A beam tracking method for a multi-base station cooperative millimeter wave sensing and communication integrated system

Abstract

The application provides a beam tracking method of a multi-base-station cooperative millimeter wave sensing and communication integrated system, comprising: multiple base stations estimating local sensing parameters of a target vehicle and uploading to a cloud processing center; the cloud processing center performing state prediction, querying a theoretical observation vector corresponding to a predicted state by using a channel knowledge graph, and determining a service base station and a predicted beam of a next time slot; after obtaining an actual observation vector in the next time slot, calculating an innovation statistic and comparing with a preset threshold; if the threshold is not exceeded, filtering update is performed; if the threshold is exceeded, a beam correction and realignment mechanism is triggered, a local candidate state set is constructed with the predicted state as the center, the theoretical observation vectors of each candidate state are queried by using the channel knowledge graph, a candidate state with the minimum consistency cost function is selected as a correction state, and the service base station and the beam direction are re-determined based on the correction state. The application can guarantee the continuity and robustness of beam tracking, and can be widely applied in the field of 6G technology.

Description

Technical Field

[0001] This application relates to the field of sixth-generation mobile communication (6G) technology, and in particular to a beam tracking method for a multi-base station cooperative millimeter-wave integrated sensing and communication (ISAC) system. Background Technology

[0002] Integrated Communication and Sensing (ISAC) is considered one of the key enabling technologies for sixth-generation mobile communication networks, providing crucial support for emerging applications such as intelligent transportation, connected vehicles, and autonomous driving. Millimeter-wave communication, with its abundant spectrum resources and high angular resolution, plays a vital role in ISAC systems. However, millimeter-wave ISAC systems typically rely on highly directional, narrow-beam transmission, which can easily lead to beam mismatch in scenarios involving high-speed vehicle movement and complex road environments, significantly increasing the difficulty of beam alignment and continuous tracking. Furthermore, single-base station ISAC systems have limitations in coverage and observation angle; when vehicles are obstructed by buildings, both communication link and sensing performance can degrade significantly. In contrast, multi-base station collaborative millimeter-wave ISAC systems can improve the system's tracking and communication performance through multi-view observation and service link switching.

[0003] In millimeter-wave ISAC systems with multi-base station cooperation, vehicle state, channel parameters, and the optimal serving base station all change dynamically over time, resulting in a strong coupling between state tracking, serving base station selection, and beam design. Traditional methods typically achieve target tracking through recursive state estimation, such as those based on extended Kalman filtering (EKF). However, these methods are prone to error accumulation in complex environments due to model mismatch and observational anomalies. Once the tracking deviation increases, the system struggles to autonomously detect and correct it, leading to beam mismatch or even tracking interruption.

[0004] On the other hand, Channel Knowledge Map (CKM), as an effective environmental awareness prior, can integrate historical propagation information within a region to construct location-related environmental priors, enabling the system to quickly infer key information such as angle, delay, Doppler, and path loss based on the target state. Existing CKM-related research mainly focuses on CKM construction methods and forward query applications (querying channel parameters based on location). Most studies use CKM for forward inference, but have not yet applied CKM to anomaly detection and loop closure correction in beam tracking.

[0005] In summary, existing technologies have the following shortcomings: In complex urban environments, beam tracking of millimeter-wave ISAC systems is easily affected by factors such as building obstruction, local observation anomalies, and sensor link degradation. Traditional recursive tracking methods lack effective means to detect tracking deviations and lack reliable correction and realignment mechanisms when beam mismatch occurs, making it difficult to guarantee the continuity and robustness of beam tracking. Summary of the Invention

[0006] The main objective of this application is to propose a beam tracking method, system, electronic device, storage medium, and program product for a multi-base station cooperative millimeter-wave sensing integrated system. This addresses the shortcomings of existing millimeter-wave ISAC systems in beam tracking in complex urban environments, such as the lack of detection methods for beam tracking anomalies and the absence of effective correction and realignment mechanisms when beam mismatch occurs, making it difficult to guarantee the continuity and robustness of beam tracking.

[0007] To achieve the above objectives, one aspect of this application proposes a beam tracking method for a multi-base station cooperative millimeter-wave sensing integrated system, the method comprising: Multiple base stations transmit sensing signals to the target vehicle and receive echo signals reflected by the target vehicle. Each base station estimates the local sensing parameters of the target vehicle relative to the base station based on the echo signals and uploads the local sensing parameters to the cloud processing center. The cloud processing center performs state prediction on the target vehicle based on the local sensing parameters uploaded by each base station, and obtains the predicted state. Based on the predicted state, the cloud processing center uses the channel knowledge graph to query the theoretical observation vector corresponding to the predicted state, and determines the serving base station and predicted beam for the next time slot. In the next time slot, the serving base station uses the predicted beam to perform downlink data transmission. Each base station receives the actual echo signal and estimates the actual observation vector, and uploads the actual observation vector to the cloud processing center. The cloud processing center calculates the innovation statistic based on the predicted state and the actual observation vector, and determines whether the innovation statistic exceeds a preset threshold. If the innovation statistic does not exceed the preset threshold, the cloud processing center filters and updates the predicted state based on the actual observation vector to obtain the updated state estimate. If the innovation statistics exceed the preset threshold, the cloud processing center determines that a beam anomaly has occurred and triggers a beam correction and realignment mechanism. The beam correction and realignment mechanism includes: constructing a local candidate state set centered on the predicted state; for each candidate state in the local candidate state set, querying the theoretical observation vector corresponding to the candidate state through the channel knowledge graph, and calculating a consistency cost function based on the theoretical observation vector and the actual observation vector; selecting the candidate state with the smallest consistency cost function as the corrected state; and re-querying the channel knowledge graph based on the corrected state to reconstruct the channel state information and redetermine the serving base station and beam direction.

[0008] In some embodiments, the cloud processing center performs state prediction on the target vehicle, specifically including: The cloud processing center obtains the posterior state sequence of the target vehicle in historical time slots and constructs a historical state window. The historical state window is input into a pre-trained long short-term memory network, which then outputs the prior state prediction result of the target vehicle in the current time slot.

[0009] In some embodiments, the cloud processing center filters and updates the predicted state based on the actual observation vector, specifically including: Under the extended Kalman filter framework, the cloud processing center uses the channel knowledge graph as a measurement function to establish a nonlinear mapping relationship between the state vector of the target vehicle and the observation vectors of each base station. The measurement functions of each base station are linearized to obtain the joint Jacobian matrix; The Kalman gain is calculated based on the joint Jacobian matrix, and the predicted state is updated based on the Kalman gain and the actual observation vector to obtain the updated state estimate.

