A mobile communication method and system based on tensor calculation

Abstract

The application discloses a kind of mobile communication method and system based on tensor calculation, comprising: obtaining user multidimensional channel observation data, constructs tensor and detects motion disturbance event;After detecting disturbance, according to sensitivity determines dimension reset area and executes ablation;Local self-calibration is executed, pilot is detected, structure is fitted and tensor is compensated;Time frequency, antenna multipath and motion coupling core are constructed, and coupling decomposition is executed;Stable structure fragment is divided, graph structure is constructed and reversible shift operator is established;According to motion parameter, channel update reconstruction is carried out by calling shift operator.The application realizes the stable acquisition and accurate reconstruction of channel structure in high-speed mobile scene by multidimensional channel tensor modeling, local reset self-calibration mechanism and cross-dimension reversible structure update.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, and particularly relates to a mobile communication method and system based on tensor calculation. BACKGROUND

[0002] With the evolution of mobile communication systems to 5G and 6G, the channel presents strong correlation and fast time-varying characteristics in multiple dimensions such as time, frequency, antenna space and multipath delay. Especially in the case of high-speed movement, direction mutation or cell switching, the channel structure will be significantly disturbed. The existing technology generally uses a multi-dimensional tensor model with fixed topology to uniformly express the channel, but this kind of model assumes that the structure of each dimension is stable and will not change significantly with the motion state, which leads to the inability to effectively capture the rearrangement of multipath clusters, the mutation of antenna correlation and the cross-dimension linkage effect. When the channel is severely disturbed, the traditional tensor model often has difficulty in maintaining structural consistency, and the estimation error will accumulate inside the tensor and further affect channel prediction and resource scheduling.

[0003] In terms of channel updating and reconstruction, the existing tensor decomposition method is still mainly based on global factor updating, and lacks targeted processing of local abnormal areas. When part of the dimensions are affected by strong disturbance, global decomposition is easily contaminated, which not only makes it difficult to suppress error propagation, but also cannot achieve rapid recovery of local areas. Existing decomposition methods usually only process the data structure of the dimension itself, and cannot express the coupling relationship between time-frequency, antenna-multipath and time-motion dimensions, nor can they identify the cross-dimension structure changes triggered by motion. With the intensification of environmental changes, these deficiencies lead to obvious limitations in channel reconstruction accuracy and real-time performance of the existing technology.

[0004] Traditional technology regards the time evolution of the channel as a continuous updating process of factors, lacks the modeling ability of reversible relationship between structure fragments, and cannot describe the evolution law of the channel with motion through the migration path between structure patterns. When the user accelerates, turns or switches coverage, the channel structure often presents a stage transition, but the existing method cannot record the change path between different structure patterns, nor can it select the optimal evolution mode based on the motion parameters, so it is difficult to maintain the stability and traceability of channel updating.

[0005] Therefore, how to provide a mobile communication method and system based on tensor calculation is a problem that those skilled in the art need to solve. SUMMARY

[0006] One purpose of the present application is to propose a mobile communication method and system based on tensor calculation, which realizes accurate expression and stable update of channel structure in high-speed mobile environment through technologies such as multi-dimensional channel tensor modeling, dimension resetting and local self-calibration, cross-dimensional coupling kernel decomposition and reversible shift compensation graph.

[0007] A mobile communication method based on tensor calculation according to an embodiment of the present application comprises:

[0008] Obtaining multi-dimensional channel observation data of a target user within a continuous time window, constructing a multi-dimensional channel tensor, and detecting a motion disturbance event;

[0009] After detecting the motion disturbance event, determining a dimension resetting area according to the disturbance sensitivity of the multi-dimensional channel tensor in each dimensional sub-area, and performing dimension ablation operation on the tensor data in the dimension resetting area;

[0010] Based on the dimension resetting area, performing local self-calibration, generating a calibrated local tensor through local pilot detection, local structure fitting and tensor compensation processing, fusing the local tensor with the multi-dimensional channel tensor, and obtaining a calibrated channel tensor;

[0011] Based on the calibrated channel tensor, constructing cross-dimensional coupling kernels including time and frequency coupling kernels, antenna and multipath coupling kernels, and time and motion characteristic coupling kernels, performing cross-dimensional coupling kernel decomposition on the calibrated channel tensor, and obtaining a coupling decomposition result;

[0012] According to the coupling decomposition result, dividing the channel structure into a plurality of stable structure fragments and constructing a reversible shift compensation graph, defining each structure fragment as a graph node, and representing the structure shift relationship between the nodes as a reversible shift operator;

[0013] When the target user moves or switches cells, selecting a target node path of the reversible shift compensation graph based on the motion parameters, calling the corresponding reversible shift operator to perform structure update on the channel tensor, generating an updated channel tensor, performing channel reconstruction, and using the channel reconstruction result for beamforming, resource allocation and modulation and coding scheme configuration of the downlink.

[0014] Optionally, the multi-dimensional channel observation data comprises time-domain channel observation data, frequency-domain channel observation data, antenna spatial-domain channel observation data, and multi-path delay-domain channel observation data.

[0015] Optionally, the constructing the multi-dimensional channel tensor comprises:

[0016] According to the time-domain channel observation data, the frequency-domain channel observation data, the antenna spatial-domain channel observation data, and the multi-path delay-domain channel observation data, channel sampling results of each dimension are arranged and combined according to preset time dimension, frequency dimension, antenna dimension, and multi-path dimension to form a multi-dimensional channel tensor.

[0017] Based on the multi-dimensional channel tensor, structural variation quantities of the tensor in adjacent time windows are calculated, including amplitude variation quantity, phase variation quantity, antenna correlation variation quantity, and multi-path cluster energy distribution variation quantity.

[0018] According to speed variation, acceleration variation, orientation variation, and cell switching events of the user, the structural variation quantities are compared with a preset disturbance threshold value, and if the structural variation quantities exceed the disturbance threshold value, it is determined that a motion disturbance event occurs.

[0019] Optionally, the determining the dimension reset area according to the disturbance sensitivity of the multi-dimensional channel tensor in each dimension sub-area comprises:

[0020] Based on the multi-dimensional channel tensor, the multi-dimensional channel tensor is segmented according to preset granularity in the time dimension, the frequency dimension, the antenna dimension, and the multi-path dimension, and the multi-dimensional channel tensor is divided into a plurality of dimension sub-areas, each of which corresponds to a group of continuous time intervals, frequency intervals, antenna sets, and multi-path sets.

[0021] For each dimension sub-area, a disturbance sensitivity index representing a motion disturbance degree of the dimension sub-area is calculated by using channel amplitude variation, phase variation, antenna correlation variation, and multi-path energy distribution variation in adjacent time windows, in combination with user speed, acceleration, orientation variation, and cell switching events.

