A deep learning-based multi-base station cooperative robust positioning method and system

Abstract

The application discloses a kind of multi-base station cooperative robust positioning method and system based on deep learning, it is related to wireless communication technical field, by gathering the change channel observation data of distributed base station, composition multi-base station multi-time slot joint channel matrix;The time slot data of multi-base station multi-time slot joint channel matrix is shifted and handled, simulates the synchronization error of real scene, obtains the multi-base station joint channel matrix under synchronization error condition;Multi-base station joint channel matrix under synchronization error condition is handled by synchronization error calibration, and the multi-base station joint channel matrix after calibration is obtained;User coordinate positioning is carried out using the multi-base station joint channel matrix after calibration and the positioning network built.This application significantly improves the performance of multi-base station cooperative positioning system while solving the problem of synchronization error.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and more specifically to a multi-base station cooperative robust positioning method and system based on deep learning. Background Technology

[0002] With the large-scale deployment of fifth-generation mobile communication (5G) technology and the gradual evolution of the sixth-generation communication system (6G) standard, high-precision positioning technology has become a key enabling technology supporting the development of an intelligent Internet of Things society. In typical scenarios such as the Industrial Internet of Things (IIoT), vehicle-to-everything (V2X) systems, smart factories, and drone swarm control, the demand for centimeter-level positioning accuracy is showing an exponential growth trend. Multi-base station collaborative positioning technology, by constructing a distributed sensing network, achieves spatiotemporal fusion of measurement data between base stations. Compared with the Global Navigation Satellite System (GNSS), it exhibits significant technical advantages in complex channel environments such as non-line-of-sight (NLOS) propagation and indoor multipath propagation, and has become a research focus in the field of ubiquitous positioning.

[0003] However, this technology faces severe synchronization challenges in practical deployment. Physical layer limitations such as inherent hardware latency and clock drift caused by multipath propagation lead to microsecond-level or even nanosecond-level clock deviations between base stations, directly affecting positioning accuracy and system performance. In traditional multi-base station positioning methods, synchronization errors cause deviations in the measurement of Time of Arrival (TOA) or Time Difference of Arrival (TDOA), which in turn causes positioning errors when using distributed algorithms for joint positioning.

[0004] Addressing the synchronization error problem in multi-base station cooperative positioning, this paper delves into its impact mechanism and proposes effective error compensation methods, which has significant theoretical and practical value. In multi-base station cooperative positioning in complex environments and highly dynamic applications, synchronization error calibration is crucial for improving positioning stability and anti-interference capabilities. It is also key to advancing multi-base station cooperative positioning technology from theoretical models to practical deployment. However, current research on multi-base station cooperative positioning is largely based on the ideal assumption of no synchronization error; therefore, filling this research gap is essential.

[0005] Therefore, how to significantly improve the performance of multi-base station cooperative positioning systems while solving the synchronization error problem is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a multi-base station cooperative robust localization method and system based on deep learning, which has both high-precision localization capability and robustness to complex channel environments.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A deep learning-based multi-base station cooperative robust localization method includes:

[0009] Collect change channel observation data from distributed base stations to form a multi-base station, multi-timeslot joint channel matrix;

[0010] The time slot data of the multi-base station multi-time slot joint channel matrix is ​​shifted to simulate the synchronization error in the real scenario and obtain the multi-base station joint channel matrix under the synchronization error condition.

[0011] Synchronization error calibration is performed on the joint channel matrix of multiple base stations under synchronization error conditions to obtain the calibrated joint channel matrix of multiple base stations.

[0012] User coordinates are located using the calibrated multi-base station joint channel matrix and the established positioning network.

[0013] Preferably, obtaining the multi-base station joint channel matrix under synchronization error conditions includes applying a cyclic shift operation to the time dimension of the channel state information data in the multi-base station multi-timeslot joint channel matrix. Assuming there are b base stations in the system, and each base station receives channel state information data within t time slots, the specific operation is expressed as follows:

[0014] ;

[0015] in, It is a channel state matrix without synchronization error. This is the channel state matrix with synchronization error, where n is the number of samples, b is the number of base stations, a is the number of antennas, and f is the number of subcarriers. This represents the time slot offset of base station b in the nth sample. After applying a cyclic shift operation, the channel state information data is input into the synchronization error calibration network. Synchronization error calibration is achieved by predicting the number of shifts in the data and performing a reverse shift.

