Beam management method and apparatus

Abstract

The application provides a beam management method and device, applied to the field of wireless communication. The beam management method comprises the following steps: obtaining first position information of a terminal device, and then predicting second position information of the terminal device in a future first preset time based on the first position information. According to the second position information, a first beam set is determined, and beam scanning is performed on the first beam set to determine a transmitting beam and / or a receiving beam between the terminal device. In the method, the position information of the terminal device in the future time is predicted in advance before the beam scanning is performed, so that the motion track of the terminal device can be determined, and the beam scanning can be performed based on the prediction result, without the need of performing omnidirectional beam scanning, thereby improving the efficiency of the beam scanning and reducing the cost.

Description

Technical Field

[0001] The present application relates to the field of wireless communications, and in particular to a beam management method and device. Background Art

[0002] In communications scenarios, especially high-frequency ones, higher operating frequencies result in greater path loss when signals are transmitted across space. To increase communication distance, two communicating devices can perform beam tracking before communication to find an optimal communication link.

[0003] To determine the beam direction for beam alignment, communication devices typically need to perform exhaustive, indiscriminate beam scanning within a certain range (such as the coverage area) and determine the beam used for beam alignment based on the measurement results of each beam. This beam alignment process incurs a large amount of beam training overhead and can easily lead to tracking instability in complex scenarios. Summary of the Invention

[0004] The embodiments of the present application provide a beam management method and apparatus, which can improve the efficiency of beam scanning and reduce overhead.

[0005] In a first aspect, a beam management method is provided. The method can be performed by a beam management device, such as a network device, or a module within the network device. The method includes: obtaining first location information of a terminal device, the first location information including the terminal device's current location information and / or historical location information; processing the first location information using a preset trajectory prediction model to obtain second location information of the terminal device within a first preset time in the future; determining a first beam set based on the second location information; and performing beam scanning on the first beam set to determine a transmit beam and / or receive beam with the terminal device.

[0006] Based on the above solution, before beam scanning is performed, the terminal device's location information for a period of time in the future is predicted in advance, thereby determining the terminal device's future location and movement trend. This allows for targeted beam scanning based on the prediction results, eliminating the need for omnidirectional beam scanning. This significantly improves beam alignment efficiency, communication stability, and resource utilization. Related technologies rely on real-time position feedback when performing beam scanning, which can lead to problems after adjustments.

[0007] In combination with the first aspect, in some possible implementations of the first aspect, the area pointed to by the first beam set covers the position indicated by the second position information.

[0008] In the above scheme, the moving trajectory of the terminal device indicated by the second position information is within the area pointed to by each beam in the first beam set, that is, the maximum opening angle θ of the first beam set is determined by the position of the terminal device indicated by the second position information. This can significantly reduce the size of the first beam set, and there is no need to perform a full-range scan when performing beam scanning, which can significantly improve the scanning efficiency and reduce resource consumption.

[0009] In some possible implementations of the first aspect, obtaining first location information of a terminal device includes: sending a linear frequency modulation wave to the terminal device and receiving a corresponding echo signal from the terminal device; and determining the first location information based on the echo signal.

[0010] Based on the above solution, the precise location of the terminal device can be obtained through sensing technology. Combining this technology with beam management devices (such as network equipment) can not only solve the limitations of discrete systems in cost, efficiency and scenario adaptability in related technologies, but also provide a technical basis for subsequent terminal device location prediction.

[0011] In some possible implementations of the first aspect, the trajectory prediction model includes a convolutional neural network layer and a converter layer, the convolutional neural network layer is used to extract features of the first position information to obtain feature data; the converter layer is used to process the feature data to obtain the second position information.

[0012] This application uses a convolutional neural network-converter hybrid model to predict the location of the terminal device, decoupling feature extraction and location prediction, which can improve the precision and accuracy of location prediction, enable the beam management device to actively pre-adjust the beam direction, reduce signal interruption or quality degradation caused by terminal device movement, reduce invalid beam scanning, and improve spatial multiplexing capabilities in multi-user scenarios.

[0013] In some possible implementations of the first aspect, the convolutional neural network layer includes a first convolutional layer and a second convolutional layer, the convolution kernel size of the first convolutional layer is smaller than a preset threshold, the second convolutional layer adopts a strided convolution structure, the first convolutional layer is used to process the time series corresponding to the first position information to extract local features, and the second convolutional layer is used to process the time series corresponding to the first position information to extract global features, and the local features and global features are used to generate feature data.

[0014] In the above solution, a dual-convolutional layer structure is designed. One convolutional layer uses a small kernel to extract local detail features, such as local dynamic behavior characteristics of terminal devices, such as sudden speed changes and angle jumps. The other convolutional layer uses strided convolution (or strided convolution) to capture global features with a larger receptive field, such as the macroscopic motion trends of terminal devices (specifically, the periodic motion trends of terminal devices). Stacking multi-scale convolutional kernels can significantly improve the representation ability of temporal features, addressing the shortcomings of general convolutional neural networks in long-term series dependency and noise sensitivity. It breaks through the bottleneck of convolutional neural networks' limited local receptive field in long-term series modeling, and effectively solves the problems of insufficient capture of long-term series dependency and noise sensitivity. This design achieves fine-grained perception and enhanced representation of position features to improve the accuracy of downstream position prediction.

[0015] In some possible implementations of the first aspect, the convolutional neural network layer also includes a convolutional block attention module, which is located after the first convolutional layer and before the second convolutional layer. The convolutional block attention module includes a channel attention submodule and a spatial attention submodule connected in series. The channel attention submodule is used to assign channel attention weights to channel features of local features based on the channel attention mechanism, and the spatial attention submodule is used to assign spatial attention weights to local features based on the spatial attention mechanism.

[0016] In the above scheme, a convolutional block attention module is inserted between two convolutional layers, and spatial attention weights are assigned to local features through the spatial attention sub-module to improve the model's ability to pay attention to features at key time points; channel attention weights are assigned to channel features of local features through the channel attention sub-module to improve the model's ability to pay attention to key feature dimensions, thereby achieving the suppression of information redundancy and enhanced discriminability.

[0017] In some possible implementations of the first aspect, the convolutional neural network layer also includes a pooling layer, the pooling layer includes a maximum pooling layer, an average pooling layer and a fusion module, the maximum pooling layer is used to perform maximum pooling operations on local features and global features to obtain maximum pooling results, the average pooling layer is used to perform average pooling operations on local features and global features to obtain average pooling results, and the fusion module is used to fuse the maximum pooling results and the average pooling results to obtain feature data.

[0018] In the above scheme, a hybrid pooling mechanism that integrates maximum pooling and average pooling is adopted to capture global statistical information while retaining local extreme value features, thereby effectively avoiding the risk of information loss brought by a single pooling strategy.

[0019] In some possible implementations of the first aspect, the fusion module is specifically used to perform channel-by-channel weighted operations on the splicing operation results and the channel weight vectors to obtain feature data, wherein the splicing operation results are obtained by splicing the maximum pooling results and the average pooling results in the channel dimension, and the channel weights are obtained based on the feature contribution distribution of the maximum pooling results and the average pooling results.

[0020] In the above scheme, the contributions of the two pooling results are automatically smoothed to improve the feature expression ability of the pooling results.

[0021] In some possible implementations of the first aspect, the converter layer includes an encoder layer, which is used to encode the feature data to obtain a feature vector; the encoder layer includes an embedding layer, which is used to convert the feature data into a representation vector; the encoder layer also includes a position encoding layer, which is used to generate a position vector for each data in the feature data based on the feature correlation between the historical position information and the position information at the current moment, the strength of the feature correlation is proportional to the weight of the corresponding position vector, and the representation vector and the position vector are used to determine the feature vector.

[0022] In the above scheme, the position encoding layer performs position encoding based on the feature correlation between the historical position information and the position information at the current moment. Since the feature correlation between the position information at different historical moments and the position information at the current moment is different, the degree of influence on the future position information is different. Therefore, considering the feature correlation between the features at different moments, different position weights are assigned to different time steps, which can highlight the features of key time points and suppress noise interference. In other words, the embodiment of the present application adopts a dynamic context encoding mechanism for position encoding, taking into account the correlation between the features at the current moment and the historical context, so that the position embedding is no longer fixed, but driven by the feature content, automatically highlighting the key time points and suppressing noise interference. Compared with other position encoding technologies (such as static sinusoidal position encoding), it can provide sufficient position perception capabilities in non-uniformly sampled or dynamically changing time series, making it more in line with the needs of non-stationary and non-periodic time series, and improving the adaptability of the model to heterogeneous time features.

[0023] In some possible implementations of the first aspect, the encoder layer includes a multi-head self-attention layer, each attention head of the multi-head self-attention layer is followed by a depthwise separable convolution layer, and the multi-head self-attention layer and the depthwise separable convolution layer are used to process the representation vector to obtain an attention score, which is used to represent the dependency between different representation vectors.

[0024] In the above scheme, a depth-wise separable convolutional layer is set behind each attention head in the multi-head self-attention layer to extract fine-grained local change features in the time dimension and enhance the sensitivity to local mutation features.