[0010] In some embodiments, the cloud processing center determines the serving base station and predicted beam for the next time slot, specifically including: Based on the predicted state, the cloud processing center obtains the set of channel parameters corresponding to the predicted state by querying the channel knowledge graph. Reconstruct the downlink channel state information between the target vehicle and each base station based on the set of channel parameters; The predicted reachability rate of each base station is calculated based on the downlink channel state information, and the base station with the highest predicted reachability rate is selected as the serving base station for the next time slot. Design a predictive beamforming vector based on the channel parameters corresponding to the serving base station.

[0011] In some embodiments, each base station estimates local perception parameters of the target vehicle relative to the base station based on the echo signal, specifically including: Each base station constructs a sample covariance matrix based on the received frequency domain echo signal, and performs eigenvalue decomposition on the sample covariance matrix to obtain the noise subspace; Based on the noise subspace, a two-dimensional MUSIC spatial spectrum function is constructed, and a peak search is performed in the two-dimensional angle domain to obtain the estimated azimuth and elevation angles of the target vehicle relative to the base station.

[0012] In some embodiments, the method of each base station estimating the local perception parameters of the target vehicle relative to the base station based on the echo signal further includes: After obtaining the estimated azimuth angle and the estimated elevation angle, each base station constructs a receiving beamforming vector along the target direction, performs spatial combining and symbol compensation on the echo signal, and obtains a scalar echo signal. Based on the scalar echo signal, construct the time-delay covariance matrix and the Doppler covariance matrix, and perform eigenvalue decomposition to obtain the time-delay noise subspace and the Doppler noise subspace, respectively. Based on the time-delay noise subspace, a time-delay MUSIC spectrum function is constructed and a peak search is performed to obtain the round-trip time delay estimate; based on the Doppler noise subspace, a Doppler MUSIC spectrum function is constructed and a peak search is performed to obtain the round-trip Doppler frequency shift estimate.

[0013] In some embodiments, linearizing the measurement functions of each base station to obtain the joint Jacobian matrix specifically includes: The central difference method is used to numerically approximate the measurement function represented by the channel knowledge graph; Near the predicted state, positive and negative perturbations are applied to each dimension of the state vector. The corresponding perturbed observation vector is obtained by querying the channel knowledge graph. The approximate value of the partial derivative corresponding to that dimension is calculated based on the ratio of the difference between the perturbed observation vector and the theoretical observation vector corresponding to the predicted state to the perturbation step size.

[0014] In some embodiments, the cloud processing center calculates innovation statistics based on the predicted state and the actual observation vector, specifically including: The cloud processing center calculates a joint theoretical observation vector based on the predicted state and the channel knowledge graph. The difference between the actual observation vector and the joint theoretical observation vector is calculated to obtain the joint innovation vector; Based on the joint innovation vector and the pre-calculated innovation covariance matrix, the normalized innovation statistic is calculated.

[0015] In some embodiments, constructing a local candidate state set centered on the predicted state specifically includes: With the position and velocity components of the predicted state as centers, candidate state spaces are constructed in the position and velocity domains respectively according to preset local search radii. The local candidate state set is obtained by sampling within the candidate state space according to a preset step size.

[0016] In some embodiments, reconstructing the downlink channel state information between the target vehicle and each base station based on the channel parameter set specifically includes: The cloud processing center queries the channel knowledge graph based on the predicted state to obtain the multipath channel parameters at the corresponding locations of each base station. The multipath channel parameters include the complex gain, angle of arrival, delay, and Doppler shift of each propagation path. Based on the multipath channel parameters, a downlink channel matrix between each base station and the target vehicle is constructed, which serves as the reconstructed channel state information.

[0017] In some embodiments, the channel knowledge graph is constructed offline using a deep learning-based method, specifically including: Collect vehicle position, speed, and corresponding multipath channel parameter data within the target area to construct a training dataset; Using vehicle position and speed as input and corresponding multipath channel parameters as output, a deep learning network is trained to learn the nonlinear mapping relationship between vehicle state and channel parameters. The trained deep learning network is used as the channel knowledge graph for fast querying during the online phase.

[0018] In some embodiments, the sensing signal is an orthogonal frequency division multiplexing signal, and the operating frequency bands of each base station are mutually orthogonal; each base station performs downlink data transmission and target sensing in parallel in each time slot, wherein the serving base station performs downlink data transmission with a predicted beam, and all base stations synchronously transmit sensing signals and receive echo signals.

[0019] In some embodiments, after updating the predicted state, the method further includes: The covariance matrix is ​​updated using the Joseph form to enhance numerical stability.

[0020] In some embodiments, the step of designing the predicted beamforming vector based on the channel parameters corresponding to the serving base station specifically includes: Extract the azimuth and elevation angles corresponding to the main path or the strongest path from the channel parameters corresponding to the serving base station; The transmission guidance vector is constructed based on the azimuth and elevation angles and used as the predicted beamforming vector.

[0021] To achieve the above objectives, another aspect of this application proposes a multi-base station cooperative millimeter-wave sensing integrated system, the system comprising: Multiple base stations, each used to transmit sensing signals to a target vehicle, receive echo signals reflected by the target vehicle, estimate local sensing parameters of the target vehicle relative to that base station based on the echo signals, and upload the local sensing parameters to a cloud processing center; and... The cloud processing center is used to predict the state of the target vehicle based on the local sensing parameters uploaded by each base station, and obtain the predicted state; based on the predicted state, it uses the channel knowledge graph to query the theoretical observation vector corresponding to the predicted state, and determines the serving base station and predicted beam for the next time slot; in the next time slot, it receives the actual observation vectors uploaded by each base station; it calculates the innovation statistic based on the predicted state and the actual observation vector, and determines whether the innovation statistic exceeds a preset threshold; if it does not exceed the preset threshold, it filters and updates the predicted state based on the actual observation vector; if it exceeds the preset threshold, it determines that a beam anomaly has occurred, triggering a beam correction and realignment mechanism: constructing a local candidate state set centered on the predicted state; for each candidate state, querying the theoretical observation vector corresponding to the candidate state through the channel knowledge graph; calculating the consistency cost function based on the theoretical observation vector and the actual observation vector; selecting the candidate state with the smallest consistency cost function as the corrected state; re-querying the channel knowledge graph based on the corrected state; reconstructing the channel state information; and re-determining the serving base station and beam direction.

