Method for realizing data communication by multi-band adaptive antenna based on 5G communication

Abstract

The invention discloses a method for realizing data communication by a multi-band adaptive antenna based on 5G communication. The method comprises the steps of dynamic sensing and multi-mode signal fusion, multi-band intelligent analysis, hybrid beam forming and dynamic reconstruction, cross-layer parameter joint tuning, real-time monitoring and self-learning compensation, multi-band adaptive switching and fault diagnosis and error calculation. The problems that in traditional 5G communication, the antenna frequency band is fixed, the beam adjusting capacity is limited, the anti-interference performance is poor, and the communication performance is unstable in a complex environment are solved. According to the method, the channel state is accurately evaluated and predicted through fusion of the extended Kalman filtering algorithm and the convolutional neural network, and frequency band resources are intelligently allocated and optimized through the deep Q network DQN and the non-orthogonal multiple access NOMA principle, so that omnibearing optimization and management of the antenna are realized, the communication efficiency and stability are improved, and the method is suitable for large-scale popularization and application. And the efficient and stable operation of the communication system in various complex environments is ensured.

Description

Technical Field

[0001] The present invention relates to the field of 5G communication technology, and more specifically to a method for implementing data communication based on a multi-band adaptive antenna for 5G communication. Background Art

[0002] With the continuous advancement of 5G communication technology and related antenna technologies, the application of multi-band adaptive antennas in the communications field has gradually become popular, the scale of construction has continued to expand, and the scope of application has also been increasingly broadened. However, multi-band adaptive antennas lack the efficiency, stability, and resource utilization of data communications, and the implementation of the corresponding technology is also not perfect.

[0003] Under current technology, multi-band adaptive antennas have developed certain evaluation methods for frequency band switching and beamforming, which have, to a certain extent, alleviated the problems of degraded communication quality and limited network capacity caused by limited spectrum resources and severe electromagnetic interference. However, they still face challenges such as high frequency band switching latency, lagging channel status monitoring, and suboptimal beamforming effectiveness.

[0004] To effectively address these issues, a comprehensive and robust multi-band adaptive antenna evaluation and optimization system is needed. By analyzing key factors such as antenna performance, spectrum utilization, and anti-interference capabilities, combined with core modules such as channel state perception and prediction and user behavior analysis, the system can adjust and optimize the operating status of multi-band adaptive antennas, thereby addressing the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies in data communication efficiency and stability of multi-band adaptive antennas, this invention discloses a method for implementing data communication using a multi-band adaptive antenna based on 5G communications. Spectrum resource allocation is evaluated using a band adaptive selection algorithm and a deep Q-network (DQN). Channel status is assessed by fusing an extended Kalman filter algorithm with a convolutional neural network. A control module then feeds the evaluation results back to the multi-band adaptive antenna and executes corresponding operations, ensuring safe and stable operation of the multi-band adaptive antenna and improving communication efficiency and stability.

[0006] The present invention adopts the following technical solutions: A method for implementing data communication based on a multi-band adaptive antenna for 5G communication, comprising the following steps: Step 1: Dynamic perception and multimodal signal fusion; Through a multimodal sensor network, the system collects terminal motion trajectories, spatial obstacle distribution, and electromagnetic interference maps in real time. It then generates a three-dimensional channel state matrix based on the fusion of the extended Kalman filter algorithm and a convolutional neural network to predict the terminal's movement path within the next 0.5 seconds. Step 2: Multi-band intelligent analysis; Adopting a frequency band adaptive selection algorithm, integrated with the Deep Q Network (DQN) to jointly evaluate the current network load, user service type, and channel quality, it dynamically allocates 3300-3600MHz, 4800-5000MHz, and millimeter wave frequency band resources. Based on the principle of non-orthogonal multiple access (NOMA), it achieves a 40% increase in spectrum utilization. Step 3: Hybrid beamforming and dynamic reconstruction; Through the collaborative design of Luneburg lens medium and electromagnetic metamaterial array, multi-beam independent electrical adjustment signals are generated. The improved genetic algorithm is used to optimize the beam width and pointing angle to achieve 3D beam coverage in the horizontal and vertical dimensions. The beam shape is dynamically reconstructed based on the CSI information fed back by the terminal, which improves the sidelobe suppression capability by 30%. Step 4: Jointly tune cross-layer parameters; The antenna tilt, power allocation, and MIMO layer number parameters are mapped into a multi-objective optimization problem. A dynamic strategy model is constructed using the deep reinforcement learning (DRL) framework. By combining offline training with online transfer learning, the interference suppression ratio (ICIC) is optimized to over 18 dB while reducing energy consumption by 15%. Step 5: Real-time monitoring and self-learning compensation; A self-learning compensation module is introduced to aggregate user QoE data from edge nodes. The module analyzes and dynamically adjusts antenna parameters, enabling rapid filling of spectrum holes in response to sudden electromagnetic interference to ensure communication continuity. Step 6: Multi-band adaptive switching Based on the user behavior prediction model, when it is predicted that the terminal is about to enter a weak coverage area, the frequency band switching protocol is triggered, and carrier aggregation technology is used to achieve seamless switching between the 3300MHz low frequency band and the millimeter wave high frequency band, with a switching delay of less than 10ms; Step 7: Fault diagnosis and error calculation; An antenna health assessment model is constructed through an improved Bayesian network model to perform real-time diagnosis of PIM intermodulation distortion and feed point impedance mismatch abnormal conditions, and the matrix dipole array is enabled to reconstruct the radiation pattern to ensure system availability ≥ 99.999%.

[0007] As a further technical solution of the present invention, the working method of the dynamic perception and multimodal signal fusion is: The inertial measurement unit (IMU) deployed on the terminal collects three-axis acceleration and angular velocity data to construct a motion trajectory matrix. LiDAR and millimeter-wave radar are used to obtain the three-dimensional coordinates of spatial obstacles. The electromagnetic spectrum sensor scans the interference power spectrum density in the 24-40GHz frequency band in real time to form a multi-dimensional original data set. A terminal motion state space model is established, and the denoised trajectory data, obstacle point cloud, and interference map are converted into a three-dimensional grid tensor. The tensor is input into a 12-layer deep CNN network. Spatial feature associations are extracted through void convolution and attention mechanism, and a channel state matrix containing path loss and multipath delay is output. The time series characteristics of the channel state matrix are modeled based on the long short-term memory network (LSTM). Combined with the terminal's historical movement speed and direction angle, a particle filter algorithm is used to predict the three-dimensional movement trajectory coordinates within the next 0.5 seconds, with a prediction root mean square error of less than 0.3 meters.

