A channel parameter prediction method, system, computer device and storage medium

Abstract

The application provides a channel parameter prediction method, system, computer device and storage medium, and belongs to the field of wireless communication, and comprises the following steps: obtaining an original satellite image of a wireless channel transceiver to be predicted; determining a predicted value of a channel parameter from the original satellite image by using a channel parameter prediction model trained by element learning; the channel parameter prediction model is updated by part parameters in advance according to a labeled sample of the wireless channel transceiver to be predicted; extracting a scatterer feature vector from the original satellite image; extracting an embedded feature vector from original system parameters of the wireless channel transceiver to be predicted; fusing the scatterer feature vector and the embedded feature vector to obtain a fused feature representation; and determining the predicted value of the channel parameter from the fused feature representation. The method uses the channel parameter prediction model trained by element learning in advance, and updates part parameters by using the labeled sample of the target scene, so that the model can quickly adapt to a new wireless propagation environment.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication, and specifically relates to a channel parameter prediction method, system, computer device, and storage medium. Background Technology

[0002] A wireless channel specifically refers to the electromagnetic wave propagation medium from the transmitter (Tx) to the receiver (Rx). Wireless channel modeling is the process of exploring and characterizing channel characteristics in real-world environments, which is crucial for the optimization and performance evaluation of wireless networks. As one of the key parameters of a wireless channel, accurate path loss prediction directly determines the accuracy of link budget calculation, which plays a key role in optimizing network layout, rationally allocating spectrum resources, and improving system capacity. However, with the increasing complexity of electromagnetic propagation environments, the core challenge facing current fifth-generation mobile communication systems (5G) is how to effectively improve the accuracy and generalization ability of channel models in complex environments. Traditional channel modeling methods are mainly divided into two categories: statistical models and deterministic models. Statistical models establish an empirical framework for characterizing channel characteristics by fitting parameters to measured data. This framework is simple to implement and intuitive to analyze, making it more suitable for rapid characterization and application compared to complex models. Currently, mainstream standardized models include the New York University Wireless Channel Simulator (NYUSIM), the mmMAGIC channel model, geometry-based stochastic models, and the COST231 Hata model. However, these statistical models are not applicable to all real-world scenarios, and the modeling accuracy requires high quality and quantity of measurement data. In contrast, deterministic modeling employs physical simulation methods based on electromagnetic wave propagation principles such as ray tracing, offering higher accuracy compared to statistical models. However, deterministic models require extensive parameter recalibration when the position or properties of scatterers in the environment change. Furthermore, due to their reliance on precise environmental configurations, deterministic models often suffer from high modeling complexity and computational costs. Overall, traditional channel modeling methods face numerous limitations in practical applications: on the one hand, acquiring large-scale, high-precision channel measurement data is difficult and costly; on the other hand, the applicability and scalability of the models remain insufficient in complex scenarios, which significantly restricts their application and development in future wireless communication systems.

[0003] With the continuous development of artificial intelligence, its ability to effectively extract the complex mapping relationship between channel characteristics and propagation environment based on limited datasets has become increasingly important. Recent research has focused on optimizing machine learning-based channel modeling methods by selecting appropriate input parameters and network architectures to achieve accurate prediction of channel parameters, opening up new avenues for channel parameter prediction. For example, path loss prediction models based on convolutional neural networks utilize satellite imagery for path loss prediction, thus eliminating the need for environmental feature extraction. However, while such path loss prediction models perform well within specific regions, their prediction accuracy drops significantly when crossing different scenarios or environmental changes, resulting in poor generalizability. Summary of the Invention

[0004] To address the problem of poor versatility in existing channel parameter prediction methods, this invention provides a channel parameter prediction method, system, computer device, and storage medium.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A channel parameter prediction method, comprising:

[0007] Obtain raw satellite images of the transceiver ends of the wireless channel to be predicted;

[0008] The channel parameter prediction model, which is pre-trained by a meta-learning mechanism, predicts the channel parameters of the wireless transceiver to be predicted from the original satellite image. The channel parameter prediction model includes an environmental feature extraction module, a system parameter embedding module, a feature fusion module, and a regression prediction module connected in sequence. The channel parameter prediction model updates some parameters in advance based on the labeled samples of the wireless transceiver to be predicted.

[0009] The scatterer feature vector is extracted from the original satellite image using the environmental feature extraction module;

[0010] The embedded feature vector is obtained by extracting the original system parameters of the wireless channel transceiver from the system parameter embedding module.

