Beam determination method, non-volatile storage medium and computer device

Abstract

This invention discloses a beamforming method, a non-volatile storage medium, and a computer device. The method includes: acquiring multiple beam selection data fed back by a target device; inputting the multiple beam selection data into a neural network model, whereby the neural network model outputs the search values ​​of multiple predetermined beams, wherein the neural network model includes multiple sub-networks corresponding one-to-one with the multiple beam selection data, and each sub-network determines a set of overall initial beam values ​​and a set of offset initial beam values ​​based on the corresponding input beam selection data; determining multiple candidate beams from the multiple beams based on their respective search values; measuring the multiple candidate beams to obtain measurement results, and selecting the target beam for communication with the target device from among the multiple candidate beams. This invention solves the technical problem of low accuracy in beam prediction due to insufficient utilization of available information.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and more specifically, to a beam determination method, a non-volatile storage medium, and a computer device. Background Technology

[0002] The optimal communication beam between a base station and a user is related to many factors, such as the relative positions of the base station and the user, the location of obstructions in the channel environment, and the positions of scatterers and reflectors. Related technologies typically use only traditional channel-related parameters for beam prediction, such as channel state information and Doppler shift, which suffers from low accuracy in beam prediction.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This invention provides a beam determination method, a non-volatile storage medium, and a computer device to at least solve the technical problem that beam prediction accuracy is low when using only traditional channel-related parameters for beam prediction.

[0005] According to one aspect of the present invention, a beam determination method is provided, comprising: acquiring multiple beam selection data fed back by a target device for beam selection, wherein the multiple beam selection data includes communication state data for describing a communication channel and perception state data for describing the position and motion state of the target device, wherein the target device includes a terminal and a base station; inputting the multiple beam selection data into a neural network model, wherein the neural network model outputs search values ​​for each of the predetermined multiple beams, wherein the neural network model includes multiple sub-networks corresponding one-to-one with the multiple beam selection data, and the multiple sub-networks respectively determine a set of overall initial beam values ​​and a set of [unclear] based on the corresponding input beam selection data. The initial value of the offset beam is determined by multiple sets of overall beam initial values ​​and multiple sets of offset beam initial values ​​output by multiple sub-networks. The overall beam initial value is the initial search value for each beam, while the offset beam initial value characterizes the influence of the corresponding beam selection data on the search values ​​of the offset beams. Based on the search values ​​of each beam, multiple candidate beams are determined. Measurements are performed on these candidate beams to obtain measurement results, which characterize the communication quality of each candidate beam. Based on the measurement results, a target beam for communication with the target device is selected from the candidate beams.

[0006] Optionally, multiple beam selection data are input into a neural network model, and the neural network model outputs the search values ​​of each of the predetermined multiple beams. This includes: inputting the multiple beam selection data into multiple sub-networks one-to-one, and having each sub-network output a set of overall initial beam values ​​and a set of offset initial beam values; inputting the multiple beam selection data into a weight selection network included in the neural network model, and having the weight selection network output the overall weights and offset weights of each of the multiple sub-networks; and determining the search values ​​of each of the multiple beams based on their initial search values ​​and the overall weights and offset weights of each of the multiple sub-networks.

[0007] Optionally, the method further includes: when the measurement results of multiple candidate beams meet the first predetermined conditions, combining multiple beam selection data, the search values ​​of each of the multiple beams and the measurement results of the multiple candidate beams into training samples, and adding them to the online training sample set.

[0008] Optionally, the method further includes: if the number of samples included in the online training sample set meets a second predetermined condition, using the online training sample set to train the neural network model online to obtain an updated neural network model, and determining the beam used for communication based on the updated neural network model.

[0009] Optionally, the neural network model is trained online using an online training sample set to obtain an updated neural network model. This includes: using the online training sample set, with the objective of minimizing the difference between the search values ​​of the multiple beams output by the neural network model and the target search value, training multiple sub-networks and a weight selection network included in the neural network model to obtain trained multiple sub-networks and a trained weight selection network, wherein the target search value is determined based on the measurement results of multiple candidate beams; and constructing an updated neural network model based on the trained multiple sub-networks and the trained weight selection network.

[0010] Optionally, the first predetermined condition is that the beam with the highest search value among multiple beams is not the same beam as the beam with the best communication quality characterized by the measurement results.

[0011] Optionally, the method further includes: marking subnetworks in a predetermined time period whose sum of overall weights and offset weights is less than a predetermined threshold as disabled, thereby obtaining an updated neural network model.

[0012] According to another aspect of the present invention, a non-volatile storage medium is also provided, the non-volatile storage medium including a stored program, wherein, when the program is running, the device where the non-volatile storage medium is located is controlled to execute any of the above-described beam determination methods.

[0013] According to another aspect of the present invention, a computer device is also provided, the computer device including a processor, the processor being configured to run a program, wherein the program executes any of the above-described beam determination methods during runtime.

[0014] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements any of the above-described beam determination methods.

[0015] In this embodiment of the invention, multiple beam selection data for beam selection are obtained from the target device. These multiple beam selection data include communication status data describing the communication channel and perception status data describing the position and motion state of the target device. The target device includes a terminal and a base station. The multiple beam selection data are input into a neural network model, which outputs the search values ​​of multiple predetermined beams. The neural network model includes multiple sub-networks corresponding one-to-one with the multiple beam selection data. Each sub-network determines a set of overall beam initial values ​​and a set of offset beam initial values ​​based on the corresponding input beam selection data. The search values ​​of each beam are jointly determined by the multiple sets of overall beam initial values ​​and multiple sets of offset beam initial values ​​output by the multiple sub-networks. The overall beam initial value is the initial search value of each beam, and the offset beam initial value is the initial search value of each beam. The initial value of the shifted beam is used to characterize the influence of the corresponding beam selection data on the search value of multiple offset beams. Based on the search value of each beam, multiple candidate beams are determined from among the multiple beams. The multiple candidate beams are measured to obtain the measurement results of the multiple candidate beams, which are used to characterize the communication quality of the multiple candidate beams. Based on the measurement results, the target beam for communication with the target device is selected from among the multiple candidate beams. This achieves the goal of utilizing communication state data used to describe the channel and sensing state data used to describe the position and motion state of the target device. By using multiple types of data for beam prediction, the neural network model can more accurately select the optimal beam in the current environment, improving the accuracy of beam prediction. This solves the technical problem of low accuracy in beam prediction due to insufficient utilization of available information. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0017] Figure 1 A hardware structure block diagram of a computer terminal for implementing a beamforming method is shown.

