Beamforming prediction method and related equipment for RIS-assisted sensor-integrated vehicle networking

Abstract

This application provides a beamforming prediction method and related equipment for a RIS-assisted integrated sensing vehicle-to-everything (V2X) network, belonging to the fields of intelligent transportation and 6G communication technology. This application proposes a two-stage prediction scheme: In the first stage, a capacitive Kalman filter algorithm is used to fuse a long short-term memory network and a multilayer perceptron model to predict the beam departure angle between the reconfigurable smart surface and the vehicle in the next time slot, thereby reconstructing the channel state information; In the second stage, based on the predicted channel state information, a soft actor-commentator reinforcement learning algorithm is used to jointly optimize the base station beamforming matrix and the reconfigurable smart surface phase shift matrix to maximize the total system throughput. This application achieves signal multiplexing through an integrated communication and sensing architecture, reducing pilot overhead; improves channel prediction accuracy through algorithm fusion; and achieves dynamic resource optimization through reinforcement learning, significantly improving the communication reliability and throughput performance of millimeter-wave V2X networks in high-speed mobile and non-line-of-sight scenarios.

Description

Technical Field

[0001] This application relates to the technical field of cross-integration of intelligent transportation systems and sixth-generation mobile communication technology, and in particular to a beamforming prediction method and related equipment for RIS-assisted integrated sensing vehicle networking. Background Technology

[0002] In current vehicle-to-everything (V2X) communication scenarios, factors such as the high-speed mobility of vehicles and road obstructions can easily lead to unstable communication links and severe signal attenuation in non-line-of-sight transmissions between base stations and vehicles. Traditional V2X communication and sensing functions are independent of each other, and sensing data and communication link status information cannot be used in synergy. This results in beamforming parameter adjustments lagging behind dynamic changes in the link, reducing the utilization rate of communication spectrum resources and increasing transmission latency.

[0003] Inductively coupled communication (ISAC) technology has become a key technology in B5G and 6G wireless networks due to its enormous potential in improving spectral efficiency. Existing research has extensively explored deep learning, channel tracking, and predictive beamforming. The first existing technical approach proposes using recurrent neural networks (LSTM+GNN) to design RIS reflection coefficients and base station beamforming matrices; the second proposes a beamforming prediction scheme for multi-user V2I networks based on extended Kalman filtering (EKF); and the third studies RIS-assisted single-user ISAC systems, employing a two-stage method or end-to-end neural networks for beamforming prediction.

[0004] However, existing technologies still have the following shortcomings: 1) There is relatively little research on beamforming prediction in V2I systems that combine RIS and ISAC, while beamforming prediction can effectively offset beam inaccuracy caused by channel time-varying and the high overhead of channel state information prediction; 2) Most existing beamforming prediction research is based on end-to-end neural networks, which has limited prediction accuracy in rapidly time-varying channel environments, making it difficult to guarantee the high reliability and low latency communication required by vehicle-to-everything (V2X) networks. Summary of the Invention

[0005] The main objective of this application is to propose a beamforming prediction method, electronic device, storage medium, and program product for a RIS-assisted integrated sensing vehicle network, in order to solve the technical problems in the prior art, such as the difficulty in efficiently integrating communication and sensing data, large prediction deviation of channel state information, and excessively complex beamforming optimization algorithms that are difficult to adapt to rapid time-varying channels.

[0006] To achieve the above objectives, one aspect of this application proposes a beamforming prediction method for a RIS-assisted integrated sensing vehicle network, the method comprising: Step S1: In the ISAC vehicle-to-everything (V2X) system assisted by the reconfigurable smart surface RIS, the base station acquires the beam angle status information between the RIS and the vehicle in historical time slots. Step S2: Based on the beam angle state information of the historical time slots, the capacitive Kalman filter (CKF) algorithm is used to fuse the long short-term memory (LSTM) network and the multilayer perceptron (MLP) model to predict the beam departure angle (AoDs) of the next time slot. Step S3: Based on the predicted AoDs of the next time slot, reconstruct the channel state information of the next time slot; Step S4: Based on the channel state information of the next time slot obtained from the reconstruction, the soft actor-commentator SAC reinforcement learning algorithm is used to jointly optimize and obtain the base station optimal beamforming matrix and RIS optimal phase shift matrix of the next time slot in order to maximize the total system throughput.

[0007] In some embodiments, step S2 includes: A dual-mode prediction module consisting of an MLP network and an LSTM network, and a fusion module based on the CKF algorithm were constructed. The MLP network is used to fit the state transition equation to propagate the volume points of the beam angle state vector; the LSTM network is used to generate the predicted AoDs vector based on historical beam angle information, which serves as the measurement update value for the CKF algorithm; the CKF algorithm is used to fuse the outputs of the MLP network and the LSTM network to obtain the final predicted AoDs for the next time slot.

