Beam channel map construction method and apparatus, electronic device, and storage medium

Abstract

This application relates to the field of wireless communication technology, and in particular to a method, apparatus, electronic device, and storage medium for constructing a beam channel map. The method includes: acquiring an environmental map of an application scenario and location information of base stations within the application scenario; obtaining sparse observations of the beam and channel gain values ​​at any location within the application scenario; constructing a training set using the environmental map, location information, and sparse observations as inputs, and channel gain values ​​as outputs; training a target neural network using joint loss as an indicator to construct a channel gain model that meets preset training conditions; and outputting a beam velocity channel map of the target application scenario based on the channel gain model. This solves the problem that as the parameter dimensionality of channel estimation increases, the demand for pilot resources increases, leading to a continuous increase in pilot occupancy, which in turn compresses time-frequency resources used for data transmission, reducing system spectral efficiency and consequently lowering overall system throughput and data transmission rate.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, and in particular to a method, apparatus, electronic device and storage medium for constructing beam channel maps. Background Technology

[0002] In related technologies, channel estimation is a crucial step in wireless communication systems, typically requiring the acquisition of channel state information based on pilot signals. In single-antenna scenarios, the channel parameters have a lower dimensionality, allowing the receiver to accurately estimate the channel gain using a small amount of pilot signals, thereby acquiring the channel state information.

[0003] However, as the number of base station antennas continues to increase, the channel expands from a scalar form to a high-dimensional matrix form. The dimensionality of the channel parameters that need to be estimated increases significantly, which increases the demand for pilot resources during the channel estimation process. The proportion of pilot occupancy continues to rise, which in turn compresses the time and frequency resources that can be used for data transmission, resulting in a significant decrease in system spectral efficiency. This affects the overall system throughput and data transmission rate, and reduces network capacity and multi-user concurrent service capabilities. Summary of the Invention

[0004] This application provides a beam channel map construction method, apparatus, electronic device, and storage medium to solve the problem in related technologies where, due to the increase in the parameter dimension of channel estimation, the demand for pilot resources increases, leading to a continuous increase in the proportion of pilot occupancy, which in turn compresses the time and frequency resources used for data information transmission, causing a decrease in system spectral efficiency, thereby reducing the overall system throughput and data transmission rate.

[0005] The first aspect of this application provides a beam channel map construction method, comprising the following steps: collecting an environmental map of at least one application scenario and location information of base stations in the application scenario; obtaining sparse observations of at least one beam and obtaining channel gain values ​​at any location in the application scenario; constructing a training set with the environmental map, location information, and sparse observations as inputs and the channel gain value as output; training a target neural network with joint loss as an indicator to construct a channel gain model that meets preset training conditions; and outputting a beam velocity channel map of the target application scenario based on the channel gain model.

[0006] Using the above techniques, a training set is constructed based on environmental maps, base station location information, and sparse observations as inputs, with channel gain as the output. This training set is then used to train a target neural network to build a channel gain model. This model can effectively capture the complex nonlinear relationship between environmental geometric features and signal propagation, thereby achieving high-precision channel gain prediction without requiring precise physical parameters. This significantly reduces pilot overhead and channel feedback burden. Simultaneously, sparse observations are used to achieve full-scene interpolation and reconstruction, enhancing adaptability to dynamic environments (such as obstacle movement and scene changes). This provides high-fidelity, low-latency channel modeling support for coverage optimization, resource allocation, base station deployment, and positioning services in wireless communication networks.

[0007] Optionally, in one embodiment of this application, outputting a wave velocity channel map of the target application scenario according to the channel gain model includes: obtaining the actual environment map of the target application scenario and the coordinates of the target base station, and generating a two-dimensional grayscale image based on the actual environment map and coordinates; inputting the two-dimensional grayscale image into the channel gain model, and combining it with sparse observations to output a wave velocity channel map of each location in the target application scenario under different discrete Fourier transform codewords.

[0008] By applying the above technical means to the channel gain model in real-world scenarios, a wave velocity channel map of each location in the real-world scenario under different discrete Fourier transform codewords can be output. This can intuitively reflect the channel distribution characteristics of different beams at various spatial locations, thereby assisting the base station in beam selection and codeword optimization. Without the need for large-scale pilot measurements or exhaustive calculations, a subset of high-gain codewords can be quickly selected, thereby reducing pilot overhead, improving system spectral efficiency and data transmission performance, and enhancing the system's adaptability and real-time performance in complex environments.

[0009] Optionally, in one embodiment of this application, before constructing a channel gain model that meets preset training conditions, the method further includes: extracting feature codes of environmental map and location information using the U-shaped network structure of the target neural network; performing global interaction on the feature codes using the Transformer structure of the target neural network to obtain the contextual association of the application scenario; extracting redundant and complementary information of sparse observations using the cross-attention mechanism of the target neural network; and determining the channel gain model based on the contextual association, redundant information, and complementary information.