[0018] Figure 2 This is a schematic flowchart of the beam determination method provided according to an embodiment of the present invention;

[0019] Figure 3 This is a schematic diagram of the structure of a neural network model provided according to an optional embodiment of the present invention;

[0020] Figure 4 This is a schematic diagram of a parallel thread provided according to an optional embodiment of the present invention;

[0021] Figure 5 This is a schematic diagram illustrating the effects provided by an optional embodiment of the present invention;

[0022] Figure 6 This is a structural block diagram of a beamforming device provided according to an optional embodiment of the present invention. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] According to an embodiment of the present invention, a method embodiment for beam determination is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0026] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1A hardware block diagram of a computer terminal for implementing a beamforming method is shown. Figure 1 As shown, the computer terminal 10 may include one or more processors (shown as 102a, 102b, ..., 102n in the figure) (the processor may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0027] It should be noted that the aforementioned one or more processors and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be implemented wholly or partially as software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be wholly or partially integrated into any other element in the computer terminal 10. As involved in the embodiments of this application, the data processing circuits serve as processor control (e.g., selection of a variable resistor termination path connected to an interface).

[0028] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the beam-finding method in this embodiment of the invention. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the beam-tracking method of the application described above. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0029] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10.

[0030] Figure 2 This is a schematic flowchart of the beam determination method provided according to an embodiment of the present invention, as shown below. Figure 2 As shown, the method includes the following steps:

[0031] Step S202: Obtain multiple beam selection data fed back by the target device for beam selection. The multiple beam selection data includes communication status data describing the communication channel and perception status data describing the position and motion status of the target device. The target device includes a terminal and a base station.

[0032] In this step, the executing entity can be a mobile communication system, especially a system employing Multiple Input Multiple Output (MIMO) technology. The multiple beam selection data refers to all data about the target device that the communication system can acquire for beam selection. Specifically, this may include communication status data related to communication signal quality, such as reference signal-to-noise ratio, data channel block error rate, channel state information, constellation diagram error vector magnitude, etc., as well as sensing status data regarding the target device's position and motion, such as the terminal's coordinates in three-dimensional space, the terminal's azimuth angle, and velocity. The target device can be a terminal or a base station.

[0033] Step S204: Input multiple beam selection data into the neural network model, and output the search values ​​of each of the predetermined multiple beams by the neural network model. The neural network model includes multiple sub-networks that correspond one-to-one with the multiple beam selection data. Each of the multiple sub-networks determines a set of overall beam initial values ​​and a set of offset beam initial values ​​based on the corresponding input beam selection data. The search values ​​of each of the multiple beams are jointly determined by the multiple sets of overall beam initial values ​​and multiple sets of offset beam initial values ​​output by the multiple sub-networks. The overall beam initial value is the initial search value of each of the multiple beams, and the offset beam initial value is used to characterize the degree of influence of the corresponding beam selection data on the search values ​​of the multiple offset beams.

[0034] In this step, a neural network model can be used to predict the search value of multiple beams based on various beam selection data. The search value can be a quantitative indicator used to evaluate and select beams. By using a neural network model to predict and compare the search values ​​of different beams, the beam with the highest search value can be effectively selected for communication.

[0035] To fully utilize various beam selection data, multiple sub-networks can be set up within the neural network model. Each sub-network is responsible for beam prediction based on one type of beam selection data, and the output includes two sets: one is the overall initial beam value, which predicts multiple predetermined beams based on the corresponding input data, determining the initial search value of each beam; the other is the initial offset beam value, which predicts multiple offset beams around the currently used beam based on the corresponding input data, determining the initial search value of each offset beam. It should be noted that the device's position and speed, among other motion states, can significantly affect the values ​​of the multiple offset beams around the current beam, causing more dramatic changes. The initial offset beam value can reflect these changes, thereby improving the accuracy of beam prediction. Among them, multiple offset beams refer to beams obtained by moving multiple steps with the current beam as the center. The specific range of the step size can be determined based on experience. For example, the farthest position that the device can move within one operation cycle and the beam used at that position can be determined based on experience. Then, the range of the step size of the multiple beams can be determined based on the step size between the beam used by the device at the farthest position and the current beam.

[0036] The neural network model can combine a set of overall beam initial values ​​and a set of offset beam initial values ​​output by multiple sub-networks, i.e., multiple sets of overall beam initial values ​​and multiple sets of offset beam initial values, to jointly determine the search values ​​of the multiple beams in the final output.

[0037] Step S206: Based on the search value of each of the multiple beams, determine multiple candidate beams among the multiple beams.

[0038] In this step, based on the search values ​​of multiple beams output by the neural network model, the beams with the highest values ​​can be selected as candidate beams. The candidate beams are the beams that are most likely to provide the best communication performance as predicted by the neural network model.

[0039] Step S208: Measure multiple candidate beams to obtain measurement results for multiple candidate beams, wherein the measurement results for multiple candidate beams are used to characterize the communication quality of multiple candidate beams.

[0040] In this step, actual measurements can be performed on multiple candidate beams to obtain the measurement results for each candidate beam. The measurement results can include indicators such as signal-to-noise ratio, power, and bit error rate, which are used to measure the communication quality of the beam.

[0041] The available information was not fully utilized for beam prediction, resulting in low accuracy of beam prediction. Step S210: Based on the measurement results, the target beam for communication with the target device is selected from multiple candidate beams.

[0042] In this step, after obtaining the actual measurement results, the optimal beam from multiple candidate beams can be selected as the target beam for communication with the target device. It should be noted that this step does not directly select the beam with the highest search value as the target beam for communication. Instead, considering that the predictions of the neural network model reflect the statistical probability distribution of the target beam's position, which involves uncertainty, multiple candidate beams are selected, and then the beam with better actual performance is chosen for communication from among these candidate beams.

[0043] Through the above steps, the goal of utilizing communication state data to describe the channel and sensing state data to describe the position and motion state of the target device can be achieved. By using multiple types of data for beam prediction, the neural network model can more accurately select the optimal beam in the current environment, thereby improving the accuracy of beam prediction. This solves the technical problem of low beam prediction accuracy caused by not fully utilizing the available information for beam prediction.