[0011] The feature fusion module fuses the scatterer feature vector and the embedded feature vector to obtain a fused feature representation;

[0012] The channel parameters are predicted by performing regression prediction on the fused feature representation through the regression prediction module.

[0013] Optionally, in the channel parameter prediction method provided by the present invention, the channel parameter prediction model is a YOLOv6 network, and the environmental feature extraction module includes a YOLOv6 backbone network, a neck network, a detection head, RoIAlign and a pooling layer connected in sequence.

[0014] Multi-scale features are extracted from the original satellite image using the YOLOv6 backbone network; the multi-scale features are fused using the neck network to obtain a fused feature map; the fused feature map is densely predicted using the detection head to obtain scatterer candidate boxes and their corresponding confidence scores; the region of interest features are extracted using RoIAlign from the set of scatterer candidate boxes with confidence scores higher than a preset threshold and the fused feature map to obtain feature map patches corresponding to the scatterer candidate boxes; the feature map patches are then globally pooled using pooling layers to obtain scatterer feature vectors.

[0015] Optionally, the neck network is constructed based on a feature pyramid network. The channel parameter prediction method provided by this invention further includes:

[0016] A fused feature map is obtained by fusing multi-scale features through a top-down path and lateral connections.

[0017] Optionally, the feature fusion module includes a vector concatenation layer, a fully connected layer, and a ReLU activation function. The channel parameter prediction method provided by this invention also includes:

[0018] The vector concatenation layer concatenates the scatterer feature vector and the embedded feature vector to obtain a joint feature vector.

[0019] The joint feature vector is subjected to a nonlinear projection transformation through a fully connected layer to obtain the nonlinear projection result;

[0020] The nonlinear projection results are processed by the ReLU activation function to obtain the fused feature representation.

[0021] Optionally, the channel parameter prediction method provided by the present invention further includes:

[0022] Obtain multiple sets of training support datasets and training query datasets for different environments;

[0023] The model parameters of the regression prediction module are updated by gradient based on the training support dataset for each environment, resulting in the updated model parameters for each environment.

[0024] Meta-loss is calculated based on the updated model parameters for each environment and the corresponding training query dataset. The initial parameters of the YOLOv6 network are then updated using gradients based on the meta-loss to obtain the channel parameter prediction model.

[0025] Optionally, the channel parameter prediction method provided by the present invention further includes:

[0026] Obtain labeled samples of the wireless channel transceivers to be predicted; divide the labeled samples into an adaptation dataset and a non-overlapping validation dataset;

[0027] The model parameters of the regression prediction module in the channel parameter prediction model are adjusted based on the adaptation dataset to obtain the scenario-adjusted channel parameter prediction model.

[0028] When the non-overlapping verification dataset is input into the scenario-adjusted channel parameter prediction model, and the root mean square error of the predicted channel parameters output by the scenario-adjusted channel parameter prediction model meets the prediction accuracy requirements, the scenario-adjusted channel parameter prediction model is determined as the channel parameter prediction model adapted to the scenario of the wireless channel transceiver to be predicted.

[0029] Optionally, the channel parameter prediction method provided by the present invention further includes:

[0030] Obtain the original satellite image of the wireless channel transceiver and the latitude and longitude of the wireless channel transceiver to be predicted;

[0031] The relative distance between the transceivers of the wireless channel to be predicted is obtained by calculating the spherical distance based on the latitude and longitude of the transceivers.

[0032] Based on the relative distance between the transceiver and receiver of the wireless channel to be predicted, the original satellite image is cropped to obtain the preprocessed original satellite image.

[0033] The present invention also provides a channel parameter prediction system, comprising:

[0034] The data acquisition module is used to acquire raw satellite images of the wireless channel transceiver to be predicted;

[0035] The parameter prediction module is used to predict the channel parameters of the wireless transceiver terminal to be predicted from the original satellite image using a channel parameter prediction model trained in advance by a meta-learning mechanism. The channel parameter prediction model includes an environment feature extraction module, a system parameter embedding module, a feature fusion module, and a regression prediction module connected in sequence. The channel parameter prediction model is pre-updated with some parameters iteratively based on labeled samples of the wireless transceiver terminal to be predicted. The environment feature extraction module extracts scatterer feature vectors from the original satellite image. The system parameter embedding module extracts embedded feature vectors from the original system parameters of the wireless transceiver terminal to be predicted. The feature fusion module fuses the scatterer feature vectors and the embedded feature vectors to obtain a fused feature representation. The regression prediction module performs regression prediction on the fused feature representation to obtain the predicted channel parameters.