[0008] In some embodiments, the fusion module based on the CKF algorithm performs the following steps: Volume point propagation steps: Generate volume points based on the beam angle state vector and its error covariance matrix of the current time slot, and propagate the volume points using the trained MLP network; State prediction steps: Take the mean of the volume points after propagation to obtain the predicted state vector and the prediction error covariance matrix; Measurement update steps: Use the output of the trained LSTM network as the actual observation value, and calculate the Kalman gain based on the predicted state vector; State fusion step: Based on the Kalman gain, the predicted state vector is fused with the output of the LSTM network to obtain the final predicted next time slot AoDs.

[0009] In some embodiments, step S4 specifically includes: Flatten the real and imaginary parts of the elements of the base station beamforming matrix and the RIS phase shift matrix to construct the action space of the SAC algorithm. The base station beamforming matrix, RIS phase shift matrix, power vectors allocated to each vehicle, and cascaded channel matrix of the current time slot are flattened and whitened to construct the state space of the SAC algorithm. The total downlink reachable throughput of the system is used as the immediate reward, and penalties are set for failing to meet the QoS constraints and SIC decoding requirements of continuous interference cancellation, thus constructing the reward function of the SAC algorithm. The optimal base station beamforming matrix and RIS phase shift matrix are output through iterative training of the policy network, value network, and target network of the SAC algorithm.

[0010] In some embodiments, the base station beamforming matrix and RIS phase shift matrix in the action space satisfy the maximum transmit power constraint of the base station and the discrete phase constraint of the RIS reflection unit; the phase codebook of the discrete phase constraint is preset according to the number of quantization bits.

[0011] In some embodiments, the system is an ISAC-based RIS-assisted millimeter-wave vehicle-to-everything (V2X) system in urban scenarios, comprising a fixed ISAC base station configured with an array of multiple transmitting antennas and multiple receiving antennas, K single-antenna vehicles, and a RIS deployed between the base station and the vehicles, containing N reflective units.

[0012] In some embodiments, the communication between the base station and the vehicle adopts a parallel transmission protocol: the base station continuously transmits downlink ISAC signals in the same time slot, which are reflected by the vehicle and used as sensing echo signals, and transmitted back to the base station via RIS; the base station extracts the beam angle information of the current time slot from the received sensing echo signals to predict the beam angle of the next time slot.

[0013] In some embodiments, the beam angle state information in step S1 is specifically the beam departure angle (AoDs) of the line-of-sight path between the RIS and the vehicle, including the azimuth angle and the elevation angle; when reconstructing the channel state information in step S3, the line-of-sight path channel is first calculated based on the predicted AoDs and the vehicle speed, and then the randomly generated non-line-of-sight path channel component is superimposed.

[0014] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0015] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0016] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0017] Compared with the prior art, this application has the following advantages: 1) System architecture innovation: A V2I system framework combined with ISAC with RIS assistance is proposed. Communication and sensing signals reuse the same time and frequency resources, reducing pilot overhead and providing a new solution for the construction of future wireless networks.

[0018] 2) Improved prediction accuracy: An MLP-LSTM dual-mode beam angle predictor with CKF algorithm is proposed. By fitting nonlinear state transitions through MLP network, generating measurement updates through LSTM network, and achieving optimal fusion through CKF algorithm, the beam angle prediction accuracy in fast time-varying channel environments is significantly improved, thereby reconstructing more accurate channel state information.

[0019] 3) Superior optimization performance: Based on the predicted channel state information, the SAC reinforcement learning algorithm is used to jointly optimize the base station beamforming matrix and the RIS phase shift matrix. The entropy regularization mechanism of the SAC algorithm gives it stronger exploration capabilities, enabling it to adapt to time-varying channel environments and effectively improve the system's achievable throughput.

[0020] 4) Low complexity and real-time performance: The two-stage method in this application decomposes the complex non-convex optimization problem into learnable prediction and optimization tasks, avoiding the high computational complexity of traditional convex optimization methods and meeting the real-time requirements of vehicle networking scenarios. Attached Figure Description

[0021] Figure 1 This is a diagram illustrating the system scenario considered in the embodiments of this application; Figure 2 This is a schematic diagram of the transmission protocol in the embodiments of this application; Figure 3 This is a framework diagram of the CKF algorithm fusion with LSTM and MLP in the embodiments of this application; Figure 4 This is a diagram illustrating the CKF algorithm fusion process in an embodiment of this application. Figure 5 This is a schematic diagram of the root mean square error of beam angle prediction for different vehicles in different time slots in the embodiments of this application; Figure 6 This is a comparison chart showing the difference between the total reachable rate of the system under different beam angle prediction schemes and the total reachable rate under an ideal channel in the embodiments of this application; Figure 7 This is a comparison chart of the average total achievable throughput of the system under different beamforming optimization schemes in the embodiments of this application; Figure 8 A flowchart of the beamforming prediction method for a RIS-assisted sensing integrated vehicle network provided in this application embodiment. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0024] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0025] 1) Reconfigurable Intelligent Surface (RIS) is a cutting-edge wireless communication technology based on programmable artificial electromagnetic materials. It can dynamically control the propagation environment of electromagnetic waves, thereby significantly improving the coverage, capacity and energy efficiency of communication systems.

