A channel prediction method, apparatus, and program based on wireless environment information

Abstract

This invention provides a channel prediction method, apparatus, and program based on wireless environment information, relating to the field of wireless technology. The channel prediction method includes: obtaining a portion of the first Channel State Information (CSI) under a first wireless scenario based on a target pilot pattern; and obtaining multi-view image data of the first wireless scenario. Based on the first CSI, the multi-view image data, and a target channel prediction model, the method predicts the complete CSI under the first wireless scenario. The target channel prediction model includes an image feature extraction neural network, a target pilot pattern, and a CSI reconstruction neural network. The image feature extraction neural network is used to extract image features from the image data, and the CSI reconstruction neural network is used to predict the complete CSI based on the image features and the portion of the CSI. This invention achieves high-precision, low-overhead channel state information prediction while considering dynamic environmental scenarios, addressing the needs of future communication systems.

Description

Technical Field

[0001] This invention relates to the field of wireless technology, and in particular to a channel prediction method, apparatus, and program based on wireless environment information. Background Technology

[0002] In sixth-generation (6G) massive multiple-input multiple-output (MIMO) systems, the size of antenna arrays and the number of antennas are constantly increasing. As an important component of wireless communication systems, the application of massive MIMO technology brings many benefits to wireless communication systems, such as high spectral efficiency, high energy efficiency, high spatial resolution, and high beamforming gain. To achieve these advantages, 6G massive MIMO systems need to acquire accurate Channel State Information (CSI).

[0003] Currently, channel prediction methods based on artificial intelligence (AI) can acquire channel index (CSI) in large-scale MIMO systems with relatively low pilot overhead. Given that channels exhibit certain correlations in spatial, temporal, and frequency dimensions, this method can uncover deeper relationships between channels in the spatiotemporal-frequency multi-domain domains, thereby reducing the need for multidimensional pilot signals.

[0004] However, current traditional AI-based channel prediction algorithms use partial antenna, time, and frequency CSI to predict the entire CSI, without considering wireless environment information. This data-driven algorithm only generates a prediction model based on training data and is difficult to generalize to other scenarios. Therefore, traditional algorithms are only suitable for relatively stable static environments and cannot achieve stable prediction performance with low overhead in dynamic scenarios. In the future, with the complex and variable nature of 6G channels, these algorithms will no longer meet the requirements. Summary of the Invention

[0005] The purpose of this invention is to provide a channel prediction method, apparatus, and program based on wireless environment information, in order to solve the problem that existing channel prediction algorithms cannot achieve high-precision, low-overhead channel prediction in dynamic environment scenarios.

[0006] To address the aforementioned technical problems, the embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, embodiments of the present invention provide a channel prediction method, comprising:

[0008] Based on the target pilot pattern, obtain partial first channel state information (CSI) of the first wireless scenario, and obtain multi-view image data of the first wireless scenario;

[0009] Based on the first CSI, the multi-view image data, and the target channel prediction model, the complete CSI in the first wireless scenario is predicted.

[0010] The target channel prediction model includes an image feature extraction neural network, the target pilot pattern, and a CSI reconstruction neural network. The image feature extraction neural network is used to extract image features from multi-view image data. The CSI reconstruction neural network is used to predict the complete CSI based on the image features and partial CSI. The image feature extraction neural network, the target pilot pattern, and the CSI reconstruction neural network are trained based on the complete second CSI in the simulation environment of the first wireless scenario and the multi-view image data in the simulation environment of the first wireless scenario.

[0011] Optionally, the method further includes:

[0012] The receiver's second CSI at the target location is obtained in the simulation environment of the first wireless scenario, and the receiver's multi-view image data at the target location is obtained in the simulation environment of the first wireless scenario, wherein the target location includes a first location and a second location.

[0013] Based on the first network hyperparameters of the first neural network, the second CSI at the first position, and the multi-view image data at the first position, the first image feature extraction parameters, the first pilot pattern, and the first CSI reconstruction parameters in the second neural network are trained to obtain the second image feature extraction parameters, the second pilot pattern, and the second CSI reconstruction parameters.

[0014] Based on the second CSI at the second position, the second pilot pattern, the second image feature extraction parameters, the first network hyperparameters, the multi-view image data at the second position, and the second CSI reconstruction parameters, the third CSI at the second position is predicted.

[0015] Obtain first error information and first similarity information between the second CSI and the third CSI;

[0016] If the first error information and the first similarity information do not meet the first preset condition, the first network hyperparameters are adjusted.

[0017] The process involves training the first image feature extraction parameters, the first pilot pattern, and the first CSI reconstruction parameters in the first neural network and the second neural network based on the first network hyperparameters of the first neural network, the second CSI at the first position, and the multi-view image data at the first position, to obtain the second image feature extraction parameters, the second pilot pattern, and the second CSI reconstruction parameters. This process continues until the first error information and the first similarity information meet the first preset condition. Based on the last adjusted first network hyperparameters and the second image feature extraction parameters corresponding to the last adjusted first network hyperparameters, the image feature extraction neural network is obtained. The second pilot pattern corresponding to the last adjusted first network hyperparameters is used as the target pilot pattern. The CSI reconstruction neural network is obtained based on the first CSI reconstruction parameters in the second neural network corresponding to the last adjusted first network hyperparameters.

