A distributed ris cooperation channel estimation and interference management method and system

Abstract

The application provides a distributed RIS cooperation channel estimation and interference management method and system, and belongs to the technical field of channel estimation and interference management. The application combines a distributed timing pilot architecture, a history data driven minimum mean square error Wiener filter interpolation technology based on strict causality, a weighted minimum mean square error iterative collaborative optimization algorithm using spatial diversity, and a Kalman filter algorithm prediction closed-loop control mechanism assisted by a deep learning neural network. Especially for a super large scale multiple input multiple output system, the application can adaptively process the spatial non-stationary characteristics of the array in channel estimation and beamforming optimization, and realizes joint resolution enhancement in the angle domain and the distance domain based on a near-field spherical wave channel model, thereby providing a set of logically rigorous, mathematically complete and engineering implementable channel acquisition and interference management overall patent solution for future 6G large-scale network deployment.

Description

Technical Field

[0001] This invention relates to the field of channel estimation and interference management technology, and in particular to a distributed RIS cooperative channel estimation and interference management method and system. Background Technology

[0002] With the large-scale commercialization of 5G mobile communication networks and the gradual formation of the vision for 6G mobile communication, future wireless communication systems are placing unprecedentedly stringent demands on network edge coverage, instantaneous data transmission rates, massive device connection density, and extremely high mobility support capabilities. To overcome the bottlenecks of traditional radio frequency physical layer technologies, reconfigurable smart surfaces, as a revolutionary wireless communication auxiliary technology, have received widespread attention from academia, industry, and the 3G Partnership standardization organization in recent years.

[0003] Reconfigurable smart surfaces typically consist of a large number of low-cost passive electromagnetic reflective units. Each reflective unit can independently adjust the phase and amplitude of the incident electromagnetic signal through a microcontroller network deployed on its backplane, thereby intelligently reshaping the wireless propagation environment in three-dimensional space. This enables directional signal enhancement, precise suppression of co-channel interference, and expanded coverage in communication dead zones. Compared to traditional active repeaters with RF processing capabilities, reconfigurable smart surfaces do not require expensive and energy-intensive RF links and power amplifiers. They can even achieve energy self-sufficiency through multi-source energy harvesting by integrating photovoltaic panels. These characteristics give them significant advantages such as low power consumption, low manufacturing cost, and ease of widespread deployment on urban building surfaces, making them widely regarded as one of the key physical layer enabling technologies for 6G networks.

[0004] However, existing reconfigurable smart surface-assisted communication systems still face many deep-seated and urgent technical challenges. These technical bottlenecks not only limit their theoretical performance but also hinder their transformation into large-scale commercial applications: First, channel estimation accuracy degrades significantly or even fails under high-mobility scenarios. Accurate Channel State Information (CSI) is a prerequisite for leveraging the capabilities of reconfigurable smart surface beamforming. Most existing smart surface beamforming channel estimation methods adhere to the block fading assumption, which assumes that CSI remains absolutely constant within a complete data frame or coherence time. However, the channel undergoes rapid and drastic changes within a very short timeframe, with significant Doppler shifts causing substantial distortion within a single data frame. Second, pilot contamination and inter-user interference are particularly prominent in multi-user high-concurrency scenarios. Existing research on smart surface beamforming channel estimation is largely limited to idealized scenarios with single or a few users. When the system expands to multi-user high-concurrency scenarios, the number of available orthogonal pilot sequences is extremely limited by the coherence time. Non-orthogonal or multiplexed pilot signals transmitted by different user devices superimpose at the receiving antenna array of the base station computing equipment, leading to severe pilot contamination and preventing accurate decoupling of channel parameters. Third, the lack of cross-base station cooperation mechanisms results in insufficient exploitation of spatial diversity gain. Most existing research and patented solutions for reconfigurable smart surface systems are based on simplified network topologies with single base station coverage. However, in actual dense deployments of cellular networks, user equipment is often located in the overlapping coverage area of ​​multiple base station computing devices, i.e., classic cell edge users. Fourth, there are challenges in electromagnetic interference and coordinated scheduling between arrays in densely deployed distributed reconfigurable smart surface networks. As communication networks evolve towards higher frequencies, such as millimeter waves and terahertz, signal penetration capabilities decrease dramatically. A single reconfigurable smart surface node is obviously insufficient to meet the requirements for wide-area coverage and signal blind spot elimination in complex urban canyon environments. Therefore, multiple distributed reconfigurable smart surface arrays will inevitably be densely deployed in the network. Under the existing research framework, when multiple reconfigurable smart surfaces are in operation simultaneously, if their unit activation modes and digital phase configurations lack the support of strict network-level timing control and collaborative joint optimization algorithms, the electromagnetic waves reflected by different reconfigurable smart surface panels will cause destructive interference in space, resulting in severe inter-array interference and severely deteriorating the signal-to-interference-plus-noise ratio at the receiver. Fifth, there is a deep-seated contradiction between the channel estimation hardware processing delay and the real-time performance of the joint optimization algorithm. The core premise for reconfigurable smart surfaces to provide performance gains is that their hardware phase deflection operation must be highly matched with extremely accurate and causal current channel state information.Since the transmission and reception of pilot signals, the separation and calculation of cascaded channel matrices, and the highly complex joint optimization algorithms all require considerable computation time, by the time these calculations are completed and the configuration instructions are sent to the reconfigurable smart surface hardware nodes via the wired control link, the actual physical channel environment has already changed in high mobility scenarios, causing the calculated "optimal configuration" to become an "outdated configuration" or even a "degraded configuration" at the time of application.