[0008] As a further technical solution of the present invention, the steps for implementing the frequency band adaptive selection algorithm are as follows: 1) Multi-dimensional state feature extraction; Based on the channel state matrix, the network load, service type and channel quality are characterized and the feature fusion formula is: In formula (1), S represents the comprehensive score of frequency band selection; Indicates the normalized network load factor, with a value range of [0,1]; Indicates the channel interference power; Indicates the ratio of the current signal-to-noise ratio to the target threshold; Indicates the service latency requirement. is the baseline delay; are the weight coefficients of load and latency, respectively, which are dynamically adjusted through DQN; 2) Dynamic Q-value modeling and action space construction; Define the state space S is a vector set containing multi-dimensional features of network load, service type and channel quality, and the action space A Covering resource allocation strategies for 3300-3600MHz, 4800-5000MHz and millimeter wave bands, including band bandwidth allocation ratio and power control parameters, building a deep Q network model with the network input being state space S The corresponding feature vector is mapped to the action value function through the multi-layer perceptron structure; 3) Frequency band allocation and NOMA power control; According to the action strategy output by DQN, the primary and auxiliary frequency bands are allocated according to priority; differential power allocation is used for NOMA users: In formula (2), is the transmission power of the kth user, K is the number of multiplexed users, is the fairness adjustment factor, is the maximum transmit power, represents the signal-to-interference-and-noise ratio adjustment coefficient, For the k Signal-to-interference-noise ratio for each user; 4) Online strategy optimization and feedback; A lightweight DQN model is deployed in the base station edge computing unit to collect channel quality indicator (CQI) and bit error rate (BER) data in real time; network weights are updated using a double-delayed deep deterministic policy gradient (TD3) algorithm. As a further technical solution of the present invention, the improved genetic algorithm is implemented as follows: In formula (3), G represents the individual fitness score; is the beam width, is the theoretical optimal pointing angle; is the weight coefficient, and its initial value is set to 0.5; Indicates the sidelobe suppression level, which is the preset maximum sidelobe suppression level; MLG is the main lobe gain, Preset maximum main lobe gain; is the weighting coefficient; 2) Dynamic crossover and mutation probability adjustment According to the number of iterations t and population diversity Dynamically adjust the crossover probability Pc and mutation probability Pm: In formula (4), T is the preset maximum number of iterations; when When , Pm is greatly increased to 0.3 to prevent the population from falling into the local optimum; when When Pm is set to 0, the convergence speed is accelerated; 3) Multi-objective elite retention and reconstruction Use non-dominated sorting to screen the Pareto optimal solution set, decode the beam parameters of the elite individuals, and output the optimal solution: In formula (5), The terminal movement direction angle predicted in step 1; the main lobe gain MLG and side lobe peak are calculated in real time through the electromagnetic simulation model; 4) closed-loop feedback optimization; Send the optimal beam parameters to the Luneburg lens controller to collect the actual sidelobe suppression ratio If the terminal receives the signal strength RSSI, , then restart the population evolution process. As a further technical solution of the present invention, the working method of cross-layer parameter joint tuning is: First, an optimization variable vector is defined, consisting of three key parameters: antenna tilt, transmit power in each frequency band, and the number of MIMO layers. The antenna tilt range is set to 0°-15°, the transmit power range is 0-46dBm, and the number of MIMO layers can be selected from 1 to 8. Next, a multi-objective optimization model is constructed around the interference suppression ratio, transmit power, and throughput targets. The optimization goal is to minimize the sum of the inverse of the interference suppression ratio and the normalized transmit power, while ensuring that the throughput does not fall below a set threshold. In this process, a deep reinforcement learning framework is adopted, taking the base station MAC layer throughput, physical layer signal-to-interference-and-noise ratio, and RRC layer connection status as state inputs, and outputting the optimal parameter combination through a competitive network architecture. Finally, a layered training strategy is implemented. In the offline phase, policy gradient training is performed based on large-scale channel simulation data. In the online phase, transfer learning is used to quickly adapt to the real-time network environment. Parameters are updated every 50ms to achieve a stable interference suppression ratio above 18dB while reducing transmit power by 15%-20%, achieving a balanced optimization between performance and energy consumption.

[0009] As a further technical solution of the present invention, the working method of the self-learning compensation is: Edge nodes periodically upload user experience quality of experience (QoE) data, including throughput, latency, and packet loss rate characteristics. Secure multi-party computing (MPC) technology is used to protect data privacy. The global model aggregates and locally updates parameters every 10 seconds. A bidirectional LSTM model with 128 hidden units is constructed. The antenna parameters and QoE data of the previous 5 minutes are input, and the parameter correction value for the next minute is output as Δ=[Δ θ ,Δ ϕ ,Δ P ], the prediction mean square error is less than 5%; when sudden electromagnetic interference is detected and the interference power is greater than -80dBm, the spectrum sensing module is triggered to scan the idle sub-bands within the 500MHz bandwidth, and the available spectrum holes are identified through fast Fourier transform (FFT). The subcarrier allocation and beam direction adjustment are completed within 2ms to ensure the continuity of the communication link.

[0010] As a further technical solution of the present invention, the multi-band adaptive switching is implemented by the following steps: 1) User behavior prediction model; Based on historical mobile trajectories and service types, a support vector machine (SVM) is used to build a prediction model for entering a weak coverage area. Input parameters include terminal movement speed, direction angle, and RSRP value of the current frequency band. The prediction accuracy is ≥95%; 2) Switching trigger criteria; When the terminal's received power in the 3300MHz band falls below -105dBm and it is predicted to enter a millimeter wave coverage blind spot, the frequency band switching protocol is initiated and a pre-configuration request is sent to the neighboring base station via the Xn interface. 3) Carrier aggregation seamless switching; A hard handover mechanism based on measurement reports is used to establish a millimeter wave link in advance in the target frequency band. The channel state is estimated using the uplink synchronization signal (SRS). An inverse fast Fourier transform (IFFT) is used to generate a multi-band aggregate signal, enabling parallel transmission of the 3300MHz low-frequency band and the 28GHz high-frequency band. The handover delay is controlled to less than 10ms using time advance compensation (TA). As a further technical solution of the present invention, the fault diagnosis and error calculation method is as follows: An improved Bayesian network model containing 12 nodes was constructed. The input parameters included intermodulation distortion (PIM) ≥ -90dBc, feed point impedance deviation > 5%, and number of array element failures > 3 abnormal features. The antenna health level was output as 0-1, and the diagnostic accuracy was ≥ 98%. The reflection coefficient S11, standing wave ratio (VSWR), and temperature data of the antenna array were collected every 20ms. Gradual faults were identified using the hidden Markov model (HMM), and the sudden fault detection delay was < 5ms. When more than three adjacent array elements were detected to have failed, the preset 15% matrix oscillator array was activated, and the phase of the radiating element was dynamically adjusted using electromagnetic metamaterials. The beamforming vector w was reconstructed based on the least squares method to ensure that the gain loss after reconstruction was ≤ 1.5dB. The system availability was guaranteed to be ≥ 99.999% through the N+1 redundancy design.