[0016] Preferably, the synchronization error calibration network includes:

[0017] The spatial processing module uses a three-layer two-dimensional convolutional network to extract the subcarrier and antenna spatial features of the channel state information data after cyclic shifting. Then, it is flattened and passed through two layers of one-dimensional convolution to obtain a one-dimensional feature vector, which is then input into the first time-series processing module. The activation function is LeakyReLU.

[0018] The first time-series processing module constructs a Transformer encoder based on the MLA mechanism and captures time-series dependencies by compressing key-value caches using low-rank latent vectors.

[0019] The classification output layer predicts the time slot offset of each base station through a fully connected layer and a Softmax function;

[0020] The spatial processing module, the first temporal processing module, and the classification output layer are connected in sequence.

[0021] Preferably, the spatial processing module includes a first three-layer 2D dilated convolutional network, with the number of channels expanded from the initial 2 to 4, 8, and 16, and the activation function selected is LeakyReLU; the first temporal processing module is a dual structure of one-dimensional convolution and Transformer encoder. The features output by the spatial processing module are first input into the first one-dimensional convolutional layer, activated by ReLU, and then input into the second one-dimensional convolutional layer and activated by ReLU again. Subsequently, the feature vector enters the TransformerEncoder module: its core MLA layer directly receives the temporal feature sequence; the output of the MLA layer is residually connected to the original input through the Dropout layer, and then input into the LayerNorm layer for normalization; the normalized data enters the feedforward network; the output of the feedforward network is also residually connected through the Dropout layer and input into the second LayerNorm layer. Finally, the temporal features are compressed into a fixed-length vector through an adaptive average pooling layer.

[0022] Preferably, the positioning network includes a second and third layer 2D dilated convolutional network, a second temporal processing module, and an output module, which are connected in sequence.

[0023] Preferably, the three-layer two-dimensional convolutional network extracts the spatial features of the channel state information data step by step, and then flattens the spatial features. The second time-series processing module includes a one-dimensional convolutional network and an xLSTM model. The flattened spatial features are received through the one-dimensional convolutional network and input into the xLSTM model. The output module receives the output of the xLSTM model through a time pooling layer, merges and fuses the output of the time pooling layer, and then passes it through three fully connected layers. The last fully connected layer is a two-neuron linear fully connected layer, which directly outputs continuous coordinate values.

[0024] Preferably, the loss function of the synchronization error calibration network is the cross-entropy loss function, which calculates the prediction error of each base station and takes the average value; the loss function of the positioning network is the mean square error loss function, which directly optimizes the coordinate prediction accuracy.

[0025] A deep learning-based multi-base station cooperative robust localization system includes:

[0026] The matrix building module collects changing channel observation data from distributed base stations and forms a multi-base station, multi-timeslot joint channel matrix.

[0027] The error synchronization module performs shift processing on the time slot data of the multi-base station multi-time slot joint channel matrix to simulate the synchronization error in the real scenario and obtain the multi-base station joint channel matrix under the synchronization error condition.

[0028] The calibration module performs synchronization error calibration on the joint channel matrix of multiple base stations under synchronization error conditions to obtain the calibrated joint channel matrix of multiple base stations.

[0029] The positioning module uses the calibrated multi-base station joint channel matrix and the established positioning network to locate the user's coordinates.