[0018] Optionally, based on the first network hyperparameters of the first neural network, the second CSI at the first position, and the multi-view image data at the first position, the first image feature extraction parameters, the first pilot pattern, and the first CSI reconstruction parameters in the second neural network are trained to obtain the second image feature extraction parameters, the second pilot pattern, and the second CSI reconstruction parameters, including:

[0019] Using the first neural network, feature extraction is performed on the multi-view image data at the first location according to the first image feature extraction parameters to obtain an environmental feature map;

[0020] Based on the first pilot pattern and the second CSI, a fourth CSI is obtained for the portion below the first position;

[0021] Based on the first CSI reconstruction parameters in the second neural network, the environmental feature map, and the fourth CSI, the complete fifth CSI in the simulation environment of the first wireless scenario is predicted;

[0022] Obtain the loss information between the fifth CSI and the second CSI;

[0023] The first image feature extraction parameters are adjusted according to the loss information, the first pilot pattern is adjusted according to the loss information and the depth probability sampling algorithm, the first CSI reconstruction parameters are adjusted according to the loss information, and the process returns to the step of using the first neural network to extract features from the multi-view image data at the first position according to the first image feature extraction parameters to obtain an environmental feature map, until the number of iterations exceeds the preset number.

[0024] The first image feature extraction parameters after the last adjustment are used as the second image feature extraction parameters, the first pilot pattern after the last adjustment are used as the second pilot pattern, and the first CSI reconstruction parameters after the last adjustment are used as the second CSI reconstruction parameters.

[0025] Optionally, based on the first CSI reconstruction parameters in the second neural network, the environmental feature map, and the fourth CSI, a complete fifth CSI in the simulation environment of the first wireless scenario is predicted, including:

[0026] Using the second neural network and the proximal gradient iteration algorithm, the fifth CSI is predicted based on the first CSI reconstruction parameters, the environmental feature map, and the fourth CSI.

[0027] Optionally, the first preset condition includes:

[0028] The first error information is less than the second error information corresponding to the first channel prediction model, and the first similarity information is greater than the second similarity information corresponding to the first channel prediction model.

[0029] The first channel prediction model includes at least one of the following:

[0030] The first sub-channel prediction model includes: a randomly generated second pilot pattern and a CSI reconstruction neural network;

[0031] The second subchannel prediction model, wherein the first subchannel prediction model includes: a randomly generated second pilot pattern, the image feature extraction neural network, and the CSI reconstruction neural network;

[0032] The third sub-channel prediction model, wherein the first sub-channel prediction model includes: the target pilot pattern and the CSI reconstruction neural network.

[0033] Optionally, the method further includes:

[0034] Obtain environmental information of the first wireless scenario;

[0035] Using modeling tools, a simulation environment for the first wireless scenario is constructed based on the environmental information.

[0036] Optionally, obtaining the second CSI of the receiver at the target location in the simulation environment of the first wireless scenario includes:

[0037] Using a ray tracing channel simulation tool, the second CSI of the receiver at the target location in the simulation environment is generated.

[0038] Optionally, acquiring multi-view image data of the receiver at the target location in the simulation environment of the first wireless scenario includes:

[0039] Using an autonomous driving simulation platform, multi-view image data of the receiver at the target location is obtained in the simulation environment of the first wireless scenario.

[0040] Optionally, the target location may further include a third location;

[0041] The method further includes:

[0042] Based on the second CSI at the third position and the target pilot pattern, the sixth CSI at the third position is obtained;

[0043] The image feature extraction neural network is used to extract image features from the multi-view image data at the third position;

[0044] By using the CSI reconstruction neural network, the complete seventh CSI at the third position is predicted based on the image features and the sixth CSI;

[0045] The accuracy of the target channel prediction model is obtained based on the third error information and the third similarity information between the seventh CSI and the second CSI at the third position.

[0046] Optionally, the method further includes:

[0047] Acquire complete eighth CSI data and multi-view image data of the receiver in the simulation environment of the second wireless scenario;

[0048] Based on the eighth CSI and the target pilot pattern, a portion of the ninth CSI in the simulation environment of the second wireless scenario is obtained;

[0049] The image feature extraction neural network is used to extract image features from the multi-view image data of the receiver in the simulation environment of the second wireless scenario;

[0050] By using the CSI reconstruction neural network, based on the image features and the ninth CSI, the complete tenth CSI in the simulation environment of the second wireless scenario is predicted.

[0051] Based on the fourth error information and fourth similarity information between the tenth CSI and the eighth CSI, the generalization applicability result of the target channel prediction model is obtained.

[0052] Secondly, embodiments of the present invention also provide a channel prediction apparatus, comprising:

[0053] The first acquisition module is used to acquire partial first channel state information (CSI) of the first wireless scenario based on the target pilot pattern and to acquire multi-view image data of the first wireless scenario.

[0054] The first processing module is used to predict the complete CSI in the first wireless scenario based on the first CSI, the multi-view image data, and the target channel prediction model.

[0055] The target channel prediction model includes an image feature extraction neural network, the target pilot pattern, and a CSI reconstruction neural network. The image feature extraction neural network is used to extract image features from multi-view image data. The CSI reconstruction neural network is used to predict the complete CSI based on the image features and partial CSI. The image feature extraction neural network, the target pilot pattern, and the CSI reconstruction neural network are trained based on the complete second CSI in the simulation environment of the first wireless scenario and the multi-view image data in the simulation environment of the first wireless scenario.

