Beam prediction method and device, equipment, storage medium and product

Abstract

The invention relates to the technical field of beam forming, and discloses a beam prediction method and device, equipment, a storage medium and a product, and the method comprises the steps: obtaining the environment information of a wide beam in a current channel, and carrying out the feature extraction of the environment information, and obtaining a channel state feature; inputting the channel state characteristics into a preset hybrid expert model, wherein the hybrid expert model comprises a gating network and a plurality of expert networks; selecting a target expert network with the highest matching degree with each channel state feature from the expert networks according to the gating network; and determining expert output corresponding to each channel state feature based on the target expert network, and determining an optimal beam index prediction result of the current channel according to the expert output. The target expert network with the highest matching degree with the current channel is dynamically selected through the gating network of the hybrid expert model, so that the model can perform specialized prediction according to the characteristics of signal environments in different scenes, the limitation of a single model is avoided, and the accuracy and scene adaptability of beam prediction are improved.

Description

Technical Field

[0001] The present application relates to the field of beamforming technology, and in particular to a beam prediction method, apparatus, device, storage medium, and product. Background Art

[0002] The development and application of massive antennas can effectively improve the energy efficiency and spectrum utilization of MIMO (Multiple Input Multiple Output) systems, but all of this relies on precise beamforming technology. As terminal mobility increases, this forces the base station to perform more frequent beam scanning, beam measurement, beam reporting, and beam alignment. However, when existing FNN (Feedforward Neural Network)-based solutions are applied to beam prediction, a single deep learning model is usually used to train and predict all channel environments. However, the changes in wireless channels are complex and highly time-varying, resulting in poor performance of the trained model in new environments. Summary of the Invention

[0003] The main purpose of this application is to provide a beam prediction method, apparatus, device, storage medium and product, aiming to solve the technical problem that when the existing FNN-based solution is applied to beam prediction, the changes in the wireless channel are complex and highly time-varying, resulting in the poor performance of the trained model in the new environment.

[0004] To achieve the above objectives, the present application proposes a beam prediction method, which includes:

[0005] Acquire environmental information of the wide beam in the current channel, and perform feature extraction on the environmental information to obtain channel state features;

[0006] Inputting the channel state feature into a preset hybrid expert model, wherein the hybrid expert model includes a gated network and multiple expert networks;

[0007] Selecting a target expert network with the highest matching degree with each of the channel state features from the expert networks according to the gating network;

[0008] An expert output corresponding to each of the channel state features is determined based on the target expert network, and an optimal beam index prediction result for the current channel is determined according to the expert output.

[0009] Optionally, the step of determining the expert output corresponding to each of the channel state features based on the target expert network, and determining the optimal beam index prediction result for the current channel according to the expert output includes:

[0010] Inputting the channel state features into the corresponding target expert network, performing forward propagation calculation, and outputting the expert output of each target expert network for the narrow beam index;

[0011] Adding the expert outputs to obtain an expert mixed output;

[0012] The expert mixture output is converted into a probability distribution of beam indices based on a normalized exponential function, and an index with the highest probability is selected from the probability distribution as a beam index prediction result.

[0013] Optionally, the environmental information includes signal reception strength, user location information, and historical beam selection data;

[0014] The step of obtaining wide beam environmental information in the current channel and performing feature extraction on the environmental information to obtain channel state features includes:

[0015] Calculate the signal reception strength of the beam receivable by the target user in the channel based on the channel vector and codebook vector of the current channel;

[0016] Performing coordinate system conversion and noise modeling on the location coordinates of the target user to obtain user location information;

[0017] Converting the beam selection information corresponding to the target user into a spatial feature sequence to generate historical beam selection data;

[0018] Feature extraction is performed on the signal reception strength, the user location information, and the historical beam selection data to obtain channel state features.

[0019] Optionally, the step of calculating the signal reception strength of the beam receivable by the target user in the channel based on the channel vector and codebook vector of the current channel includes:

[0020] constructing a channel vector between the network terminal and the target user based on the complex gain, azimuth, and elevation of each transmission path in the current channel;

[0021] Determine a codebook vector according to the spacing between adjacent antenna elements and the wavelength of the millimeter wave carrier in the current channel;

[0022] A matrix operation is performed on the channel vector and the codebook vector to obtain the signal reception strength of each beam that can be received by the target user.

[0023] Optionally, before the step of inputting the channel state feature into a preset hybrid expert model, the method further includes:

[0024] Build an initial expert model based on the gating network and the expert network;

[0025] Acquiring historical environment information of the current channel, wherein the historical environment information includes historical signal reception strength of the wide beam and historical user location information;

[0026] Determining an optimal beam index tag of a narrow beam in a current channel according to the historical signal reception strength and the historical user location information;

[0027] The initial expert model is trained based on the historical environment information and the optimal beam index label to obtain a hybrid expert model.