[0011] In a further embodiment, the improved Bayesian network model comprises a three-layer node structure, a 6-node bottom observation layer, a 3-node intermediate feature layer, and a 3-node top decision layer. The improved Bayesian network model comprises an online parameter update algorithm, and the function is: In formula (6), is the dynamic weight factor, is the posterior probability of failure given the evidence, is a normalization constant. Compared with the prior art, the present invention has the following positive effects: The present invention uses a frequency band adaptive selection algorithm and a deep Q-network (DQN) to evaluate spectrum resource allocation, and uses an extended Kalman filter algorithm fused with a convolutional neural network to evaluate channel status. It also optimizes the configuration of the parts related to beamforming and antenna parameters. The control module feeds back the evaluation results to the multi-band adaptive antenna system and prompts the execution of related operations, thereby ensuring that the multi-band adaptive antenna can operate stably and efficiently. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which: Figure 1 This is a flow chart of a method for realizing data communication based on a multi-band adaptive antenna for 5G communication according to the present invention; Figure 2 This is a flow chart of a frequency band adaptive selection algorithm implementation method for realizing data communication based on a multi-band adaptive antenna for 5G communication according to the present invention; Figure 3 This is a flowchart of an improved genetic algorithm for implementing a method for realizing data communication using a multi-band adaptive antenna based on 5G communication according to the present invention; Figure 4 This is a flowchart of a method for realizing data communication using a multi-band adaptive antenna based on 5G communication according to the present invention, which shows multi-band adaptive switching implementation. DETAILED DESCRIPTION

[0013] The following, combined with the accompanying drawings, provides a clear and complete description of the technical solutions in the embodiments herein. Obviously, the described embodiments represent only a portion of the embodiments herein, and not all of them. It should be understood that these descriptions are merely illustrative and are not intended to limit the scope of the present invention. Furthermore, descriptions of well-known structures and technologies are omitted in this description to avoid unnecessary confusion regarding the concepts of the present invention.

[0014] like Figure 1-4 As shown, a method for implementing data communication based on a multi-band adaptive antenna for 5G communication includes the following steps: Step 1: Dynamic perception and multimodal signal fusion; Through a multimodal sensor network, the system collects terminal motion trajectories, spatial obstacle distribution, and electromagnetic interference maps in real time. It then generates a three-dimensional channel state matrix based on the fusion of the extended Kalman filter algorithm and a convolutional neural network to predict the terminal's movement path within the next 0.5 seconds. Step 2: Multi-band intelligent analysis; Adopting a frequency band adaptive selection algorithm, integrated with the Deep Q Network (DQN) to jointly evaluate the current network load, user service type, and channel quality, it dynamically allocates 3300-3600MHz, 4800-5000MHz, and millimeter wave frequency band resources. Based on the principle of non-orthogonal multiple access (NOMA), it achieves a 40% increase in spectrum utilization. Step 3: Hybrid beamforming and dynamic reconstruction; Through the collaborative design of Luneburg lens medium and electromagnetic metamaterial array, multi-beam independent electrical adjustment signals are generated. The improved genetic algorithm is used to optimize the beam width and pointing angle to achieve 3D beam coverage in the horizontal and vertical dimensions. The beam shape is dynamically reconstructed based on the CSI information fed back by the terminal, which improves the sidelobe suppression capability by 30%. Step 4: Jointly tune cross-layer parameters; The antenna tilt, power allocation, and MIMO layer number parameters are mapped into a multi-objective optimization problem. A dynamic strategy model is constructed using the deep reinforcement learning (DRL) framework. By combining offline training with online transfer learning, the interference suppression ratio (ICIC) is optimized to over 18 dB while reducing energy consumption by 15%. Step 5: Real-time monitoring and self-learning compensation; A self-learning compensation module is introduced to aggregate user QoE data from edge nodes. The module analyzes and dynamically adjusts antenna parameters, enabling rapid filling of spectrum holes in response to sudden electromagnetic interference to ensure communication continuity. Step 6: Multi-band adaptive switching Based on the user behavior prediction model, when it is predicted that the terminal is about to enter a weak coverage area, the frequency band switching protocol is triggered, and carrier aggregation technology is used to achieve seamless switching between the 3300MHz low frequency band and the millimeter wave high frequency band, with a switching delay of less than 10ms; Step 7: Fault diagnosis and error calculation; An antenna health assessment model is constructed through an improved Bayesian network model to perform real-time diagnosis of PIM intermodulation distortion and feed point impedance mismatch abnormal conditions, and the matrix dipole array is enabled to reconstruct the radiation pattern to ensure system availability ≥ 99.999%.

[0015] Furthermore, the working method of the dynamic perception and multimodal signal fusion is: The inertial measurement unit (IMU) deployed on the terminal collects three-axis acceleration and angular velocity data to construct a motion trajectory matrix. LiDAR and millimeter-wave radar are used to obtain the three-dimensional coordinates of spatial obstacles. The electromagnetic spectrum sensor scans the interference power spectrum density in the 24-40GHz frequency band in real time to form a multi-dimensional original data set. A terminal motion state space model is established, and the denoised trajectory data, obstacle point cloud, and interference map are converted into a three-dimensional grid tensor. The tensor is input into a 12-layer deep CNN network. Spatial feature associations are extracted through void convolution and attention mechanism, and a channel state matrix containing path loss and multipath delay is output. The time series characteristics of the channel state matrix are modeled based on the long short-term memory network (LSTM). Combined with the terminal's historical movement speed and direction angle, a particle filter algorithm is used to predict the three-dimensional movement trajectory coordinates within the next 0.5 seconds, with a prediction root mean square error of less than 0.3 meters.