[0026] 2) Integrated Sensing and Communication (ISAC) is a technology that integrates communication and sensing functions into the same wireless system, using the same spectrum, hardware and signal resources to achieve synergistic effects between information transmission and environmental perception.

[0027] 3) Vehicle-to-Infrastructure (V2I) is an important component of vehicle-to-everything (V2X) technology. It refers to the two-way information exchange between vehicles and road infrastructure through wireless communication to improve traffic safety, efficiency and intelligence.

[0028] The immense potential of ISAC technology in improving spectral efficiency makes it a promising technology for B5G and 6G wireless networks. Academia and industry have conducted extensive research on RIS-assisted ISAC systems, focusing on key technologies such as deep learning, channel tracking, and predictive beamforming.

[0029] The first existing technical solution considers a user-initiated RIS-assisted ISAC system and proposes a deep learning framework using recurrent neural networks. It uses the time-varying CSI obtained from the received pilot as the state vector and employs LSTM and GNN to design the uplink and downlink RIS reflection coefficients and the base station beamforming matrix. The second existing technical solution proposes a beamforming prediction scheme based on a multi-user V2I network for dual-function radar communication. In the considered ISAC system, an extended Kalman filter is used to track and predict the beam angles of multiple vehicles. Under given sum rate thresholds and power constraints, the power allocation strategy for multiple vehicles is optimized with the goal of minimizing the posterior Cramer-Rao bound of the angle estimation. The third existing technical solution studies a RIS-assisted single-user ISAC system and proposes two beamforming prediction methods with the goal of maximizing the user's received power: the first method is a two-stage method, first predicting beam angle information based on echoes, and then obtaining the beamforming matrix through maximum ratio transmission and manifold optimization techniques; the second method is an end-to-end prediction method, directly utilizing neural networks to predict the beamforming matrix based on historical echo information.

[0030] However, existing technical solutions have the following drawbacks: 1) Current research in this field mainly focuses on beam tracking in ISAC systems or beamforming optimization in RIS-assisted ISAC systems, but there is little research on beamforming prediction in V2I systems that combine RIS and ISAC. Beamforming prediction technology can effectively offset the beam inaccuracy caused by time-varying channels in V2I systems and the high overhead of estimating channel state information, thus ensuring high reliability and low latency communication.

[0031] 2) Currently, there is limited research on beamforming prediction in RIS-assisted ISAC systems, and most existing beamforming prediction studies are based on end-to-end neural networks, utilizing historical time slot information to predict the beamforming matrix. However, the prediction performance is limited in accuracy under rapidly changing channel conditions. Therefore, it is necessary to propose a better beamforming prediction method to ensure the acquisition of a better beamforming matrix under more accurate channel state information, thereby maximizing the downlink throughput of the system.

[0032] In view of this, this application embodiment considers a RIS-assisted millimeter-wave vehicle-to-everything (V2X) system based on ISAC, aiming to maximize the downlink communication throughput of the system. The core technical problems to be solved in achieving this goal include: First, under the ISAC architecture, communication and sensing data are difficult to integrate efficiently, making it impossible to accurately capture the rapidly changing channel characteristics caused by vehicle movement and dynamic road environment occlusion. This results in large prediction deviations and insufficient timeliness of channel state information, failing to provide reliable support for beamforming adjustments. Second, existing beamforming and RIS phase shift matrix optimization methods have significant drawbacks. Most of them focus on transforming the optimization problem into a convex optimization problem for solution. However, convex optimization modeling often requires simplifying assumptions about complex real-world scenarios, leading not only to excessively complex optimization algorithms that are difficult to adapt to the real-time requirements of vehicle movement scenarios, but also to optimization results that are usually suboptimal, failing to fully exploit the system's performance potential. To address the above issues, the solution relies on ISAC technology. First, the calculus Kalman filter (CKF) algorithm is fused with a long short-term memory (LSTM) network and a multilayer perceptron (MLP) to accurately predict the beam angle information of the next time slot and reconstruct accurate channel state information. Then, based on the predicted channel state information, the Soft Actor-Critic (SAC) reinforcement learning method is used to obtain the base station optimized beamforming and RIS phase shift matrix for the next time slot.

[0033] like Figure 8 As shown in the figure, this application provides a beamforming prediction method for a RIS-assisted sensor-integrated vehicle network, which specifically includes the following steps: Step S1: In the ISAC vehicle-to-everything (V2X) system assisted by the reconfigurable smart surface RIS, the base station acquires the beam angle status information between the RIS and the vehicle in historical time slots. Step S2: Based on the beam angle state information of the historical time slots, the capacitive Kalman filter (CKF) algorithm is used to fuse the long short-term memory (LSTM) network and the multilayer perceptron (MLP) model to predict the beam departure angle (AoDs) of the next time slot. Step S3: Based on the predicted AoDs of the next time slot, reconstruct the channel state information of the next time slot; Step S4: Based on the channel state information of the next time slot obtained from the reconstruction, the soft actor-commentator SAC reinforcement learning algorithm is used to jointly optimize and obtain the base station optimal beamforming matrix and RIS optimal phase shift matrix of the next time slot in order to maximize the total system throughput.