[0025] In some possible implementations of the first aspect, the converter layer includes a decoder layer, which is configured to perform parallel decoding processing on the feature vector using a query vector to obtain the second position information, wherein the query vector embeds a display spatiotemporal prior pattern.

[0026] In related technologies, the decoder of the converter typically uses an autoregressive approach to generate predictions step by step, which can lead to cumulative error propagation and suffer from drawbacks such as inability to parallelize computation. In this approach, parallel decoding is achieved using query vectors that explicitly encode the spatiotemporal pattern of future moments. All prediction steps are computed synchronously, allowing the generation of a complete prediction sequence in parallel at once, significantly accelerating inference speed.

[0027] In some possible implementations of the first aspect, the decoder layer includes a step size mapping unit, which is used to match a corresponding window size according to a prediction mode, and data corresponding to the window is used to determine the second position information.

[0028] In the above scheme, a dynamic range attention range control mechanism is used to regulate the size of the window, that is, to regulate the attention range: in the short-term prediction mode, focus on the local window, and in the long-term prediction mode, focus on the global trend. In this way, the perception of temporal locality can be enhanced and the computational complexity can be reduced.

[0029] In some possible implementations of the first aspect, the trajectory prediction model also includes an online learning module, which is used to process the second position information based on a Q-learning mechanism to determine reward values ​​under different states, and adjust model parameters of the convolutional neural network layer and the converter layer according to the reward values.

[0030] In the above scheme, in order to improve the continuous prediction performance of the hybrid model and make the model more suitable for the application scenario of this application, an online learning module is introduced to enable the model to continuously learn and update, thereby improving the trajectory prediction capability for terminal devices in different states.

[0031] In some possible implementations of the first aspect, the online learning module is used to process the second position information based on the Q learning mechanism to determine the Q value, including: the online learning module is specifically used to process the data in the processing window based on the Q learning mechanism to determine the Q value, the data in the processing window is part of the data in the second position information, the processing window includes multiple sub-windows, each sub-window includes multiple data, the weight corresponding to each sub-window is different, and the weight corresponding to each sub-window is determined based on the channel semantic attention mechanism.

[0032] In the above scheme, by assigning different attention weights to subsequences with different time spans, we can cope with the situation where long-term and short-term behaviors are mixed, increase the attention weight on key targets, and thus enhance the directionality of reasoning.

[0033] In a second aspect, a beam management device is provided, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the beam management device performs any one of the methods of the first aspect.

[0034] In a third aspect, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the method described in any aspect of the first aspect are implemented.

[0035] In a fourth aspect, a computer program product is provided. When the computer program product is run on an electronic device, the electronic device executes the method described in any one of the first to third aspects above.

[0036] In a fifth aspect, a chip system is provided, which includes a processor coupled to a memory, and the processor executes a computer program stored in the memory to implement the method described in any one of the first aspects above.

[0037] The chip system may be a single chip or a chip module composed of multiple chips.

[0038] It can be understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of the architecture of a communication system 100 applicable to an embodiment of the present application is shown;

[0040] Figure 2 A schematic diagram of the structure of a terminal device and a network device is shown;

[0041] Figure 3A A schematic diagram of a scanning process of the beam tracking process provided in an embodiment of the present application;

[0042] Figure 3B A schematic diagram of a scanning process of the beam tracking process provided in an embodiment of the present application;

[0043] Figure 3C A schematic diagram of a scanning process of the beam tracking process provided in an embodiment of the present application;

[0044] Figure 3DA schematic diagram of a scanning process of the beam tracking process provided in an embodiment of the present application;

[0045] Figure 3E A schematic diagram of a scanning process of the beam tracking process provided in an embodiment of the present application;

[0046] Figure 4A A possible scenario applicable to the embodiments of the present application is shown;

[0047] Figure 4B Another possible scenario applicable to the embodiments of the present application is shown;

[0048] Figure 5 An exemplary flow chart of a beam tracking method 200 provided in an embodiment of the present application is shown;

[0049] Figure 6 An exemplary flow chart of an environment perception solution provided by an embodiment of the present application is shown;

[0050] Figure 7 A schematic diagram of a beam tracking process provided by an embodiment of the present application is shown;

[0051] Figure 8A A schematic diagram of a scanning process of the beam tracking process provided in an embodiment of the present application;

[0052] Figure 8B A schematic diagram of a scanning process of the beam tracking process provided in an embodiment of the present application;

[0053] Figure 8C A schematic diagram of a scanning process of the beam tracking process provided in an embodiment of the present application;

[0054] Figure 9 A possible structural diagram of a trajectory prediction model provided in an embodiment of the present application is shown;

[0055] Figure 10 A possible structural diagram of a convolutional neural network layer 310 is shown;

[0056] Figure 11 A possible structural diagram of the encoder layer 321 in the converter layer 320 is shown;

[0057] Figure 12 A possible structural diagram of the decoder layer 322 in the converter layer 320 is shown;

[0058] Figure 13 A schematic diagram of a processing flow of the online learning module 330;

[0059] Figure 14is a schematic block diagram of a communication device 10 provided in an embodiment of the present application;

[0060] Figure 15 is a schematic diagram of another communication device 20 provided in an embodiment of the present application;

[0061] Figure 16 FIG. 3 is a schematic diagram of a chip system 30 provided in an embodiment of the present application. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0063] The embodiments of the present application are applicable to beam-based multi-carrier communication systems, such as: global system for mobile communications (GSM) system, code division multiple access (CDMA) system, wideband code division multiple access (WCDMA) system, general packet radio service (GPRS), long term evolution (LTE) system, LTE frequency division duplex (FDD) system, LTE time division duplex (TDD) system, universal mobile telecommunication system (UMTS), world-wide interoperability for microwave access (WiMAX) communication system, fifth generation (5G) system or new radio (NR), sixth generation (6G) system, etc.

[0064] The above-mentioned communication system applicable to the present application is only an example, and the communication system applicable to the present application is not limited to this. It is described uniformly here and will not be repeated below.

[0065] Figure 1 FIG1 shows a schematic diagram of a communication system 100 applicable to an embodiment of the present application. As shown in the figure, the communication system 100 may include at least one network device, such as Figure 1 The network device 110 shown; the communication system 100 may also include at least one terminal device, such as Figure 1The terminal device 120 ( Figure 1 Two terminal devices are exemplarily shown in FIG. ). The network device 110 and the terminal device 120 may communicate via a wireless link.

[0066] Various communication devices, such as Figure 1 The network device 110 or terminal device 120 in the example may be configured with multiple antennas. These multiple antennas may include at least one transmit antenna for sending signals and at least one receive antenna for receiving signals. Furthermore, each communication device also includes a transmitter chain and a receiver chain. Those skilled in the art will appreciate that these may include multiple components related to signal transmission and reception (e.g., processors, modulators, multiplexers, demodulators, demultiplexers, or antennas). Therefore, communication between the network and terminal devices can be achieved using multi-antenna technology.

[0067] It should be understood that the network 110 in the wireless communication system can be any device with wireless transceiver functions, especially a wireless access network device. For example, the network device 110 includes but is not limited to: evolved NodeB (eNB), radio network controller (RNC), node B (NB), base station controller (BSC), base transceiver station (BTS), home base station (for example, Home evolved NodeB, or Home Node B, HNB), baseband unit (BBU), access point (AP) in a wireless fidelity (WIFI) system, wireless relay node, wireless backhaul node, transmission point (TP) or transmission and reception point (TP) The network device 110 may also be a gNB in ​​a 5G system, such as a NR system, or a transmission point (TRP or TP), one or a group of antenna panels (including multiple antenna panels) of a base station in a 5G system, or a network node constituting a gNB or a transmission point, such as a baseband unit (BBU) or a distributed unit (DU). For convenience, the following example uses a gNB as the network device 110.

[0068] In some deployments, a gNB may include a centralized unit (CU) and a DU. The gNB may also include a radio unit (RU). The CU implements some gNB functions, while the DU implements some gNB functions. For example, the CU implements radio resource control (RRC) and packet data convergence protocol (PDCP) layer functions, while the DU implements radio link control (RLC), media access control (MAC), and physical (PHY) layer functions. Because RRC layer information ultimately becomes PHY layer information, or is converted from PHY layer information, in this architecture, higher-layer signaling, such as RRC layer signaling, can also be considered to be sent by the DU, or by both the DU and the CU. It should be understood that a radio access network device can be a CU node, a DU node, or a device that includes both a CU node and a DU node. In addition, the CU can be divided into a radio access network device in the access network (radio access network, RAN), and the CU can also be divided into a radio access network device in the core network (core network, CN), which is not limited in this application.

[0069] It should also be understood that the terminal device 120 in the communication system 100 may also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent, or user device. The terminal device in the embodiments of the present application may be a mobile phone, a tablet computer, a computer with wireless transceiver function, a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal in industrial control, a wireless terminal in self-driving, a wireless terminal in remote medical care, a wireless terminal in smart grid, a wireless terminal in transportation safety, a wireless terminal in smart city, a wireless terminal in smart home, etc. The embodiments of the present application do not limit the application scenarios.