[0044] As an optional embodiment, multiple beam selection data are input into a neural network model, and the neural network model outputs the search values ​​of each of the predetermined multiple beams. This includes: inputting the multiple beam selection data into multiple sub-networks one-to-one, and having each sub-network output a set of overall beam initial values ​​and a set of offset beam initial values; inputting the multiple beam selection data into a weight selection network included in the neural network model, and having the weight selection network output the overall weights and offset weights of each of the multiple sub-networks; and determining the search values ​​of each of the multiple beams based on their initial search values ​​and the overall weights and offset weights of each of the multiple sub-networks.

[0045] Optionally, the neural network model includes a beam selection network, specifically comprising multiple sub-networks, which can output the search values ​​of multiple beams individually. These sub-networks can make predictions based on different types of beam selection data; that is, each sub-network is responsible for a specific type of state data, and each sub-network can predict the overall initial beam value and the initial offset beam value based solely on its own input state data.

[0046] In addition to the beam selection network, the neural network model also includes a weight selection network, which outputs the overall weights and offset weights for each of the multiple sub-networks. The overall weights measure the initial value of the overall beam, while the offset weights measure the initial value of the offset beam. Since each sub-network outputs a set of overall beam initial values ​​and a set of offset beam initial values, the weight selection network determines the weights of the values ​​output by each sub-network when combining these search values. That is, based on the overall weights and offset weights of each sub-network, the initial search values ​​of the multiple beams output by each sub-network can be combined to obtain the final search value of each beam.

[0047] Optionally, each sub-network can be a dual-output structure, meaning each sub-network has two output ports: one outputting the overall initial beam value (the search value for each beam) and the other outputting the initial value of the offset beam (the search value for each offset beam). It's important to note that because each sub-network processes a specific type of beam selection data, and different types of information sources have different relationships to the optimal beam selection problem—for example, the terminal's 3D coordinates are directly related to the beam angle but not to the beam offset per unit time, while the terminal's rotation speed is unrelated to the beam angle but strongly correlated with the beam offset per unit time—it is crucial that each sub-network employs a dual-output structure. For each type of state information, a network with a dual-output structure is needed to provide the search value for each beam and the search value for each offset beam, thereby fully utilizing all the information contained in the state information. Furthermore, given that the subnetwork has a dual-output structure, the weight selection network can assign a corresponding weight value to each output port of each subnetwork, that is, assign an overall weight and an offset weight to each subnetwork. The weight of each subnetwork can be the average of the weights of the two output ports.

[0048] The optional embodiments of the present invention do not limit the specific structure of the subnetwork, and can be common network structures such as fully connected networks, partially connected networks, convolutional networks, recurrent networks, and attention mechanisms.

[0049] It should be noted that the offset beam is the index of the communication beam in the current time slot n [b x,n b y,n Based on this, the beam is moved several steps in the x-axis or y-axis direction, so the output of the second output port is (2R+1). 2 The search value of each offset beam, specifically the beam [b x,n +r x b y,n +r y ], r x and r yAll are integers in the range from -R to R, where R refers to the maximum movement step size, x represents the x-axis of the beam array arrangement, y represents the y-axis of the beam array arrangement, and the index of each beam is represented as [b x,n b y,n ].

[0050] Since the offset beam does not correspond one-to-one with all beams, it is necessary to output (2R+1). 2 The search value of each offset beam is mapped to the search value of all beams, and then combined with the search values ​​of all beams output from the first output port. Specifically, the mapping method is divided into two cases, for those included in the above (2R+1) 2 For beams within a given beam range, the search value is the corresponding output value. For beams not included in the above (2R+1)... 2 The search value of a beam within a beam is set to 0.

[0051] As an optional embodiment, the method further includes: marking subnetworks in a predetermined time period whose sum of overall weights and offset weights is less than a predetermined threshold as disabled, thereby obtaining an updated neural network model.

[0052] Optionally, the overall weights and offset weights assigned to each subnetwork by the weight selection network over a relatively long period can be summed or averaged. For subnetworks whose summed or average weights are lower than a preset threshold, the state information corresponding to that subnetwork is considered to have weak relevance to the beam tracking problem. This state information is then deleted, and the corresponding beam selection subnetwork is shut down. During this process, the input and output dimensions of the weight selection network remain unchanged, all inputs to the deleted state information are set to zero, and the weights of the subnetwork corresponding to the deleted state information are no longer used. Automatically identifying and deleting state information with weak relevance to the beam tracking problem can effectively filter out key information from a vast amount of information sources, thereby reducing the overhead of state information acquisition, simplifying the neural network model, and reducing the system's computational resource consumption.

[0053] The advantage of this optional embodiment is that it can adaptively filter state information, excluding state information that is weakly related to the optimal beam selection problem, thereby reducing the overhead of state information acquisition, simplifying the neural network model, and reducing the system's computational resource consumption.

[0054] Furthermore, due to the complexity of changes in the external environment, when collecting communication and perception state information, it is not always possible to search for all state information as expected, and some state information may be missing. The subnetwork corresponding to the missing state information does not output the corresponding search value, and in the input of the weight selection network, the state information is set to zero, and the weights corresponding to the output subnetwork are not used.

[0055] As an optional embodiment, the method further includes: when the measurement results of multiple candidate beams meet a first predetermined condition, combining multiple beam selection data, the search values ​​of each of the multiple beams, and the measurement results of the multiple candidate beams into training samples, and adding them to an online training sample set.

[0056] Optionally, after measuring the beam and obtaining the actual measurement results, if the measurement results meet the first predetermined conditions, the actual measurement results, along with the input data (i.e., multiple beam selection data) and output data (i.e., the search values ​​of each of the multiple beams) of the neural network model, can be recorded and added to the online training sample set as training samples. In this case, when training the neural network model using the online training sample set, the neural network model can adjust its parameters according to the gap between the actual measurement results and its own output search value, so that the output of the neural network model matches the actual situation as closely as possible. This achieves the goal of adjusting the parameters of the neural network model in a timely manner according to the beam change patterns during the application process, thereby enhancing the adaptability of the beam determination method and enabling the neural network model to quickly adapt to the current environment and more accurately select the optimal beam under the current environment.