[0034] The solutions of the embodiments of this application will be described in detail below with reference to the accompanying drawings and specific application examples.

[0035] (1) System Model See Figure 1 This embodiment considers an ISAC-based RIS-assisted millimeter-wave vehicle-to-everything (V2X) system in an urban scenario to maximize the downlink communication speed. The system includes a configured... Meta transmitting antenna and Fixed ISAC base station with element receiving antenna array and K A single-antenna vehicle. In dense urban environments, due to building obstruction, a stable line-of-sight path cannot be established between the base station and the vehicle, resulting in severe damage to the wireless communication link. Therefore, a single-antenna vehicle is deployed between the base station and the vehicle. N The RIS of each reflector unit reconstructs the wireless channel between the base station and the vehicle, thereby establishing a stable communication transmission and sensing information interaction link.

[0036] 1.1) Channel Modeling Considering the dominant role of the direct path in millimeter-wave communication scenarios, a Ricean channel model is used to characterize the channel. Let... This indicates the number of reflection units in the RIS. and These represent the number of reflective units in the horizontal and vertical directions, respectively. Therefore, RIS is related to the base station and RIS is related to the... k The channels between each vehicle are modeled as follows: , (1) in, and These are respectively RIS and base station and RIS and the first k The path gain between vehicles can be uniformly expressed as , Indicates the carrier frequency. and The Rice attenuation components can be expressed as follows: , (2) in, This represents the Rice coefficient of the corresponding channel. and This represents the line-of-sight (LoS) component of the corresponding channel. and The steering vector representing the corresponding channel can be expressed as: (3) (4) in, and These represent the azimuth and elevation angles associated with RIS, respectively. It is the departure angle related to the base station. and These represent the non-line-of-sight (NLoS) components of the corresponding channel, each of which is an independent and identically distributed complex Gaussian random variable with zero mean and 1 variance. Furthermore, since the locations of the base station and the RIS are fixed, the BS-RIS transmission channel is quasi-static and can be pre-acquired before system operation.

[0037] 1.2) Transmission Protocol To reduce pilot overhead and acquire channel state information in dynamic scenarios based on sensing information, this embodiment adopts the following... Figure 2 The parallel transmission protocol shown can simultaneously support downlink communication and uplink channel state information prediction and sensing. (In time slots) t Taking -1 as an example, the base station continuously transmits downlink ISAC signals within this time slot. These signals, reflected by the vehicle, serve as a sensing and detection signal. Considering the round-trip transmission delay of the ISAC signal, the RIS receives the echo signal reflected by the vehicle and transmits it to the base station only a short time slot after the start of this time slot. Subsequently, the base station obtains information from the received reflected signal. t -1 time slot beam angle information for each vehicle, and use this angle information to predict t The time slot beam angle is calculated and the channel state information is reconstructed. Based on the predicted channel state information, the base station can pre-calculate the time slot beam angle and reconstruct the channel state information. t The optimized transmit beamforming matrix and RIS phase shift matrix of the time-slot base station can be used without channel state information prediction. t Time slots maximize downlink data transmission. The specific communication and sensing data transmission process is represented as follows: Data transmission model: The base station transmits a signal vector as follows: ,satisfy .set up The beamforming matrix for the base station is used, therefore the composite signal transmitted by the base station can be represented as follows: .in satisfy , This represents the maximum transmit power of the base station.

[0038] The RIS auxiliary base station transmits downlink communication signals to the vehicle, therefore the first k The signal received by each vehicle is: (5) in, To enable base stations to reach the first stage with RIS assistance k The equivalent channel for each vehicle. The RIS phase shift matrix in the downlink transmission link can be represented as: ,in and They represent the first n The amplitude coefficient and phase shift of each reflecting element This is the phase shift matrix of the RIS. To maximize the reflection energy of the RIS, let... The value is 1. In actual deployment and application scenarios, due to the constraints of hardware physical conditions, the phase values ​​of each reflection unit of the RIS are limited to a discrete finite set. Therefore, the codebook for the RIS phase shift is set to 1. , b This represents the number of quantization bits. Indicates the first k The additive white Gaussian noise (AWGN) received by each vehicle.

[0039] To reduce inter-user interference, Continuous Interference Cancellation (SIC) is used to decode the received signal. It is assumed that the equivalent channel gain satisfies... , No. k The communication rate of each vehicle can be calculated as follows: (6) For any two vehicles j and k satisfy , No. j The vehicle decodes the first signal before decoding its own signal. k For a given number of vehicles, the decoding rate can be expressed as: (7) To ensure successful SIC decoding, the decoding rate must meet the following requirements. Therefore, the total throughput of the system can be expressed as: .

