Metasurface-assisted MIMO precoding and beam tracking methods, apparatus, equipment and media
By using topology-adaptive hierarchical codebook design and multi-agent collaborative decision-making of intelligent metasurfaces, combined with fuzzy logic control and physical information neural networks, the problems of high channel information acquisition overhead, high computational complexity, privacy leakage and insufficient beam tracking accuracy are solved, realizing efficient and real-time distributed metasurface-assisted MIMO precoding and beam tracking.
Patent Information
- Application Number
- CN202611042868.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-14
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies suffer from huge overhead in acquiring channel information, high computational complexity in joint optimization, and inability to meet real-time requirements. Centralized training poses a risk of privacy leakage, beam tracking accuracy is insufficient in mobile scenarios, and the hierarchical codebook design does not consider distributed metasurface geometric constraints.
Initial beam alignment is performed based on the installation topology information of the smart metasurface. Cooperative search is carried out using a topology-adaptive hierarchical codebook design. Multi-agent cooperative decision-making and deep reinforcement learning are combined. Precoding optimization is performed using a fuzzy logic controller. Dynamic beam tracking is performed through a physical information neural network. This enables distributed metasurface-assisted MIMO precoding and beam tracking.
It reduces the overhead of acquiring channel information, improves computational efficiency and real-time performance, protects data privacy, and enhances beam tracking accuracy and collaborative tracking efficiency in mobile scenarios.
Smart Images

Figure CN123122592A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wireless communication technology, and in particular to metasurface-assisted MIMO precoding and beam tracking methods, apparatuses, devices, and media. Background Technology
[0002] With the increasing demand for ultra-high spectral efficiency and full coverage in 6G networks, decellularized massive MIMO and reconfigurable smart surface (RIS)-assisted communication technologies have become important research directions. In RIS-assisted decellularized massive MIMO systems, a large number of distributed access points (APs) and RISs, under the coordination of a central processing unit (CPU), utilize the same time-frequency resources to provide services to users. To maximize system performance and data rate, joint optimization of the AP precoding matrix and RIS reflection phase is required; in mobile scenarios, dynamic beam tracking of the RIS beam is also necessary. In existing technologies, joint precoding often employs alternating optimization (AO), semi-definite relaxation (SDR), or deep reinforcement learning (DRL) methods; beam tracking often uses methods based on direction-of-arrival (DOA) estimation, deep learning prediction, and hierarchical codebook search.
[0003] However, existing technologies have the following drawbacks: First, the channel information acquisition overhead is huge: the existing system needs to acquire the instantaneous CSI of all AP-RIS-user cascaded channels. Under the distributed multi-RIS and multi-AP deployment, the pilot overhead increases exponentially, making it difficult to support large-scale system expansion.
[0004] Second, the joint optimization computation is highly complex and cannot meet the real-time requirements: the algorithm complexity based on AO and SDR is as high as O(N³), and the computation delay is on the order of seconds, which is far greater than the channel coherence time of about 120μs in the 6G terahertz band (such as 0.3THz), and cannot achieve real-time beam adjustment.
[0005] Third, centralized training has privacy and communication bottlenecks: existing multi-agent reinforcement learning (MARL) based solutions still need to aggregate global channel information during the training phase, resulting in huge communication overhead and data privacy leakage risks in scenarios where multiple operators share infrastructure.
[0006] Fourth, beam tracking accuracy is insufficient in mobile scenarios: Existing methods that rely on instantaneous DOA estimation or pure data-driven methods have large angle estimation errors (5°-10°) in high dynamic scenarios, resulting in severe beam gain loss. Furthermore, they do not consider multi-RIS collaborative tracking, which can easily lead to ping-pong switching and discontinuous experience.
[0007] Fifth, the hierarchical codebook does not consider the topological constraints of distributed RIS: the existing hierarchical codebook is mainly designed for single RIS scenarios and does not consider the topological differences in the position, orientation and size of each RIS in a distributed system. The unified codebook cannot adapt to the geometric features of each node, resulting in low efficiency of collaborative tracing. Summary of the Invention
[0008] The technical problem to be solved by this invention is to provide a metasurface-assisted MIMO precoding and beam tracking method, apparatus, device and medium to solve the technical problems in the prior art, such as huge overhead in channel information acquisition, high computational complexity of joint optimization and inability to meet real-time requirements, privacy leakage and large communication overhead in centralized training architecture, insufficient beam tracking accuracy in mobile scenarios, and failure to consider distributed metasurface geometric constraints in hierarchical codebook design.
[0009] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a metasurface-assisted MIMO precoding and beam tracking method, comprising: Based on the installation topology information of each smart metasurface, the beams of each smart metasurface are initially aligned to obtain the initial beam configuration. Based on the initial beam configuration, joint precoding optimization is performed on each access point and each smart metasurface to obtain a steady-state precoding configuration. Based on the steady-state precoding configuration, the beam offset caused by user movement is dynamically tracked to obtain the updated beam configuration.
[0010] Based on the above technical solution, the present invention can be further improved as follows.
[0011] Furthermore, the initial beam alignment of each smart metasurface based on its installation topology information to obtain an initial beam configuration includes: Based on the installation location and orientation information of each smart metasurface, the transformation relationship between the local coordinate system and the global coordinate system of each smart metasurface is constructed; Based on the transformation relationship, a codebook adapted to the installation orientation of each smart metasurface is generated; Based on the codebook, the beams of each smart metasurface are searched to obtain the initial beam configuration.
[0012] Furthermore, the codebook is a hierarchical codebook, which includes a coarse-grained codebook and a fine-grained codebook; The beam search of each smart metasurface includes: Parallel coarse search is performed based on the coarse-grained codebook to determine the master intelligent metasurface based on the received power fed back by the user terminal; fine search is performed on the master intelligent metasurface based on the fine-grained codebook to determine the optimal beam index; the other auxiliary intelligent metasurfaces calculate their respective optimal beam directions based on their relative positions and orientation differences with the master intelligent metasurface.