[0022] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0023] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0024] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0025] The embodiments of this application include at least the following beneficial effects: First, by estimating the local sensing parameters of the target vehicle through each base station and uploading them to the cloud processing center, the cloud processing center performs state prediction and theoretical observation vector query based on the channel knowledge graph (CKM). By comparing the innovation statistics with a preset threshold, quantitative detection of beam tracking anomalies is achieved, solving the problem of the lack of anomaly judgment methods in the prior art. Second, when the innovation statistics are detected to exceed the threshold, a local candidate state set is constructed with the predicted state as the center, and the theoretical observation vector corresponding to each candidate state is queried using CKM. The optimal corrected state is searched based on the minimum consistency cost function, so that the system can quickly complete beam deviation correction after a single observation without relying on the gradual convergence under the traditional filtering framework. This effectively avoids beam mismatch and tracking interruption caused by occlusion enhancement, local observation anomalies, or sensing link degradation. Finally, after obtaining the corrected state, the CKM is queried again based on the corrected state and the channel state information is reconstructed, which can quickly restore beam alignment without re-performing a complete beam scan or pilot training. This significantly reduces the overhead and time delay of anomaly recovery and ensures the continuity and robustness of beam tracking. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the system model considered in the embodiments of this application; Figure 2 This is a schematic diagram of the transmission protocol used in the embodiments of this application; Figure 3 This is a comparison diagram of the position errors of different methods within the tracking time slot in the embodiments of this application; Figure 4 This is a comparison diagram of the velocity errors of different methods within the tracking time slot in the embodiments of this application; Figure 5 This is a comparison chart of the average effective rates of different methods in the embodiments of this application under different numbers of base station transmit antennas; Figure 6 This is a flowchart illustrating the steps of a beam tracking method for a multi-base station cooperative millimeter-wave sensing integrated system provided in this application embodiment. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0029] Current research on beam tracking in millimeter-wave ISAC systems suffers from the following shortcomings: First, there is the mismatch between the motion model and the measurement model. Most existing technologies use an approximate constant velocity model to describe vehicle motion and establish a mapping relationship between vehicle state and measurement parameters based on ideal geometric relationships. However, in complex urban road environments, vehicle motion is easily affected by factors such as turning, acceleration / deceleration, traffic congestion, and road constraints, exhibiting significant time-varying and maneuverability. Simultaneously, the millimeter-wave propagation environment is complex, and there is often a strong nonlinear relationship between vehicle state and observed parameters. The simplified motion model and ideal measurement model used in existing methods cannot accurately characterize these characteristics, easily leading to motion model mismatch and measurement model mismatch problems, thereby reducing the accuracy of vehicle state prediction and beam tracking stability.

[0030] Second, acquiring real-time channel state information (CSI) incurs high overhead. Currently, downlink rate calculation and beam selection in millimeter-wave ISAC systems typically rely on accurate CSI. Most existing methods acquire CSI through real-time pilot training or online channel estimation and then use this information for beam adjustment. However, continuous beam tracking scenarios have high real-time requirements. Frequent pilot training and online channel estimation introduce significant time and signaling overhead, reducing effective communication time and making it difficult to meet the real-time beam update requirements of high-speed moving vehicle scenarios.

[0031] Third, there is a lack of effective beam correction and realignment mechanisms. In complex urban millimeter-wave ISAC scenarios, vehicle beam tracking is easily affected by factors such as building obstruction, local observation anomalies, sensing link degradation, and multipath propagation changes. Existing beam tracking methods typically rely on current observation information for state updates and beam adjustments, lacking effective beam correction and realignment mechanisms for abnormal observations and link degradation. When measurement information is unreliable or local link quality deteriorates, the system struggles to promptly identify and effectively correct tracking deviations, easily leading to beam mismatch, tracking interruptions, and degraded communication performance.

[0032] In view of this, this application provides a beam tracking method, system, electronic device, storage medium, and program product for a multi-base station cooperative millimeter-wave sensing integrated system based on channel knowledge graphs, to solve one of the following prior art problems: 1) Solving the problems of vehicle motion model mismatch, measurement model mismatch, and insufficient continuous beam tracking accuracy in millimeter-wave ISAC systems in complex urban environments. 2) Reducing the pilot training overhead and online channel estimation complexity caused by real-time CSI acquisition during continuous beam tracking, and improving the real-time performance of beam updates in high-speed moving scenarios. 3) Alleviating beam mismatch and tracking interruption problems under conditions such as enhanced occlusion, local observation anomalies, and sensing link degradation, and improving the communication stability and beam tracking robustness of millimeter-wave ISAC systems in complex urban environments.

[0033] like Figure 6 As shown, this embodiment provides a beam tracking method for a multi-base station cooperative millimeter-wave sensing integrated system, including the following steps: Step S1: Multiple base stations transmit sensing signals to the target vehicle and receive the echo signals reflected by the target vehicle. Each base station estimates the local sensing parameters of the target vehicle relative to the base station based on the echo signals and uploads the local sensing parameters to the cloud processing center. Step S2: The cloud processing center performs state prediction on the target vehicle based on the local sensing parameters uploaded by each base station to obtain the predicted state; Step S3: Based on the predicted state, the cloud processing center uses the channel knowledge graph to query the theoretical observation vector corresponding to the predicted state, and determines the serving base station and predicted beam for the next time slot; Step S4: In the next time slot, the serving base station uses the predicted beam to perform downlink data transmission, each base station receives the actual echo signal and estimates the actual observation vector, and uploads the actual observation vector to the cloud processing center; Step S5: The cloud processing center calculates the innovation statistic based on the predicted state and the actual observation vector, and determines whether the innovation statistic exceeds a preset threshold; Step S6: If the innovation statistic does not exceed the preset threshold, the cloud processing center filters and updates the predicted state based on the actual observation vector to obtain the updated state estimate; Step S7: If the innovation statistic exceeds the preset threshold, the cloud processing center determines that a beam anomaly has occurred and triggers a beam correction and realignment mechanism. The beam correction and realignment mechanism includes: constructing a local candidate state set centered on the predicted state; for each candidate state in the local candidate state set, querying the theoretical observation vector corresponding to the candidate state through the channel knowledge graph, and calculating a consistency cost function based on the theoretical observation vector and the actual observation vector; selecting the candidate state with the smallest consistency cost function as the corrected state; and re-querying the channel knowledge graph based on the corrected state to reconstruct the channel state information and redetermine the serving base station and beam direction.

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

[0035] (1) System Model This embodiment considers a multi-base station cooperative millimeter-wave ISAC system, such as... Figure 1 As shown. The system has a total of M There are base stations, and the index set is denoted as . These base stations work collaboratively to provide communication and sensing services to a high-speed, single-antenna vehicle. Each base station is equipped with a transmit uniform planar array (UPA) and a receive UPA, with the number of transmit antennas and receive antennas respectively. and ,in and These represent the number of antennas in the horizontal and vertical directions of the transmitting array, respectively. and These represent the number of antennas in the receiving array in the horizontal and vertical directions, respectively. Let the total observation time be... T and divide it into A length of Discrete time slots. For ease of analysis, it is assumed that the vehicle state and channel parameters remain constant within each time slot.