[0056] Thirdly, embodiments of the present invention also provide a channel prediction device, comprising: a transceiver, a processor, a memory, and a program or instructions stored in the memory and executable on the processor; the processor, when executing the program or instructions, implements the steps of the channel prediction method as described in any one of the first aspects.

[0057] Fourthly, embodiments of the present invention also provide a readable storage medium having a program or instructions stored thereon, which, when executed by a processor, implement the steps of the channel prediction method as described in any one of the first aspects.

[0058] Fifthly, embodiments of the present invention also provide a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the channel prediction method as described in any one of the first aspects.

[0059] The beneficial effects of the above-described technical solution of the present invention are as follows:

[0060] The channel prediction method provided by this invention obtains a portion of the first CSI in a first wireless scenario based on the target pilot pattern, extracts image features from multi-view image data of the first wireless scenario using an image feature extraction neural network in the target channel model, and predicts the complete CSI in the first wireless scenario based on the image features and the portion of the first CSI using a CSI reconstruction neural network in the target channel model. That is, under the premise of considering dynamic environment scenarios, it achieves high-precision and low-overhead channel state information prediction to meet the needs of future communication systems. Attached Figure Description

[0061] Figure 1 A flowchart of the channel prediction method provided in the embodiments of the present invention;

[0062] Figure 2 This is a schematic diagram of the error information comparison results provided in an embodiment of the present invention;

[0063] Figure 3 This is a schematic diagram of similarity comparison results provided in an embodiment of the present invention;

[0064] Figure 4 A schematic diagram illustrating the prediction of complete CSI provided in an embodiment of the present invention;

[0065] Figure 5 This is a schematic diagram of the neural network structure in the target channel prediction model provided in an embodiment of the present invention;

[0066] Figure 6 This is a schematic diagram of the channel prediction device provided in an embodiment of the present invention;

[0067] Figure 7 This is a schematic diagram of the channel prediction device provided in an embodiment of the present invention. Detailed Implementation

[0068] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0069] To address the problem that existing channel prediction algorithms cannot achieve high-precision, low-overhead channel prediction in dynamic environments, this invention provides a channel prediction method, apparatus, and program based on wireless environment information.

[0070] like Figure 1 As shown, an embodiment of the present invention provides a channel prediction method, including:

[0071] Step 101: Obtain partial first channel state information (CSI) of the first wireless scenario based on the target pilot pattern, and obtain multi-view image data of the first wireless scenario.

[0072] The target pilot pattern is trained based on the complete second CSI under the simulation environment of the first wireless scenario and the multi-view image data under the simulation environment of the first wireless scenario.

[0073] It should be noted that each scenario has an optimal pilot pattern to facilitate the selection of pilot positions. The target pilot pattern is trained based on the complete second CSI and multi-view image data in the simulation environment of the first wireless scenario. This target pilot pattern is the optimal pilot pattern in the first wireless scenario.

[0074] After the target pilot pattern is determined, the transmitter and receiver become aware of the target pilot pattern and obtain a portion of the CSI, i.e., the first CSI, based on the target pilot pattern.

[0075] The first wireless scenario is any wireless communication scenario, which includes scenario size information, the location of scatterers in the scenario, scatterer size information, transmitter (Tx) location, receiver (Rx) location, and other information.

[0076] In this step, the multi-view image data in the first wireless scenario can be understood as the multi-view image data of the receiver in the first wireless scenario. The multi-view image data includes full-view image data, such as the full-view image data including the image data of the receiver from the east, west, south, and north. So, the multi-view image data includes the image data from at least one of the east, west, south, and north directions. That is, the multi-view image data can also be understood as image data from different directions.

[0077] Step 102: Based on the first CSI, the multi-view image data, and the target channel prediction model, predict the complete CSI for the first wireless scenario.

[0078] The target channel prediction model includes an image feature extraction neural network, the target pilot pattern, and a CSI reconstruction neural network. The image feature extraction neural network is used to extract image features from multi-view image data. The CSI reconstruction neural network is used to predict the complete CSI based on the image features and partial CSI. The image feature extraction neural network and the CSI reconstruction neural network are trained based on the complete second CSI in the simulation environment of the first wireless scenario and the multi-view image data in the simulation environment of the first wireless scenario.

[0079] Specifically, in this step, an image feature extraction neural network is used to extract image features from the multi-view image data, and a CSI reconstruction neural network is used to predict the complete CSI under the first wireless scenario based on the first CSI and the environmental features.

[0080] In this step, the image feature extraction neural network in the target channel prediction model is used to extract image features (also known as environmental features) in the first wireless communication scenario. These image features are obtained by constructing the relationship between the environment and the channel.

[0081] In this step, environmental features of multi-view image data are extracted through an image feature extraction neural network. The first CSI, environmental features, and CSI of this part are used to reconstruct the neural network, which can predict the complete CSI (or the overall CSI). In this embodiment of the invention, environmental information is introduced into the prediction, which significantly improves the accuracy of the overall CSI prediction and reduces pilot overhead.

[0082] Furthermore, this method is more suitable for the diverse application scenarios required by future mobile communication systems, and has greater applicability to various scenarios.

[0083] Furthermore, before training the target channel prediction model, the method further includes:

[0084] The receiver's second CSI at the target location is acquired in the simulation environment of the first wireless scenario, and the receiver's multi-view image data at the target location is acquired in the simulation environment of the first wireless scenario, wherein the target location includes a first location and a second location.