[0036] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement any of the steps in a channel parameter prediction method.

[0037] The present invention also provides a computer-readable storage medium storing a computer program that, when loaded by a processor, can execute any step of a channel parameter prediction method.

[0038] The channel parameter prediction method provided by this invention has the following beneficial effects:

[0039] This invention employs a pre-trained channel parameter prediction model using meta-learning, and rapidly iterates and updates some parameters of the model using a small number of labeled samples from the target scenario—the labeled samples of the transceivers in the wireless channel to be predicted. This allows the model to quickly adapt to new and unseen wireless propagation environments, effectively overcoming the problems of significantly decreased prediction accuracy and poor generalization ability of existing path loss prediction models when crossing scenarios or environmental changes. Simultaneously, through the collaborative work of the environmental feature extraction module, system parameter embedding module, and feature fusion module, the model can comprehensively utilize visual environmental features and physical system parameters to construct a more representative unified feature representation, providing high-quality input for the regression prediction module and further ensuring the accuracy and reliability of channel parameter prediction in different scenarios. Attached Figure Description

[0040] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is one of the schematic diagrams of a channel parameter prediction method provided in an embodiment of the present invention;

[0042] Figure 2 This is a second schematic diagram of a channel parameter prediction method provided in an embodiment of the present invention;

[0043] Figure 3 This is an example of the overall process of the prediction model provided in the embodiments of the present invention;

[0044] Figure 4 This is an example of a meta-learning process provided in an embodiment of the present invention;

[0045] Figure 5 This is an example of scatterer feature vector extraction provided in an embodiment of the present invention. Detailed Implementation

[0046] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0047] Because scatterers in the environment significantly alter the reflection and diffraction paths of wireless point signals, channel parameter prediction models based on real-world environments are receiving increasing attention. To obtain more comprehensive environmental information, most existing studies use images as input to channel prediction models, such as satellite images, elevation maps, and building maps, and employ path loss prediction models based on convolutional neural networks. These models utilize satellite images for path loss prediction, thus eliminating the need for environmental feature extraction. Furthermore, to further improve the prediction performance of convolutional neural networks, a small-area satellite image or feature map is used as input, for example, focusing on an area within 250 meters of the transceiver, and a distance grayscale image is used as auxiliary input, thereby reducing the impact of redundant information in large-area images on prediction accuracy. However, such high-precision original pixel-level images have too high a dimension. Path loss prediction models based on convolutional neural networks usually require a large amount of training data to achieve excellent performance. This results in slow network convergence and requires a lot of computing resources. Furthermore, existing predictions mainly focus on the impact of regular scatterers such as buildings on the channel propagation process, while insufficient consideration is given to the impact of irregular scatterers such as vegetation. This leads to insufficient versatility in predicting channel parameters such as path loss when facing complex environments.

[0048] To address the aforementioned shortcomings, the channel parameter prediction method provided by this invention identifies and extracts key scatterer features from raw satellite images, optimizes input features and network structure, reduces the model's dependence on large-scale training data, and achieves accurate prediction of channel parameters. Furthermore, the channel parameter prediction method provided by this invention improves the model's generalization ability under different geographical environments and scatterer distributions by incorporating a meta-learning mechanism, reducing the need for retraining or parameter calibration, and ensuring the model's prediction accuracy across different scenarios or environmental changes.

[0049] Example 1

[0050] This application provides a channel parameter prediction method, specifically as follows: Figure 1 and Figure 2 As shown, it includes the following steps:

[0051] Step 11: Obtain the original satellite image of the wireless channel transceiver to be predicted.

[0052] Step 11 includes:

[0053] Step 111: Obtain the original satellite image of the wireless channel transceiver to be predicted and the latitude and longitude of the wireless channel transceiver to be predicted.

[0054] Step 112: Calculate the spherical distance based on the latitude and longitude of the transceiver terminals of the wireless channel to be predicted, and obtain the relative distance between the transceiver terminals of the wireless channel to be predicted.

[0055] Step 113: Based on the relative distance between the transceiver and receiver of the wireless channel to be predicted, crop the original satellite image to obtain the preprocessed original satellite image.