[0005] More importantly, to meet the extreme demand for higher spectral efficiency in 6G networks, base station antenna arrays are rapidly evolving from traditional massive MIMO to ultra-massive MIMO (Giga MIMO). While Giga MIMO brings huge performance gains, it also introduces new technical challenges: First, the number of antennas surges to the thousands, resulting in extremely high channel matrix dimensions. The computational complexity of traditional channel estimation and precoding algorithms increases exponentially, making it difficult to meet real-time processing requirements. Second, the massive array physical aperture fundamentally changes the electromagnetic wave propagation model, moving from the far-field plane wave assumption to the near-field spherical wave domain, severely reducing the accuracy of traditional angle-domain-based far-field beamforming methods. Third, the spatial non-stationarity of the channel—that is, different parts of the array may observe completely different sets of scatterers—becomes a significant characteristic of Giga MIMO systems. Traditional channel estimation methods based on the full array statistical stationarity assumption will completely fail due to model mismatch. These new challenges are deeply coupled with the aforementioned five major problems, posing unprecedentedly higher requirements for the design of RIS-assisted ultra-massive MIMO communication systems. Summary of the Invention

[0006] To address the problems in the prior art, this invention provides a distributed RIS collaborative channel estimation and interference management method and system. This invention integrates a distributed time-series pilot architecture, a Wiener filter interpolation technique driven by historical data with strict causal relationships, an iterative collaborative optimization algorithm utilizing spatial diversity and weighted minimum mean square error (WMMSE), and a Kalman filter algorithm prediction closed-loop control mechanism assisted by deep learning neural networks. In channel estimation and beamforming optimization, it can adaptively handle the spatial non-stationary characteristics of Giga MIMO arrays and achieve joint resolution enhancement in the angle and range domains based on an accurate near-field spherical wave channel model. This provides a logically rigorous, mathematically sound, and engineering-feasible overall patented solution for channel acquisition and interference management for future large-scale 6G network deployments. To achieve the above objectives, the technical solution is as follows: On one hand, this invention provides a distributed RIS cooperative channel estimation and interference management method, the method comprising: S1. Based on a multi-user, multi-base station ultra-large-scale MIMO system, multiple pilot subframes are obtained through distributed pilot design; S2. Based on the multiple pilot subframes, perform MMSE interpolation estimation for direct links and reflected links based on the pilots to obtain the local channel estimation matrix and the local estimation mean square error matrix. S3. Based on the local channel estimation matrix and the local estimation mean square error matrix, the fused global channel estimation is obtained through weighted fusion. S4. Based on the fused global channel estimation, an end-to-end equivalent channel is constructed based on the near-field spherical wave channel model. Through a joint optimization algorithm, the optimized precoding matrix and the optimized RIS phase configuration are obtained. S5. Based on the optimized precoding matrix and the optimized RIS phase configuration, the direct link channel vector and the reflection link concatenated channel coefficients are obtained through the neural network residual compensation Kalman filter algorithm. S6. Based on the direct link channel vector and the reflection link cascaded channel coefficients, ensure that the user rate does not change abruptly through configuration and selection.

[0007] Optionally, in S1, based on a multi-user, multi-base station ultra-large-scale MIMO system, multiple pilot subframes are obtained through distributed pilot design, including: S11. Based on the multi-user, multi-base station ultra-large-scale MIMO system, assign a ZC sequence of length L to each base station to obtain the pilot sequence of each base station; S12. Based on the pilot sequence of each base station, multiple pilot subframes are obtained by dividing the large-scale path loss of the user.

[0008] Optionally, in S2, based on the multiple pilot subframes, MMSE interpolation estimation of direct links and reflection links is performed based on the pilots to obtain the local channel estimation matrix and the local estimation mean square error matrix, including: S21. Based on the multiple pilot subframes, all RIS units in the multi-user multi-base station ultra-large-scale MIMO system are configured to be off, and the user's direct link pilot signal is obtained. S22. Based on the user's direct link pilot signal, MMSE interpolation estimation is used to obtain the local channel estimation matrix; S23. Based on the multiple pilot subframes, the RIS unit is configured to be in the open state, resulting in the superposition of the direct link signal and the reflected link signal; S24. Based on the superposition of the direct link signal and the reflected link signal and the user's direct link pilot signal, obtain the net observation value of the reflected link; S25. Based on the local channel estimation matrix and the net observation of the reflection link, MMSE interpolation estimation is used to obtain the local estimation mean square error matrix.

[0009] Optionally, the formula for the weighted fusion includes: (1) In the formula, For base stations The local channel estimation matrix, This is the corresponding local estimation mean square error matrix. This is the global channel estimation after fusion.

[0010] Optionally, in S4, based on the fused global channel estimate, an end-to-end equivalent channel is constructed based on the near-field spherical wave channel model. Through a joint optimization algorithm, the optimized precoding matrix and the optimized RIS phase configuration are obtained, including: S41. Based on the fused global channel estimation, an end-to-end equivalent channel is constructed based on the near-field spherical wave channel model to obtain the initialized precoding matrix and the initialized RIS phase configuration. S42. By fixing the initial RIS phase configuration, the WMMSE algorithm is used to iteratively update the initial precoding matrix to obtain the precoding matrix of the first stage. S43. By fixing the precoding matrix of the first stage, the initial RIS phase configuration is iteratively updated using the block coordinate descent method to obtain the RIS phase configuration of the first stage. S44. Repeat S42~S43 and make a judgment. If convergence is achieved or the maximum number of iterations is reached, output the optimized precoding matrix and the optimized RIS phase configuration.