[0085] In this context, one receiver corresponds to one target location, or the location of the receiver in the first wireless scenario is the target location.

[0086] Specifically, in this step, for each receiver in the first wireless scenario, the corresponding channel parameters at the target location are defined as the second CSI. For each receiver in the first wireless scenario, the RGB images in the east, west, south, and north directions at the target location constitute the multi-view image data of the target location.

[0087] Align the CSI and multi-view image data according to the target location to obtain the channel-multi-view image dataset. The channel-multi-view image dataset includes the second CSI and multi-view image data corresponding to each target location. The target location includes the first location and the second location. That is, the channel-multi-view image dataset includes the second CSI and multi-view image data corresponding to the first location and the second CSI and multi-view image data corresponding to the second location. It should also be noted that the target location may also include the third location. That is, the channel-multi-view image dataset may also include the second CSI and multi-view image data corresponding to the third location.

[0088] This channel-multiview image dataset provides a data foundation for training subsequent target channel prediction models.

[0089] Optionally, the second CSI and multi-view image data corresponding to the first position can be used as the training dataset, the second CSI and multi-view image data corresponding to the second position can be used as the validation dataset, and the second CSI and multi-view image data corresponding to the third position can be used as the test dataset, respectively for training, validation, and testing of the target channel prediction model. The ratio of the first, second, and third positions in the target position can be 0.7:0.1:0.2, respectively.

[0090] In one alternative embodiment, the process of building the simulation environment includes:

[0091] The environmental information of the first wireless scene is obtained. Specifically, the three-dimensional wireless propagation environment of the first wireless scene is collected. The environmental information of the three-dimensional wireless propagation environment is automatically collected through actual measurement, map-assisted acquisition, and computer vision-related technologies. The environmental information includes the overall layout of the wireless scene, scatterer information, transceiver information, etc., such as scene size information, scatterer position in the scene, scatterer size information, transmitter position, receiver position, etc.

[0092] Using modeling tools, a simulation environment for the first wireless scenario is constructed based on the environmental information. Specifically, using 3D modeling tools such as Blender and Google SketchUp, a wireless propagation environment dataset of scatterer distribution is constructed in a coordinate system of the same size as the actual scene, based on the environmental information. This dataset is the simulation environment for the first wireless scenario.

[0093] In an optional embodiment, obtaining the second CSI of the receiver at the target location in the simulated environment of the first wireless scenario includes:

[0094] Using a ray tracing channel simulation tool, the second CSI of the receiver at the target location in the simulation environment is generated.

[0095] Using a ray-tracing channel simulation tool, the second CSI of the receiver at the target location in the simulation environment is generated. Specifically, the channel parameters corresponding to each Rx location are generated using a ray-tracing (RT) channel simulation tool, such as WirelessInsite, to generate the second CSI of the target location.

[0096] In an optional embodiment, acquiring multi-view image data of the receiver at the target location in the simulation environment of the first wireless scenario includes:

[0097] Using an autonomous driving simulation platform, the receiver acquires multi-view image data at the target location in the simulation environment of the first wireless scenario. Specifically, an autonomous driving simulation platform, such as CALA, is used to collect RGB images in the four directions of east, west, south, and north for each Rx location.

[0098] The following specific example illustrates the process of constructing a channel-multi-view image dataset:

[0099] A large outdoor urban scene measuring 200m long and 200m wide was constructed in RoadRunner, and a base station was set up to cover the scene. The scene contains four building clusters and four roads. Different types of vehicles are randomly placed on the roads to ensure sufficient diversity in the scene across different roads. The building clusters, roads, and vehicles can all be considered as scattering bodies.

[0100] Import the scene into CALA, and obtain multi-view image data at different Rx positions by deploying cameras at 0.26m intervals on the road and capturing multi-view images in sequence.

[0101] Since the lack of surface details has a limited impact on channel data, the entire scene was simplified in Blender software, where buildings and vehicles were replaced with simple cubes. The simplified scene model was then imported into WirelessInsite for RT simulation, generating channels at different Rx locations and extracting CSI data. The number of transmitting antennas M used in the simulation to generate CSI data was... t =128, Number of receiving antennas N r =1, the number of subcarriers N c =69, the number of symbols N in Orthogonal Frequency Division Multiplexing (OFDM) in the time domain. T =3, and detailed RT simulation data are shown in Table 1 below.

[0102] Table 1

[0103]

[0104] The training process of the target channel prediction model is explained in detail below:

[0105] Based on the first network hyperparameters of the first neural network, the second CSI at the first position, and the multi-view image data at the first position, the first image feature extraction parameters, the first pilot pattern, and the first CSI reconstruction parameters in the second neural network are trained to obtain the second image feature extraction parameters, the second pilot pattern, and the second CSI reconstruction parameters.

[0106] Wherein, the first network hyperparameter is a randomly generated initial network hyperparameter, the first image feature extraction parameter is a randomly generated initial image feature extraction parameter, and the first CSI reconstruction parameter is a randomly generated CSI reconstruction parameter.

[0107] The first neural network can be understood as the image feature extraction neural network before it has been trained and optimized. The CSI reconstruction neural network includes the second neural network before it has been trained and optimized. The type of the first neural network should be consistent with the type of the image feature extraction neural network, and the type of the CSI reconstruction neural network should be consistent with the type of the second neural network.

[0108] The first neural network can be a convolutional neural network (CNN), a self-attention layer, etc. It is understood that the image feature extraction neural network is of the same type as the first neural network, and the image feature extraction neural network is trained from the first neural network.