[0036] The transmission protocol in this embodiment is as follows: Figure 2 As shown. Specifically, communication and sensing are executed in parallel within each time slot. In the... n Within each time slot, the serving base station determined by the previous time slot... Downlink data transmission continues, while all base stations simultaneously perform target perception. The perception process consists of two phases. In phase one, each base station transmits perception signals and receives target echoes, then processes the echo signals to obtain local parameter estimation results. In phase two, each base station uploads its local parameters to the Cloud Processing Center (CPC), where the CPC performs multi-base station data fusion and vehicle state prediction. Based on the predicted state, the CPC uses CKM to reconstruct the first... n +1 time slot CSI, and further complete the serving base station selection and predictive beamforming design. In the first n In the +1 time slot, each base station performs communication and sensing based on the predicted beam. After obtaining the actual observation of the time slot, the system determines whether a tracking anomaly has occurred based on the degree of mismatch between the predicted result and the actual observation; if an anomaly is detected, a beam correction and realignment mechanism is further triggered.

[0037] (2) Signal Model This embodiment uses OFDM as the ISAC waveform. This is because OFDM is not only a typical modulation method commonly used in millimeter-wave systems, but also facilitates joint processing of time delay and Doppler information in both the frequency and time domains, thus making it suitable for subsequent parameter estimation and beam tracking. Assume a base station... The carrier frequency is bandwidth is B This frequency band is divided into K If there are subcarriers, then the subcarrier spacing is . Each OFDM block contains L A series of consecutive OFDM symbols, each with a duration of [duration missing]. ,in and These are the duration of the effective symbol and the duration of the cyclic prefix, respectively.

[0038] Set up base station m In the n The first time slot, the first l The OFDM symbol, the first k The data symbols transmitted on each subcarrier are Base station m In the n The transmit beamforming vector for each time slot is defined as follows: ,in and Indicates the first n Each time slot vehicle relative to the base station m azimuth and elevation angles It is the launch steering vector, and Assuming the antenna spacing is half a wavelength, the direction cosine is defined as... , Then the launch steering vector can be expressed as ,in (1) (2) Therefore, the base station's transmitted signal in the frequency domain is represented as ,in This represents the base station's transmit power. Further, after K-point inverse discrete Fourier transform and cyclic prefix insertion, the base station... m In the n Time slot, moment t The transmitted baseband time-domain signal is represented as follows: (3) in This indicates a transmit pulse shaping filter. Subsequently, the baseband signal is up-converted to the carrier frequency. And it is launched by the launch array.

[0039] (3) Perception model OFDM sensing signals transmitted by each base station are reflected by the vehicle and received by their corresponding receiving arrays. Since different base stations operate in mutually orthogonal, non-overlapping frequency bands, signal components from other base stations can be effectively suppressed by bandpass filters. Furthermore, it is assumed that clutter caused by stationary scatterers has been pre-filtered using clutter cancellation methods. Under the far-field assumption, the vehicle's transmit and receive angles relative to the base station are approximately the same, and therefore uniformly expressed as azimuth-elevation pairs. Therefore, the sensing channel of base station m in the nth time slot can be expressed as: (4) in P Indicates the number of perception paths. , , , and The first p The reflection coefficient, round-trip delay, round-trip Doppler shift, azimuth angle, and elevation angle of the path. This is for receiving the steering vector. In real-world vehicle perception scenarios, the line-of-sight (LoS) path typically dominates the echo power, while the echo power of other reflected paths is usually much weaker. Therefore, for ease of analysis, unless otherwise specified, only the dominant path with the strongest power will be retained in subsequent analyses. Thus, the base station... m In the n The baseband echo signal received in each time slot can be expressed as: (5) in It is additive white Gaussian noise (AWGN). After performing analog-to-digital conversion, cyclic prefix removal, and K-point discrete Fourier transform on the received signal, the... l The OFDM symbol, the first k The frequency domain echo signal on each subcarrier can be represented as: (6) (4) Communication model Considering the finite scattering characteristics of millimeter-wave channels, this embodiment uses the Saleh-Valenzuela geometric channel model to model it. Let the... n The index of base stations providing communication services to vehicles in each time slot is: Then the vehicle is in the 1st l The OFDM symbol, the first k The downlink communication channel on each subcarrier can be represented as: (7) in I Indicates base station The total number of propagation paths between vehicles, For the first i The complex gain of the path. Therefore, the first path... n The signal received by a vehicle in each time slot is represented as follows: (8) in Let be the vehicle's AWGN. Then, the downlink speed at the vehicle is expressed as: (9) (5) CKM construction The channel knowledge graph (CKM) is a key tool for beam tracking in this embodiment, used to quickly obtain channel parameters related to the propagation environment based on vehicle status. Specifically, the CKM can be viewed as a mapping from vehicle position and speed to a set of multipath channel parameters. For the base station... m The corresponding CKM is defined as: (10) in and Let represent the vehicle's position vector and velocity vector, respectively. Further, the CKM of the entire network can be represented as... .

[0040] In this embodiment, the CKM is mainly used to quickly query the multipath channel parameters at the corresponding location based on the predicted state of the vehicle, thereby providing environmental prior support for beam tracking, CSI reconstruction, and beam correction. Therefore, the modeling accuracy of the CKM directly affects the beam tracking performance of the proposed method. Therefore, this embodiment uses a deep learning (DL)-based method to construct the CKM offline. Compared to traditional interpolation methods, the DL-based method can more effectively learn the complex nonlinear mapping relationship between vehicle state and channel parameters, and is more suitable for characterizing the environment-dependent propagation characteristics in complex road environments. Since the training of the CKM is completed offline, only a fast query needs to be performed in the online stage, which will not significantly increase the computational overhead of real-time beam tracking.

[0041] (6) Problem Description The objective of this embodiment is to maximize the average downlink rate of the vehicle through predictive beam tracking within continuous tracking time slots. Therefore, the optimization problem can be formulated as (P1): (11) Solving the optimization problem (P1) faces the following challenges. First, this embodiment considers the continuous beam tracking problem for high-speed moving vehicles. The vehicle state and channel parameters change dynamically over time, resulting in significant coupling of beam design in the time dimension. Second, in complex road environments, vehicle motion is often affected by factors such as turning, acceleration / deceleration, and obstruction, exhibiting obvious time-varying and environment-dependent characteristics. Traditional near-constant velocity motion models and ideal geometric measurement models are difficult to accurately characterize real motion and propagation patterns. Third, the calculation of downlink rate for each time slot depends on accurate CSI, but real-time CSI estimation is usually accompanied by high training and feedback overhead, and statistical CSI is difficult to meet the requirements of continuous and accurate tracking. Finally, due to factors such as prediction errors, sudden maneuvers, and local obstruction, beam anomalies or even tracking interruptions may occur during continuous tracking. Therefore, the system also needs to have the ability to detect and correct anomalies in a timely manner. To address these challenges, this embodiment proposes a two-stage CKM-enabled predictive beam tracking method, which achieves robust beam tracking of continuous time slots by jointly performing temporal state prediction, CKM-assisted CSI reconstruction, and beam correction.