[0011] Optionally, in S5, based on the optimized precoding matrix and the optimized RIS phase configuration, a neural network residual compensation Kalman filter algorithm is used to obtain the direct link channel vector and the reflection link concatenated channel coefficients, including: S51. Based on the optimized precoding matrix and the optimized RIS phase configuration, the enhanced prediction value is obtained through a neural network algorithm. S52. Based on the enhanced prediction value, the posterior state estimate and posterior error covariance are obtained by calculating the Kalman gain. S53. Based on the optimized precoding matrix, the optimized RIS phase configuration, the posterior state estimation, and the posterior error covariance, the direct link channel vector and the reflection link concatenated channel coefficients are obtained.

[0012] On the other hand, the present invention provides a distributed RIS cooperative channel estimation and interference management system, which is applied to a distributed RIS cooperative channel estimation and interference management method. The system includes: The pilot subframe acquisition module is used to obtain multiple pilot subframes based on a multi-user, multi-base station ultra-large-scale MIMO system through distributed pilot design. The local estimation module is used to perform MMSE interpolation estimation of direct link and reflection link based on the multiple pilot subframes, and obtain the local channel estimation matrix and the local estimation mean square error matrix. The global channel estimation module is used to obtain the fused global channel estimate by weighted fusion based on the local channel estimation matrix and the local estimation mean square error matrix. The joint optimization module is used to construct an end-to-end equivalent channel based on the near-field spherical wave channel model according to the fused global channel estimate, and obtain the optimized precoding matrix and the optimized RIS phase configuration through the joint optimization algorithm. The compensation module is used to obtain the direct link channel vector and the reflection link concatenated channel coefficients based on the optimized precoding matrix and the optimized RIS phase configuration using a neural network residual compensation Kalman filter algorithm. The configuration module is used to ensure that the user rate does not change abruptly by configuring and selecting the direct link channel vector and the reflection link cascaded channel coefficients.

[0013] Compared with the prior art, the technical solution of the present invention has at least the following beneficial effects: The above-mentioned scheme effectively addresses the challenges of extremely high dimensional computational complexity and spatial non-stationarity in ultra-large-scale MIMO systems. First, it introduces a user clustering mechanism based on large-scale path loss and spatial covariance similarity. Second, it adopts a per-user, per-antenna distributed MMSE channel estimation framework, which does not make mandatory assumptions about the spatial stationarity of the array channel and can adaptively learn the channel statistical characteristics of each antenna port and the spatial non-stationary channel characteristics of Giga MIMO, thus avoiding the serious performance loss caused by model mismatch in traditional methods. Furthermore, based on an accurate near-field spherical wave channel model, an end-to-end equivalent channel is constructed, fully utilizing the joint spatial resolution capability of the Giga MIMO array in the angle and range domains to achieve more refined spatial cancellation of multi-user interference at the same frequency. This is a technical effect that traditional large-scale MIMO methods under the far-field plane wave assumption cannot achieve. Secondly, by using pilot-based MMSE interpolation estimation for direct and reflected links, Doppler distortion is effectively overcome, improving the accuracy of direct link channel estimation in high-dynamic environments. Thirdly, through distributed pilot design, multi-user pilot pollution and RIS cascaded reflection interference are completely eliminated. Fourthly, through serial interference cancellation, clean extraction and estimation of reflected link channels are achieved, overcoming the passive characteristics of RIS. Fifthly, by adopting a hybrid algorithm of WMMSE and coordinate descent under the block coordinate descent framework, global weighting and rate maximization are achieved with controllable computational complexity. Sixthly, through neural network residual compensation Kalman filter algorithm, channel prediction is performed, fully compensating for processing and transmission delays, and achieving "synchronization control" in high-dynamic environments. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of an embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention; Figure 2 This is a flowchart illustrating the process of obtaining multiple pilot subframes in an embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention. Figure 3 This is a flowchart illustrating the process of obtaining the local channel estimation matrix and the local estimation mean square error matrix in an embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention. Figure 4This is a flowchart of the optimized precoding matrix and optimized RIS phase configuration obtained in an embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention; Figure 5 This is a flowchart illustrating the process of obtaining the direct link channel vector and the concatenated channel coefficients of the reflection link in an embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention. Figure 6 This is a schematic diagram of the physical topology of the communication system in an embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention; Figure 7 This is a system block diagram of an embodiment of the distributed RIS cooperative channel estimation and interference management system of the present invention. Detailed Implementation

[0016] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0017] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] like Figure 6 The diagram shown illustrates the physical topology of the communication system in an embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention. The physical topology of the communication system involved in this embodiment includes the following key entities: (1) Base Station Computing Device: K in total. Each base station computing device is equipped with N radio frequency antennas, adopting a very large-scale multiple-input multiple-output (Giga MIMO) radio frequency architecture. The number of antenna elements N is usually 1024 or more, and the physical size of the array is significantly increased. It supports fully digital or hybrid beamforming calculations in the baseband processing unit. All base station computing devices are connected to the central control device located in the network hub through a high-bandwidth, low-latency fiber backhaul link for real-time reporting of status information and receiving coordinated scheduling instructions.