[0040] Perception Model: In the system considered in this embodiment, vehicle perception is performed by the base station. Specifically, the ISAC signal sent by the base station first passes through RIS transmission before reaching the vehicle, and is then transmitted back to the base station along the same path after being transmitted by the vehicle. Therefore, the ISAC signal can be regarded as a detection signal used by the base station to estimate the beam angle, and the echo signal received at the base station is represented as: (8) in, This represents the channel coefficient of the reflected echo between the RIS and the vehicle. After receiving this echo signal, the base station can obtain the beam angle information between the RIS and the vehicle through channel estimation. This invention mainly studies the prediction of the beam angle, therefore it is assumed that the base station can accurately estimate the beam angle information between the RIS and the vehicle.

[0041] 1.3) Optimization Problem Based on the above analysis, this invention aims to jointly optimize the base station beamforming matrix. Downward phase shift matrix of RIS Maximize the total achievable throughput of the system. The mathematical expression for the optimization problem is (P1): (9) Constraint C1 guarantees the minimum achievable communication rate, constraint C2 represents the total transmit power limit of the base station, constraint C3 guarantees the smooth implementation of SIC decoding, and C4 represents the phase shift constraint of each RIS unit.

[0042] (2) Two-stage beamforming prediction method In the beamforming prediction problem under consideration, the ultimate goal is to obtain the optimal base station beamforming matrix and RIS phase shift matrix by solving a joint optimization problem (P1) based on the predicted equivalent channel state information between the base station and the vehicle. To this end, this embodiment proposes a two-stage method for predicting the beamforming matrix. In the first stage, based on the beam angle information of historical time slots, the capacitive Kalman filter (CKF) algorithm is fused with a long short-term memory (LSTM) network and a multilayer perceptron (MLP) to accurately predict the beam angle information of the next time slot, i.e., the beam departure angle (AoDs) between the RIS and the vehicle in the next time slot, thereby reconstructing the channel prediction to obtain the channel state information of the next time slot. In the second stage, based on the channel state information predicted in the previous stage, the soft actor-critic (SAC) algorithm is used to optimize the base station beamforming matrix and RIS phase shift matrix to maximize the total system throughput. The specific process is as follows: 2.1) Stage 1: Predicting Beam Angle Information In this system, since the base station and RIS deployment locations are fixed, the channel between the BS and RIS can be considered quasi-static; therefore, the channel... It can be fully obtained with low pilot signal overhead. However, since the vehicle is in motion, the channel between the RIS and the vehicle is time-varying. The ultimate goal of this stage is to predict the channel state information between the RIS and the vehicle. First, the angle of the line-of-sight path between the RIS and the vehicle is estimated. Then, the distance between them is calculated based on the vehicle's speed, thereby obtaining the channel state information of this line-of-sight path. Afterwards, the channel state information of multiple random non-line-of-sight paths is superimposed, ultimately achieving accurate reconstruction of the overall channel state information. Therefore, this embodiment focuses on how to quickly and accurately predict the beam angle of the line-of-sight path between the RIS and the vehicle. To solve this problem, see [link to relevant documentation]. Figure 3 This embodiment proposes an MLP-LSTM dual-mode prediction architecture based on the CKF algorithm, which improves the accuracy of AoDs prediction by integrating dual-mode data streams and filtering techniques. The proposed architecture mainly consists of two parts: a prediction module based on MLP and LSTM networks and a fusion module based on the CKF algorithm. In this framework, the CKF algorithm serves as the key fusion algorithm, integrating the outputs of the two neural networks to achieve unified angle state information prediction, as detailed below: Updating the beam angle is typically a complex nonlinear process, which can be simplified into a state transition process by extracting the time-varying features of the angle information. Deep learning possesses powerful feature extraction and nonlinear mapping capabilities, making it widely used in regression tasks. Therefore, we model the evolution of RIS to vehicle AoDs in each time slot as a state transition process and employ a deep neural network to approximate the corresponding state transition equation. The corresponding beam angle state vector... Defined as ,in and These represent the azimuth and elevation angles, respectively. The following section details the process of predicting the beam angle state vector using an MLP-LSTM dual-mode module and fusing the output through the CKF algorithm to obtain the predicted AoDs for the next time slot.

[0043] MLP-LSTM dual-modal module: In the proposed fusion method, an MLP network is trained to approximate the state transition equations used to propagate the beam angle state vector. The volume point. Specifically, the MLP receives the first... t The volume point of the wave velocity angle state vector in each time slot After calculating the volume point after propagation in the next time slot, the propagation process of the MLP network can be expressed as the following expression: (10) in Represents MLP networks, This represents the set of weight parameters of the trained MLP network. In the fusion algorithm proposed in this paper, the trained MLP network is used to fit the state transition equation of volumetric point propagation. The details of the fusion of MLP and CKF algorithms will be described later in the CKF fusion module.

[0044] In the CKF algorithm fusion process, the output of the LSTM prediction network is used as the measurement update value. The LSTM's memory units are utilized to process long-term dependencies in the time-series angle state vector data, enabling the generation of reliable predicted AoDs vectors based on historical beam angle information. Specifically, the LSTM network receives... S The input sequence consisting of AoDs vectors from consecutive historical time slots is denoted as The process of generating the AoDs prediction vector for the next time slot can be represented as: (11) in Representing LSTM networks, This represents the set of weight parameters of the trained LSTM network.