[0070] Figure 1 The communication between the network device 110 and the terminal device 120 in the communication system shown can also be represented in another form, such as Figure 2 As shown, terminal device 120 includes a processor 121, a memory 122, and a transceiver 123. Transceiver 123 includes a transmitter 1231, a receiver 1232, and an antenna 1233. Network device 110 includes a processor 111, a memory 112, and a transceiver 113. Transceiver 113 includes a transmitter 1131, a receiver 1132, and an antenna 1133. Receiver 1132 can be configured to receive transmission control information via antenna 1133, and transmitter 1131 can be configured to send transmission feedback information to network device 110 via antenna 1133. Transmitter 1131 can be configured to send transmission control information to terminal device 120 via antenna 1133, and receiver 1132 can be configured to receive transmission feedback information sent by terminal device 120 via antenna 1133.

[0071] In communication system 100, network device 110 can communicate with terminal device 120 via electromagnetic waves. Lower-frequency electromagnetic waves (e.g., those with frequencies less than or equal to 6 GHz) have lower path loss and longer transmission distances. Higher-frequency electromagnetic waves (e.g., those with frequencies greater than 6 GHz) have higher path loss and shorter transmission distances. Therefore, when communicating with terminal device 120 via lower-frequency electromagnetic waves, network device 110 transmits signals omnidirectionally, allowing the terminal to receive the signals. However, when communicating with terminal device 120 via higher-frequency electromagnetic waves, network device 110 still transmits signals omnidirectionally, resulting in a shorter transmission distance and potential inability for terminal device 120 to receive the signals. In this scenario, beamforming or other techniques can be used to form more directional, narrower beams of electromagnetic waves emitted by network device 110 and terminal device 120 toward each other. This allows for longer propagation distances for electromagnetic waves, enabling communication between the network device and the terminal.

[0072] Beamforming (also known as beamforming) is a signal preprocessing technology based on antenna arrays. Network equipment adjusts the weighting coefficients of each element in the antenna array to generate a directional beam, thereby achieving significant array gain. Beamforming technologies can be specifically categorized as digital beamforming, analog beamforming, and hybrid digital / analog beamforming.

[0073] A beam is a communication resource. A beam can be wide, narrow, or any other type of beam. Different beams can be considered different resources. Different beams can transmit the same or different information. Alternatively, multiple beams with the same or similar communication characteristics can be considered a single beam. A beam can include one or more antenna ports for transmitting data channels, control channels, and sounding signals. It should be understood that the one or more antenna ports that form a beam can also be considered an antenna port set.

[0074] When using low-frequency or medium-frequency bands, signals can be sent omnidirectionally or through a wider angle. When using high-frequency bands, thanks to the smaller carrier wavelength of high-frequency communication systems, antenna arrays consisting of many antenna elements can be arranged at the transmitting and receiving ends. The transmitting end sends signals with certain beamforming weights, so that the transmitted signals form a beam with spatial directionality. At the same time, the receiving end uses an antenna array with certain beamforming weights to receive the signals, which can improve the received signal power at the receiving end and combat path loss.

[0075] Beams include transmit beams (Tx beams) and receive beams (Rx beams). A transmit beam refers to the distribution of signal strength in different directions after a signal is transmitted by an antenna. A receive beam refers to the distribution of wireless signal strength in different directions across the space as a result of the antenna array strengthening or weakening the signal.

[0076] In the current NR protocol, beams can be represented by the quasi colocation (QCL) relationship between antenna ports. Specifically, two signals in the same beam have a QCL relationship with respect to spatial receive parameters. Beams can be represented in the protocol by various signal identifiers, such as the resource index of the channel state information reference signal (CSI-RS), the index of the synchronous signal / physical broadcast channel block (SS / PBCH block or SSB), the resource index of the sounding reference signal (SRS), and the resource index of the tracking reference signal (TRS).

[0077] Beams generally correspond to resources. For example, during beam measurement, network device 110 transmits different resources via different beams. Terminal device 120 then provides feedback on the measured resource quality, allowing the radio access network to determine the quality of the corresponding beam. During data transmission, beam information is also indicated by its corresponding resource. For example, the radio access network uses the Transmission Configuration Indication (TCI) field in the Downlink Control Information (DCI) to indicate to the terminal the beam information for receiving the Physical Downlink Shared Channel (PDSCH).

[0078] In directional communications, an antenna array forms one or more narrow beams directed toward a specific target or direction. When the target moves or the environment changes, causing the communication link quality to degrade, the antenna array can be dynamically adjusted to keep the beam pointed toward the target, maintaining high-quality communication. This technology is called beam management, beam tracking, real-time beam alignment, or adaptive beam steering.

[0079] The following combination Figure 3A to Figure 3E A possible beam tracking process involved in an embodiment of the present application is exemplified.

[0080] Please refer to Figure 3A , the network device 110 performs beam sweeping on the first beam set set B, that is, the network device 110 transmits wireless signals (or wireless resources) to the terminal device 120 through the beams in set B. Since the network device 110 and the terminal device 120 do not know each other's positions, the network device 110 performs beam sweeping in all directions, that is, the beams in set B cover all directions. The terminal device 120 receives the wireless signals transmitted by the network device through the beams in set B, and then performs beam measurement to obtain measurement results #1 of each beam, such as the layer 1 reference signal received power (L1-RSRP) corresponding to each beam, which is used to reflect the signal strength corresponding to each beam. The terminal device 120 reports the measurement information to the network device 110 (Beam reporting).

[0081] Please refer to Figure 3B, the network device 110 determines the beams whose signal strength is within a preset range based on the measurement result #1 fed back by the terminal device 120. The coverage of these beams points to the general direction of the terminal device 120. It can be understood that the closer the direction of a beam is to the direction pointed by the network device 110 to the terminal device 120, the better the signal quality of the corresponding wireless signal received by the terminal device 120. Assume Figure 3B The signal strengths corresponding to the beam in the middle shaded portion and the beam in the middle portion are within a preset range.

[0082] Please refer to Figure 3C , the network device 110 determines the second beam set set A. As an example, the network device 110 determines the second beam set set A according to Figure 3B The beams in the shaded area are determined as set A. For example, the network device 110 Figure 3B Set A is generated based on the direction of the beam in the shaded area. The beams in set A can be narrower than the beams in set B, or they can have the same width as the beams in set B. This is not a limitation in this application. Furthermore, network device 110 performs beam scanning on set A. That is, network device 110 transmits wireless signals (or wireless resources) to terminal device 120 via the beams in set A. Accordingly, terminal device 120 feeds back corresponding measurement result #2 to network device 110.

[0083] Please refer to Figure 3D Network device 110 determines a third beam set, top K, based on measurement result #2 of set A fed back by terminal device 120. For example, network device 110 uses an artificial intelligence (AI) / machine learning (ML) model to predict the top K transmit beams in set A based on the L1-RSRP measurement results, forming the top K beams. Network device 110 scans the top K beams. Terminal device 120 determines a target transmit beam based on measurement result #3 of the top K beams. The target transmit beam may be the beam with the best measurement result among the top K beams.

[0084] Please refer to Figure 3E Terminal device 120 reports the identification information (beam ID) of the target transmit beam to network device 110 and scans different receive beams to find the target receive beam that pairs with the target transmit beam. This allows network device 110 and terminal device 120 to obtain the currently optimal transmit and receive beams, thereby improving communication quality.

[0085] From the above Figure 3A to Figure 3EThe example process demonstrates that beam tracking is a key air interface technology for improving spectrum efficiency and signal transmission quality. However, this process relies on a wide range of directional searches and feedback mechanisms, resulting in high overhead and extended latency. Furthermore, in some highly dynamic or complex scenarios, beam tracking suffers from poor timeliness and is prone to tracking instability.

[0086] Figure 4A A possible scenario applicable to the embodiment of the present application is shown. Figure 4A As shown in the figure, in this scenario, the terminal device is a moving car, so the terminal device moves at a high speed. The above beam tracking process has a long latency. By the time beam tracking is completed, the terminal device's position may have changed significantly. Therefore, it cannot match the wireless channel in real time, which may lead to beam failure or communication interruption.

[0087] Figure 4B Another possible scenario applicable to the embodiment of the present application is shown. Figure 4B As shown in the figure, in this scenario, the terminal device is carried by a moving pedestrian, so the movement trajectory of the terminal device is extremely uncertain, so it is necessary to repeatedly Figure 3A to Figure 3E The above process is given to perform beam tracking. Each beam tracking requires a full range of beam scanning, resulting in a large amount of resource waste.

[0088] It should be noted that, in order to facilitate the description and better illustrate the embodiments of the present application, Figure 4A and Figure 4B As an example, the applicable scenarios of the embodiments of the present application are briefly introduced. However, the applicable scenarios of the embodiments of the present application include but are not limited to Figure 4A as well as Figure 4B .

[0089] In view of the above problems, an embodiment of the present application provides a beam tracking method, which predicts the location information of the terminal device in the future period in advance before performing beam scanning, so as to determine the movement trajectory of the terminal device. In this way, beam scanning can be performed in a targeted manner based on the prediction results without the need for omnidirectional beam scanning, thereby improving the efficiency of beam scanning and reducing overhead.