[0042] (7) CKM-enabled predictive beam tracking method This embodiment proposes a two-stage CKM-enabled predictive beam tracking method for continuous beam tracking of millimeter-wave ISAC systems in complex urban environments. In the first stage, each base station estimates the local sensing parameters of the vehicle based on the received echo signals and uploads the estimation results to the central processing unit (CPC). In the second stage, the CPC, based on the local estimation information uploaded by each base station and combined with CKM, completes vehicle state prediction, channel state information reconstruction, communication base station selection, and predictive beam design to improve the real-time performance, accuracy, and stability of beam tracking. The specific process is as follows: 7.1) Phase 1: Local Parameter Estimation In Phase 1, each base station performs local parameter estimation based on the received target echo, thereby obtaining the vehicle's spatial angle, round-trip time delay, and round-trip Doppler shift relative to each base station. Subsequently, each base station uploads its local estimation results to the CPC for subsequent multi-base station data fusion, vehicle state prediction, and predictive beamforming design.

[0043] A) Angle Estimation: Assuming the initial beam acquisition phase has been completed and the vehicle target has been reliably detected through initial detection and beam scanning, this embodiment focuses on the parameter estimation problem during the continuous tracking phase. Let the base station... m In the n The frequency domain echo signal received in each time slot is Its expression is given by equation (6). First, the sample covariance matrix is ​​constructed using echo samples from all subcarriers and OFDM symbols, i.e.: (12) Secondly, for Perform eigenvalue decomposition to obtain ,in It is an eigenvalue diagonal matrix. Let be the corresponding feature vector matrix. Since this embodiment considers a single-target tracking scenario, the signal subspace has a dimension of 1, and the corresponding noise subspace is denoted as . Then, a spatial spectral function is constructed based on the two-dimensional music: (13) By performing a peak search in the two-dimensional angle domain, the vehicle's position relative to the base station can be obtained. m The azimuth and elevation angles are estimated, i.e. .

[0044] B) Time Delay and Doppler Shift Estimation: This involves obtaining a two-dimensional angle estimate. After that, the base station m A receiving beam is constructed along the target direction, and spatial combining and signal compensation are performed on the echo signal. The receiving beamforming vector is defined as follows: Therefore, the scalar echo after spatial merging and compensation can be expressed as: (14) in: (15) This represents the noise term after merging. As can be seen from equation (14), after spatial merging and symbol compensation, the echo signal exhibits the same characteristics as the time delay in the subcarrier dimension and OFDM symbol dimension, respectively. and Doppler shift The related two-dimensional exponential structure. Further, a two-dimensional data matrix is ​​constructed. , of which Each element is defined as Based on equation (14), It can be represented as: (16) in For the noise matrix, For delay steering vector, This is the Doppler steering vector. Because... and Both exhibit a Vandermonde structure, thus the MUSIC algorithm can be used to estimate the time delay and Doppler frequency shift separately. Specifically, the covariance matrices of the Doppler dimension and the time delay dimension are defined as follows: (17) (18) To each and After performing eigenvalue decomposition, the corresponding noise subspace can be obtained. and Therefore, the MUSIC spectral functions for Doppler and time delay can be expressed as follows: (19) (20) By searching for the peak values ​​of the spectral function, estimates of the Doppler frequency shift and time delay can be obtained, i.e. , At this point, base station m has obtained the local parameter estimation results for the vehicle in the nth time slot. .

[0045] 7.2) Phase Two: Predictive Beam Tracking Enabled by CKM In Phase Two, each base station first uploads its local parameter estimation results to the CPC. The CPC then performs multi-base station data fusion to complete vehicle state estimation and state prediction for the next time slot. Based on the predicted state, the CPC reconstructs the CSI using CKM and accordingly completes the selection of the serving base station and the predicted beam design for the next time slot. If a tracking anomaly is detected, a CKM-assisted beam correction and realignment mechanism is further activated.

[0046] A) Predictive Beam Tracking: The goal of beam tracking is to accurately estimate the vehicle's motion state in consecutive time slots and pre-generate appropriate beamforming for the next time slot to improve the average downlink rate throughout the tracking process. Since EKF (Expected Kinematics) can recursively estimate nonlinear dynamic systems under conditions of system and observation noise, it has been widely used in tasks such as target tracking and autonomous navigation. In the system under consideration, beam tracking based on observations from M base stations can essentially be modeled as a multi-sensor information fusion EKF problem, where the observations include the vehicle's angle, time delay, and Doppler shift relative to each base station, and the state variables are the vehicle's position and velocity. Therefore, this embodiment uses the EKF framework to implement multi-base station cooperative beam tracking.

[0047] However, traditional EKF beam tracking methods typically employ near-constant velocity models to describe target motion and rely on idealized geometric relationships to explicitly construct the measurement model. This modeling approach often has two limitations in complex urban road environments. First, vehicle motion is usually affected by factors such as turning, acceleration / deceleration, lane changing, and traffic flow disturbances, exhibiting significant time-varying and highly maneuverable characteristics. Simple near-constant velocity models cannot accurately depict the real motion patterns. Second, in complex propagation environments, the mapping relationship between vehicle states and parameters such as angles, time delays, and Doppler shifts has significant environmental dependence and nonlinear characteristics. If an idealized geometric model is used, model mismatch is easily generated, thereby reducing fusion accuracy and beam prediction performance. Therefore, we propose an EKF beam tracking method that combines a temporal prediction network and CKM. Specifically, an LSTM temporal network is used to learn the temporal evolution of vehicle motion from historical state sequences to generate prior state predictions that better reflect actual maneuverability. Meanwhile, by using CKM to establish a nonlinear environmental perception mapping between vehicle status and multi-base station observation parameters, angle, time delay and Doppler frequency shift are uniformly incorporated into the EKF framework for fusion and updating.

[0048] The beam tracking framework based on EKF mainly includes a state evolution model, a measurement model, and beam tracking.

[0049] State evolution model: Define the vehicle in the state evolution stage. n The state vector of each time slot is Considering the significant temporal correlation of vehicle motion, this embodiment no longer employs an approximately constant speed model, but instead utilizes a historical state sequence of length S to construct an LSTM predictor. Define the... The historical status window corresponding to each time slot is: (twenty one) Then, the vehicle was in n The prior state prediction for each time slot can be expressed as: ,in This represents a nonlinear state prediction function implemented by LSTM. This is state evolution noise, and .