[0013] Furthermore, the joint precoding optimization of each access point and each smart metasurface based on the initial beam configuration to obtain a steady-state precoding configuration includes: Obtain local channel status information for each access point and each smart metasurface; Based on the local channel state information, the precoding matrix of each access point and the reflection phase of each smart metasurface are obtained through multi-agent collaborative decision-making. Based on expert knowledge, the precoding matrix and the reflection phase are corrected to obtain a steady-state precoding configuration.
[0014] Furthermore, the multi-agent collaborative decision-making is achieved through a deep reinforcement learning policy network, and the expert knowledge is achieved through a fuzzy logic controller. The fuzzy logic controller converts the output of the deep reinforcement learning policy network into fuzzy linguistic variables based on fuzzy rules, and fuses the deep reinforcement learning output and fuzzy logic correction results through a membership function to obtain a steady-state precoding configuration.
[0015] Furthermore, each access point and each smart metasurface independently trains a policy network based on local channel state information, and uploads the policy network parameters to the central processing unit for aggregation. After obtaining the global policy network parameters, they are then distributed to each access point and each smart metasurface. During the inference phase, each access point and each smart metasurface makes independent decisions based on local observations and does not transmit raw channel state information.
[0016] Furthermore, the step of dynamically tracking the beam offset caused by user movement based on the steady-state precoding configuration to obtain the updated beam configuration includes: Based on the historical channel state information sequence, the user's beam angle of arrival is predicted by a prediction model to obtain the predicted beam direction; The predicted beam direction is corrected based on the physical constraints of electromagnetic wave propagation to obtain an updated beam configuration. The dynamic tracking includes: running the prediction model in parallel on multiple smart metasurfaces, collecting the prediction results and prediction variance of each smart metasurface, calculating the fusion weight based on the prediction variance, and performing a weighted average of the prediction angles of each smart metasurface to obtain the global optimal angle estimate.
[0017] Secondly, the present invention provides a metasurface-assisted MIMO precoding and beam tracking device, comprising: The initial alignment module is used to perform initial beam alignment on each smart metasurface based on the installation topology information of each smart metasurface to obtain the initial beam configuration. The joint precoding module is used to perform joint precoding optimization on each access point and each smart metasurface based on the initial beam configuration to obtain a steady-state precoding configuration; The dynamic tracking module is used to dynamically track the beam offset caused by user movement based on the steady-state precoding configuration, and obtain the updated beam configuration. The collaborative scheduling module is used to schedule the initial alignment module, joint precoding module, and dynamic tracking module in a time sequence according to the initial alignment, joint precoding optimization, and dynamic tracking to form a collaborative closed loop.
[0018] Thirdly, the present invention provides a computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the metasurface-assisted MIMO precoding and beam tracking method.
[0019] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the metasurface-assisted MIMO precoding and beam tracking method. Attached Figure Description
[0020] Figure 1 A flowchart illustrating the metasurface-assisted MIMO precoding and beam tracking method; Figure 2 Flowchart of TA-HC codebook search for metasurface-assisted MIMO precoding and beam tracking method; Figure 3 A flowchart of the Fuzzy_MARL method for metasurface-assisted MIMO precoding and beam tracking. Figure 4 PINN beam tracking flowchart for metasurface-assisted MIMO precoding and beam tracking method; Figure 5 A schematic diagram of a metasurface-assisted MIMO precoding and beam tracking device; Figure 6 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation
[0021] The technical solutions of the embodiments of the present invention will be described below with reference to the accompanying drawings. In the description of the present invention, unless otherwise stated, " / " indicates that the objects before and after are in an "or" relationship. For example, A / B can represent A or B. The "or" in the present invention is merely a description of the relationship between the related objects, indicating that three relationships can exist. For example, A or B can represent: A alone, A and B simultaneously, and B alone. A and B can be singular or plural. Furthermore, in the description of the present invention, unless otherwise stated, "multiple" refers to two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items.
[0022] Furthermore, to facilitate a clear description of the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.
[0023] In this embodiment of the invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in this embodiment of the invention should not be construed as superior or more advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner for ease of understanding.
[0024] Example 1, an embodiment of the present invention, provides a metasurface-assisted MIMO precoding and beam tracking system. The system architecture includes the following core components: Central Processing Unit (CPU): Responsible for coordination and management, but does not participate in real-time channel estimation and precoding calculations, thus protecting data privacy.
[0025] Distributed Access Point (AP) Cluster: Each AP is equipped with a small number of antennas (typically 8-16) to serve users in a specific area. The APs are connected to the CPU via wireless fronthaul.
[0026] Distributed RIS clusters: Deployed on building walls, light poles, traffic facilities, etc. Each RIS contains hundreds to thousands of adjustable reflective units, requiring no dedicated power supply, and is powered by environmental backscattering or radio frequency energy harvesting.
[0027] Edge AI Controller: Deployed on the AP or RIS edge server, responsible for running the Fuzzy-MARL algorithm and PINN beam tracking algorithm to achieve distributed real-time decision-making.
[0028] User terminal (UE): Equipped with multiple antennas and supports multi-stream parallel reception.
[0029] Example 2, as Figure 1 As shown, this is an embodiment of the present invention, providing a metasurface-assisted MIMO precoding and beam tracking method. Specifically, the three steps S1, S2, and S3 are not independent of each other, but form an organically cooperative whole, achieving optimal system-level performance through a four-layer mechanism of temporal coordination, information interaction, resource sharing, and performance gain superposition.
[0030] S1: Based on the installation topology information of each smart metasurface, the beams of each smart metasurface are initially aligned to obtain the initial beam configuration. It should be noted that in distributed RIS cooperative tracing, initial beam alignment (IBA) is a crucial step in determining the system access success rate. Traditional hierarchical codebook designs assume a regular geometric topology for a single RIS, making it difficult to adapt to situations where the positions, orientations, and sizes of each RIS in a distributed RIS system vary. This application proposes a Topology-Adaptive Hierarchical Codebook (TA-HC) design method.