[0090] Figure 5 FIG2 shows an exemplary flow chart of a beam tracking method 200 provided in an embodiment of the present application. It is understood that the method 200 can be applied to Figure 1 The communication system 100 shown can be specifically executed by the network architecture 110 in the communication system 100, or by some modules or units in the network architecture 110, and this application is not limited thereto. For convenience, the following description uses the network device executing the method 200 as an example. The following exemplary description of each execution step in the method 200 is provided.

[0091] S210: Obtain first location information of the terminal device.

[0092] Exemplarily, the network device obtains the first location information of the terminal device.

[0093] As an example, the location information in the embodiment of the present application includes the location coordinate information of the terminal device. By sensing the location of the terminal device in advance before performing beam tracking, the range of the beam scanning can be narrowed in the subsequent beam tracking process.

[0094] The location coordinate information indicates the physical coordinate position of the terminal device. Optionally, the location information in the embodiment of the present application also includes a terminal device velocity vector and a direction angle, wherein the velocity vector indicates the speed magnitude and direction of the terminal device, and the direction angle indicates the azimuth angle of the terminal device relative to the network device.

[0095] In a possible implementation, the first location information includes the location information of the terminal device at the current moment, such as the location coordinate information, velocity vector, and direction angle at the current moment.

[0096] This application does not limit the specific implementation method of the network device obtaining the current location information of the terminal device. As an example, the network device can obtain the current location information of the terminal device through satellite assistance, or through mobility management technology, or through multi-base station triangulation technology. This application does not limit this. As a specific example, the network device can use Figure 6 The environment perception scheme shown is used to obtain the current location information of the terminal device. The specific implementation process can be found in the subsequent Figure 6 For the sake of brevity, the corresponding specific description is not repeated here.

[0097] In one possible implementation, the first location information also includes the historical location information of the terminal device. For example, the first location information includes the location information of the terminal device at one or more moments in the past second preset time. In a possible exemplary scenario, the network device senses the location information of the terminal device at a preset frequency, such as every 50ms, and records the location information of the terminal device at different moments in a time series. Therefore, the network device can not only obtain the first location information at the current moment, but also obtain the historical location information obtained before that. The network device can obtain the historical location information according to the length of the time series, such as obtaining the historical location information of 15 time series lengths; or, the network device obtains the historical location information according to the divided time lengths, such as obtaining the historical location information of the terminal device in the past 3 minutes.

[0098] S220: Use a preset trajectory prediction model to process the first location information to obtain second location information of the terminal device within a first preset time in the future.

[0099] Exemplarily, after obtaining the first location information, the network device uses a preset trajectory prediction model to process the first location information to obtain second location information. The second location information is the location information of the terminal device predicted by the network device at one or more future moments, that is, the second location information can indicate the movement trajectory of the terminal device within a first preset time in the future.

[0100] This application does not limit the length of the time series corresponding to the second position information. The specific length is determined by the pre-trained prediction trajectory model.

[0101] In one possible implementation, the time series length corresponding to the second location information is greater than or equal to a first threshold and less than or equal to a second threshold. The first threshold is, for example, 10, and the second threshold is, for example, 20. The second location information includes, for example, location information of 16 time series lengths. By controlling the number of predicted second location information to be greater than or equal to the first threshold, it is possible to avoid the situation where the range covered by the first beam set subsequently determined based on the second location information is too small due to the location indicated by the second location information being close to the location indicated by the first location information. Since the second location information is predicted location information, it may differ from the actual situation. During the actual movement process, the terminal device may move outside the coverage range of the first beam set, ultimately resulting in beam tracking failure. The embodiment of the present application increases the number of predicted location information in the second location information to indirectly expand the range covered by the first beam set (for specific solutions related to the first beam set, please refer to the subsequent step S230) to reduce the situation of beam tracking failure. On the other hand, by controlling the amount of second position information obtained by prediction to be less than or equal to the second threshold, it is possible to prevent excessive predicted data from causing serious distortion of the results, thereby causing the subsequently determined first beam set to be inaccurate, thereby affecting the beam tracking process.

[0102] It can be understood that the trajectory prediction model described in the embodiment of the present application is an artificial intelligence model for predicting the movement trajectory of the terminal device, and the present application does not limit the training process and specific structure of the model.

[0103] S230. Determine a first beam set according to the second position information.

[0104] Exemplarily, after acquiring the second location information, the network device determines the first beam set based on the second location information. Optionally, in a specific implementation, the terminal device determines the first beam set based on the current location information of the terminal device in the first location information and the second location information.

[0105] In one possible embodiment, the first beam set includes one or more beams. The first beam set is used to perform beam tracking between the network device and the terminal device. In other words, the first beam set is used to determine the transmit beam and / or receive beam between the network device and the terminal device. The specific process can be found in the description of step S240 and is not further described here.

[0106] In a possible embodiment, the area pointed to by the first beam set covers the location indicated by the second location information. In other words, the movement trajectory of the terminal device indicated by the second location information is within the area pointed to by each beam in the first beam set.

[0107] Figure 8A A possible schematic diagram of the first beam set is shown. Figure 8A As shown in the figure, the first beam set includes multiple beams. The coverage of these beams is indicated by two arrows in the image. The angle between the two arrows is denoted as θ, which is the maximum angle of the first beam set. The movement trajectory of the terminal device indicated by the second position information is shown as the dotted line in the figure. As can be seen from the figure, the movement trajectory of the terminal device is within the coverage of the first beam set. In other words, the maximum angle θ of the first beam set is determined by the position of the terminal device indicated by the second position information.

[0108] S240. Perform beam scanning on the first beam set to determine a transmit beam and / or receive beam with the terminal device.

[0109] Exemplarily, after determining the first beam set, the network device scans the first beam set to perform beam tracking, or in other words, to determine a transmit beam and / or receive beam between the network device and the terminal device.

[0110] This application does not limit the specific process of beam tracking. Figure 7 The process in the following example illustrates a possible implementation method. Figure 7 The method flow in can be regarded as a possible implementation of step S240.

[0111] S241. The network device performs beam scanning on the first beam set.

[0112] For example, the network device may adopt step S230 in method 200 to determine a first beam set, where the area pointed to by the first beam set covers the location of the terminal device indicated by the second location information. Figure 8A Since the first beam set is determined by the predicted terminal device location information, the maximum angle θ of the first beam set is often much smaller than 360°. Figure 3AIn the beam scanning process shown, the number of first beam sets in this embodiment is greatly reduced, which helps to save resources.

[0113] After determining the first beam set, the network device performs beam scanning on the first beam set, that is, the network device transmits a wireless signal (or wireless resource) to the terminal device through the beams in the first beam set.

[0114] S242: The terminal device executes the first measurement process.

[0115] S243. The terminal device sends the first measurement result to the network device.

[0116] Exemplarily, the terminal device receives the wireless signal transmitted by the network device through the beam within the first beam set, and then performs beam measurement to obtain the measurement results of each beam, which are recorded as the first measurement results. The first measurement results, for example, include the L1-RSRP corresponding to each beam, which is used to reflect the signal strength corresponding to each beam.

[0117] After obtaining the first measurement result, the terminal device feeds back the first measurement result to the network device.

[0118] S244. The network device determines a second beam set.

[0119] S245. The network device performs beam scanning on the second beam set.

[0120] Exemplarily, after receiving the first measurement result fed back by the terminal device, the network device determines the second beam set according to the first measurement result. Figure 8B A schematic diagram of the second beam set is shown, wherein the coverage of the second beam set is smaller than the coverage of the first beam set.

[0121] In one example, the second beam set is a subset of the first beam set, for example, the second beam set is a beam in the first beam set whose signal strength meets a preset requirement.

[0122] In another example, the second beam set is a beam set re-determined based on the first measurement result. For example, based on the first measurement result, the network device determines a first direction and a second direction, where the first direction and the second direction are determined by beams in the second beam set whose measurement results are at a critical value (e.g., just equal to or just greater than a preset threshold). The second beam set is then determined between the first direction and the second direction. Optionally, in this example, the beams in the second beam set are narrower than those in the first beam set, which helps improve measurement accuracy. That is, when performing a large-scale scan, using a relatively wide beam can reduce resource consumption; when performing a small-scale scan, using a relatively narrow beam can improve scanning accuracy.

[0123] After determining the second beam set, the network device scans the first beam set, that is, uses the beams in the first beam set to send wireless signals to the terminal device.

[0124] S246: The terminal device executes the second measurement process.

[0125] S247. The terminal device sends the second measurement result to the network device.

[0126] Exemplarily, the terminal device receives the wireless signal transmitted by the network device through the beam within the second beam set, and then performs beam measurement to obtain the measurement results of each beam, which are recorded as the second measurement results. The second measurement results, for example, include the L1-RSRP corresponding to each beam, which is used to reflect the signal strength corresponding to each beam.

[0127] After obtaining the second measurement result, the terminal device feeds back the second measurement result to the network device.

[0128] S248. The network device determines a third beam set.

[0129] S249. The network device performs beam scanning on the third beam set.

