MIMO antenna array beam control method and system based on metamaterial and predictive optimization

Abstract

The application relates to a MIMO antenna array beam control method and system based on metamaterials and prediction optimization, wherein the method adjusts a sampling frequency according to a current channel change rate, collects a multi-antenna channel state information matrix in real time, and performs pretreatment; inputs the pretreated channel state information matrix in a preset time window into a trained time sequence prediction model to obtain a predicted channel matrix in a future preset time length; takes the predicted channel matrix as input, calculates bias distribution parameters of a metamaterial decoupling layer and excitation weight parameters of an antenna array through a multi-objective genetic algorithm; applies the bias distribution parameters to the metamaterial decoupling layer, controls the dielectric constant in front of an antenna array radiation aperture to be in a non-uniform gradient distribution, and applies the excitation weight parameters to the antenna array to control the beam pointing of a radiation pattern, so that the purpose of beam control with high isolation and high dynamic tracking precision is achieved.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a method and system for beam control of MIMO antenna arrays based on metamaterials and predictive optimization. Background Technology

[0002] In 5G and future 6G wireless communication systems, massive MIMO antenna arrays are core components for improving system capacity and spectral efficiency. To meet the demands of miniaturization and integration, the spacing between antenna elements is often reduced to less than half a wavelength. The strong mutual coupling effect caused by surface wave propagation between elements directly deteriorates port isolation, resulting in a significant decrease in system channel capacity. Existing decoupling schemes based on eigenmode compensation matrices and fixed electromagnetic bandgap structures exhibit a sharp decline in decoupling effectiveness when the element spacing is less than 0.3 times the wavelength, and cannot adapt to the dynamic changes in coupling state during beam scanning. In high-speed mobile scenarios, the coherence time of wireless channels is significantly shortened. Traditional beamforming algorithms are affected by channel state information feedback and the inherent delay of the computation link, causing parameter updates to lag behind channel changes, which easily leads to beam pointing inaccuracies. Existing solutions struggle to achieve a balance between low computational complexity and high tracking accuracy. In multi-band collaborative operation scenarios, traditional fixed resonant antennas cannot be dynamically reconfigured between Sub-6GHz and millimeter-wave bands. Deploying multiple arrays will lead to a doubling of volume and cost and an increase in electromagnetic compatibility risks. Existing metamaterial decoupling solutions are mostly fixed structures that can only adapt to a single frequency band and fixed beam pointing. A few tunable solutions can only independently adjust the decoupling parameters and cannot achieve global collaborative optimization with digital beamforming, making it difficult to adapt to the needs of complex communication scenarios. Summary of the Invention

[0003] The main objective of this invention is to provide a beam control method and system for MIMO antenna arrays based on metamaterials and predictive optimization. By constructing channel prediction, multi-objective genetic algorithm joint optimization, and dynamic control of metamaterial decoupling layer, the aim is to achieve beam control with high isolation and high dynamic tracking accuracy.

[0004] To achieve the above objectives, this invention provides a MIMO antenna array beam control method based on metamaterials and predictive optimization, comprising the following steps: The sampling frequency is adjusted according to the current channel change rate, and the multi-antenna channel state information matrix is ​​collected in real time. The collected channel state information matrix is ​​then preprocessed to obtain the preprocessed channel state information matrix. The preprocessed channel state information matrix within a preset time window is input into a trained time series prediction model to obtain a predicted channel matrix for a future preset duration. Using the predicted channel matrix as input, the bias distribution parameters of the metamaterial decoupling layer and the excitation weight parameters of the antenna array are calculated by a multi-objective genetic algorithm. The optimization objectives of the calculation include maximizing the equivalent array gain, minimizing the port coupling power, and minimizing the gain deviation between frequency bands. The bias distribution parameter is applied to the metamaterial decoupling layer to adjust the dielectric constant in front of the antenna array's radiation aperture to have a non-uniform gradient distribution, and the excitation weight parameter is applied to the antenna array to adjust the beam pointing of the radiation pattern.

[0005] Further, the steps of adjusting the sampling frequency according to the current channel change rate, acquiring a multi-antenna channel state information matrix in real time, and preprocessing the acquired channel state information matrix to obtain a preprocessed channel state information matrix include: Obtain the user terminal's movement speed parameters in the current communication environment; The minimum threshold of the sampling frequency is determined based on the moving speed parameter. The sampling frequency corresponding to the minimum threshold is higher than the rate of change of the channel state information under multipath time-varying conditions, and the effective sample size collected within the preset time window meets the input requirements of the time series prediction model. The channel state information corresponding to each antenna unit is collected through the analog-to-digital conversion circuit of the radio frequency front end at a sampling frequency not lower than the minimum threshold, forming a channel state information matrix. The channel state information matrix is ​​subjected to noise cancellation processing; The channel state information matrix after noise cancellation is subjected to phase calibration to obtain a preprocessed channel state information matrix.

[0006] Furthermore, the training steps of the trained time series prediction model include: Construct a training dataset containing historical channel state information sequences under different movement speed scenarios; The time series prediction model is subjected to offline supervised learning using the training dataset to learn the evolution of the channel state under multipath time-varying conditions. After training is completed, the time series prediction model is loaded into the channel prediction module. The time series prediction model is used to perform forward inference on the input preprocessed channel state information matrix and output the predicted channel matrix.

[0007] Furthermore, the steps of calculating the bias distribution parameters of the metamaterial decoupling layer and the excitation weight parameters of the antenna array using a multi-objective genetic algorithm include: Construct a fitness function, which includes maximizing the equivalent array gain, minimizing the antenna port coupling power, and minimizing the array gain deviation between different operating frequency bands; The bias distribution parameters and the excitation weight parameters are used as variables to be optimized and combined into the same solution vector. Under the constraints of the upper limit of the total transmit power of the antenna array and the dynamic range of the bias voltage of the metamaterial decoupling layer, an initial solution set is generated. The initial solution set is subjected to selection, crossover, and mutation operations. After multiple generations of iterative evolution, the bias distribution parameter and the excitation weight parameter are selected from the Pareto optimal frontier solution set.

[0008] Furthermore, after calculating the bias voltage distribution parameters and excitation weight parameters, the following steps are also included: The bias distribution parameters are encoded into metamaterial decoupling layer control commands, and the metamaterial decoupling layer control commands are sent to the liquid crystal metamaterial control circuit. The excitation weight parameters are encoded into antenna excitation control commands, and the antenna excitation control commands are sent to the digital beamformer. Based on the response delay of the liquid crystal metamaterial control circuit and the response delay of the digital beamformer, a corresponding trigger offset is set. After confirming that both the liquid crystal metamaterial control circuit and the digital beamformer have received the corresponding control commands, the dielectric constant distribution modulation action and the radiation pattern modulation action are synchronously triggered according to the trigger offset.

[0009] Further, the step of applying the bias distribution parameters to the metamaterial decoupling layer to adjust the dielectric constant in front of the antenna array radiation aperture to exhibit a non-uniform gradient distribution includes: The bias distribution parameters are converted into bias voltage values ​​for each liquid crystal cell in the metamaterial decoupling layer, wherein each liquid crystal cell is arranged in an array. The bias voltage value is output to the corresponding liquid crystal cell through a multiplexer digital-to-analog converter. The deflection state of the molecules inside each liquid crystal cell is controlled by adjusting the amplitude of the bias voltage. The molecules maintain their initial alignment state when the bias voltage value is below a threshold voltage, and begin to deflect after the bias voltage value reaches the threshold voltage. The deflection angle increases with the increase of the bias voltage value and tends to stabilize after the bias voltage value reaches the saturation voltage. By adjusting the bias voltage amplitude of each liquid crystal cell, a spatial gradient distribution with different dielectric constant values ​​is formed in the metamaterial decoupling layer between the central region and the edge region.