[0050] Measurement Model: Unlike traditional methods that explicitly construct the analytical geometric relationship between the vehicle and sensing parameters, this embodiment uses CKM to establish the measurement function. For base stations m The corresponding measurement function is defined as follows: (twenty two) in , , and Let represent the mapping functions of the CKM corresponding to the m-th base station in terms of azimuth, elevation, time delay, and Doppler shift, respectively. Therefore, the vehicle measurement model is: (twenty three) in To measure the noise vector, and Since the measurement model in equation (22) is nonlinear, it needs to be linearized. In the predicted state... Nearby, the linearized form of equation (22) can be expressed as: (twenty four) in Represents measurement function Regarding the Jacobian matrix of the state vector. Since the CKM is constructed using a deep learning approach, it is difficult to directly obtain the closed derivative of the measurement function. Therefore, this embodiment uses the central difference method to numerically approximate the Jacobian matrix. Specifically, in predicting the state... Nearby, base station m The partial derivatives of each column can be expressed as: (25) in: (26) (27) (28) (29) in , , and They are respectively with , , and The corresponding standard basis vectors, , , and This indicates a sufficiently small perturbation step size.

[0051] Beam tracking based on EKF: in the first n In each time slot, CPC first constructs an input window based on the historical posterior state sequence obtained from the previous time slot. And use an LSTM network to complete the prior state prediction, that is The corresponding prediction error covariance matrix is ​​calculated as follows: (30) in This indicates that the LSTM prediction function has respect to the latest posterior state. The local Jacobian matrix. In calculating... In this case, the state estimates for earlier times within the historical window are treated as known constants, and only a small perturbation is applied to the last state component. Therefore, the Jacobian of the i-th column can be approximated using the central difference: (31) in and These represent the last state in the history window. The i Add or subtract disturbances The two input sequences were then obtained. Next, the local observation vectors from all base stations were stacked to form the overall observation vector. And define the joint observation function as Accordingly, the joint Jacobian matrix and the joint noise covariance matrix are expressed as follows: , Subsequently, the Kalman gain can be calculated as: (32) Furthermore, the state update can be represented as: (33) To enhance numerical stability, this embodiment uses the Joseph form to update the covariance matrix, expressed as: (34) After completing the first n After the state update of the first time slot, CPC further predicts the state of the second time slot based on the latest posterior state sequence. n The vehicle status of +1 time slot, i.e. Subsequently, for any base station m CPC is based on predicted state The set of channel parameters at this location is obtained by querying CKM, and is represented as follows: Then, CPC designs a predicted beamforming vector for each base station. Without loss of generality, this embodiment designs the beamforming using the angle corresponding to the principal diameter (or the strongest diameter), i.e. Each base station performs beam sensing and communication services for the next time slot based on the predicted beamforming.

[0052] B) Selection of communication service base stations: In the... Before each time slot begins, the CPC needs to determine the base stations that will provide communication services to the vehicles. We use the achievable speed of each base station as the criterion for selecting communication service base stations. Specifically, the CPC first bases on... Reconstructing the first n CSI with +1 slot is represented as: (35) Then calculate the first using equation (9). n +1 time slot of predicted average rate Therefore, the first The communication service base station for each time slot is selected as .

[0053] C) CKM-assisted beam correction and realignment mechanism: Although the proposed two-stage CKM-enabled predictive beam tracking method can generate beams in advance for the next time slot, when the vehicle makes a sharp turn, accelerates or decelerates rapidly, experiences enhanced local obstruction, or suffers degradation in the observation quality of some base stations, the predicted beam obtained based on the information from the previous time slot may still deviate from the actual target direction, thus causing beam mismatch and even beam tracking interruption. Therefore, this embodiment proposes a CKM-assisted beam correction and realignment mechanism.

[0054] Specifically, in the n After the n+1th time slot ends, the system completes state prediction, serving base station selection, and beam pre-configuration for the n+1th time slot. Subsequently, in the n+1th time slot... n In the +1 time slot, the serving base station uses predicted beams to perform downlink data transmission, while all base stations receive target echoes and complete local parameter estimation. Then, each base station uploads its local parameter estimation results to the CPC to form the actual observation vector. Furthermore, CPC utilizes predicted states Compared with the actual observed vector Constructing a joint innovation vector The corresponding innovation covariance matrix is: Therefore, the normalized innovation statistic is defined as follows: .when Exceeding the preset threshold When the system determines that there is a beam anomaly or state mismatch in the (n+1)th time slot, it triggers a correction mechanism.

[0055] During the correction phase, the standard EKF update results are no longer used directly; instead, the predicted state is used. Centered on a given point, a set of local candidate states is constructed within its neighborhood. The nearby local candidate region is defined as: (36) in and These represent the local search radii in the position and velocity domains, respectively. Subsequently, the system uses the observation information from the current time slot to perform consistency matching on the candidate states. For any candidate state... The corresponding joint theoretical observation vectors can be obtained through CKM query. Therefore, the consistency cost function is defined as follows: (37) This cost function characterizes the degree of mismatch between the theoretical observations corresponding to the candidate states and the current actual observations. Therefore, the optimal corrected state can be expressed as: After obtaining the corrected state estimate, CPC further... The CKM is re-queried, the corrected CSI is reconstructed, and the beamforming vector is recalculated to complete the beam realignment.

[0056] (8) Simulation experiment In this embodiment, the effectiveness of the proposed CKM-enabled predictive beam tracking algorithm is verified through simulation analysis. To comprehensively evaluate the performance of this method, it is compared with the following four typical methods: Algorithm S1: Utilizes LSTM as the state evolution model, CKM as the measurement function, and employs multiple base stations for collaborative tracking.

[0057] Algorithm S2: It uses a constant velocity model as the state evolution model, CKM as the measurement function, and multiple base stations for collaborative tracking.

[0058] Algorithm S3: It uses a constant velocity model as the state evolution model, a geometric model as the measurement function, and multiple base stations for collaborative tracking.

[0059] Algorithm S4: Uses a constant velocity model as the state evolution model, a geometric model as the measurement function, and a single base station for beam tracking.

[0060] Figure 3 The paper demonstrates the position errors of different methods within the tracking time slot. It can be seen that the proposed method maintains the lowest and most stable error throughout the entire tracking process, with a root mean square error (RMS) of only 1.019 m, significantly outperforming other comparative algorithms. Algorithm S1 performs second best, with an RMS error of 2.231 m. This indicates that under the same CKM measurement function and multi-base station cooperative tracking conditions, the LSTM state evolution model can more accurately characterize the vehicle motion process. However, due to the lack of a beam correction mechanism, Algorithm S1 struggles to restore beam pointing in time when occlusion or local tracking failure occurs, thus its performance remains inferior to the proposed method. Although Algorithm S2 introduces the CKM measurement function, it still uses a constant velocity state model, making it difficult to adapt to the dynamic changes in vehicle motion under complex road scenarios, thus further increasing the position error. Algorithm S3 shows significant error accumulation in the later stages of tracking, indicating that in complex urban environments and under random occlusion conditions, relying solely on a constant velocity state model and geometric measurement model is insufficient to accurately reflect the true state evolution and observation relationship, easily leading to error accumulation. It is worth noting that the error of algorithm S3 is even greater than that of algorithm S4. This indicates that when both the state model and the measurement model are mismatched, multi-base station fusion not only fails to bring performance gains, but may also exacerbate the error due to the amplification of observation biases from each base station. In contrast, the proposed method integrates the LSTM state evolution model, the CKM measurement function, and the CKM-assisted correction mechanism, thus significantly improving position tracking accuracy.