[0022] According to a current motion state, a disturbance sensitivity threshold value is set, dimension sub-areas with a disturbance sensitivity index not lower than the disturbance sensitivity threshold value are marked as candidate reset sub-areas, adjacent or partially overlapping candidate reset sub-areas in the time dimension, the frequency dimension, the antenna dimension, or the multi-path dimension are merged to form a plurality of candidate dimension reset area clusters.

[0023] For each candidate dimension reset cluster, the expected benefits of performing dimension reset in the candidate dimension reset cluster are evaluated according to the local channel reconstruction error prediction, the available pilot resource constraint, and the dimension reset calculation overhead, and under the condition of meeting a preset reset resource budget, one or more candidate dimension reset clusters with the highest expected benefits are selected as the final dimension reset region;

[0024] The corresponding multi-dimensional channel tensor data in the final dimension reset region is subjected to dimension ablation processing, the channel estimation result in the final dimension reset region is cleared or marked as a reconstruction required state, and reset marking information corresponding to the multi-dimensional channel tensor position is generated.

[0025] Optionally, the fusing of the local tensor and the multi-dimensional channel tensor to obtain the calibrated channel tensor comprises:

[0026] Based on the final dimension reset region and the division results of the time dimension, the frequency dimension, the antenna dimension, and the multipath dimension, a pilot resource set used by the local self-calibration is determined.

[0027] The local pilot signal is transmitted or received at the time position, the frequency position, and the antenna position corresponding to the pilot resource set, and the local channel observation data corresponding to the position of the final dimension reset region is obtained.

[0028] Based on the local channel observation data and the dimension structure of the multi-dimensional channel tensor, the local channel estimation is performed on each dimension sub-region in the final dimension reset region to obtain the local channel estimation result of each dimension sub-region.

[0029] The local channel estimation result of each dimension sub-region is subjected to local structure fitting to generate local tensor data consistent with the time dimension, the frequency dimension, the antenna dimension, and the multipath dimension, and the multiple local tensor data is combined according to the division form of the final dimension reset region to generate a local calibration tensor.

[0030] Each element in the local calibration tensor is written into the corresponding position of the multi-dimensional channel tensor according to the index position, replacing the original channel estimation result in the final dimension reset region, to obtain the calibrated channel tensor.

[0031] Optionally, the cross-dimension coupling kernel decomposition is performed on the calibrated channel tensor to obtain a coupling decomposition result, comprising:

[0032] Based on the calibrated channel tensor, the time and frequency dimension combination, the antenna and multipath dimension combination, and the time and motion feature dimension combination are determined as the target dimension combination of the cross-dimension coupling processing, and the joint observation data set corresponding to each target dimension combination is extracted from the calibrated channel tensor.

[0033] constructing a time-frequency coupling kernel, an antenna-multipath coupling kernel and a time-motion feature coupling kernel based on the joint observation data set, wherein the time-frequency coupling kernel is composed of multiple kernel units corresponding to different time intervals and frequency subbands, the antenna-multipath coupling kernel is composed of multiple kernel units corresponding to different antenna subarrays and multipath clusters, and the time-motion feature coupling kernel is composed of multiple kernel units corresponding to different motion states and time intervals;

[0034] processing the joint observation data of the calibrated channel tensor in the time dimension and the frequency dimension by using the time-frequency coupling kernel to separate the time-frequency joint variation patterns corresponding to different kernel units into multiple time-frequency coupling components, and generating a first cross-dimension coupling result;

[0035] processing the joint observation data of the first cross-dimension coupling result in the antenna dimension and the multipath dimension by using the antenna-multipath coupling kernel to separate the spatial propagation structures corresponding to different antenna subarrays and multipath clusters into multiple antenna-multipath coupling components, and generating a second cross-dimension coupling result;

[0036] processing the joint observation data of the second cross-dimension coupling result in the time dimension and the motion feature dimension by using the time-motion feature coupling kernel and the motion parameters to generate multiple time-motion feature coupling components, and correlating and aggregating the multiple time-frequency coupling components, the multiple antenna-multipath coupling components and the multiple time-motion feature coupling components to form a coupling decomposition result.

[0037] Optionally, the dividing the channel structure into multiple stable structure segments and constructing a reversible drift compensation graph comprises:

[0038] based on the coupling decomposition result and the calibrated channel tensor, segmenting according to a preset stability determination condition to obtain multiple stable structure segments, each stable structure segment corresponding to continuous time intervals, frequency intervals, antenna sets, multipath sets and motion feature sets;

[0039] allocating a structure segment identifier to each stable structure segment, establishing a corresponding relationship between the structure segment identifier and the corresponding time-frequency coupling components, antenna-multipath coupling components and time-motion feature coupling components to form a structure segment set;

[0040] based on the structure segment set and the motion parameters, analyzing the adjacent relationship, the overlapping relationship and the coupling component variation pattern of any two structure segments in the time intervals, the frequency intervals, the antenna sets, the multipath sets and the motion feature sets to generate a reversible drift operator for describing the channel structure variation between the two structure segments;

[0041] Using structural segment identifiers and reversible drift operators, each stable structural segment is constructed as a graph node, and each reversible drift operator is constructed as a graph edge connecting two graph nodes, forming a reversible drift compensation graph. In the reversible drift compensation graph, the forward drift path information from the source graph node to the target graph node and the reverse drift path information from the target graph node to the source graph node are recorded for each graph edge.

[0042] In the reversible drift compensation graph, weight information is assigned to each edge of the graph. The weight information is determined based on the channel reconstruction error caused by the corresponding reversible drift operator, the computational cost of channel structure update, and the degree of consistency with motion parameters, thus forming a weighted reversible drift compensation graph.

[0043] Optionally, the step of using the channel reconstruction results for downlink beamforming, resource allocation, and modulation and coding scheme configuration includes:

[0044] When the target user enters the current scheduling cycle, obtain the motion parameters and the graph node identifiers corresponding to the structural segments of the previous time step in the weighted reversible drift compensation graph;

[0045] Based on motion parameters and weighted reversible drift compensation graph edge weight information, according to preset path selection rules, the target graph node is selected from the adjacent graph nodes connected to the graph node identifier of the previous time step, and the reversible drift path sequence from the graph node of the previous time step to the target graph node is determined.

[0046] Based on the reversible drift path sequence, the reversible drift operators corresponding to each graph edge in the reversible drift path sequence are called sequentially to update the structure fragment of the previous time step in the calibrated channel tensor. The structure update result is written into the structure fragment position corresponding to the target graph node identifier to obtain the updated channel tensor.