[0010] Furthermore, the antenna array is a dual-frequency reconfigurable antenna array, with each antenna element integrating a tunable element. Prior to the step of adjusting the beam pointing of the radiation pattern, the method further includes: Obtain the operating frequency band indication signal currently allocated by the communication system; According to the operating frequency band indication signal, the resonant frequency of each antenna element is adjusted by the tunable element, and the antenna array is switched to the corresponding target frequency band for operation; The operating frequency band indication signal is used as the input parameter of the multi-objective genetic algorithm. The operating frequency band indication signal is used to calculate the frequency band characteristics of the current antenna array when the bias distribution parameters and the excitation weight parameters are matched.

[0011] Furthermore, the steps for adjusting the beam pointing of the radiation pattern include: The excitation weighting parameters are decomposed into the feed amplitude coefficient and feed phase coefficient of each antenna element; The feed amplitude coefficient and the feed phase coefficient are written into the complex multiplier of each antenna element in the digital beamformer; By exciting each antenna element according to its corresponding feed amplitude coefficient and feed phase coefficient, the radiation pattern beam of the antenna array is controlled to point towards the target user terminal.

[0012] Furthermore, after the beam pointing of the radiation pattern is adjusted and the antenna array enters a stable operating state, the method further includes: Obtain the port isolation and beam pointing deviation generated during the actual operation of the regulated antenna array; If the port isolation deviates from the preset threshold range, the port isolation is used as a feedback signal to calibrate the fitness function of the multi-objective genetic algorithm online. If the beam pointing deviation deviates from the preset threshold range, the beam pointing deviation is used as a feedback signal to correct the network weight parameters of the time series prediction model online, and the corrected network weight parameters are reloaded into the time series prediction model for model update.

[0013] This invention also provides a MIMO antenna array beam control system based on metamaterials and predictive optimization, comprising: The channel acquisition module is used to adjust the sampling frequency according to the current channel change rate, acquire the multi-antenna channel state information matrix in real time, and preprocess the acquired channel state information matrix to obtain the preprocessed channel state information matrix. The channel prediction module is used to input the preprocessed channel state information matrix within a preset time window into the trained time series prediction model to obtain the predicted channel matrix for a preset future duration. The joint calculation module is used to simultaneously calculate the bias distribution parameters of the metamaterial decoupling layer and the excitation weight parameters of the antenna array using the predicted channel matrix as input and a multi-objective genetic algorithm. The optimization objectives of the calculation include maximizing the equivalent array gain, minimizing the port coupling power, and minimizing the gain deviation between frequency bands. The execution control module is used to apply the bias distribution parameters to the metamaterial decoupling layer, adjust the dielectric constant in front of the antenna array radiation aperture to form a non-uniform gradient distribution, and apply the excitation weight parameters to the antenna array to adjust the beam pointing of the radiation pattern.

[0014] The MIMO antenna array beam control method and system based on metamaterials and predictive optimization provided by this invention have the following beneficial effects: This invention provides a low-coupling near-field environment for beamforming by setting a gradient-tunable liquid crystal metamaterial decoupling layer in front of the array's radiation aperture, thereby disrupting surface wave propagation conditions with a spatially non-uniform dielectric constant distribution; it also utilizes an offline-trained predictive model to predict future channel states in advance, providing an advance time window for optimization calculations, ensuring that beam control commands are synchronized with the real channel during execution, and maintaining a high beam alignment success rate in high-speed moving scenarios; furthermore, it employs a multi-objective genetic algorithm to jointly optimize the metamaterial bias parameters and antenna excitation parameters globally, jointly optimizing the decoupling and beamforming effects, improving the system's throughput in dual-frequency operating mode, and reducing power consumption. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a MIMO antenna array beam control method based on metamaterials and predictive optimization in one embodiment of the present invention. Figure 2 This is a structural block diagram of a MIMO antenna array beam control system based on metamaterials and predictive optimization in one embodiment of the present invention.

[0016] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Reference Figure 1 This is a flowchart illustrating a MIMO antenna array beam control method based on metamaterials and predictive optimization proposed in this invention, including the following steps: S1, adjust the sampling frequency according to the current channel change rate, collect the multi-antenna channel state information matrix in real time, and preprocess the collected channel state information matrix to obtain the preprocessed channel state information matrix. S2, input the preprocessed channel state information matrix within the preset time window into the trained time series prediction model to obtain the predicted channel matrix for the future preset duration; S3, using the predicted channel matrix as input, calculate the bias distribution parameters of the metamaterial decoupling layer and the excitation weight parameters of the antenna array through a multi-objective genetic algorithm. The optimization objectives of the calculation include maximizing the equivalent array gain, minimizing the port coupling power, and minimizing the gain deviation between frequency bands. S4, apply the bias distribution parameter to the metamaterial decoupling layer to adjust the dielectric constant in front of the antenna array radiation aperture to form a non-uniform gradient distribution, and apply the excitation weight parameter to the antenna array to adjust the beam pointing of the radiation pattern.

[0019] As described in step S1 above, in a mobile communication environment, the user terminal's moving speed directly determines the time-varying rate of the wireless channel, specifically manifested as the magnitude of the Doppler frequency shift. When the user terminal moves at a high speed, the channel state information changes significantly in a very short time. If the sampling frequency is too low, the collected channel state information will not accurately reflect the channel's changing characteristics, leading to the complete failure of subsequent channel prediction and beam control. Therefore, this step dynamically binds the sampling frequency to the channel change rate, adaptively determining the minimum threshold of the sampling frequency based on the user terminal's current moving speed to ensure that the number of samples collected per unit time can fully capture the time-varying details of the channel state information. Simultaneously, in actual communication systems, the raw channel state information collected by the RF front-end inevitably contains thermal noise, quantization noise, and random phase offsets between antenna channels. If these interference components are directly input into the prediction model without preprocessing, the model will treat them as valid signals for learning and deduction, resulting in severely degraded prediction accuracy. To address this, this step sequentially performs noise cancellation and phase calibration on the channel state information matrix, filtering out interference components and unifying the phase reference of each channel, resulting in a clean and phase-aligned preprocessed channel state information matrix.

[0020] As described in step S2 above, in high-speed mobile scenarios where channel coherence time is shortened, traditional beamforming algorithms suffer from issues such as delayed channel state information feedback and computation, resulting in updates lagging behind channel changes and a high probability of beam inaccuracy. This step addresses this by using a time-series prediction model to predict future channel states in advance, providing a lead time for beam control. This step utilizes an offline-trained time-series prediction model to perform time-series modeling and analysis on a sequence of continuously collected channel state information within a preset time window. It extracts the evolutionary patterns of channel states under multipath time-varying conditions and uses this to deduce the predicted channel matrix for the future preset time period. This predicted channel matrix provides an advance time window for subsequent beam control optimization, enabling optimization calculations to be performed in advance based on future channel states. This ensures synchronization between the actual control command execution and the real channel state, effectively avoiding beam inaccuracies caused by computational delays.