[0056] Specifically, such as Figure 3 As shown, firstly, the original satellite image and the latitude and longitude of the transceiver are obtained, the geographic coordinates are mapped to the image pixel coordinates, and the spherical distance between the transceiver and the transceiver is calculated, as shown in formula (1):

[0057] (1)

[0058] in, For the Earth's radius, and These are the latitudes of the transmitting and receiving ends, respectively. and These are the longitude degrees of the transmitting and receiving ends, respectively. Let be the spherical distance between the transceiver and receiver. Then, based on the spherical distance or relative position of the transceiver and receiver, and the image coverage requirements during channel parameter prediction, a multi-scale cropping input is constructed to capture environmental information as comprehensively as possible while reducing image size. The cropping scales are shown in formulas (2) and (3):

[0059] (2)

[0060] (3)

[0061] in, and For a predefined scaling factor, and These are the minimum width and length of the image, respectively. and These represent the cropping width and cropping length, respectively. Then, the original satellite image is cropped based on the cropping scale to obtain the preprocessed original satellite image. .

[0062] Step 12: Obtain multiple sets of training support datasets and training query datasets for different environments.

[0063] Step 13: Update the model parameters of the regression prediction module using gradients based on the training support dataset for each environment to obtain the updated model parameters for each environment.

[0064] Step 14: Calculate the meta-loss based on the updated model parameters of each environment and the corresponding training query dataset. Update the initial parameters of the YOLOv6 network using gradients based on the meta-loss to obtain the channel parameter prediction model.

[0065] Step 15: Obtain labeled samples of the wireless channel transceivers to be predicted; divide the labeled samples into adaptation datasets and non-overlapping verification datasets.

[0066] Step 16: Adjust the model parameters of the regression prediction module in the channel parameter prediction model according to the adaptation dataset to obtain the channel parameter prediction model after scenario adjustment.

[0067] Step 17: Input the non-overlapping verification dataset into the scenario-adjusted channel parameter prediction model. When the root mean square error of the predicted channel parameters output by the scenario-adjusted channel parameter prediction model meets the prediction accuracy requirements, the scenario-adjusted channel parameter prediction model is determined as the channel parameter prediction model adapted to the scenario of the wireless channel transceiver to be predicted.

[0068] Specifically, such as Figure 4 As shown, this invention performs meta-training based on the idea of ​​meta-learning during model training. The specific meta-training process consists of an inner loop and an outer loop. Its core idea is to achieve a generalized representation of model parameters across multiple tasks through repeated optimization. The first step is the inner loop or internal task update phase, which begins by dividing the training support dataset from the training set containing satellite images and corresponding channel parameter values. and training query dataset And based on the training support dataset for each set of environments. Model training is performed separately, loss functions are calculated, and model parameters are updated once or multiple times with the goal of optimizing the loss value, thereby obtaining the environment-specific internal parameters for each set of models, as shown in Equations (4) and (5):

[0069] (4)

[0070] (5)

[0071] in, These are the initial parameters for the model's internal tasks. These are the model parameters after the (i-1)th iteration. For each meta-task, the total number of meta-learning tasks is . , For internal task learning rate, Let be the batch size of the resamples for the i-th task. To support the actual channel parameter values ​​in the dataset, These are the predicted values ​​for the channel parameters. In summary, through multiple gradient updates, the optimal intrinsic parameters are determined for the meta-learning task of each environment group. In the outer loop phase, the training query dataset is based on all environments or all tasks. Calculate the loss function and update the extrinsic parameters of the model's meta-learner to minimize the sum of the loss functions for all tasks. As shown in formula (6):

[0072] (6)

[0073] in, These are the extrinsic parameters of the meta-learner. In summary, by combining the task-inner loop and the task-outer loop, the optimal global model parameters that reflect the internal correlation between multiple meta-learning tasks are determined, namely the optimal model internal parameters and model extrinsic parameters. This enables the pre-trained model to quickly adapt to new prediction tasks in different scenarios.