[0031] S1.1: Based on the installation location and orientation information of each smart metasurface, construct the transformation relationship between the local coordinate system and the global coordinate system of each smart metasurface; S1.2: Based on the transformation relationship, generate a codebook that adapts to the installation orientation of each smart metasurface; S1.3: Based on the codebook, the beams of each smart metasurface are searched to obtain an initial beam configuration. The codebook is a hierarchical codebook, which includes a coarse-grained codebook and a fine-grained codebook. The beam search of each smart metasurface includes: Parallel coarse search is performed based on the coarse-grained codebook to determine the master intelligent metasurface based on the received power fed back by the user terminal; a fine search is then performed on the master intelligent metasurface based on the fine-grained codebook to determine the optimal beam index; the remaining auxiliary intelligent metasurfaces calculate their respective optimal beam directions based on their relative positions and orientation differences with the master intelligent metasurface. The specific implementation method is as follows: Let the first The physical parameters of the RIS are: installation height Orientation angle (Relative to due north) Effective aperture (Unit: wavelength multiple, i.e.) The embodiments of this application dynamically generate the hierarchical codebook specific to the RIS based on these parameters.
[0032] Based on the installation height, orientation angle, and effective aperture of each RIS, a hierarchical codebook is dynamically generated for that RIS. The codebook is divided into three layers: the first layer has four wide beams covering a 360-degree range, the second layer has 16 medium beams covering the selected sector, and the third layer has 64 narrow beams for precise alignment.
[0033] Where l = 1, 2, ..., L represents the codebook level (typically L = 3), M l The number of codewords in the l-th layer: Layer 1 (coarse search): M1 = 4 wide beams, covering the angular range [0°, 360°], beamwidth Ω1 ≈ 90° / D n .
[0034] Layer 2 (Mid-search): M2 = 16 mid-beams, covering the selected sector in Layer 1, beamwidth Ω2 ≈ 22.5° / D n .
[0035] Furthermore, the cooperative search method based on topology-adaptive hierarchical codebook (TA-HC) includes two stages: offline codebook generation and online search, achieving cooperative beam alignment of multiple RIS through geometric deduction. For example... Figure 2 As shown, the specific implementation method is as follows: Step 101: Offline codebook generation; Obtaining geometric parameters for each RIS: During system initialization, the CPU collects the physical parameters of each RIS, including the installation height h. n Orientation angle ψ n (Relative to true north) and effective aperture D n .
[0036] Calculate the coordinate transformation matrix: based on the orientation angle ψ of each RIS. n Calculate the rotation transformation matrix R(ψ) from the global coordinate system to the RIS local coordinate system. n ), used for subsequent beam pointing mapping.
[0037] Generating a three-layer codebook: For the nth RIS, dynamically generate a three-layer hierarchical codebook based on its effective aperture Dn: It has 4 wide beams (covering a 90° / Dn range). It has 16 medium-wave beams. It has 64 narrow beams.
[0038] Store codeword index: Store the generated codeword and its corresponding beam pointing angle (elevation angle θ, azimuth angle φ) index in the local memory of each RIS, occupying about 1MB of storage space.
[0039] Step 102: Parallel coarse search; CPU synchronously sends L1 index: The CPU synchronously sends the L1 codeword index command to all RIS, and each RIS configures the phase response of its reflection unit according to the index to form the corresponding wide beam pointing.
[0040] Independent scanning of each RIS: Each RIS is configured with 4 wide beams in the codebook order, with each beam lasting for a duration of Tbeam. The user terminal measures the received signal power RSP during each beam period.
[0041] Received power measurement: The user terminal measures the received power of the reflected signals from each RIS through the Rx-BeamSweeping process and records the power value of each beam for each RIS.
[0042] Feedback signal strength: The user sets the received power vector of all RIS to RSP=[RSP1,RSP2,...,RSP] N Feedback is sent to the CPU via the PUCCH channel.
[0043] Step 103: Select the main control RIS; Comparing the signal strength of each RIS: The CPU analyzes the coarse search results for each RIS and compares the maximum received power value corresponding to each RIS. .
[0044] Select the strongest signal RIS: Select The largest RIS serves as the master RIS (denoted as RIS). The RIS has optimal coverage conditions at the current user location.
[0045] Mark master and auxiliary: Mark the selected master RIS as performing fine search tasks, and mark the other RIS as auxiliary roles to perform geometric calculation tasks.
[0046] Step 104: Master RIS fine search; Perform L2 layer scan: Within the sector selected by the coarse search, the main control RIS performs a L2 scan search of 16 mid-wave beams to determine the optimal direction at the mid-wave beam level.
[0047] Perform L3 layer scan: Based on the selected sub-sectors in L2, perform a fine search of 64 narrow beams in the third layer to determine the final narrow beam index. .
[0048] Determine the optimal beam index: Record the optimal beam index of the master RIS. and its corresponding spatial frequency This is used for subsequent geometric calculations.
[0049] Step 105: Geometric deduction coordination; Calculating the relative position of the auxiliary RIS: Based on the installation coordinates of each RIS, the CPU calculates the positional offset of the auxiliary RIS relative to the main control RIS. , and orientation angle difference .
[0050] Calculating the optimal phase of the auxiliary RIS: Based on the optimal beam direction of the master RIS, the optimal phase configuration of the auxiliary RIS is calculated through geometric relationships. .
[0051] Multi-RIS joint configuration: The CPU broadcasts the optimal phase configuration of each RIS, and all RIS simultaneously configure the reflection phase to form a cooperative beam pointing towards the user.
[0052] Beamforming complete: The user receives coherent composite signals from each RIS in the overlapping area of multiple RIS coverage, realizing the complete beamforming process, and the system enters the steady-state communication stage.