[0030] As can be seen from the above technical solutions, compared with existing technologies, this invention discloses a multi-base station collaborative robust positioning method and system based on deep learning. Through data-driven error calibration and spatiotemporal feature fusion, it significantly improves positioning performance in complex environments. Simulation results show that the xLSTM model achieves a mean absolute positioning error of 0.0347 meters, while the MLA model maintains an error of 0.2626 meters, achieving a 6.2-fold increase in training efficiency. The method proposed in this invention overcomes the dependence on high-precision clock synchronization equipment and supports robust positioning in dynamic scenarios. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0032] Figure 1 The simulation channel model scenario diagram provided by this invention;

[0033] Figure 2 The CSI channel model for a multi-base station receiver provided by this invention;

[0034] Figure 3 The simulation parameter settings diagram for the simulation scenario and channel model provided by this invention;

[0035] Figure 4 A model structure diagram of the synchronization error calibration network provided by this invention;

[0036] Figure 5 The main structure diagram of the transformer encoder used in the model provided by this invention;

[0037] Figure 6 This is a diagram of the multi-base station cooperative positioning system architecture provided by the present invention;

[0038] Figure 7aA graph showing the loss variation during the training process of the synchronization error calibration network provided by this invention;

[0039] Figure 7b A graph showing the accuracy change trend during the training process of the synchronous error calibration network provided by this invention;

[0040] Figure 8a This invention provides a comparison chart of xLSTM model positioning errors before and after synchronization error compensation.

[0041] Figure 8b The comparison diagram of MLA model positioning error before and after synchronization error compensation provided by the present invention;

[0042] Figure 9 A flowchart of the method steps provided by the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] This invention mainly comprises two stages: a synchronization error detection and compensation stage, and a multi-base station cooperative positioning stage after compensation. This method aims to solve the data offset problem caused by clock asynchrony among multiple base stations on CSI (Channel State Information) data, improving the positioning accuracy of wireless users indoors or in complex scenarios. This invention provides a deep learning-based multi-base station cooperative robust positioning method. First, considering the clock synchronization error characteristics between base stations, an error model based on CSI time-dimensional cyclic shift is established. Then, an MLA network is used to capture temporal dependencies and adaptively estimate error parameters. Simulation results show that the displacement prediction accuracy of the calibration network is close to 99.9%, significantly improving the reliability of time slot alignment. In the positioning network design, the architecture integrating spatial convolution modules and temporal processing modules (such as xLSTM and MLA) effectively extracts the spatiotemporal features of channel state information data. The xLSTM model, after calibration, achieves an average absolute positioning error of 0.0347 meters, verifying the superiority of the method. Furthermore, by comparing the performance of different temporal modules, the advantages of xLSTM in accuracy and the balance between the efficiency of MLA are revealed, providing theoretical support for model selection in practical scenarios. This not only provides a new technical approach for multi-base station collaborative positioning under synchronization error conditions, but also verifies the applicability of data-driven methods in complex channel environments, and has important engineering value for high-precision positioning scenarios such as intelligent transportation and indoor navigation.

[0045] This invention discloses a multi-base station cooperative robust localization method based on deep learning, such as... Figure 9 As shown, it includes:

[0046] Collect change channel observation data from distributed base stations to form a multi-base station, multi-timeslot joint channel matrix;

[0047] The time slot data of the multi-base station multi-time slot joint channel matrix is ​​shifted to simulate the synchronization error in the real scenario and obtain the multi-base station joint channel matrix under the synchronization error condition.

[0048] Synchronization error calibration is performed on the joint channel matrix of multiple base stations under synchronization error conditions to obtain the calibrated joint channel matrix of multiple base stations.

[0049] User coordinates are located using the calibrated multi-base station joint channel matrix and the established positioning network.

[0050] In a specific embodiment, obtaining the joint channel matrix of multiple base stations under synchronization error conditions includes applying a cyclic shift operation to the time dimension of the channel state information data in the joint channel matrix of multiple base stations and multiple time slots. Assuming there are b base stations in the system, and each base station receives channel state information data within t time slots, the specific operation is expressed as follows:

[0051] ;

[0052] in, It is a channel state matrix without synchronization error. This is the channel state matrix with synchronization error, where n is the number of samples, b is the number of base stations, a is the number of antennas, and f is the number of subcarriers. This represents the time slot offset of base station b in the nth sample. After applying a cyclic shift operation, the channel state information data is input into the synchronization error calibration network. Synchronization error calibration is achieved by predicting the number of shifts in the data and performing a reverse shift.