[0028] Optionally, after the step of determining the expert output corresponding to each of the channel state features based on the target expert network, and determining the optimal beam index prediction result for the current channel according to the expert output, the step further includes:

[0029] generating a beam adjustment control signal according to the beam index prediction result;

[0030] Adjusting the beam direction of the network terminal in the current channel based on the beam adjustment control signal to establish a target narrow beam index;

[0031] Record beam selection data corresponding to the target narrow beam index, and update the environment information of the current channel.

[0032] In addition, to achieve the above objectives, the present application also proposes a beam prediction device, which includes:

[0033] A feature extraction module is used to obtain the environmental information of the wide beam in the current channel and perform feature extraction on the environmental information to obtain channel state features;

[0034] A model input module, configured to input the channel state characteristics into a preset hybrid expert model, wherein the hybrid expert model includes a gated network and multiple expert networks;

[0035] A network selection module, configured to select a target expert network having the highest matching degree with each of the channel state features from the expert networks according to the gating network;

[0036] A beam prediction module is used to determine the expert output corresponding to each of the channel state features based on the target expert network, and determine the optimal beam index prediction result of the current channel according to the expert output.

[0037] In addition, to achieve the above-mentioned purpose, the present application also proposes a beam prediction device, which includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the computer program is configured to implement the steps of the beam prediction method as described above.

[0038] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the beam prediction method described above are implemented.

[0039] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the beam prediction method as described above.

[0040] This application discloses obtaining environmental information of a wide beam in the current channel, performing feature extraction on the environmental information, and obtaining channel state features; inputting the channel state features into a preset hybrid expert model, the hybrid expert model comprising a gating network and multiple expert networks; selecting a target expert network from the expert network with the highest degree of match for each of the channel state features based on the gating network; determining the expert output corresponding to each of the channel state features based on the target expert network, and determining the optimal beam index prediction result for the current channel based on the expert output. By dynamically selecting the target expert network with the highest degree of match for the current channel through the gating network in the hybrid expert model, the model can perform specialized predictions for the features of the signal environment in different scenarios, avoiding the limitations of a single model, and thereby improving the accuracy of beam prediction and scenario adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0043] Figure 1 This is a flowchart of the first embodiment of the beam prediction method of the present application;

[0044] Figure 2 This is a schematic diagram of the model training of the expert mixture model of this application;

[0045] Figure 3 This is a flowchart of the second embodiment of the beam prediction method of the present application;

[0046] Figure 4 This is a flowchart of the third embodiment of the beam prediction method of the present application;

[0047] Figure 5 This is a schematic diagram of the architecture of the expert hybrid model for this application;

[0048] Figure 6 This is a schematic diagram of the module structure of the beam prediction device according to an embodiment of the present application;

[0049] Figure 7 Schematic diagram of the device structure of the hardware operating environment involved in the beam prediction method in the embodiment of the present application.

[0050] The purpose, features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0051] It should be understood that the specific embodiments described herein are merely used to explain the technical solutions of the present application and are not intended to limit the present application.

[0052] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0053] In 5G, 6G, and future wireless communication systems, communications in the mmWave and THz bands experience significant signal propagation losses due to their high-frequency characteristics. Beamforming technology is crucial for improving communication quality. Increased terminal mobility compels base stations to perform more frequent beam scanning, beam measurement, beam reporting, and beam alignment. The surge in the number of antennas further increases scanning overhead. AI-based beam prediction has been widely researched. The base station activates a subset of antennas to generate wide beams. By scanning only the signal reception strength of these wide beams, the optimal beam index across all beams is predicted, thereby reducing beam scanning overhead. However, existing beam management systems primarily rely on traditional beam scanning or static prediction methods based on historical data, which struggle to meet certain practical requirements. For example, in complex environments such as urban high-rise buildings and multipath indoor environments, wireless signal propagation exhibits highly dynamic variations. Existing AI models have poor generalization across different environments, resulting in reduced prediction accuracy. Therefore, an intelligent beam prediction method that can adapt to diverse channel environments is needed to improve system robustness and generalization.

[0054] Therefore, this application provides a beam prediction method based on a mixture of experts model. This method uses multiple expert networks to collaboratively learn the characteristics of different channel environments, and a gating network dynamically selects the optimal expert for prediction, improving the model's adaptability. This method solves the aforementioned problems and enables communication systems to maintain stable and efficient signal transmission even in complex environments.

[0055] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a computer, or an electronic device capable of implementing the above functions. This embodiment and the following embodiments are described below using a beam prediction system as an example.

[0056] Based on this, the embodiment of the present application provides a beam prediction method, referring to Figure 1 , Figure 1 This is a flowchart of the first embodiment of the beam prediction method of the present application.

[0057] In this embodiment, the beam prediction method includes:

[0058] Step S10: Acquire the wide beam environment information in the current channel, and perform feature extraction on the environment information to obtain channel state features.