[0047] Based on the updated channel tensor, the channel information in the time dimension, frequency dimension, antenna dimension and multipath dimension is jointly reconstructed to obtain the downlink effective channel information in the current scheduling period;

[0048] Based on the downlink effective channel information, the beamforming weights, time and frequency resource allocation schemes, and modulation and coding scheme configuration parameters corresponding to the target user are determined. The beamforming weights are applied to the base station antenna port, and the time and frequency resource allocation schemes and modulation and coding scheme configuration parameters are used to generate downlink transmission parameters.

[0049] A mobile communication system based on tensor computation according to an embodiment of the present invention includes the following modules:

[0050] The channel construction and detection module is used to acquire multi-dimensional channel observation data, construct multi-dimensional channel tensors, and perform motion perturbation detection on the multi-dimensional channel tensors.

[0051] The reset region determination and ablation module is used to determine the dimension reset region based on the multidimensional channel tensor after detecting motion disturbance, and to perform dimension ablation on the dimension reset region.

[0052] The self-calibration and fusion module is used to perform local self-calibration based on the dimension reset region to obtain the calibrated channel tensor;

[0053] The coupling kernel decomposition module is used to construct a cross-dimensional coupling kernel, perform cross-dimensional coupling kernel decomposition on the calibrated channel tensor, and obtain the coupling decomposition result.

[0054] The structural fragment and compensation graph module is used to generate stable structural fragments based on the coupling decomposition results and construct a reversible drift compensation graph.

[0055] The structure update and downlink configuration module is used to select the target path of the reversible drift compensation map based on motion parameters, update the channel tensor, and perform channel reconstruction and downlink configuration based on the updated channel tensor.

[0056] The beneficial effects of this invention are:

[0057] This invention enables the system to quickly identify structural changes in high-speed moving environments by using multidimensional channel tensor modeling combined with a motion disturbance detection mechanism. Based on the disturbance sensitivity, the system performs dimension reset and local self-calibration on the affected area, effectively suppressing the spread of local disturbances to the global environment and keeping the channel structure stable in complex scenarios. This overcomes the problem that traditional fixed topology models are prone to estimation shift and structural distortion under time-varying conditions.

[0058] This invention utilizes cross-dimensional coupling kernel decomposition to simultaneously extract the dynamic relationships between time-frequency, antenna-multipath, and time-motion features, enabling channel modeling to reflect multi-dimensional linkage characteristics. It can maintain high decomposition accuracy even when the structure changes drastically, making the calibrated channel structure more consistent and predictable, effectively making up for the shortcomings of traditional decomposition methods in handling cross-dimensional disturbances and multi-dimensional coupling.

[0059] This invention further constructs a reversible drift compensation graph, using reversible paths to describe the evolutionary relationships between channel structure segments, thus elevating channel updates from continuous factor iteration to structural path selection. During user movement or handover, the system can select the optimal path based on motion parameters, achieving stable channel updates and reconstruction. Reversibility and traceability reduce the risk of update error accumulation, thereby improving the overall accuracy of downlink beamforming, resource allocation, and modulation / coding configuration, as well as system performance. Attached Figure Description

[0060] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0061] Figure 1 This is a flowchart of a mobile communication method based on tensor computation proposed in this invention;

[0062] Figure 2 This is a schematic diagram of the structure of a mobile communication system based on tensor computation proposed in this invention. Detailed Implementation

[0063] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0064] refer to Figure 1 A mobile communication method based on tensor computation, comprising:

[0065] Acquire multidimensional channel observation data of the target user within a continuous time window, construct a multidimensional channel tensor, and detect motion disturbance events;

[0066] After detecting a motion disturbance event, the dimension reset region is determined based on the disturbance sensitivity of the multidimensional channel tensor in each dimension sub-region, and the dimension ablation operation is performed on the tensor data in the dimension reset region.

[0067] Based on the dimension reset region, local self-calibration is performed. A calibrated local tensor is generated through local pilot detection, local structure fitting, and tensor compensation processing. The local tensor is then fused with the multidimensional channel tensor to obtain the calibrated channel tensor.

[0068] Based on the calibrated channel tensor, cross-dimensional coupling kernels are constructed, including time-frequency coupling kernels, antenna-multipath coupling kernels, and time-motion feature coupling kernels. Cross-dimensional coupling kernel decomposition is performed on the calibrated channel tensor to obtain the coupling decomposition results.

[0069] Based on the coupling decomposition results, the channel structure is divided into multiple stable structural segments and a reversible drift compensation graph is constructed. Each structural segment is defined as a graph node, and the structural drift relationship between nodes is represented as a reversible drift operator.

[0070] When a target user moves or a cell handover occurs, the target node path of the reversible drift compensation graph is selected based on motion parameters, and the corresponding reversible drift operator is called to perform a structural update on the channel tensor, generating an updated channel tensor. Channel reconstruction is then performed, and the channel reconstruction results are used for downlink beamforming, resource allocation, and modulation and coding scheme configuration.

[0071] In this embodiment, the multidimensional channel observation data includes time-domain channel observation data, frequency-domain channel observation data, antenna spatial-domain channel observation data, and multipath delay-domain channel observation data.

[0072] In this embodiment, the construction of a multidimensional channel tensor for detecting motion disturbance events includes:

[0073] Based on the channel observation data in the time domain, frequency domain, antenna spatial domain, and multipath delay domain, the channel sampling results of each dimension are arranged and combined according to the preset time dimension, frequency dimension, antenna dimension, and multipath dimension to form a multidimensional channel tensor.

[0074] Based on the multidimensional channel tensor, the structural changes of the tensor within adjacent time windows are calculated, including amplitude changes, phase changes, antenna correlation changes, and multipath cluster energy distribution changes. Specifically, the calculation of the structural changes of the tensor within adjacent time windows involves:

[0075] Amplitude change: Extract the complex amplitude values ​​of the tensors in the two windows respectively, and then subtract the amplitudes at the same position to obtain a difference map. Summarize the difference maps as a whole, take the average of the absolute values, and the resulting value is the overall amplitude change.

[0076] Phase change: Convert the channel tensors of the two time windows into phase representations respectively, then subtract them element by element and limit the result to the standard range of −π to π. Take the absolute value of all differences and then average them to quantify the overall phase change.

[0077] Antenna correlation change: Calculate the covariance or correlation matrix of each tensor in the antenna dimension in two time windows, subtract the two matrices to obtain the error matrix reflecting the correlation change, and summarize the error matrices to obtain the antenna correlation change.