[0053] Layer 3 (Fine Search): M3 = 64 narrow beams, covering the selected sub-sectors in Layer 2, beamwidth... .
[0054] Typing The array steering vector has the following spatial frequency: Beam pointing parameters: The spatial frequency of the codeword is determined by the elevation and azimuth angles of the beam, which in turn determine the physical pointing direction of the beam. .
[0055] In some embodiments, in a multi-RIS collaborative scenario, the user may be located in an overlapping area covered by multiple RIS systems. This invention proposes a fast search strategy based on distributed coordination: Parallel coarse search: All RIS perform Level 1 coarse search simultaneously, and the user measures and reports the received power of the reflected signal from each RIS.
[0056] Centralized optimization: The CPU collects coarse search results from each RIS and selects the RIS with the least path loss (i.e. the highest receiving power) as the master RIS, with the other RIS serving as auxiliary RIS.
[0057] Layered Fine-Grade Search: After the master RIS is determined, Layer 2 and Layer 3 fine-grained searches are performed. The Layer 2 search performs a multi-choice search within the sector range determined by Layer 1. The user measures and reports the received signal power corresponding to each codeword. The CPU selects the codeword with the highest power as the optimal codeword for Layer 2 and determines the corresponding sector range. The Layer 3 search performs a more refined search within the sub-sector range determined by Layer 2. Similar to Layer 2, the user measures the received signal power corresponding to each codeword, and the CPU selects the codeword with the highest power as the final optimal codeword. The beam pointing angle corresponding to this codeword is the final beam alignment result. The other RISs, based on the search results of the master RIS, calculate their own optimal codewords through geometric relationships (without repeating the search).
[0058] Geometric calculation mechanism: After the master RIS completes the fine search, the auxiliary RIS calculates its optimal beam direction based on its relative position and orientation difference with the master RIS through geometric relationships, without the need for repeated searches.
[0059] in, For the first The geometrical offset of each RIS relative to the main RIS. This is the distance from the RIS to the main RIS. This represents the relative angular difference between the two orientations.
[0060] Search complexity analysis: For a single RIS, the total number of searches in a hierarchical search is the sum of the number of codewords at each layer, which is significantly more efficient than exhaustive search. However, in multi-RIS collaborative scenarios, the advantages of hierarchical search are even more pronounced. If each RIS performs a complete search independently, the total number of searches is directly proportional to the number of RISes. By adopting the collaborative search strategy of this application embodiment, the total number of searches is significantly reduced. When the system contains multiple RISes, the number of searches required for independent searches is far greater than the number of searches required for collaborative searches. By adopting the collaborative search strategy of this invention, the total number of searches is significantly reduced, and search efficiency is improved several times over.
[0061] The trade-off between search time and accuracy: A key parameter in hierarchical codebook search is the resolution ratio between layers. The resolution ratio employed in this invention achieves the optimal balance between search efficiency and positioning accuracy. Simulation verification shows that this design achieves extremely high access success rate in the initial access phase and precise tracking performance in the beam tracking phase.
[0062] Furthermore, the technical effects of the embodiments of this application compared with the prior art are as follows: First, existing hierarchical codebooks are designed for single-AP scenarios and do not consider the multi-source collaboration problem of distributed RIS. This invention proposes for the first time a distributed collaborative search strategy based on RIS geometric topology, reducing the search time from O(M·N) to O(M+N).
[0063] Second, existing solutions assume that all RIS have the same physical parameters, ignoring individual differences. The TA-HC of this invention independently generates the optimal codebook based on the actual topology parameters of each RIS.
[0064] Third, this invention uses a geometric calculation mechanism to enable the auxiliary RIS to obtain a near-optimal beam direction without repeated searches, saving 60% of the search time compared to the full search strategy.
[0065] S2: Based on the initial beam configuration, joint precoding optimization is performed on each access point and each smart metasurface to obtain a steady-state precoding configuration; This invention models the joint optimization problem of RIS-assisted CF large-scale MIMO as a distributed partially observable Markov decision process (Dec-POMDP).
[0066] S2.1: Obtain local channel status information for each access point and each smart metasurface; S2.2: Based on the local channel state information, the precoding matrix of each access point and the reflection phase of each smart metasurface are obtained through multi-agent collaborative decision-making.
[0067] S2.3: Based on expert knowledge, the precoding matrix and the reflection phase are corrected to obtain a steady-state precoding configuration. The specific implementation method is as follows: Suppose the system contains M APs, N RISs, and K users. The number of antennas in each AP is N m , No. The number of units in each RIS is .user The received signal model is as follows: user The received signal consists of three parts: the direct signal from the AP, the target signal reflected by the RIS, and the interference signal from other users.
[0068] Among them: local channel state information, including the direct channel vector from AP to user (dimension consistent with the number of AP antennas) and the channel vector reflected by RIS (dimension consistent with the number of RIS elements).
[0069] For the first The RIS is reflected by the first AP to user Channel vector (reflection path) For the first AP to user Direct channel vector (direct path) For the first The reflection phase matrix of each RIS For the first An AP sends to the user precoding vector For users The transmitted symbol, The sum rate (sum SE) optimization problem for an additive white Gaussian noise system is formulated as follows: Optimization objective: Maximize the sum rate (sum SE) of all users in the system while satisfying the total transmit power constraint and the RIS phase constraint. The optimization variables are the precoding matrix W of all APs and the reflection phase matrix of all RISs.
[0070] Constraint: The total transmit power of all APs shall not exceed the maximum power budget. The phase amplitude of each RIS reflector is always 1 (i.e., pure phase adjustment).
[0071] in For users The aggregated channel vector, This is for aggregating precoded vectors.
[0072] The multi-agent collaborative decision-making is achieved through a deep reinforcement learning policy network. This invention models each AP and each RIS as an agent, forming a collaborative decision-making system of (M+N) agents. The state space, action space, and reward function of each agent are defined as follows: State space: intelligent agent status Includes: local channel vector Equivalent interference levels of adjacent APs / RIS Remaining power budget Current time slot t.