[0109] Network hyperparameters are variables that determine the network structure (e.g., the number of hidden units) and how the network is trained (e.g., the learning rate). In the context of machine learning, network hyperparameters are parameters whose values ​​are set before the learning process begins, rather than parameters obtained through training data. Typically, network hyperparameters need to be optimized to select an optimal set of hyperparameters for the learning machine to improve learning performance and effectiveness.

[0110] The first image feature extraction parameters are those primarily related to feature extraction. These parameters include weight parameters, bias parameters, etc.

[0111] The target image feature extraction parameters are obtained by training and optimizing the first image feature extraction parameters using the training data mentioned above, including the optimal weight parameters, bias parameters, etc.

[0112] The second type of neural network includes CNN, Recurrent Neural Network (RNN), etc.

[0113] CSI reconstruction parameters include relevant parameters used for CSI estimation and design parameters for the reconstruction algorithm.

[0114] Based on the first network hyperparameters of the first neural network, the second CSI at the first position, and the multi-view image data at the first position, the first image feature extraction parameters, the first pilot pattern, and the first CSI reconstruction parameters in the second neural network are trained to obtain the second image feature extraction parameters, the second pilot pattern, and the second CSI reconstruction parameters. Specifically, based on the second CSI at the second position and the second pilot pattern, the CSI corresponding to the second position is obtained. Based on the first neural network and the first network hyperparameters and the second image feature extraction parameters, the image features of the multi-view image data at the second position are extracted. Based on the image features, the aforementioned CSI, the second neural network, and the first CSI reconstruction parameters of the second neural network, the complete third CSI at the second position is predicted.

[0115] Obtain first error information and first similarity information between the second CSI and the third CSI, wherein the first error information is normalized mean square error (NMSE) and the first similarity information is cosine similarity.

[0116] Determine whether the first error information and the first similarity information meet the first preset conditions, wherein the first preset conditions include: the first error information is less than the second error information corresponding to the first channel prediction model, and the first similarity information is greater than the second similarity information corresponding to the first channel prediction model;

[0117] The first channel prediction model includes at least one of the following:

[0118] The first sub-channel prediction model includes: a randomly generated second pilot pattern and a CSI reconstruction neural network. That is, the first sub-channel prediction model does not perform pilot pattern optimization and does not include an image feature extraction neural network for extracting image features from multi-view image data (referred to as environment-free random sampling).

[0119] The second sub-channel prediction model includes: a randomly generated second pilot pattern, the image feature extraction neural network, and a CSI reconstruction neural network. That is, the first sub-channel prediction model does not perform pilot pattern optimization, but includes an image feature extraction neural network (referred to as environmental random sampling).

[0120] The third sub-channel prediction model includes the target pilot pattern and the CSI reconstruction neural network. That is, the third sub-channel prediction model optimizes the pilot pattern but does not include the image feature extraction neural network (abbreviated as environment-free DPS) used to extract image features from multi-view image data.

[0121] The target channel prediction model is simply referred to as environment-dependent DPS.

[0122] All of the above models were trained for 200 epochs and used the same mean-square error (MSE) loss function.

[0123] Specifically, the first error information is compared with the second error information corresponding to the first sub-channel prediction model, the second sub-channel prediction model, and the third sub-channel prediction model, respectively. The comparison results are as follows: Figure 2 As shown, the first similarity information is compared with the second similarity information corresponding to the first sub-channel prediction model, the second sub-channel prediction model, and the third sub-channel prediction model, respectively. The comparison results are as follows. Figure 3 As shown, Figure 2 and Figure 3 In Chinese, "random+withoutenvironment" means "random sampling without environment", "random+environment" means "random sampling with environment", "DPS+without environment" means "DPS without environment", and "DPS+withenvironment" means "DPS with environment".

[0124] according to Figure 2 and Figure 3 It can be determined whether the first error information and the first similarity information meet the first preset conditions.

[0125] If the first error information and the first similarity information do not meet the first preset condition, the first network hyperparameters are adjusted, and the process returns to the step of training the first image feature extraction parameters and the first pilot pattern in the first neural network based on the first network hyperparameters of the first neural network, the second CSI at the first position, and the multi-view image data at the first position to obtain the second image feature extraction parameters and the second pilot pattern. The process then determines whether the adjusted first error information and the first similarity information meet the first preset condition again. If they do not meet the first preset condition, the process returns; if they do meet the condition, the training ends. Based on the first network hyperparameters adjusted last time and the second image feature extraction parameters corresponding to the first network hyperparameters adjusted last time, the image feature extraction neural network is obtained, and the second pilot pattern corresponding to the first network hyperparameters adjusted last time is used as the target pilot pattern.

[0126] Specifically, after training, the hyperparameters of the first neural network are set to the last adjusted hyperparameters, and the image feature extraction parameters of the first neural network are set to the second image feature extraction parameters corresponding to the last adjusted hyperparameters, thus obtaining the trained image feature extraction neural network. The last adjusted hyperparameters are the optimal hyperparameters, and the second image feature extraction parameters corresponding to the last adjusted hyperparameters are the optimal image feature extraction parameters.