[0021] As described in step S3 above, existing solutions for decoupling control and beamforming optimization are performed step-by-step, failing to converge to the global optimum and struggling to dynamically adapt between Sub-6GHz and millimeter-wave bands, leading to reduced system throughput and energy efficiency. This step employs a multi-objective genetic algorithm to jointly optimize the metamaterial bias and antenna excitation weights globally. Traditional solutions typically first determine the structural parameters of the metamaterial decoupling layer to improve port isolation, and then independently optimize the excitation weights of the antenna array to achieve beamforming. However, the bias distribution parameters of the metamaterial decoupling layer directly determine the spatial distribution of the dielectric constant in front of the antenna array's radiating aperture. This distribution alters the effective radiation pattern of each antenna element, the mutual coupling strength between ports, and the array's manifold matrix, which is precisely the direct mathematical basis for calculating the excitation weight beamforming. Therefore, there is an inherently strong coupling relationship between the bias distribution parameters and the excitation weight parameters, making step-by-step optimization unable to converge to the global optimum. Step S3 employs a multi-objective genetic algorithm, grouping the bias distribution parameters and excitation weight parameters into a single solution vector. In the joint solution space, it simultaneously performs Pareto optimization for three objectives: maximizing the equivalent array gain, minimizing port coupling power, and minimizing inter-band gain deviation. The fitness function is used to simultaneously evaluate the decoupling and beamforming effects in each iteration. Finally, the optimal combination of bias distribution parameters and excitation weight parameters that achieves equilibrium across the three objectives is selected from the Pareto optimal frontier solution set, thereby achieving synergistic optimization of decoupling and beamforming.

[0022] As described in step S4 above, the bias distribution parameters are applied to the metamaterial decoupling layer to adjust the dielectric constant in front of the antenna array's radiating aperture to exhibit a non-uniform gradient distribution. The excitation weight parameters are then applied to the antenna array to adjust the beam pointing of the radiation pattern. The bias distribution parameters and excitation weight parameters obtained in step S3 are converted into the driving signal for the metamaterial decoupling layer and the feeding signal for the antenna array, respectively, making the optimization results effective at the physical level. These two adjustment actions are executed synchronously after time delay compensation. The non-uniform gradient dielectric constant distribution of the metamaterial decoupling layer not only suppresses mutual coupling between antenna elements in the near field but also changes the equivalent radiation boundary of the antenna elements, optimizing the far-field radiation pattern of the elements. The optimization calculation of the excitation weight parameters fully considers the influence of the metamaterial decoupling layer on the array manifold matrix. The two work synergistically to ensure accurate beam pointing of the antenna array under high isolation conditions, avoiding performance mismatch caused by traditional step-by-step optimization.

[0023] In one embodiment, step S1, which involves adjusting the sampling frequency according to the current channel change rate, acquiring a multi-antenna channel state information matrix in real time, and preprocessing the acquired channel state information matrix to obtain a preprocessed channel state information matrix, includes: Obtain the user terminal's movement speed parameters in the current communication environment; The minimum threshold of the sampling frequency is determined based on the moving speed parameter. The sampling frequency corresponding to the minimum threshold is higher than the rate of change of the channel state information under multipath time-varying conditions, and the effective sample size collected within the preset time window meets the input requirements of the time series prediction model. The channel state information corresponding to each antenna unit is collected through the analog-to-digital conversion circuit of the radio frequency front end at a sampling frequency not lower than the minimum threshold, forming a channel state information matrix. The channel state information matrix is ​​subjected to noise cancellation processing; The channel state information matrix after noise cancellation is subjected to phase calibration to obtain a preprocessed channel state information matrix.

[0024] In practice, the user terminal's movement speed parameters in the current communication environment are obtained through one or more combinations of various methods, such as positioning calculations at the base station, GPS information or inertial sensor data reported by the user terminal, and Doppler frequency shift estimation based on the channel state information's own change period. Movement speed parameters are a core physical quantity characterizing the degree of time-varying channel dynamics. When the user terminal is walking at low speed or stationary, the multipath structure and fading characteristics of the wireless channel remain relatively stable over a longer period, resulting in a longer channel coherence time. However, when the user terminal is moving at high speed, such as on a highway or in a high-speed rail scenario, the channel coherence time shortens drastically, and the channel state information changes significantly within a millisecond timescale.

[0025] After obtaining the movement speed parameters, a minimum threshold for the sampling frequency is determined based on these parameters. The determination of the minimum threshold follows two parallel constraints: the first is a physical sampling constraint, whereby the sampling frequency corresponding to the minimum threshold must be higher than the rate of change of channel state information under multipath time-varying conditions. Specifically, according to the Nyquist sampling theorem generalized to time-varying channels, the sampling frequency should be at least twice the channel Doppler shift to ensure that the discrete sampling sequence can completely retain the continuous change characteristics of the channel state information and avoid information loss due to sampling aliasing. The second constraint is a sample size constraint at the model input level. Within the preset time window, the effective sample size obtained by sampling at a frequency not lower than the minimum threshold must meet the minimum requirement of the time series prediction model for the input sequence length. The time series prediction model relies on channel state information samples from multiple consecutive time points to extract time-dependent features. If the number of samples within the time window is insufficient to constitute an effective input sequence length, the model's prediction accuracy will significantly decrease. By comprehensively calculating the physical sampling constraint and the model sample size constraint, the larger of the two values ​​is taken as the minimum threshold for the sampling frequency, ensuring that the sampling frequency meets both the physical requirements for channel change tracking and the data supply requirements of the prediction model. In this embodiment, the preset time window is set to 10ms. The length of this time window is determined by combining the channel coherence time and the input time series span of the prediction model, which can achieve a balance between capturing the time-varying characteristics of the channel and maintaining the real-time performance of the model inference.

[0026] After determining the minimum threshold, the analog-to-digital converter (ADC) circuit of the RF front-end acquires channel state information corresponding to each antenna element at a sampling frequency not lower than the minimum threshold. The RF front-end contains multiple parallel receiving channels, each corresponding to one antenna element. The ADC circuit of each channel discretizes the received RF signal according to the sampling frequency uniformly set by the system, extracting complex channel state information reflecting the current channel amplitude attenuation and phase rotation. The channel state information obtained by all antenna elements at the same sampling time is arranged by antenna index and sampling time to form a multi-antenna channel state information matrix. The dimension of this matrix is ​​determined by the number of antenna elements and the number of sampling times, completely recording the overall channel changes at the array level during the acquisition period.

[0027] After forming the channel state information matrix, noise cancellation and phase calibration are performed sequentially. Noise cancellation addresses thermal noise, quantization noise introduced by the RF front-end, and random electromagnetic interference from the environment. It employs filtering, smoothing, or statistical model-based noise suppression methods to separate the signal subspace from the noise subspace, extracting the effective signal components that reflect the true channel characteristics. Phase calibration addresses random phase shifts introduced between antenna receiving channels due to hardware differences, transmission line length variations, and asynchronous local oscillator signals. By comparing the response differences of each channel to a known reference signal, the phase compensation factor for each channel is calculated, and the data of each antenna element in the channel state information matrix is ​​phase-rotated to ensure uniform alignment of the phase references for all channels. After the cascaded processing of noise cancellation and phase calibration, a preprocessed channel state information matrix is ​​obtained. This matrix retains accurate information reflecting the multipath fading characteristics of the channel while suppressing errors introduced by noise interference and phase inconsistencies between channels.

[0028] In one embodiment, for step S2, the training steps of the trained time series prediction model include: Construct a training dataset containing historical channel state information sequences under different movement speed scenarios; The time series prediction model is subjected to offline supervised learning using the training dataset to learn the evolution of the channel state under multipath time-varying conditions. After training is completed, the time series prediction model is loaded into the channel prediction module. The time series prediction model is used to perform forward inference on the input preprocessed channel state information matrix and output the predicted channel matrix.

[0029] In specific implementation, the time series prediction model described in step S2 needs to be trained offline before deployment to obtain the ability to accurately model the evolution of channel state under multipath time-varying conditions. This includes constructing a training dataset containing historical channel state information sequences under different moving speed scenarios; using the training dataset to perform offline supervised learning on the time series prediction model to learn the evolution of channel state under multipath time-varying conditions; after training is completed, the time series prediction model is loaded into the channel prediction module to perform forward inference on the input preprocessed channel state information matrix and output the predicted channel matrix.