[0130] Exemplarily, after receiving the second measurement result fed back by the terminal device, the network device determines a third beam set according to the second measurement result. Figure 8C A schematic diagram of a third beam set is shown, wherein the third beam set is composed of some beams in the second beam set. Specifically, the third beam set is composed of K' beams (denoted as Top K') with the highest measurement results in the second beam set, or K' beams determined based on other methods. This application does not impose any restrictions on this.

[0131] After determining the third beam set, the network device scans the third beam set.

[0132] S250: The terminal device executes the third measurement process.

[0133] S251. The terminal device sends identification information of the target transmission beam to the network device.

[0134] Exemplarily, the terminal device receives a wireless signal transmitted by the network device via a beam within the third beam set, then performs beam measurement to obtain a measurement result for each beam, which is recorded as a third measurement result. Furthermore, the terminal device determines a target transmit beam based on the third measurement result. The target transmit beam may be the beam with the best measurement result in the third beam set.

[0135] The terminal device reports the identification information of the target transmit beam to the network device, and then scans different receive beams to find the target receive beam that matches the target transmit beam. This allows the network device and terminal device to obtain the current optimal transmit and receive beams, thereby improving communication quality.

[0136] In summary, the embodiments of the present application provide a beam management method for beam tracking (or beam scanning). In this method, before beam scanning is performed, the location information of the terminal device in the future is predicted in advance, so as to determine the future location and movement trend of the terminal device. In this way, beam scanning can be performed in a targeted manner based on the prediction results without the need for omnidirectional beam scanning, thereby significantly improving the efficiency of beam alignment, communication stability and resource utilization. When beam scanning is performed in related technologies, it relies on real-time position feedback, and there are problems after adjustment. For example, Figure 3A and Figure 8A By comparison, it can be found that the solution provided in the embodiment of the present application can significantly reduce the size of the first beam set, that is, setB is reduced to setB'; Figure 3C and Figure 8B By comparison, it can be found that the solution provided in the embodiment of the present application can significantly reduce the size of the second beam set, that is, setA is reduced to setA'; Figure 3D and Figure 8C By comparison, it can be found that the solution provided by the embodiment of the present application can significantly reduce the size of the third beam set, that is, topK is reduced to topK'. By reducing the size of the beam set in the beam scanning process, the scanning efficiency can be significantly improved and resource consumption can be reduced.

[0137] Figure 6 An exemplary flow chart of a location awareness solution provided by an embodiment of the present application is shown. The solution can be implemented independently or in combination with method 200, such as a specific implementation of step S210 of method 200. It is understandable that Figure 6 The scheme shown can be applied to Figure 1 The communication system 100 shown in FIG. 1 may be executed by the network architecture 110 in the communication system 100, or by some modules or units in the network architecture 110, which is not limited in this application. Figure 6 The following is an example of the solution shown in the figure. Figure 6 Each step in the following is an exemplary description of an implementation process of the solution.

[0138] S211 , transmitting a linear frequency modulation wave and receiving a corresponding echo signal.

[0139] Exemplarily, a network device performs environmental perception by transmitting wave signals. For example, the network device transmits a linear frequency modulated continuous wave signal, at least a portion of which is received by a terminal device, which then reflects a corresponding echo signal back to the network device. The echo signal can be understood as an electromagnetic wave signal reflected by the terminal device. It is essentially a delayed, frequency-shifted, and modulated copy of the original signal, carrying motion characteristic parameters related to the terminal device. This echo signal is used to determine the current first location information of the terminal device.

[0140] S212: Generate a difference frequency signal.

[0141] For example, after receiving the echo signal reflected by the terminal device, the network device mixes the echo signal with the originally transmitted wave signal to generate a difference frequency signal carrying target distance information, so that the distance between the terminal device and the network device can be subsequently analyzed.

[0142] S213: Generate RDM.

[0143] For example, after the beat frequency signal is generated, the beat frequency signal is subjected to spectral analysis to construct a range-doppler matrix (RDM), where the RDM is a matrix of target range and velocity information, which can provide a data basis for subsequent target detection, classification, and tracking.

[0144] S214a: Determine the distance between the terminal device and the terminal device.

[0145] Exemplarily, a cell-averaging constant false alarm rate (CA-CFAR) algorithm is used to generate a detection threshold to distinguish valid information from noise interference in the RDM, that is, to filter out valid information and determine the target distance between the network device and the terminal device.

[0146] S214b. Determine the direction corresponding to the terminal device.

[0147] Exemplarily, the direction corresponding to the terminal device is found by analyzing the characteristics of the echo signal. For example, the network device analyzes the echo signal using a multiple signal classification (MUSIC) algorithm to calculate the angle (including azimuth and / or elevation) of the terminal device.

[0148] S215: Determine the first location information.

[0149] Exemplarily, by fusing the target distance, angle and three-dimensional coordinates of the network device, the first position information of the terminal device is calculated, including parameters of three aspects: position coordinates, velocity vector and direction angle.

[0150] Based on the above solution, the network device can obtain the precise location of the terminal device. Specifically, the network device in the above embodiment has environmental perception capabilities. For example, the network device integrates synaesthesia technology. This not only solves the limitations of discrete systems in cost, efficiency, and scenario adaptability in related technologies, but also provides a technical foundation for the evolution of networks towards intelligence and scenario-based services.

[0151] Will Figure 6 The embodiment shown is applied to method 200, which enables the network device in method 200 to accurately locate the position of the terminal device, thereby narrowing the scanning range in advance in the subsequent beam scanning process, reducing the number of unnecessary beam scans, speeding up the beam tracking process, saving resources, and improving communication efficiency.

[0152] As can be seen from the above scheme, the trajectory prediction model in the embodiment of the present application is used to predict the user's location information, that is, the trajectory prediction model is used to process the first location information to obtain the second location information, wherein the first location information includes the current location information and / or historical location information of the terminal device, and the second location information includes the second location information of the terminal device within a first preset time in the future. The first location information and the second location information are usually time series (that is, a sequence composed of location information corresponding to different time points). Therefore, the location information in the embodiment of the present application can also be referred to as trajectory information in some application scenarios. This application does not limit the specific structure of the trajectory prediction model. A possible example is given below.

[0153] Figure 9 FIG. 1 shows a possible structural diagram of a trajectory prediction model provided by an embodiment of the present application. Figure 9 As shown, the trajectory prediction model includes a convolutional neural network (CNN) layer 310 , a transformer 320 , and optionally, an online learning module 330 .

[0154] A convolutional neural network is a network structure used to process grid topology data. The basic structure of a convolutional neural network generally consists of two layers: a feature extraction layer, where the input of each neuron is connected to the local receptive field of the previous layer, extracting local features. Once a local feature is extracted, its positional relationship with other features is determined. The second layer is a feature mapping layer. Each computational layer of the network consists of multiple feature maps, each representing a plane where all neurons have equal weights. The feature mapping structure can use a sigmoid function as the activation function of the convolutional network, making the feature map shift-invariant. Furthermore, since neurons on a mapping plane share weights, the number of free parameters in the network is reduced. Each convolutional layer in the convolutional neural network is followed by a computational layer for local averaging and secondary extraction. This unique two-step feature extraction structure reduces feature resolution. In one possible implementation of the present embodiment, convolutional neural network layer 310 is used to extract features from the first position information to obtain feature data, i.e., the convolutional neural network layer is used for feature extraction. The converter layer 320 is used to process the feature data to obtain the second position information. Specifically, the converter layer 320 is used to perform position prediction to determine the second position information. In one example, the converter 320 includes an encoder layer 321 and a decoder layer 322. The online learning module 330 is used to perform online learning based on the output of the converter layer 320 (i.e., the second position information) to adjust the model parameters of the convolutional neural network layer 310 and the converter layer 320. The following is an exemplary description of the structure and function of each layer.

[0155] Figure 10 A possible structural diagram of the convolutional neural network layer 310 is shown. Figure 10As shown, the convolutional neural network layer 310 includes a first convolutional layer 311 and a second convolutional layer 313. It should be understood that the convolutional layer is the core component of the convolutional neural network. Its function is to extract multi-level abstract features from the input data through local perception and weight sharing mechanisms. In the embodiment of the present application, the convolution kernel size of the first convolutional layer 311 is smaller than a preset threshold, that is, the first convolutional layer 311 uses a small kernel, such as a (3×1) kernel; the second convolutional layer uses a strided convolution structure, such as a (5×1) kernel. In one design, the second convolutional layer 313 is located after the first convolutional layer 311. The first convolutional layer 311 processes the time series corresponding to the first location information to extract local features. When the first location information includes current location information and historical location information, the first location information can be considered a time series, i.e., a data sequence consisting of time and location information. For example, the first location information includes three feature data (position coordinates, velocity vector, and azimuth angle, a total of 9 dimensions) with a time series length of 128. The second convolutional layer 313 has a larger receptive field and processes the time series corresponding to the first location information to extract global features. These local and global features are used to generate feature data, or in other words, they form the basis for generating feature data.

[0156] The above one-dimensional convolution process can be expressed as:

[0157] ;

[0158] in, For output channels At time step The response, Indicates that the convolution kernel is in the channel Upper and lower The weight of the position, is the input time series, is the convolution kernel length, is the activation function, is the channel bias.