[0053] In one specific embodiment, the synchronization error calibration network includes:

[0054] The spatial processing module uses a three-layer two-dimensional convolutional network to extract the subcarrier and antenna spatial features of the channel state information data after cyclic shifting. Then, it is flattened and passed through two layers of one-dimensional convolution to obtain a one-dimensional feature vector, which is then input into the first time-series processing module. The activation function is LeakyReLU.

[0055] The first timing processing module, namely Figure 4 The Transformer encoder in the text is built based on the MLA mechanism. It reduces inference memory usage and captures temporal dependencies by compressing key-value caches through low-rank latent vectors.

[0056] The classification output layer uses a fully connected layer and a Softmax function to predict the time slot offset of each base station;

[0057] The spatial processing module, the first temporal processing module, and the classification output layer are connected in sequence.

[0058] Note that both the aforementioned spatial processing module and the first timing processing module target the signal matrix of a single base station for the same target. Therefore, after timing processing, the signal matrices of different base stations for the same target are first merged, and then the output layer is classified.

[0059] In one specific embodiment, the spatial processing module includes a first three-layer 2D dilated convolutional network, with the number of channels expanded from the initial 2 to 4, 8, and 16, and LeakyReLU activation function selected. The first temporal processing module is a dual structure of one-dimensional convolution and Transformer encoder. The features output by the spatial processing module are first input into the first one-dimensional convolutional layer, activated by ReLU, and then input into the second one-dimensional convolutional layer and activated by ReLU again. Subsequently, the feature vector enters the TransformerEncoder module: its core MLA layer directly receives the temporal feature sequence; the output of the MLA layer is residually connected to the original input through the Dropout layer, and then input into the LayerNorm layer for normalization; the normalized data enters the feedforward network; the output of the feedforward network is also residually connected through the Dropout layer and input into the second LayerNorm layer. Finally, the temporal features are compressed into a fixed-length vector through an adaptive average pooling layer.

[0060] In one specific embodiment, the localization network includes a second and third layer 2D dilated convolutional network, a second temporal processing module, and an output module, which are connected sequentially.

[0061] In one specific embodiment, a three-layer two-dimensional convolutional network extracts the spatial features of the channel state information data step by step, and then flattens the spatial features. The second temporal processing module includes a one-dimensional convolutional network and an xLSTM model. The flattened spatial features are received through the one-dimensional convolutional network and input into the xLSTM model, which adopts an exponential gating and matrix memory structure and achieves parallel processing through a covariance update rule. The output module receives the output of the xLSTM model through a temporal pooling layer, merges and fuses the output of the temporal pooling layer, and then passes it through three fully connected layers. The last fully connected layer adopts a two-neuron linear fully connected layer to directly output continuous coordinate values.

[0062] In one specific embodiment, the loss function of the synchronization error calibration network is the cross-entropy loss function, which calculates the prediction error of each base station and takes the average value; the loss function of the positioning network is the mean square error loss function, which directly optimizes the coordinate prediction accuracy.

[0063] A deep learning-based multi-base station cooperative robust localization system includes:

[0064] The matrix building module collects changing channel observation data from distributed base stations and forms a multi-base station, multi-timeslot joint channel matrix.

[0065] The error synchronization module performs shift processing on the time slot data of the multi-base station multi-time slot joint channel matrix to simulate the synchronization error in the real scenario and obtain the multi-base station joint channel matrix under the synchronization error condition.

[0066] The calibration module performs synchronization error calibration on the joint channel matrix of multiple base stations under synchronization error conditions to obtain the calibrated joint channel matrix of multiple base stations.

[0067] The positioning module uses the calibrated multi-base station joint channel matrix and the established positioning network to locate the user's coordinates.

[0068] The simulation verification conditions of this invention are as follows: the operating frequency band is set to 3.5 GHz, the bandwidth to 40 MHz, and the number of subcarriers to 64; a CSI dataset containing random time slot offsets is generated; multipath propagation is simulated using a ray tracing algorithm to construct a time-varying channel impulse response and convert it into frequency domain CSI data; the positioning errors before and after calibration are compared and analyzed to verify the effectiveness of synchronization error compensation. Simultaneously, for dynamic scenarios, resource allocation is optimized by combining 6G Sensor-Integrated Communication (ISAC) technology to improve the environmental adaptability of the positioning system.