[0030] In practice, the construction of the training dataset must cover multiple representative mobile speed scenarios, ranging from low to high speeds. For each speed scenario, channel state information sequences are continuously collected over a specific time period through actual channel measurements or simulations based on standardized channel models, recording the multi-antenna channel state information matrix at each sampling moment. The channel state information sequences collected in chronological order constitute the input part of the training samples, while the corresponding future real channel state information matrices serve as the label part of the training samples. Together, they constitute a complete supervised learning sample. By repeating the above data collection and sample construction process under multiple speed scenarios, a training dataset containing various channel change patterns is accumulated.

[0031] During training, the input portion of each sample in the training dataset is sequentially fed into the time series prediction model. The model, through its internal time-series modeling structure, encodes and extracts features from the channel state information at each moment in the input sequence layer by layer, gradually capturing the channel change trends, time-varying weights of multipath components, and evolution patterns of coupling states between antenna elements. The predicted channel matrix output by the model is compared element-wise with the label portion of the training samples using a pre-defined loss function to calculate the prediction error. This prediction error is propagated layer by layer to each parameter node of the model through a backpropagation mechanism, and the network weight parameters within the model are iteratively updated according to the gradient descent principle, allowing the model's output in the next prediction round to gradually approach the true value. By traversing the entire training dataset and repeating multiple training cycles, the network weight parameters within the model are gradually adjusted to a stable optimal state. After training is complete, the time series prediction model is loaded into the channel prediction module. This module, as the system's online inference unit, receives the preprocessed channel state information matrix output from step S1 in real time during actual operation. Based on the trained fixed network weight parameters, it performs unidirectional layer-by-layer calculations on the input data, directly outputting the predicted channel matrix for a preset future duration, without performing any parameter updates or backpropagation calculations. The preset future duration is set to 3ms, which is determined based on the channel coherence time and beam control command execution delay in high-speed mobile scenarios, achieving a balance between prediction accuracy and control real-time performance.

[0032] In one embodiment, step S3, which involves calculating the bias distribution parameters of the metamaterial decoupling layer and the excitation weight parameters of the antenna array using a multi-objective genetic algorithm, includes: Construct a fitness function, which includes maximizing the equivalent array gain, minimizing the antenna port coupling power, and minimizing the array gain deviation between different operating frequency bands; The bias distribution parameters and the excitation weight parameters are used as variables to be optimized and combined into the same solution vector. Under the constraints of the upper limit of the total transmit power of the antenna array and the dynamic range of the bias voltage of the metamaterial decoupling layer, an initial solution set is generated. The initial solution set is subjected to selection, crossover, and mutation operations. After multiple generations of iterative evolution, the bias distribution parameter and the excitation weight parameter are selected from the Pareto optimal frontier solution set.

[0033] In specific implementation, a fitness function is constructed, which includes maximizing the equivalent array gain, minimizing the antenna port coupling power, and minimizing the array gain deviation between different operating frequency bands. These three objectives are simultaneously incorporated into the evaluation framework of the fitness function through weighted combination or Pareto dominance ranking. In each generation of population evolution, the fitness function calculates a comprehensive fitness value or non-dominated level for each solution individual to guide the population towards the global optimum. After the fitness function is constructed, the bias distribution parameters and the excitation weight parameters are used as variables to be optimized and combined into a single solution vector. Under the constraints of the upper limit of the total transmit power of the antenna array and the dynamic range of the bias voltage of the metamaterial decoupling layer, an initial solution set is generated. The bias distribution parameters are the set of bias voltage values ​​corresponding to each liquid crystal unit in the metamaterial decoupling layer, and their dimension is consistent with the array size of the liquid crystal unit in the metamaterial decoupling layer. The excitation weight parameters are the set of complex feed coefficients of each antenna unit in the antenna array, including the excitation amplitude and excitation phase of each unit. These two sets of parameters are concatenated in a preset order to form a single solution vector. All elements in this solution vector constitute a complete candidate solution. When generating the initial solution set, two constraints must be satisfied simultaneously: first, the upper limit constraint on the total transmit power of the antenna array, meaning the sum of the squares of the excitation amplitudes of all antenna elements must not exceed the rated total power of the RF front-end power amplifier; second, the dynamic range constraint of the bias voltage of the metamaterial decoupling layer, meaning the bias voltage of each liquid crystal cell must be between the minimum and maximum output voltage of the driving circuit, and should be within the actual responsive threshold and saturation voltage range of the liquid crystal material. Both of these constraints are inherent physical limitations of the communication system hardware; violating either constraint will prevent the optimization results from being physically realized. Under these constraints, an initial solution set containing multiple individual solutions is generated through random sampling. This initial solution set is the starting population for the genetic algorithm's evolution.

[0034] The initial solution set undergoes selection, crossover, and mutation operations. After multiple generations of iterative evolution, the bias distribution parameters and the excitation weight parameters are selected from the Pareto optimal front solution set. In each generation of evolution, each individual solution in the current solution set is evaluated according to the fitness function. Individuals with better performance are retained for the next generation. The diversity of the solution set is maintained by information recombination and random perturbation of the solution vectors. The overall fitness level of the population gradually increases with the number of generations and converges towards the Pareto optimal front. After multiple generations of iterative evolution reaching the preset termination condition, the Pareto optimal front solution set is extracted from the final evolved population. This front solution set consists of all non-dominated individuals in the population, meaning that no individual is superior to another individual in at least one of the three optimization objectives without being inferior to that individual in the other objectives. One solution is selected from the frontier solution set as the final output solution. The selection strategy can be based on the trade-off requirements between equivalent array gain, port isolation and inter-band gain deviation in the actual deployment scenario. The decision method based on ideal point distance or the weighted ranking method based on engineering preferences can be adopted to select the combination of bias distribution parameters and excitation weight parameters that best meets the current communication quality requirements.

[0035] In one embodiment, after calculating the bias distribution parameters and excitation weight parameters, the method further includes: The bias distribution parameters are encoded into metamaterial decoupling layer control commands, and the metamaterial decoupling layer control commands are sent to the liquid crystal metamaterial control circuit. The excitation weight parameters are encoded into antenna excitation control commands, and the antenna excitation control commands are sent to the digital beamformer. Based on the response delay of the liquid crystal metamaterial control circuit and the response delay of the digital beamformer, a corresponding trigger offset is set. After confirming that both the liquid crystal metamaterial control circuit and the digital beamformer have received the corresponding control commands, the dielectric constant distribution modulation action and the radiation pattern modulation action are synchronously triggered according to the trigger offset.

[0036] In practical implementation, after calculating the bias distribution parameters and excitation weight parameters in step S3, and before actually applying these parameters to the metamaterial decoupling layer and antenna array in step S4, a set of control command encoding, distribution, and timing synchronization operations need to be performed. In this implementation, the bias distribution parameters exist in numerical form as optimized variable values. These parameters need to be encapsulated and encoded according to the instruction frame format specified by the liquid crystal metamaterial control circuit. The bias voltage value of each liquid crystal cell is mapped to the corresponding field in the instruction frame, and auxiliary information necessary for the communication protocol, such as device address and checksum, is added to generate complete metamaterial decoupling layer control instructions. After encoding, the control instructions are sent to the liquid crystal metamaterial control circuit via the control bus. After correctly receiving and parsing the instructions, the liquid crystal metamaterial control circuit stores the bias voltage value carried in the instructions in its internal register, and outputs it to each liquid crystal cell after the trigger signal arrives. Simultaneously, the excitation weight parameters are encoded into antenna excitation control instructions in parallel. The excitation weight parameters include the feed amplitude coefficient and feed phase coefficient of each antenna element. These need to be encoded into binary control words according to the register mapping rules of the digital beamformer and assembled into antenna excitation control commands, which are then sent to the digital beamformer. The digital beamformer loads the coefficient values ​​of each channel into the corresponding complex multiplier register, and waits for the trigger signal to arrive before taking effect.