[0020] (2) User Equipment (UE): U in total. All of them are intelligent terminals with high mobility features, such as vehicle communication modules and low-altitude aircraft data links. To simplify the radio frequency complexity of the terminals and reduce power consumption, each user equipment is equipped with a single omnidirectional transceiver antenna.

[0021] (3) Reconfigurable smart surfaces: There are a total of R, which are distributed and deployed on the exterior walls of buildings, street light poles and other infrastructure within the network coverage area. Each reconfigurable smart surface contains M passive electromagnetic reflection units, which adopt a uniform planar array (UPA) physical layout. To achieve control, each array is equipped with a dedicated reconfigurable smart surface controller (RIS Controller, RISC). The controller is directly connected to the central control device through a low-speed control link based on Ethernet or a dedicated leased line. It is responsible for receiving binary digital phase configuration commands issued by the central control device and converting them into bias voltages to drive PIN diodes or varactor diodes.

[0022] (4) Central Control Device: 1 in total. Deployed on the core network side or implemented using a multi-access edge computing (MEC) node, it has powerful matrix operation and parallel processing accelerators, such as graphics processing unit (GPU) or tensor processor (TPU), as well as massive data storage capabilities.

[0023] In this embodiment, the wireless channel modeling between all base stations and user equipment, and between base stations and RIS, is based on an accurate near-field spherical wavefront model to fully reflect the spatial resolution capability of the Giga MIMO array in the range domain. Simultaneously, this method does not impose a mandatory assumption of overall array stationarity regarding the potential spatial non-stationarity of the channel—that is, different antenna elements of the array may experience different propagation environments. Instead, it adaptively addresses this through a user-by-user, antenna-by-antenna distributed MMSE processing framework in subsequent steps, thereby ensuring that the channel estimation accuracy does not degrade in the Giga MIMO scenario.

[0024] like Figure 1 The flowchart shown is an embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention. The present invention provides a distributed RIS cooperative channel estimation and interference management method, which is implemented by a distributed RIS cooperative channel estimation and interference management system. The method includes: S1. Based on a multi-user, multi-base station ultra-large-scale MIMO system, multiple pilot subframes are obtained through distributed pilot design; Specifically, such as Figure 2The flowchart shown in this embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention obtains multiple pilot subframes. In S1, based on a multi-user, multi-base station ultra-large-scale MIMO system, multiple pilot subframes are obtained through distributed pilot design, including: S11. Based on the multi-user, multi-base station ultra-large-scale MIMO system, assign a ZC sequence of length L to each base station to obtain the pilot sequence of each base station; S12. Based on the pilot sequence of each base station, multiple pilot subframes are obtained by dividing the large-scale path loss of the user.

[0025] Furthermore, the central control equipment, at the software level, divides continuous frequency bands and time resources into segments with a length of [missing information]. Physical layer communication data frames (milliseconds in duration). The header of each communication frame contains a pilot subframe region for downlink channel sensing and uplink feedback synchronization; the duration of this subframe is... It contains several consecutive Orthogonal Frequency Division Multiplexing (OFDM) symbols. The number of symbols contained is determined by... Confirmed, among which To round up, The number of bundled control units for each group of reconfigurable smart surface reflective units that are pre-defined and physically adjacent to the system.

[0026] (1) Sequence allocation to eliminate interference between base stations: The central control device generates a complex pseudo-random sequence library in its internal memory and calculates the sequence for each base station. Assign a fixed length The Zadoff-Chu (ZC) constant envelope zero autocorrelation sequence is used as a dedicated probe pilot signal vector. Any two different base station computing devices and The sequences satisfy the time-domain orthogonality constraint: .

[0027] (2) User clustering based on large-scale characteristics: The central control equipment continuously evaluates the large-scale path loss values ​​reported by each user equipment and the similarity characteristics of the spatial channel covariance matrix. Based on these physical attributes, the clustering is performed... Each user equipment is divided into Users are divided into independent user groups. Users with high covariance similarity due to proximity in physical space or overlapping reflection paths are forcibly assigned to different groups. User equipment within different groups uses completely orthogonal independent pilot sequence resources; user equipment within the same group that is spatially far apart uses the same pilot base sequence, but the central control equipment assigns different cyclic shift offset factors to them for the baseband processor to distinguish.

[0028] (3) Time-division multiplexed reconfigurable smart surface activation timing schedule: The central control device issues millisecond-level time-division activation command schedules to each reconfigurable smart surface controller through an independent low-latency control network. Within the first OFDM symbol period of the pilot subframe, all controllers in the central control device's command system... All reflective units of the reconfigurable smart surface are set to a "silent off (OFF)" state, meaning all PIN diodes are adjusted so that the panel's reflection modulus for a specific frequency of electromagnetic waves is approximately 0. From the second OFDM symbol to the... Within one OFDM symbol period, the central control device takes turns directing different physical groups of different reconfigurable smart surfaces in the network, specifying that each group contains Physically adjacent units (the last group may be insufficient) Each packet is sequentially and non-overlappingly activated to the "ON" state on the time axis (the reflection coefficient amplitude jumps to 1). This ensures that within any OFDM symbol period, only one packet block on a specific smart surface reflects the signal, thus avoiding cascaded signal overlap and mixing.