[0069] This invention constructs a realistic ray-tracing scene of urban blocks in the Sionna simulation platform, such as... Figure 1 As shown, the CSI data between the base station and user equipment was acquired via ray tracing, and the dataset covers multiple synchronization error levels and user motion states. By comparing the error performance of the CSI data before and after calibration in the positioning network, the effectiveness of the synchronization error calibration method of this invention is significantly verified.

[0070] like Figure 2 As shown, this invention establishes a CSI channel model for multiple base station receivers in the system model and constructs a five-dimensional CSI tensor with a time dimension. To simulate the synchronization error between different base stations, the CSI tensor is quantitatively shifted in the time dimension to simulate the reception timing differences between base stations caused by clock offsets in the actual environment, thereby constructing a dataset with offset labels for subsequent error detection network training.

[0071] During the training of the synchronization error calibration network, the time slot offset is set to a discrete value of ±1 to ±4 to simulate microsecond-level clock deviation. The number of classification head parameters increases linearly with the offset range, and the total memory usage is 294.55~294.57 MB.

[0072] The main parameter configurations of this invention are as follows: Figure 3 As shown. The core parameter configuration of the simulation system is mainly based on two factors: the actual needs of the wireless communication scenario and the algorithm verification objective. The system operating frequency band is set to 3.5 GHz to cover typical application scenarios of mid-frequency wireless communication. The total system bandwidth is 40 MHz, which is divided into 64 subcarriers using orthogonal frequency division multiplexing (OFDM) technology. The subcarrier spacing is calculated to be 625 kHz by uniformly distributing the total bandwidth.

[0073] To simulate the multipath effect in complex propagation environments, the maximum reflection number is set to 5, covering physical phenomena such as reflection and scattering experienced by the signal during propagation.

[0074] The antenna configuration of the base station and user equipment is a crucial component of the simulation scenario. The base station receiver employs a 2x4 planar array antenna with a vertical and horizontal spacing of 0.5 wavelengths and vertical polarization to enhance spatial diversity. The user equipment uses a single-antenna transmit configuration to simulate a simplified hardware design of a typical mobile terminal.

[0075] To accurately capture the Doppler effect caused by user equipment movement, the sampling frequency was set to 625 kHz in the simulation (matching the subcarrier allocation bandwidth calculated above), and the time step was 14. The time difference between adjacent time slots can be calculated by the reciprocal relationship between period and frequency.

[0076] This invention discloses a synchronization error calibration network model, such as... Figure 4 As shown, the structure of the transformer encoder used is as follows: Figure 5 As shown.

[0077] This neural network model consists of a spatial processing module, a first temporal processing module, and an output layer. These modules work collaboratively through specific connections. The spatial processing module employs a three-dimensional convolutional structure. First, the input data is processed by a first two-dimensional convolutional layer (input channel 2, output channel 4, 3×3 kernels), followed by a non-linear transformation using the LeakyReLU activation function (negative slope 0.1). Its output is directly fed into a second two-dimensional convolutional layer (input channel 4, output channel 8, 3×3 kernels), also connected to the LeakyReLU activation function. Subsequently, it enters a third two-dimensional convolutional layer (input channel 8, output channel 16, 3×3 kernels) and is again activated by the LeakyReLU function. Finally, the output of this module is flattened to compress the spatial dimension into a one-dimensional feature vector.