[0045] Since both MLP and LSTM networks use similar training processes to predict the AoDs vector for the next time slot, the gradient of their loss function relative to the weight parameters can be uniformly expressed as: (12) Where the loss function Used for calculation t+ The difference between the actual beam angle vector and the predicted beam angle vector in time slot 1. The parameters represent the training parameters for the MLP and LSTM networks. Here, the mean squared error loss function is used to quantify the deviation between the predicted and true values. It is important to note that the MLP network is trained using the original state vector to extract features of the state transition process, while when the trained MLP network is applied in the fusion algorithm, the input is the volume points of the state vector. Essentially, it uses the Gaussian distribution of the state vector to better handle the nonlinearity of the state transition.

[0046] CKF Algorithm Fusion Module: The CKF algorithm can handle nonlinear problems by transforming the state vector volume points. Its core process consists of two stages: state prediction and measurement update. (See also...) Figure 4 In this embodiment, the MLP network and LSTM network are used for the propagation and measurement update of AoDs volume points, respectively. The specific fusion process is as follows: Step 1: Volume Point Propagation. First, calculate the beam angle vector. The volume point is Here, the error covariance matrix... Performing the Choleski decomposition yields , Representation matrix The i List, n The dimension of the AoDs vector is represented. The calculated volume points are propagated using equation (9).

[0047] Step 2: Predict the state variables and their covariance. The predicted state vector is obtained by taking the mean of the propagated volume points: (13) Next, calculate the error covariance matrix: (14) here, It is the process noise covariance.

[0048] Step 3: Calculate the measured values ​​and their covariance. Based on the predicted state. and the predicted covariance matrix Generate new volume points: (15) The observation equation employs identity transformations; similarly, the new volume points are transformed using identity transformations and then averaged to obtain the observed values. The measurement covariance can be expressed as: (16) here, It measures the noise covariance.

[0049] Step 4: Calculate the state-measurement cross-covariance and Kalman gain. Based on the predicted and measured states obtained in the first three steps, calculate the state-measurement cross-covariance, expressed as: (17) The Kalman gain is used to measure the weight between the predicted and measured values. The Kalman gain can be expressed as: (18) Step 5: Update the forecast. The predicted value calculated using equation (10) is... The predicted state is corrected using the actual observations to obtain the final predicted state vector: (19) 2.2) Phase Two: Maximizing the total achievable throughput of the system Based on the predicted beam angle information between the RIS and the vehicle obtained in Phase 1, the predicted channel between the RIS and the vehicle in the next time slot can be reconstructed. Since the optimization problem (P1) involves multiple optimization variables and constraints C1, C3, and C4 are non-convex, solving this optimization problem becomes a non-convex optimization problem. To solve this complex non-convex optimization problem, this embodiment uses the SAC reinforcement learning algorithm to find the optimal beamforming matrix policy. Deep reinforcement learning includes four core elements: agent, action, reward, and policy, modeling the optimization problem as a Markov decision process. In each step, the agent continuously observes the state of the environment. Then through strategy Execute action Interact with the environment and receive rewards. and new status The entire sequence can be simplified to In V2I scenarios, vehicle movement patterns exhibit periodicity within specific timeframes. Therefore, this invention defines one round of the environment as... The initial time slot is denoted as The following section will construct the actions, states, rewards, and algorithm execution, with relevant details as follows: Action: Flatten the beamforming matrix and RIS phase shift matrix This serves as the action vector for the policy network. Since neural networks cannot directly handle complex numbers, the action network generates them separately. and The real and imaginary parts of the equation. To satisfy the transmit power constraint in constraint C2 and the phase constraint in constraint C4, the agent normalizes the action output, and the final action space is constructed as follows: (20) State: The state vector contains the base station beamforming matrix. RIS phase shift matrix A power vector encompassing the transmit power allocated to each vehicle. and the cascaded channel matrix for each vehicle Similarly, these matrices will be flattened, and the state space will be represented as: (twenty one) In addition, to reduce the correlation between state dimensions and introduce redundant information that could lead to suboptimal decision-making, the state vector is whitened after each step of the environment.

[0050] Reward: In each step, the immediate reward is determined by the total achievable downlink throughput, while also satisfying the Quality of Service constraint C1 and the continuous interference cancellation decoding requirement of constraint C3. Therefore, the reward function is defined as: (twenty two) in, , (twenty three) In equation (21), the first term is at time step t The total achievable throughput, the second item is the penalty for not meeting QoS and SIC decoding requirements, It is a control factor.