[0037] After both sets of control commands have been issued, the response delays of the liquid crystal metamaterial control circuit and the digital beamformer differ by orders of magnitude. The response delay of the liquid crystal metamaterial control circuit mainly stems from the physical deflection process of the liquid crystal molecules. After applying a bias voltage, the liquid crystal molecules need a certain amount of time to rotate from their current orientation state to the target orientation state; this response delay is typically on the order of milliseconds. In contrast, the digital beamformer, as an all-electronic device, has a response delay on the order of nanoseconds from receiving the control command to updating the register value and then to the complex multiplier output signal. If a trigger signal is given simultaneously after issuing both sets of commands, the digital beamformer will update the excitation weights and change the radiation pattern of the antenna array within nanoseconds. However, at this time, the dielectric constant distribution of the liquid crystal metamaterial layer is still in a slowly changing transitional state and has not yet reached the target gradient distribution. This causes the antenna array to radiate in a near-field environment where the dielectric constant distribution is unstable, resulting in beam pointing and sidelobe characteristics that are significantly different from the expected optimized design. To address this timing mismatch issue, this embodiment sets corresponding trigger offsets for both the response delay of the liquid crystal metamaterial control circuit and the response delay of the digital beamformer. Specifically, the trigger signal of the digital beamformer is delayed relative to the trigger signal of the liquid crystal metamaterial control circuit by a preset offset time. This offset time is the difference between the response delays of the liquid crystal metamaterial control circuit and the digital beamformer. This ensures that the moment when the liquid crystal molecules in the liquid crystal metamaterial layer complete deflection and the non-uniform dielectric constant distribution stabilizes is precisely aligned with the moment when the digital beamformer updates its excitation weights and the antenna array radiation pattern changes. After confirming that both have correctly received the corresponding control commands and completed loading of their internal registers, a trigger signal is issued according to the set trigger offset, synchronously triggering the dielectric constant distribution adjustment action and the radiation pattern adjustment action. Here, synchronous triggering refers to the alignment of the physical effects of the two actions after time delay compensation, rather than their starting at the same absolute time.

[0038] In one embodiment, step S4, which involves applying the bias distribution parameters to the metamaterial decoupling layer to adjust the dielectric constant in front of the antenna array's radiating aperture to exhibit a non-uniform gradient distribution, includes: The bias distribution parameters are converted into bias voltage values ​​for each liquid crystal cell in the metamaterial decoupling layer, wherein each liquid crystal cell is arranged in an array. The bias voltage value is output to the corresponding liquid crystal cell through a multiplexer digital-to-analog converter. The deflection state of the molecules inside each liquid crystal cell is controlled by adjusting the amplitude of the bias voltage. The molecules maintain their initial alignment state when the bias voltage value is below a threshold voltage, and begin to deflect after the bias voltage value reaches the threshold voltage. The deflection angle increases with the increase of the bias voltage value and tends to stabilize after the bias voltage value reaches the saturation voltage. By adjusting the bias voltage amplitude of each liquid crystal cell, a spatial gradient distribution with different dielectric constant values ​​is formed in the metamaterial decoupling layer between the central region and the edge region.

[0039] In practical implementation, the metamaterial decoupling layer consists of multiple liquid crystal cells arranged in an array. Each liquid crystal cell physically contains a layer of nematic liquid crystal material, with transparent electrodes deposited on its upper and lower substrates. By applying a voltage between the electrodes, the orientation state of the liquid crystal molecules can be controlled, thereby changing the equivalent dielectric constant of the cell. The bias voltage distribution parameters are given in numerical form in step S3, with dimensions consistent with the size of the liquid crystal cell array. Each value corresponds to the target bias voltage value of a liquid crystal cell. According to the row and column index mapping relationship of the array, the bias voltage parameters of each cell are extracted one by one and converted into the corresponding bias voltage values, which are used as the input for subsequent digital-to-analog conversion. The multiplexer includes multiple independent channels, each corresponding to a liquid crystal cell, which can accurately convert the digitally represented voltage value into an analog voltage signal and apply it to the electrodes of the liquid crystal cell.

[0040] There is a nonlinear dependence between the deflection state of liquid crystal molecules and the applied bias voltage: when the bias voltage is below the threshold voltage of the liquid crystal material, the electric field torque on the liquid crystal molecules is insufficient to overcome the elastic restoring torque and surface anchoring energy between molecules, and the molecules maintain their initial alignment, with the equivalent dielectric constant of the unit remaining at its initial value; when the bias voltage reaches the threshold voltage, the molecules begin to deflect, and the deflection angle gradually increases with the increase of the bias voltage, and the equivalent dielectric constant of the unit also changes continuously from its initial value towards the target direction; when the bias voltage is further increased to the saturation voltage, the molecular deflection angle reaches its physical limit, and even if the voltage continues to increase thereafter, the deflection angle tends to stabilize, and the equivalent dielectric constant also stabilizes at the saturation value. This continuously adjustable range from the threshold to saturation constitutes the effective working range for dynamic control of the equivalent dielectric constant of the liquid crystal unit. The threshold voltage refers to the minimum applied voltage required for the liquid crystal molecules to undergo a perceptible deflection from their initial alignment state. Its value depends on the dielectric anisotropy, elastic constant, and thickness of the liquid crystal material. The saturation voltage refers to the applied voltage at which the liquid crystal molecules reach their physical limit and the deflection angle no longer changes significantly with further increases in voltage. Its value is also determined by the physical properties of the liquid crystal material. In typical nematic liquid crystal materials, the threshold voltage is usually between 1V and 3V, the saturation voltage is usually between 10V and 20V, and the equivalent dielectric constant varies continuously between 2.5 and 12. This embodiment achieves independent and continuous adjustment of the equivalent dielectric constant of each liquid crystal cell by precisely controlling the bias voltage of each cell between the threshold voltage and the saturation voltage.

[0041] By adjusting the bias voltage amplitude of each liquid crystal cell, the equivalent dielectric constant at different spatial locations in the metamaterial decoupling layer exhibits a predetermined spatial gradient distribution. Specifically, a relatively low bias voltage is applied to the liquid crystal cells located in the central region of the metamaterial decoupling layer, resulting in a larger equivalent dielectric constant; a relatively high bias voltage is applied to the liquid crystal cells located in the edge region of the metamaterial decoupling layer, resulting in a smaller equivalent dielectric constant, thus forming a spatial gradient of dielectric constant that decreases from the center to the edge. This non-uniform gradient dielectric constant distribution alters the propagation boundary conditions of surface waves on the plane where the metamaterial decoupling layer is located, effectively suppressing coupling waves propagating along the surface of the dielectric substrate between antenna elements, reducing the mutual coupling strength between antenna ports, improving port isolation, and providing a purified near-field electromagnetic environment for beamforming of the antenna array.

[0042] In one embodiment, the antenna array is a dual-frequency reconfigurable antenna array, and each antenna element integrates a tunable element. Prior to the step of adjusting the beam pointing of the radiation pattern, the method further includes: Obtain the operating frequency band indication signal currently allocated by the communication system; According to the operating frequency band indication signal, the resonant frequency of each antenna element is adjusted by the tunable element, and the antenna array is switched to the corresponding target frequency band for operation; The operating frequency band indication signal is used as the input parameter of the multi-objective genetic algorithm. The operating frequency band indication signal is used to calculate the frequency band characteristics of the current antenna array when the bias distribution parameters and the excitation weight parameters are matched.