[0078] The first temporal processing module comprises a dual structure of one-dimensional convolution and a Transformer encoder. The features output from the spatial processing module are first fed into a first one-dimensional convolutional layer (input channels are automatically calculated spatial feature dimensions, output channels are 512, kernel size is 3), activated by ReLU, and then fed into a second one-dimensional convolutional layer (input channels are 512, output channels are 64, kernel size is 3) and activated again by ReLU. The feature vectors then enter the TransformerEncoder module: its core MLA layer directly receives the temporal feature sequence; the MLA output is residually connected to the original input through a Dropout layer (dropout rate 0.1), and then input into a LayerNorm layer for normalization; the normalized data enters the Feedforward Network (FFN), which consists of linear layers (input / output dimensions 64 / 128), ReLU activation, a Dropout layer (dropout rate 0.1), and a linear layer (128 / 64) connected sequentially; the FFN output is also residually connected through a Dropout layer and fed into the second LayerNorm layer. Finally, the temporal features are compressed into fixed-length vectors through an adaptive average pooling layer (AdaptiveAvgPool1d).

[0079] The output layer, designed for discrete classification of time slot offsets from six base stations, first concatenates the feature vectors of the six base stations and inputs them into the classifier sequence: a first linear layer (input dimension 64×6, output 128) connected to a ReLU activation function, and a second linear layer (128 dimensions to 6×num_classes output). The final output tensor is reconstructed into a three-dimensional structure (batch × 6 stations × number of classes), completing the end-to-end mapping from the original spatiotemporal input to station-level prediction. Throughout the processing, spatial convolution extracts local features, temporal convolution captures short-range dependencies, the Transformer encoder models long-range temporal relationships, and finally, a fully connected layer achieves joint prediction across multiple stations. The simulation sets the time slot shift range to cover ±1 to ±4, corresponding to microsecond-level synchronization errors, and accurate error compensation is achieved through classification prediction.

[0080] This invention discloses a localization network model, such as... Figure 6 As shown, the model first extracts the spatial features of CSI data step by step through a second and third layer of 2D dilated convolutional networks. Multi-level convolutional operations are used to capture the spatial correlation of signals in the subcarrier and path dimensions. After flattening the spatial features, a one-dimensional convolutional network is used to process the time series, reducing computational load and achieving feature integration. The second time series processing module is directly connected to the xLSTM model, eliminating the need for a transformer encoder. The multi-base station cooperative positioning network, as a regression task, needs to predict the two-dimensional spatial coordinates of user equipment. Therefore, the output layer uses a two-neuron linear fully connected layer to directly output continuous coordinate values. This design avoids unnecessary interference from nonlinear activation functions on the regression values ​​while ensuring the numerical continuity of the coordinate prediction.

[0081] Figure 7a The figure shows the loss changes during the training process of the four models. Figure 7b The figure shows the accuracy trends during the training of the four models. The loss functions of all models decreased with the number of iterations and stabilized after 100 training rounds, indicating good model convergence.

[0082] Figure 8a and Figure 8b The performance of different network structures under various synchronization error conditions was demonstrated. Experimental results show that: 1. The calibration network can accurately identify and correct the time offset between base stations, and the recovered CSI data significantly reduces the positioning error. Especially in high-error scenarios (±4 time slot shifts), the calibrated positioning accuracy can be basically restored to the error-free level; 2. The positioning network can achieve high-precision (centimeter-level) positioning in the current scenario and has a certain tolerance to the impact of synchronization errors.

[0083] In summary, the multi-base station cooperative localization method and system provided by this invention under synchronization error conditions comprehensively utilizes deep learning techniques such as cyclic shift modeling, multi-head latent attention (MLA) mechanism, and extended long short-term memory network (xLSTM) to achieve accurate compensation for synchronization errors between multiple base stations and a significant improvement in user positioning accuracy. Through spatiotemporal feature extraction and regression optimization of CSI data, this invention exhibits good robustness and engineering adaptability in complex multipath environments and dynamic scenarios.