[0010] By employing the above technical means, and based on the U-shaped network structure of the target neural network and the cross-attention mechanism of the Transformer structure, a channel gain model can be constructed. This can enhance the model's ability to express the channel propagation characteristics, achieve high-precision prediction of channel gain under different spatial locations and different codewords, reduce dependence on pilot signals, improve the model's generalization ability and prediction accuracy, and provide reliable support for beam selection and communication performance optimization.

[0011] Optionally, in one embodiment of this application, obtaining the channel gain value at any location in the application scenario includes: determining the codeword of the beam velocity according to a preset discrete Fourier transform codebook; calculating the propagation path and received power of the corresponding beam in the application scenario according to the codeword; and determining the channel gain value according to the propagation path and received power.

[0012] Using the above techniques, the beam direction and antenna weighting parameters corresponding to each codeword are determined based on a preset discrete Fourier transform codebook. Combined with the channel gain values ​​of different codewords at various spatial locations in the application scenario, channel gain sample data covering multiple beams and multiple locations is constructed. This data can be used to train the channel gain model, enabling the model to learn the mapping relationship between different codewords, environmental information, and location information. This improves the model's ability to represent channel characteristics and its prediction accuracy, providing reliable data support for subsequent fast channel estimation and beam selection.

[0013] Optionally, in one embodiment of this application, the formula for calculating the joint loss is: , in, Indicates the total joint loss, This represents the loss due to the mean square error between the constrained predicted value and the true value of the ray tracing. This represents the loss function used to fit propagation textures at different spatial frequencies using Laplacian pyramid decomposition at multiple scales. This represents the loss obtained by combining the Sobel gradient operator to enhance gradient constraints at the edges of the main lobe and abrupt changes in shadow. This represents the loss in improving the visual fidelity of the predicted image from a structural perception perspective. The weight parameters represent the loss.

[0014] By employing the above techniques, a joint loss function is constructed based on the loss between predicted and true values, multi-scale fitting loss, gradient constraint loss, and visual fidelity loss. This function can constrain the model from multiple dimensions, such as global numerical accuracy, local detail consistency, and structural similarity. As a result, the model can not only accurately fit the overall distribution of channel gain during training, but also maintain the continuity of spatial structure and edge variation characteristics. This improves the accuracy and stability of channel gain prediction results, enhances the model's adaptability to complex propagation environments, and increases the realism and usability of the generated channel gain map.

[0015] Optionally, in one embodiment of this application, the calculation formula for the channel gain model is: , in, Represents the beam velocity channel map. Represents the channel gain model. The parameters representing the target neural network, This represents the number of pixels in the environment map. Indicates the first The coordinates of each user in the environment map. Represents the precoding vector. For environmental maps.

[0016] By using the above techniques, the input environmental map, location information and sparse observations can be modeled and inferred based on the calculation formula of the channel gain model. This allows the model output to gradually approximate the real channel gain distribution, thereby achieving high-precision fitting of channel characteristics under complex propagation environments while ensuring computational efficiency.

[0017] A second aspect of this application provides a beam channel map construction apparatus, comprising: a data acquisition module for acquiring an environmental map of at least one application scenario and location information of a base station in the application scenario; an acquisition module for acquiring sparse observations of at least one beam and acquiring channel gain values ​​at any location in the application scenario; and a construction module for constructing a training set with the environmental map, location information, and sparse observations as input and the channel gain value as output, training a target neural network with joint loss as an indicator to construct a channel gain model that meets preset training conditions, and outputting a beam velocity channel map of the target application scenario based on the channel gain model.

[0018] Optionally, in one embodiment of this application, the construction module includes: a generation unit, used to obtain the actual environment map of the target application scenario and the coordinates of the target base station, and generate a two-dimensional grayscale image based on the actual environment map and coordinates; and an output unit, used to input the two-dimensional grayscale image into the channel gain model and combine it with sparse observations to output a wave velocity channel map of each location in the target application scenario under different discrete Fourier transform codewords.

[0019] Optionally, in one embodiment of this application, the beam channel map construction device further includes: a first extraction module, used to extract feature codes of environmental map and location information using the U-shaped network structure of the target neural network; a second extraction module, used to perform global interaction on the feature codes using the Transformer structure of the target neural network to obtain the contextual association of the application scenario; a third extraction module, used to extract redundant and complementary information of sparse observations using the cross-attention mechanism of the target neural network; and a determination module, used to determine the channel gain model based on the contextual association, redundant information, and complementary information.