[0073] Action space: The actions of the AP agent are precoded vectors sent to each user; the actions of the RIS agent are the phase configurations of each reflection unit.
[0074] Reward function: agent Instant rewards Defined as: in, For users The achievable rate, , , These are the weighting coefficients. for The user set of the service Total transmission power, This represents the level of interference between small segments.
[0075] The expert knowledge is implemented through a fuzzy logic controller. The fuzzy logic controller converts the output of the deep reinforcement learning policy network into fuzzy linguistic variables based on fuzzy rules, and fuses the deep reinforcement learning output and fuzzy logic correction results through a membership function to obtain a steady-state pre-encoding configuration. In traditional DRLs, the Q-value or policy gradient output of the neural network suffers from large estimation bias and unstable convergence. This invention introduces a fuzzy logic controller (FLC) to correct the action output of the DRL. The core idea is to use fuzzy rules to map the continuous state space into coarse-grained actions guided by expert knowledge, and then have the DRL network learn fine-grained compensation. Fuzzy logic enhancement mechanism: The policy network considers both the actions output by DRL and the expert rules of FLC, and integrates the two through the membership function to balance exploration efficiency and expert knowledge guidance.
[0076] in, The membership function chosen for the action is derived from the expert knowledge in the FLC rule base from the following physical constraints: The RIS reflection phase should be such that the reflected signal and the direct signal are superimposed in phase at the user.
[0077] The reflected beam of the RIS should be directed towards the user whose direct path is blocked.
[0078] The precoding of adjacent APs should be spatially orthogonal to reduce interference.
[0079] In some embodiments, such as Figure 3 As shown, the implementation process based on Fuzzy-MARL joint precoding and phase optimization includes the following five core stages, each of which performs a specific function and provides the necessary input data for the next stage.
[0080] Step 201: System initialization phase; Initial AP power configuration: CPU based on total system power budget. Based on the number of users served by each AP, allocate the initial transmit power proportionally to generate a power allocation vector. .
[0081] Initial RIS phase configuration: Set the phase of each RIS reflection unit to 0 degrees, i.e. This ensures that the initial reflection is omnidirectional scattering.
[0082] User association establishment: Based on user geographical location and channel conditions, establish an AP-user association matrix. Determine the user set for each AP service.
[0083] Policy network parameter initialization: Initialize the DRL policy network using the Xavier method. and value network The parameters are initialized simultaneously with the FLC membership function parameters.
[0084] Step 202: Local data acquisition phase; Each AP independently receives SRS pilot signals: In the uplink pilot time slot of the TDD system, each AP receives the Sounding Reference Signal (SRS) sent by its serving user.
[0085] Constructing the local channel state vector: Each AP performs channel estimation based on the received SRS and constructs a local channel state vector. This includes estimates for both the direct channel and the RIS reflection channel.
[0086] Calculate the local interference level: Based on the set of serving users of neighboring APs, calculate the equivalent interference level = , as one of the state inputs of the DRL agent.
[0087] Step 203: Fuzzy-MARL Inference Phase; DRL policy network output actions: Each AP / RIS agent outputs actions based on its local state. Through policy network Output initial action (Precoded vector or phase configuration).
[0088] Fuzzy Logic Controller Correction: The actions output by the DRL are converted into fuzzy linguistic variables (high / medium / low), corrected using expert knowledge from the FLC rule base, and the corrected fuzzy actions are output. .
[0089] Motion fusion and output: using membership functions Perform motion fusion and output the final motion. And configure it to AP or RIS.
[0090] Step 204: Federated Aggregation Update Phase; It should be noted that each access point and each smart metasurface independently trains a policy network based on local channel state information, and uploads the policy network parameters to the central processing unit for aggregation. After obtaining the global policy network parameters, they are then distributed to each access point and each smart metasurface. Local gradient calculation: Every τ time slots (τ=50), each AP / RIS samples from the local experience replay pool and calculates the policy gradient. , where J is the objective function (and rate maximization).
[0091] Gradient upload to CPU: Each AP / RIS will upload the encrypted gradient. The data is uploaded to the CPU via a secure channel, with a transmission volume of approximately 10KB per transmission, which is more than 80% lower than the original CSI data.
[0092] FedAvg aggregation: The CPU executes the FedAvg algorithm to aggregate global gradients. Where M is the number of APs and N is the number of RISs.
[0093] Parameter delivery: The CPU executes gradient descent to update the global policy network parameters. The updated policy parameters are then sent to the edge controllers of each AP / RIS.
[0094] Step 205: Convergence Judgment and Configuration Phase; Convergence condition check: Determine whether the change in reward |Δr| over N consecutive iterations is less than the threshold ε (typical value ε=0.01). If it is satisfied, it is considered converged.
[0095] Precoding vector configuration: If the convergence state is determined, the CPU will send the globally optimal precoding matrix W* to each AP, and the AP will lock the precoding configuration and transmit the data.
[0096] RIS phase configuration: Simultaneously optimize the globally optimal RIS phase configuration. *The data is sent to each RIS, and the RIS completes phase locking and enters the steady-state transmission phase.
[0097] It should be noted that during the inference phase, each access point and each intelligent metasurface makes independent decisions based on local observations and does not transmit original channel state information. To reduce communication overhead during the training phase, this invention adopts a hybrid architecture combining federated learning and MARL: Local training: Each AP / RIS independently trains its policy network on an edge server based on local CSI samples. .
[0098] Gradient aggregation: Each AP / RIS only aggregates the gradients of the policy network parameters. Upload to the CPU, but do not transmit raw channel data.
[0099] Policy Update: The CPU aggregates global gradients using the FedAvg algorithm and updates the global policy network. The updated policy will then be distributed to each AP / RIS.