[0016] The specific implementation examples include the key steps of hardware deployment and data collection, data fusion and feature extraction, and time series modeling and trajectory prediction. During the hardware deployment and data collection phase, it is necessary to integrate inertial measurement units (IMUs), LiDARs, millimeter-wave radars, and electromagnetic spectrum sensors on mobile terminals such as drones, autonomous vehicles, and intelligent robots. Ensure that the sensors are properly installed and calibrated to ensure the accuracy and reliability of data collection. The IMU collects three-axis acceleration and angular velocity data at a high frequency, and after removing noise and deviation through data filtering and compensation algorithms, constructs an accurate motion trajectory matrix. The LiDAR and millimeter-wave radar collaboratively scan the surrounding environment to obtain the three-dimensional coordinate information of spatial obstacles. The collected point cloud data is subjected to noise reduction and data format conversion for subsequent processing. The electromagnetic spectrum sensor scans the 24-40GHz frequency band in real time, monitors the interference power spectrum density, and normalizes and extracts features from the collected spectrum data to form an interference map. During data fusion and feature extraction, a spatial model describing the terminal's position, velocity, acceleration, and posture state is established based on the terminal's dynamics and kinematic equations. The preprocessed motion trajectory data, obstacle point cloud data, and interference map data are converted into a three-dimensional grid tensor, which can be fused and analyzed in a unified spatial model. The fused three-dimensional grid tensor data is then input into a pre-trained 12-layer deep CNN network. The network's receptive field is expanded through dilated convolution to capture broader contextual information. The attention mechanism is used to highlight and extract spatial feature associations related to the channel state. The network outputs a channel state matrix containing path loss and multipath delay information, providing a basis for subsequent time series modeling. In the time series modeling and trajectory prediction stage, the time series characteristics of the channel state matrix are modeled based on the long short-term memory network LSTM, fully considering the dynamic change characteristics of the channel state, and combining the terminal's historical movement speed and angular information. These time series data are input into the LSTM network for joint modeling to learn the complex time series relationship between terminal movement and channel state. Then, the particle filter algorithm is used to predict the three-dimensional movement trajectory coordinates of the terminal within the next 0.5 seconds based on the output results of the LSTM model. By generating a large number of particle samples in the state space, the particles are weighted and resampled according to the prediction model and observation data to obtain the predicted trajectory coordinates. After continuous optimization and adjustment of the prediction error, the root mean square error is controlled within a range of less than 0.3 meters, ensuring that the accuracy of the prediction results can meet the actual application scenarios of precise flight control of drones, safe obstacle avoidance of autonomous vehicles, and path planning of intelligent robots. Furthermore, the implementation steps of the frequency band adaptive selection algorithm are as follows: In formula (1), S represents the comprehensive score of frequency band selection; Indicates the normalized network load factor, with a value range of [0,1]; Indicates the channel interference power; Indicates the ratio of the current signal-to-noise ratio to the target threshold; Indicates the service latency requirement. is the baseline delay; are the weight coefficients of load and latency, respectively, which are dynamically adjusted through DQN; 2) Dynamic Q-value modeling and action space construction; Define the state space S is a vector set containing multi-dimensional features of network load, service type and channel quality, and the action space A Covering resource allocation strategies for 3300-3600MHz, 4800-5000MHz and millimeter wave bands, including band bandwidth allocation ratio and power control parameters, building a deep Q network model with the network input being state space S The corresponding feature vector is mapped to the action value function through the multi-layer perceptron structure; 3) Frequency band allocation and NOMA power control; According to the action strategy output by DQN, the primary and auxiliary frequency bands are allocated according to priority; differential power allocation is used for NOMA users: In formula (2), is the transmission power of the kth user, K is the number of multiplexed users, is the fairness adjustment factor, is the maximum transmit power, represents the signal-to-interference-and-noise ratio adjustment coefficient, For the k Signal-to-interference-noise ratio for each user; 4) Online strategy optimization and feedback; Deploy a lightweight DQN model in the edge computing unit of the base station to collect channel quality indication CQI and bit error rate BER data in real time; update the network weight through the dual-delay deep deterministic policy gradient TD3 algorithm. In a specific implementation instance, the frequency band adaptive selection algorithm can accurately select the appropriate frequency band by analyzing the multi-dimensional state characteristics of network load, service type and channel quality, providing a more efficient and stable communication environment for the wireless communication system. This method considers frequency band selection not only based on the current network status, but also dynamically adjusts according to business needs and channel changes, which provides strong technical support for communication service providers and helps them optimize resource allocation and improve service quality. The frequency band adaptive selection algorithm first extracts multi-dimensional state features, decomposes the network load, service type and channel quality based on the channel state matrix, and then performs feature fusion according to the feature fusion formula. Calculate the comprehensive score of frequency band selection and dynamically adjust it through DQN. Then perform dynamic Q value modeling and action space construction to define the state space S is a vector set containing multi-dimensional features of network load, service type and channel quality, and the action space ACovering resource allocation strategies for 3300-3600MHz, 4800-5000MHz and millimeter wave bands, including band bandwidth allocation ratio and power control parameters, building a deep Q network model with the network input being state space S The corresponding feature vector is mapped to the action value function through the multi-layer perceptron structure. Then the frequency band allocation and NOMA power control are performed. According to the action strategy output by DQN, the main frequency band and auxiliary frequency band are allocated according to priority; differential power allocation is used for NOMA users according to the formula Calculate the k The transmit power of each user is calculated, and finally online strategy optimization and feedback are performed. A lightweight DQN model is deployed in the edge computing unit of the base station to collect channel quality indication CQI and bit error rate BER data in real time. The network weights are updated through the double-delay deep deterministic policy gradient TD3 algorithm. The calculated data is put into the actual operation of the communication system, and the algorithm parameters are modified based on the feedback results. Table 1 shows the data comparison using the frequency band adaptive selection algorithm. Table 1 shows that the load and delay weighting coefficients in the adaptive frequency band selection algorithm influence channel utilization. Appropriate weighting coefficients can more effectively balance network load and service latency requirements, optimizing resource allocation. By monitoring network status and channel quality in real time, operators can obtain detailed information about the current state of the communication system, including network load, service suitability, and channel interference. This precise monitoring not only identifies potential issues promptly but also provides clear guidance to the network optimization team, enabling the development of targeted optimization strategies.

[0017] Furthermore, the improved genetic algorithm is implemented as follows: In formula (3), G represents the individual fitness score; is the beam width, is the theoretical optimal pointing angle; is the weight coefficient, and its initial value is set to 0.5; Indicates the sidelobe suppression level, which is the preset maximum sidelobe suppression level; MLG is the main lobe gain, Preset maximum main lobe gain; is the weighting coefficient; 2) Dynamic crossover and mutation probability adjustment According to the number of iterations t and population diversity Dynamically adjust the crossover probability Pc and mutation probability Pm: In formula (4), T is the preset maximum number of iterations; when When , Pm is greatly increased to 0.3 to prevent the population from falling into the local optimum; when When Pm is set to 0, the convergence speed is accelerated; 3) Multi-objective elite retention and reconstruction Use non-dominated sorting to screen the Pareto optimal solution set, decode the beam parameters of the elite individuals, and output the optimal solution: In formula (5), The terminal movement direction angle predicted in step 1; the main lobe gain MLG and side lobe peak are calculated in real time through the electromagnetic simulation model; 4) closed-loop feedback optimization; Send the optimal beam parameters to the Luneburg lens controller to collect the actual sidelobe suppression ratio If the terminal receives the signal strength RSSI, , then restart the population evolution process. In a specific implementation example, the improved genetic algorithm can effectively improve the performance of the beam by optimizing the beam parameters. This method not only optimizes the detailed beam state at the theoretical level, but can also be dynamically adjusted according to the actual scenario. This provides strong data support for operators of wireless communication systems, helping them to improve communication quality and efficiency. The improved genetic algorithm first performs beam parameter encoding and population initialization, and then adjusts the beam width based on the current CSI information. θ and pointing angle ϕ Perform binary coding to generate the initial population. According to the fitness function Calculate the individual fitness score, then adjust the dynamic crossover and mutation probability according to the number of iterations t and population diversity Dynamically adjust crossover probability and mutation probability , the formula is in is the preset maximum number of iterations. When, a significant increase Pm To 0.3, to avoid the population falling into the local optimum; when When Set to 0 to speed up the convergence. Then perform multi-objective elite retention and reconstruction, use non-dominated sorting to screen the Pareto optimal solution set, decode the beam parameters of the elite individuals, and output the optimal solution. is the terminal moving direction angle predicted in step 1, the main lobe gain MLG The sidelobe peak is calculated in real time through the electromagnetic simulation model. Finally, closed-loop feedback optimization is performed and the optimal beam parameters are sent to the Luneburg lens controller to collect the actual sidelobe suppression ratio. The terminal receives the signal strength RSSI ,like The population evolution process is restarted. The calculated data is put into the actual operation of the wireless communication system, and the algorithm parameters are modified based on the feedback results. Table 2 shows the data comparison using the improved genetic algorithm. From Table 2, we can see that in the improved genetic algorithm, the weighted coefficient γ 1 and γ The value of 2 affects the degree of communication efficiency improvement. Appropriate weighting coefficients can more effectively balance mainlobe gain and sidelobe suppression, optimizing beam performance. The algorithm monitors beam status in real time, providing operators with detailed information on the current state of the communication system, including beamwidth, pointing angle, and sidelobe suppression ratio. This precise monitoring not only identifies potential issues promptly but also provides clear guidance to the network optimization team, enabling them to develop targeted optimization strategies.