[0074] Subsequently, when facing a new wireless channel transceiver environment to be predicted, samples of partially labeled channel parameters in the new scenario are first collected and a dataset is constructed. The dataset is then divided into an adaptation dataset for the new task and a non-overlapping validation dataset. The internal parameters of the pre-trained model meta-learner are first fine-tuned based on the adaptation dataset so that the model can be quickly converged and generalized using the SGD method, as shown in formula (7):

[0075] (7)

[0076] in, These are the internal task parameters of the model before a certain iteration update in the meta-adaptation phase. This refers to the internal task parameters of the model after this iteration in the meta-adaptation phase. These internal task parameters are continuously updated until convergence. This is the dataset used for fine-tuning the model during the meta-adaptation phase. The calculation and formula (5) are not detailed in this application. The model extrinsic parameters determined in the previous meta-learning stage do not need to be adjusted at this time. Then, the root mean square error (RMSE) of the model prediction values ​​is calculated based on the non-overlapping validation dataset to evaluate the prediction accuracy of the model in the new scenario with sparse dataset. When the RMSE meets the prediction accuracy requirements, the channel parameter prediction model trained by the pre-meta-learning is determined.

[0077] Step 18: The channel parameter prediction model trained by pre-meta-learning determines the predicted values ​​of the channel parameters of the wireless channel transceiver to be predicted from the original satellite image. The channel parameter prediction model includes an environmental feature extraction module, a system parameter embedding module, a feature fusion module, and a regression prediction module connected in sequence. The channel parameter prediction model updates some parameters in advance based on the labeled samples of the wireless channel transceiver to be predicted.

[0078] Step 18 includes:

[0079] Step 181: Extract scatterer feature vectors from the original satellite image using the environmental feature extraction module. The channel parameter prediction model is a YOLOv6 network. The environmental feature extraction module includes a YOLOv6 backbone network, a neck network, a detector head, RoIAlign, and a pooling layer connected in sequence. Step 181 includes:

[0080] Step 1811: Extract multi-scale features from the original satellite images using the YOLOv6 backbone network.

[0081] Step 1812: Perform feature fusion on multi-scale features through the neck network to obtain a fused feature map. Specifically, based on the feature pyramid network, multi-scale features are fused through a top-down path and lateral connections to obtain the fused feature map.

[0082] Step 1813: Perform dense prediction on the fused feature map using the detection head to obtain the candidate boxes of scatterers and their corresponding confidence scores.

[0083] Step 1814: Extract region of interest features from the set of scatterer candidate boxes with confidence scores higher than a preset threshold and the fused feature map using RoIAlign to obtain the feature map patches corresponding to the scatterer candidate boxes.

[0084] Step 1815: Perform global pooling on the feature map blocks through a pooling layer to obtain the scatterer feature vector.

[0085] Specifically, such as Figure 5 As shown, this invention mainly predicts channel parameters based on the improved YOLOv6 model. After the preprocessing of the original satellite image is completed, multi-scale feature extraction is performed based on the backbone network of the YOLOv6 model, and the fusion of multi-scale features is achieved based on the feature pyramid network added to the neck network of the YOLOv6 model, as shown in formulas (8) and (9):

[0086] (8)

[0087] (9)

[0088] in, This is the feature map output by the l-th convolutional layer of the YOLOv6 model backbone network. The main feature extraction function is... It is the backbone network parameter set. This is a function for multi-scale feature fusion, consisting of a feature pyramid network and a path aggregation network. It enhances spatial details and semantic consistency through top-down and bottom-up paths. This is the set of trainable parameters for convolutional kernels, weighted fusion layers, upsampling layers, etc., in the neck network of the YOLOv6 model. To fuse feature maps.

[0089] Then, the detection head of the YOLOv6 model performs dense prediction on the fused feature map through a multi-scale convolutional structure, and the regression module in the detection head estimates the spatial location parameters of the candidate boxes. The classification module in the detection head calculates the confidence pre-class probability, thereby obtaining the set of scatterer candidate boxes, as shown in formula (10):

[0090] (10)

[0091] in, Let i be the prediction result for the i-th candidate box. For detecting header functions, Let N be the set of trainable parameters for the detector head, and N be the number of predicted scatterer candidate boxes.

[0092] Then, for the prediction results of multiple candidate boxes, multiple prediction results with a confidence level higher than a preset threshold are selected as the detection box set. RoIAlign is applied to extract feature map patches corresponding to candidate scatterer boxes of consistent size. As shown in formula (11):

[0093] (11)

[0094] Furthermore, feature map patches The scatterer feature vector is obtained through global pooling or flattening. As shown in formula (12):

[0095] (12)

[0096] in, This is a global average pooling function used to perform spatial averaging on each channel of the feature map, resulting in a channel mean vector. This is a flattening function used to convert the channel mean vector into a one-dimensional vector form, which facilitates processing by subsequent fully connected layers.