[0029] S2. Based on the multiple pilot subframes, perform MMSE interpolation estimation for direct links and reflected links based on the pilots to obtain the local channel estimation matrix and the local estimation mean square error matrix. Specifically, such as Figure 3 The flowchart shown in the embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention illustrates the process of obtaining the local channel estimation matrix and the local estimation mean square error matrix. In step S2, based on the multiple pilot subframes, MMSE interpolation estimation of direct links and reflected links is performed to obtain the local channel estimation matrix and the local estimation mean square error matrix, including: S21. Based on the multiple pilot subframes, all RIS units in the multi-user multi-base station ultra-large-scale MIMO system are configured to be off, and the user's direct link pilot signal is obtained. S22. Based on the user's direct link pilot signal, MMSE interpolation estimation is used to obtain the local channel estimation matrix; S23. Based on the multiple pilot subframes, the RIS unit is configured to be in the open state, resulting in the superposition of the direct link signal and the reflected link signal; S24. Based on the superposition of the direct link signal and the reflected link signal and the user's direct link pilot signal, obtain the net observation value of the reflected link; S25. Based on the local channel estimation matrix and the net observation of the reflection link, MMSE interpolation estimation is used to obtain the local estimation mean square error matrix.

[0030] Furthermore, during the first or first few start symbols of the pilot subframe, all reconfigurable smart surface units (SSUs) across the network are configured to be forcibly turned off (OFF) by the central control device. At this time, the antenna arrays of each base station computing device only receive pilot signals directly propagated from the target user equipment, serving as the basic observation samples for the direct link. The base station computing device extracts historical frame observation data stored in its local cache, constructs and solves the Wiener-Hoff equation reflecting Doppler characteristics, calculates the optimal interpolation weight matrix for the MMSE (Multi-Level Multi-Segment), and then uses historical causal data to achieve high-precision continuous reconstruction of the direct link channel state information (CSI) at any time within the current frame.

[0031] During the first OFDM symbol period of the pilot subframe of each data frame, all smart surfaces are in an off state, and the base station computing equipment... Received from user equipment The baseband signal can be represented as: (2) in, For base stations Received by the user The transmitted observation signal vector is obtained through ADC sampling; For users The transmitted pilot symbols are known constants; It is additive white Gaussian noise; Let be the direct link channel response vector, and be an unknown quantity.

[0032] To address the Doppler effect caused by high-speed movement, the base station computing device not only uses the instantaneous sample of the current frame, but also calls upon a length of... Historical observation FIFO queue For any time within the current frame Channel prediction value Calculated using the following discrete filtering: (3) in, This is the filter weight matrix. The weights are obtained by solving the two-dimensional Wiener-Hough equations: (4) in, The autocorrelation covariance matrix of the received signal. Let be the cross-correlation matrix between the channel and the received signal. Using the Jakes scattering model, the channel time correlation matrix is ​​described by a zeroth-order Bessel function of the first kind: (5) in, Doppler frequency ( For user relative speed, For carrier frequency, (speed of light) For Rice Factors. The base station computing equipment solves this symmetric positive definite equation system through Gaussian elimination or singular value decomposition to obtain the local channel estimation matrix.

[0033] In the subsequent time slots of the pilot subframe, the timing tables generated by each reconfigurable smart surface segmentally activate its reflection unit groups, and the base station computing device synchronously receives the signal. At this time, the received signal is a superposition of the direct link signal and the RIS cascaded reflection link signal. The local channel estimation matrix obtained by the base station computing device and the known pilot symbols are used to remove the direct link component from the total received signal through signal cancellation, obtaining the net observation value of the reflection link. On this basis, the MMSE filter framework is used again to perform time-domain interpolation estimation on the element-by-element cascaded channel in the product form of "base station-RIS-user". This mechanism decouples the mixed electromagnetic signals through serial interference cancellation, separating the independent channel component controlled by the RIS, thereby solving the hardware limitation that the RIS, as a passive device, cannot actively transmit pilot sequences.

[0034] According to the activation sequence, in the numbered Only the first one on RIS Each reflection unit is in the active state. The corresponding observation symbol time within the pilot subframe. The signal received by the base station computing device is: (6) use After cancellation, the net observation value of the reflection link is obtained: (7) This net observation corresponds to the first-hop channel vector. With the second hop channel scalar The product of these factors is then added, along with residual noise and estimation error. The base station computing device establishes a cyclic array of observations spanning the time dimension for each reflective unit. Then, the MMSE time-domain interpolation filtering framework is called again to obtain the local estimated mean square error matrix.

[0035] S3. Based on the local channel estimation matrix and the local estimation mean square error matrix, the fused global channel estimation is obtained through weighted fusion. Specifically, all base station computing devices report their locally estimated channel estimation matrix and local estimated mean square error matrix to the storage cluster of the central control device via fiber optic backhaul links. The fusion module within the central control device performs multi-source information fusion based on inverse variance weighting, integrating observation data from multiple spatial locations into a fused global channel estimate. Utilizing the macroscopic spatial diversity formed by the physical deployment spacing between different base stations, the system weights the estimation results based on the mean square error reported by each base station, thereby mitigating the problem of local estimation failure at a single base station caused by building obstruction or deep shadow fading.