[0059] It should be noted that in wireless communications, a wide beam refers to a beam with a wider beam width in the antenna radiation pattern, which is usually used to cover a larger area to achieve initial signal capture or large-scale user access. Compared with a narrow beam, a wide beam has lower gain but a wider coverage range, and is suitable for the channel environment detection phase. Environmental information represents the physical layer data of the current communication channel state, including but not limited to the signal reception strength of the wide beam, the spatial position coordinates of the user equipment, historical beam selection records, etc., reflecting the channel propagation characteristics, user distribution and beam usage patterns. Channel state features are abstract feature vectors used to describe the current channel transmission characteristics, which can be directly input into the prediction model to characterize the channel state.

[0060] It is understandable that through the fusion and feature extraction of multi-dimensional environmental information, the physical characteristics of the channel (such as multipath fading and shadow effects) can be more accurately characterized, avoiding the misjudgment of the channel state caused by a single parameter (such as only signal strength). The extracted channel state features remove the noise and redundant information in the original data, making the subsequent training and reasoning of the hybrid expert model more efficient, reducing the model convergence time and computational overhead.

[0061] Step S20: inputting the channel state characteristics into a preset hybrid expert model, where the hybrid expert model includes a gated network and multiple expert networks.

[0062] It's important to note that the hybrid expert model is an integrated learning architecture consisting of multiple expert networks dedicated to specific tasks and a gating network responsible for scheduling. In beam prediction scenarios, each expert network focuses on learning the characteristic beam mapping relationship under different channel conditions, while the gating network dynamically assigns weights based on the input channel state characteristics to determine the contribution of each expert network.

[0063] As you can see, the gating network is the core component of the hybrid expert model. It receives channel state features as input and dynamically selects the expert network by calculating the matching weights of each expert network. This can be done using a fully connected layer or an attention mechanism. The output weight vector represents the probability that each expert network should process the current input.

[0064] It can be understood that the expert network is a subnetwork within the hybrid expert model, responsible for specialized processing of specific types of channel characteristics. Each expert network can be optimized for different scenarios (such as static users, high-speed mobile users) or channel conditions (such as line-of-sight, non-line-of-sight), and generates a predicted output of the beam index through forward propagation calculations.

[0065] Furthermore, in order to train the MOE (Mixture of Experts) model using supervised learning so that it can adapt to different channel environments and improve generalization ability and prediction accuracy, before step S20, the following steps are further included:

[0066] An initial expert model is constructed based on a gated network and an expert network; historical environmental information of the current channel is obtained, wherein the historical environmental information includes historical signal reception strength of a wide beam and historical user location information; an optimal beam index label of a narrow beam in the current channel is determined based on the historical signal reception strength and the historical user location information; and the initial expert model is trained based on the historical environmental information and the optimal beam index label to obtain a hybrid expert model.

[0067] It is understandable that the initial expert model is a hybrid expert model framework constructed before training, which contains the initial structure of the gated network and multiple expert networks, but the parameters are not optimized. Historical environmental information is the channel state data that records the historical communication process between the base station and the user equipment, which may include historical signal reception strength and historical user location information. The optimal beam index label is a narrow beam index that can achieve the best communication quality (such as the highest signal strength, the lowest bit error rate) under specific channel conditions based on historical environmental information. This label serves as the goal of supervised learning to guide the model to learn the mapping relationship between channel characteristics and the optimal beam.

[0068] In one example, reference Figure 2 , Figure 2 This is a diagram showing the model training for the Mixed Experts model used in this application. After the Mixed Experts (MoE) model is constructed, the model remains untrained, with the weight parameters of the expert model and gating network randomly initialized. To enable the model to effectively perform beam prediction, supervised learning can be used for training and optimization of model parameters, as shown in the figure. The training data consists of wide-beam signal reception strength, user location information, historical beam selection data, and its corresponding true beam labels.

[0069] The wide beam intensity, user location, historical beam selection data and labels in wireless communication data are used as input. The features of wide beam intensity, user location and historical beam selection data are extracted and sent to the hybrid expert model. Different expert networks in the model respectively produce expert outputs for wide beam, user location and historical beam. Finally, the outputs are fused with the labels, and the model is trained using the fused data.

[0070] When training a neural network, different types of loss functions may be selected depending on the task. Since beam prediction is a multi-classification task, the cross-entropy loss (Cross-Entropy Loss) in the classification task can be used. The cross-entropy loss function is suitable for measuring the gap between the predicted probability distribution and the true category distribution. Its mathematical formula is as follows:

[0071]

[0072] Where N is the number of beam categories (i.e. the total number of beams supported by the base station), y i is the true beam label, using one-hot encoding, which is 1 for the correct category and 0 for other categories. is the model's predicted probability for the i-th beam. The goal of the cross-entropy loss is to maximize the probability output corresponding to the true class and minimize the output probabilities of other classes. By adjusting the network parameters through backpropagation, the predicted results are closer to the true beam selection, improving the accuracy of beam prediction.