[0020] Optionally, in one embodiment of this application, the acquisition module includes: a determining unit, configured to determine the codeword of the wave velocity according to a preset discrete Fourier transform codebook; a calculation unit, configured to calculate the propagation path and received power of the corresponding beam in the application scenario according to the codeword; and an acquisition unit, configured to determine the channel gain value according to the propagation path and received power.

[0021] Optionally, in one embodiment of this application, the formula for calculating the joint loss is: , in, Indicates the total joint loss, This represents the loss due to the mean square error between the constrained predicted value and the true value of the ray tracing. This represents the loss function used to fit propagation textures at different spatial frequencies using Laplacian pyramid decomposition at multiple scales. This represents the loss obtained by combining the Sobel gradient operator to enhance gradient constraints at the edges of the main lobe and abrupt changes in shadow. This represents the loss in improving the visual fidelity of the predicted image from a structural perception perspective. The weight parameters represent the loss.

[0022] Optionally, in one embodiment of this application, the calculation formula for the channel gain model is: , in, Represents the beam velocity channel map. Represents the channel gain model. The parameters representing the target neural network, This represents the number of pixels in the environment map. Indicates the first The coordinates of each user in the environment map. Represents the precoding vector. For environmental maps.

[0023] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the beam channel map construction method as described in the above embodiments.

[0024] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the beam channel map construction method described above.

[0025] A fifth aspect of this application provides a computer program product, including a computer program that, when executed, is used to implement the beam channel map construction method described above.

[0026] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0027] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a beam channel map construction method provided according to an embodiment of this application; Figure 2 This is a schematic diagram illustrating the principle of a target neural network according to an embodiment of this application; Figure 3 This is a schematic diagram showing the performance comparison of performance parameters as a percentage of sparse observations in one embodiment of this application. Figure 4 This is a schematic diagram showing the performance comparison of the performance parameters of one embodiment of this application with the number of beams in the observed values; Figure 5 This is a block diagram of a beam channel map construction device according to an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application.

[0028] Figure label: 10-Beam channel map construction device; 100-Acquisition module, 200-Acquisition module, 300-Construction module; 601-Memory, 602-Processor, 603-Communication interface. Detailed Implementation

[0029] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0030] The beam channel map construction method, apparatus, electronic device, and storage medium of this application are described below with reference to the accompanying drawings. Addressing the technical problem mentioned in the background art, where the increasing dimensionality of channel estimation parameters leads to a greater demand for pilot resources, resulting in a continuously increasing pilot occupancy ratio, and consequently compressing time-frequency resources used for data transmission, thus reducing system spectral efficiency and lowering overall system throughput and data transmission rate, this application provides a beam channel map construction method aimed at constructing a high-precision channel gain prediction model to achieve rapid acquisition of beam velocity channel maps under different Fourier transform codewords, thereby avoiding reliance on large-scale pilot measurements or highly complex calculations.

[0031] Specifically, Figure 1 This is a flowchart illustrating a beam channel map construction method provided in an embodiment of this application.

[0032] like Figure 1 As shown, the beam channel map construction method includes the following steps: In step S101, an environmental map of at least one application scenario and the location information of base stations in the application scenario are collected.

[0033] Methods for acquiring environmental maps include, but are not limited to: scanning the surrounding environment with LiDAR to obtain 3D point cloud data and using image information acquired by cameras for visual mapping; or obtaining location information through GPS (Global Positioning System) and combining it with inertial measurement unit data for fusion processing to create an environmental map of the application scenario. Environmental maps include, but are not limited to, environmental information affecting channel propagation such as building distribution and building material characteristics. They can describe phenomena such as reflection, obstruction, and diffraction that may occur in the signal propagation path. These propagation mechanisms affect the channel's path loss, multipath effect, and signal fading characteristics, thus allowing for a direct and convenient expression and analysis of channel propagation characteristics using environmental maps. For example, when a signal encounters the surface of an obstacle such as a building, reflection occurs, resulting in multipath propagation and causing superposition enhancement or destructive fading of the received signal; when the signal propagation path is blocked by an obstacle, an obstruction effect occurs, leading to attenuation or even interruption of the direct path, significantly reducing signal strength.

[0034] Methods for collecting location information include, but are not limited to: obtaining base station location information via GPS; or obtaining the deployment location and related parameter information of base stations through network management systems or base station configuration databases. Base station location information can be used to determine the starting point of signal propagation, thus providing basic data support for signal propagation path analysis and channel modeling.

[0035] For example, in this embodiment of the application, an environmental map E of the application scenario can be obtained via GPS, where locations with buildings are set to 1 and locations without buildings are set to 0, thereby generating a corresponding binary grayscale image. Simultaneously, the location of the base station is set to 1, and other points are set to 0, to represent the spatial distribution of the base station. Furthermore, in this embodiment of the application, software such as Blender or Sionna can be used to simulate the application scenario and the base station, constructing a corresponding three-dimensional simulation scenario and simulating the signal propagation process, thereby providing basic data support for subsequent analysis of channel gain and beam characteristics.