[0097] Step 182: Extract the embedded feature vector from the original system parameters of the wireless channel transceiver to be predicted through the system parameter embedding module.

[0098] Step 183: The scatterer feature vector and the embedded feature vector are fused using the feature fusion module to obtain the fused feature representation. The feature fusion module includes a vector concatenation layer, a fully connected layer, and a ReLU activation function. Step 183 includes:

[0099] Step 1831: The scatterer feature vector and the embedded feature vector are concatenated by a vector concatenation layer to obtain a joint feature vector.

[0100] Step 1832: Perform a nonlinear projection transformation on the joint feature vector through a fully connected layer to obtain the nonlinear projection result.

[0101] Step 1833: Process the nonlinear projection results using the ReLU activation function to obtain the fused feature representation.

[0102] Specifically, while acquiring the raw satellite images of the transceiver terminal of the wireless channel to be predicted, the raw system parameters of the transceiver terminal of the wireless channel to be predicted can also be acquired. The system parameter embedding module then uses a fully connected multi-layer perceptron (MLP) to embed the raw system parameters such as transmit power, altitude, and frequency. Mapping to the same semantic space as the image features yields the embedded feature vector. As shown in formula (13):

[0103] (13)

[0104] in, This is the first-level weight matrix, used to weight the original system parameters. The vectors are linearly mapped to the hidden layer space, controlling the transformation relationship between the input dimension and the intermediate feature dimension. This is the second-layer weight matrix, used to map the hidden layer output to the final embedded feature space. This is the first layer bias term corresponding to the additive bias in the hidden layer, used to improve the non-linear expressive power of the model. This is the second-level bias term, used to adjust the center of the mapping result.

[0105] Subsequently, in order to fully explore the potential correlation between visual scatterer features and system parameter features, a concatenated projection fusion mechanism is adopted to achieve efficient fusion of cross-modal information. For example, the system features are embedded into vectors by a vector concatenation layer. and scatterer eigenvectors The features are concatenated along the feature dimension to form a joint feature representation vector. Then, a nonlinear projection transformation is performed through one or more fully connected layers to capture the high-order interaction between the spatial features of the scatterer and the semantic features of the system parameters. For example, linear transformation and ReLU activation function are used to obtain the final fused feature representation. As shown in formula (14):

[0106] (14)

[0107] in, The weight matrix is ​​a learnable matrix. This is a learnable bias term.

[0108] Step 184: Perform regression prediction on the fused feature representation using the regression prediction module to obtain the predicted values ​​of the channel parameters.

[0109] Specifically, to calculate the channel propagation characteristics, key scatterer features and other system parameters in the environment surrounding the transmitter and receiver are used as inputs to the subsequent feature fusion layer. Then, a regression prediction module, such as the regression head of a YOLOv6 model, calculates the fused feature representation. Predicted values ​​of output channel parameters As shown in formula (15):

[0110] (15)

[0111] in, For detecting header functions, This is the set of trainable parameters for the regression head. Furthermore, the prediction accuracy of the model can be evaluated by calculating the RMSE, enabling fine-tuning of internal parameters for new scenarios or tasks. The RMSE evaluation is shown in formula (16):

[0112] (16)

[0113] in, These are the actual values ​​of channel parameters measured from a small number of samples in the new scenario. This is the number of predicted channel parameters.

[0114] It is important to emphasize that this invention introduces a meta-learning mechanism into the YOLOv6 model framework to enhance the model's rapid adaptability to new environments and its generalization performance with few samples. Specifically, the backbone network parameter set... The set of trainable parameters for convolutional kernels, weighted fusion layers, upsampling layers, etc. in the neck network. and the set of trainable parameters for the detection head. The set of trainable parameters for the regression head consists of cross-task shared extrinsic parameters pre-determined based on meta-learning training. To address task-specific internal parameters, when faced with new wireless channel transceiver scenarios to be predicted, the parameters are first fine-tuned using labeled samples, thereby ensuring the model's general applicability.

[0115] In summary, the channel parameter prediction method provided by this invention introduces an optimized YOLOv6 target detection structure and combines it with a mask segmentation module to accurately identify and label major scatterers such as buildings and vegetation in the original satellite images. Compared to traditional scatterer extraction methods based on grayscale features or manual thresholding, this invention can automatically generate high-resolution scatterer feature maps, effectively reducing the interference of invalid background noise such as roads and open ground on channel prediction, thereby improving the representativeness of input features and achieving high-precision identification and extraction of key scatterer features in the environment, thus optimizing the prediction network input.