[0073] During model training, it is also necessary to select a suitable optimizer to adjust the model parameters. This embodiment uses the Adam optimizer and combines it with a learning rate adjustment strategy to improve training efficiency and convergence speed. The initial learning rate of the Adam optimizer is set to 0.001, the total number of training rounds is 200 epochs, and an exponential decay strategy is adopted, that is, the learning rate decays by 10% after every 50 epochs. In addition, to prevent the gradient explosion problem during training, it is necessary to introduce gradient clipping to limit the gradient norm to less than 5.

[0074] Step S30 : selecting a target expert network having the highest matching degree with each of the channel state features from the expert networks according to the gating network.

[0075] It is understandable that the gating network quantifies the degree of matching between each expert network and the current channel state features based on preset matching rules (such as calculating the similarity and weight of the channel state features and each expert network in the feature space through fully connected layers and attention mechanisms). For example, the matching scores of each expert network are converted into a probability distribution using the Softmax function, where the higher the probability, the higher the matching degree. From multiple expert networks, the expert network with the highest matching degree (such as the highest probability) is selected and determined as the target expert network. Subsequently, it performs targeted beam prediction processing on the channel state features, allowing the model to flexibly call the most suitable expert network capabilities based on the actual channel conditions to improve prediction accuracy.

[0076] In one example, the MoE model contains multiple expert networks, each consisting of a two-layer multilayer perceptron, and each expert network focuses on different types of data to improve the robustness and adaptability of beam prediction.

[0077]

[0078] in, and are the outputs of the first and second layers of the k-th expert network, X represents the feature input after data extraction, ReLU is the activation function, and are the weight matrices of the first and second layers of the k-th expert network, and is the bias term.

[0079] The gating network is used to dynamically select the optimal expert model by calculating the weight of each expert model as the matching degree between each expert and the input feature:

[0080] g=Softmax(W G ·X+b G )

[0081] Where X represents the feature input after data extraction, W G and b G represents the weight and bias parameters of the gating network, the matching degree between the input features and all expert models [g1,g2,…,g k ]satisfy:

[0082]

[0083] In the traditional MoE method, the final prediction result is usually the weighted sum of all experts’ outputs. However, this method requires calculating the forward propagation of all experts, which has a high computational complexity. To reduce the computational overhead, this embodiment adopts the Top-1 selection strategy, that is, only selects the expert model k with the highest matching degree.* Perform the calculation:

[0084]

[0085]

[0086] That is, the expert with the highest matching score is the final output for that feature type. Compared to the weighted summation of all experts, the Top-1 selection strategy significantly reduces computational effort without sacrificing performance, improving inference speed and making it suitable for devices with limited computing resources or low-latency communication scenarios.

[0087] Step S40: determining expert outputs corresponding to the channel state features based on the target expert network, and determining an optimal beam index prediction result for the current channel according to the expert outputs.

[0088] It should be noted that the expert output is the result generated by the target expert network after forward propagation calculation of the channel state characteristics, which is usually the probability distribution of the beam index.

[0089] Furthermore, after the beam index is predicted, the beam direction can be adjusted by generating a control signal, a target narrow beam index is established, and environmental information is updated, thereby optimizing the beam pointing of the base station or terminal in real time to ensure the stability of the communication link. Meanwhile, new beam selection data is recorded and environmental information is updated, enabling the system to continuously learn channel variation patterns. After step S40, the following steps are further included:

[0090] A beam adjustment control signal is generated according to the beam index prediction result; a beam direction of a network terminal in a current channel is adjusted based on the beam adjustment control signal, and a target narrow beam index is established; beam selection data corresponding to the target narrow beam index is recorded, and environmental information of the current channel is updated.

[0091] It should be noted that the beam adjustment control signal is generated based on the beam index prediction results and is used to instruct the network terminal to adjust the beam direction. It can precisely control parameters such as the phase and amplitude of the terminal antenna array to achieve beam pointing changes. Network terminals are devices that participate in signal transmission and reception in wireless communications, such as base stations and user terminals (mobile phones, IoT devices, etc.). Here, the device that performs beam adjustment is responsible for changing the beam direction according to the control signal.

[0092] It can be understood that the target narrow beam index is the index corresponding to the narrow beam actually used by the network terminal to establish communication after adjustment. It represents a specific beam direction and is used to identify and manage different beams. The beam selection data records information related to the beam adjustment, including the adjustment basis (beam index prediction result), the adjusted target narrow beam index, and the channel environment (signal strength, user location) at the time of adjustment.

[0093] Specifically, using the beam index prediction results, a control signal containing key information such as the target narrow beam index can be generated in an instruction format recognizable by the network terminal (such as a specific protocol frame, parameter configuration instruction), clearly informing the terminal of the beam direction to be adjusted. After receiving the beam adjustment control signal, the network terminal changes the beam radiation direction by adjusting the phase, amplitude amplifier and other hardware modules of its own antenna array according to the target narrow beam index in the signal, completes the establishment of the target narrow beam, and switches the communication link to the direction corresponding to the beam. The terminal or network-side device records and stores the beam selection data involved in this beam adjustment, including the target narrow beam index, adjustment trigger conditions (such as prediction results), channel signal strength during adjustment, user location, etc.; and integrates this new data into the channel environment information library, updates the original environment information, and provides more real-time and accurate training and inference data for subsequent beam prediction models.