[0036] In step S102, sparse observations of at least one beam are obtained, and channel gain values ​​at any location in the application scenario are obtained.

[0037] Sparse observations refer to observational data obtained by measuring only some key locations or directions in an overall application scenario, representing partial characteristic information of the channel. They can be acquired through sampling of the application scenario by sensors, or by transmitting pilot signals to a portion of the beam and measuring them at the receiving end. Sparse observations only require observation of a small number of sampling points, reducing data acquisition costs and system overhead without requiring full measurement.

[0038] Channel gain is a quantitative representation of the amplitude change of a signal during propagation between the transmitter and receiver, reflecting the attenuation or enhancement effect of the channel on the signal. It can be determined by identifying the starting position of signal propagation based on the base station's location information and combining this with building distribution information from an environmental map. Software such as Sionna is used to simulate and calculate the reflection, obstruction, and diffraction paths generated during signal propagation, and multipath signals are superimposed to determine the channel gain value at the target location. While this method can accurately calculate the channel gain value, it involves significant computational complexity and poor real-time performance. Therefore, this application constructs a channel gain model to replace the highly complex calculation process, enabling rapid prediction of channel gain, thereby reducing system computational overhead and improving real-time performance.

[0039] As a specific example, in one embodiment of this application, obtaining the channel gain value at any location in the application scenario includes: determining the codeword of the beam velocity according to a preset discrete Fourier transform codebook; calculating the propagation path and received power of the corresponding beam in the application scenario according to the codeword; and determining the channel gain value according to the propagation path and received power.

[0040] The preset Discrete Fourier Transform codebook refers to a set of weighted vectors generated based on the Discrete Fourier Transform according to the antenna array structure. Each weighted vector is used to perform phase control on the antenna array, thereby forming discrete beams covering different spatial directions. Each codeword corresponds to a beam in a specific direction, which is used for beam scanning and beam selection.

[0041] According to the embodiments of this application, the codeword can be determined based on the preset discrete Fourier transform codebook, thereby determining the transmission direction and antenna weighting parameters of the corresponding beam. Based on the base station location information and environmental map, ray tracing technology is used to simulate the signal propagation path of the beam in the application scenario, and the path loss and phase change of each propagation path are calculated. The multipath signals are superimposed to obtain the received power of the corresponding beam at the target location, thereby determining the channel gain value.

[0042] As a concrete example, the predefined discrete Fourier transform codebook can be written as: , in, This indicates a preset discrete Fourier transform codebook. This indicates the number of antennas at the base station.

[0043] In this embodiment of the application, after setting up the antenna array of the base station, different codewords of the DFT (Discrete Fourier Transform) codebook can be used as the precoding vectors of the base station to generate channel gain values ​​at different locations in the entire application scenario, so as to construct channel gain sample data for training the channel gain model and realize rapid prediction and optimization of channel gain.

[0044] In step S103, a training set is constructed with environmental map, location information and sparse observation values ​​as inputs and channel gain value as output. The target neural network is trained with joint loss as an indicator to construct a channel gain model that meets preset training conditions, and the wave speed channel map of the target application scenario is output based on the channel gain model.

[0045] The target neural network may include, but is not limited to, an input layer, a feature encoding layer, a cross-attention fusion layer, and an output layer. The input layer is used to receive environmental map information and channel gain values, the feature encoding layer is used to extract input feature representations, the cross-attention fusion layer is used to realize the interaction and fusion of features of different modes, and the output layer is used to output the channel gain or beam selection results of the actual application scenario.

[0046] Joint loss can include, but is not limited to, channel gain error loss, beam classification loss, or position consistency loss. By optimizing different task objectives through joint loss, the deviation between the output of the channel gain model and the true label can be effectively measured, so as to coordinate constraints on different tasks and improve the accuracy and robustness of the target neural network in channel gain prediction.

[0047] Preset training conditions can be set when the joint loss reaches a set threshold or when the model's error on the validation set is below a set range. These conditions can be set according to the actual situation.

[0048] The embodiments of this application can utilize a target neural network to construct a channel gain model, thereby learning features from the input environmental information and channel gain values ​​to achieve rapid prediction of channel gain in actual application scenarios without having to perform exhaustive calculations point by point for the entire application scenario. This significantly reduces computational complexity and improves the real-time performance of the wireless communication system.