[0100] Distributed execution: During the inference phase, each AP / RIS independently selects actions based on local observations, without requiring real-time CPU involvement. This architecture reduces channel information exchange overhead by more than 80% while ensuring data privacy.
[0101] It should be noted that the beneficial effects of the embodiments of this application compared to the prior art are as follows: First, while existing solutions employ MARL, the training phase still relies on centralized global channel information aggregation. This invention, through a hybrid architecture combining FL-MARL and federated learning, achieves, for the first time, a fully distributed training phase.
[0102] Second, existing solutions only handle steady-state optimization in static scenarios and do not address dynamic beam tracking for mobile users. This invention models joint precoding and beam tracking as a unified temporal decision problem.
[0103] Third, the introduction of the fuzzy logic layer improves the convergence speed of the algorithm by more than 3 times, and maintains stable performance even in scenarios with channel estimation errors as high as 20%.
[0104] S3: Based on the steady-state precoding configuration, the beam offset caused by user movement is dynamically tracked to obtain the updated beam configuration. To address the dynamic beam tracking problem for mobile users in a distributed RIS-assisted CF system, this invention proposes a Physical Information Neural Network (PINN) fusion method. Unlike traditional purely data-driven deep learning methods, PINN embeds the physical laws of electromagnetic wave propagation as hard or soft constraints into the loss function of the neural network, significantly improving the model's generalization ability in scenarios with few samples and high dynamics.
[0105] Based on the historical channel state information sequence, the user's beam angle of arrival is predicted by a prediction model to obtain the predicted beam direction; the predicted beam direction is corrected based on the physical constraints of electromagnetic wave propagation to obtain the updated beam configuration. In some embodiments, the path loss of the RIS-assisted wireless link follows the Extended Radar Range Equation: Physical constraints: The path loss of the RIS auxiliary link is related to transmit power, antenna gain, operating wavelength, effective aperture of the RIS, and the distance between the base station RIS and the RIS user. This physical model ensures that the neural network predictions conform to the laws of electromagnetic wave propagation.
[0106] in: For transmission power, For transmit / receive antenna gain; For the operating wavelength, The path loss index (usually 2-4); , For base station-RIS and RIS-user distances; For the effective aperture area of RIS, For RIS reflection efficiency; PINN's loss function consists of three parts: in: Data-driven loss (supervised learning term); This represents a loss of physical consistency. This is for boundary condition loss (ensuring the predicted location is within a reasonable range); , Trade-off coefficient (typical value) =0.5、 =0.1).
[0107] In some embodiments, the prediction model employs a bidirectional long short-term memory (Bi-LSTM) network architecture to fully exploit the temporal correlation of user movement trajectories. Input layer: The user CSI matrix of the past T time slots Convert to a sequence feature vector.
[0108] Bi-LSTM layer: Two independent LSTMs perform forward and backward temporal modeling respectively, extracting long-term dependency features: , .
[0109] Feature fusion layer: This will integrate the forward feed... and backward Hidden state splicing: .
[0110] In some embodiments, such as Figure 4 As shown, the dynamic beam tracking method based on PINN fusion includes two stages: offline training and online tracking. It achieves high-precision mobile user location prediction through the constraint embedding of physical information neural networks.
[0111] The dynamic tracking includes: running the prediction model in parallel on multiple smart metasurfaces, collecting the prediction results and prediction variance of each smart metasurface, calculating the fusion weight based on the prediction variance, and performing a weighted average of the prediction angles of each smart metasurface to obtain the global optimal angle estimate.
[0112] Step 301: Offline training phase; Collect user CSI sequence data during system operation to construct a training dataset. ,in For a moment, For the channel matrix, This refers to the user's actual location coordinates.
[0113] The original CSI data is denoised and normalized, and time-series features including user movement speed, acceleration and steering angle are extracted to construct the input sequence of Bi-LSTM.
[0114] Construct a loss function that includes physical constraints: in, This is the consistency loss term in the radar range equation.
[0115] After testing the model performance on the validation set and ensuring that the angle prediction error is less than 3°, the model parameters are deployed to the edge AI controllers of each RIS.
[0116] Step 302: Online tracking initialization; Each RIS controller reads the trained PINN model parameters from memory and initializes the network weights of the Bi-LSTM encoder and physical constraint layer.
[0117] The user terminal initiates a random access procedure, the RIS receives the uplink reference signal and performs initial channel estimation to establish a communication link.
[0118] Based on the initial channel estimation results, a wide beam is used for coverage to ensure that users can successfully receive downlink signals in the initial stage.
[0119] Step 303: Bi-LSTM feature extraction; The RIS controller collects channel state information from the past T time slots (T=10) and constructs a time-series CSI matrix. .
[0120] Convert the CSI matrix into a sequence of eigenvectors. Each It includes channel gain, phase information, and Doppler characteristics.
[0121] Bi-LSTM performs both forward and backward time series modeling simultaneously, outputting the predicted angle of arrival for the user at the next time step. and prediction variance .
[0122] Step 304: PINN physical constraint correction; According to the radar range equation Calculate the theoretical received power corresponding to the predicted location. .
[0123] Actual measured received power Compared with theoretical value The deviation is used as the physical constraint loss, and the prediction output of the Bi-LSTM is corrected by backpropagation.
[0124] After physical constraint correction, the final angle prediction value is output. This ensures that the prediction conforms to the physical laws of electromagnetic wave propagation.
[0125] Step 305: Multi-RIS Collaborative Tracing; All RIS in the system run the PINN tracker in parallel, and each RIS independently estimates the user's angle of arrival relative to that RIS. and prediction variance .
[0126] The edge controller collects the prediction results of each RIS, selects the RIS with the smallest prediction variance as the master RIS (denoted as n*), and marks the other RIS as auxiliary roles.
[0127] fusion weights are calculated based on the prediction variance. We then perform a weighted average of the prediction angles of each RIS to obtain the globally optimal angle estimate.