[0051] The SAC algorithm is suitable for continuous control tasks; therefore, this invention uses this algorithm to solve for the optimal beamforming matrix and RIS phase shift. During the process of obtaining the optimal strategy, the neural network operates in a high-dimensional state. The input is the probability distribution of actions, and the output is the probability distribution of actions. Then, specific actions are obtained from the probability distribution of actions through sampling, and the network parameters are adjusted based on the reward after the action is executed, thereby approximating the optimal policy. The SAC agent consists of five neural networks: one policy network, two Q-value networks, and a target network. Here, to reduce overestimation of the Q-value, two Q-value networks are used. The Q-value networks calculate the estimated Q-values ​​using the state and actions as inputs, and their loss function and parameter update method are expressed as follows: (twenty four) (25) in, These are small batches of state transition samples obtained from the experience replay buffer. The batch size is denoted as . The target Q value is denoted as . ,in, It's the discount rate. Corresponding to the j The parameters of the Q network, and the temperature coefficient It is the control entropy term. The strategy is the next state after observation. The action to choose from, The parameter is The policy network, and These represent the gradient of the loss function and the learning rate, respectively.

[0052] Similarly, the policy network takes the state vector provided by the environment as input and generates action vectors. The loss function and parameter update strategy of the policy network are expressed as follows: (26) (27) Among them, policy gradient The parameters are calculated from the gradient of the stochastic policy and updated based on gradient ascent. Finally, a soft update mechanism is used to update the parameters of the target Q-value network. , is represented as: (28) in It is to satisfy The constant.

[0053] (3) Simulation results In this embodiment, simulation analysis verifies the effectiveness of the proposed two-stage beamforming matrix prediction method in improving the system's achievable throughput. It is compared with the first-stage method based on Extended Kalman Filter (EKF) and LSTM prediction, and the second-stage method based on DDPG and alternating iterative optimization algorithms. In the EKF and LSTM-based scheme, the beam angle vectors between the RIS and vehicles are predicted using EKF and LSTM, respectively. The optimization of the beamforming matrix and RIS phase shift in the second stage is solved using the SAC reinforcement learning algorithm proposed in this embodiment. In the DDPG and alternating iterative optimization algorithm-based scheme, the beam angle vector prediction in the first stage uses the algorithm proposed in this invention, and the second stage uses DDPG and alternating optimization algorithms to optimize the beamforming matrix and RIS phase shift.

[0054] The simulation parameters are set as follows: total number of vehicles: 4; number of RF links: 4; number of transmit and receive antennas at the base station: 8 each. The RIS is a uniform planar array of 16 elements with 3 phase quantization bits. The number of consecutive historical sequences input to the LSTM network is 4; the noise power at the vehicle end is -60 dBm; and the minimum achievable communication rate is 1 bps / Hz.

[0055] Figure 5 This demonstrates the error in beam angle prediction between the RIS and the vehicle when using the proposed beam angle prediction scheme in a V2I system with RIS assistance and ISAC under consideration. The straight-line speeds of the four vehicles are... The root mean square error between the predicted beam angle vector and the actual beam angle is defined as: It can be seen that, in the vehicle beam angle prediction at different speeds, as the number of time slots increases, the vehicle first approaches and then moves away from the RIS. Due to the increase in tangential velocity, the predicted beam angle error first increases and then decreases. The results show that the beam angle prediction algorithm proposed in this application can obtain a beam angle closer to the actual value, especially in the near-field region close to the RIS.

[0056] Figure 6 This paper demonstrates the difference between the total achievable throughput and the throughput under ideal channel conditions when using the proposed two-stage beamforming prediction in a V2I system with RIS assistance and ISAC. The results are compared with prediction schemes based on extended Kalman filtering and long short-term memory networks. Simulation results show that the total achievable throughput of the system under the proposed two-stage beamforming prediction scheme is closest to the ideal condition, highlighting the crucial role of effective beam angle prediction methods in improving the total system throughput.

[0057] Figure 7 This demonstrates the average achievable throughput across all time slots when employing the proposed two-stage beamforming prediction scheme in a V2I system with RIS assistance and ISAC under consideration. The average achievable throughput is defined as: The proposed two-order beamforming prediction scheme is compared with schemes based on the DDPG algorithm and alternating iterative algorithm. Simulation results show that the average achievable throughput of the system under this scheme is significantly better than the other two schemes, demonstrating that optimized resource allocation is crucial for improving system throughput.

[0058] (4) Advantages and beneficial effects Current research in this field mainly focuses on beam tracking in ISAC systems or beamforming optimization in RIS-assisted ISAC systems, but there is relatively little research on beamforming prediction in V2I systems combining RIS and ISAC. Furthermore, most current research on beamforming prediction directly predicts based on historical channel information in an end-to-end manner, but its prediction performance is limited in rapidly changing channel environments. Considering the current state of research, this application proposes a V2I system framework combining RIS and ISAC, along with a corresponding two-stage beamforming prediction method. This application has the following main advantages: 1) This application proposes a V2I system combined with ISAC under RIS assistance, which provides a new solution for the construction of future wireless networks, and studies the beamforming matrix prediction of the system.

[0059] 2) In order to improve the accuracy of beam angle prediction, this application proposes an MLP-LSTM dual-mode beam angle predictor that integrates the CKF algorithm. It can obtain better beam angle prediction results by fusing the prediction results of the two models through the CKF algorithm.