[0043] In this implementation, the antenna array is a dual-band reconfigurable antenna array, with each antenna element integrating a tunable component. This embodiment addresses the difficulty of dynamic coordination and adaptation between Sub-6GHz and millimeter-wave bands in existing solutions by using electrically controlled tunable components to dynamically switch the antenna resonant frequency and synchronously incorporates frequency band information into the optimization process, achieving adaptive matching across frequency bands. Specifically, the base station or access point dynamically allocates one or more operating frequency bands to the current communication link based on network load, user terminal capabilities, channel quality, and spectrum resource allocation strategies, extracting the operating frequency band indication signal from the communication protocol stack through the control channel or upper-layer signaling interface. The tunable component can be a PIN diode, varactor diode, or microelectromechanical switch, etc., with electrically controllable adjustable characteristics. These components are integrated into the radiating structure or feed network of each antenna element. By controlling the state of the tunable component, the electrical length of the parasitic structure or extension arm loaded on the antenna element is changed, thereby altering the resonant frequency of the antenna element. Through this frequency band switching operation, the antenna array can flexibly switch between different frequency bands without replacing the physical antenna.

[0044] Using the operating frequency band indication signal as the input parameter for the multi-objective genetic algorithm is problematic because different operating frequency bands correspond to different free-space wavelengths. The electrical dimensions of each element in the antenna array, and the electrical distance between elements (measured in wavelength), all change with the operating wavelength. This leads to significant differences in the radiation pattern characteristics, inter-element coupling strength, and array manifold matrix of the antenna array across different frequency bands. If the multi-objective genetic algorithm fails to know the actual target frequency band of the current antenna array when solving for the bias distribution parameters and excitation weight parameters, its constructed fitness function and array manifold matrix model will be calculated based on incorrect frequency assumptions. Consequently, the obtained bias distribution parameters will be unable to effectively suppress coupling in the corresponding frequency band, and the excitation weight parameters will be unable to accurately form the beam pointing in that frequency band. Therefore, in this embodiment, after the aforementioned frequency band switching operation is completed, the operating frequency band indication signal is fed back to the input interface of the multi-objective genetic algorithm. This allows the algorithm to load the analytical model or numerical interpolation table of the antenna element radiation pattern template, inter-element coupling coefficient matrix, and array manifold matrix corresponding to the current frequency band characteristics of the antenna array in the next optimization cycle. This ensures that the synchronously calculated bias distribution parameters and excitation weight parameters can accurately match the actual operating frequency band of the current antenna array. During system operation, if the system detects a change in the operating frequency band after step S4 of the current cycle is executed, the new operating frequency band indication signal will be incorporated into the optimization constraints in step S3 of the next cycle, achieving continuous adaptation under multi-frequency band operating conditions.

[0045] In one embodiment, the step of adjusting the beam pointing of the radiation pattern includes: The excitation weighting parameters are decomposed into the feed amplitude coefficient and feed phase coefficient of each antenna element; The feed amplitude coefficient and the feed phase coefficient are written into the complex multiplier of each antenna element in the digital beamformer; By exciting each antenna element according to its corresponding feed amplitude coefficient and feed phase coefficient, the radiation pattern beam of the antenna array is controlled to point towards the target user terminal.

[0046] In specific implementation, the excitation weight parameters are output as a complete complex vector in step S3. Each complex element in this vector corresponds to an antenna element in the antenna array. The modulus of the complex number represents the feed amplitude of the element, and the argument of the complex number represents the feed phase of the element. By traversing each element in the complex vector, the feed amplitude coefficient of each antenna element is extracted through modulus operation, and the feed phase coefficient of each antenna element is extracted through argument operation, thereby decomposing the excitation weight parameters into two independent control quantities: amplitude dimension and phase dimension. The digital beamformer is located between the baseband processing unit and the RF front end. It integrates a number of parallel complex multiplier channels equal to the number of antenna elements. Each channel is responsible for independently performing amplitude weighting and phase rotation processing on the signal of one antenna element. According to the index order of the antenna elements, the feed amplitude coefficient of each antenna element is written into the amplitude register of the corresponding channel in the form of a binary control word, and the feed phase coefficient of the element is written into the phase register of the corresponding channel in the form of a phase control word. Each antenna element uses the amplitude-weighted and phase-rotated signal output by the digital beamformer as an excitation source to radiate electromagnetic waves into space. The electromagnetic waves radiated by each element interfere and superimpose with each other in space, forming constructive interference in the target direction to generate a high-gain main lobe, and forming destructive interference in the non-target direction to suppress the sidelobe level. By precisely setting the excitation weight of each antenna element, the direction of the maximum value of the main lobe of the radiation pattern is precisely aligned with the spatial angle of the target user terminal, thus completing the beam pointing control.

[0047] In one embodiment, after the beam pointing of the radiation pattern is adjusted and the antenna array enters a stable operating state, the method further includes: Obtain the port isolation and beam pointing deviation generated during the actual operation of the regulated antenna array; If the port isolation deviates from the preset threshold range, the port isolation is used as a feedback signal to calibrate the fitness function of the multi-objective genetic algorithm online. If the beam pointing deviation deviates from the preset threshold range, the beam pointing deviation is used as a feedback signal to correct the network weight parameters of the time series prediction model online, and the corrected network weight parameters are reloaded into the time series prediction model for model update.

[0048] In its implementation, this embodiment addresses the issues of beam misalignment and degraded decoupling effects caused by dynamic changes in channel characteristics and long-term drift of metamaterial hardware characteristics in high-speed mobile scenarios. It achieves long-term adaptive and stable system operation by calibrating the optimization algorithm and prediction model through dual independent feedback channels. Specifically, performance data acquisition must be performed after confirming that the antenna array has entered a stable operating state, rather than immediately after the control action is triggered. This is because the deflection of liquid crystal molecules in the metamaterial decoupling layer is a physical relaxation process with a response delay on the order of milliseconds. After the bias voltage is applied, the liquid crystal molecules need a transition phase to gradually deflect from the initial alignment state to the target orientation state, thereby allowing the dielectric constant distribution of the metamaterial decoupling layer to evolve from the pre-control distribution to the target gradient distribution and tend to stabilize. If port isolation and beam pointing deviation are acquired before this transition phase is completed, the obtained data reflects the system performance in an intermediate state where the dielectric constant distribution is not yet in place, rather than the steady-state performance that the system should achieve after the control action is completed. Using this invalid data as the basis for subsequent feedback correction will introduce systematic errors. Therefore, in this embodiment, after triggering the control action, a waiting period is set according to the response delay of the liquid crystal metamaterial control circuit. After this waiting period ends and the antenna array enters a stable working state, the parameters between each antenna port are obtained through the coupling measurement circuit, and the port isolation index is extracted from it. At the same time, the beam pointing monitoring module compares the difference between the actual pointing angle of the main lobe of the current radiation pattern and the spatial angle of the target user terminal to calculate the beam pointing deviation. The preset threshold ranges are set for both port isolation and beam pointing deviation. The preset threshold range for port isolation is determined based on the system design target value and takes into account hardware drift tolerance. The preset threshold range for beam pointing deviation is determined based on the beamwidth and the angular resolution corresponding to the operating frequency band.

[0049] Based on the deviations of port isolation and beam pointing deviation, two independent and non-overlapping feedback correction channels are activated. The first feedback channel processes the deviation of port isolation. If the obtained port isolation deviates from a preset threshold range, the port isolation is used as a feedback signal to calibrate the fitness function of the multi-objective genetic algorithm online. The deviation of port isolation typically stems from device parameter drift that may occur in the metamaterial decoupling layer during long-term operation. Examples include dielectric anisotropy changes in liquid crystal materials after prolonged bias application, charge accumulation effects of transparent electrodes, and shifts in liquid crystal viscosity coefficient and threshold voltage caused by changes in ambient temperature. These physical variations result in differences in the actual dielectric constant gradient distribution generated by the same bias distribution parameters at different operating stages. The root cause of this error lies in the changes in the physical properties of the metamaterial decoupling layer, rather than the modeling deviation of the channel prediction model in the wireless propagation environment. Therefore, its correction operation should be applied to the optimization stage of the multi-objective genetic algorithm. Specifically, the deviation between the current measured port isolation and the target isolation is used as a correction factor to dynamically adjust the weight coefficients or penalty terms in the fitness function that are related to minimizing port coupling power. This makes the multi-objective genetic algorithm tend to select bias distribution parameters with a larger margin in the port isolation dimension in subsequent optimization solutions, in order to compensate for the impact of the slow changes in metamaterial hardware properties on the decoupling effect.

