array signal vector obtained by the k-th sampling, with a dimension of N×1; [·] H represents the conjugate transpose of a matrix or vector, and E(·) represents the mathematical expectation; Represents the matrix R x Seek inverse; Indicates that the antenna array is Array of direction manifold vectors, of dimension N×1.

[0175]

[0176] represents the desired beam direction, θ0 is the horizontal azimuth, that is, the angle between the X axis and the XY plane, is the vertical pitch angle, that is, the angle with the XY plane; A unit vector representing the desired beam direction:

[0177]

[0178] exp(·) represents the exponential function with the natural constant e as the base, j represents the imaginary part of the complex number, π represents the circumference of the circle, f represents the signal frequency, c represents the speed of light, Ω i =[x i ,y i ,z i ] T represents the 3D rectangular coordinates of the i-th surface matrix, τ i represents the fiber delay from the i-th curved array to the processing center, i = 1, 2, ..., N, N represents the number of curved arrays, sin represents the sine function, cos represents the cosine function, [·] T Represents the transpose of a matrix or vector.

[0179] (8) The processing center will w MVDR As a training sample to train the RBF neural network, so that the output of the neural network is close to w MVDR .

[0180] The 3D rectangular coordinates of N curved surface arrays Ω, fiber delay τ, received array signal vector x(k), the angle of the desired beam The input neural network, which is subjected to multi-layer weighting and nonlinear processing, finally outputs the antenna array weight vector w NN .

[0181] (8-1) Input layer processing:

[0182] The input parameters and signals are combined into a large real vector as the input layer of the RBF neural network:

[0183]

[0184] wherein Input represents the vector of the input layer of the neural network, and the dimension is Qx1; Re(·) represents taking the real part, Im(·) represents taking the imaginary part, and Q=3N+N+2N+2=6N+2.

[0185] (8-2) Hidden layer processing:

[0186]

[0187] wherein represents the output value of the mth neuron node of the hidden layer, m=1,2,…,M, and M is the number of neuron nodes of the hidden layer;

[0188] W (1) represents the connection weighting matrix from the input layer to the hidden layer, and the dimension is QxM;

[0189] represents the Gaussian function center of the mth neuron node, and the dimension is Qx1;

[0190] b (1) represents the bias vector from the input layer to the hidden layer, and the dimension is Mx1;

[0191] a (1) represents the neuron node vector of the hidden layer, and the dimension is Mx1;

[0192] Ψ(·) represents the radial basis function, and in the embodiment, the Gaussian function is adopted, that is, Ψ(x)=exp(-x 2 );

[0193] ||x,y|| represents the distance between vectors x and y, and in the embodiment, the cosine distance is adopted, that is,

[0194]

[0195] (8-3) Output layer processing:

[0196] a (2) = W (2) a (1) + b (2) ;

[0197] wherein a (2) represents the node vector of the output layer, the training result of which is the antenna weighting vector, and the dimension is 2N x 1, N represents the number of curved arrays, and since the antenna weighting vector is a complex number, the real part and the imaginary part are separated;

[0198] W (2) represents the connection weighting matrix from the hidden layer to the output layer, and the dimension is 2N x M;

[0199] b (2) represents the bias vector from the hidden layer to the output layer, and the dimension is 2N x 1.

[0200] (8-4) Output antenna weighting vector

[0201] Finally, a (2) is converted into a complex vector to obtain the antenna weighting vector w:

[0202]

[0203] wherein w NN represents the antenna weighting vector, and the dimension is N x 1, and j represents the imaginary part of the complex number.

[0204] (9) The processing center calculates the training error vector e = w NN -w MVDR , and determines whether to end the training.

[0205] The 2-order norm ||e|| 2 of the error vector is calculated. wherein e i represents the i-th element in the error vector, and ∑(·) represents the summation function.

[0206] If the 2-order norm ||e|| 2 of the error vector is greater than the preset threshold Threshold1, the training error vector e is fed back to the neural network for correcting the parameters of the neural network, and the iteration training is continued by jumping to step (7); otherwise, if the 2-order norm ||e|| 2 of the error vector is less than or equal to the preset threshold Threshold1, the training is ended.