[0128] The fused angle is sent to each RIS controller, and the RIS adjusts the reflection phase according to the predicted angle to achieve dynamic beam tracking and alignment.
[0129] Output the predicted user angle value at the next time step. and prediction confidence .
[0130] In some embodiments, when a user moves within the coverage area of multiple RIS, the present invention proposes a collaborative tracking strategy based on weighted geometric centers: Each RIS operates an independent PINN tracker to estimate the user's angle of arrival relative to that RIS.
[0131] The edge controller collects angle estimates from each RIS and obtains the globally optimal tracking result through weighted fusion. In some embodiments, during multi-RIS collaborative tracking, a weighted average is calculated based on the confidence level of each RIS's prediction results. RIS with higher confidence levels (smaller prediction variance) receive greater fusion weights, achieving globally optimal tracking.
[0132] in, For the first Fusion weights of each RIS (with prediction variance) (inversely proportional) To prevent small constants from being divided by zero.
[0133] It should be noted that the beneficial effects of the embodiments of this application compared to the prior art are as follows: First, existing DOA-MUSIC tracking schemes suffer from a sharp drop in estimation accuracy in high-dynamic scenarios and are only applicable to single RIS scenarios. This invention, for the first time, embeds physical constraints into a neural network, maintaining effective tracking even with an estimation error within 20%.
[0134] Second, existing LSTM tracking schemes rely on offline training with large amounts of historical data, resulting in poor generalization ability in new deployment environments. PINN significantly reduces the amount of training data required by introducing physical priors (convergence is achieved with only 100 samples).
[0135] Third, the multi-RIS collaborative tracking mechanism is the first in the industry to consider the impact of different RIS geometric topologies on tracking accuracy, achieving adaptive weighted fusion.
[0136] It should be noted that steps S1-S3 form a closed-loop collaboration according to the temporal logic of initial alignment, steady-state optimization, and dynamic tracking. During the user access phase, the TA-HC codebook collaborative search first completes the initial beam alignment (taking approximately 5ms), providing the basic beam direction for subsequent joint precoding and beam tracking. During the steady-state transmission phase, the Fuzzy-MARL algorithm continuously optimizes the AP precoding matrix and RIS reflection phase based on the initial beam established by TA-HC, maximizing system performance and rate. When the user moves, PINN beam tracking predicts the user's future position in real time and triggers Fuzzy-MARL to readjust the precoding and phase configuration, while TA-HC provides alternative beam directions for rapid switching. These three steps form a closed-loop iteration of alignment, optimization, tracking, and re-optimization, ensuring the system is always in its optimal operating state.
[0137] Furthermore, steps S1-S3 achieve deep collaboration through multi-dimensional information interaction. Fuzzy-MARL provides PINN with the current optimal precoding matrix W* and RIS phase configuration Θ*, which PINN uses to calculate the user-perceived signal quality as feature input for trajectory prediction. PINN feeds back the user's future location prediction and movement speed estimate to Fuzzy-MARL, which adjusts the interference weight parameters α, β, and γ in the reward function accordingly, achieving prediction-driven precoding optimization. TA-HC provides Fuzzy-MARL and PINN with the geometric topology parameters (installation height h) of each RIS. n Orientation angle ψ n Effective aperture D nThis is used to construct a customized beamspace. Fuzzy-MARL and PINN feed back beam tracking success rates and failure scenarios to TA-HC for dynamically updating the codebook resolution configuration. All interaction information is managed uniformly through the CPU's shared memory, with interaction latency controlled within milliseconds.
[0138] Furthermore, steps S1-S3 share the computing and storage resources of the edge AI controller, achieving unified scheduling and management through the CPU. Computational resource sharing: The Fuzzy-MARL policy network, PINN's Bi-LSTM prediction network, and TA-HC's codebook generation algorithm are all deployed on the same edge server, sharing GPU / NPU computing power. The CPU dynamically allocates resources based on the computational load of each step, ensuring that PINN prediction, with the highest real-time requirements, receives priority scheduling. Storage resource sharing: The three steps share the user's CSI historical database, RIS topology parameter library, and precoding policy library, avoiding redundant data storage. Unified scheduling and management: The CPU maintains a cooperative scheduling state machine, defining the triggering conditions and execution priorities of the three steps to avoid resource conflicts and deadlocks.
[0139] In summary, the collaborative operation of steps S1-S3 produces a significant cumulative performance gain. Using Fuzzy-MARL alone: system and rate increase by approximately 40%, but performance decreases by over 20% in mobile scenarios. Using PINN alone: beam tracking error is reduced by 60%, but the lack of joint precoding optimization leads to spectral efficiency loss. Using TA-HC alone: initial access time is reduced by 60%, but steady-state performance is not optimized. The synergy of the three: timing coordination ensures optimal performance at each stage, information exchange enables prediction-driven dynamic optimization, and resource sharing reduces deployment costs. The system-level performance gain after collaboration: system and rate increase of approximately 65% (compared to traditional solutions), performance fluctuation less than 5% in mobile scenarios, initial access time reduced by 70%, system computational resource consumption reduced by 40%, achieving a synergistic effect.
[0140] Example 2, as Figure 5 As shown, this embodiment provides a metasurface-assisted MIMO precoding and beam tracking device, including: The initial alignment module is used to perform initial beam alignment on each smart metasurface based on the installation topology information of each smart metasurface to obtain the initial beam configuration. The joint precoding module is used to perform joint precoding optimization on each access point and each smart metasurface based on the initial beam configuration to obtain a steady-state precoding configuration; The dynamic tracking module is used to dynamically track the beam offset caused by user movement based on the steady-state precoding configuration, and obtain the updated beam configuration. The collaborative scheduling module is used to schedule the initial alignment module, joint precoding module, and dynamic tracking module in a time sequence according to the initial alignment, joint precoding optimization, and dynamic tracking to form a collaborative closed loop.