[0084] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0085] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A deep learning-based multi-base station cooperative robust positioning method, characterized in that, The method comprises the following steps: Collecting the changing channel observation data of the distributed base stations to form a multi-base station multi-time slot joint channel matrix; Performing shift processing on the time slot data of the multi-base station multi-time slot joint channel matrix to simulate the synchronization error of the real scene and obtain a multi-base station joint channel matrix under the condition of the synchronization error; Performing synchronization error calibration processing on the multi-base station joint channel matrix under the condition of the synchronization error to obtain a calibrated multi-base station joint channel matrix; Using the calibrated multi-base station joint channel matrix and a positioning network built to position the user coordinates; The step of obtaining the multi-base station joint channel matrix under the condition of the synchronization error comprises applying a cyclic shift operation on the time dimension of the channel state information data in the multi-base station multi-time slot joint channel matrix, and the specific operation is represented as: wherein b represents the total number of base stations in the system, t represents the number of time slots in which each base station receives the channel state information data, and represents the cyclic shift operation on the time dimension of the channel state information data. ； wherein, is a channel state matrix without synchronization error, is a channel state matrix with synchronization error, n is a sample number, b is a base station number, a is an antenna number, and f is a subcarrier number, denotes a time slot offset of the base station b in the n-th sample, and the channel state information data is input into a synchronization error calibration network after a cyclic shift operation is applied, and synchronization error calibration is realized by predicting the number of shifts of the data and performing reverse shifting. The synchronization error calibration network comprises: A spatial processing module that extracts the subcarrier and antenna dimension spatial features of the channel state information data subjected to the cyclic shift operation by using a three-layer two-dimensional convolution network, then flattens the features, and obtains a one-dimensional feature vector by passing through two one-dimensional convolution layers, and inputs the one-dimensional feature vector into a first time sequence processing module, and the activation function is LeakyReLU; The first time sequence processing module is constructed based on the MLA mechanism to build a Transformer encoder, and the low-rank latent vector is used to compress the key-value cache to capture the time sequence dependency; A classification output layer that mainly predicts the time slot offset of each base station by using a fully connected layer and a Softmax function; The spatial processing module, the first time sequence processing module and the classification output layer are sequentially connected.

2. The deep learning based multi-base station cooperative robust positioning method according to claim 1, characterized in that, The spatial processing module comprises a first three-layer two-dimensional dilated convolution network, the number of channels is expanded from the initial 2 to 4, 8 and 16, and the activation function is LeakyReLU; the first time sequence processing module is a dual structure of one-dimensional convolution and Transformer encoder, the features output by the spatial processing module are first input into the first one-dimensional convolution layer, activated by ReLU, then input into the second one-dimensional convolution layer and activated by ReLU again, and then the feature vector enters the TransformerEncoder module: the core MLA layer directly receives the time sequence feature sequence; the output of the MLA layer is connected in residual connection with the original input through the Dropout layer, and then input into the LayerNorm layer for normalization; the normalized data enters the feedforward network; the output of the feedforward network is also connected in residual connection through the Dropout layer and input into the second LayerNorm layer, and finally the time sequence features are compressed into a fixed length vector through the adaptive average pooling layer.

3. The deep learning based multi-base station cooperative robust positioning method according to claim 1, wherein, The positioning network comprises a second three-layer two-dimensional dilated convolution network, a second time sequence processing module and an output module, and the second three-layer two-dimensional dilated convolution network, the second time sequence processing module and the output module are sequentially connected.

4. The deep learning based multi-base station cooperative robust positioning method according to claim 1, characterized in that, The loss function of the synchronization error calibration network is a cross-entropy loss function, and the prediction error of each base station is calculated and averaged; the loss function of the positioning network is a mean square error loss function, and the coordinate prediction accuracy is directly optimized.

5. A deep learning based multi-base station cooperative robust positioning system, applying the deep learning based multi-base station cooperative robust positioning method of any one of claims 1-4, characterized in that, The method comprises the following steps: The matrix establishing module collects the changed channel observation data of the distributed base stations to form a multi-base station multi-time slot joint channel matrix. The error synchronization module performs shift processing on the time slot data of the multi-base station multi-time slot joint channel matrix to simulate the synchronization error of a real scene and obtain a multi-base station joint channel matrix under a synchronization error condition. The calibration module performs synchronization error calibration processing on the multi-base station joint channel matrix under the synchronization error condition to obtain a calibrated multi-base station joint channel matrix. The positioning module uses the calibrated multi-base station joint channel matrix and a built positioning network to position the user coordinates.

Images