[0057] In addition, when determining online training samples, since each sub-network outputs two sets of initial search values, two sets of training samples can be generated for adjustment. One set of training samples is labeled with the search values ​​that multiple beams should have, determined based on beam measurement results. The other set of training samples is labeled with the search values ​​that multiple beams should have, determined based on the search values ​​that multiple beams should have, and the beams currently used for communication, determined with the search values ​​that multiple offset beams should have.

[0058] It should be noted that the neural network model in this step is trained online in practical applications. This means that during application, the parameters of the neural network model are further adjusted based on the data from the application process to adapt to the complex and ever-changing application environment. Specifically, during application, various beam selection data obtainable from the communication system are input into the neural network model. The network outputs the search value of all beams and measures the measurement results of all beams, selecting the communication beam with the optimal measurement results. During training, the network parameters are continuously updated based on real-time feedback data from the application process, aiming to narrow the gap between the network's output search value and the actual measurement results. The updated parameters are then fed back to the neural network model during application, enabling the model to provide more reasonable search values. Through continuous interaction and iteration between these two processes, the neural network can gradually adapt to the current communication environment, providing increasingly accurate beam search values.

[0059] As an optional embodiment, the first predetermined condition is that the beam with the highest search value among multiple beams is not the same beam as the beam with the best communication quality characterized by the measurement results.

[0060] Optionally, if the beam with the highest search value among multiple beams is the same beam as the beam with the best communication quality represented by the measurement results, it indicates that the prediction result of the neural network model is relatively accurate, and in this case, the data does not need to be input into the training samples. Conversely, if the beam with the highest search value among multiple beams is not the same beam as the beam with the best communication quality represented by the measurement results, it indicates that the prediction result of the neural network model is not accurate enough, and the parameters of the neural network model need to be adjusted with the current data. Therefore, under the condition of meeting the first predetermined condition, the corresponding data can be used as training samples and added to the online training sample set.

[0061] As an optional embodiment, the method further includes: if the number of samples included in the online training sample set meets a second predetermined condition, using the online training sample set to train the neural network model online to obtain an updated neural network model, and determining the beam used for communication based on the updated neural network model.

[0062] Optionally, since each training iteration requires computational resources to adjust the model, a second condition can be pre-set to limit the frequency of training the neural network model. The neural network model is only trained online using the online training sample set when the number of samples in the online training sample set reaches a certain threshold, resulting in an updated neural network model. The samples in the online training sample set are acquired and continuously updated in real-time during the application process. Furthermore, starting online training only when a certain number of samples is reached ensures that the sample set is large enough to represent the true distribution of the data, thus making online training effective. During online training, the parameters of the neural network model are adjusted according to the new sample set to better reflect the current data characteristics. After the model is updated, this new model is immediately used to determine the beam used for communication; this updated model is more suitable for the current external environment than the original model.

[0063] As an optional embodiment, an online training sample set is used to train the neural network model online to obtain an updated neural network model. This includes: using the online training sample set, with the objective of minimizing the difference between the search values ​​of the multiple beams output by the neural network model and the target search value, training multiple sub-networks and a weight selection network included in the neural network model to obtain trained multiple sub-networks and a trained weight selection network, wherein the target search value is determined based on the measurement results of multiple candidate beams; and constructing an updated neural network model based on the trained multiple sub-networks and the trained weight selection network.

[0064] Optionally, when training the neural network model using an online training sample set, the "search value of each of the multiple beams" in the training sample set is the predicted data output by the neural network model based on the previously learned content, while the "target search value" is the value that should actually be assigned to the beam after actual measurement of multiple candidate beams. Therefore, the neural network model can adjust its parameters to minimize the gap between the search value of each of the multiple beams and the target search value, so that the output of the neural network model matches the actual situation as closely as possible.

[0065] During training, the neural network model can be trained based on the measurement results of multiple candidate beams. To facilitate model operation, the measurement results can be processed and transformed into the benefits obtained by using the corresponding beam. Specifically, measurement results positively correlated with benefits, such as signal-to-noise ratio, can be defined, with the benefit t′(i) = t(i) / max(t). Conversely, for measurement results negatively correlated with benefits, such as bit error rate, the benefit t′(i) = min(t) / t(i) can be defined, where t(i) represents the i-th element in the measurement result vector t, and max(t) and min(t) represent the largest and smallest elements in the measurement result vector t, respectively. If the measurement result t(i) is 0, making the formula t′(i) = min(t) / t(i) impossible to calculate, the benefit of that beam can be directly set to 1. The benefit of unmeasured beams is 0. After defining the benefits, they can be used in place of the measurement results in the model training process.

[0066] First, multiple sub-networks can be trained using an online training sample set, and the search value x output by the neural network model can be used as the basis for the training. m Construct the target search value of the m-th sample in the online training sample set β is a predetermined constant satisfying 0 ≤ β ≤ 1, t′ m Let m be the profit for the m-th sample. After obtaining the target search value for all M samples, the minimum mean square error loss function is then applied. Training is performed on all subnetworks and the weight selection network. It's important to note that during training, if the effective subnetwork set for the m-th sample does not contain a certain subnetwork (i.e., the state information corresponding to that network was not obtained during sample acquisition, or although the state information was obtained during sample acquisition, the network was disabled during training because its state information has weak relevance to the beam tracking problem), then the loss value for that sample is... It will not be used to train this subnetwork.

[0067] After training, samples that were stored in the online training sample pool for an earlier period are also deleted. For example, an expiration time threshold is set, and samples that have been stored in the sample pool for a longer period than the threshold are deleted. This ensures that the neural network is trained with newer samples each time, thereby ensuring that the neural network can learn and adapt to the latest changes in the communication environment in real time.

[0068] As an optional embodiment, the neural network model is trained online using an online training sample set to obtain an updated neural network model, including: setting up parallelizable training threads and application threads; performing beam prediction using the neural network model based on the application thread; and training the neural network model online using an online training sample set based on the training thread to obtain an updated neural network model.

[0069] Optionally, during the application process, parallel training threads and application threads can be started to perform beam prediction (i.e., the application process) and model training, respectively. In the application thread, the neural network model is used to perform the beam prediction task, specifically using the method of the embodiments of the present invention to determine the search value of multiple beams based on the collected beam selection data.