[0061] Figure 4The velocity errors of different methods within the tracking time slot are shown. It can be seen that the proposed method achieves the best performance, with a root mean square error of only 1.212 m / s. Algorithm S1 outperforms the other comparative methods, indicating that accurate state evolution modeling is crucial for velocity estimation during beam tracking. Although Algorithm S1 does not introduce a correction mechanism, its LSTM effectively characterizes the dynamic changes in the vehicle's historical state, thus exhibiting stronger velocity tracking capability compared to methods based on constant velocity models. Algorithm S2's constant velocity state model struggles to adequately adapt to the dynamic changes in vehicle speed under complex traffic conditions, and its prediction error is further propagated to the filtering update process, thus limiting the overall velocity estimation accuracy. Algorithm S3 exhibits the largest velocity error, indicating that when both the state model and measurement model employ traditional idealized modeling, the system struggles to achieve stable and reliable velocity tracking under complex propagation environments and random occlusion conditions. In contrast, the method proposed in this application simultaneously improves the accuracy of state prediction and measurement modeling, and suppresses error propagation through a correction mechanism when observations degrade, thus maintaining a low and stable velocity error throughout the entire tracking process.

[0062] Figure 5 The average effective rate (ARR) of different methods under varying numbers of transmit antennas at base stations is presented. As the number of transmit antennas increases, the ARR of all methods gradually improves. This is because larger transmit arrays can provide higher beamforming gain, thereby improving the receive signal-to-noise ratio (SNR) and spectral efficiency. The proposed method consistently achieves the highest ARR across all antenna configurations. When the number of transmit antennas is 64, its ARR approaches 7 bps / Hz, significantly outperforming other comparative methods. This performance gain is primarily attributed to the proposed method's ability to directly reconstruct the downlink CSI using CKM, reducing pilot training overhead, and maintaining more stable beam alignment under obstruction or measurement quality degradation through the CKM-assisted correction mechanism. In contrast, Algorithm S1 lacks a correction mechanism and still requires pilot estimation of CSI during the communication phase, resulting in a lower ARR than the proposed method. Algorithm S2, employing a constant-velocity state model, struggles to accurately describe vehicle dynamics in complex urban scenarios, leading to further deterioration in beam prediction performance. Algorithm S3 further employs a geometric measurement model, which fails to adequately characterize non-ideal propagation characteristics in complex environments, thus further reducing its rate performance. Algorithm S4 relies on a single base station for service and tracking, lacking multi-base station collaboration and selection gain, thus resulting in the lowest average effective rate.

[0063] In summary, compared with the prior art, this application has at least the following advantages and beneficial effects: 1) To address the issues of vehicle motion model mismatch and measurement model mismatch in complex urban environments, this application proposes a predictive beam tracking framework for a multi-base station cooperative millimeter-wave ISAC system powered by CKM. This framework acquires vehicle echo information collaboratively through multiple base stations, and the cloud processing center uniformly completes state prediction, observation fusion, channel reconstruction, serving base station selection, and predictive beam design, thereby improving the overall performance of continuous beam tracking for vehicles.

[0064] 2) To improve the accuracy of vehicle state prediction, this application introduces a time-series prediction network to learn the historical motion patterns of vehicles, capturing dynamic changes caused by factors such as vehicle turning, acceleration / deceleration, and traffic disturbances. Compared to traditional near-constant velocity models, this method can better adapt to the time-varying and maneuverable nature of vehicle motion in complex urban roads, thereby reducing the impact of motion model mismatch on beam tracking accuracy.

[0065] 3) To alleviate the difficulty in accurately modeling the nonlinear mapping between vehicle state and observation parameters in complex propagation environments, this application utilizes CKM to construct an environmental perception mapping relationship between vehicle state and observation parameters such as spatial angle, time delay, and Doppler frequency shift. Furthermore, it implements multi-base station observation fusion and state update within the extended Kalman filter framework. This method can fully utilize prior environmental information, improving state estimation accuracy and tracking stability.

[0066] 4) To reduce the pilot training overhead and online channel estimation complexity caused by real-time CSI acquisition during continuous beam tracking, this application reconstructs CSI using CKM based on predicted vehicle state, and further completes the selection of the serving base station and the predicted beam design for the next time slot. Compared with traditional methods that rely on frequent pilot training, this method can reduce channel acquisition overhead, improve beam update real-time performance, and increase the effective communication rate of the system.

[0067] 5) To address the issues of beam mismatch and tracking interruption that can easily occur under conditions such as enhanced occlusion, local observation anomalies, and sensing link degradation, this application designs a CKM-assisted beam correction and realignment mechanism. When the system detects an anomaly in the sensing link, it can search for the candidate state that best matches the current observation within the predicted state neighborhood, and readjust the serving base station and beam direction accordingly, thereby improving beam tracking robustness and communication service continuity under abnormal scenarios.

[0068] 6) This application, through a multi-base station cooperative sensing and communication mechanism, can maintain more stable service base station handover and a smoother effective rate in environments with random obstruction and complex urban propagation. Compared with existing beam tracking methods, this application has superior performance in terms of position error, velocity error, and average effective rate, and can provide a more stable, reliable, and efficient continuous beam tracking scheme for millimeter-wave ISAC systems in complex urban environments.

[0069] This application also provides a multi-base station cooperative millimeter-wave sensing integrated system that can implement the above-mentioned method. The system includes: Multiple base stations, each used to transmit sensing signals to a target vehicle, receive echo signals reflected by the target vehicle, estimate local sensing parameters of the target vehicle relative to that base station based on the echo signals, and upload the local sensing parameters to a cloud processing center; and... The cloud processing center is used to predict the state of the target vehicle based on the local sensing parameters uploaded by each base station, and obtain the predicted state; based on the predicted state, it uses the channel knowledge graph to query the theoretical observation vector corresponding to the predicted state, and determines the serving base station and predicted beam for the next time slot; in the next time slot, it receives the actual observation vectors uploaded by each base station; it calculates the innovation statistic based on the predicted state and the actual observation vector, and determines whether the innovation statistic exceeds a preset threshold; if it does not exceed the preset threshold, it filters and updates the predicted state based on the actual observation vector; if it exceeds the preset threshold, it determines that a beam anomaly has occurred, triggering a beam correction and realignment mechanism: constructing a local candidate state set centered on the predicted state; for each candidate state, querying the theoretical observation vector corresponding to the candidate state through the channel knowledge graph; calculating the consistency cost function based on the theoretical observation vector and the actual observation vector; selecting the candidate state with the smallest consistency cost function as the corrected state; re-querying the channel knowledge graph based on the corrected state; reconstructing the channel state information; and re-determining the serving base station and beam direction.