[0159] In the above scheme, a dual convolutional layer structure is designed, in which one convolutional layer uses a small kernel to extract local detail features, such as local dynamic behavior characteristics of terminal devices such as speed changes and angle jumps; the other convolutional layer uses strided convolution (or strided convolution) to use a larger receptive field to capture global features, such as the macroscopic motion trend of terminal devices (specifically, the periodic motion trend of terminal devices).

[0160] In a possible implementation provided in the present application, the convolutional neural network layer 310 also includes a convolutional block attention module (CBAM) 312, and the convolutional block attention module 312 is located after the first convolutional layer 311 and before the second convolutional layer 313. In one example, the convolutional block attention module includes a channel attention submodule and a spatial attention submodule (not shown in the figure) connected in series. The channel attention submodule is used to assign channel attention weights to the channel features of local features based on the channel attention mechanism, and the spatial attention submodule is used to assign spatial attention weights to local features based on the spatial attention mechanism. It can be understood that the present application does not limit the attention fusion method of the channel attention submodule and the spatial attention submodule. In one example, element-by-element multiplication can be used for attention fusion, and the present application does not impose any restrictions.

[0161] In the above scheme, a convolution block attention module 312 is inserted between two convolutional layers, and spatial attention weights are assigned to local features through the spatial attention sub-module to improve the model's ability to pay attention to features at key time points; channel attention weights are assigned to channel features of local features through the channel attention sub-module to improve the model's ability to pay attention to key feature dimensions, thereby achieving suppression of information redundancy and enhanced discriminability.

[0162] In one possible implementation provided by the present application, the convolutional neural network layer 310 further includes a pooling layer 314. In one design, the hybrid pooling layer 314 is disposed after the second convolutional layer 313. The pooling layer 314 includes a maximum pooling layer 314a, an average pooling layer 314b, and a fusion module 314c.

[0163] The maximum pooling layer 314a is used to perform a maximum pooling operation on the local features and the global features to obtain a maximum pooling result, and the average pooling layer 314b is used to perform an average pooling operation on the local features and the global features to obtain an average pooling result. As an example, the process of hybrid pooling and average pooling can be expressed as:

[0164] ;

[0165] in represents the features after convolution, represents the maximum pooling operation, represents the average pooling operation, is the maximum pooling result, This is the average pooling output result.

[0166] The fusion module 314c is used to fuse the maximum pooling result and the average pooling result to obtain feature data. In a specific example, the fusion module is specifically used to obtain a splicing operation result by splicing the maximum pooling result and the average pooling result in the channel dimension, and then perform a channel-by-channel weighted operation on the splicing operation result and the channel weight vector to obtain feature data, wherein the channel weight is obtained based on the feature contribution distribution of the maximum pooling result and the average pooling result. As an example, the process of splicing the operation results of two pooling layers in the channel dimension can be expressed as:

[0167] ;

[0168] in Represents the result of splicing the two pooling layer channels, Represents feature splicing.

[0169] The two pooling outputs are spliced ​​in the channel dimension, and the channel attention mechanism is used to adaptively balance the feature contribution of the two pooling strategies. The process is shown as follows:

[0170] ;

[0171] in, represents the channel weight vector, represents channel-by-channel weighting, is a small feedforward neural network that models the relationship between channels. Represents global average pooling of the input tensor.

[0172] In the above scheme, a hybrid pooling mechanism that integrates maximum pooling and average pooling is adopted to capture global statistical information while retaining local extreme value features, thereby effectively avoiding the risk of information loss brought by a single pooling strategy.

[0173] In summary, the convolutional neural network layer 310 of the trajectory prediction model provided in the embodiment of the present application uses a convolutional neural network with two convolutional layers and employs multi-scale convolution kernel stacking, which significantly improves the ability to represent temporal features. This addresses the shortcomings of general convolutional neural networks in long-term series dependency and noise sensitivity, and overcomes the bottleneck of convolutional neural networks in modeling long sequences, where local receptive fields are limited. This effectively addresses the issues of insufficient capture of long-term series dependency and noise sensitivity. This design achieves fine-grained perception and enhanced representation of position features to improve the accuracy of downstream position predictions.

[0174] In addition, the above-mentioned convolutional neural network layer also adopts a hybrid pooling (maximum pooling plus average pooling) mechanism, which can balance local extreme values ​​and global trends, avoid information loss caused by single pooling, and make predictions more accurate.

[0175] Figure 11 FIG. 3 shows a possible structural diagram of the encoder layer 321 in the converter layer 320. Figure 11 As shown, the encoder layer is used to encode the feature data to obtain a feature vector. The encoder layer 321 includes an embedding layer 3211 and a position encoding layer. The embedding layer 3211 is used to convert the feature data into a representation vector. The position encoding layer 3212 is used to generate a position vector for each data point in the feature data based on the feature correlation between the historical position information and the current position information. The strength of the feature correlation is proportional to the weight of the corresponding position vector. The representation vector and the position vector are used to determine the feature vector.

[0176] In the above scheme, the position encoding layer 3212 performs position encoding based on the feature correlation between the historical position information and the position information at the current moment. Since the feature correlation between the position information at different historical moments and the position information at the current moment is different, the degree of influence on the future position information is different. Therefore, considering the feature correlation between the features at different moments, different position weights are assigned to different time steps, which can highlight the features of key time points and suppress noise interference. In other words, the embodiment of the present application adopts a dynamic context encoding mechanism for position encoding, taking into account the correlation between the features at the current moment and the historical context, so that the position embedding is no longer fixed, but driven by the feature content, automatically highlighting the key time points and suppressing noise interference. Compared with other position encoding technologies (such as static sinusoidal position encoding), it can provide sufficient position perception capabilities in non-uniformly sampled or dynamically changing time series, making it more in line with the needs of non-stationary and non-periodic time series, and improving the adaptability of the model to heterogeneous time features.

[0177] In one example, a gating mechanism is used to dynamically adjust the position weights at each time step. A gating mechanism is a neural network design paradigm that controls the flow of information through dynamically learned weight coefficients. Its core concept is to automatically determine the contribution ratio of different information sources through trainable parameters, enabling feature selection, path switching, or task adaptation. An example of the weight determination process is as follows:

[0178] ;

[0179] in, Represents the current input features, represents the historical context aggregation representation, 、 represents the gating weight parameter, Represents the dynamic gating vector, which represents the position importance weight of this time step.

[0180] In one possible implementation, the encoder layer 321 includes a multi-head self-attention layer 3213, and each attention head of the multi-head self-attention layer 3213 is followed by a depth-wise separable convolution layer 3214. The multi-head self-attention layer 3213 and the depth-wise separable convolution layer 3214 are used to process the representation vector to obtain an attention score, which is used to represent the dependency between different representation vectors.

[0181] In the above scheme, each attention head in the multi-head self-attention layer 3213 is followed by a depthwise separable convolutional layer (DSC) 3214, for example, of size 3×1, to extract fine-grained local variation features in the temporal dimension. A gating mechanism can also dynamically balance the global attention output of the multi-head self-attention layer 3213 with the convolutional-aware features of the DSC 3214, forming a composite attention representation with contextual awareness. This design enhances sensitivity to local mutation features while maintaining the ability to model global dependencies.

[0182] Figure 12 FIG. 3 is a schematic diagram showing a possible structure of the decoder layer 322 in the converter layer 320. The decoder layer 322 in the embodiment of the present application is used to perform parallel decoding processing on the feature vector to obtain the second position information. Figure 12 , the decoder layer 322 includes an encoding unit 3221 and a step mapping unit 3222. The encoding unit 3221 is used to perform parallel decoding processing on the feature vector through the query vector. The query vector in the embodiment of the present application embeds a display of the spatiotemporal prior pattern. It is understandable that in the related art, the decoder of the converter generally uses an autoregressive method to gradually generate predictions, which will lead to cumulative error propagation and defects such as inability to parallelize calculations. The embodiment of the present application uses a query vector to achieve parallel decoding. The query vector explicitly encodes the spatiotemporal pattern of the future moment, and all prediction steps are calculated synchronously, so it supports the one-time parallel generation of a complete prediction sequence, which can significantly improve the inference speed. An example of the query vector initialization process is as follows:

[0183] ;

[0184] Among them, Q represents the initialized query vector, each are trainable parameters, which include spatiotemporal priors.

[0185] Combine Figure 12, the decoder layer 322 also includes a step length mapping unit 3222, which is used to match the corresponding window size according to the prediction mode, and the data corresponding to the window is used to determine the second position information. The prediction mode in the embodiment of the present application is used to indicate the prediction duration, for example, including a short-term prediction mode and a long-term prediction mode. The solution provided by the embodiment of the present application can dynamically adjust the attention range, that is, the window size, according to the prediction mode. The window size determines the amount of data used in the current prediction step. For example, in the short-term prediction mode, it focuses on high-resolution perception of the local window range to capture fine-grained short-term dynamic features. The window function example is:

[0186] ;

[0187] in is the desired window function, is a learnable dynamic window size function, is the time interval function, which is the target time step With historical time steps distance.