[0141] In some embodiments, the metasurface-assisted MIMO precoding and beam tracking device of the present invention can be implemented in a combination of hardware and software. As an example, the metasurface-assisted MIMO precoding and beam tracking device of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the metasurface-assisted MIMO precoding and beam tracking method of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.
[0142] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.
[0143] An electronic device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned metasurface-assisted MIMO precoding and beam tracking methods. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store the computer program; the processor is used to execute the metasurface-assisted MIMO precoding and beam tracking method shown in any embodiment of the present invention by calling the computer program.
[0144] In one alternative embodiment, an electronic device is provided, such as Figure 6 As shown, Figure 6The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present invention.
[0145] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.
[0146] Bus 4002 may include a path for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The bus 4002 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.
[0147] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.
[0148] The memory 4003 stores application code (computer program) for executing the present invention, and its execution is controlled by the processor 4001. The processor 4001 executes the application code stored in the memory 4003 to implement the content shown in the foregoing method embodiments.
[0149] Among them, electronic devices can also be terminal devices, which can be any device that can install applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.
[0150] It should be noted that, Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.
[0151] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-described metasurface-assisted MIMO precoding and beam tracking methods.
[0152] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.
[0153] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned metasurface-assisted MIMO precoding and beam tracking method.
[0154] Computer program code for performing the operations of this invention can be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, and C++, and conventional procedural programming languages such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0155] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0156] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EEPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
[0157] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.
[0158] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.
[0159] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.
[0160] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.
[0161] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.
Claims
1. A metasurface-assisted MIMO precoding and beam tracking method, characterized in that, include: Based on the installation topology information of each smart metasurface, the beams of each smart metasurface are initially aligned to obtain the initial beam configuration. Based on the initial beam configuration, joint precoding optimization is performed on each access point and each smart metasurface to obtain a steady-state precoding configuration. Based on the steady-state precoding configuration, the beam offset caused by user movement is dynamically tracked to obtain the updated beam configuration.
2. The metasurface-assisted MIMO precoding and beam tracking method according to claim 1, characterized in that, The initial beam alignment of each smart metasurface based on its installation topology information, to obtain an initial beam configuration, includes: Based on the installation location and orientation information of each smart metasurface, the transformation relationship between the local coordinate system and the global coordinate system of each smart metasurface is constructed; Based on the transformation relationship, a codebook adapted to the installation orientation of each smart metasurface is generated; Based on the codebook, the beams of each smart metasurface are searched to obtain the initial beam configuration.
3. The metasurface-assisted MIMO precoding and beam tracking method according to claim 2, characterized in that, The codebook is a hierarchical codebook, which includes a coarse-grained codebook and a fine-grained codebook. The beam search of each smart metasurface includes: Parallel coarse search is performed based on the coarse-grained codebook to determine the master intelligent metasurface based on the received power fed back by the user terminal; fine search is performed on the master intelligent metasurface based on the fine-grained codebook to determine the optimal beam index; the other auxiliary intelligent metasurfaces calculate their respective optimal beam directions based on their relative positions and orientation differences with the master intelligent metasurface.
4. The metasurface-assisted MIMO precoding and beam tracking method according to claim 1, characterized in that, The step of jointly precoding and optimizing each access point and each smart metasurface based on the initial beam configuration to obtain a steady-state precoding configuration includes: Obtain local channel status information for each access point and each smart metasurface; Based on the local channel state information, the precoding matrix of each access point and the reflection phase of each smart metasurface are obtained through multi-agent collaborative decision-making. Based on expert knowledge, the precoding matrix and the reflection phase are corrected to obtain a steady-state precoding configuration.
5. The metasurface-assisted MIMO precoding and beam tracking method according to claim 4, characterized in that, The multi-agent collaborative decision-making is achieved through a deep reinforcement learning policy network, and the expert knowledge is achieved through a fuzzy logic controller. The fuzzy logic controller converts the output of the deep reinforcement learning policy network into fuzzy linguistic variables based on fuzzy rules, and fuses the deep reinforcement learning output and fuzzy logic correction results through a membership function to obtain a steady-state precoding configuration.
6. The metasurface-assisted MIMO precoding and beam tracking method according to claim 4, characterized in that, Each access point and each smart metasurface independently trains a policy network based on local channel state information, and uploads the policy network parameters to the central processing unit for aggregation. After obtaining the global policy network parameters, they are then distributed to each access point and each smart metasurface. During the inference phase, each access point and each smart metasurface makes independent decisions based on local observations and does not transmit raw channel state information.
7. The metasurface-assisted MIMO precoding and beam tracking method according to claim 1, characterized in that, The step of dynamically tracking beam offset caused by user movement based on the steady-state precoding configuration to obtain an updated beam configuration includes: Based on the historical channel state information sequence, the user's beam angle of arrival is predicted by a prediction model to obtain the predicted beam direction; The predicted beam direction is corrected based on the physical constraints of electromagnetic wave propagation to obtain an updated beam configuration. The dynamic tracking includes: running the prediction model in parallel on multiple smart metasurfaces, collecting the prediction results and prediction variance of each smart metasurface, calculating the fusion weight based on the prediction variance, and performing a weighted average of the prediction angles of each smart metasurface to obtain the global optimal angle estimate.
8. A metasurface-assisted MIMO precoding and beam tracking device, applied to the method of any one of claims 1-7, characterized in that, include: The initial alignment module is used to perform initial beam alignment on each smart metasurface based on the installation topology information of each smart metasurface to obtain the initial beam configuration. The joint precoding module is used to perform joint precoding optimization on each access point and each smart metasurface based on the initial beam configuration to obtain a steady-state precoding configuration; The dynamic tracking module is used to dynamically track the beam offset caused by user movement based on the steady-state precoding configuration, and obtain the updated beam configuration. The collaborative scheduling module is used to schedule the initial alignment module, joint precoding module, and dynamic tracking module in the order of initial alignment, joint precoding optimization, and dynamic tracking.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.