[0070] A training thread can be a thread specifically designed for online training. This refers to training the model using new data samples after the model has been deployed and begun processing data. Through online training, the parameters of the neural network model are adjusted based on the new sample data. This update can be incremental, meaning that only small adjustments are made to the model each time to maintain its stability.

[0071] The advantage of this alternative embodiment lies in its ability to enable continuous learning and adaptation of the model, which is particularly important for applications that need to handle constantly changing data and environments. Through parallel processing, training and application can be performed simultaneously, thereby improving the system's responsiveness and efficiency. Furthermore, this approach can reduce performance degradation caused by outdated models, as the model can learn from new data in real time.

[0072] After training to obtain the latest neural network model, the training thread can feed the new network parameters back to the application thread. The application thread then uses the new network to provide beam search values. Through multiple interactive iterations between the application thread and the training thread, the training thread, by repeatedly learning from the measurement results fed back by the application thread, can gradually adjust the neural network parameters to provide beam search values ​​that match the actual measurement results. Meanwhile, the application thread, by utilizing the new neural network parameters fed back by the training thread, can gradually and accurately predict the beam measurement results in advance, thereby gradually reducing the number of beams in the candidate beam set, reducing the overhead of beam determination, allocating more time-frequency resources to data transmission, and improving the overall system throughput.

[0073] As a specific embodiment, the method provided by the present invention may include the following steps.

[0074] The system determines all available communication and sensing state information (multiple beam selection data). Any communication system contains at least two types of state information: the current beam index value and the current beam index difference sequence. The current beam index value refers to the sequence number of the current beam in all available beam sets. The current beam index difference sequence is the difference between the current beam index and the beam index at the previous moment, the difference between the beam index at the previous moment and the beam index at the moment before that, and so on, to obtain a sequence of index differences at several consecutive moments.

[0075] Assume there are P types of communication and sensing state information {u1 u2 … u P The above information is combined and represented as a beam tracking state vector s = [u1…u2…u3…u4…u5…u6…u7…u8…u9 ... P ].

[0076] For ease of description, an example is given here. For instance, a base station antenna array has 128 beams, each beam with an average coverage angle of 7.5°. There are 16 beam angles along the x-axis and 8 beam angles along the y-axis. Therefore, all 128 beams can cover a 120° range horizontally and a 60° range vertically. The index of each beam is represented as... 0≤b x <16, 0≤b y <8. Similarly, the user's beam index is represented as Assuming the beam tracking period is T, during the time intervals nT to (n+1)T, the base station and the user respectively use the beam. and During communication, at time (n+1)T, since the user's location may have changed, the base station needs to update the beam again. The current beam index value at this time is represented as... The current beam index differential sequence is represented as follows:

[0077]

[0078] It should be further explained that the types of communication and sensing status information can change dynamically. For example, a communication terminal can accurately sense three-dimensional positioning information in an open outdoor environment. However, when encountering strong interference sources or entering an obstructed environment, the quality of the sensing signal will deteriorate significantly, making it impossible to obtain positioning information. In this case, the number of status information types will decrease by 1. When the sensing signal quality recovers, the number of status information types will return to its original value.

[0079] like Figure 3As shown, a beam tracking neural network (neural network model) is constructed based on the types of state information that the system can acquire. Due to the different usage environments of base stations and users—for example, base station antennas are fixed, while user terminals may be handheld and their posture is constantly changing in altitude; or base station antenna arrays may be larger with more elements, while user terminal arrays may be smaller with fewer elements—independent beam tracking neural networks should be constructed for beam tracking of the base station and user, respectively. Each beam tracking neural network contains P subnetworks f1, ..., f... P And a weighted network w. Where the p-th subnetwork f p The input is the p-th state information u p The first output head (output port) outputs the search value x for all beams. 2p-1 The search value of all beams of the second output head (output port) x 2p Regarding the mapping of a subnetwork from its input to its output, this invention does not limit its specific structure; it can be a fully connected network, a partially connected network, a convolutional network, a recurrent network, an attention mechanism, or other common network structures. For the mapping of a subnetwork from its input to its second output, assuming there are N layers, the mapping from the input layer to the (N-1)th layer and from the (N-1)th layer to the output layer is described as follows.

[0080] Because the second output head provides the index [b] of the current communication beam. x,n b y,n Based on this, the search value of the beam after moving several steps in the x-axis or y-axis direction is considered. Therefore, the output of the (N-1)th layer of the second output head is (2R+1). 2 The search value of each beam, specifically the beam [b x,n +r x b y,n +r y ], r x and r y All are integers in the range from -R to R. From the (N-1)th layer to the output layer, the above (2R+1)... 2 The search value of a single beam is mapped to the search value of all beams. All beams can be divided into two cases: those included in the above (2R+1). 2 For beams within a given beam range, the search value is the output value corresponding to the (N-1)th layer. For beams not included in the above (2R+1)... 2 The search value of a beam within a beam is set to 0.

[0081] Obtain the search value x1, ..., x of all subnetworks 2P Then, a weighted average of all search values ​​is calculated to obtain the final search value x. The weight of each output head of all subnetworks is represented as [α1α2…α]. 2P ], The weights are given through a weight network w (weight selection network), and the input to the weight network w is the beam tracking state vector s = [u1…u2]. P The output is the weights [α1α2…α]. 2P For the weighted network, this invention does not limit its specific structure. Since the weights are based on the beam tracking state vector s = [u1…u2]... P Therefore, the weights of the subnetworks change dynamically for different beam tracking state vectors. When the user is in different communication environments, each subnetwork has different weights, which makes the algorithm very flexible and adaptable to different scenarios.

[0082] When a data source temporarily fails, such as when the perception state information cannot be obtained as mentioned earlier, the corresponding beam selection subnetwork will have no input and will no longer calculate the output value. The weighted averaging stage will also no longer consider the output of this network. Therefore, in this case, the final search value can also be expressed as... in This represents the set of all valid subnetworks.