[0188] For example, in long-term prediction mode, the attention mechanism automatically expands to a wider historical context, increasing the window size to model global behavioral trends and cyclical structures. The adjustment of the attention range is dynamically guided by the time interval function.

[0189] In the above scheme, a dynamic range attention range control mechanism is used to regulate the size of the window, that is, to regulate the attention range: in the short-term prediction mode, focus on the local window, and in the long-term prediction mode, focus on the global trend. In this way, the perception of temporal locality can be enhanced and the computational complexity can be reduced.

[0190] Optionally, the trajectory prediction model provided in an embodiment of the present application further includes an online learning module 330, which has an adaptive online learning mechanism for inversely adjusting the model parameters of the convolutional neural network layer 310 and the converter layer 320 based on the output of the converter layer 320 (i.e., the second position information). In one example, the online learning module 330 employs an adaptive online learning mechanism based on TS-TrajGen, or other mechanisms may be employed, which are not limited in this application.

[0191] Figure 13 Schematic diagram of a processing flow of the online learning module 330. Figure 13First, the time series data corresponding to the second position information is processed in a window. The processing window is a sliding window that limits the data range involved in updating the model parameters. For example, according to the volatility of the input time series data, the length of the processing window is automatically expanded or compressed. Specifically, when the time series data fluctuates greatly, the length of the processing window is appropriately increased, while when the time series data fluctuates slightly, the length of the processing window is appropriately reduced. After each window slide, the model parameters are fine-tuned only based on the data in the new window, avoiding the full training data and improving training efficiency and real-time performance.

[0192] In a specific example, the processing window can be defined as follows:

[0193] ;

[0194] in To process the window length, is the input time series (ie, the time series corresponding to the second position information).

[0195] In a specific example, according to data volatility An example of the process of dynamically adjusting the window length is as follows:

[0196] ;

[0197] in is a learnable function, is the minimum window length, is the maximum window length.

[0198] Furthermore, an adaptive weight allocator adaptively assigns weights to different data segments within the processing window, dynamically adjusting the model's focus on different prediction targets (such as displacement accuracy and velocity estimation), enhancing inference accuracy. For example, the processing window can be divided into multiple subwindows, each containing multiple data points, or corresponding to a different time subsequence. A channel-wise semantic attention mechanism is used to assign corresponding weights to each subwindow. A specific implementation is illustrated below.

[0199] Designing temporal granularity attention using triple factor attention mechanism:

[0200] ;

[0201] in Indicates the The attention weight of the time scale, Indicates the The learnable weight vector corresponding to the time scale, Indicates the Subsequences at different time scales, fusion output It can be expressed as:

[0202] ;

[0203] Through the channel semantic attention mechanism, the contribution of each dimension of input features is automatically measured and attention weights are assigned based on the contribution. For example, each input channel The corresponding attention weight is:

[0204] ;

[0205] in Indicates the The attention weight of each channel, represents the learnable weight vector, Indicates the input feature All data of each channel.

[0206] In the above scheme, by assigning different attention weights to subsequences with different time spans, we can cope with the situation where long-term and short-term behaviors are mixed, increase the attention weight on key targets, and thus enhance the directionality of reasoning.

[0207] Furthermore, the model parameters of the convolutional neural network layer 310 and / or the converter layer 320 are determined through a Q-learning mechanism. For example, the second position information is processed based on the Q-learning mechanism to determine reward values ​​under different states, and the model parameters of the convolutional neural network layer and the converter layer are adjusted based on the reward values. A possible implementation process is exemplified below.

[0208] Q-learning is a model-free reinforcement learning algorithm whose core goal is to learn an optimal policy through interaction with the environment, so that the agent maximizes the cumulative reward by selecting actions in the state space. In one implementation, the state space, action space, and reward function are designed based on the Q-learning mechanism. The state space defines all possible states of the agent's perception of the environment and is a basic component of reinforcement learning. In the application scenario of the embodiment of this application, trajectory prediction error, historical scheduling feedback, data complexity (coefficient of variation, curvature change), etc. are used as state vectors to construct the state space; the action space defines the set of operations that the agent can perform in a specific state; the reward function is the "compass" of the agent's learning, guiding strategy optimization by quantifying the quality of actions. In the application scenario of the embodiment of this application, the reward function is constructed by considering three-dimensional indicators: prediction accuracy, computational overhead, and response delay.

[0209] By continuously interacting with the environment, it learns the optimal scheduling strategy, achieves a balance between accuracy and efficiency, maximizes adaptive adjustment capabilities, and optimizes the update path. An example of the Q-learning update mechanism is as follows:

[0210] ;

[0211] in for Momentary state-action value, is the learning rate, For the current reward, is the discount factor, is the objective function value.

[0212] In one possible implementation, the value table is updated according to the Q-learning update rules. It can be understood that updating the Q-value table refers to adjusting the Q-value of each state-action pair according to the Q-learning algorithm's update rules to reflect changes in the expected cumulative reward under the current strategy. Specifically, after each action is executed, the Q-value of the current state-action pair is updated based on the reward value and the maximum Q-value of the next state. This process is directly related to dynamically optimizing the inter-stimulus interval, because after the Q-value table is updated, the Q-learning algorithm will select the action with the highest Q-value, thereby gradually approaching the set target.

[0213] Corresponding to the methods provided in the above-mentioned method embodiments, the embodiments of the present application also provide corresponding devices, which include modules for executing the corresponding methods in the above-mentioned method embodiments. The modules can be software, hardware, or a combination of software and hardware. It is understood that the technical features described in the above-mentioned method embodiments are also applicable to the following device embodiments. Therefore, for the content not described in detail, please refer to the above method embodiments. For the sake of brevity, they will not be repeated here.

[0214] Figure 14 1 is a schematic block diagram of a communication device 10 provided in an embodiment of the present application. The device 10 includes a transceiver module 11 and a processing module 12. The transceiver module 11 can implement corresponding communication functions, and the processing module 12 is used to process data. In other words, the transceiver module 11 is used to perform operations related to receiving and sending, while the processing module 12 is used to perform operations other than receiving and sending. The transceiver module 11 can also be referred to as a communication interface or a communication unit.

[0215] Optionally, the device 10 may also include a storage module 13, which can be used to store instructions and / or data. The processing module 12 can read the instructions and / or data in the storage module so that the device can implement the actions of the equipment or network element in the aforementioned method embodiments.

[0216] In a first design, the apparatus 10 may correspond to the network device in the above method embodiment, or a component (such as a chip) of the network device.

[0217] The device 10 can implement the steps or processes executed by the network device in the above method embodiment, wherein the transceiver module 11 can be used to perform the transceiver-related operations of the network device in the above method embodiment, and the processing module 12 can be used to perform the processing-related operations of the network device in the above method embodiment.

[0218] In one possible implementation, the transceiver module 11 is used to obtain first location information of the terminal device, where the first location information includes the current location information and / or historical location information of the terminal device; the processing module 12 is used to process the first location information using a preset trajectory prediction model to obtain second location information of the terminal device within a first preset time in the future; the processing module 12 is also used to determine a first beam set based on the second location information, and to perform beam scanning on the first beam set to determine a transmitting beam and / or receiving beam between the terminal device and the terminal device.

[0219] It should be understood that the specific process of each module executing the above corresponding steps has been described in detail in the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0220] It should also be understood that the device 10 herein is embodied in the form of a functional module. The term "module" herein may refer to an application-specific integrated circuit (ASIC), an electronic circuit, a processor (e.g., a shared processor, a dedicated processor, or a group processor, etc.) and memory for executing one or more software or firmware programs, combined logic circuits, and / or other suitable components that support the described functionality. In an optional example, those skilled in the art will appreciate that the device 10 may specifically be the terminal device described in the above-described embodiments, and may be used to execute the various processes and / or steps corresponding to the terminal device in the above-described method embodiments. To avoid repetition, these are not further described here.

[0221] The apparatus 10 of each of the above-described solutions has the function of implementing the corresponding steps performed by the terminal device in the above-described method. This function can be implemented via hardware, or by hardware executing corresponding software implementations. This hardware or software includes one or more modules corresponding to the above-described functions; for example, the transceiver module can be replaced by a transceiver (e.g., the transmitting unit in the transceiver module can be replaced by a transmitter, and the receiving unit in the transceiver module can be replaced by a receiver). Other units, such as the processing module, can be replaced by a processor to respectively perform the transmitting and receiving operations and related processing operations in each method embodiment.

[0222] In addition, the transceiver module 11 may also be a transceiver circuit (for example, may include a receiving circuit and a sending circuit), and the processing module may be a processing circuit.

[0223] Figure 15 FIG2 is a schematic diagram of another communication device 20 provided in an embodiment of the present application. The device 20 includes a processor 21, which is configured to execute computer programs or instructions stored in a memory 22, or read data / signaling stored in the memory 22, to perform the methods described in the above method embodiments. Optionally, there are one or more processors 21.

[0224] Alternatively, as Figure 7 As shown, the device 20 further includes a memory 22, which is used to store computer programs or instructions and / or data. The memory 22 can be integrated with the processor 21, or can be separately provided. Optionally, there are one or more memories 22.