[0127] like Figure 2 and Figure 3 As shown, the proposed method (target channel prediction model) achieved the highest correlation coefficient and the lowest NMSE during network model training. Comparing the results with and without the pilot optimization module (e.g., green and dark black curves or blue and red curves), under the same environment, the optimal pilot pattern is beneficial to improving CSI prediction accuracy, reducing NMSE by approximately 76.4% and 64.7%, respectively. Furthermore, under the same pilot optimization scheme, without environment information (e.g., green and blue curves or dark black and red curves), the environmental information provides a gain, leading to a reduction in NMSE of approximately 64.2% and 46.4%, respectively. Moreover, compared to 1 / 5 pilots with random sampling without environment (light black curve), 1 / 8 pilots with environmental DPS exhibit better CSI prediction performance. Pilot overhead is reduced by at least 37.5%.

[0128] Specifically, the overall network of the target neural channel prediction model consists of three modules: an image feature extraction module, a pilot optimization module, and a CSI reconstruction neural network.

[0129] Specifically, based on the first network hyperparameters of the first neural network, the second CSI at the first position, and the multi-view image data at the first position, the first image feature extraction parameters, the first pilot pattern, and the first CSI reconstruction parameters in the second neural network are trained to obtain the second image feature extraction parameters, the second pilot pattern, and the second CSI reconstruction parameters, including:

[0130] Using the first neural network, an environmental feature map is obtained by extracting features from the multi-view image data at the first location according to the first image feature extraction parameters. Specifically, multi-view images are acquired and input into an image feature extraction module, which includes an image feature extraction neural network. The image feature extraction neural network extracts image features from the multi-view image data.

[0131] In the image feature extraction module, CNNs excel in visual tasks because they can extract hierarchical features from multi-view image data. Therefore, we utilize three convolutional layers and one pooling layer of a CNN to extract environmental feature maps from panoramic images, which can reflect the scene environment information around the receiver.

[0132] In the pilot optimization module, based on the first pilot pattern and the second CSI, the fourth CSI for the portion at the first position is obtained, using the following formula: H partial =H m,n A, where H partial Indicates the fourth CSI, H m，n Indicates the second CSI, and A indicates the first pilot pattern;

[0133] Based on the first CSI reconstruction parameters in the second neural network, the environmental feature map, and the fourth CSI, the complete fifth CSI in the simulation environment of the first wireless scenario is predicted;

[0134] Obtain the loss information between the fifth CSI and the second CSI;

[0135] The first image feature extraction parameters are adjusted according to the loss information, the first pilot pattern is adjusted according to the loss information and the depth probability sampling algorithm, the first CSI reconstruction parameters are adjusted according to the loss information, and the process returns to the step of using the first neural network to extract features from the multi-view image data at the first position according to the first image feature extraction parameters to obtain an environmental feature map, until the number of iterations exceeds the preset number.

[0136] The first image feature extraction parameters after the last adjustment are used as the second image feature extraction parameters, the first pilot pattern after the last adjustment are used as the second pilot pattern, and the first CSI reconstruction parameters after the last adjustment are used as the second CSI reconstruction parameters.

[0137] Specifically, low-overhead optimal pilot pattern design can be described as a task-adaptive compressed sensing problem, the goal of which is to select the optimal subset of signal samples to achieve end-to-end optimization. For each channel state information H corresponding to the m-th transmit antenna and the n-th receive antenna... m,n (i.e., the second CSI) The deep probabilistic subsampling (DPS) algorithm generates an optimized pilot pattern in the time-frequency dimension based on the loss information. (i.e., the adjusted first pilot pattern). It is the time-frequency dimension CSI (partial CSI) under a given pilot pattern, which can be represented as H. partial =H m,n A. The number of elements with a '1' in the optimized sampling matrix A is equal to the number of pilot signals N. p The optimal pilot position is obtained during the training process of the neural network's forward propagation. The pilot optimization module improves channel prediction accuracy while minimizing pilot overhead. The pilot sampling ratio is set to 1 / 8, meaning 1 / 8 (N) of A is used. c *N T (8 = 26) The element is 1, and the rest are 0. c N represents the number of subcarriers. T This represents the number of OFDM symbols in the time domain. After a preset number of iterations (i.e., a preset number of iterations), the final optimized pilot pattern becomes the target pilot pattern. This preset number of iterations is 200.

[0138] In the CSI reconstruction neural network, the environmental feature map and the CSI of that part are used as inputs. Based on the first CSI reconstruction parameters, the complete CSI prediction value, i.e., the fifth CSI, is obtained.

[0139] The loss information between the fifth CSI and the second CSI can be obtained through a preset loss function.

[0140] After each acquisition of loss information, the first image feature extraction parameters, the first pilot pattern, and the first CSI reconstruction parameters are adjusted according to the loss information. Then, the next iteration is performed until the number of iterations exceeds 200, at which point the iteration process ends.

[0141] Further, based on the first CSI reconstruction parameters in the second neural network, the environmental feature map, and the fourth CSI, a complete fifth CSI is predicted under the simulated environment of the first wireless scenario, including:

[0142] Using the second neural network and the proximal gradient iteration algorithm, the fifth CSI is predicted based on the first CSI reconstruction parameters, the environmental feature map, and the fourth CSI.

[0143] Specifically, in the CSI reconstruction neural network, the environmental feature map and the fourth CSI are used as inputs, and a second neural network (which includes the first CSI reconstruction parameters) is used in conjunction with a proximal gradient iterative algorithm to predict the complete channel matrix {H} (i.e., the fifth CSI), thereby accelerating the convergence of the model. The iterative steps are expressed as follows:

[0144]

[0145] X k+1 :=prox k (S k+1 ).