[0018] Furthermore, the cross-layer parameter joint tuning method works as follows: First, an optimization variable vector is defined, including three key parameters: antenna tilt, transmit power in each frequency band, and the number of MIMO layers. The antenna tilt range is set to 0°-15°, the transmit power range is 0-46dBm, and the number of MIMO layers can be selected from 1 to 8. Next, a multi-objective optimization model is constructed around the interference suppression ratio, transmit power, and throughput targets, with the optimization goal of minimizing the sum of the inverse of the interference suppression ratio and the normalized transmit power, while ensuring that the throughput does not fall below a set threshold. In this process, a deep reinforcement learning framework is adopted, taking the base station MAC layer throughput, physical layer signal-to-interference-and-noise ratio, and RRC layer connection status as state inputs, and outputting the optimal parameter combination through a competitive network architecture. Finally, a layered training strategy is implemented. In the offline phase, policy gradient training is performed based on large-scale channel simulation data. In the online phase, transfer learning is used to quickly adapt to the real-time network environment. Parameters are updated every 50ms to achieve a stable interference suppression ratio above 18dB while reducing the transmit power by 15%-20%, achieving a balanced optimization of performance and energy consumption.

[0019] In a specific implementation example, the implementation of the cross-layer parameter joint optimization technology solution is mainly divided into four key stages: parameter initialization and system integration, deep reinforcement learning framework deployment, layered training and online optimization, and performance evaluation and continuous iteration. During the parameter initialization and system integration stage, the base station integrates the multi-band antenna array hardware module and seamlessly connects with the base station operating system and communication protocol stack. The network management system configures the antenna with an initial tilt angle of 7.5°, a transmit power of 30dBm, and MIMO layers of 4 as the optimization starting point. At the same time, an API interface is developed to collect cross-layer information such as average throughput, SINR distribution, and handover success rate from the MAC layer, physical layer, and RRC layer. When deploying the deep reinforcement learning framework, the state vector S is defined to include the total MAC cell throughput, the average SINR of each UE, and the RRC connection establishment success rate. The action vector A includes the antenna tilt adjustment, the power increment for each frequency band, and the increase or decrease in the number of MIMO layers. A master policy network and a target network are constructed using a five-layer fully connected network. An experience replay buffer is used to store the most recent 10,000 state-action-reward-next-state quadruple pairs. During the layered training and online optimization phases, offline training uses the 3GPP TR 38.901 channel model to generate 100,000 channel samples. The parallel policy gradient algorithm (PPO) is used for 72 hours of offline training. A / B testing determines the weighted ratio of interference rejection ratio, power consumption, and throughput to be 0.4:0.3:0.3. The model inference engine is deployed online in the existing base station network. Transfer learning is used to fine-tune the offline trained model in the new area for 24 hours, achieving a 50ms parameter update cycle. This means that every 20ms is used to collect state information, 10ms to complete inference calculations, and 20ms to perform parameter adjustments. During the performance evaluation and continuous iteration phase, a network performance monitoring system was deployed to collect real-time KPI statistics. A digital twin model was established to analyze the contribution of parameter adjustments. Genetic algorithms were used to optimize hyperparameters weekly, triggering active learning processes to expand the data set for new scenarios. This solution stabilized the network's interference suppression ratio at 20.3dB, a 2.1dB improvement over traditional DAS systems. The solution also reduced average base station transmit power by 17.8%, saving approximately 23,000 kWh annually. The average user-plane throughput reached 926Mbps, exceeding the optimization target by 18.6%. Integration with the existing network management system enabled full automation of the optimization process, reducing network optimization labor costs by over 40%.

[0020] Furthermore, the working method of the self-learning compensation is: Edge nodes periodically upload user experience quality of experience (QoE) data, including throughput, latency, and packet loss rate characteristics. Secure multi-party computing (MPC) technology is used to protect data privacy. The global model aggregates and locally updates parameters every 10 seconds. A bidirectional LSTM model with 128 hidden units is constructed. The antenna parameters and QoE data of the previous 5 minutes are input, and the parameter correction value for the next minute is output as Δ=[Δ θ ,Δ ϕ,Δ P ], the prediction mean square error is less than 5%; when sudden electromagnetic interference is detected and the interference power is greater than -80dBm, the spectrum sensing module is triggered to scan the idle sub-bands within the 500MHz bandwidth, and the available spectrum holes are identified through fast Fourier transform (FFT). The subcarrier allocation and beam direction adjustment are completed within 2ms to ensure the continuity of the communication link.