[0078] Multipath cluster energy distribution change: First, according to the preset multipath cluster division rules, calculate the total energy vector of each cluster within two time windows; then compare the differences between the two vectors in each cluster. The Euclidean distance or information entropy difference index can be used for overall measurement. The result is the multipath cluster energy distribution change.

[0079] Based on changes in user speed, acceleration, orientation, and cell handover events, the amount of structural change is compared with a preset disturbance threshold. If the change exceeds the disturbance threshold, a motion disturbance event is determined to have occurred.

[0080] In this embodiment, the step of determining the dimension reset region based on the perturbation sensitivity of the multidimensional channel tensor in each dimensional sub-region, and performing a dimension ablation operation on the tensor data within the dimension reset region, includes:

[0081] Based on the multidimensional channel tensor, the multidimensional channel tensor is segmented according to a preset granularity in the time dimension, frequency dimension, antenna dimension, and multipath dimension. The multidimensional channel tensor is divided into multiple dimensional sub-regions. Each dimensional sub-region corresponds to a set of continuous time intervals, frequency intervals, antenna sets, and multipath sets. The preset granularity is as follows: time dimension is an interval every 1ms, frequency dimension is a subcarrier block every 180kHz, antenna dimension is a group of 4 adjacent physical antennas, and multipath dimension is a multipath set divided with a delay interval of 0.25μs.

[0082] For each dimensional sub-region, using the channel amplitude variation, phase variation, inter-antenna correlation variation, and multipath energy distribution variation within adjacent time windows, combined with user velocity, acceleration, azimuth variation, and cell handover events, a disturbance sensitivity index characterizing the degree of motion disturbance in the dimensional sub-region is calculated. Specifically, the calculation of the disturbance sensitivity index characterizing the degree of motion disturbance in the dimensional sub-region is as follows:

[0083] For the dimension sub-region, select the channel snapshots of the current time window and the immediately preceding time window, and respectively calculate the mean difference of amplitude, the mean phase drift, the difference of antenna correlation coefficient, and the multipath energy distribution variation rate, and normalize the four differences according to a unified dimension.

[0084] The user speed, acceleration, azimuth change angle and cell handover trigger flag recorded in the same time window are mapped to four weight coefficients, and the weights increase monotonically with the intensity of the movement.

[0085] The normalized channel difference value is multiplied by the corresponding weight coefficients one by one and then the weighted sum is obtained to obtain a single-scale disturbance sensitivity index. The larger the value, the more significant the impact of motion disturbance on the dimensional sub-region.

[0086] Based on the current motion state, a disturbance sensitivity threshold is set, and dimensional sub-regions with disturbance sensitivity indices not lower than the disturbance sensitivity threshold are marked as candidate reset sub-regions. Candidate reset sub-regions that are adjacent or partially overlapping in the time dimension, frequency dimension, antenna dimension, or multipath dimension are merged to form multiple candidate dimensional reset region clusters.

[0087] For each candidate dimension reset region cluster, based on local channel reconstruction error prediction, available pilot resource constraints, and dimension reset computational overhead, the expected benefit of performing dimension reset within the candidate dimension reset region cluster is evaluated. If a preset reset resource budget condition is met, the candidate dimension reset region cluster or multiple candidate dimension reset region clusters with the highest expected benefit are selected as the final dimension reset region. Specifically, the evaluation of the expected benefit of performing dimension reset within the candidate dimension reset region cluster involves:

[0088] For each candidate dimension, the local channel reconstruction error reduction magnitude is predicted based on historical calibration results and current perturbation sensitivity, and the reduction magnitude is quantified as a reconstruction gain value.

[0089] The number of pilot symbols available for the candidate region cluster in this frame and the processing clock cycle requirements are counted. Combined with the preset upper limit of pilot resources and upper limit of computation delay, the resource consumption cost is calculated.

[0090] The expected revenue is the difference between the reconstruction revenue and the resource consumption cost. All candidate clusters are sorted by their expected revenue, and one or more clusters are selected as the final dimension reset area according to the budget threshold from high to low.

[0091] Dimension ablation processing is performed on the multidimensional channel tensor data corresponding to the final dimension reset region. The channel estimation results in the final dimension reset region are cleared or marked as needing reconstruction, and reset mark information corresponding one-to-one with the position of the multidimensional channel tensor is generated.

[0092] In this embodiment, fusing the local tensor with the multidimensional channel tensor to obtain the calibrated channel tensor includes:

[0093] Based on the final dimension reset region and the partitioning results of time dimension, frequency dimension, antenna dimension and multipath dimension, the set of pilot resources used for local self-calibration is determined;

[0094] Send or receive local pilot signals at the time, frequency and antenna positions corresponding to the pilot resource set, and obtain local channel observation data corresponding to the final dimension reset area position;

[0095] Based on local channel observation data and the dimensional structure of the multidimensional channel tensor, local channel estimation is performed on each dimensional sub-region within the final dimensional reset region to obtain the local channel estimation result for each dimensional sub-region. Specifically, the local channel estimation for each dimensional sub-region within the final dimensional reset region is performed as follows:

[0096] Trigger local pilot transmission and reception, and collect high-resolution pilot observations only within the time interval, frequency sub-block, antenna set and multipath delay window corresponding to the target dimension sub-region to form a local channel sampling set;

[0097] Based on this sampling set, the local minimum mean square error-constrained least squares hybrid estimation algorithm is invoked. Combining the low-rank features of neighboring sub-regions and historical priors, the amplitude, phase, and correlation parameters of the current frame of the dimensional sub-region are calculated to generate the initial local channel estimate.

[0098] Perform time-frequency-space joint smoothing and consistency verification on the initial local channel estimate, remove outliers, and output the final local channel estimate result.

[0099] For each dimension sub-region, the local channel estimation results are fitted with local structure to generate local tensor data consistent with the time dimension, frequency dimension, antenna dimension and multipath dimension. Then, multiple local tensor data are combined according to the division form of the final dimension reset region to generate a local calibration tensor.

[0100] Each element in the local calibration tensor is written into the corresponding position of the multidimensional channel tensor according to its index position, replacing the original channel estimation result in the final dimension reset region, and the calibrated channel tensor is obtained.

[0101] In this embodiment, performing cross-dimensional coupling kernel decomposition on the calibrated channel tensor to obtain the coupling decomposition result includes:

[0102] Based on the calibrated channel tensor, the combination of time and frequency dimension, antenna and multipath dimension, and time and motion feature dimension are determined as the target dimension combination for cross-dimensional coupling processing. The joint observation data set corresponding to each target dimension combination is extracted from the calibrated channel tensor.