[0207] Normal working phase:

[0208] (10) The processing center continuously transmits the 3D rectangular coordinates Ω, fiber delay τ, receiving array signal vector x(k), and the angle of the desired beam of N curved arrays. Input to the trained neural network, after multi-layer neural network weighting and nonlinear processing, update the output antenna array weight vector w′ NN , the dimension is N×1.

[0209] (11) The processing center uses the updated antenna array weight vector w′ NN Perform weighted summation on the received array signal x(k) to obtain the composite signal y(k):

[0210]

[0211] Thus, the desired adaptive receiving beam is formed.

[0212] (12) The processing center uses the updated antenna array weight vector w′ NN The single-channel transmission signal s(k) is weighted and output to obtain the transmission array signal vector s(k):

[0213]

[0214] The N elements in s(k) are distributed to N curved arrays for transmission, thereby forming the desired adaptive transmission beam.

[0215] (13) During normal operation, if the processing center finds a new abnormal surface array h, the surface array h is removed from the valid surface array set, and then jumps to step (7) to retrain the neural network; otherwise, jumps to step (10) to continue updating parameters and iterating.

[0216] Example 1 uses a two-stage weighted approach to implement the beamforming of the curved array itself and the distributed collaborative beamforming between the curved arrays, thereby reducing the complexity of signal processing implementation of large-scale distributed array antennas. At the same time, a radial basis function-based neural network is used to approximate the MVDR algorithm to generate antenna array weight vectors. Compared with the traditional MVDR algorithm, the computational complexity and amount of computation are reduced, and real-time updating of the antenna array weight vector is supported. Dynamic adaptation to scenarios where the array manifold changes in real time is supported, dynamic tracking is supported, interference suppression in different directions is suppressed, and partial anti-destruction capabilities are possessed.

[0217] The technical solution of the first embodiment has the following advantages:

[0218] 1) In the technical solution of the first embodiment, the beamforming of the curved array itself and the distributed collaborative beamforming between the curved arrays are respectively implemented by a two-stage weighting method, thereby reducing the signal processing implementation complexity of the large-scale distributed array antenna.

[0219] 2) The technical solution of the first embodiment, in the first weighting stage, each curved array respectively generates an internal element weight vector by using the CBF algorithm, which reduces the implementation complexity of the internal element weight calculation of the curved array.

[0220] 3) The technical solution of the first embodiment, in the second weighting stage, the processing center generates the antenna array weight vector by using the neural network fitting, without complex operations such as matrix inversion, which reduces the operation complexity, reduces the operation amount, and reduces the demand for computing resources, thereby supporting real-time updating of the antenna array weight vector.

[0221] 4) The technical solution of the first embodiment, the processing center takes the coordinate positions of the curved arrays and the fiber delay as inputs of the neural network, and calculates and updates the antenna array weight vector in real time, which can dynamically adapt to the scene of real-time change of the array manifold, such as the scene of dynamic change of the curved array position.

[0222] 5) The technical solution of the first embodiment, the processing center takes the received array signal vector as an input of the neural network, and calculates and updates the antenna array weight vector in real time, which can dynamically adapt to the change of the interference source direction in the received signal, and adjust the null point of the beam to align the interference in real time, thereby realizing dynamic tracking and interference suppression.

[0223] 6) The technical solution of the first embodiment, each curved array sends its state information to the processing center, the processing center selects N normal working curved arrays as an effective curved array set, when part of the curved arrays are damaged, the processing center removes the invalid curved arrays from the effective curved array set, re-trains the neural network and updates the antenna array weight vector in time, and continues to maintain the expected beam pattern, which has a partial damage resistance.

[0224] The units described in the embodiments of the present application can be implemented in the form of software, or can be implemented in the form of hardware, and the described units can also be arranged in a processor. In some cases, the names of these units do not constitute a limitation on the units themselves.

[0225] According to an aspect of an embodiment of the present application, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method provided in the various optional implementation manners described above.