[0021] In specific implementation examples, the implementation of the self-learning compensation technology solution primarily involves three key steps: data collection and privacy protection, prediction model training and deployment, and dynamic spectrum management and interference mitigation. During the data collection and privacy protection phase, intelligent gateway devices and edge servers are deployed at the network edge, integrating a QoE data collection module to monitor terminal throughput, latency, and packet loss rate in real time at a rate of once per second. Enterprise campus networks deploy edge servers in each office area to collect Wi-Fi 6 / 7 terminal data, while cellular networks integrate this collection function into 5G base station distribution units. During prediction model training and deployment, a bidirectional LSTM model with 128 hidden units is constructed, outputting three-dimensional parameter corrections. This model is pre-trained in the cloud for two weeks using historical data using the PyTorch framework, employing the Adam optimizer, a learning rate of 0.001, and a batch size of 64. The pre-trained model is then quantized and compressed, then distributed to edge nodes via secure channels. Every 10 minutes, each node performs federated learning fine-tuning using the latest 5 minutes of local data, updating only the LSTM layer weights. During dynamic spectrum management and interference response, the base station integrates a broadband active antenna array and an electromagnetic spectrum monitoring module to scan the interference power spectrum density in the 24-40GHz frequency band in real time. Spectrum sensing is triggered when the interference power in any sub-band exceeds -80dBm / 10MHz. The spectrum sensing module uses an FFT algorithm to decompose the 500MHz bandwidth signal into 2500 200kHz-spaced subcarriers, scanning them at a 0.5ms period. After identifying spectrum holes, the beamforming algorithm completes beam direction adjustment and antenna parameter updates within 2ms. The solution addresses electromagnetic interference from industrial equipment. When a welding robot in the AGV operating area causes communication interruption, the system completes spectrum switching and beam adjustment within 2ms, shortening link recovery time by 87%, increasing network throughput by 14.2%, and reducing packet loss by 31.5%. In telemedicine workshop testing, it ensures interference-free communication for ultrasound diagnostic equipment within a 300-meter range, reducing the freezing rate of surgical video transmission from 8.3% to 1.7%, verifying its effectiveness in high-reliability scenarios. Furthermore, the implementation steps of the multi-band adaptive switching are as follows: 1) User behavior prediction model; Based on historical mobile trajectories and service types, a support vector machine (SVM) is used to build a prediction model for entering a weak coverage area. Input parameters include terminal movement speed, direction angle, and RSRP value of the current frequency band. The prediction accuracy is ≥95%; 2) Switching trigger criteria; When the terminal's received power in the 3300MHz band falls below -105dBm and it is predicted to enter a millimeter wave coverage blind spot, the frequency band switching protocol is initiated and a pre-configuration request is sent to the neighboring base station via the Xn interface. 3) Carrier aggregation seamless switching; A hard switching mechanism based on measurement reports is adopted to establish a millimeter wave link in the target frequency band in advance. The channel state is estimated through the uplink synchronization signal SRS, and the inverse fast Fourier transform (IFFT) is used to generate a multi-band aggregation signal to achieve parallel transmission of the 3300MHz low frequency band and the 28GHz high frequency band. The switching delay is controlled within 10ms through time advance compensation (TA).

[0022] In a specific implementation example, the implementation of a multi-band adaptive switching technology solution in a 5G network environment mainly includes three core steps: user behavior prediction model construction and application, switching trigger criteria implementation, and carrier aggregation seamless switching. During the user behavior prediction model construction and application stage, the 5G base station side collects the terminal's mobile speed, direction angle, and 3300MHz frequency band reference signal received power (RSRP) value through the wireless interface. After preliminary processing and normalization of the user service type data by the baseband unit, it serves as the input feature of the support vector machine (SVM) model. The SVM model trained with historical data is deployed in the baseband unit of the base station, predicting the terminal status every second. In a certain commercial area, the prediction accuracy for pedestrian users with an average moving speed of 5km / h and car users with an average moving speed of 30km / h entering the millimeter wave coverage blind spot reached 97.3% and 96.1%, respectively. During the handover triggering process, the base station continuously monitors the RSRP value in the 3300MHz band. When it falls below -105dBm, an SVM model evaluation is triggered. For example, if a terminal in an industrial park is near metal processing equipment, RSRP monitoring and the model work together to identify the risk of entering a blind spot. When the conditions are met, the base station sends a pre-configuration request with the terminal context information to the neighboring millimeter-wave base station via the Xn interface. When seamless handover is implemented using carrier aggregation, the terminal sends an uplink synchronization signal after receiving the handover preparation message. The millimeter-wave base station uses this information to estimate the channel state information and aggregate the 3300MHz and 28GHz band signals using an inverse fast Fourier transform (IFFT). This increased the downlink peak rate by 3.2 times to 2.8Gbps in a large shopping mall. A timing advance (TA) compensation mechanism is used to control handover latency to less than 10ms, with an average latency of 8.3ms in actual tests. The solution has achieved remarkable results in different scenarios: the switching success rate for high-definition video streaming users in commercial areas entering blind spots in shopping malls reached 99.4%, the video freeze rate was reduced to 0.8%, and the service interruption time for VR / AR device switching was controlled within 5ms; the switching success rate for industrial IoT devices in industrial parks remained at 98.7%, and the transmission delay jitter was controlled within 1ms; the switching success rate in high-speed mobile scenarios at transportation hubs remained at 97.6%, and the downlink throughput fluctuation range was controlled within ±15%, fully meeting diverse business needs.

[0023] Furthermore, the working method of the fault diagnosis and error calculation is: An improved Bayesian network model containing 12 nodes was constructed. The input parameters included intermodulation distortion (PIM) ≥ -90dBc, feed point impedance deviation > 5%, and number of array element failures > 3 abnormal features. The antenna health level was output as 0-1, and the diagnostic accuracy was ≥ 98%. The reflection coefficient S11, standing wave ratio (VSWR), and temperature data of the antenna array were collected every 20ms. Gradual faults were identified using the hidden Markov model (HMM), and the sudden fault detection delay was < 5ms. When more than three adjacent array elements were detected to have failed, the preset 15% matrix oscillator array was activated, and the phase of the radiating element was dynamically adjusted using electromagnetic metamaterials. The beamforming vector w was reconstructed based on the least squares method to ensure that the gain loss after reconstruction was ≤ 1.5dB. The system availability was guaranteed to be ≥ 99.999% through the N+1 redundancy design.

[0024] In a specific implementation example, when implementing fault diagnosis and error calculation technology solutions in a large-scale 5G base station antenna array system, it mainly covers four aspects: monitoring system deployment and data collection, fault diagnosis model construction and training, fault detection and fault tolerance mechanism implementation, and system availability assurance and maintenance. During the monitoring system deployment and data collection phase, a micro sensor is installed in each array element to monitor the reflection coefficient S in real time. 11 , VSWR, and temperature. High-precision impedance monitoring modules are installed at each feed point, and intermodulation distortion monitors are deployed at key nodes to detect PIM values. High-speed data acquisition cards collect data synchronously every 20ms and filter and normalize preprocessing to remove noise outliers. When building and training the fault diagnosis model, an improved Bayesian network model with 12 nodes is constructed, encompassing intermodulation distortion (PIM), feed point impedance deviation, and the number of element failures. The conditional probability distribution of the nodes is determined based on prior knowledge and historical data. After extensive training and cross-validation optimization using historical data, the model achieves diagnostic accuracy exceeding 98%. During the implementation of the fault detection and fault tolerance mechanism, a hidden Markov model (HMM) is used to identify gradual faults based on the preprocessed data observation sequence and regularly update the model parameters. A fast detection algorithm is designed to minimize the delay in detecting sudden faults to less than 5ms. When more than three adjacent element failures are detected, 15% of the matrix oscillator array is automatically activated. The phase of the radiating elements is dynamically adjusted using electromagnetic metamaterials. The beamforming vector is reconstructed using the active elements using the least squares method, keeping the post-reconstruction gain loss within 1.5dB. To ensure system availability and maintenance, we implemented an N+1 redundancy design and conducted regular inspections and maintenance. We continuously monitored operational status and performed regular maintenance and optimization based on data, raising system availability to 99.999%. This solution, through full-process technical implementation, effectively improved antenna system reliability, reduced communication interruptions, and ensured a secure user experience.

