Dual-channel fingerprint-driven cell-free MIMO low-overhead beam tracking method

Abstract

This invention provides a dual-channel fingerprint-driven, low-overhead beam tracking method for cellular-free MIMO, comprising a Kalman filter module and a dual-channel fingerprint fusion neural network beam prediction module. The implementation steps are as follows: at the beginning of the algorithm, the terminal position and attitude state information for the current time slot are predicted using Kalman filtering and navigation information from GPS and IMU; after obtaining the real-time prediction information, a coarse narrow beam selection probability is learned using the navigation prediction information, while another set of narrow beam prediction information is learned using sensing beam detection information; the two narrow beam selection probability information are fused to obtain the final beam prediction probability; and a gating network is introduced to reduce the complexity of the neural network. Compared with existing beam tracking schemes, this dual-channel fingerprint-driven scheme has stronger robustness, higher average performance, and lower beam scanning overhead.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, and particularly relates to a dual-channel fingerprint-driven cellular MIMO low-overhead beam tracking method. Background Technology

[0002] Beam tracking in wireless communication systems is a key technology for ensuring the stability of millimeter-wave communication links. In highly mobile scenarios, the terminal position and channel state change rapidly, placing higher demands on the real-time performance and accuracy of beam alignment. Traditional beam tracking methods typically rely on geometric or channel models, updating the beam through periodic beam scanning or model-driven prediction. However, in complex dynamic environments, channel non-stationarity increases, and model mismatch and measurement errors are unavoidable, resulting in high beam training overhead and high tracking latency, making it difficult to meet the requirements of low latency and high reliability communication.

[0003] To reduce beam tracking overhead and improve system performance, fingerprint-based beam tracking methods have gradually attracted attention. Navigation information such as pose can serve as geometric fingerprints, providing spatial priors for beam selection; the perceived beam measurement response can serve as a channel fingerprint, directly reflecting the characteristics of the current propagation environment. However, single-fingerprint-driven beam tracking schemes still have significant limitations: relying solely on pose fingerprints, navigation prediction errors can cause beam mismatch; relying solely on beam detection fingerprints requires additional measurement resources and lacks robustness under low signal-to-noise ratio or occlusion conditions.

[0004] In view of the above problems, it is necessary to construct a beam tracking mechanism that integrates multi-source channel fingerprint information. The dual-channel fingerprint beam tracking algorithm establishes a complementary relationship between geometric prior and real-time channel awareness by jointly utilizing pose fingerprint and beamfinding fingerprint: pose fingerprint is used to narrow the candidate beam search space and obtain a coarse beam selection, while beamfinding fingerprint is used to capture instantaneous channel changes for fine-grained tracking. Through the effective fusion of dual-channel fingerprint information, this method can improve beam tracking accuracy and robustness while reducing training overhead, providing an effective technical solution for high-speed wireless communication in highly maneuverable scenarios. Summary of the Invention

[0005] Technical Problem: Existing beam tracking methods rely on only a single information source in highly maneuverable scenarios, making them susceptible to navigation prediction errors or beam measurement instability, leading to beam mismatch, high training overhead, and degraded tracking performance. To improve beam tracking accuracy and robustness while reducing beam training overhead, this invention proposes a dual-channel fingerprint-driven, cellular-free MIMO low-overhead beam tracking method.

[0006] The present invention provides a dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking method, comprising the following steps: Step 1, System Modeling: Construct a downlink scenario model for joint transmission in a cellular-free system, including a channel model and a transmission model, and establish an optimization problem to maximize the achievable rate of joint transmission; Step 2, Overall Scheme: The overall scheme design of dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking is presented, including a navigation information prediction module based on Kalman filtering and a dual-channel fingerprint fusion neural network module based on multilayer perceptron (MLP) and convolutional neural network (CNN).

[0007] Step 3: Navigation Information Prediction Module: The navigation information prediction module obtains the necessary information through GPS (Global Positioning System) and IMU (Inertial Measurement Unit). Based on the observation information from GPS and IMU, it estimates the terminal state information for each beam tracking time slot using Kalman filtering.

[0008] Step 4: Dual-channel fingerprint fusion neural network module: The beam tracking problem is modeled using the prediction information from the first navigation information prediction module and the detection beam scanning information. Based on this, a gating network, a navigation information extraction module, a sensing beam detection information extraction module, and an adaptive fusion module are designed to construct a complete dual-channel fingerprint fusion neural network module.