[0146] Among them, X k ,prox k α k X represents the forward value, the near-end gradient operator, and the channel prediction parameters for the k-th iteration, respectively. k+1 and S k+1 These are the backward value and intermediate variable of the (k+1)th iteration, respectively. The proximal operator prox is executed using a second neural network. k and channel prediction parameter α k The algorithm iteration count is set to 2 (i.e., k equals 0, 1, 2). The CSI data obtained after iteration is combined with the number of channels and image feature mapping, and the complete CSI prediction value (i.e., the fifth CSI) is obtained through the second neural network.

[0147] Predicting the complete process of the fifth CSI as follows Figure 4 As shown in the diagram. A schematic diagram of the neural network structure in the target channel prediction model is shown below. Figure 5 As shown. Figure 5 The actual CSI in the data is the second CSI mentioned above, and some CSIs are the fourth CSI.

[0148] Furthermore, the method also includes:

[0149] Based on the second CSI at the third position and the target pilot pattern, the sixth CSI at the third position is obtained;

[0150] The image feature extraction neural network is used to extract image features from the multi-view image data at the third position;

[0151] By using the CSI reconstruction neural network, the complete seventh CSI at the third position is predicted based on the image features and the sixth CSI;

[0152] The accuracy of the target channel prediction model is obtained based on the third error information and the third similarity information between the seventh CSI and the second CSI at the third position.

[0153] In this embodiment of the invention, the accuracy of the target channel prediction model is tested using the test data described above. The third error information is the normalized mean square error, and the third similarity information is the cosine similarity information.

[0154] That is, using the test data, the sixth CSI is obtained according to the fourth CSI acquisition process described above, and the seventh CSI is obtained according to the fifth CSI acquisition process described above. The specific process will not be repeated here.

[0155] Furthermore, the method also includes:

[0156] Acquire complete eighth CSI data and multi-view image data of the receiver in the simulation environment of the second wireless scenario;

[0157] The second wireless scenario can be understood as a new scenario different from the first wireless scenario. For example, modifying the building and vehicle layout of the original scenario to create a new scenario. The dataset (simulation environment) of the new scenario contains 1601 Rx values ​​for the two middle roads.

[0158] Based on the eighth CSI and the target pilot pattern, a portion of the ninth CSI in the simulation environment of the second wireless scenario is obtained;

[0159] The image feature extraction neural network is used to extract image features from the multi-view image data of the receiver in the simulation environment of the second wireless scenario;

[0160] By using the CSI reconstruction neural network, based on the image features and the ninth CSI, the complete tenth CSI in the simulation environment of the second wireless scenario is predicted.

[0161] Based on the fourth error information and fourth similarity information between the tenth CSI and the eighth CSI, the generalization applicability result of the target channel prediction model is obtained.

[0162] The ninth CSI is obtained by following the acquisition process of the fourth CSI described above, and the tenth CSI is obtained by following the acquisition process of the fifth CSI described above. The specific process will not be repeated here.

[0163] The fourth error information is the normalized mean square error, and the fourth similarity information is the cosine similarity information.

[0164] Overall, the cosine similarity and NMSE in the new scenario are similar to those in the original scenario (the first wireless scenario), indicating that the target channel prediction model generalizes well.

[0165] Furthermore, the cosine similarity and NMSE of the original scene and the new scene under the target channel prediction model, the first sub-channel prediction model, the second sub-channel prediction model and the third sub-channel prediction model can be compared to obtain the target channel prediction model. The comparison and results are shown in Table 2 below.

[0166] Table 2

[0167]

[0168] In summary, this invention utilizes both the powerful feature extraction and nonlinear mapping capabilities of AI and the environmental feature information provided by images. It can improve CSI prediction accuracy while saving pilot overhead, solving the problem of high CSI acquisition overhead in large-scale MIMO systems. The optimized pilot arrangement scheme increases prediction accuracy compared to random pilot schemes. It can perform efficient and accurate CSI prediction for specific practical application scenarios. By introducing environmental information into the CSI prediction process, it increases the adaptability of the AI ​​model to the environment, making it more suitable for the diverse application scenarios required by future mobile communication systems and more universally applicable.

[0169] like Figure 6 As shown, embodiments of the present invention also provide a channel prediction apparatus, comprising:

[0170] The first acquisition module 601 is used to acquire partial first channel state information (CSI) of the first wireless scenario based on the target pilot pattern, and to acquire multi-view image data of the first wireless scenario.

[0171] The first processing module 602 is used to predict the complete CSI in the first wireless scenario based on the first CSI, the multi-view image data and the target channel prediction model.

[0172] The target channel prediction model includes an image feature extraction neural network, the target pilot pattern, and a CSI reconstruction neural network. The image feature extraction neural network is used to extract image features from multi-view image data. The CSI reconstruction neural network is used to predict the complete CSI based on the image features and partial CSI. The image feature extraction neural network, the target pilot pattern, and the CSI reconstruction neural network are trained based on the complete second CSI in the simulation environment of the first wireless scenario and the multi-view image data in the simulation environment of the first wireless scenario.

[0173] Optionally, the device further includes:

[0174] The second acquisition module is used to acquire the second CSI of the receiver at the target location in the simulation environment of the first wireless scenario, and to acquire multi-view image data of the receiver at the target location in the simulation environment of the first wireless scenario, wherein the target location includes a first location and a second location.