[0049] As one possible implementation, firstly, this application embodiment considers the channel from the base station to the user, assuming the number of antennas of the base station is... Assume the first The channel between a user and a base station can be represented as follows: At a specific moment, the base station sends a signal to the first... An instance of a user sending a signal transmission can be represented by the following formula: , in, For precoded vectors, For information symbols, It is Gaussian white noise with a mean of 0 and a variance of . Assume the first The coordinates of a user in the environment map can be represented as: Then the user's gain value can be expressed as Further consider the set of all possible locations within the entire environment map. Then the wave velocity channel map can be represented as: , in, Represents the beam velocity channel map. The method for obtaining the beam velocity channel map can be an interpolation method based on a mathematical model, or it can be a machine learning method, a channel gain model, or other similar methods. This is an environmental diagram.

[0050] In one embodiment of this application, the entire process is for training a suitable To generate high-precision wave velocity channel maps under different codewords.

[0051] The goal of constructing wave velocity channel maps under different codewords can be summarized as follows: by taking sparse observations, environmental maps, and base station location information as inputs, the target neural network can output results that are as close as possible to the channel gain values ​​obtained through ray tracing, thereby obtaining a channel gain model.

[0052] In actual implementation, the formula for calculating the channel gain model can be expressed as: , in, Represents the beam velocity channel map. Represents the channel gain model. The parameters representing the target neural network, This represents the number of pixels in the environment map. Indicates the first The coordinates of each user in the environment map. Represents the precoding vector. For environmental maps.

[0053] For example, in one embodiment of this application, before constructing a channel gain model that meets preset training conditions, the method further includes: extracting feature codes of environmental map and location information using the U-shaped network structure of the target neural network; performing global interaction on the feature codes using the Transformer structure of the target neural network to obtain the contextual association of the application scenario; extracting redundant and complementary information of sparse observations using the cross-attention mechanism of the target neural network; and determining the channel gain model based on the contextual association, redundant information, and complementary information.

[0054] Feature encoding refers to the process of mapping the environmental map of the application scenario, the location information of the base station, and sparse observations into a high-dimensional feature representation through a target neural network to extract key semantic information. Context association refers to introducing spatial structure information and neighborhood positional relationships into the feature encoding process to establish associations between different spatial locations in order to reflect the overall structural characteristics of the signal propagation environment. Redundant information in sparse observations refers to observation data that are repetitive or highly correlated between different observation locations or different beam directions, while complementary information refers to the non-overlapping information provided by different observations. The two are used together to improve the ability to express and reconstruct the complete channel characteristics.

[0055] like Figure 2 As shown, embodiments of this application can extract spatial features from the input using a U-shaped network structure, Transformer structure, cross-attention mechanism, etc., of the target neural network; the specific implementation process is as follows: First, this application embodiment performs preliminary extraction of spatial features and multi-scale perception based on the backbone architecture of the target neural network UNet (U-shaped network), and transforms the binarized environment map. Location information of base stations As initial feature inputs, during the encoding stage, the target neural network performs nonlinear mapping on physical information such as building outlines and base station relative coordinates in the environment through cascaded residual convolutional blocks, compressing the complex spatial topology into a high-dimensional latent space to generate deep feature vectors characterizing the electromagnetic propagation environment. During the decoding and reconstruction stage, this embodiment utilizes symmetrical cascaded upsampling operations to restore image resolution and directly fuses the shallow spatial details captured by the encoder with the semantic features in the decoder through cross-layer skip connections. This ensures that the channel gain model can still accurately preserve the geometric fidelity of building edges and shadow fading areas when processing environmental features after significant downsampling, laying a solid spatial feature foundation for subsequent high-precision pixel-level prediction.

[0056] To address the problem of long-distance, nonlocal channel dependency modeling caused by building reflections and diffraction in complex urban environments, this application introduces a Transformer architecture in the latent space to capture the contextual relationships across the entire scene. The channel gain model serializes the latent features extracted by the CNN (Convolutional Neural Network) in the UNet framework into multiple linear embedding vectors, which are then input into a Transformer module containing several hidden layers. MHSA (Multi-Head Self-Attention) is used to calculate the interaction weights between any two points in the space, thereby learning the global propagation patterns of electromagnetic waves under specific environmental layouts. This process achieves deep feature evolution through a combination of LN (Layer Normalization) and MLP (Multi-Layer Perceptron) residuals. The inter-layer computation logic can be formally defined as follows: , , in, For the first The output of the layer, This is the intermediate vector output by the multi-head attention mechanism. This design can effectively overcome the limitation of the receptive field in traditional convolutional neural networks, enabling the channel gain model to accurately predict the channel gain intensity of each pixel in the non-line-of-sight region based on the position information of the far-end reflector.