[0009] Furthermore, the system modeling described in step 1 includes the following specific steps: Step 1.1: Considering a scenario where B base stations simultaneously serve one terminal in a non-cellular system for joint downlink transmission, define a cooperative base station set. Each base station is equipped with Uniform planar arrays with a uniform number of antennas, each terminal equipped with A uniform array with a certain number of antennas, wherein This refers to the number of antennas in the horizontal direction of the base station. This refers to the number of antennas in the vertical direction of the base station; This refers to the number of antennas in the horizontal direction of the terminal. The number of antennas in the vertical direction of the terminal; the base station is connected to the CPU via a forward link for unified information processing; the channel between base station b and the terminal. That is, the channel matrix is ​​a A complex matrix of dimension 1 is represented as: Where L represents the number of propagation paths, For the gain coefficients of the base station and terminal antenna arrays, It is the product of the number of antennas. This represents the complex gain of the l-th path, where l = 0 indicates a Loss of Suppression (LoS) path. It is the antenna array response vector on the terminal side. It is the antenna array response vector on the base station side. express The conjugate transpose of; Indicates the azimuth and elevation angles of the base station. The azimuth and elevation angles of the terminal are indicated. The base station and terminal antenna arrays adopt a uniform surface array with an adjacent antenna spacing of half a wavelength.

[0010] Step 1.2: For downlink transmission without cellular connectivity, configure the base station side to use single-stream transmission, and the channel between the terminal and all cooperating base stations is... ,in Representing the terminal and the first Channels between base stations Considering that the base station adopts a hybrid beamforming architecture, each base station uses a standard DFT codebook. Select one beam; the system's analog beamset is... , For the corresponding number The selected analog beam for each base station The terminal side adopts an analog beamforming architecture, and the terminal side uses the standard DFT codebook. Select one beam Beam-domain equivalent channel matrix constructed based on analog reception If the CPU uses MRT for digital precoding, then the digital precoding on the base station side is represented as follows: ; in , Let be the conjugate transpose of the equivalent channel matrix and the Frobenius norm of the corresponding matrix, respectively. The received signal of the terminal is represented as: ; in This represents the average normalized transmit power. This indicates the transmitted signal of the base station, satisfying... , Indicates received noise, satisfying , The noise power is represented by the terminal's SNR. ; Without considering the beam training time of the terminal, the achievable rate of the terminal is expressed as: ; Step 1.3: With the transmit beam fixed at the base station, different beam selections at the terminal will result in different achievable rates. The optimization problem of beam tracking in a cellular-free system to maximize the achievable rate is expressed as follows: .

[0011] That is, selecting the optimal terminal beam to maximize the achievable rate.

[0012] Furthermore, the overall design of the dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking scheme in step 2 includes the following specific steps: Step 2.1: Acquisition and time alignment of multi-source sensor information. The terminal acquires position information, attitude information and its rate of change through GPS and IMU. Since the sampling frequencies of GPS, IMU and beam tracking are inconsistent, in each beam tracking time slot, the position, velocity, attitude and angular velocity of the terminal are estimated and predicted in real time using a Kalman filter to obtain the predicted navigation state information of the current time slot.

[0013] Step 2.2: Feature selection and interference suppression driven by the gating network. The gating network learns the nonlinear mapping from the navigation state to the soft attention distribution on the sensing beam, generates a binary mask, selects an appropriate sensing beam for detection, and filters the output of the navigation information extraction module, thereby accelerating the network convergence of the navigation information extraction module.

[0014] Step 2.3: Coarse-grained beam prediction based on navigation fingerprint. A navigation information extraction module based on MLP is constructed using the predicted navigation state output by Kalman filter. The nonlinear mapping relationship between the navigation state and the optimal narrow beam is learned, and the narrow beam selection probability based on navigation fingerprint is learned.

[0015] Step 2.4: Fine-grained feature extraction based on perceptual beam fingerprint. Using the detection values ​​of the perceptual beam after screening by the gated network, a CNN-based perceptual beam feature extraction module is constructed to learn the spatial features related to the narrow beam direction and to learn the narrow beam selection probability based on perceptual beam sensing.

[0016] Step 2.5: Establish an adaptive probability fusion module driven by dual-channel fingerprint collaboration, normalize and concatenate the two narrow beam selection probabilities obtained, adaptively learn the importance weights of the two types of fingerprints under different conditions through the attention module, and input the weighted features into the fusion network to obtain the final narrow beam selection probability.