[0025] In a further embodiment, the improved Bayesian network model comprises a three-layer node structure, a 6-node bottom observation layer, a 3-node intermediate feature layer, and a 3-node top decision layer. The improved Bayesian network model comprises an online parameter update algorithm, and the function is: In formula (6), is the dynamic weight factor, is the posterior probability of failure given the evidence, is the normalization constant. The bottom observation layer (6 nodes) of the three-layer node structure is shown in Table 3. The intermediate feature layer (3 nodes) is shown in Table 4. The top decision layer (3 nodes) is shown in Table 5. The performance indicators are shown in Table 6. Table 6 Performance indicators Through the deep integration of a three-layer node structure and an online parameter update algorithm, an improved Bayesian network achieves high accuracy, low latency, and adaptability in 5G base station antenna array fault diagnosis. The core advantage of this solution lies in its integration of domain knowledge (such as electromagnetic characteristics and 5G protocols) with probabilistic reasoning models. Using a dynamic weighting mechanism, it adapts to complex and changing wireless environments, providing a solid foundation for the reliable operation of 5G networks.

[0026] Although specific embodiments of the present invention have been described above, those skilled in the art will appreciate that these specific embodiments are merely illustrative, and that those skilled in the art may omit, substitute, and modify the details of the methods and systems described above without departing from the principles and spirit of the present invention. For example, combining the above method steps to perform substantially the same functions in substantially the same manner to achieve substantially the same results falls within the scope of the present invention. Accordingly, the scope of the present invention is limited solely by the appended claims.

Claims

1. A method for implementing data communication using a multi-band adaptive antenna based on 5G communication, characterized in that: The following steps are involved: Step 1: Dynamic perception and multimodal signal fusion; Through a multimodal sensor network, the system collects terminal motion trajectories, spatial obstacle distribution, and electromagnetic interference maps in real time. It then generates a three-dimensional channel state matrix based on the fusion of the extended Kalman filter algorithm and a convolutional neural network to predict the terminal's movement path within the next 0.5 seconds. Step 2: Multi-band intelligent analysis; Adopting a frequency band adaptive selection algorithm, integrated with the Deep Q Network (DQN) to jointly evaluate the current network load, user service type, and channel quality, it dynamically allocates 3300-3600MHz, 4800-5000MHz, and millimeter wave frequency band resources. Based on the principle of non-orthogonal multiple access (NOMA), it achieves a 40% increase in spectrum utilization. Step 3: Hybrid beamforming and dynamic reconstruction; Through the collaborative design of Luneburg lens dielectrics and electromagnetic metamaterial arrays, multi-beam independent electrically regulated signals are generated. An improved genetic algorithm is used to optimize beam width and pointing angle, achieving 3D beam coverage in both horizontal and vertical dimensions. The beam shape is dynamically reconstructed based on CSI information fed back by the terminal, improving sidelobe suppression by 30%. Step 4: Jointly tune cross-layer parameters; The antenna tilt, power allocation, and MIMO layer number parameters are mapped into a multi-objective optimization problem. A dynamic strategy model is constructed using the deep reinforcement learning (DRL) framework. By combining offline training with online transfer learning, the interference suppression ratio (ICIC) is optimized to over 18 dB while reducing energy consumption by 15%. Step 5: Real-time monitoring and self-learning compensation; A self-learning compensation module is introduced to aggregate user QoE data from edge nodes. The module analyzes and dynamically adjusts antenna parameters, enabling rapid filling of spectrum holes in response to sudden electromagnetic interference to ensure communication continuity. Step 6: Multi-band adaptive switching; Based on the user behavior prediction model, when it is predicted that the terminal is about to enter a weak coverage area, the frequency band switching protocol is triggered, and carrier aggregation technology is used to achieve seamless switching between the 3300MHz low frequency band and the millimeter wave high frequency band, with a switching delay of less than 10ms; Step 7: Fault diagnosis and error calculation; An antenna health assessment model is constructed through an improved Bayesian network model to perform real-time diagnosis of PIM intermodulation distortion and feed point impedance mismatch abnormal conditions, and the matrix dipole array is enabled to reconstruct the radiation pattern to ensure system availability ≥ 99.999%.

2. The method for realizing data communication using a multi-band adaptive antenna based on 5G communication according to claim 1, characterized in that: The working method of the dynamic perception and multimodal signal fusion is: The inertial measurement unit (IMU) deployed on the terminal collects three-axis acceleration and angular velocity data to construct a motion trajectory matrix, and uses LiDAR and millimeter-wave radar to obtain the three-dimensional coordinates of spatial obstacles. The electromagnetic spectrum sensor scans the interference power spectrum density in the 24-40GHz frequency band in real time to form a multi-dimensional original data set. A terminal motion state space model is established, and the denoised trajectory data, obstacle point cloud, and interference map are converted into a three-dimensional grid tensor. The tensor is input into a 12-layer deep CNN network, and spatial feature associations are extracted through void convolution and attention mechanism. The channel state matrix containing path loss and multipath delay is output. The time series characteristics of the channel state matrix are modeled based on the long short-term memory network (LSTM). Combined with the terminal's historical movement speed and direction angle, a particle filter algorithm is used to predict the three-dimensional movement trajectory coordinates within the next 0.5 seconds, with a prediction root mean square error of less than 0.3 meters.

3. The method for realizing data communication using a multi-band adaptive antenna based on 5G communication according to claim 1, characterized in that: The steps for implementing the frequency band adaptive selection algorithm are as follows: 1) Multi-dimensional state feature extraction; Based on the channel state matrix, the network load, service type and channel quality are characterized and the feature fusion formula is: In formula (1), S represents the comprehensive score of frequency band selection; Indicates the normalized network load factor, with a value range of [0,1]; Indicates the channel interference power; Indicates the ratio of the current signal-to-noise ratio to the target threshold; Indicates the service latency requirement. is the baseline delay; are the weight coefficients of load and delay, which are dynamically adjusted by DQN; 2) Dynamic Q-value modeling and action space construction; Define the state space S is a vector set containing multi-dimensional features of network load, service type and channel quality, and the action space A Covering resource allocation strategies for 3300-3600MHz, 4800-5000MHz and millimeter wave bands, including band bandwidth allocation ratio and power control parameters, building a deep Q network model with the network input being state space S The corresponding feature vector is mapped to the action value function through the multi-layer perceptron structure; 3) Frequency band allocation and NOMA power control; According to the action strategy output by DQN, the primary and auxiliary frequency bands are allocated according to priority; differential power allocation is adopted for NOMA users: In formula (2), is the transmission power of the kth user, K is the number of multiplexed users, is the fairness adjustment factor, is the maximum transmit power, represents the signal-to-interference-and-noise ratio adjustment coefficient, For the k Signal-to-interference-noise ratio for each user; 4) Online strategy optimization and feedback; A lightweight DQN model is deployed on the edge computing unit of the base station to collect channel quality indicator (CQI) and bit error rate (BER) data in real time; the network weights are updated using the double-delayed deep deterministic policy gradient (TD3) algorithm.