[0094] This embodiment discloses obtaining environmental information of a wide beam in the current channel, performing feature extraction on the environmental information to obtain channel state features; inputting the channel state features into a preset hybrid expert model, the hybrid expert model comprising a gating network and multiple expert networks; selecting a target expert network from the expert networks that best matches each of the channel state features based on the gating network; determining expert outputs corresponding to each of the channel state features based on the target expert network, and determining the optimal beam index prediction result for the current channel based on the expert outputs. By dynamically selecting the target expert network that best matches the current channel through the gating network in the hybrid expert model, the model can perform specialized predictions for the characteristics of the signal environment in different scenarios, avoiding the limitations of a single model and thereby improving the accuracy and scenario adaptability of beam predictions.

[0095] Reference Figure 3 , Figure 3 This is a flow chart of the second embodiment of the beam prediction method of the present application. Based on the above-mentioned first embodiment, the second embodiment of the beam prediction method of the present application is proposed.

[0096] In the second embodiment, the environmental information includes signal reception strength, user location information, and historical beam selection data; and step S10 includes:

[0097] Step S101 : calculating the signal reception strength of a beam receivable by a target user in the channel based on a channel vector and a codebook vector of a current channel.

[0098] It should be noted that the channel vector is a mathematical vector representing the transmission characteristics of a wireless channel. It is constructed based on a multipath propagation model and includes parameters such as the complex gain (amplitude and phase information), azimuth (horizontal angle of the signal's arrival direction), and elevation (vertical angle of the signal's arrival direction) of each transmission path in the channel. It is used to quantify the propagation loss and spatial characteristics of the signal from the transmitter to the target user. The codebook vector describes the beam direction of the antenna array. It is determined by the spacing between adjacent antenna elements in the current channel and the millimeter-wave carrier wavelength. Weighting coefficients for different beam directions are expressed in matrix or vector form, allowing the antenna array to radiate a beam in a specific direction.

[0099] Furthermore, in order to make the obtained signal reception strength closer to the actual communication scenario based on the calculation method of the physical layer channel model and avoid the error caused by the simplified model, the step S101 may include:

[0100] Based on the complex gain, azimuth, and elevation of each transmission path in the current channel, a channel vector between the network terminal and the target user is constructed; a codebook vector is determined based on the spacing between adjacent antenna units and the millimeter wave carrier wavelength in the current channel; and a matrix operation is performed on the channel vector and the codebook vector to obtain the signal reception strength of each beam that can be received by the target user.

[0101] Understandably, to ensure that beam prediction models in wireless communication systems can accurately learn the characteristics of different channel environments, a high-quality dataset is required. Data collection includes three types of input features: wide beam signal reception strength, user location information, and historical beam selection. Wide beam signal reception strength is a key input, characterizing the communication quality between each wide beam and the user when used as a transmission beam.

[0102] In one example, consider a millimeter wave channel with p multipaths, where each path l has a complex gain of α l , the azimuth and elevation angles are φ l ,θ l , the M-dimensional beam pointing vector at these angles is a(φ l ,θ l ). The channel vector between the base station and the user can be expressed as:

[0103]

[0104] Next, let d and γ represent the distance between adjacent antenna units and the wavelength of the millimeter wave carrier, respectively, and generally satisfy d = γ / 2. ULA For a uniform linear array of antenna elements, the beam steering vector with a spatial angle of θ can be expressed as:

[0105]

[0106] The codebook matrix is ​​defined as:

[0107]

[0108] Based on the channel vector and codebook vector, the base station can calculate the received signal strength (RSS) of all beams received by the user. The RSS of the i-th beam can be expressed as:

[0109]

[0110] Among them, the codebook matrix W ULA The i-th column is the codebook vector w i , P is the transmission power.

[0111] Step S102 : performing coordinate system conversion and noise modeling on the location coordinates of the target user to obtain user location information.

[0112] It is understandable that user relative position information is one of the important input features for beam prediction in wireless communication systems, especially in millimeter wave communications and massive MIMO systems, where the user's position relative to the base station is highly correlated with the beam direction.

[0113] In one example, GPS data is used to obtain the location of the user device by recording its latitude and longitude coordinates (lat, lon, alt). Subsequently, the location information is preprocessed to convert the GPS latitude and longitude coordinates into a Cartesian coordinate system (X, Y, Z) relative to the base station so that it can be aligned with the wide-beam signal reception strength data. Considering the measurement errors between GPS and base station ranging, Gaussian noise is used to model the location data to simulate the errors in the real environment. The formula for GPS error modeling is expressed as:

[0114]

[0115] Here, σ is the standard deviation. For example, at 2.04 meters, the probability that the noise amplitude exceeds 5 meters is less than 5%.

[0116] Step S103: convert the beam selection information corresponding to the target user into a spatial feature sequence to generate historical beam selection data.