[0017] Furthermore, the navigation information prediction module described in step 3 includes the following specific steps: Step 3.1: Upon obtaining the observation information, the state transition equation and observation equation are used to predict the terminal's position and attitude. Consider that the changes in the terminal's position and attitude follow a linear pattern: , in , , , These represent the position, velocity, attitude, and angular velocity of the time slot t terminal along the o-xyz axes, respectively. For beam tracking time slot length, Here, represents the beam tracking frequency. Taking position and velocity state prediction as an example, assume the sampling frequency for acquiring observation information is . The sampling frequency of the observation information is the same as the beam tracking frequency. times, when When, the corresponding transfer equation and observation equation are: , , in, It is the identity matrix. , The process and observation of Gaussian white noise are represented.

[0018] Step 3.2, within the observation information acquisition interval, i.e. Assuming the terminal undergoes uniform linear motion, the state transition equation is used to predict the terminal state. , Real-time predicted values ​​of attitude and angular velocity are obtained through the transfer equation and observation equation. , , Furthermore, step 4 involves constructing a complete dual-channel fingerprint fusion neural network module, including the following specific steps: Step 4.1: Design the gating network. The input to the gating network is the navigation prediction information at time t-1. and the set of detection sensing beam indices at time t-1 The output is the detection beam index at time t. And generate the corresponding binary mask. .

[0019] Step 4.2: Navigation information feature extraction. The input to the navigation module is the navigation prediction information at time t. The output is the narrow beam selection probability based on navigation prediction information. ,in This represents the selection probability of the k-th beam. After filtering with a binary mask, the result is obtained .

[0020] Step 4.3, Sensing Beam Detection Information Extraction: At time t, the terminal extracts information from the sensing beam DFT codebook. Select the first one. A sensing beam The received sensing beam detection signal from the b-th base station is represented as: , in The number of antennas used for beam training on the terminal side. Let be the channel corresponding to time t. This is the pilot transmission power. Based on the output of the gating network. The input of the sensing beam module can be obtained. for ; After the spatial features are extracted by the sensing beam module, the narrow beam selection probability is obtained. , This represents the selection probability of the k-th beam. .

[0021] Step 4.4, Adaptive probability fusion, will , After concatenation and normalization, the results are fed into the attention module to generate weights. The weighted probabilities are then fed into the feature fusion network for fusion, yielding the final narrow beam selection probability. , This represents the selection probability of the k-th beam. The terminal side performs beam tracking based on this probability.

[0022] Beneficial effects: The dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking method proposed in this invention, compared with existing technologies that rely solely on navigation information or solely on sensing beam detection information, can adaptively adjust feature weights through an attention mechanism in scenarios with high attitude observation noise or limited pilot power, thereby improving the robustness of beam selection and maintaining a high communication achievable rate.

[0023] Meanwhile, this invention uses a gating network to select more representative sensing beams for detection, avoiding invalid measurements caused by random beam detection and improving beam detection efficiency. While ensuring communication performance, this invention can significantly reduce beam training overhead, achieving an effective balance between communication performance and system overhead, making it suitable for high-speed mobile terminals and dynamic wireless environments. Attached Figure Description

[0024] Figure 1 This is a flowchart of the dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking method of the present invention; Figure 2 This is a schematic diagram of the overall design process of the dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking scheme of the present invention. Figure 3 This is a schematic diagram of the timing flow of the navigation information prediction module of the present invention; Figure 4 This is a performance comparison chart of the fusion neural network proposed in this invention with a low-overhead benchmark scheme under different pose observation noise conditions; Figure 5 This is a performance comparison chart between the fusion neural network scheme proposed in this invention and the full-sensing beam scanning input scheme under different attitude observation noise conditions; Figure 6 This is a performance comparison chart of the fusion neural network scheme proposed in this invention with other beam tracking methods under different pilot power. Detailed Implementation

[0025] To better understand the purpose, structure, and function of this invention, the following detailed description of the dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking method of this invention, in conjunction with the accompanying drawings, is provided.