[0116] Furthermore, the channel parameter prediction method provided by this invention fuses the scatterer feature map with system parameter vectors such as transmitter and receiver position, altitude, frequency, and distance in the channel dimension to form an optimized multimodal input structure. The defined multimodal input retains environmental spatial features and introduces physically interpretable parameters, significantly improving the prediction stability and robustness of the network, thereby realizing multimodal information fusion and improving the model's adaptability to environmental changes.

[0117] Furthermore, the channel parameter prediction method provided by this invention addresses the problem of poor generalization of existing models in cross-environment prediction. This invention integrates a meta-learning algorithm into the main network framework. By distinguishing between the training mechanisms of "inner layer parameters or task-specific parameters" and "outer layer parameters or cross-task shared parameters", the model can quickly migrate to new environments with a small number of adaptive samples, thereby integrating a meta-learning mechanism to achieve rapid adaptation across scenarios.

[0118] Furthermore, the channel parameter prediction method provided by this invention can be widely applied to channel modeling, site planning, and link budget optimization in 5G / 6G communication networks. Through more accurate path loss prediction, it can assist operators in achieving more reasonable resource allocation and coverage planning in the early stages of network deployment, thereby improving spectrum utilization and yielding certain economic and social benefits.

[0119] Example 2

[0120] This application also provides a channel parameter prediction system, including:

[0121] The data acquisition module is used to acquire raw satellite images of the wireless channel transceiver to be predicted;

[0122] The parameter prediction module is used to predict the channel parameters of the wireless transceiver terminal to be predicted from the original satellite image using a channel parameter prediction model trained in advance by a meta-learning mechanism. The channel parameter prediction model includes an environment feature extraction module, a system parameter embedding module, a feature fusion module, and a regression prediction module connected in sequence. The channel parameter prediction model is pre-updated with some parameters iteratively based on labeled samples of the wireless transceiver terminal to be predicted. The environment feature extraction module extracts scatterer feature vectors from the original satellite image. The system parameter embedding module extracts embedded feature vectors from the original system parameters of the wireless transceiver terminal to be predicted. The feature fusion module fuses the scatterer feature vectors and the embedded feature vectors to obtain a fused feature representation. The regression prediction module performs regression prediction on the fused feature representation to obtain the predicted channel parameters.

[0123] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps in an embodiment of a channel parameter prediction method. Specific implementation methods can be found in the method embodiments, and will not be repeated here.

[0124] Furthermore, the present invention also provides a non-transitory computer-readable storage medium containing instructions on which a computer program is stored. For example, a memory containing instructions that can be executed by a processor of a computer device to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. When the computer program is executed by the processor, it can implement the steps in an embodiment of a channel parameter prediction method. Specific implementation methods can be found in the method embodiments, which will not be repeated here.

[0125] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0126] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0127] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0128] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0129] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the present invention patent. No reference numerals in the claims should be construed as limiting the scope of the claims. Any simple variations or equivalent substitutions of technical solutions that can be readily obtained by those skilled in the art within the scope of the technology disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A method of channel parameter prediction, characterized by, include: Obtain raw satellite images of the transceiver ends of the wireless channel to be predicted; The channel parameter prediction model, trained in advance by a meta-learning mechanism, predicts the channel parameters of the wireless transceiver to be predicted from the original satellite image. The channel parameter prediction model includes an environmental feature extraction module, a system parameter embedding module, a feature fusion module, and a regression prediction module connected in sequence. The channel parameter prediction model updates some parameters in advance based on the labeled samples of the wireless transceiver to be predicted. The scatterer feature vector is extracted from the original satellite image using the environmental feature extraction module; The embedded feature vector is obtained by extracting the original system parameters of the wireless channel transceiver to be predicted through the system parameter embedding module; The feature fusion module fuses the scatterer feature vector and the embedded feature vector to obtain a fused feature representation; The channel parameters are predicted by performing regression prediction on the fused feature representation through the regression prediction module.