[0117] It should be noted that beam selection information records the beam index used by the target user during historical communications, reflecting the historical record of beam matching between the user and the base station. The spatial feature sequence is an ordered sequence of the spatial positions (such as latitude and longitude, relative base station coordinates) and beam indices in beam selection information, arranged in chronological order. It is used to characterize the relationship between user movement trajectory and beam selection.

[0118] In one example, in order to optimize the beam prediction model, it is necessary to collect the beam selection information of the base station for the user over a period of time. The beam selection history B can be expressed as:

[0119] B=[b1,b2,…,b T ]

[0120] Among them, b t represents the beam index of the t-th frame, and T is the time window length. At the same time, due to the close spatial connection between the wide beam and the narrow beam, it is possible to predict the optimal index in the narrow beam through the wide beam. That is, through M b wide beams to predict N b The optimal index i in a narrow beam is expressed as:

[0121] Step S104: extracting features from the signal reception strength, the user location information, and the historical beam selection data to obtain channel state features.

[0122] It is understandable that before beam prediction, the MOE model first needs to perform independent feature extraction on different types of input data to ensure a variety of data types so that each expert can focus on processing specific data categories.

[0123] In one example, the received signal strength of all wide beams is used as a feature vector to characterize the communication quality between each wide beam and the user when it is used as a transmission beam:

[0124]

[0125] Next, the latitude and longitude coordinates (lat, lon, alt) recorded by the GPS user device are converted to a Cartesian coordinate system (X, Y, Z) to align with the other data. In order to simulate the measurement error in the real environment, Gaussian noise can be used to model the GPS position:

[0126]

[0127] Secondly, assume that the base station selects a beam index b for the user equipment at each time frame t t , then the historical beam selection sequence can be expressed as:

[0128] X history =[b t-T ,b t-T+1 ,...,b t ]

[0129] Finally, the three types of features are input into the model in sequence to obtain three types of expert outputs.

[0130] In this embodiment, the signal reception strength of the beam receivable by the target user in the channel is calculated based on the channel vector and codebook vector of the current channel; the position coordinates of the target user are subjected to coordinate system conversion and noise modeling to obtain user position information; the beam selection information corresponding to the target user is converted into a spatial feature sequence to generate historical beam selection data; and feature extraction is performed on the signal reception strength, the user position information, and the historical beam selection data to obtain channel state features. By calculating the signal strength based on the channel vector and codebook vector, performing coordinate system conversion and noise modeling on the user position, and converting the historical beam selection data into a spatial feature sequence, the feature data of the input model can accurately reflect the physical characteristics of the channel, the spatial distribution of users, and the historical beam usage patterns, providing high-quality input for the hybrid expert model.

[0131] Reference Figure 4 , Figure 4 This is a flow chart of the third embodiment of the beam prediction method of the present application. Based on the above embodiments, the third embodiment of the beam prediction method of the present application is proposed.

[0132] In the third embodiment, step S40 includes:

[0133] Step S401: input the channel state features into the corresponding target expert network, perform forward propagation calculation, and output the expert output of each target expert network for narrow beam index.

[0134] It can be understood that when the channel state features are input into the corresponding target expert network, there are three types of features. After entering the MOE model, three types of expert outputs will be obtained:

[0135] H wide =MOE(X wide )

[0136] H location =MOE(X location )

[0137] H history =MOE(X history )

[0138] If the target expert network is designed for the "high-speed mobile user" scenario, its forward propagation will pay more attention to the speed parameters and historical switching frequency in the features; while the expert network for "static users" focuses on the signal strength stability characteristics to achieve scenario-specific beam prediction.

[0139] Step S402: Add the expert outputs to obtain an expert mixed output.

[0140] It should be understood that by integrating the knowledge of multiple experts, the model can adapt to a wider range of channel scenarios and avoid the "cognitive bias" of a single expert. The features of the three types of experts are added together to obtain the expert mixed output H fusion for:

[0141] H fusion =H wide +H location +H history

[0142] Then the expert mix output H fusion Input into the softmax function to calculate the probability:

[0143] S=Softmax(W predict H fusion +b predict )

[0144] Step S403: Convert the expert mixture output into a probability distribution of beam indices based on a normalized exponential function, and select the index with the highest probability from the probability distribution as the beam index prediction result.

[0145] It should be noted that the normalized exponential function can convert any real number vector into a mathematical function of probability distribution. In beam prediction, it is used to convert the non-normalized scores output by the expert mixture into probability values ​​for each beam index, ensuring that the sum of all probabilities is 1.

[0146] It is understandable that the expert mixture output is converted into a probability distribution of beam indices:

[0147] S=[p 1 ,p 2 ,…,p N ]

[0148] Among them, p j represents the probability of predicting beam j, where N is the total number of beams. The differentiability of the Softmax function supports backpropagation during model training (e.g., optimizing parameters through the cross-entropy loss function), enabling the mixture of experts model to learn how to adjust weights to improve the prediction accuracy of high-probability beams. Using a probability distribution, the system can quantify the reliability of prediction results rather than simply outputting a single index. Compared to directly using the raw scores output by the mixture of experts (e.g., selecting the highest-scoring index), the probability distribution transformed by Softmax can suppress the influence of outliers.