[0036] The formula for weighted fusion includes: (1) In the formula, For base stations The local channel estimation matrix, This is the corresponding local estimation mean square error matrix. This is the global channel estimation after fusion.

[0037] S4. Based on the fused global channel estimation, an end-to-end equivalent channel is constructed based on the near-field spherical wave channel model. Through a joint optimization algorithm, the optimized precoding matrix and the optimized RIS phase configuration are obtained. Specifically, such as Figure 4 The flowchart shown in this embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention obtains the optimized precoding matrix and the optimized RIS phase configuration. In step S4, based on the fused global channel estimation, an end-to-end equivalent channel is constructed based on the near-field spherical wave channel model. Through a joint optimization algorithm, the optimized precoding matrix and the optimized RIS phase configuration are obtained, including: S41. Based on the fused global channel estimation, an end-to-end equivalent channel is constructed based on the near-field spherical wave channel model to obtain the initialized precoding matrix and the initialized RIS phase configuration. S42. By fixing the initial RIS phase configuration, the WMMSE algorithm is used to iteratively update the initial precoding matrix to obtain the precoding matrix of the first stage. S43. By fixing the precoding matrix of the first stage, the initial RIS phase configuration is iteratively updated using the block coordinate descent method to obtain the RIS phase configuration of the first stage. S44. Repeat S42~S43 and make a judgment. If convergence is achieved or the maximum number of iterations is reached, output the optimized precoding matrix and the optimized RIS phase configuration.

[0038] Furthermore, the central control unit uses the fused global channel estimate as input to establish a non-convex optimization model with the objective function of maximizing the system's global weighted sum rate (WSR). The optimization process is subject to two physical constraints: the maximum transmit power limit of the base station radio frequency, and the constant modulus constraint that the RIS reflection units must satisfy due to their passive characteristics. The central control unit employs an alternating iterative optimization framework based on block coordinate descent, sequentially updating the active transmit precoding matrix of the base station and the passive phase matrix of each RIS in each iteration until the increment of the weighted sum rate between two iterations is lower than a preset convergence threshold. This step, through multi-dimensional collaborative optimization, overcomes the limitations of independent processing by a single base station, enabling the computation of the globally optimal beamforming strategy for the entire network.

[0039] For Giga MIMO systems with a large number of antennas, the near-field spherical wavefront effect between the base station and the user / RIS cannot be ignored. The end-to-end equivalent channel in this optimization model... Based on an accurate near-field spherical wave channel model, rather than the traditional far-field plane wave assumption, beam focusing with higher spatial resolution is achieved.

[0040] The optimization model is expressed as follows: (8) Constraints: Constant modulus constraint: ; Power constraints: .

[0041] in, , For user weight, For the equivalent end-to-end channel, For the first The phase diagonal matrix of each RIS For base stations For users The precoded vector.

[0042] The central control unit invokes a hybrid framework combining WMMSE (Wide Mode, Machine, and Spinning) alternating optimization and block coordinate descent. The iterative process includes: (1) Fixed RIS phase matrix Optimize active precoding vectors The original problem is transformed into a multi-user mean square error (MSE) minimization problem through an equivalent transformation. In the first... In this iteration, the following calculations are performed sequentially: Receiver virtual filter coefficients: ; Error amplification weights: ; Transmit precoding vector The power constraint optimization problem with Lagrange multipliers is obtained by solving the problem, where the multipliers are... A binary search is used to determine the power constraint. This yields the precoding matrix for the first stage.

[0043] (2) Fixed active precoding vector Optimize the phase matrix of each RIS Since different RISes are physically far apart, near-field coupling effects are negligible, therefore the optimization of each RIS can be performed independently and in parallel. For each RIS... The coordinate descent method was used to optimize each of its reflection units. With the phases of other elements fixed, the objective function degenerates into one containing only the phases of the other elements. The trigonometric function form can be used to directly obtain the closed-form solution. This ultimately yields the first-stage RIS phase configuration.

[0044] After each macroscopic iteration, check the weighted sum rate increment. If the increment is less than a preset threshold... The iteration stops when the time is right. The final optimized precoding matrix and optimized RIS phase configuration are distributed by the central control device to each base station and RIS controller through fiber optic backhaul and low-speed control links.

[0045] S5. Based on the optimized precoding matrix and the optimized RIS phase configuration, the direct link channel vector and the reflection link concatenated channel coefficients are obtained through the neural network residual compensation Kalman filter algorithm. Specifically, such as Figure 5 The flowchart shown in this embodiment of the distributed RIS cooperative channel estimation and interference management method of the present invention obtains the direct link channel vector and the concatenated channel coefficients of the reflection link. In step S5, based on the optimized precoding matrix and the optimized RIS phase configuration, the direct link channel vector and the concatenated channel coefficients of the reflection link are obtained through a neural network residual compensation Kalman filter algorithm, including: S51. Based on the optimized precoding matrix and the optimized RIS phase configuration, the enhanced prediction value is obtained through a neural network algorithm. S52. Based on the enhanced prediction value, the posterior state estimate and posterior error covariance are obtained by calculating the Kalman gain. S53. Based on the optimized precoding matrix, the optimized RIS phase configuration, the posterior state estimation, and the posterior error covariance, the direct link channel vector and the reflection link concatenated channel coefficients are obtained.