[0057] Furthermore, addressing the beam propagation differences caused by different precoding vectors in a multi-antenna system, this embodiment employs a multimodal cross-attention mechanism to achieve deep complementarity and collaborative enhancement of information between beams. In this mechanism, firstly, an index vector is embedded into the observation values; then, the features of the target predicted beam are used as a query vector, while real-time observation features of adjacent or non-adjacent auxiliary beams are extracted as key and value vectors. By calculating the similarity weights between the target beam and auxiliary beams in spatial coverage, redundant and complementary information is adaptively extracted from the multimodal observation data. Finally, the total features are output, which can be represented as: , in, Features extracted from multiple beam observations, where Represents the target beam characteristic. and Representing the first The key vector and value vector of each auxiliary beam. The scaling factor is used. This process not only utilizes the local smoothing characteristics of adjacent beams to fill the coverage gaps of the target beam under sparse sampling, but also enhances the prediction robustness of the channel gain model under environmental changes or positioning error interference through cross-beam information transfer. As a result, the predicted channel gain map can simultaneously reflect the joint gain effect brought about by spatial location, environmental contours, and different precoding vectors.

[0058] In order to guide the above network parameters to converge to the physical propagation characteristics, the embodiments of this application adopt a carefully designed joint loss function to guide the iterative optimization process of the channel gain model. This function integrates four dimensions: pixel accuracy, multi-scale detail, structural consistency and edge sharpness.

[0059] Optionally, in one embodiment of this application, the formula for calculating the joint loss can be expressed as: ， in, The loss is used to constrain the mean square error between the predicted value and the ray-traced true value, ensuring the accuracy of the basic prediction. The loss function utilizes Laplacian pyramid decomposition to perform multi-scale fitting of propagation textures at different spatial frequencies; The loss function, combined with the Sobel gradient operator, strengthens the gradient constraints at the edges of the main lobe and abrupt changes in shadow, resulting in sharper edge prediction; while The loss function improves the visual fidelity of the predicted image from a structural perception perspective. The weight parameters of the loss function... This is achieved through adjustments during the training process. Specifically, in this embodiment, the target neural network is trained by minimizing the joint loss. The resulting channel gain model can output a wave velocity channel map with real-time accuracy superior to existing deep learning schemes, while having a significantly lower computational cost than ray tracing algorithms.

[0060] In actual execution, the wave velocity channel map of the target application scenario is output according to the channel gain model, including: obtaining the actual environment map of the target application scenario and the coordinates of the target base station, and generating a two-dimensional grayscale map based on the actual environment map and coordinates; inputting the two-dimensional grayscale map into the channel gain model, and combining it with sparse observations, to output the wave velocity channel map of each location in the target application scenario under different discrete Fourier transform codewords.

[0061] Specifically, firstly, the embodiments of this application can collect the actual geographical environment information and base station deployment parameters of the target application scenario, and use rasterization processing technology to convert the building topology into a corresponding two-dimensional grayscale environment map. The physical location coordinates of the base station are mapped to a specific pixel value of 1, forming a multidimensional input tensor consistent with the format used in the training phase. To address dynamic environmental changes or model generalization errors, this embodiment introduces real-time sparse observations during the prediction process. As a calibration aid, among which For the first The observation results of each beam are used. By inputting these measured data from low-overhead sensors or mobile terminals into the trained channel gain model, the cross-attention mechanism within the channel gain model is activated, thereby using the local features of the observation points to make real-time corrections to the prediction distribution of the entire scene. Subsequently, the trained channel gain model processes the spatial and observation features of the input in parallel through its built-in hybrid convolution and Transformer inference structure, and performs decoding on each precoded vector in the DFT codebook. Perform pixel-level inference calculations. The channel gain model utilizes prior knowledge of electromagnetic wave propagation learned during the training phase to automatically complete the gain values ​​for building obstruction areas and multipath reflection areas, ultimately generating a wave velocity channel map reflecting the complete coverage characteristics of the target scene under different codewords. .

[0062] Through the above process, the embodiments of this application can not only significantly reduce the huge channel estimation overhead required by the relevant beam scanning scheme, but also provide key decision-making basis for adaptive beam selection and spectrum resource allocation on the base station side by outputting a high-fidelity beam velocity channel map.

[0063] Combination Figure 3 , Figure 4 The following examples illustrate the performance of the beam channel map construction method of this application.

[0064] In the embodiments of this application, the simulation parameters are set as follows: the number of base station antennas is... The dataset covers multiple cities, including Beijing and Shanghai. After generating XML files using Blender, it was imported into Sionna. Channel gain values ​​were generated using ray tracing technology in Sionna. The carrier frequency used was 3GHz, and the resolution of the environmental map was [resolution missing]. Each pixel represents an actual 1m². For example... Figure 3As shown, compared with UNet, RadioWNet, and RME-GAN methods, the beam channel map construction method of this application shows a decrease in MSE (Mean Squared Error) and NMSE (Normalized Mean Squared Error), while the PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity Index Measure) are increased, demonstrating the superior performance of the beam channel map construction method. Meanwhile, as... Figure 4 As shown, the performance of the beam channel map construction method improves with the increase of non-target beams used, which proves that when using non-target beam observation, the cross-attention complementary mechanism in this application can play a gain role.