[0149] Finally, the system selects the beam index b* with the highest probability as the final prediction result:

[0150]

[0151] At the same time, the system sends b* to the base station for beamforming.

[0152] In one example, reference Figure 5 , Figure 5 This is a schematic diagram of the architecture of the expert hybrid model for this application. First, the system activates some antennas to obtain a wide beam. The base station scans the user's signal reception strength, user location information, and historical beam selection data under different wide beams as model input. Feature extraction is performed on the three types of original data. The three types of extracted features are then input into the hybrid expert model to obtain the corresponding expert input (such as wide beam expert output, user location expert output, historical beam expert output, etc.). Finally, the three types of expert outputs are fused, and beam prediction is performed based on the fusion results. Among them, the hybrid expert model includes a gating network and multiple expert networks (such as Expert 1, Expert 2, and Expert 3, etc.). The gating network calculates the gating value, and determines which expert network is most suitable for processing the current feature through Top-1 selection, and activates the corresponding expert network for training. The expert network selected by the gating network will output its own prediction results based on the input features.

[0153] This embodiment discloses inputting the channel state features into the corresponding target expert network, performing forward propagation calculations, and outputting the expert outputs of each target expert network for narrow beam indices. These expert outputs are summed to obtain a mixed expert output. This mixed expert output is converted into a probability distribution of beam indices based on a normalized exponential function, and the index with the highest probability from this probability distribution is selected as the beam index prediction result. Only the expert model with the highest matching degree is selected for calculation, i.e., the expert output with the highest matching degree becomes the final output for this type of feature. Compared to the weighted summation of all experts, this method significantly reduces the amount of computation without sacrificing performance, improving inference speed and making it suitable for devices with limited computing resources or low-latency communication scenarios.

[0154] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the beam prediction method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0155] This application also provides a beam prediction device, please refer to Figure 6 , the beam prediction device includes:

[0156] A feature extraction module 10 is used to obtain the wide beam environment information in the current channel and perform feature extraction on the environment information to obtain channel state features;

[0157] A model input module 20 is used to input the channel state characteristics into a preset hybrid expert model, wherein the hybrid expert model includes a gated network and multiple expert networks;

[0158] A network selection module 30 is configured to select a target expert network having the highest matching degree with each of the channel state features from the expert networks according to the gating network;

[0159] The beam prediction module 40 is configured to determine the expert output corresponding to each of the channel state features based on the target expert network, and determine the optimal beam index prediction result for the current channel according to the expert output.

[0160] The beam prediction device provided in this application, employing the beam prediction method of the aforementioned embodiment, can address the technical issue with existing FNN-based solutions for beam prediction, where complex and highly time-varying wireless channel variations lead to poor performance of trained models in new environments. Compared to the prior art, the beam prediction device provided in this application has the same beneficial effects as the beam prediction method provided in the aforementioned embodiment, and other technical features of the beam prediction device are the same as those disclosed in the aforementioned embodiment method and are not further elaborated here.

[0161] The present application provides a beam prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the beam prediction method in the above-mentioned embodiment one.

[0162] Reference below Figure 7 , which shows a schematic diagram of the structure of a beam prediction device suitable for implementing the embodiments of the present application. The beam prediction device in the embodiments of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 7 The beam prediction device shown is merely an example and should not limit the functionality and scope of use of the embodiments of the present application.

[0163] like Figure 7As shown, the beam prediction device may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the beam prediction device. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: an input device 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, hard disk, etc.; and a communication device 1009. Communication device 1009 can allow the beam prediction device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a beam prediction device with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented or have alternatively.

[0164] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device 1003, or installed from a read-only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are performed.

[0165] The beam prediction device provided in this application, employing the beam prediction method of the aforementioned embodiment, can address the technical issue of existing FNN-based solutions for beam prediction, where complex and highly time-varying wireless channel variations lead to poor performance of trained models in new environments. Compared to the prior art, the beam prediction device provided in this application has the same beneficial effects as the beam prediction method provided in the aforementioned embodiment, and other technical features of the beam prediction device are the same as those disclosed in the aforementioned embodiment, and are not further elaborated here.

[0166] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0167] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0168] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, a computer program) stored thereon, wherein the computer-readable program instructions are used to execute the beam prediction method in the above-mentioned embodiment.

[0169] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared or semiconductor systems, systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0170] The computer-readable storage medium may be included in the beam prediction device; or may exist independently without being assembled into the beam prediction device.

[0171] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the beam prediction device, the beam prediction device performs the beam prediction method described above.

[0172] Computer program code for performing the operations of the present application may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0173] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of code, and the module, program segment or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0174] The modules described in the embodiments of the present application may be implemented in software or hardware, wherein the name of a module does not necessarily limit the unit itself.