[0083] The average weight of each subnetwork is obtained by averaging the weights of the two output heads of all subnetworks k time steps backward from the current time. For subnetworks with average weights less than a preset threshold, the state information corresponding to that subnetwork is considered to have weak relevance to the beam tracking problem. Therefore, the state information is deleted, and the subnetwork is shut down. During this process, the input and output dimensions of the weight network remain unchanged. In the input of the weight network, the input nodes corresponding to the deleted state information are all set to zero, and the weights of the subnetworks corresponding to the deleted state information are no longer used or trained.

[0084] In the specific embodiments provided by this invention, two threads can run in parallel: a beam tracking thread (application thread) and a beam training thread (training thread), such as... Figure 4As shown in the diagram. In the beam tracking thread, relevant state information obtainable from the communication system is input into the beam tracking network. The network outputs the search value of all beams, selects beams with search values ​​greater than a threshold to form a candidate beam set, measures each beam in the candidate beam set, selects the communication beam with the best measurement result, and feeds the measurement result and relevant state information input to the neural network back to the beam training thread. In the beam training thread, based on the information fed back from the beam tracking thread, the network parameters are continuously updated with the goal of narrowing the gap between the network's output search value and the actual measurement results. The updated parameters are then fed back to the beam tracking thread, enabling the beam tracking thread to provide more reasonable search values. Through continuous interaction and iteration between the two threads, the neural network can gradually adapt to the current communication environment and provide increasingly accurate beam search values.

[0085] Specifically, in the beam tracking thread, the aforementioned beam tracking network and beam tracking state vector s = [u1…u] are used. P After obtaining the search value x of all beams, sort the search values ​​of all beams in descending order, and select those with a search value greater than the threshold x. th The beams form a candidate beam set. For candidate beam sets The beams in the circuit are measured one by one to obtain the measurement result t. The communication quality of the beam can be measured by indicators such as signal-to-interference-plus-noise ratio, power, bit error rate, and spectral efficiency, but this invention does not limit these indicators. Based on the measurement results, the beam with the best communication quality is selected as the communication beam between the base station and the user. Threshold x th The beam tracking state vector can be determined by multiplying the average value of all beam search values ​​by a constant factor, or by multiplying the maximum value of all beam search values ​​by a constant factor. After determining the communication beam, the beam tracking state vector is s = [u1…u... ......u......u......u......u......u......u......u......u......u......u......u......u......u......u......u......u......u......u......u......u......u.........u.........u.........u.........u............u...............u.................. P Candidate beam set Effective subnetwork set The measurement results t of all beams are fed back to the beam training thread, where t(i) represents the i-th element of vector t, i.e., the candidate beam set. Measurement results for the i-th beam. Effective subnetwork set. This refers to the subnetworks corresponding to all valid state information obtained when collecting the sample, and the subnetworks are not shut down because their corresponding state information has a weak correlation with the beam tracking problem.

[0086] In the beam training thread, the information fed back from the beam tracking thread is first saved as a sample. This invention uses reinforcement learning to train the neural network. The advantage of reinforcement learning is that it does not require offline pre-training of the neural network; it can be tracked and trained synchronously online, and it has good adaptability to dynamically changing communication environments. The main principle of reinforcement learning in training the neural network is to associate the search value of each beam output by the neural network with the "reward" that each beam can obtain. In this invention, the reward is represented by the measurement result t. For example, the reward of a beam with a high signal-to-noise ratio should be greater than the reward of a beam with a low signal-to-noise ratio, or the reward of a beam with a low bit error rate should be greater than the reward of a beam with a high bit error rate. When the beam training thread receives a new sample, it needs to determine whether the beam with the maximum search value in the sample (i.e., the first beam in the candidate beam set) is the same as the beam with the best measurement result. If they are the same, it means that for the beam tracking state vector s of this sample, the neural network can give a beam search value consistent with the actual measurement result, so the sample is no longer saved; if they are different, it means that for the beam tracking state vector s of this sample, the neural network cannot correctly give a beam search value consistent with the actual measurement result, so the sample is saved to the sample pool.

[0087] When the number of samples in the sample pool exceeds a certain threshold, several samples are randomly selected from the pool to train the beam tracking neural network. During training, the gain for each beam in the sample needs to be calculated. For measurements positively correlated with gain, such as signal-to-noise ratio, the gain t′(i) = t(i) / max(t) is defined. Conversely, for measurements negatively correlated with gain, such as bit error rate, the gain t′(i) = min(t) / t(i) is defined, where t(i) represents the i-th element of vector t, and max(t) and min(t) represent the largest and smallest elements in vector t, respectively. If the measurement result t(i) is 0, making the formula t′(i) = min(t) / t(i) impossible to calculate, the gain for that beam can be directly set to 1. Unmeasured beams have a gain of 0.

[0088] Suppose that M samples are drawn from the sample pool. During training, for the m-th sample, first, based on the sample's measurement result t... m Obtain the gain t′ for each beam corresponding to this sample. m Then sample state s m The value x of the neural network output is obtained by inputting the input into the neural network. m And construct the target search value of the m-th sample. β is a constant satisfying 0 ≤ β ≤ 1. After obtaining the target search value of all M samples, the minimum mean square error loss function is used. Train all subnetworks and the weight selection network sequentially. Note that during training, if the effective subnetwork set for the m-th sample is... If a sample does not contain a certain subnetwork, then the loss value is... It will not be used to train this subnetwork.

[0089] After the latest neural network is trained, the new network parameters are fed back to the beam tracking thread. The beam tracking thread then uses the new network to provide beam search values. Through multiple interactive iterations between the beam tracking thread and the beam training thread, the beam training thread can gradually adjust the neural network parameters by repeatedly learning from the measurement results fed back by the beam tracking thread, providing beam search values ​​that match the actual measurement results. Meanwhile, the beam tracking thread, by utilizing the new neural network parameters fed back by the beam training thread, can gradually and accurately predict the beam measurement results in advance, thereby gradually reducing the number of beams in the candidate beam set, reducing beam tracking overhead, allocating more time-frequency resources to data transmission, and improving the overall system throughput.

[0090] After training, samples with earlier retention times are deleted from the sample pool. For example, an expiration time threshold is set, and samples with a retention time longer than the threshold are deleted. This ensures that the neural network is trained with newer samples each time, thereby guaranteeing that the neural network can learn and adapt to the latest changes in the communication environment in real time.

