An adaptive beamforming method based on a spaceborne movable antenna

Abstract

This invention relates to the field of integrated air-space-ground information networks and satellite communication technology, specifically to an adaptive beamforming method based on a spaceborne movable antenna. The method includes: preprocessing broadcast automatic correlation surveillance data to obtain airspace traffic service data; establishing an end-to-end joint optimization neural network to receive the airspace traffic service data, sequentially performing feature extraction, antenna position optimization, and beam weight optimization, outputting the movable antenna position and beamforming vector; performing unsupervised joint training based on the airspace traffic service data and a composite loss function; obtaining the current user angle and communication service weights, inputting them into the trained end-to-end joint optimization neural network to obtain the current movable antenna position and beamforming vector; and controlling the movable antenna using the current movable antenna position and beamforming vector. This invention can improve the performance of antenna communication.

Description

Technical Field

[0001] This invention relates to the field of integrated air-space-ground information networks and satellite communication technology, specifically to an adaptive beamforming method based on a spaceborne movable antenna. Background Technology

[0002] In recent years, with the digital transformation of the global civil aviation industry, air traffic has experienced exponential growth. To ensure the safe and efficient flight of aircraft globally, building an integrated "air traffic system" encompassing communication, navigation, and surveillance (CNS) has become an inevitable trend in the industry. Satellite communication subsystems, with their inherent advantages of wide coverage and lack of geographical limitations, serve as the "main artery" connecting aircraft to the ground core network, and are a key link in achieving seamless all-weather, all-airspace interconnection.

[0003] However, analysis based on Automatic Dependent Surveillance-True Broadcast (ADS-B) data reveals that global aviation traffic exhibits strong heterogeneity and dynamism in its spatiotemporal distribution. Busy air corridors (such as the North Atlantic routes) and areas surrounding hub airports often form high-density "communication hotspots," while areas such as the ocean floor are sparsely populated. This uneven distribution of traffic poses a severe challenge to the resource scheduling capabilities of satellite communication systems.

[0004] Current airborne satellite communication systems primarily rely on fixed-position antenna (FPA) arrays and static resource allocation strategies. While this traditional approach is technologically mature, it still suffers from drawbacks when facing increasingly complex aerospace communication demands, including resource supply and demand mismatch, limited spatial freedom, poor real-time performance of optimization algorithms, and a lack of service awareness. Summary of the Invention

[0005] In view of the above problems, the present invention provides an adaptive beamforming method based on a spaceborne movable antenna, which solves the technical problems of insufficient efficiency of fixed-position antenna arrays and static resource allocation in the prior art.

[0006] This invention provides an adaptive beamforming method based on a spaceborne movable antenna, comprising the following steps: Step S1: Preprocess the broadcast automatic correlation surveillance data to obtain airspace traffic service data; the airspace traffic service data includes user perspective and communication service weights; Step S2: Establish an end-to-end joint optimization neural network to receive airspace traffic service data. The network sequentially goes through feature extraction, antenna position optimization, and beam weight optimization, and outputs the movable antenna position and beamforming vector. Step S3: Establish a composite loss function, and perform unsupervised joint training based on the airspace traffic service data and the composite loss function to obtain a trained end-to-end joint optimization neural network. Step S4: Real-time acquisition of the current user angle and communication service weights, input into the trained end-to-end joint optimization neural network, and acquisition of the current movable antenna position and beamforming vector; The movable antenna is controlled by the current movable antenna position and beamforming vector.

[0007] Preferably, step S1 specifically includes: Step S1-1: Perform data cleaning and spatiotemporal alignment on the ADS-B data; Step S1-2: Divide the time window and grid the airspace. Based on the spatiotemporally aligned ADS-B data, count the number of aircraft and traffic demand within the grid. Steps S1-3: Cluster the number of aircraft and communication traffic demand within the grid of the time window to obtain multiple user clusters, and obtain the user perspective and communication service weight for each user cluster.

[0008] Preferably, step S1-1 specifically includes: Collect historical ADS-B data of the target airspace and remove outliers, then convert all timestamps of the ADS-B data to UTC time. Steps S1-3 specifically include: using the K-Means clustering algorithm to cluster the number of aircraft and communication traffic requirements within the grid, dividing them into... A user cluster; The normalized density of the communication traffic demand of each user cluster is used as the communication service weight of the user cluster, and the expression is as follows:

[0009] in, Indicates the first Communication service weights for each user cluster Indicates the first Communication traffic requirements of individual user clusters Indicates the first Communication traffic requirements of individual user clusters This represents the total number of user clusters.