[0070] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0071] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0072] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0073] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0074] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0075] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0076] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0077] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented in the embodiments of this program product are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments. The executable computer program code or "code" used to perform the various embodiments can be written in high-level programming languages ​​such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0078] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0079] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0080] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0081] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0082] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0083] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0084] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0085] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0086] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0087] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0088] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A beam tracking method for a multi-base station cooperative millimeter-wave sensing integrated system, characterized in that, The method includes the following steps: Multiple base stations transmit sensing signals to the target vehicle and receive echo signals reflected by the target vehicle. Each base station estimates the local sensing parameters of the target vehicle relative to the base station based on the echo signals and uploads the local sensing parameters to the cloud processing center. The cloud processing center performs state prediction on the target vehicle based on the local sensing parameters uploaded by each base station, and obtains the predicted state. Based on the predicted state, the cloud processing center uses the channel knowledge graph to query the theoretical observation vector corresponding to the predicted state, and determines the serving base station and predicted beam for the next time slot. In the next time slot, the serving base station uses the predicted beam to perform downlink data transmission. Each base station receives the actual echo signal and estimates the actual observation vector, and uploads the actual observation vector to the cloud processing center. The cloud processing center calculates the innovation statistic based on the predicted state and the actual observation vector, and determines whether the innovation statistic exceeds a preset threshold. If the innovation statistic does not exceed the preset threshold, the cloud processing center filters and updates the predicted state based on the actual observation vector to obtain the updated state estimate. If the innovation statistics exceed the preset threshold, the cloud processing center determines that a beam anomaly has occurred and triggers a beam correction and realignment mechanism. The beam correction and realignment mechanism includes: constructing a local candidate state set centered on the predicted state; for each candidate state in the local candidate state set, querying the theoretical observation vector corresponding to the candidate state through the channel knowledge graph, and calculating a consistency cost function based on the theoretical observation vector and the actual observation vector; selecting the candidate state with the smallest consistency cost function as the corrected state; and re-querying the channel knowledge graph based on the corrected state to reconstruct the channel state information and redetermine the serving base station and beam direction.

2. The method according to claim 1, characterized in that, The cloud processing center performs state prediction on the target vehicle, specifically including: The cloud processing center obtains the posterior state sequence of the target vehicle in historical time slots and constructs a historical state window. The historical state window is input into a pre-trained long short-term memory network, which then outputs the prior state prediction result of the target vehicle in the current time slot.

3. The method according to claim 1, characterized in that, The cloud processing center filters and updates the predicted state based on the actual observation vector, specifically including: Under the extended Kalman filter framework, the cloud processing center uses the channel knowledge graph as a measurement function to establish a nonlinear mapping relationship between the state vector of the target vehicle and the observation vectors of each base station. The measurement functions of each base station are linearized to obtain the joint Jacobian matrix; The Kalman gain is calculated based on the joint Jacobian matrix, and the predicted state is updated based on the Kalman gain and the actual observation vector to obtain the updated state estimate.

4. The method according to claim 1, characterized in that, The cloud processing center determines the serving base station and predicted beam for the next time slot, specifically including: Based on the predicted state, the cloud processing center obtains the set of channel parameters corresponding to the predicted state by querying the channel knowledge graph. Reconstruct the downlink channel state information between the target vehicle and each base station based on the set of channel parameters; The predicted reachability rate of each base station is calculated based on the downlink channel state information, and the base station with the highest predicted reachability rate is selected as the serving base station for the next time slot. Design a predictive beamforming vector based on the channel parameters corresponding to the serving base station.

5. The method according to claim 1, characterized in that, Each base station estimates the local perception parameters of the target vehicle relative to that base station based on the echo signal, specifically including: Each base station constructs a sample covariance matrix based on the received frequency domain echo signal, and performs eigenvalue decomposition on the sample covariance matrix to obtain the noise subspace; Based on the noise subspace, a two-dimensional MUSIC spatial spectrum function is constructed, and a peak search is performed in the two-dimensional angle domain to obtain the estimated azimuth and elevation angles of the target vehicle relative to the base station.

6. The method according to claim 5, characterized in that, The method of each base station estimating the local perception parameters of the target vehicle relative to that base station based on the echo signal further includes: After obtaining the estimated azimuth angle and the estimated elevation angle, each base station constructs a receiving beamforming vector along the target direction, performs spatial combining and symbol compensation on the echo signal, and obtains a scalar echo signal. Based on the scalar echo signal, construct the time-delay covariance matrix and the Doppler covariance matrix, and perform eigenvalue decomposition to obtain the time-delay noise subspace and the Doppler noise subspace, respectively. Based on the time-delay noise subspace, a time-delay MUSIC spectrum function is constructed and a peak search is performed to obtain the round-trip time delay estimate; based on the Doppler noise subspace, a Doppler MUSIC spectrum function is constructed and a peak search is performed to obtain the round-trip Doppler frequency shift estimate.

7. The method according to claim 3, characterized in that, The linearization of the measurement functions of each base station to obtain the joint Jacobian matrix specifically includes: The central difference method is used to numerically approximate the measurement function represented by the channel knowledge graph; Near the predicted state, positive and negative perturbations are applied to each dimension of the state vector. The corresponding perturbed observation vector is obtained by querying the channel knowledge graph. The approximate value of the partial derivative corresponding to that dimension is calculated based on the ratio of the difference between the perturbed observation vector and the theoretical observation vector corresponding to the predicted state to the perturbation step size.

8. The method according to claim 1, characterized in that, The cloud processing center calculates innovation statistics based on the predicted state and the actual observation vector, specifically including: The cloud processing center calculates a joint theoretical observation vector based on the predicted state and the channel knowledge graph. The difference between the actual observation vector and the joint theoretical observation vector is calculated to obtain the joint innovation vector; Based on the joint innovation vector and the pre-calculated innovation covariance matrix, the normalized innovation statistic is calculated.

9. The method according to claim 1, characterized in that, The construction of a local candidate state set centered on the predicted state specifically includes: With the position and velocity components of the predicted state as centers, candidate state spaces are constructed in the position and velocity domains respectively according to preset local search radii. The local candidate state set is obtained by sampling within the candidate state space according to a preset step size.

10. The method according to claim 4, characterized in that, The step of reconstructing the downlink channel state information between the target vehicle and each base station based on the channel parameter set specifically includes: The cloud processing center queries the channel knowledge graph based on the predicted state to obtain the multipath channel parameters at the corresponding locations of each base station. The multipath channel parameters include the complex gain, angle of arrival, delay, and Doppler shift of each propagation path. Based on the multipath channel parameters, a downlink channel matrix between each base station and the target vehicle is constructed, which serves as the reconstructed channel state information.

Images