[0091] In one embodiment, a method for applying the present invention to a practical system is provided. Assume a UAV communication system with a base station located at a fixed ground station, providing 128 beams, and a UAV carrying a user terminal, providing 64 beams. First, it is necessary to determine the types of data sources the system can provide. Assuming that the communication system can provide the following types of status information: UAV GPS coordinates (x, y, z axes), attitude angles (pitch, roll, azimuth), and the current received signal power intensity, four sub-networks can be set up for the base station, respectively inputting the base station's current beam index, UAV GPS coordinates, base station's current beam index differential sequence, and the current received signal power intensity. Five sub-networks can be set up for the user, respectively inputting the user's current beam index, UAV GPS coordinates, UAV attitude angles, user's current beam index differential sequence, and the current received signal power intensity.

[0092] In the initial stage, the weights of the beam tracking neural networks for both the base station and the user are randomly initialized, so the candidate beam set is also, in effect, random. However, as measurement and training progress, the neural network gradually learns from the measurement results which beams are more likely to have high received power and which are less likely to have high received power in the current state, thus gradually converging to a relatively stable tracking state. When the communication environment changes and the original rules no longer apply to the new environment, the neural network can quickly learn of this situation through the measurement results and adjust the weights based on the measurement data in the new environment, quickly adapting to the environmental changes. Figure 5 The performance of the algorithm of this invention was compared with that of the comparison algorithm in related technologies. The horizontal axis of the coordinate system represents time, and the vertical axis represents the probability that the tracking beam given by the neural network is the same as the actual optimal beam, referred to as the beam alignment rate. In the simulation, the number of beams searched by the two algorithms was kept the same. The results show that, in the simulation environment, the beam alignment rate of the proposed algorithm can be stabilized at around 90% with small fluctuations. The comparison method only searches a few beams around the current beam, cannot handle complex scenarios such as beam jumps, and the beam alignment rate changes drastically, making it impossible to maintain the stability of the communication link.

[0093] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0094] Through the above description of the embodiments, those skilled in the art can clearly understand that the beam determination method according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0095] According to embodiments of the present invention, a beam-determining apparatus for implementing the above-described beam-determining method is also provided. Figure 6 This is a structural block diagram of a beamforming device provided according to an embodiment of the present invention, such as... Figure 6As shown, the beamforming device includes: an acquisition module 61, a prediction module 62, a determination module 63, a measurement module 64, and a selection module 65. The beamforming device will be described below.

[0096] The acquisition module 61 is used to acquire various beam selection data fed back by the target device for beam selection. The various beam selection data include communication status data for describing the communication channel and perception status data for describing the position and motion status of the target device. The target device includes a terminal and a base station.

[0097] The prediction module 62, connected to the acquisition module 61, is used to input multiple beam selection data into a neural network model, and the neural network model outputs the search values ​​of each of the predetermined multiple beams. The neural network model includes multiple sub-networks that correspond one-to-one with the multiple beam selection data. Each sub-network determines a set of overall beam initial values ​​and a set of offset beam initial values ​​based on the corresponding input beam selection data. The search values ​​of each of the multiple beams are jointly determined by the multiple sets of overall beam initial values ​​and multiple sets of offset beam initial values ​​output by the multiple sub-networks. The overall beam initial value is the initial search value of each of the multiple beams, and the offset beam initial value is used to characterize the degree of influence of the corresponding beam selection data on the search values ​​of the multiple offset beams.

[0098] The determination module 63, connected to the prediction module 62, is used to determine multiple candidate beams among multiple beams based on the search values ​​of each beam.

[0099] The measurement module 64, connected to the determination module 63, is used to measure multiple candidate beams and obtain measurement results of multiple candidate beams. The measurement results of multiple candidate beams are used to characterize the communication quality of multiple candidate beams.

[0100] Selection module 65, connected to measurement module 64, is used to select the target beam for communication with the target device from multiple candidate beams based on the measurement results.

[0101] It should be noted that the acquisition module 61, prediction module 62, determination module 63, measurement module 64, and selection module 65 mentioned above correspond to steps S202 to S210 in the embodiments. Multiple modules and their corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in the above embodiments. It should also be noted that the above modules, as part of the device, can run on the computer terminal 10 provided in the embodiments.

[0102] Embodiments of the present invention may provide a computer device. Optionally, in this embodiment, the computer device may be located in at least one of a plurality of network devices in a computer network. The computer device includes a memory and a processor.

[0103] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the beamforming method and apparatus in this embodiment of the invention. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned beamforming method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to a computer terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0104] The processor can access information and application programs stored in memory via a transmission device to perform the following steps: acquiring multiple beam selection data fed back by the target device for beam selection, wherein the multiple beam selection data includes communication status data describing the communication channel and perception status data describing the position and motion status of the target device, wherein the target device includes a terminal and a base station; inputting the multiple beam selection data into a neural network model, wherein the neural network model outputs the search values ​​of multiple predetermined beams, wherein the neural network model includes multiple sub-networks corresponding one-to-one with the multiple beam selection data, and each sub-network determines a set of initial overall beam values ​​based on the corresponding input beam selection data. A set of initial offset beam values ​​is used, and the search values ​​of multiple beams are jointly determined by multiple sets of overall beam initial values ​​and multiple sets of offset beam initial values ​​output by multiple sub-networks. The overall beam initial value is the initial search value of each beam, and the offset beam initial value characterizes the influence of the corresponding beam selection data on the search values ​​of the multiple offset beams. Based on the search values ​​of each beam, multiple candidate beams are determined from the multiple beams. Measurements are performed on the multiple candidate beams to obtain measurement results, which characterize the communication quality of the multiple candidate beams. Based on the measurement results, the target beam for communication with the target device is selected from the multiple candidate beams.