[0065] According to the beam channel map construction method proposed in this application, an environmental map, location information, and sparse observations are used as inputs, and a training set is constructed with channel gain as the output. The joint loss is used as an indicator to train a target neural network to obtain a channel gain model, which outputs a beam velocity channel map for the target application scenario. This effectively realizes the mapping relationship from the environmental map, base station location information, and sparse observations to the channel gain value under any DFT codeword, thereby assisting the base station in beam selection and beamforming optimization. Especially in scenarios with large pilot overhead, it can reduce the complexity of channel estimation while quickly filtering high-gain codeword subsets through model prediction, thereby significantly reducing pilot resource consumption, improving system spectral efficiency, and having a positive promoting effect on green and low-power communication.

[0066] Next, the beam channel map construction apparatus proposed according to the embodiments of this application is described with reference to the accompanying drawings.

[0067] Figure 5 This is a block diagram of a beam channel map construction device according to an embodiment of this application.

[0068] like Figure 5 As shown, the beam channel map construction device 10 includes: a data acquisition module 100, an acquisition module 200, and a construction module 300.

[0069] The acquisition module 100 is used to acquire an environmental map of at least one application scenario and the location information of base stations in the application scenario.

[0070] The acquisition module 200 is used to acquire sparse observations of at least one beam and acquire channel gain values ​​at any location in the application scenario.

[0071] The construction module 300 is used to construct a training set with environmental map, location information and sparse observation values ​​as input and channel gain value as output, and to train the target neural network with joint loss as the indicator, so as to construct a channel gain model that meets the preset training conditions, and to output the wave speed channel map of the target application scenario based on the channel gain model.

[0072] Optionally, in one embodiment of this application, the construction module 300 includes a generation unit and an output unit.

[0073] The generation unit is used to obtain the actual environment map of the target application scenario and the coordinates of the target base station, and generate a two-dimensional grayscale image based on the actual environment map and coordinates.

[0074] The output unit is used to input a two-dimensional grayscale image into the channel gain model and combine it with sparse observations to output a wave velocity channel map of each location in the target application scenario under different discrete Fourier transform codewords.

[0075] Optionally, in one embodiment of this application, the beam channel map construction device 10 further includes: a first extraction module, a second extraction module, a third extraction module, and a determination module.

[0076] The first extraction module is used to extract feature codes of environmental maps and location information using the U-shaped network structure of the target neural network.

[0077] The second extraction module is used to perform global interaction on feature encoding using the Transformer structure of the target neural network to obtain the contextual association of the application scenario.

[0078] The third extraction module is used to extract redundant and complementary information from sparse observations using the cross-attention mechanism of the target neural network.

[0079] The determination module is used to determine the channel gain model based on contextual association, redundancy information, and complementary information.

[0080] Optionally, in one embodiment of this application, the acquisition module 200 includes: a determining unit, a calculation unit, and an acquisition unit.

[0081] The determining unit is used to determine the codeword for wave velocity based on a preset discrete Fourier transform codebook.

[0082] The calculation unit is used to calculate the propagation path and received power of the corresponding beam in the application scenario based on the codeword.

[0083] The acquisition unit is used to determine the channel gain value based on the propagation path and the received power.

[0084] Optionally, in one embodiment of this application, the formula for calculating the joint loss is: , in, Indicates the total joint loss, This represents the loss due to the mean square error between the constrained predicted value and the true value of the ray tracing. This represents the loss function used to fit propagation textures at different spatial frequencies using Laplacian pyramid decomposition at multiple scales. This represents the loss obtained by combining the Sobel gradient operator to enhance gradient constraints at the edges of the main lobe and abrupt changes in shadow. This represents the loss in improving the visual fidelity of the predicted image from a structural perception perspective. The weight parameters represent the loss.

[0085] Optionally, in one embodiment of this application, the calculation formula for the channel gain model is: , in, Represents the beam velocity channel map. Represents the channel gain model. The parameters representing the target neural network, This represents the number of pixels in the environment map. Indicates the first The coordinates of each user in the environment map. Represents the precoding vector. For environmental maps.

[0086] It should be noted that the foregoing explanation of the beam channel map construction method embodiment also applies to the beam channel map construction device of this embodiment, and will not be repeated here.

[0087] According to the beam channel map construction device proposed in this application, an environmental map, location information, and sparse observations are used as inputs, and channel gain is used as the output to construct a training set. The joint loss is used as an indicator to train a target neural network to obtain a channel gain model, which outputs a beam velocity channel map of the target application scenario. This effectively realizes the mapping relationship from the environmental map, base station location information, and sparse observations to the channel gain value under any DFT codeword, thereby assisting the base station in beam selection and beamforming optimization. Especially in scenarios with large pilot overhead, it can reduce the complexity of channel estimation while quickly screening a subset of high-gain codewords through model prediction, thereby significantly reducing pilot resource consumption, improving system spectral efficiency, and having a positive promoting effect on green and low-power communication.