[0026] like Figure 1 As shown, the dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking method proposed in this invention includes the following specific steps in its implementation: Step S101: Considering the scenario of B base stations simultaneously serving one terminal in a non-cellular system joint transmission downlink scenario, define a cooperative base station set. Each base station is equipped with Uniform planar arrays with a uniform number of antennas, each terminal equipped with A uniform array with a certain number of antennas, wherein , This refers to the number of antennas in the horizontal and vertical directions for the base station and terminal. The base station is connected to the CPU via a forward link for unified information processing. The channel between base station b and the terminal... That is, the channel matrix is ​​a A complex matrix of dimension 1 can be represented as: Where L represents the number of propagation paths, For the gain coefficients of the base station and terminal antenna arrays, It is the product of the number of antennas. This represents the complex gain of the l-th path, where l = 0 indicates a Loss of Suppression (LoS) path. It is the antenna array response vector on the terminal side. It is the antenna array response vector on the base station side. express The conjugate transpose of; Indicates the azimuth and elevation angles of the base station. The azimuth and elevation angles of the terminal are indicated. The base station and terminal antenna arrays adopt a uniform surface array with an adjacent antenna spacing of half a wavelength.

[0027] Step S102: For downlink transmission without cellular connectivity, assuming single-stream transmission is used on the base station side, the channel between the terminal and all cooperating base stations is... ,in Representing the terminal and the first Channels between base stations Considering that the base station adopts a hybrid beamforming architecture, each base station uses a standard DFT codebook. Select one beam; the system's analog beamset is... , For the corresponding number The selected analog beam for each base station The terminal side adopts an analog beamforming architecture, and the terminal side uses the standard DFT codebook. Select one beam Beam-domain equivalent channel matrix constructed based on analog reception. If the CPU uses MRT for digital precoding, then the digital precoding on the base station side can be expressed as: ; in , Given the conjugate transpose of the equivalent channel matrix and the Frobenius norm of the corresponding matrix, respectively, the received signal of the terminal can be expressed as: in This represents the average normalized transmit power. This indicates the transmitted signal of the base station, satisfying... , Indicates received noise, satisfying , This represents noise power. The SNR of a terminal can be expressed as: Without considering the beam training time of the terminal, the achievable rate of the terminal can be expressed as: Step S103, taking terminal-side beam tracking as an example, with the base station-side transmit beam fixed, different beam selections on the terminal side will result in different achievable rates. This chapter aims to select a suitable receive analog beam to maximize the achievable rate. Therefore, the optimization problem of beam tracking in a cellular system to maximize the achievable rate can be expressed as follows: ; That is, selecting the optimal terminal beam to maximize the achievable rate.

[0028] Step 2, Overall Scheme: The overall scheme design of dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking is presented, including a navigation information prediction module based on Kalman filtering and a dual-channel fingerprint fusion neural network module based on multilayer perceptron (MLP) and convolutional neural network (CNN).

[0029] The overall design of the dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking scheme described in step 2 is as follows: Figure 2 As shown, the specific steps include the following: Step S201: Acquisition and time alignment of multi-source sensor information. The terminal acquires position information, attitude information and their rate of change through GPS and IMU. Since the sampling frequencies of GPS, IMU and beam tracking are inconsistent, in each beam tracking time slot, the position, velocity, attitude and angular velocity of the terminal are estimated and predicted in real time using a Kalman filter to obtain the predicted navigation state information of the current time slot.

[0030] Step S202: Feature selection and interference suppression driven by the gating network. The gating network learns the nonlinear mapping from the navigation state to the soft attention distribution on the sensing beam, generates a binary mask, selects an appropriate sensing beam for detection, and filters the output of the navigation information extraction module to accelerate network convergence.

[0031] Step S203: Coarse-grained beam prediction based on navigation fingerprint. Construct a navigation information extraction module based on MLP using the predicted navigation state output by Kalman filter, learn the nonlinear mapping relationship between navigation state and optimal narrow beam, and learn the narrow beam selection probability based on navigation fingerprint.

[0032] Step S204: Based on the fine-grained feature extraction of the sensing beam fingerprint, a CNN-based sensing beam feature extraction module is constructed using the detection values ​​of the sensing beam after screening by the gated network. The module learns the spatial features related to the narrow beam direction and learns the narrow beam selection probability based on sensing beam perception.

[0033] Step S205: Establish an adaptive probability fusion module driven by dual-channel fingerprint collaboration, normalize and concatenate the two narrow beam selection probabilities obtained, adaptively learn the importance weights of the two types of fingerprints under different conditions through the attention module, and input the weighted features into the fusion network to obtain the final narrow beam selection probability.