[0105] This invention provides a beam determination scheme. It acquires various beam selection data fed back by a target device for beam selection. This data includes communication status data describing the communication channel and perception status data describing the position and motion state of the target device, which includes a terminal and a base station. The various beam selection data are input into a neural network model, which outputs predetermined search values ​​for multiple beams. The neural network model includes multiple sub-networks corresponding one-to-one with the various beam selection data. Each sub-network determines a set of overall beam initial values ​​and a set of offset beam initial values ​​based on the corresponding input beam selection data. The search values ​​for each beam are jointly determined by the multiple sets of overall beam initial values ​​and offset beam initial values ​​output by the multiple sub-networks. The overall beam initial value is the initial search value for each beam, and the offset beam initial value is... The value is used to characterize the influence of the corresponding beam selection data on the search value of multiple offset beams; based on the search value of each beam, multiple candidate beams are determined from the multiple beams; the multiple candidate beams are measured to obtain the measurement results of the multiple candidate beams, where the measurement results of the multiple candidate beams are used to characterize the communication quality of the multiple candidate beams; based on the measurement results, the target beam for communication with the target device is selected from the multiple candidate beams, achieving the goal of utilizing communication state data used to describe the channel and sensing state data used to describe the position and motion state of the target device, making full use of all available data for beam prediction, enabling the neural network model to more accurately select the optimal beam in the current environment, improving the accuracy of beam prediction, and thus solving the technical problem of low beam prediction accuracy caused by not fully utilizing the available information for beam prediction.

[0106] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a non-volatile storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc.

[0107] Embodiments of the present invention also provide a non-volatile storage medium. Optionally, in this embodiment, the aforementioned non-volatile storage medium can be used to store the program code executed by the beam determination method provided in the above embodiments.

[0108] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0109] Optionally, in this embodiment, the non-volatile storage medium is configured to store program code for performing the following steps: acquiring multiple beam selection data fed back by the target device for beam selection, wherein the multiple beam selection data includes communication state data for describing the communication channel and sensing state data for describing the position and motion state of the target device, wherein the target device includes a terminal and a base station; inputting the multiple beam selection data into a neural network model, wherein the neural network model outputs the search values ​​of each of the predetermined multiple beams, wherein the neural network model includes multiple sub-networks corresponding one-to-one with the multiple beam selection data, and the multiple sub-networks determine a set of initial overall beams according to the corresponding input beam selection data. The search value of multiple beams is determined by multiple sets of overall beam initial values ​​and multiple sets of offset beam initial values ​​output by multiple sub-networks. The overall beam initial value is the initial search value for each beam, and the offset beam initial value characterizes the influence of the corresponding beam selection data on the search values ​​of the multiple offset beams. Based on the search values ​​of each beam, multiple candidate beams are determined. Measurements are performed on the candidate beams to obtain measurement results, which characterize the communication quality of the candidate beams. Based on the measurement results, a target beam for communication with the target device is selected from the candidate beams.

[0110] Embodiments of the present invention also provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the beam determination method in various embodiments of the present application.

[0111] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0112] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0113] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0114] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0115] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0116] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a non-volatile storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0117] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A beam determination method, characterized in that, include: Acquire multiple beam selection data fed back by the target device for beam selection, wherein the multiple beam selection data includes communication status data for describing the communication channel and sensing status data for describing the position and motion status of the target device, wherein the target device includes a terminal and a base station; The various beam selection data are input into a neural network model, which outputs the search values ​​of each of the predetermined multiple beams. The neural network model includes multiple sub-networks that correspond one-to-one with the various beam selection data. Each of the multiple sub-networks determines a set of overall beam initial values ​​and a set of offset beam initial values ​​based on the corresponding input beam selection data. The search values ​​of each of the multiple beams are jointly determined by the multiple sets of overall beam initial values ​​and multiple sets of offset beam initial values ​​output by the multiple sub-networks. The overall beam initial value is the initial search value of each of the multiple beams, and the offset beam initial value is used to characterize the degree of influence of the corresponding beam selection data on the search values ​​of the multiple offset beams. Based on the search value of each of the multiple beams, multiple candidate beams are determined among the multiple beams; The plurality of candidate beams are measured to obtain the measurement results of the plurality of candidate beams, wherein the measurement results of the plurality of candidate beams are used to characterize the communication quality of the plurality of candidate beams; Based on the measurement results, a target beam for communicating with the target device is selected from the plurality of candidate beams.

2. The method according to claim 1, characterized in that, The step of inputting the multiple beam selection data into a neural network model, and having the neural network model output the search values ​​of each of the predetermined multiple beams, includes: The various beam selection data are input into the multiple sub-networks one by one, and the multiple sub-networks output a set of overall beam initial values ​​and a set of offset beam initial values ​​respectively. The various beam selection data are input into the weight selection network included in the neural network model, and the weight selection network outputs the overall weight and offset weight of each of the multiple sub-networks. The search value of each of the multiple beams is determined based on their initial search value and the overall weight and offset weight of each of the multiple sub-networks.

3. The method according to claim 1, characterized in that, Also includes: If the measurement results of the multiple candidate beams meet the first predetermined conditions, the multiple beam selection data, the search values ​​of the multiple beams, and the measurement results of the multiple candidate beams are combined into training samples and added to the online training sample set.

4. The method according to claim 3, characterized in that, Also includes: If the number of samples included in the online training sample set meets the second predetermined condition, the neural network model is trained online using the online training sample set to obtain an updated neural network model, and the beam used for communication is determined based on the updated neural network model.

5. The method according to claim 4, characterized in that, The step of training the neural network model online using the online training sample set to obtain an updated neural network model includes: Using the online training sample set, with the goal of minimizing the difference between the search value of each of the multiple beams output by the neural network model and the target search value, the multiple sub-networks and weight selection network included in the neural network model are trained to obtain the trained multiple sub-networks and the trained weight selection network, wherein the target search value is determined based on the measurement results of the multiple candidate beams; The updated neural network model is constructed based on the trained subnetworks and the trained weight selection network.

6. The method according to claim 3, characterized in that, Also includes: In the plurality of subnetworks, the subnetworks whose sum of overall weight and offset weight assigned within a predetermined time period is less than a predetermined threshold are marked as disabled, thus obtaining the updated neural network model.

7. The method according to any one of claims 3 to 6, characterized in that, The first predetermined condition is that the beam with the highest search value among the plurality of beams is not the same beam as the beam with the best communication quality characterized by the measurement results.

8. A non-volatile storage medium, characterized in that, The non-volatile storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the non-volatile storage medium to perform the beam determination method according to any one of claims 1 to 7.

9. A computer device, characterized in that, include: Memory and processor The memory stores computer programs; The processor is configured to execute a computer program stored in the memory, wherein when the computer program is executed, the processor performs the beam determination method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that, The computer instructions are executed by the processor according to the beam determination method of any one of claims 1 to 7.

Images