[0226] As another aspect, the embodiments of the present application also provide a computer readable medium, which can be included in the electronic device described in the above embodiments, or can exist independently without being assembled into the electronic device. The computer readable medium carries one or more programs, which, when executed by the electronic device, enable the electronic device to implement the method described in the above embodiments.

Claims

1. A distributed multi-curved array full-space collaborative beamforming method, characterized in that: The system comprises multiple distributed curved arrays and a processing center. The multiple distributed curved arrays are connected to the processing center to transmit parameters, control signals, received signals, and transmitted signal data. The processing center is used to control the multiple distributed curved arrays to complete their own beamforming and collaborative beamforming. Specifically, the system comprises the following steps: S1, each curved array performs first-level weighting on its internal antenna elements and synthesizes primary beams respectively; S2, the processing center performs secondary weighting on the signals from the distributed curved array to synthesize an equivalent large-aperture phased array antenna and form the desired adaptive beam; In step S1, each curved array performs primary weighting on its internal antenna elements and synthesizes primary beams, which specifically includes the following sub-steps: in the primary weighting stage, each curved array uses the CBF algorithm to generate its internal element weight vector; Each curved array uses the CBF algorithm to generate its own internal element weight vector, which specifically includes the following sub-steps: S1-1, the processing center will expect the beam angle Send to each surface array, is the horizontal azimuth, is the vertical pitch angle; S1-2, each curved array performs first-level weighting on its own internal antenna array elements and synthesizes primary beams respectively; A curved array is activated according to the angle of the desired beam antenna array elements, establish a three-dimensional rectangular coordinate system with the phase center of the curved array antenna as the origin, The coordinates of the antenna array elements are expressed as , Indicates the antenna element number, , Represents antenna array elements The x-axis coordinate, Represents antenna array elements The y-axis coordinate, Represents antenna array elements The z-axis coordinate of Represents the transpose of a matrix or vector; The surface array is based on The three-dimensional rectangular coordinates of the active antenna elements and the angle of the desired beam , calculate the antenna array weight vector according to the following formula ; ； in, , For input The linear function of weight vector Synthesize primary beam and output receive composite signal ; In step S2, the processing center performs secondary weighting on the signals of the distributed curved array, specifically including: in the secondary weighting stage, the processing center uses neural network fitting to generate weight vectors between the curved arrays, which can support real-time updating of the antenna array weight vectors; In step S2, the processing center uses neural network fitting to generate weight vectors between curved arrays, which can support real-time updating of antenna array weight vectors. Specifically, the processing center includes the following sub-steps: In the secondary weighting process, the The curved array will transmit its own position coordinate information, delay and receive signals Sent to the processing center, the delay Including fiber optic delay; The processing center receives the parameters, data and desired beam directions of multiple curved arrays. Combined into a real number vector and input to the input layer of the neural network, the neural network undergoes multi-layer weighting and nonlinear processing and outputs the antenna array weight vector in real time ; Reuse at processing center Perform weighted processing on the signals of multiple curved arrays.

2. The distributed multi-curved array full-space-domain collaborative beamforming method according to claim 1, characterized in that: In step S2, the processing center performs secondary weighting on the signals of the distributed curved array, including an initialization phase, in which the following sub-steps are specifically performed: S2-3, the processing center receives The position coordinate information and delay of the curved array , receive signal and node status information; S2-4, the processing center determines the number N of curved arrays required for collaborative beamforming based on the measurement and control link budget requirement or antenna gain requirement; S2-5, the processing center determines whether the current curved array is normal based on the received parameters of each curved array, the received signal and the node status information, and selects N curved arrays that are working normally as the valid curved array set; S2-6, after receiving the parameters and reception signals of all normal N curved arrays, the processing center arbitrarily designates a spatial reference point as the coordinate origin and establishes a custom three-dimensional rectangular coordinate system. The processing center converts the position coordinates of each curved array into the custom three-dimensional rectangular coordinate system coordinates, which are expressed as , represents the surface array number, Represents a surface array The x-axis coordinate, Represents a surface array The y-axis coordinate, Represents a surface array The z-axis coordinate of T Indicates transpose.