[0010] Preferably, in step S2, the end-to-end joint optimization neural network includes a feature extraction module, an antenna position optimization module, and a beam weight optimization module; The feature extraction module specifically employs a multilayer perceptron to extract vectors representing user perspectives and communication service weights. and The weighted concatenation is used as the input feature vector to the multilayer perceptron; the multilayer perceptron outputs a high-dimensional semantic feature vector. .

[0011] Preferably, in step S2, the antenna position optimization module will optimize the semantic feature vector. As input, the prediction is first obtained A non-negative spacing increment and total length scaling factor ;according to Calculate the total effective movement length and normalize all spacing increments; then generate the final antenna position through a cumulative summation operation, expressed as:

[0012] in, Indicates the first The location of each movable antenna element This indicates the minimum spacing constraint for antennas. This represents the i-th spacing increment; The beam weighting optimization module will use semantic feature vectors The amplitude and phase components of the beam are predicted separately. The predicted amplitude and phase components are combined to form an initial beam vector. The initial beam vector is then normalized to the L2 norm through a power normalization layer to obtain the beamforming vector.

[0013] Preferably, step S3 specifically includes: Step S3-1: Establish a physical channel model for the movable antenna. The physical channel model is used to calculate the amplitude and phase gain of the channel based on the antenna position vector. Step S3-2: Determine the overall system efficiency based on service weights and signal-to-interference-plus-noise ratio (SINR); the SINR is determined by the physical channel model. Step S3-3: Determine the sidelobe suppression penalty and fairness regularization term, and establish a composite loss function based on the total system effectiveness, sidelobe suppression penalty, and fairness regularization term; Step S3-4: Perform unsupervised joint training based on the airspace traffic service data and the composite loss function to obtain a trained end-to-end joint optimization neural network.

[0014] Preferably, in step S3-1, the expression for the physical channel model is:

[0015] in, Represents antenna position variables The corresponding number Channel vectors for each user cluster For the first Channel fading coefficients for each user cluster It is a natural constant. The imaginary unit, For carrier wavelength, They represent the 1st, The location of each movable antenna element Indicates the first From the perspective of individual user clusters; In step S3-2, the expression for the overall system efficiency is:

[0016] in, Indicates the overall system efficiency. Indicates the first Communication service weights for user clusters Indicates the first Signal-to-interference-plus-noise ratio of user clusters; In step S3-3, the composite loss function is the sum of the main utility loss, the sidelobe suppression penalty, and the fairness regularization term, wherein the main utility loss is set as the negative of the total system efficiency.

[0017] Preferably, step S4 specifically includes: Step S4-1: Receive the current ADS-B data stream in real time and generate the user perspective and communication service weight at the current moment; Step S4-2: Input the real-time acquired current airspace user angle and communication service weight into the trained end-to-end joint optimization neural network to obtain the optimal antenna position and beamforming vector; Step S4-3: Control the movable antenna system according to the optimal antenna position and beamforming vector.

[0018] Compared with the prior art, the present invention has at least the following beneficial effects: (1) Significantly improves communication capacity in hotspot areas: By introducing movable antenna technology, this method can physically adjust the antenna position to reconstruct the channel environment based on the "communication hotspots" identified by ADS-B data, and accurately focus signal energy on high service density areas. Compared with traditional fixed antennas, it significantly improves the signal-to-interference-plus-noise ratio and total system throughput in hotspot areas, and solves the "supply and demand mismatch" problem.

[0019] (2) Balancing the fairness of system services: By introducing business weights and fairness constraints (such as proportional fairness) into the optimization objectives, this method can prioritize hotspot areas while also taking into account the basic communication needs of non-hotspot areas. This effectively avoids sacrificing the user experience of edge users in pursuit of maximizing the total rate, and enhances the inclusiveness and stability of air traffic control system services.

[0020] (3) Real-time decision-making at the millisecond level: To address the pain point of high computational complexity and difficulty in real-time solution of mixed integer nonlinear programming (MINLP) problems, this invention adopts a deep learning-based "offline training + online inference" mode. Once the model training is completed, it can output the optimal antenna position and beam weight within milliseconds based on the real-time input of the business situation, meeting the stringent requirements of the air traffic communication system for high dynamic response. Attached Figure Description

[0021] The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of the invention.