[0034] Step 3, Navigation Information Prediction Module: Based on GPS and IMU observation information, this module predicts and estimates the terminal state information for each beam tracking time slot. The timing flow of the navigation information prediction module is as follows: Figure 3 As shown, the specific steps are as follows.

[0035] Step S301: Upon obtaining the observation information, the state transition equation and observation equation are used to predict the terminal's position and attitude. Consider that the changes in the terminal's position and attitude conform to linear changes: , in , , , These represent the position, velocity, attitude, and angular velocity of the time slot t terminal along the o-xyz axes, respectively. For beam tracking time slot length, Here, represents the beam tracking frequency. Taking position and velocity state prediction as an example, assume the sampling frequency for acquiring observation information is . The sampling frequency of the observation information is the same as the beam tracking frequency. times, when When, the corresponding transfer equation and observation equation are: , , in, It is the identity matrix. Let t be the terminal location observed by GPS. Let t be the terminal velocity observed by GPS. , The process and observation of Gaussian white noise are represented.

[0036] Step S302: Within the observation information acquisition interval, i.e. Assuming the terminal undergoes uniform linear motion, the state transition equation is used to predict the terminal state. , The real-time predicted values ​​of attitude and angular velocity are obtained through the following equations. , , Step 4: Model the beam tracking problem using the prediction information from the first module and the detection beam scanning information. Based on this, design a gating network, a navigation information extraction module, a sensing beam detection information extraction module, and an adaptive fusion module to construct a complete dual-channel fingerprint fusion neural network.

[0037] Step S401: Design the gating network. The input to the gating network is the navigation prediction information at time t-1. and the set of detection sensing beam indices at time t-1 The output is the detection beam index at time t. And generate the corresponding binary mask. .

[0038] Step S402: Navigation information feature extraction. The input to the navigation module is the navigation prediction information at time t. The output is the narrow beam selection probability based on navigation prediction information. ,in This represents the selection probability of the k-th beam. After filtering with a binary mask, the result is obtained .

[0039] Step S403: Extraction of sensing beam detection information. At time t, the terminal extracts information from the sensing beam DFT codebook. Select the first one. A sensing beam The received sensing beam detection signal from the b-th base station is represented as: , in The number of antennas used for beam training on the terminal side. Let be the channel corresponding to time t. This is the pilot transmission power. Based on the output of the gating network. The input of the sensing beam module can be obtained. for After the spatial features are extracted by the sensing beam module, the narrow beam selection probability is obtained. , This represents the selection probability of the k-th beam. .

[0040] Step S404, Adaptive probability fusion, will , After concatenation and normalization, the results are fed into the attention module to generate weights. The weighted probabilities are then fed into the feature fusion network for fusion, yielding the final narrow beam selection probability. , This represents the selection probability of the k-th beam. The terminal side performs beam tracking based on this probability.

[0041] Experimental results: The beam tracking method based on dual-channel fingerprinting in this embodiment is compared with the beam tracking method described below: A beam tracking method based on real-time pose information: This method uses real-time navigation information estimated by Kalman filtering and a neural network to predict narrow beams for beam tracking. This scheme relies solely on single-slot navigation estimation information for beam prediction.

[0042] The LSTM beam tracking method based on temporal pose information uses navigation estimation information from the past 16 time slots obtained by Kalman filtering and LSTM neural network to predict narrow beams and perform beam tracking.

[0043] A sensor beam detection and beam tracking method based on gating networks is proposed: This method utilizes sensor beam detection values ​​selected by a gating network and narrow beam prediction by a neural network. Navigation pose information is only used for filtering the sensor beam set and is not used for joint feature fusion of sensor beam measurement information and pose information.

[0044] The dual-channel fingerprint beam tracking method based on random sensing beam selection: The gated network only performs masking on the output results of the navigation information branch, while the set of sensing beams used by the sensing beam detection module is determined by a random strategy, thus forming a dual-channel fingerprint beam tracking method based on the joint processing of random sensing beam input and pose information.

[0045] A dual-channel fingerprint beam tracking method based on full-sensing beam scanning: The input to the sensing beam module is the detection information of all sensing beams. This scheme requires scanning all sensing beams, resulting in a significant increase in the training sample size and computational cost.