[0103] Based on the joint observation data set, time and frequency coupling kernel, antenna and multipath coupling kernel, and time and motion feature coupling kernel are constructed. The time and frequency coupling kernel is composed of multiple kernel units corresponding to different time intervals and frequency sub-bands. The antenna and multipath coupling kernel is composed of multiple kernel units corresponding to different antenna subarrays and multipath clusters. The time and motion feature coupling kernel is composed of multiple kernel units corresponding to different motion states and time intervals.

[0104] By using the time-frequency coupling kernel to process the joint observation data of the calibrated channel tensor in the time and frequency dimensions, the time-frequency joint change patterns corresponding to different kernel units are separated into multiple time-frequency coupling components, generating the first cross-dimensional coupling result.

[0105] The joint observation data of the first cross-dimensional coupling result in the antenna dimension and multipath dimension are processed by the antenna and multipath coupling kernel. The spatial propagation structure corresponding to different antenna subarrays and multipath clusters is separated into multiple antenna and multipath coupling components to generate the second cross-dimensional coupling result.

[0106] The joint observation data of the second cross-dimensional coupling result in the time and motion feature dimensions are processed using a time-motion feature coupling kernel and motion parameters to generate multiple time-motion feature coupling components. These multiple time-frequency coupling components, multiple antenna-multipath coupling components, and multiple time-motion feature coupling components are then correlated and aggregated to form a coupling decomposition result. Specifically, the generation of these multiple time-motion feature coupling components involves:

[0107] Based on the target user’s speed, acceleration, heading angle and cell handover label obtained in the current window, a motion feature vector is constructed, and the vector is concatenated with the time series index in a one-to-one correspondence relationship to form a joint observation entry.

[0108] Principal component correlation analysis based on a sliding window is performed on the joint observation items to extract the main correlation directions between the time series and each motion feature, and each main correlation direction is mapped to a time-motion coupling component;

[0109] Orthogonality tests and energy normalization are performed on the coupling components in all main correlation directions. The coupling components with a contribution greater than the threshold are retained as the final set of coupling components for time and motion features.

[0110] In this embodiment, dividing the channel structure into multiple stable structural segments and constructing a reversible drift compensation map includes:

[0111] Based on the coupling decomposition results and the calibrated channel tensor, the structure is segmented according to a preset stability criterion to obtain multiple stable structural segments. Each stable structural segment corresponds to a continuous time interval, frequency interval, antenna set, multipath set, and motion feature set. The segmentation according to the preset stability criterion specifically involves:

[0112] In the coupling decomposition results, the amplitude drift amplitude, phase drift amplitude, coupled component energy fluctuation amplitude, and drift compensation map node switching number are calculated sequentially for each time-frequency-space-multipath-motion joint unit, and the four indicators are compared with the corresponding thresholds.

[0113] Observation units in which all frequency sub-blocks, antenna subarrays and multipath clusters within the continuous time index have all four indicators not exceeding the threshold are merged into the same candidate segment. If any indicator exceeds the limit, the segment boundary is broken at that position.

[0114] The obtained candidate segments are filtered by duration length. Segments with a continuous time interval length greater than the minimum stable duration and containing a complete frequency range, antenna set and multipath set are retained and determined to be stable structure segments.

[0115] Each stable structural segment is assigned a structural segment identifier, and a correspondence is established between the structural segment identifier and the corresponding time and frequency coupling components, antenna and multipath coupling components, and time and motion feature coupling components to form a set of structural segments.

[0116] Based on the set of structural segments and motion parameters, the adjacency, overlap and coupling component change patterns of any two structural segments in time interval, frequency interval, antenna set, multipath set and motion feature set are analyzed, and a reversible drift operator is generated to describe the channel structure change between two structural segments.

[0117] Using structural segment identifiers and reversible drift operators, each stable structural segment is constructed as a graph node, and each reversible drift operator is constructed as a graph edge connecting two graph nodes, forming a reversible drift compensation graph. In the reversible drift compensation graph, the forward drift path information from the source graph node to the target graph node and the reverse drift path information from the target graph node to the source graph node are recorded for each graph edge.

[0118] In the reversible drift compensation graph, weight information is assigned to each edge of the graph. The weight information is determined based on the channel reconstruction error caused by the corresponding reversible drift operator, the computational cost of channel structure update, and the degree of consistency with motion parameters, thus forming a weighted reversible drift compensation graph.

[0119] In this embodiment, the step of using the channel reconstruction result for downlink beamforming, resource allocation, and modulation and coding scheme configuration includes:

[0120] When a target user enters the current scheduling cycle, obtain motion parameters and the graph node identifiers corresponding to the previous time segment in the weighted reversible drift compensation graph. The motion parameters include absolute velocity, absolute acceleration, azimuth rate of change, position coordinate increment, cell handover flag, and expected driving trajectory identifier.

[0121] Based on motion parameters and weighted reversible drift compensation graph edge weights, and following a preset path selection rule, a target graph node is selected from adjacent graph nodes connected to the graph node identifier of the previous time step. This determines the reversible drift path sequence from the previous graph node to the target graph node. The preset path selection rule is as follows:

[0122] Minimize the motion perturbation matching degree cost, minimize the cumulative drift weight cost, satisfy the stable structure preservation threshold, satisfy the real-time update delay constraint, and satisfy the resource budget;

[0123] Based on the reversible drift path sequence, the reversible drift operators corresponding to each graph edge in the reversible drift path sequence are called sequentially to update the structure of the previous time step within the calibrated channel tensor. The structure update result is written to the structure step position corresponding to the target graph node identifier to obtain the updated channel tensor. Specifically, the structure update of the previous time step within the calibrated channel tensor is performed as follows:

[0124] The reversible drift operator of the first graph edge in the reversible drift path sequence is analyzed. The time index, frequency index, antenna index and multipath index of the structure segment at the previous time step are mapped to the drift coordinate system corresponding to the operator, and the coordinate transformation and index displacement are completed.

[0125] The phase correction, power gain and energy redistribution coefficients carried by the reversible drift operator are superimposed on the intermediate structural segment after the coordinate transformation. The phase synchronization, amplitude leveling and energy normalization are performed sequentially on the cross-dimensional coupling relationship to obtain the structural segment after a single drift compensation.

[0126] The structural fragment after single drift compensation is used as input. The reversible drift operators of the remaining graph edges are called in sequence. The coordinate transformation, parameter superposition and energy normalization process is repeated until all operators in the drift path are executed. The final structural fragment is then written to the corresponding position of the target graph node identifier, and the index mapping table of the channel tensor and the fragment state flag are updated to complete the structural update.