[0022] Figure 1 The flowchart shows the adaptive beamforming method based on a spaceborne movable antenna provided by the present invention.

[0023] Figure 2 The overall flowchart of the adaptive beamforming method based on a spaceborne movable antenna provided by the present invention is shown below. Detailed Implementation

[0024] To better understand the above-described objectives, features, and advantages of the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other. Furthermore, the present invention can be implemented in other ways different from those described herein; therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0025] To illustrate the effectiveness of the method proposed in this invention, the following detailed description of the above technical solution is provided through a specific embodiment, such as... Figure 1 , Figure 2 As shown, this invention discloses an adaptive beamforming method based on a spaceborne movable antenna, and the specific implementation steps are as follows: Step S1: Preprocess the broadcast automatic correlation surveillance data to obtain airspace traffic service data; the airspace traffic service data includes user perspective and communication service weights; Automatic Dependent Surveillance-Broadcast (ADS-B) data is a type of digital status information automatically broadcast by aircraft (such as airplanes, helicopters, and drones) to ground stations and surrounding aircraft. This digital status information includes time-stamped data such as position, angle, operating parameters, and identification information.

[0026] This invention utilizes broadcast automatic dependent surveillance data to transform unstructured flight trajectories into structured airspace traffic service data, which serves as the basis for subsequent beamforming decision-making. The specific description is as follows.

[0027] Step S1-1: Perform data cleaning and spatiotemporal alignment on the ADS-B data; In this step, the present invention collects historical ADS-B data of the target airspace and removes outliers, such as abrupt changes in latitude and longitude, altitude anomalies, etc.; then, all timestamps of the ADS-B data are uniformly converted to UTC time. Using geofencing technology, flight data points located within the satellite service coverage area are selected.

[0028] Step S1-2: Divide the time window and grid the airspace. Based on the spatiotemporally aligned ADS-B data, count the number of aircraft and traffic demand within the grid. In this step, the invention divides the continuous time stream into fixed time windows, such as 15 minutes. Within each time window, the airspace is gridded, and the number of aircraft and instantaneous communication traffic demand within each grid are statistically analyzed. The communication traffic demand is used to reflect the aircraft's workload.

[0029] Steps S1-3: Cluster the number of aircraft and communication traffic demand within the grid of the time window to obtain multiple user clusters, and obtain the user perspective and communication service weight for each user cluster.

[0030] In this step, for each time window, the K-Means clustering algorithm is used to cluster the number of aircraft and communication traffic demand within the grid, dividing them into... There are 1 user cluster, and each user cluster represents a communication hotspot.

[0031] First, the user angle for each user cluster is determined. In some embodiments, the angles of each aircraft within a user cluster can be statistically analyzed, and the corresponding user angles can be obtained based on the statistical values.

[0032] The normalized density of communication traffic demands for all aircraft in the user cluster is calculated and used as the communication service weight for the user cluster. The expression is as follows:

[0033] in, Indicates the first Communication service weights for each user cluster Indicates the first Communication traffic requirements of individual user clusters Indicates the first Communication traffic requirements of individual user clusters This represents the total number of user clusters.

[0034] Communication service weight This directly reflects the urgency of the region's communication needs.

[0035] Performing the above processing on each time window yields spatial user perspective and communication service weight data with multiple time stamps and including multiple user clusters. This data serves as the spatial traffic service data, where each spatial user perspective and communication service weight data can be a vector composed of information from multiple user clusters. This provides input data for subsequent neural network optimization.

[0036] Step S2: Establish an end-to-end joint optimization neural network to receive airspace traffic service data. The network sequentially goes through feature extraction, antenna position optimization, and beam weight optimization, and outputs the movable antenna position and beamforming vector. In this step, the present invention constructs an end-to-end joint optimization neural network based on deep learning. This network can receive spatial traffic service data and sequentially pass through three modules: feature extraction, antenna position optimization, and beam weight optimization, ultimately outputting the movable antenna position and beamforming vector. The entire network architecture adopts an end-to-end design concept, enabling joint optimization throughout the entire process from input to output.

[0037] Feature extraction module The feature extraction module uses a multilayer perceptron (MLP) as the feature extractor. The input to the feature extraction module is the user perspective and communication service weights obtained in the preceding steps; specifically, it can be a user perspective vector. and business weight vector .