2. The channel parameter prediction method according to claim 1, characterized in that, The channel parameter prediction model is a YOLOv6 network, and the environmental feature extraction module includes a YOLOv6 backbone network, a neck network, a detection head, RoIAlign, and a pooling layer connected in sequence. Multi-scale features are extracted from the original satellite image using the YOLOv6 backbone network. The multi-scale features are fused using the neck network to obtain a fused feature map; the fused feature map is densely predicted using the detection head to obtain scatterer candidate boxes and their corresponding confidence scores; region of interest features are extracted from the set of scatterer candidate boxes with confidence scores higher than a preset threshold and the fused feature map using RoIAlign to obtain feature map patches corresponding to the scatterer candidate boxes; and global pooling is performed on the feature map patches using the pooling layer to obtain scatterer feature vectors.

3. The channel parameter prediction method according to claim 2, characterized in that, The neck network is constructed based on a feature pyramid network. The multi-scale features are fused using the neck network to obtain a fused feature map, including: The fused feature map is obtained by fusing the multi-scale features through a top-down path and lateral connections.

4. The channel parameter prediction method according to claim 2, characterized in that, The feature fusion module includes a vector concatenation layer, a fully connected layer, and a ReLU activation function. The feature fusion module fuses the scatterer feature vector and the embedded feature vector to obtain a fused feature representation, including: The vector concatenation layer concatenates the scatterer feature vector and the embedded feature vector to obtain a joint feature vector. The joint feature vector is subjected to a nonlinear projection transformation through the fully connected layer to obtain the nonlinear projection result; The nonlinear projection result is processed by the ReLU activation function to obtain the fused feature representation.

5. The channel parameter prediction method according to claim 2, characterized in that, Before the channel parameter prediction model is pre-trained using a meta-learning mechanism, it also includes: Obtain multiple sets of training support datasets and training query datasets for different environments; The model parameters of the regression prediction module are updated using gradients based on the training support dataset for each environment, resulting in updated model parameters for each environment. Meta-loss is calculated based on the updated model parameters of each environment and the corresponding training query dataset. The initial parameters of the YOLOv6 network are then updated using gradients based on the meta-loss to obtain the channel parameter prediction model.

6. The channel parameter prediction method according to claim 5, characterized in that, After obtaining the channel parameter prediction model, the method further includes: Obtain labeled samples of the wireless channel transceivers to be predicted; divide the labeled samples into an adaptation dataset and a non-overlapping verification dataset; The model parameters of the regression prediction module in the channel parameter prediction model are adjusted according to the adaptation dataset to obtain the scenario-adjusted channel parameter prediction model. The non-overlapping verification dataset is input into the scenario-adjusted channel parameter prediction model. When the root mean square error of the predicted channel parameters output by the scenario-adjusted channel parameter prediction model meets the prediction accuracy requirement, the scenario-adjusted channel parameter prediction model is determined as the channel parameter prediction model adapted to the scenario of the wireless channel transceiver to be predicted.

7. The channel parameter prediction method according to claim 1, characterized in that, Obtaining raw satellite images of the transceiver terminals of the wireless channel to be predicted includes: Obtain the original satellite image of the wireless channel transceiver to be predicted and the latitude and longitude of the wireless channel transceiver to be predicted; The relative distance between the transceivers of the wireless channel to be predicted is obtained by calculating the spherical distance based on the latitude and longitude of the transceivers. Based on the relative distance between the transceiver terminals of the wireless channel to be predicted, the original satellite image is cropped to obtain the preprocessed original satellite image.

8. A channel parameter prediction system, characterized in that, include: The data acquisition module is used to acquire raw satellite images of the wireless channel transceiver to be predicted; The parameter prediction module is used to predict the channel parameters of the wireless transceiver to be predicted from the original satellite image using a channel parameter prediction model trained in advance by a meta-learning mechanism. The channel parameter prediction model includes an environmental feature extraction module, a system parameter embedding module, a feature fusion module, and a regression prediction module connected in sequence. The channel parameter prediction model is pre-updated with some parameters iteratively based on labeled samples of the wireless transceiver to be predicted. The environmental feature extraction module extracts scatterer feature vectors from the original satellite image. The system parameter embedding module extracts the embedded feature vector from the original system parameters of the wireless channel transceiver to be predicted; the feature fusion module fuses the scatterer feature vector and the embedded feature vector to obtain a fused feature representation; and the regression prediction module performs regression prediction on the fused feature representation to obtain the predicted value of the channel parameters.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the channel parameter prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded by the processor, it is able to execute the steps of the channel parameter prediction method according to any one of claims 1 to 7.

Images