[0046] Furthermore, the central control unit establishes a state-space prediction model with learning capabilities. The optimized precoding matrix and the optimized RIS phase configuration are input into a Kalman filter; simultaneously, a lightweight deep residual neural network is introduced to infer and correct nonlinear dynamic residuals that the Kalman filter cannot capture. After filtering and network correction, the system predicts the target CSI at the start time of the next data frame. Subsequently, calculations are performed in advance based on this prediction. Finally, during the data transmission phase of the current frame, the central control unit sends the pre-calculated phase configuration command and the precoding matrix in parallel to each RIS controller and base station computing device. This compensates for the inherent time delay caused by signal processing and command transmission from the system architecture perspective, ensuring that the RIS phase configuration remains synchronized with the arrival time of electromagnetic waves in a high-dynamic channel environment.

[0047] Furthermore, the central control device establishes a state-space model for each channel: (1) Equation of state:

[0048] (2) Observation equation:

[0049] in, The actual channel state is the hidden variable. Here is the state transition matrix. The channel fusion estimate is the output of step four. Choose a matrix for observation (usually 1) ), and These are process noise and observation noise, respectively.

[0050] At the end of each data frame processing, the central control device performs the following operations: (1) Prediction time update: Calculate the prior state estimate based on the state equation. The prior error covariance is updated. Simultaneously, the pilot data and Doppler features of the most recent frames are input into the deep residual neural network, and the nonlinear residual compensation term is output and superimposed on the prior estimate to obtain the enhanced prediction value.

[0051] (2) Optimization using prediction values: The enhanced prediction values ​​are used as input to calculate the optimal precoding and RIS phase configuration for the next frame in advance.

[0052] (3) Measurement update: When the real pilot signal of the next frame arrives and the observation value is obtained. Then, the Kalman gain is calculated, and the posterior state estimate is updated. And posterior error covariance.

[0053] Based on the optimized precoding matrix, the optimized RIS phase configuration, the posterior state estimate, and the posterior error covariance, the direct link channel vector and the reflection link concatenated channel coefficients are obtained.

[0054] S6. Based on the direct link channel vector and the reflection link cascaded channel coefficients, ensure that the user rate does not change abruptly through configuration and selection.

[0055] like Figure 7 The diagram shown is a system block diagram of an embodiment of the distributed RIS cooperative channel estimation and interference management system of the present invention. The present invention provides a distributed RIS cooperative channel estimation and interference management system, which is applied to a distributed RIS cooperative channel estimation and interference management method. The system includes: a pilot subframe acquisition module, a local estimation module, a global channel estimation module, a joint optimization module, a compensation module, and a configuration module. Specifically, The pilot subframe acquisition module is used to obtain multiple pilot subframes based on a multi-user, multi-base station ultra-large-scale MIMO system through distributed pilot design. The local estimation module is used to perform MMSE interpolation estimation of direct link and reflection link based on the multiple pilot subframes, and obtain the local channel estimation matrix and the local estimation mean square error matrix. The global channel estimation module is used to obtain the fused global channel estimate by weighted fusion based on the local channel estimation matrix and the local estimation mean square error matrix. The joint optimization module is used to construct an end-to-end equivalent channel based on the near-field spherical wave channel model according to the fused global channel estimate, and obtain the optimized precoding matrix and the optimized RIS phase configuration through the joint optimization algorithm. The compensation module is used to obtain the direct link channel vector and the reflection link concatenated channel coefficients based on the optimized precoding matrix and the optimized RIS phase configuration using a neural network residual compensation Kalman filter algorithm. The configuration module is used to ensure that the user rate does not change abruptly by configuring and selecting the direct link channel vector and the reflection link cascaded channel coefficients.

[0056] This invention provides a distributed RIS (Recognition-Induced Streaming) cooperative channel estimation and interference management method. First, it effectively overcomes Doppler distortion and improves the accuracy of direct link channel estimation in high-dynamic environments by using pilot-based MMSE interpolation estimation for both direct and reflected links. Second, through distributed pilot design, it completely eliminates multi-user pilot pollution and RIS cascaded reflection interference. Third, it achieves clean extraction and estimation of the reflected link channel through serial interference cancellation, overcoming the passive characteristics of RIS. Then, based on a near-field spherical wave channel model, it achieves global weighted sum maximization and rate maximization using a hybrid algorithm of WMMSE and coordinate descent within a block coordinate descent framework, with controllable computational complexity. Finally, it predicts the channel using a neural network residual compensation Kalman filter algorithm, completely compensating for processing and transmission delays, and achieving "synchronization control" in high-dynamic environments. This method can adaptively handle the spatial non-stationary characteristics of Giga MIMO arrays and is a complete solution for 6G ultra-large-scale antenna systems.

[0057] It is understood that the present invention has been described through the above embodiments and should not be construed as limiting the implementation and scope of the present invention. Those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the present invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A distributed RIS cooperative channel estimation and interference management method, characterized in that, The method includes: S1. Based on a multi-user, multi-base station ultra-large-scale MIMO system, multiple pilot subframes are obtained through distributed pilot design; S2. Based on the multiple pilot subframes, perform MMSE interpolation estimation for direct links and reflected links based on the pilots to obtain the local channel estimation matrix and the local estimation mean square error matrix. S3. Based on the local channel estimation matrix and the local estimation mean square error matrix, the fused global channel estimation is obtained through weighted fusion. S4. Based on the fused global channel estimation, an end-to-end equivalent channel is constructed based on the near-field spherical wave channel model. Through a joint optimization algorithm, the optimized precoding matrix and the optimized RIS phase configuration are obtained. S5. Based on the optimized precoding matrix and the optimized RIS phase configuration, the direct link channel vector and the reflection link concatenated channel coefficients are obtained through the neural network residual compensation Kalman filter algorithm. S6. Based on the direct link channel vector and the reflection link concatenated channel coefficients, ensure that the user rate does not change abruptly through configuration and selection.