[0038] To enhance feature representation, an "angle-weight coupled coding" approach is adopted, that is... The sine / cosine values ​​and The weighted concatenation is used as the input feature vector to the multilayer perceptron, and the multilayer perceptron outputs a high-dimensional semantic feature vector. Semantic feature vector It contains in-depth feature information on airspace traffic distribution, providing a shared feature representation for the two subsequent optimization modules.

[0039] Antenna Position Optimizer (APV Optimizer) Used for prediction A movable antenna element in a given area The optimal physical location within the area. The antenna location optimization module receives the semantic feature vector. As input, the output antenna position vector ,in, They represent the 1st, 2nd, and 3rd respectively. The location of each movable antenna element This indicates transpose.

[0040] To strictly meet the minimum antenna spacing constraint Due to physical limitations, this invention employs an incremental accumulation strategy to generate antenna positions. The specific implementation process is as follows: The antenna position optimization module first predicts A non-negative spacing increment and a total length scaling factor According to the scaling factor The total effective movement length is calculated, and all spacing increments are normalized to ensure that the total length of the antenna array is within the allowable range. Finally, the final antenna position is generated through a cumulative summation operation, expressed as:

[0041] in, Indicates the first The location of each movable antenna element This indicates the minimum spacing constraint for antennas. This represents the i-th spacing increment.

[0042] The above steps ensure that the output antenna position naturally meets the physical constraints, eliminating the need for subsequent pruning.

[0043] Beam Weight Optimizer (AWV Optimizer) The beam weighting optimization module and the antenna position optimization module of this invention are arranged side by side, and both receive semantic feature vectors. As input, the output is a complex beamforming vector w.

[0044] The beamforming vector is in complex form, and the beam weighting optimization module predicts the amplitude and phase components of the beam, respectively.

[0045] The amplitude component is ensured to be non-negative through an activation function, while the phase component ranges from [0, 2π]. After the predicted amplitude and phase are combined to form an initial complex beam vector, it undergoes L2 norm normalization through a power normalization layer to ensure the beamforming vector meets the satellite's maximum transmit power constraint. The normalized beamforming vector w enables flexible beam direction control while also satisfying hardware power limitations.

[0046] Step S3: Establish a composite loss function, and perform unsupervised joint training based on the airspace traffic service data and the composite loss function to obtain a trained end-to-end joint optimization neural network. This step establishes a composite loss function based on the physical channel model and uses an unsupervised learning strategy to train the end-to-end joint optimization neural network, as detailed below.

[0047] Step S3-1: Establish a physical channel model for the movable antenna. The physical channel model is used to calculate the amplitude and phase gain of the channel based on the antenna position vector. Unlike traditional fixed antennas, this invention constructs an explicit antenna position variable... The channel vector is a complex number used to represent the amplitude and phase gain of the channel. For the ... A user cluster, whose channel vector Represented as:

[0048] in, Represents antenna position variables The corresponding number Channel vectors for each user cluster For the first Channel fading coefficients for each user cluster It is a natural constant. The imaginary unit, For carrier wavelength, They represent the 1st, The location of each movable antenna element Indicates the first From the perspective of individual user clusters.

[0049] Step S3-2: Determine the overall system efficiency based on service weights and signal-to-interference-plus-noise ratio (SINR); the SINR is determined by the physical channel model. In this step, based on the predicted antenna position x and beamforming vector w, combined with the channel vector... Calculate the received signal power of the k-th user cluster. The received signal of user cluster k contains the desired signal and interference signals from other users, and the signal-to-interference-plus-noise ratio (SIR) is calculated accordingly. .

[0050] The overall system performance of this invention is defined as the weighted sum rate, expressed as:

[0051] in, Indicates the overall system efficiency. Indicates the first Communication service weights for user clusters Indicates the first Signal-to-interference-plus-noise ratio of user clusters.

[0052] Step S3-3: Determine the sidelobe suppression penalty and fairness regularization term, and establish a composite loss function based on the total system effectiveness, sidelobe suppression penalty, and fairness regularization term; In this step, the present invention establishes a composite loss function to balance multiple optimization objectives, including communication capacity and beam quality. The composite loss function is the sum of the main utility loss, sidelobe suppression penalty, and fairness regularization term.

[0053] Wherein, the main utility loss is set as the negative value of the total system utility ( ), used to increase capacity.