[0127] Based on the updated channel tensor, the channel information in the time dimension, frequency dimension, antenna dimension and multipath dimension is jointly reconstructed to obtain the downlink effective channel information in the current scheduling period;

[0128] Based on the downlink effective channel information, the beamforming weights, time and frequency resource allocation schemes, and modulation and coding scheme configuration parameters corresponding to the target user are determined. The beamforming weights are applied to the base station antenna port, and the time and frequency resource allocation schemes and modulation and coding scheme configuration parameters are used to generate downlink transmission parameters.

[0129] refer to Figure 1 A mobile communication system based on tensor computation includes the following modules:

[0130] The channel construction and detection module is used to acquire multi-dimensional channel observation data, construct multi-dimensional channel tensors, and perform motion perturbation detection on the multi-dimensional channel tensors.

[0131] The reset region determination and ablation module is used to determine the dimension reset region based on the multidimensional channel tensor after detecting motion disturbance, and to perform dimension ablation on the dimension reset region.

[0132] The self-calibration and fusion module is used to perform local self-calibration based on the dimension reset region to obtain the calibrated channel tensor;

[0133] The coupling kernel decomposition module is used to construct a cross-dimensional coupling kernel, perform cross-dimensional coupling kernel decomposition on the calibrated channel tensor, and obtain the coupling decomposition result.

[0134] The structural fragment and compensation graph module is used to generate stable structural fragments based on the coupling decomposition results and construct a reversible drift compensation graph.

[0135] The structure update and downlink configuration module is used to select the target path of the reversible drift compensation map based on motion parameters, update the channel tensor, and perform channel reconstruction and downlink configuration based on the updated channel tensor.

[0136] Example 1:

[0137] To verify the feasibility of this invention in practice, it was applied to a 5G-A mobile communication network on a high-speed railway. An actual operating train traveling at 302 km / h was selected as the test platform, covering a distance of approximately 32.5 km. During the test, the train passed through multiple curves and bridges, and the environment exhibited significant multipath variations and obstruction abrupt changes, representing a typical high-speed time-varying channel scenario. The system described in this invention was deployed at the base station side to verify the effectiveness of the channel reset, self-calibration, coupling kernel decomposition, and reversible update mechanisms of this invention under high-speed motion scenarios.

[0138] After the train enters the test section, the system first collects multi-dimensional channel observation data within a continuous time window, including time-domain channel samples with a time resolution of 0.5 ms, frequency-domain channel response with a 15 kHz subcarrier spacing, spatial-domain sampling data of a 64-port antenna array, and multipath delay-domain data extracted based on a 24-channel Rake structure. The system arranges and combines these observations according to four dimensions: time, frequency, antenna, and multipath, forming a multi-dimensional channel tensor with a size of approximately 100×52×64×24. During the train's acceleration to 280 km / h, the amplitude change of the tensor in adjacent windows increases by an average of 47%, antenna correlation decreases by 18%, and the multipath energy distribution shows a significant drift, indicating that the system has entered a disturbance state.

[0139] As the train enters the curve region, the channel experiences strong fluctuations in both the antenna and multipath dimensions. Based on the disturbance sensitivity model, the system automatically identifies subarrays 12–24 in the antenna dimension and path components 6–10 in the multipath dimension as the dimension reset region. Within this region, the original channel estimation can no longer maintain structural consistency; therefore, the system performs dimension ablation on this block, marking it as requiring reconstruction. Based on the local pilot overhead budget (approximately 2.4% of total time resources), the base station transmits local pilot signals at the corresponding time-frequency location in this region. Using measured local channel observation data, the system reconstructs the ablated sub-region using local tensor fitting and interpolation compensation methods. After self-calibration, the local error decreases from 5.8 dB to 1.6 dB, allowing the region to be reintegrated into the global tensor structure.

[0140] The system then performs cross-dimensional coupling kernel decomposition on the calibrated channel tensor. A time-frequency coupling kernel, an antenna-multipath coupling kernel, and a time-motion feature coupling kernel are constructed based on joint observation data. During testing, the time-frequency coupling kernel successfully separated three main Doppler-frequency domain spread modes with center spread coefficients of 31Hz, 54Hz, and 63Hz, respectively. The beam main lobe directions corresponding to the three spatial propagation modes extracted by the antenna-multipath coupling kernel are -18°, 5°, and 27°, respectively, highly consistent with the actual train direction changes. The time-motion feature coupling kernel captured a significant time-motion linkage enhancement phenomenon as the train speed increased from 256km / h to 302km / h, increasing the amplitude of the second cross-dimensional coupling component by 22%.

[0141] After decomposing the coupling kernel, the system divides the channel into multiple stable structural segments based on the coupling results and constructs a reversible drift compensation map. During testing, 14 stable segments were identified, with structural changes between segments described by 29 reversible drift operators. For example, the drift operator between segments S3 and S5 can recover to the other's structure within ±3.5ms, with a reconstruction error of only 1.1dB. When the train passes through the bridge entrance (where a LOS→NLOS transition occurs), the system selects the optimal migration path S2→S4→S7 based on motion parameters, achieving predictable structural updates and successfully using the updated channel tensor for downlink configuration.

[0142] Table 1. Performance Comparison of the Invention and Traditional Methods in High-Speed ​​Railway Scenarios

[0143]

[0144] As shown in Table 1, this invention improves the accuracy and stability of channel processing in a high-speed railway scenario of 302 km / h. Relying on dimensionality reset and local self-calibration mechanisms, this invention effectively suppresses local disturbances caused by high-speed motion, reducing the channel estimation error from 0.184 to 0.113 and the beam direction deviation from 7.4° to 5.4°. This invention can maintain a more stable channel structure representation in complex environments such as multipath drift and decreased antenna correlation, overcoming the shortcomings of traditional fixed topology models that are prone to failure under severe disturbances.

[0145] Regarding link adaptive performance, this invention extracts the dynamic relationships between time-frequency, spatial, and motion features more accurately through a cross-dimensional coupling kernel, reducing configuration errors. Specifically, the MCS misclassification rate is reduced from 6.8% to 2.1%, the reconstruction failure rate caused by multipath energy drift is reduced from 12.3% to 3.7%, and the channel segment recovery time is shortened to 6.8ms. The coupling kernel decomposition and reversible drift compensation graph of this invention can effectively improve the structural pattern recognition and structural change response capabilities, enabling the system to maintain higher link reliability in dynamic environments.

[0146] In terms of overall system performance, this invention increases downlink throughput from 312 Mbps to 372 Mbps under the same bandwidth, an increase of 19.3%. Local structure calibration error is reduced from 5.8 dB to 1.6 dB, and structure pattern recognition accuracy is improved from 82.5% to 94.8%. This invention not only recovers local structures affected by disturbances but also achieves stable channel evolution through reversible paths, comprehensively improving the accuracy and efficiency of downlink resource allocation, beamforming, and link adaptation. In summary, this invention demonstrates significant performance advantages over traditional methods in high-speed, complex multipath, and abrupt structural change scenarios.