[0225] Alternatively, as Figure 15 As shown, the device 20 further includes a transceiver 23, which is used to receive and / or send signals. For example, the processor 21 is used to control the transceiver 23 to receive and / or send signals.

[0226] As a solution, the apparatus 20 is used to implement the operations performed by the network device in each of the above method embodiments.

[0227] It should be understood that the processor mentioned in the embodiments of the present application may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0228] It should also be understood that the memory mentioned in the embodiments of the present application can be volatile memory and / or non-volatile memory. Among them, the non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM). For example, RAM can be used as an external cache. By way of example and not limitation, RAM includes the following forms: static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronized link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0229] It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, the memory (storage module) can be integrated into the processor.

[0230] It should also be noted that the memory described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0231] Figure 16 FIG2 is a schematic diagram of a chip system 30 provided in an embodiment of the present application. The chip system 30 (or also referred to as a processing system) includes a logic circuit 31 and an input / output interface 32 .

[0232] The logic circuit 31 may be a processing circuit in the chip system 30. The logic circuit 31 may be coupled to a storage unit and call instructions in the storage unit so that the chip system 30 can implement the methods and functions of the various embodiments of the present application. The input / output interface 32 may be an input / output circuit in the chip system 30, outputting information processed by the chip system 30 or inputting data or signaling information to be processed into the chip system 30 for processing.

[0233] As another solution, the chip system 30 is used to implement the operations performed by the terminal device in the above various method embodiments.

[0234] For example, the logic circuit 31 is used to implement the processing-related operations performed by the network device in the above method embodiment, such as: Figures 5 to 7 The network device in the embodiment shown performs processing-related operations; the input / output interface 32 is used to implement the sending and / or receiving-related operations performed by the network device in the above method embodiment, such as Figures 5 to 7 The network devices in the illustrated embodiment perform operations related to transmission and / or reception.

[0235] An embodiment of the present application further provides a computer-readable storage medium on which computer instructions for implementing the methods executed by the network device in the above-mentioned method embodiments are stored.

[0236] For example, when the computer program is executed by a computer, the computer can implement the method performed by the network device in each embodiment of the above method.

[0237] For another example, when the computer program is executed by a computer, the computer can implement the methods performed by the network device in each embodiment of the above method.

[0238] An embodiment of the present application further provides a computer program product comprising instructions, which, when executed by a computer, implement the methods performed by the network device in the above-mentioned method embodiments.

[0239] An embodiment of the present application also provides a communication system, including the aforementioned network device.

[0240] The explanation of the relevant contents and beneficial effects of any of the above-mentioned devices can be referred to the corresponding method embodiments provided above, which will not be repeated here.

[0241] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0242] In the above embodiments, all or part of the embodiments may be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product comprises one or more computer instructions. When loaded and executed on a computer, the computer program instructions fully or partially perform the processes or functions described in the embodiments of this application. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. For example, the computer may be a personal computer, a server, or a network device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium accessible by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, or a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state disk (SSD)). For example, the available medium includes, but is not limited to, various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0243] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0244] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0245] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0246] It should be understood that the “embodiment” mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, the various embodiments in the entire specification do not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in the various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0247] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

[0248] In addition, it should be noted that the various numerical references used in this application (such as the terms "first," "second," "third," "fourth," and other terminology (if any) in the specification, claims, and accompanying drawings) are provided for descriptive purposes only and are not intended to limit the scope of this application. The sequence numbers of the various processes do not imply a specific order of execution; the execution order of the processes should be determined by their functionality and inherent logic.

[0249] The terms "including" and "having" and any variations thereof mean "including but not limited to," unless specifically stated otherwise. For example, a process, method, system, product, or apparatus that includes a series of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, product, or apparatus.

[0250] In the embodiments of this application, words such as "exemplarily" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplarily" or "for example" in the embodiments of this application should not be construed as preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplarily" or "for example" is intended to present the relevant concepts in a concrete manner.

[0251] In the various embodiments of this application, unless otherwise specified or logically conflicting, the terms and / or descriptions between different embodiments are consistent and can be referenced from each other. The technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships. The specific operating methods in the method embodiments of this application can also be applied to the device embodiments or system embodiments.

Claims

1. A beam management method, characterized in that: The method comprises: Acquire first location information of a terminal device, where the first location information includes current location information and / or historical location information of the terminal device; Processing the first location information using a preset trajectory prediction model to obtain second location information of the terminal device within a first preset time in the future; determining a first beam set according to the second position information; Performing beam scanning on the first beam set to determine a transmit beam and / or a receive beam with the terminal device; The trajectory prediction model includes a convolutional neural network layer and a converter layer, the convolutional neural network layer is used to extract features from the first position information to obtain feature data, and the converter layer is used to process the feature data to obtain the second position information; The convolutional neural network layer includes a first convolutional layer, a convolutional block attention module and a second convolutional layer. The convolutional block attention module is located after the first convolutional layer and before the second convolutional layer. The convolution kernel size of the first convolutional layer is smaller than a preset threshold. The second convolutional layer adopts a stride convolution structure. The first convolutional layer is used to process the time series corresponding to the first position information to extract local features. The second convolutional layer is used to process the time series corresponding to the first position information to extract global features. The local features and the global features are used to generate the feature data. The convolutional block attention module includes a channel attention submodule and a spatial attention submodule connected in series. The channel attention submodule is used to assign channel attention weights to the channel features of the local features based on the channel attention mechanism. The spatial attention submodule is used to assign spatial attention weights to the local features based on the spatial attention mechanism.

2. The method according to claim 1, characterized in that The area pointed to by the first beam set covers the position indicated by the second position information.

3. The method according to claim 1 or 2, characterized in that The obtaining of the first location information of the terminal device includes: sending a linear frequency modulation wave to the terminal device and receiving a corresponding echo signal from the terminal device; The first position information is determined according to the echo signal.

4. The method according to claim 1, wherein The convolutional neural network layer also includes a pooling layer, which includes a maximum pooling layer, an average pooling layer and a fusion module. The maximum pooling layer is used to perform a maximum pooling operation on the local features and the global features to obtain a maximum pooling result. The average pooling layer is used to perform an average pooling operation on the local features and the global features to obtain an average pooling result. The fusion module is used to fuse the maximum pooling result and the average pooling result to obtain the feature data.

5. The method according to claim 4, characterized in that The fusion module is specifically used to perform a channel-by-channel weighted operation on the splicing operation result and the channel weight vector to obtain the feature data, wherein the splicing operation result is obtained by splicing the maximum pooling result and the average pooling result in the channel dimension, and the channel weight is obtained based on the feature contribution distribution of the maximum pooling result and the average pooling result.

6. The method according to claim 1, characterized in that The converter layer includes an encoder layer, and the encoder layer is used to encode the feature data to obtain a feature vector; The encoder layer includes an embedding layer, the embedding layer being configured to convert the feature data into a representation vector; The encoder layer also includes a position encoding layer, which is used to generate a position vector for each data in the feature data based on the feature correlation between the historical position information and the position information at the current moment. The strength of the feature correlation is proportional to the weight of the corresponding position vector. The representation vector and the position vector are used to determine the feature vector.

7. The method according to claim 6, characterized in that The encoder layer includes a multi-head self-attention layer, each attention head of the multi-head self-attention layer is followed by a depth-wise separable convolution layer, the multi-head self-attention layer and the depth-wise separable convolution layer are used to process the representation vector to obtain an attention score, and the attention score is used to represent the dependency relationship between different representation vectors.

8. The method according to claim 6, characterized in that The converter layer includes a decoder layer, and the decoder layer is used to perform parallel decoding processing on the feature vector using a query vector to obtain the second position information, and the query vector has an embedded display spatiotemporal prior pattern.

9. The method according to claim 8, characterized in that The decoder layer includes a step length mapping unit, which is used to match a corresponding window size according to a prediction mode, and data corresponding to the window is used to determine the second position information.

10. The method according to claim 1, characterized in that The trajectory prediction model also includes an online learning module, which is used to process the second position information based on a Q-learning mechanism to determine reward values ​​under different states, and adjust model parameters of the convolutional neural network layer and the converter layer according to the reward values.

11. The method according to claim 10, characterized in that The online learning module is configured to process the second position information based on a Q-learning mechanism to determine reward values ​​under different states, including: The online learning module is specifically used to process the data in the processing window based on the Q learning mechanism to determine the reward value, the data in the processing window is part of the data in the second position information, the processing window includes multiple sub-windows, each of the sub-windows includes multiple data, and the weight corresponding to each sub-window is different. The weight corresponding to each sub-window is determined based on the channel semantic attention mechanism.

12. A beam management device, characterized in that: The apparatus comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the apparatus implements the method according to any one of claims 1 to 11.

13. A chip system, characterized in that: The chip system is applied to a beam management device, and the chip system includes one or more processors, and the one or more processors are used to call computer instructions to enable the beam management device to execute the method according to any one of claims 1-11.

14. A computer program product, characterized in that The invention comprises a computer program, which, when being executed, enables the method according to any one of claims 1 to 11 to be performed.

Images