[0054] Sidelobe suppression penalty is used to calculate the radiated power of the beam in the non-user direction. If it exceeds a preset threshold, a penalty is applied to reduce interference to nearby satellites or ground systems. Fairness regularization terms include proportional fairness logarithms, which can prevent the network from excessively ignoring low-weight users.

[0055] Step S3-4: Perform unsupervised joint training based on the airspace traffic service data and the composite loss function to obtain a trained end-to-end joint optimization neural network.

[0056] In this step, airspace traffic service data is used as training data. During the training process, there is no need to manually label the optimal labels. Instead, the optimization goals are to maximize the system weighted sum rate, suppress sidelobe interference, and ensure user fairness.

[0057] Using automatic differentiation technology, the gradient of the composite loss function with respect to the network parameters is calculated. Since both the antenna position generation process and the channel calculation process adopt differentiable design, the gradient signal can be directly fed back from the physical performance index to the various parameters of the neural network.

[0058] In some embodiments, the Adam optimizer is used to iteratively update the parameters, and the network weights of the feature extraction module, antenna position optimization module and beam weight optimization module are gradually adjusted through the backpropagation algorithm; the iteration continues until the loss function converges or reaches the preset number of training rounds, and finally the trained end-to-end joint optimization neural network model is obtained.

[0059] Step S4: Real-time acquisition of the current user angle and communication service weights, input into the trained end-to-end joint optimization neural network, and acquisition of the current movable antenna position and beamforming vector; The movable antenna is controlled by the current movable antenna position and beamforming vector.

[0060] After the model training is completed, the trained end-to-end joint optimization neural network is deployed at a satellite ground station or onboard processing unit for real-time online inference and resource reconfiguration control, as detailed below. Step S4-1: Receive the current ADS-B data stream in real time and generate the user perspective and communication service weight at the current moment; Step S4-2: Input the real-time acquired current airspace user angle and communication service weight into the trained end-to-end joint optimization neural network to obtain the optimal antenna position and beamforming vector; In this step, the network sequentially performs forward calculations through the feature extraction module, antenna position optimization module, and beam weight optimization module, directly outputting the current optimal movable antenna position x* and beamforming vector w*. The output result is a real-time optimization scheme based on the current airspace traffic distribution and service requirements.

[0061] Step S4-3: Control the movable antenna system according to the optimal antenna position and beamforming vector.

[0062] Based on the received control commands, the mechanical device moves the antenna elements to the designated position x*, achieving physical reconfiguration of the antenna array. Simultaneously, the digital beamforming system refreshes the electronic beam weights to w*, adjusting the signal amplitude and phase of each antenna element. Through the synergistic effect of the physical movement of the antenna positions and the electronic adjustment of the beam weights, adaptive coverage targeting current airspace hotspots is achieved, enabling precise beam pointing and power allocation in high-traffic areas. The system continuously cycles through the sensing, decision-making, and execution processes, adjusting the movable antenna configuration in real time according to dynamic changes in airspace traffic to maintain optimal communication service performance.

[0063] While the specific embodiments of the present invention depict actions or steps in a particular order, this should be understood as requiring such actions or steps to be performed in the shown specific order or sequential order, or requiring all illustrated actions or steps to be performed to achieve the desired result. In certain environments, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations. The above descriptions are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention.

[0064] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive beamforming method based on a spaceborne movable antenna, characterized in that, Includes the following steps: Step S1: Preprocess the broadcast automatic correlation surveillance data to obtain airspace traffic service data; the airspace traffic service data includes user perspective and communication service weights. Step S2: Establish an end-to-end joint optimization neural network to receive airspace traffic service data. The network sequentially goes through feature extraction, antenna position optimization, and beam weight optimization, and outputs the movable antenna position and beamforming vector. Step S3: Establish a composite loss function, and perform unsupervised joint training based on the airspace traffic service data and the composite loss function to obtain a trained end-to-end joint optimization neural network. Step S4: Real-time acquisition of the current user angle and communication service weights, input into the trained end-to-end joint optimization neural network, and acquisition of the current movable antenna position and beamforming vector; The movable antenna is controlled by the current movable antenna position and beamforming vector.