[0147] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A mobile communication method based on tensor computation, characterized in that, include: Acquire multidimensional channel observation data of the target user within a continuous time window, construct a multidimensional channel tensor, and detect motion disturbance events; After detecting a motion disturbance event, the dimension reset region is determined based on the disturbance sensitivity of the multidimensional channel tensor in each dimension sub-region, and the dimension ablation operation is performed on the tensor data in the dimension reset region. Based on the dimension reset region, local self-calibration is performed. A calibrated local tensor is generated through local pilot detection, local structure fitting, and tensor compensation processing. The local tensor is then fused with the multidimensional channel tensor to obtain the calibrated channel tensor. Based on the calibrated channel tensor, cross-dimensional coupling kernels are constructed, including time-frequency coupling kernels, antenna-multipath coupling kernels, and time-motion feature coupling kernels. Cross-dimensional coupling kernel decomposition is performed on the calibrated channel tensor to obtain the coupling decomposition results. Based on the coupling decomposition results, the channel structure is divided into multiple stable structural segments and a reversible drift compensation graph is constructed. Each structural segment is defined as a graph node, and the structural drift relationship between nodes is represented as a reversible drift operator. When a target user moves or a cell handover occurs, the target node path of the reversible drift compensation map is selected based on motion parameters, and the corresponding reversible drift operator is called to perform structural updates on the channel tensor, generating an updated channel tensor. Channel reconstruction is then performed, and the channel reconstruction results are used for downlink beamforming, resource allocation, and modulation and coding scheme configuration. The structural update of the channel tensor is specifically performed as follows: The reversible drift operator of the first graph edge in the reversible drift path sequence is analyzed. The time index, frequency index, antenna index and multipath index of the structure segment at the previous time step are mapped to the drift coordinate system corresponding to the operator, and the coordinate transformation and index displacement are completed. The phase correction, power gain and energy redistribution coefficients carried by the reversible drift operator are superimposed on the intermediate structural segment after the coordinate transformation. The phase synchronization, amplitude leveling and energy normalization are performed sequentially on the cross-dimensional coupling relationship to obtain the structural segment after a single drift compensation. The structural fragment after single drift compensation is taken as input, and the reversible drift operators of the remaining graph edges are called in sequence. The coordinate transformation, parameter superposition and energy normalization processes are repeated until all operators in the drift path are executed. The final structural fragment is then written to the corresponding position of the target graph node identifier, and the index mapping table of the channel tensor and the fragment state flag are updated to complete the structural update.

2. The mobile communication method based on tensor computation according to claim 1, characterized in that, The multidimensional channel observation data includes time-domain channel observation data, frequency-domain channel observation data, antenna spatial-domain channel observation data, and multipath delay-domain channel observation data.

3. The mobile communication method based on tensor computation according to claim 1, characterized in that, The construction of the multidimensional channel tensor for detecting motion disturbance events includes: Based on the channel observation data in the time domain, frequency domain, antenna spatial domain, and multipath delay domain, the channel sampling results of each dimension are arranged and combined according to the preset time dimension, frequency dimension, antenna dimension, and multipath dimension to form a multidimensional channel tensor. Based on the multidimensional channel tensor, the structural changes of the tensor within adjacent time windows are calculated, including amplitude changes, phase changes, antenna correlation changes, and multipath cluster energy distribution changes. Based on changes in user speed, acceleration, orientation, and cell handover events, the amount of structural change is compared with a preset disturbance threshold. If the change exceeds the disturbance threshold, a motion disturbance event is determined to have occurred.

4. The mobile communication method based on tensor computation according to claim 1, characterized in that, The step of determining the dimension reset region based on the perturbation sensitivity of the multidimensional channel tensor in each dimensional sub-region, and performing a dimension ablation operation on the tensor data within the dimension reset region, includes: Based on the multidimensional channel tensor, the multidimensional channel tensor is segmented according to a preset granularity in the time dimension, frequency dimension, antenna dimension and multipath dimension, and the multidimensional channel tensor is divided into multiple dimensional sub-regions. Each dimensional sub-region corresponds to a set of continuous time intervals, frequency intervals, antenna sets and multipath sets. For each dimensional sub-region, the disturbance sensitivity index characterizing the degree of motion disturbance in the dimensional sub-region is calculated by utilizing the channel amplitude change, phase change, inter-antenna correlation change and multipath energy distribution change within adjacent time windows, combined with user velocity, acceleration, azimuth change and cell handover events. Based on the current motion state, a disturbance sensitivity threshold is set, and dimensional sub-regions with disturbance sensitivity indices not lower than the disturbance sensitivity threshold are marked as candidate reset sub-regions. Candidate reset sub-regions that are adjacent or partially overlapping in the time dimension, frequency dimension, antenna dimension, or multipath dimension are merged to form multiple candidate dimensional reset region clusters. For each candidate dimension reset region cluster, based on local channel reconstruction error prediction, available pilot resource constraints, and dimension reset computational overhead, the expected benefit of performing dimension reset within the candidate dimension reset region cluster is evaluated. If a preset reset resource budget condition is met, the candidate dimension reset region cluster or multiple candidate dimension reset region clusters with the highest expected benefit are selected as the final dimension reset region. Specifically, the evaluation of the expected benefit of performing dimension reset within the candidate dimension reset region cluster involves: For each candidate dimension, the local channel reconstruction error reduction magnitude is predicted based on historical calibration results and current perturbation sensitivity, and the reduction magnitude is quantified as a reconstruction gain value. The number of pilot symbols available in this frame for the candidate dimension reset area cluster and the processing clock cycle requirement are statistically analyzed. Combined with the preset pilot resource limit and computation delay limit, the resource consumption cost is calculated. The expected revenue is the difference between the reconstruction revenue and the resource consumption cost. All candidate clusters are sorted by their expected revenue, and one or more clusters are selected as the final dimension reset area according to the budget threshold from high to low. Dimension ablation processing is performed on the multidimensional channel tensor data corresponding to the final dimension reset region. The channel estimation results in the final dimension reset region are cleared or marked as needing reconstruction, and reset mark information corresponding one-to-one with the position of the multidimensional channel tensor is generated.