3. The distributed multi-curved array full-space-domain collaborative beamforming method according to claim 2, characterized in that: In step S2-5, the process of determining whether the current curved array is normal and selecting N curved arrays that are normally working as a valid curved array set includes the following sub-steps: If any of the following three conditions is met, it means that the surface matrix h has been destroyed or damaged, and the surface matrix h is discarded without further processing; otherwise, the surface matrix h is added to the valid surface matrix set; Case a): The processing center does not receive the data of the surface array h or the received data is incomplete; Case b): the processing center receives abnormal parameters or signal data of the curved array h; Case c), the processing center receives abnormal node status of the surface array h.

4. The distributed multi-curved array full-space collaborative beamforming method according to claim 1, characterized in that: In step S2, the processing center performs secondary weighting on the signals of the distributed curved array, including a training phase, in which the following sub-steps are specifically performed: S2-7, the processing center receives the three-dimensional rectangular coordinates of the N curved arrays , Delay , receive array signal vector , the angle of the desired beam Calculate the antenna array weight vector according to the adaptive beamforming algorithm ; ； in, Represents the three-dimensional rectangular coordinates of N surface arrays, Indicates the The three-dimensional rectangular coordinates of a curved array; Represents the delay vector from N curved arrays to the processing center, with a dimension of , Indicates the The delay from the surface array to the processing center, , Indicates the number of surface arrays; Represents the array signal vector obtained by the kth sampling, the dimension is ; represents the desired beam direction, is the horizontal azimuth, is the vertical pitch angle; For input Nonlinear function of S2-8, the processing center will 、 As training samples to train the neural network so that the neural network output is close to , and finally output the antenna array weight vector ; S2-9, the processing center calculates the training error vector , and determine whether to end training; calculate the 2nd order norm of the error vector , ,in, represents the i-th element in the error vector, represents the summation function; If the 2nd order norm of the error vector If it is greater than the preset threshold Threshold1, the training error vector Feedback to the neural network is used to modify the parameters of the neural network, and jump to step S2-7 to continue iterative training; on the contrary, if the 2nd order norm of the error vector If the value is less than or equal to the preset threshold Threshold1, the training ends.

5. The distributed multi-curved array full-space collaborative beamforming method according to claim 4, characterized in that: In step S2-8, the processing center 、 As training samples to train the neural network so that the neural network output is close to , and finally output the antenna array weight vector , specifically including the following sub-steps: S2-8-1, the three-dimensional rectangular coordinates of N curved arrays , Delay , receive array signal vector , the angle of the desired beam Enter the neural network; S2-8-2, the neural network undergoes multi-layer weighting and nonlinear processing, and finally outputs the antenna array weight vector , the dimension is .

6. The distributed multi-curved array full-space collaborative beamforming method according to claim 4, characterized in that: In step S2, the processing center performs secondary weighting on the signals of the distributed curved array, including a normal working stage, and specifically performs the following sub-steps in the normal working stage: S2-10, the processing center continues to convert the three-dimensional rectangular coordinates of N curved arrays , Delay , receive array signal vector , the angle of the desired beam Input to the trained neural network, after multi-layer neural network weighting and nonlinear processing, update the output antenna array weight vector , the dimension is ; S2-11, the processing center uses the updated antenna array weight vector Receive array signal Perform weighted summation to obtain the composite signal : ； Thus forming the desired adaptive receiving beam; S2-12, the processing center uses the updated antenna array weight vector For single-channel transmission signals Perform weighted output to obtain the transmit array signal vector : ； and will The N elements in are distributed to N curved arrays for transmission, thereby forming the desired adaptive transmit beam; S2-13, during normal operation, if the processing center finds a new abnormal surface array h, the surface array h will be removed from the valid surface array set, and then jump to step S2-7 to retrain the neural network; otherwise, jump to step S2-10 to continue updating parameters and iterating.

7. A distributed multi-curved array full-space collaborative beamforming system, comprising a computer device, characterized in that: A computer program is stored in the memory of the computer device, and when the computer program is loaded by the processor of the computer device, the method according to any one of claims 1 to 6 is executed.

Images