2. The adaptive beamforming method based on a spaceborne movable antenna according to claim 1, characterized in that, Step S1 specifically includes: Step S1-1: Perform data cleaning and spatiotemporal alignment on the ADS-B data; Step S1-2: Divide the time window and grid the airspace. Based on the spatiotemporally aligned ADS-B data, count the number of aircraft and traffic demand within the grid. Steps S1-3: Cluster the number of aircraft and communication traffic demand within the grid of the time window to obtain multiple user clusters, and obtain the user perspective and communication service weight for each user cluster.

3. The adaptive beamforming method based on a spaceborne movable antenna according to claim 2, characterized in that, Step S1-1 specifically includes: Collect historical ADS-B data of the target airspace and remove outliers, then convert all timestamps of the ADS-B data to UTC time. Steps S1-3 specifically include: using the K-Means clustering algorithm to cluster the number of aircraft and communication traffic requirements within the grid, dividing them into... A user cluster; The normalized density of the communication traffic demand of each user cluster is used as the communication service weight of the user cluster, and the expression is as follows: in, Indicates the first Communication service weights for each user cluster Indicates the first Communication traffic requirements of individual user clusters Indicates the first Communication traffic requirements of individual user clusters This represents the total number of user clusters.

4. The adaptive beamforming method based on a spaceborne movable antenna according to claim 3, characterized in that, In step S2, the end-to-end joint optimization neural network includes a feature extraction module, an antenna position optimization module, and a beam weight optimization module; The feature extraction module specifically employs a multilayer perceptron to extract vectors representing user perspectives and communication service weights. and The weighted concatenation is used as the input feature vector to the multilayer perceptron; the multilayer perceptron outputs a high-dimensional semantic feature vector. .

5. The adaptive beamforming method based on a spaceborne movable antenna according to claim 4, characterized in that, In step S2, the antenna position optimization module will optimize the semantic feature vector. As input, the prediction is first obtained A non-negative spacing increment and total length scaling factor ;according to Calculate the total effective movement length and normalize all spacing increments; then generate the final antenna position through a cumulative summation operation, expressed as: in, Indicates the first The location of each movable antenna element This indicates the minimum spacing constraint for antennas. This represents the i-th spacing increment; The beam weighting optimization module will use semantic feature vectors The amplitude and phase components of the beam are predicted separately. The predicted amplitude and phase components are combined to form an initial beam vector. The initial beam vector is then normalized to the L2 norm through a power normalization layer to obtain the beamforming vector.

6. The adaptive beamforming method based on a spaceborne movable antenna according to claim 5, characterized in that, Step S3 specifically includes: Step S3-1: Establish a physical channel model for the movable antenna. The physical channel model is used to calculate the amplitude and phase gain of the channel based on the antenna position vector. Step S3-2: Determine the overall system efficiency based on service weights and signal-to-interference-plus-noise ratio (SINR); the SINR is determined by the physical channel model. Step S3-3: Determine the sidelobe suppression penalty and fairness regularization term, and establish a composite loss function based on the total system effectiveness, sidelobe suppression penalty, and fairness regularization term; Step S3-4: Perform unsupervised joint training based on the airspace traffic service data and the composite loss function to obtain a trained end-to-end joint optimization neural network.

7. The adaptive beamforming method based on a spaceborne movable antenna according to claim 6, characterized in that, In step S3-1, the expression for the physical channel model is: in, Represents antenna position variables The corresponding number Channel vectors for each user cluster For the first Channel fading coefficients for each user cluster It is a natural constant. The imaginary unit, For carrier wavelength, They represent the 1st, The location of each movable antenna element Indicates the first From the perspective of individual user clusters; In step S3-2, the expression for the overall system efficiency is: in, Indicates the overall system efficiency. Indicates the first Communication service weights for user clusters Indicates the first Signal-to-interference-plus-noise ratio of user clusters; In step S3-3, the composite loss function is the sum of the main utility loss, the sidelobe suppression penalty, and the fairness regularization term, wherein the main utility loss is set as the negative of the total system efficiency.

8. The adaptive beamforming method based on a spaceborne movable antenna according to claim 7, characterized in that, Step S4 specifically includes: Step S4-1: Receive the current ADS-B data stream in real time and generate the user perspective and communication service weight at the current moment; Step S4-2: Input the real-time acquired current airspace user angle and communication service weight into the trained end-to-end joint optimization neural network to obtain the optimal antenna position and beamforming vector; Step S4-3: Control the movable antenna system according to the optimal antenna position and beamforming vector.

Images