5. The mobile communication method based on tensor computation according to claim 1, characterized in that, The process of fusing the local tensor with the multidimensional channel tensor to obtain the calibrated channel tensor includes: Based on the final dimension reset region and the partitioning results of time dimension, frequency dimension, antenna dimension and multipath dimension, the set of pilot resources used for local self-calibration is determined; Send or receive local pilot signals at the time, frequency and antenna positions corresponding to the pilot resource set, and obtain local channel observation data corresponding to the final dimension reset area position; Based on local channel observation data and the dimensional structure of the multidimensional channel tensor, local channel estimation is performed on each dimensional sub-region within the final dimensional reset region to obtain the local channel estimation results for each dimensional sub-region. For each dimension sub-region, the local channel estimation results are fitted with local structure to generate local tensor data consistent with the time dimension, frequency dimension, antenna dimension and multipath dimension. Then, multiple local tensor data are combined according to the division form of the final dimension reset region to generate a local calibration tensor. Each element in the local calibration tensor is written into the corresponding position of the multidimensional channel tensor according to its index position, replacing the original channel estimation result in the final dimension reset region, and the calibrated channel tensor is obtained.

6. The mobile communication method based on tensor computation according to claim 1, characterized in that, The process of performing cross-dimensional coupling kernel decomposition on the calibrated channel tensor to obtain the coupling decomposition results includes: Based on the calibrated channel tensor, the combination of time and frequency dimension, antenna and multipath dimension, and time and motion feature dimension are determined as the target dimension combination for cross-dimensional coupling processing. The joint observation data set corresponding to each target dimension combination is extracted from the calibrated channel tensor. Based on the joint observation data set, time and frequency coupling kernel, antenna and multipath coupling kernel, and time and motion feature coupling kernel are constructed. The time and frequency coupling kernel is composed of multiple kernel units corresponding to different time intervals and frequency sub-bands. The antenna and multipath coupling kernel is composed of multiple kernel units corresponding to different antenna subarrays and multipath clusters. The time and motion feature coupling kernel is composed of multiple kernel units corresponding to different motion states and time intervals. By using the time-frequency coupling kernel to process the joint observation data of the calibrated channel tensor in the time and frequency dimensions, the time-frequency joint change patterns corresponding to different kernel units are separated into multiple time-frequency coupling components, generating the first cross-dimensional coupling result. The joint observation data of the first cross-dimensional coupling result in the antenna dimension and multipath dimension are processed by the antenna and multipath coupling kernel. The spatial propagation structure corresponding to different antenna subarrays and multipath clusters is separated into multiple antenna and multipath coupling components to generate the second cross-dimensional coupling result. The joint observation data of the second cross-dimensional coupling result in the time and motion feature dimensions are processed using the time and motion feature coupling kernel and motion parameters to generate multiple time and motion feature coupling components. Multiple time and frequency coupling components, multiple antenna and multipath coupling components, and multiple time and motion feature coupling components are correlated and aggregated to form the coupling decomposition result.

7. The mobile communication method based on tensor computation according to claim 1, characterized in that, The process of dividing the channel structure into multiple stable structural segments and constructing a reversible drift compensation map includes: Based on the coupling decomposition results and the calibrated channel tensor, the structure is segmented according to the preset stability judgment conditions to obtain multiple stable structural segments. Each stable structural segment corresponds to a continuous time interval, frequency interval, antenna set, multipath set and motion feature set. Each stable structural segment is assigned a structural segment identifier, and a correspondence is established between the structural segment identifier and the corresponding time and frequency coupling components, antenna and multipath coupling components, and time and motion feature coupling components to form a set of structural segments. Based on the set of structural segments and motion parameters, the adjacency, overlap and coupling component change patterns of any two structural segments in time interval, frequency interval, antenna set, multipath set and motion feature set are analyzed, and a reversible drift operator is generated to describe the channel structure change between two structural segments. Using structural segment identifiers and reversible drift operators, each stable structural segment is constructed as a graph node, and each reversible drift operator is constructed as a graph edge connecting two graph nodes, forming a reversible drift compensation graph. In the reversible drift compensation graph, the forward drift path information from the source graph node to the target graph node and the reverse drift path information from the target graph node to the source graph node are recorded for each graph edge. In the reversible drift compensation graph, weight information is assigned to each edge of the graph. The weight information is determined based on the channel reconstruction error caused by the corresponding reversible drift operator, the computational cost of channel structure update, and the degree of consistency with motion parameters, thus forming a weighted reversible drift compensation graph.

8. The mobile communication method based on tensor computation according to claim 1, characterized in that, The application of channel reconstruction results to downlink beamforming, resource allocation, and modulation and coding scheme configuration includes: When the target user enters the current scheduling cycle, obtain the motion parameters and the graph node identifiers corresponding to the structural segments of the previous time step in the weighted reversible drift compensation graph; Based on motion parameters and weighted reversible drift compensation graph edge weight information, according to preset path selection rules, the target graph node is selected from the adjacent graph nodes connected to the graph node identifier of the previous time step, and the reversible drift path sequence from the graph node of the previous time step to the target graph node is determined. Based on the reversible drift path sequence, the reversible drift operators corresponding to each graph edge in the reversible drift path sequence are called sequentially to update the structure fragment of the previous time step in the calibrated channel tensor. The structure update result is written into the structure fragment position corresponding to the target graph node identifier to obtain the updated channel tensor. Based on the updated channel tensor, the channel information in the time dimension, frequency dimension, antenna dimension and multipath dimension is jointly reconstructed to obtain the downlink effective channel information in the current scheduling period; Based on the downlink effective channel information, the beamforming weights, time and frequency resource allocation schemes, and modulation and coding scheme configuration parameters corresponding to the target user are determined. The beamforming weights are applied to the base station antenna port, and the time and frequency resource allocation schemes and modulation and coding scheme configuration parameters are used to generate downlink transmission parameters.

9. A mobile communication system based on tensor computation, comprising executing the mobile communication method based on tensor computation as described in any one of claims 1 to 8, characterized in that, Includes the following modules: The channel construction and detection module is used to acquire multi-dimensional channel observation data, construct multi-dimensional channel tensors, and perform motion perturbation detection on the multi-dimensional channel tensors. The reset region determination and ablation module is used to determine the dimension reset region based on the multidimensional channel tensor after detecting motion disturbance, and to perform dimension ablation on the dimension reset region. The self-calibration and fusion module is used to perform local self-calibration based on the dimension reset region to obtain the calibrated channel tensor; The coupling kernel decomposition module is used to construct a cross-dimensional coupling kernel, perform cross-dimensional coupling kernel decomposition on the calibrated channel tensor, and obtain the coupling decomposition result. The structural fragment and compensation graph module is used to generate stable structural fragments based on the coupling decomposition results and construct a reversible drift compensation graph. The structure update and downlink configuration module is used to select the target path of the reversible drift compensation map based on motion parameters, update the channel tensor, and perform channel reconstruction and downlink configuration based on the updated channel tensor.

Images