[0050] The second feedback channel handles beam pointing deviation. If the acquired beam pointing deviation deviates from a preset threshold range, it is used as a feedback signal to correct the network weight parameters of the time series prediction model online. The corrected network weight parameters are then reloaded into the time series prediction model for model updates. Beam pointing deviation primarily arises from changes in the statistical characteristics of the wireless propagation environment that exceed the current modeling capabilities of the prediction model. Examples include sudden changes in user terminal movement speed, redistribution of channel multipath structure due to changes in environmental scatterers, or drift in channel coherence time due to atmospheric conditions. This leads to a systematic deviation between the channel evolution patterns learned by the time series prediction model based on historical training data and the actual time-varying patterns of the current channel. The root cause of this error lies in the prediction model's lagging modeling of the channel's time-varying dynamics; therefore, its correction operation should directly apply to the prediction model itself. In practice, the beam pointing deviation is used as an error metric signal. This is used to construct the increment of the loss function between the prediction model output and the actual channel state. The network weight parameters of the time series prediction model are incrementally adjusted using a small-batch online gradient descent method, allowing the model to gradually adapt to changes in channel statistical characteristics. After the correction is completed, the corrected network weight parameters are reloaded into the channel prediction module for use in the next prediction and optimization cycle, thus forming a closed-loop adaptive tuning from beam execution effect to prediction model parameters.

[0051] Reference Figure 2 Here is a structural block diagram of a MIMO antenna array beam control system based on metamaterials and predictive optimization in one embodiment of the present invention, comprising: The channel acquisition module is used to adjust the sampling frequency according to the current channel change rate, acquire the multi-antenna channel state information matrix in real time, and preprocess the acquired channel state information matrix to obtain the preprocessed channel state information matrix. The channel prediction module is used to input the preprocessed channel state information matrix within a preset time window into the trained time series prediction model to obtain the predicted channel matrix for a preset future duration. The joint calculation module is used to simultaneously calculate the bias distribution parameters of the metamaterial decoupling layer and the excitation weight parameters of the antenna array using the predicted channel matrix as input and a multi-objective genetic algorithm. The optimization objectives of the calculation include maximizing the equivalent array gain, minimizing the port coupling power, and minimizing the gain deviation between frequency bands. The execution control module is used to apply the bias distribution parameters to the metamaterial decoupling layer, adjust the dielectric constant in front of the antenna array radiation aperture to form a non-uniform gradient distribution, and apply the excitation weight parameters to the antenna array to adjust the beam pointing of the radiation pattern.

[0052] In this embodiment, addressing the issues of drastically deteriorating decoupling performance of existing decoupling schemes when the element spacing of traditional MIMO antenna arrays is reduced to below half a wavelength, and the inability to dynamically adjust decoupling characteristics with beam scanning and frequency band switching, as well as insufficient port isolation, a hardware-software co-engineered system architecture is designed. The core utilizes a gradient-tunable liquid crystal metamaterial decoupling layer to achieve dynamic decoupling, combined with channel prediction and joint optimization algorithms to achieve precise beam control. The MIMO antenna array beam control system based on metamaterials and prediction optimization includes a dual-frequency reconfigurable MIMO antenna array, a gradient-tunable liquid crystal metamaterial decoupling layer, an RF front-end module, and a baseband processing unit (integrating a channel acquisition module, a channel prediction module, a joint calculation module, and an execution control module). The gradient-tunable liquid crystal metamaterial decoupling layer is the core decoupling functional component. The metamaterial decoupling layer is composed of 16×16 arrayed liquid crystal cells, each with a size of λ / 8×λ / 8 (based on the 28GHz millimeter-wave band, corresponding to a physical size of 3.57mm×3.57mm), covering the entire radiating aperture of the antenna array. The decoupling layer is positioned λ / 4 in front of the antenna array's radiating aperture. The upper and lower substrates are low-loss microwave dielectric substrates, with ITO transparent electrodes deposited on the inner side of the substrates. Each liquid crystal cell corresponds to a set of independent electrodes, forming an independent electrically controlled tuning channel. The liquid crystal cells are filled with nematic liquid crystal material, whose dielectric constant can be continuously controlled by the bias voltage between the electrodes, with an effective control range of 2.5-12. The threshold voltage of the liquid crystal material is 1-3V, and the saturation voltage is 10-20V. Between the threshold voltage and the saturation voltage, the deflection angle of the liquid crystal molecules increases with the increase of the bias voltage amplitude, and the equivalent dielectric constant changes continuously with the deflection angle. Beyond this range, the dielectric constant tends to stabilize. The metamaterial decoupling layer is equipped with a liquid crystal metamaterial control circuit and a multi-channel digital-to-analog converter (DAC). The number of DAC channels corresponds one-to-one with the number of liquid crystal cells, converting digital bias distribution parameters into analog bias voltages of 0-20V, which are independently applied to the electrodes of each liquid crystal cell. The liquid crystal metamaterial control circuit communicates with the execution control module via a control bus, receiving control commands and completing the output and latching of bias voltages. By adjusting the bias voltage of each liquid crystal cell, a spatial gradient distribution of dielectric constants with different dielectric constants is formed in the metamaterial decoupling layer between the central and edge regions. This changes the electromagnetic boundary conditions of the antenna array aperture surface, suppresses surface waves propagating along the surface of the dielectric substrate between antenna cells, and blocks near-field coupling paths between cells, thereby significantly improving the isolation between adjacent antenna ports. Simultaneously, the dielectric constant gradient distribution can be dynamically adjusted according to the current operating frequency band and beam direction, achieving adaptive decoupling across the entire beam scanning angle and operating frequency band.

[0053] In summary, this invention adjusts the sampling frequency according to the current channel change rate, collects multi-antenna channel state information matrices in real time, and preprocesses the collected channel state information matrices to obtain preprocessed channel state information matrices. The preprocessed channel state information matrices within a preset time window are input to a trained time series prediction model to obtain a predicted channel matrix for a preset future duration. Using the predicted channel matrix as input, a multi-objective genetic algorithm is used to calculate the bias distribution parameters of the metamaterial decoupling layer and the excitation weight parameters of the antenna array. The optimization objectives include maximizing the equivalent array gain, minimizing port coupling power, and minimizing inter-band gain deviation. The bias distribution parameters are applied to the metamaterial decoupling layer to regulate the dielectric constant in front of the antenna array's radiating aperture to exhibit a non-uniform gradient distribution. The excitation weight parameters are then applied to the antenna array to regulate the beam pointing of the radiation pattern, achieving beam control with high isolation and high dynamic tracking accuracy.

[0054] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A beam control method for MIMO antenna arrays based on metamaterials and predictive optimization, characterized in that, Includes the following steps: The sampling frequency is adjusted according to the current channel change rate, and the multi-antenna channel state information matrix is ​​collected in real time. The collected channel state information matrix is ​​then preprocessed to obtain the preprocessed channel state information matrix. The preprocessed channel state information matrix within a preset time window is input into a trained time series prediction model to obtain a predicted channel matrix for a future preset duration. Using the predicted channel matrix as input, the bias distribution parameters of the metamaterial decoupling layer and the excitation weight parameters of the antenna array are calculated by a multi-objective genetic algorithm. The optimization objectives of the calculation include maximizing the equivalent array gain, minimizing the port coupling power, and minimizing the gain deviation between frequency bands. The bias distribution parameter is applied to the metamaterial decoupling layer to adjust the dielectric constant in front of the antenna array's radiation aperture to have a non-uniform gradient distribution, and the excitation weight parameter is applied to the antenna array to adjust the beam pointing of the radiation pattern.