2. The distributed RIS cooperative channel estimation and interference management method according to claim 1, characterized in that, In S1, based on a multi-user, multi-base station ultra-large-scale MIMO system, multiple pilot subframes are obtained through distributed pilot design, including: S11. Based on the multi-user, multi-base station ultra-large-scale MIMO system, assign a ZC sequence of length L to each base station to obtain the pilot sequence of each base station; S12. Based on the pilot sequence of each base station, multiple pilot subframes are obtained by dividing the large-scale path loss of the user.

3. The distributed RIS cooperative channel estimation and interference management method according to claim 1, characterized in that, In step S2, based on the multiple pilot subframes, MMSE interpolation estimation is performed on the pilots for direct links and reflected links to obtain the local channel estimation matrix and the local estimation mean square error matrix, including: S21. Based on the multiple pilot subframes, all RIS units in the multi-user multi-base station ultra-large-scale MIMO system are configured to be off, and the user's direct link pilot signal is obtained. S22. Based on the user's direct link pilot signal, MMSE interpolation estimation is used to obtain the local channel estimation matrix; S23. Based on the plurality of pilot subframes, the RIS unit is configured to be in the open state to obtain the superposition of the direct link signal and the reflected link signal; S24. Based on the superposition of the direct link signal and the reflected link signal and the user's direct link pilot signal, obtain the net observation value of the reflected link; S25. Based on the local channel estimation matrix and the net observations of the reflection link, MMSE interpolation estimation is used to obtain the local estimation mean square error matrix.

4. The distributed RIS cooperative channel estimation and interference management method according to claim 1, characterized in that, The formula for weighted fusion includes: （1） In the formula, For base stations The local channel estimation matrix, This is the corresponding local estimation mean square error matrix. This is the global channel estimation after fusion.

5. The distributed RIS cooperative channel estimation and interference management method according to claim 1, characterized in that, In step S4, based on the fused global channel estimate, an end-to-end equivalent channel is constructed using a near-field spherical wave channel model. Through a joint optimization algorithm, the optimized precoding matrix and optimized RIS phase configuration are obtained, including: S41. Based on the fused global channel estimation, construct an end-to-end equivalent channel based on the near-field spherical wave channel model to obtain the initialized precoding matrix and the initialized RIS phase configuration. S42. By fixing the initial RIS phase configuration, the initial precoding matrix is ​​iteratively updated using the WMMSE algorithm to obtain the precoding matrix of the first stage. S43. By fixing the precoding matrix of the first stage, the initial RIS phase configuration is iteratively updated using the block coordinate descent method to obtain the RIS phase configuration of the first stage. S44. Repeat S42~S43 and make a judgment. If convergence is achieved or the maximum number of iterations is reached, output the optimized precoding matrix and the optimized RIS phase configuration.

6. The distributed RIS cooperative channel estimation and interference management method according to claim 1, characterized in that, In step S5, based on the optimized precoding matrix and the optimized RIS phase configuration, a neural network residual compensation Kalman filter algorithm is used to obtain the direct link channel vector and the reflection link concatenated channel coefficients, including: S51. Based on the optimized precoding matrix and the optimized RIS phase configuration, an enhanced prediction value is obtained through a neural network algorithm; S52. Based on the enhanced prediction value, the posterior state estimate and posterior error covariance are obtained by calculating the Kalman gain. S53. Based on the optimized precoding matrix, the optimized RIS phase configuration, the posterior state estimation, and the posterior error covariance, the direct link channel vector and the reflection link concatenated channel coefficients are obtained.

7. A distributed RIS cooperative channel estimation and interference management system, used to implement the distributed RIS cooperative channel estimation and interference management method as described in any one of claims 1-6, characterized in that, The system includes: The pilot subframe acquisition module is used to obtain multiple pilot subframes based on a multi-user, multi-base station ultra-large-scale MIMO system through distributed pilot design. The local estimation module is used to perform MMSE interpolation estimation of direct link and reflection link based on the multiple pilot subframes to obtain the local channel estimation matrix and the local estimation mean square error matrix. The global channel estimation module is used to obtain the fused global channel estimate by weighted fusion based on the local channel estimation matrix and the local estimation mean square error matrix. The joint optimization module is used to construct an end-to-end equivalent channel based on the near-field spherical wave channel model according to the fused global channel estimation, and obtain the optimized precoding matrix and the optimized RIS phase configuration through the joint optimization algorithm. The compensation module is used to obtain the direct link channel vector and the reflection link concatenated channel coefficients based on the optimized precoding matrix and the optimized RIS phase configuration using a neural network residual compensation Kalman filter algorithm. The configuration module is used to ensure that the user rate does not change abruptly by configuring and selecting the direct link channel vector and the reflection link concatenated channel coefficients.

Images