[0046] Ideal optimal beam like Figure 4 As shown, the beam tracking method based on dual-channel fingerprinting exhibits stronger performance and greater stability compared to other schemes when there is a high attitude observation error.

[0047] like Figure 5 As shown, with a 75% reduction in beam training overhead, the performance difference between the beam tracking method based on dual-channel fingerprinting and the scheme that does not use a gating network for sensing beam selection is very small.

[0048] like Figure 6 As shown, in low pilot power scenarios, the beam tracking method based on dual-channel fingerprinting exhibits strong robustness and shows a significant performance improvement compared to the beam tracking method that only utilizes sensing beam detection.

[0049] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A dual-channel fingerprint-driven, cellular-free MIMO low-overhead beam tracking method, characterized in that, Includes the following steps: Step 1: Construct a downlink scenario model for joint transmission in a cellular-free system, including a channel model and a transmission model, and establish an optimization problem to maximize the achievable rate of joint transmission. Step 2: The overall design of the dual-channel fingerprint-driven non-cellular MIMO low-overhead beam tracking scheme is given, including a navigation information prediction module based on Kalman filtering and a dual-channel fingerprint fusion neural network module based on multilayer neural network MLP and convolutional neural network CNN. Step 3: The navigation information prediction module obtains the information required by GPS and IMU, and estimates the terminal status information of each beam tracking time slot by Kalman filtering based on the observation information of GPS and IMU. Step 4: Model the beam tracking problem using the prediction information from the first navigation information prediction module and the detection beam scanning information. Based on this, design a gating network, a navigation information extraction module, a sensing beam detection information extraction module, and an adaptive fusion module to construct a complete dual-channel fingerprint fusion neural network module.

2. The dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking method according to claim 1, characterized in that: Step 1 includes the following specific steps: Step 1.1: Considering a scenario where B base stations simultaneously serve one terminal in a non-cellular system for joint downlink transmission, define a cooperative base station set. Each base station is equipped with A uniform array of antennas, each terminal equipped with A uniform array with a certain number of antennas, wherein This refers to the number of antennas in the horizontal direction of the base station. This refers to the number of antennas in the vertical direction of the base station; This refers to the number of antennas in the horizontal direction of the terminal. The number of antennas in the vertical direction of the terminal; the base station is connected to the CPU via a forward link for unified information processing; the channel between base station b and the terminal. That is, the channel matrix is ​​a A complex matrix of dimension 1 is represented as: ； Where L represents the number of propagation paths, For the gain coefficients of the base station and terminal antenna arrays, It is the product of the number of antennas. This represents the complex gain of the l-th path, where l = 0 indicates a Loss of Suppression (LoS) path. It is the antenna array response vector on the terminal side. It is the antenna array response vector on the base station side. express The conjugate transpose of; Indicates the azimuth and elevation angles of the base station. The azimuth and elevation angles of the terminal are indicated. The base station and terminal antenna arrays adopt a uniform surface array with an adjacent antenna spacing of half a wavelength. Step 1.2: For downlink transmission without cellular connectivity, configure the base station side to use single-stream transmission, and the channel between the terminal and all cooperating base stations is... ,in Representing the terminal and the first Channels between base stations ; Considering that the base station adopts a hybrid beamforming architecture, each base station uses a standard DFT codebook. Select one beam; the system's analog beamset is... , For the corresponding number The selected analog beam for each base station The terminal side adopts an analog beamforming architecture, and the terminal side uses the standard DFT codebook. Select one beam Beam-domain equivalent channel matrix constructed based on analog reception If the CPU uses MRT for digital precoding, then the digital precoding on the base station side is represented as follows: ； in , Let be the conjugate transpose of the equivalent channel matrix and the Frobenius norm of the corresponding matrix, respectively. The received signal of the terminal is represented as: ； in This represents the average normalized transmit power. This indicates the transmitted signal of the base station, satisfying... , Indicates received noise, satisfying , The noise power is represented by the terminal's SNR. ； Without considering the beam training time of the terminal, the achievable rate of the terminal is expressed as: ； Step 1.3: With the transmit beam fixed at the base station, different beam selections at the terminal will result in different achievable rates. The optimization problem of beam tracking in a cellular-free system to maximize the achievable rate is expressed as follows: ； That is, selecting the optimal terminal beam to maximize the achievable rate.