[0060] 3) Based on the proposed beam angle prediction scheme, this application further studies how to improve the overall achievable throughput of the system. Under the premise of ensuring the quality of communication services, this application proposes a method that uses the SAC algorithm in reinforcement learning to jointly optimize the beamforming matrix and the RIS phase shift matrix. This method can adapt to systems with time-varying channel states and effectively improve the achievable throughput performance of the system.

[0061] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0062] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0063] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0064] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0065] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0066] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0067] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented in the embodiments of this program product are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments. The executable computer program code or "code" used to perform the various embodiments can be written in high-level programming languages ​​such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0068] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0069] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0070] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0071] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0072] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0073] It should be understood that in this application, "at least one (item)" means one or more, and "more than one" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0074] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0075] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0076] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0077] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0078] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A beamforming prediction method for a RIS-assisted sensor-integrated vehicle network, characterized in that, The method includes the following steps: Step S1: In the ISAC vehicle-to-everything (V2X) system assisted by the reconfigurable smart surface RIS, the base station acquires the beam angle status information between the RIS and the vehicle in historical time slots. Step S2: Based on the beam angle state information of the historical time slots, the capacitive Kalman filter (CKF) algorithm is used to fuse the long short-term memory (LSTM) network and the multilayer perceptron (MLP) model to predict the beam departure angle (AoDs) of the next time slot. Step S3: Based on the predicted AoDs of the next time slot, reconstruct the channel state information of the next time slot; Step S4: Based on the channel state information of the next time slot obtained from the reconstruction, the soft actor-commentator SAC reinforcement learning algorithm is used to jointly optimize and obtain the base station optimal beamforming matrix and RIS optimal phase shift matrix of the next time slot in order to maximize the total system throughput.

2. The method according to claim 1, characterized in that, Step S2 includes: A dual-mode prediction module consisting of an MLP network and an LSTM network, and a fusion module based on the CKF algorithm were constructed. The MLP network is used to fit the state transition equation to propagate the volume points of the beam angle state vector; the LSTM network is used to generate the predicted AoDs vector based on historical beam angle information, which serves as the measurement update value for the CKF algorithm; the CKF algorithm is used to fuse the outputs of the MLP network and the LSTM network to obtain the final predicted AoDs for the next time slot.

3. The method according to claim 2, characterized in that, The fusion module based on the CKF algorithm performs the following steps: Volume point propagation steps: Generate volume points based on the beam angle state vector and its error covariance matrix of the current time slot, and propagate the volume points using the trained MLP network; State prediction steps: Take the mean of the volume points after propagation to obtain the predicted state vector and the prediction error covariance matrix; Measurement update steps: Use the output of the trained LSTM network as the actual observation value, and calculate the Kalman gain based on the predicted state vector; State fusion step: Based on the Kalman gain, the predicted state vector is fused with the output of the LSTM network to obtain the final predicted next time slot AoDs.

4. The method according to claim 1, characterized in that, Step S4 specifically includes: Flatten the real and imaginary parts of the elements of the base station beamforming matrix and the RIS phase shift matrix to construct the action space of the SAC algorithm. The base station beamforming matrix, RIS phase shift matrix, power vectors allocated to each vehicle, and cascaded channel matrix of the current time slot are flattened and whitened to construct the state space of the SAC algorithm. The total downlink reachable throughput of the system is used as the immediate reward, and penalties are set for failing to meet the QoS constraints and SIC decoding requirements of continuous interference cancellation, thus constructing the reward function of the SAC algorithm. The optimal base station beamforming matrix and RIS phase shift matrix are output through iterative training of the policy network, value network, and target network of the SAC algorithm.

5. The method according to claim 4, characterized in that, The base station beamforming matrix and RIS phase shift matrix in the action space satisfy the maximum transmit power constraint of the base station and the discrete phase constraint of the RIS reflection unit; the phase codebook of the discrete phase constraint is preset according to the number of quantization bits.

6. The method according to claim 1, characterized in that, The system is an ISAC-based RIS-assisted millimeter-wave vehicle networking system for urban scenarios, including a fixed ISAC base station configured with an array of multiple transmitting antennas and multiple receiving antennas, K single-antenna vehicles, and a RIS deployed between the base station and the vehicles, containing N reflective units.

7. The method according to claim 6, characterized in that, The communication between the base station and the vehicle adopts a parallel transmission protocol: the base station continuously sends downlink ISAC signals in the same time slot. After being reflected by the vehicle, the signals are used as sensing echo signals and transmitted back to the base station via RIS. The base station extracts the beam angle information of the current time slot from the received sensing echo signals to predict the beam angle of the next time slot.

8. The method according to claim 1, characterized in that, The beam angle state information in step S1 is specifically the beam departure angle AoDs of the line-of-sight path between the RIS and the vehicle, including the azimuth angle and the elevation angle; when reconstructing the channel state information in step S3, the line-of-sight path channel is first calculated based on the predicted AoDs and the vehicle speed, and then the randomly generated non-line-of-sight path channel components are superimposed.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 8.

Images