[0088] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 601, the processor 602, and the computer program stored on the memory 601 and capable of running on the processor 602.

[0089] When the processor 602 executes the program, it implements the beam channel map construction method provided in the above embodiments.

[0090] Furthermore, electronic devices also include: Communication interface 603 is used for communication between memory 601 and processor 602.

[0091] The memory 601 is used to store computer programs that can run on the processor 602.

[0092] The memory 601 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0093] If the memory 601, processor 602, and communication interface 603 are implemented independently, then the communication interface 603, memory 601, and processor 602 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0094] Optionally, in a specific implementation, if the memory 601, processor 602, and communication interface 603 are integrated on a single chip, then the memory 601, processor 602, and communication interface 603 can communicate with each other through an internal interface.

[0095] The processor 602 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0096] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the beam channel map construction method described above.

[0097] This application also provides a computer program product, including a computer program that can run computer instructions. When the computer instructions are executed by a processor, they implement the beam channel map construction method provided in this application.

[0098] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0099] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0100] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0101] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0102] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0103] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0104] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0105] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method of beam channel map construction, the method comprising: Includes the following steps: Collect an environmental map of at least one application scenario and the location information of base stations in the application scenario; Obtain sparse observations of at least one beam and obtain the channel gain value at any location in the application scenario; A training set is constructed using the environmental map, the location information, and the sparse observations as inputs, and the channel gain as the output. The target neural network is trained using joint loss as an indicator to construct a channel gain model that meets preset training conditions. Based on the channel gain model, a wave speed channel map of the target application scenario is output.

2. The method according to claim 1, characterized in that, The step of outputting the wave velocity channel map of the target application scenario based on the channel gain model includes: Obtain the actual environment map of the target application scenario and the coordinates of the target base station, and generate a two-dimensional grayscale image based on the actual environment map and the coordinates; The two-dimensional grayscale image is input into the channel gain model and combined with the sparse observations to output the wave velocity channel map of each location in the target application scenario under different discrete Fourier transform codewords.

3. The method of claim 1, wherein, Before constructing the channel gain model that meets the preset training conditions, the following steps are also included: The U-shaped network structure of the target neural network is used to extract feature codes of the environmental map and the location information; The feature encoding is globally interacted with using the Transformer structure of the target neural network to obtain the contextual association of the application scenario; The cross-attention mechanism of the target neural network is used to extract redundant and complementary information from the sparse observations; The channel gain model is determined based on the context association, the redundancy information, and the complementary information.

4. The method of claim 1, wherein, Obtaining the channel gain value at any location in the application scenario includes: The codeword for the wave velocity is determined according to a preset discrete Fourier transform codebook; Calculate the propagation path and received power of the corresponding beam in the application scenario based on the codeword; The channel gain value is determined based on the propagation path and the received power.

5. The method of claim 4, wherein, The formula for calculating the joint loss is: ， in, Indicates the total joint loss, This represents the loss due to the mean square error between the constrained predicted value and the true value of the ray tracing. This represents the loss function used to fit propagation textures at different spatial frequencies using Laplacian pyramid decomposition at multiple scales. This represents the loss obtained by combining the Sobel gradient operator to enhance gradient constraints at the edges of the main lobe and abrupt changes in shadow. This represents the loss in improving the visual fidelity of the predicted image from a structural perception perspective. The weight parameters represent the loss.

6. The method according to claim 1, characterized in that, The calculation formula for the channel gain model is as follows: ， in, Represents the beam velocity channel map. Represents the channel gain model. The parameters representing the target neural network, This represents the number of pixels in the environment map. Indicates the first The coordinates of each user in the environment map. Represents the precoding vector. For environmental maps.

7. A beam channel map construction device, characterized in that, include: The data acquisition module is used to acquire an environmental map of at least one application scenario and the location information of base stations in the application scenario; The acquisition module is used to acquire sparse observations of at least one beam and acquire the channel gain value at any location in the application scenario. The construction module is used to construct a training set with the environment map, the location information and the sparse observation values ​​as inputs and the channel gain value as outputs, and to train the target neural network with joint loss as an indicator, so as to construct a channel gain model that meets preset training conditions, and to output a wave speed channel map of the target application scenario based on the channel gain model.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, the processor executing the program to implement the beam channel map construction method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the beam channel map construction method as described in any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed for implementing the beam channel map construction method according to any one of claims 1-6. The computer program is executed for implementing the beam channel map construction method according to any one of claims 1-6.

Images