[0175] The computer-readable storage medium provided in this application is a computer-readable storage medium storing computer-readable program instructions (i.e., a computer program) for executing the aforementioned beam prediction method. This computer-readable storage medium can address the technical issue of existing FNN-based solutions for beam prediction, where complex and highly time-varying wireless channel variations lead to poor performance of trained models in new environments. Compared to the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the beam prediction method provided in the aforementioned embodiments and are not further elaborated here.

[0176] The present application also provides a computer program product, comprising a computer program, which implements the steps of the above-mentioned beam prediction method when executed by a processor.

[0177] The computer program product provided in this application can address the technical issue that, when applied to beam prediction using existing FNN-based solutions, the complex and highly time-varying nature of wireless channels leads to poor performance of trained models in new environments. Compared to existing technologies, the beneficial effects of the computer program product provided in this application are similar to those of the beam prediction methods provided in the aforementioned embodiments and are not further elaborated here.

[0178] The above description is only part of the embodiments of the present application and does not limit the patent scope of the present application. All equivalent structural transformations made by using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect application in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A beam prediction method, characterized in that: The beam prediction method comprises: Acquire environmental information of the wide beam in the current channel, and perform feature extraction on the environmental information to obtain channel state features; Inputting the channel state feature into a preset hybrid expert model, wherein the hybrid expert model includes a gated network and multiple expert networks; Selecting a target expert network with the highest matching degree with each of the channel state features from the expert networks according to the gating network; An expert output corresponding to each of the channel state features is determined based on the target expert network, and an optimal beam index prediction result for the current channel is determined according to the expert output.

2. The beam prediction method according to claim 1, wherein: The step of determining the expert output corresponding to each of the channel state features based on the target expert network, and determining the optimal beam index prediction result of the current channel according to the expert output, includes: Inputting the channel state features into the corresponding target expert network, performing forward propagation calculation, and outputting the expert output of each target expert network for the narrow beam index; Adding the expert outputs to obtain an expert mixed output; The expert mixture output is converted into a probability distribution of beam indices based on a normalized exponential function, and an index with the highest probability is selected from the probability distribution as a beam index prediction result.

3. The beam prediction method according to claim 1, wherein: The environmental information includes signal reception strength, user location information and historical beam selection data; The step of obtaining wide beam environmental information in the current channel and performing feature extraction on the environmental information to obtain channel state features includes: Calculate the signal reception strength of the beam receivable by the target user in the channel based on the channel vector and codebook vector of the current channel; Performing coordinate system conversion and noise modeling on the location coordinates of the target user to obtain user location information; Converting the beam selection information corresponding to the target user into a spatial feature sequence to generate historical beam selection data; Feature extraction is performed on the signal reception strength, the user location information, and the historical beam selection data to obtain channel state features.

4. The beam prediction method according to claim 3, wherein: The step of calculating the signal reception strength of the beam receivable by the target user in the channel based on the channel vector and the codebook vector of the current channel includes: constructing a channel vector between the network terminal and the target user based on the complex gain, azimuth, and elevation of each transmission path in the current channel; Determine a codebook vector according to the spacing between adjacent antenna elements and the wavelength of the millimeter wave carrier in the current channel; A matrix operation is performed on the channel vector and the codebook vector to obtain the signal reception strength of each beam that can be received by the target user.

5. The beam prediction method according to claim 1, wherein: Before the step of inputting the channel state feature into a preset hybrid expert model, the method further includes: Build an initial expert model based on the gating network and the expert network; Acquiring historical environment information of the current channel, wherein the historical environment information includes historical signal reception strength of the wide beam and historical user location information; Determining an optimal beam index tag of a narrow beam in a current channel according to the historical signal reception strength and the historical user location information; The initial expert model is trained based on the historical environment information and the optimal beam index label to obtain a hybrid expert model.

6. The beam prediction method according to any one of claims 1 to 5, characterized in that: After the steps of determining the expert output corresponding to each of the channel state features based on the target expert network, and determining the optimal beam index prediction result of the current channel according to the expert output, the method further includes: generating a beam adjustment control signal according to the beam index prediction result; Adjusting the beam direction of the network terminal in the current channel based on the beam adjustment control signal to establish a target narrow beam index; Record beam selection data corresponding to the target narrow beam index, and update the environment information of the current channel.

7. A beam prediction device, characterized in that: The device comprises: A feature extraction module is used to obtain the environmental information of the wide beam in the current channel and perform feature extraction on the environmental information to obtain channel state features; A model input module, configured to input the channel state characteristics into a preset hybrid expert model, wherein the hybrid expert model includes a gated network and multiple expert networks; A network selection module, configured to select a target expert network having the highest matching degree with each of the channel state features from the expert networks according to the gating network; A beam prediction module is used to determine the expert output corresponding to each of the channel state features based on the target expert network, and determine the optimal beam index prediction result of the current channel according to the expert output.

8. A beam prediction device, characterized in that The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the beam prediction method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the beam prediction method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the beam prediction method according to any one of claims 1 to 6 are implemented.

Images