2. The MIMO antenna array beam control method based on metamaterials and predictive optimization according to claim 1, characterized in that, The steps of adjusting the sampling frequency according to the current channel change rate, acquiring a multi-antenna channel state information matrix in real time, and preprocessing the acquired channel state information matrix to obtain a preprocessed channel state information matrix include: Obtain the user terminal's movement speed parameters in the current communication environment; The minimum threshold of the sampling frequency is determined based on the moving speed parameter. The sampling frequency corresponding to the minimum threshold is higher than the rate of change of the channel state information under multipath time-varying conditions, and the effective sample size collected within the preset time window meets the input requirements of the time series prediction model. The channel state information corresponding to each antenna unit is collected through the analog-to-digital conversion circuit of the radio frequency front end at a sampling frequency not lower than the minimum threshold, forming a channel state information matrix. The channel state information matrix is ​​subjected to noise cancellation processing; The channel state information matrix after noise cancellation is subjected to phase calibration to obtain a preprocessed channel state information matrix.

3. The MIMO antenna array beam control method based on metamaterials and predictive optimization according to claim 1, characterized in that, The training steps of the trained time series prediction model include: Construct a training dataset containing historical channel state information sequences under different movement speed scenarios; The time series prediction model is subjected to offline supervised learning using the training dataset to learn the evolution of the channel state under multipath time-varying conditions. After training is completed, the time series prediction model is loaded into the channel prediction module. The time series prediction model is used to perform forward inference on the input preprocessed channel state information matrix and output the predicted channel matrix.

4. The MIMO antenna array beam control method based on metamaterials and predictive optimization according to claim 1, characterized in that, The step of calculating the bias distribution parameters of the metamaterial decoupling layer and the excitation weight parameters of the antenna array using a multi-objective genetic algorithm includes: Construct a fitness function, which includes maximizing the equivalent array gain, minimizing the antenna port coupling power, and minimizing the array gain deviation between different operating frequency bands; The bias distribution parameters and the excitation weight parameters are used as variables to be optimized and combined into the same solution vector. Under the constraints of the upper limit of the total transmit power of the antenna array and the dynamic range of the bias voltage of the metamaterial decoupling layer, an initial solution set is generated. The initial solution set is subjected to selection, crossover, and mutation operations. After multiple generations of iterative evolution, the bias distribution parameter and the excitation weight parameter are selected from the Pareto optimal frontier solution set.

5. The MIMO antenna array beam control method based on metamaterials and predictive optimization according to claim 1, characterized in that, After calculating the bias voltage distribution parameters and excitation weight parameters, the following steps are also included: The bias distribution parameters are encoded into metamaterial decoupling layer control commands, and the metamaterial decoupling layer control commands are sent to the liquid crystal metamaterial control circuit. The excitation weight parameters are encoded into antenna excitation control commands, and the antenna excitation control commands are sent to the digital beamformer. Based on the response delay of the liquid crystal metamaterial control circuit and the response delay of the digital beamformer, a corresponding trigger offset is set. After confirming that both the liquid crystal metamaterial control circuit and the digital beamformer have received the corresponding control commands, the dielectric constant distribution modulation action and the radiation pattern modulation action are synchronously triggered according to the trigger offset.

6. The MIMO antenna array beam control method based on metamaterials and predictive optimization according to claim 1, characterized in that, The step of applying the bias distribution parameters to the metamaterial decoupling layer to adjust the dielectric constant in front of the antenna array radiating aperture to exhibit a non-uniform gradient distribution includes: The bias distribution parameters are converted into bias voltage values ​​for each liquid crystal cell in the metamaterial decoupling layer, wherein each liquid crystal cell is arranged in an array. The bias voltage value is output to the corresponding liquid crystal cell through a multiplexer digital-to-analog converter. The deflection state of the molecules inside each liquid crystal cell is controlled by adjusting the amplitude of the bias voltage. The molecules maintain their initial alignment state when the bias voltage value is below a threshold voltage, and begin to deflect after the bias voltage value reaches the threshold voltage. The deflection angle increases with the increase of the bias voltage value and tends to stabilize after the bias voltage value reaches the saturation voltage. By adjusting the bias voltage amplitude of each liquid crystal cell, a spatial gradient distribution with different dielectric constant values ​​is formed in the metamaterial decoupling layer between the central region and the edge region.

7. The MIMO antenna array beam control method based on metamaterials and predictive optimization according to claim 1, characterized in that, The antenna array is a dual-frequency reconfigurable antenna array, and each antenna element integrates a tunable element. Before the step of adjusting the beam pointing of the radiation pattern, the method further includes: Obtain the operating frequency band indication signal currently allocated by the communication system; According to the operating frequency band indication signal, the resonant frequency of each antenna element is adjusted by the tunable element, and the antenna array is switched to the corresponding target frequency band for operation; The operating frequency band indication signal is used as the input parameter of the multi-objective genetic algorithm. The operating frequency band indication signal is used to calculate the frequency band characteristics of the current antenna array when the bias distribution parameters and the excitation weight parameters are matched.

8. The MIMO antenna array beam control method based on metamaterials and predictive optimization according to claim 1, characterized in that, The step of adjusting the beam pointing of the radiation pattern includes: The excitation weighting parameters are decomposed into the feed amplitude coefficient and feed phase coefficient of each antenna element; The feed amplitude coefficient and the feed phase coefficient are written into the complex multiplier of each antenna element in the digital beamformer; By exciting each antenna element according to its corresponding feed amplitude coefficient and feed phase coefficient, the radiation pattern beam of the antenna array is controlled to point towards the target user terminal.

9. The MIMO antenna array beam control method based on metamaterials and predictive optimization according to claim 1, characterized in that, After the beam pointing of the radiation pattern is adjusted and the antenna array enters a stable operating state, the following steps are also included: Obtain the port isolation and beam pointing deviation generated during the actual operation of the regulated antenna array; If the port isolation deviates from the preset threshold range, the port isolation is used as a feedback signal to calibrate the fitness function of the multi-objective genetic algorithm online. If the beam pointing deviation deviates from the preset threshold range, the beam pointing deviation is used as a feedback signal to correct the network weight parameters of the time series prediction model online, and the corrected network weight parameters are reloaded into the time series prediction model for model update.

10. A MIMO antenna array beam control system based on metamaterials and predictive optimization, characterized in that, include: The channel acquisition module is used to adjust the sampling frequency according to the current channel change rate, acquire the multi-antenna channel state information matrix in real time, and preprocess the acquired channel state information matrix to obtain the preprocessed channel state information matrix. The channel prediction module is used to input the preprocessed channel state information matrix within a preset time window into the trained time series prediction model to obtain the predicted channel matrix for a preset future duration. The joint calculation module is used to simultaneously calculate the bias distribution parameters of the metamaterial decoupling layer and the excitation weight parameters of the antenna array using the predicted channel matrix as input and a multi-objective genetic algorithm. The optimization objectives of the calculation include maximizing the equivalent array gain, minimizing the port coupling power, and minimizing the gain deviation between frequency bands. The execution control module is used to apply the bias distribution parameters to the metamaterial decoupling layer, adjust the dielectric constant in front of the antenna array radiation aperture to form a non-uniform gradient distribution, and apply the excitation weight parameters to the antenna array to adjust the beam pointing of the radiation pattern.

Images