3. The dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking method according to claim 2, characterized in that: Step 2, the overall design of the dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking scheme, includes the following specific steps: Step 2.1: Acquisition and time alignment of multi-source sensor information. The terminal acquires position information, attitude information and its rate of change through GPS and IMU. Since the sampling frequencies of GPS, IMU and beam tracking are inconsistent, in each beam tracking time slot, the position, velocity, attitude and angular velocity of the terminal are estimated and predicted in real time using a Kalman filter to obtain the predicted navigation state information of the current time slot. Step 2.2: Feature selection and interference suppression driven by gating network. The gating network learns the nonlinear mapping from the navigation state to the soft attention distribution on the sensing beam, generates a binary mask, selects an appropriate sensing beam for detection, and filters the output of the navigation information extraction module to accelerate the network convergence of the navigation information extraction module. Step 2.3: Coarse-grained beam prediction based on navigation fingerprint. A navigation information extraction module based on MLP is constructed using the predicted navigation state output by Kalman filter. The nonlinear mapping relationship between the navigation state and the optimal narrow beam is learned, and the narrow beam selection probability based on navigation fingerprint is learned. Step 2.4: Fine-grained feature extraction based on perceptual beam fingerprint. Using the detection values ​​of the perceptual beam after screening by the gated network, a CNN-based perceptual beam feature extraction module is constructed to learn the spatial features related to the narrow beam direction and to learn the narrow beam selection probability based on perceptual beam sensing. Step 2.5: Establish an adaptive probability fusion module driven by dual-channel fingerprint collaboration, normalize and concatenate the two narrow beam selection probabilities obtained, adaptively learn the importance weights of the two types of fingerprints under different conditions through the attention module, and input the weighted features into the fusion network to obtain the final narrow beam selection probability.

4. The dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking method according to claim 3, characterized in that: The navigation information prediction module described in step 3 includes the following specific steps: Step 3.1: Upon obtaining the observation information, use the state transition equation and observation equation to predict the terminal's position and attitude; consider that the changes in the terminal's position and attitude conform to linear changes: ， in , , , These represent the position, velocity, attitude, and angular velocity of the time slot t terminal along the o-xyz axes, respectively. For beam tracking time slot length, Set the beam tracking frequency; set the sampling frequency for acquiring observation information to [value]. The sampling frequency of the observation information is the same as the beam tracking frequency. times, when When the time is right, the corresponding recurrence relation is: ； in The identity matrix has the following observation equation: ， in, Let t be the terminal location observed by GPS. Let t be the terminal velocity observed by GPS. The process represents Gaussian white noise. This indicates observed Gaussian white noise; Step 3.2, within the observation information acquisition interval, i.e. The terminal is set to move at a constant linear velocity, and the state transition equation is used to predict the terminal state: ， The real-time predicted values ​​of attitude and angular velocity are obtained through the following transfer equations and observation equations: The recurrence relation is: , The observation equation is: .

5. The dual-channel fingerprint-driven cellular-free MIMO low-overhead beam tracking method according to claim 4, characterized in that: Step 4 involves constructing a complete dual-channel fingerprint fusion neural network module, including the following specific steps: Step 4.1: Design the gating network. The input to the gating network is the navigation prediction information at time t-1. and the set of detection sensing beam indices at time t-1 The output is the detection beam index at time t. And generate the corresponding binary mask. ; Step 4.2: Navigation information feature extraction. The input to the navigation module is the navigation prediction information at time t. The output is the narrow beam selection probability based on navigation prediction information. ,in This represents the selection probability of the k-th beam. After filtering with a binary mask, the result is obtained ; Step 4.3, Sensing Beam Detection Information Extraction: At time t, the terminal extracts information from the sensing beam DFT codebook. Select the first one. A sensing beam The received sensing beam detection signal from the b-th base station is represented as: ， in The number of antennas used for beam training on the terminal side. Let be the channel corresponding to time t. For pilot transmission power; based on the output of the gating network The input of the sensing beam module is obtained. for ； After the spatial features are extracted by the sensing beam module, the narrow beam selection probability is obtained. , This represents the selection probability of the k-th beam. ; Step 4.4, Adaptive probability fusion, will , After concatenation and normalization, the results are fed into the attention module to generate weights. The weighted probabilities are then fed into the feature fusion network for fusion, yielding the final narrow beam selection probability. , This represents the selection probability of the k-th beam. The terminal side performs beam tracking based on this probability.

Images