4. The method for realizing data communication using a multi-band adaptive antenna based on 5G communication according to claim 1, characterized in that: The implementation steps of the improved genetic algorithm are: 1) Beam parameter encoding and population initialization Based on the current CSI information, the beam width θ and the pointing angle ϕ are binary-coded to generate the initial population. The fitness function is defined as: In formula (3), G represents the individual fitness score; is the beam width, is the theoretical optimal pointing angle; is the weight coefficient, and its initial value is set to 0.5; Indicates the sidelobe suppression level, which is the preset maximum sidelobe suppression level; MLG main lobe gain, Preset maximum main lobe gain; is the weighting coefficient; 2) Dynamic crossover and mutation probability adjustment According to the number of iterations t and population diversity Dynamically adjust the crossover probability Pc and mutation probability Pm: In formula (4), T is the preset maximum number of iterations; when When , Pm is greatly increased to 0.3 to prevent the population from falling into the local optimum; when When Pm is set to 0, the convergence speed is accelerated; 3) Multi-objective elite retention and reconstruction Use non-dominated sorting to screen the Pareto optimal solution set, decode the beam parameters of the elite individuals, and output the optimal solution: In formula (5), The terminal movement direction angle predicted in step 1; the main lobe gain MLG and side lobe peak are calculated in real time through the electromagnetic simulation model; 4) closed-loop feedback optimization; Send the optimal beam parameters to the Luneburg lens controller to collect the actual sidelobe suppression ratio If the terminal receives the signal strength RSSI, , then restart the population evolution process.

5. The method for realizing data communication based on a multi-band adaptive antenna of 5G communication according to claim 1, characterized in that: The working method of the cross-layer parameter joint tuning is: First, an optimization variable vector is defined, consisting of three key parameters: antenna tilt, transmit power in each frequency band, and the number of MIMO layers. The antenna tilt range is set to 0°-15°, the transmit power range is 0-46dBm, and the number of MIMO layers can be selected from 1 to 8. Next, a multi-objective optimization model is constructed around the interference suppression ratio, transmit power, and throughput targets. The optimization goal is to minimize the sum of the inverse of the interference suppression ratio and the normalized transmit power, while ensuring that the throughput does not fall below a set threshold. In this process, a deep reinforcement learning framework is adopted, taking the base station MAC layer throughput, physical layer signal-to-interference-and-noise ratio, and RRC layer connection status as state inputs, and outputting the optimal parameter combination through a competitive network architecture. Finally, a layered training strategy is implemented. In the offline phase, policy gradient training is performed based on large-scale channel simulation data. In the online phase, transfer learning is used to quickly adapt to the real-time network environment. Parameters are updated every 50ms to achieve a stable interference suppression ratio above 18dB while reducing transmit power by 15%-20%, achieving a balanced optimization between performance and energy consumption.

6. The method for realizing data communication based on a multi-band adaptive antenna of 5G communication according to claim 1, characterized in that: The working method of the self-learning compensation is: Edge nodes periodically upload user experience quality of experience (QoE) data, including throughput, latency, and packet loss rate characteristics. Secure multi-party computing (MPC) technology is used to protect data privacy. The global model aggregates and locally updates parameters every 10 seconds. A bidirectional LSTM model with 128 hidden units is constructed. The antenna parameters and QoE data of the previous 5 minutes are input, and the parameter correction value for the next minute is output as Δ=[Δ θ ,Δ ϕ ,Δ P ], the prediction mean square error is less than 5%; when sudden electromagnetic interference is detected and the interference power is greater than -80dBm, the spectrum sensing module is triggered to scan the idle sub-bands within the 500MHz bandwidth, and the available spectrum holes are identified through fast Fourier transform (FFT). The subcarrier allocation and beam direction adjustment are completed within 2ms to ensure the continuity of the communication link.

7. The method for realizing data communication based on a multi-band adaptive antenna of 5G communication according to claim 1, characterized in that: The multi-band adaptive switching is implemented in the following steps: 1) User behavior prediction model; Based on historical mobile trajectories and service types, a support vector machine (SVM) is used to build a prediction model for entering a weak coverage area. Input parameters include terminal movement speed, direction angle, and RSRP value of the current frequency band. The prediction accuracy is ≥95%; 2) Switching trigger criteria; When the terminal's received power in the 3300MHz band falls below -105dBm and it is predicted to enter a millimeter wave coverage blind spot, the terminal initiates the frequency band switching protocol and sends a pre-configuration request to the neighboring base station through the Xn interface. 3) Carrier aggregation seamless switching; A hard switching mechanism based on measurement reports is adopted to establish a millimeter wave link in the target frequency band in advance. The channel state is estimated through the uplink synchronization signal SRS, and the inverse fast Fourier transform (IFFT) is used to generate a multi-band aggregation signal to achieve parallel transmission of the 3300MHz low frequency band and the 28GHz high frequency band. The switching delay is controlled within 10ms through time advance compensation (TA).

8. The method for realizing data communication based on a multi-band adaptive antenna of 5G communication according to claim 1, characterized in that: The working method of the fault diagnosis and error calculation is as follows: An improved Bayesian network model with 12 nodes was constructed. Input parameters included intermodulation distortion (PIM) ≥ -90 dBc, feed point impedance deviation > 5%, and number of element failures > 3 abnormal features. The antenna health level was output as 0-1, with a diagnostic accuracy of ≥ 98%. The antenna array's reflection coefficient (S11), standing wave ratio (VSWR), and temperature data were collected every 20 ms. Gradual faults were identified using a hidden Markov model (HMM), and the sudden fault detection delay was < 5 ms. When more than three adjacent element failures were detected, a preset 15% matrix oscillator array was activated. The radiating element phase was dynamically adjusted using electromagnetic metamaterials, and the beamforming vector w was reconstructed using the least squares method to ensure that the gain loss after reconstruction was ≤ 1.5 dB. System availability was guaranteed to be ≥ 99.999% through an N+1 redundancy design. The improved Bayesian network model includes a three-layer node structure, a 6-node bottom observation layer, a 3-node intermediate feature layer, and a 3-node top decision layer. The improved Bayesian network model includes an online parameter update algorithm, and the function is: In formula (6), is the dynamic weight factor, is the posterior probability of failure given the evidence, is the normalization constant.

Images