[0175] The second processing module is used to train the first image feature extraction parameters, the first pilot pattern, and the first CSI reconstruction parameters in the first neural network based on the first network hyperparameters of the first neural network, the second CSI at the first position, and the multi-view image data at the first position, to obtain the second image feature extraction parameters, the second pilot pattern, and the second CSI reconstruction parameters in the second neural network.

[0176] The third processing module is used to predict the third CSI at the second position based on the second CSI at the second position, the second pilot pattern, the second image feature extraction parameters, the first network hyperparameters, the multi-view image data at the second position, and the second CSI reconstruction parameters.

[0177] The third acquisition module is used to acquire the first error information and the first similarity information between the second CSI and the third CSI;

[0178] The fourth processing module is used to adjust the hyperparameters of the first network when the first error information and the first similarity information do not meet the first preset conditions.

[0179] The fifth processing module is used to return the steps of training the first image feature extraction parameters, the first pilot pattern, and the first CSI reconstruction parameters in the first neural network and the second neural network based on the first network hyperparameters of the first neural network, the second CSI at the first position, and the multi-view image data at the first position, to obtain the second image feature extraction parameters, the second pilot pattern, and the second CSI reconstruction parameters, until the first error information and the first similarity information meet the first preset condition. The module then obtains the image feature extraction neural network based on the last adjusted first network hyperparameters and the second image feature extraction parameters corresponding to the last adjusted first network hyperparameters. It also uses the second pilot pattern corresponding to the last adjusted first network hyperparameters as the target pilot pattern, and obtains the CSI reconstruction neural network based on the first CSI reconstruction parameters in the second neural network corresponding to the last adjusted first network hyperparameters.

[0180] Optionally, the second processing module includes:

[0181] The first processing unit is used to extract environmental feature maps from multi-view image data at the first location using the first neural network and based on the first image feature extraction parameters.

[0182] The second processing unit is configured to obtain a fourth CSI for a portion of the first position based on the first pilot pattern and the second CSI.

[0183] The third processing unit is used to predict the complete fifth CSI in the simulation environment of the first wireless scenario based on the first CSI reconstruction parameters in the second neural network, the environmental feature map, and the fourth CSI.

[0184] The first acquisition unit is used to acquire the loss information between the fifth CSI and the second CSI;

[0185] The fourth processing unit is configured to adjust the first image feature extraction parameters according to the loss information, adjust the first pilot pattern according to the loss information and the depth probability sampling algorithm, adjust the first CSI reconstruction parameters according to the loss information, and return to the step of using the first neural network to extract features from the multi-view image data at the first position according to the first image feature extraction parameters to obtain an environmental feature map, until the number of iterations exceeds a preset number.

[0186] The fifth processing unit is configured to use the last adjusted first image feature extraction parameters as the second image feature extraction parameters, the last adjusted first pilot pattern as the second pilot pattern, and the last adjusted first CSI reconstruction parameters as the second CSI reconstruction parameters.

[0187] Optionally, the third processing unit is specifically used for:

[0188] Using the second neural network and the proximal gradient iteration algorithm, the fifth CSI is predicted based on the first CSI reconstruction parameters, the environmental feature map, and the fourth CSI.

[0189] Optionally, the first preset condition includes:

[0190] The first error information is less than the second error information corresponding to the first channel prediction model, and the first similarity information is greater than the second similarity information corresponding to the first channel prediction model.

[0191] The first channel prediction model includes at least one of the following:

[0192] The first sub-channel prediction model includes: a randomly generated second pilot pattern and a CSI reconstruction neural network;

[0193] The second sub-channel prediction model includes: a randomly generated second pilot pattern, the image feature extraction neural network, and the CSI reconstruction neural network;

[0194] The third sub-channel prediction model includes the target pilot pattern and the CSI reconstruction neural network.

[0195] Optionally, the device further includes:

[0196] The fourth acquisition module is used to acquire environmental information of the first wireless scenario;

[0197] The sixth processing module is used to construct a simulation environment for the first wireless scenario based on the environmental information using modeling tools.

[0198] Optionally, the second acquisition module includes:

[0199] The second acquisition unit is used to generate the second CSI of the receiver at the target location in the simulation environment using a ray tracing channel simulation tool.

[0200] Optionally, the second acquisition module includes:

[0201] The third acquisition unit is used to acquire multi-view image data of the receiver at the target location in the simulation environment of the first wireless scenario using an autonomous driving simulation platform.

[0202] Optionally, the target location may further include a third location;

[0203] The device further includes:

[0204] The seventh processing module is used to obtain the sixth CSI of the third position based on the second CSI at the third position and the target pilot pattern;

[0205] The eighth processing module is used to extract image features from the multi-view image data at the third position using the image feature extraction neural network.

[0206] The ninth processing module is used to reconstruct the neural network through the CSI, and predict the complete seventh CSI at the third position based on the image features and the sixth CSI.

[0207] The tenth processing module is used to obtain the accuracy of the target channel prediction model based on the third error information and third similarity information between the seventh CSI and the second CSI at the third position.

[0208] Optionally, the device further includes:

[0209] The fifth acquisition module is used to acquire the complete eighth CSI in the simulation environment of the second wireless scenario and the multi-view image data of the receiver in the simulation environment of the second wireless scenario;

[0210] The eleventh processing module is used to obtain a portion of the ninth CSI in the simulation environment of the second wireless scenario based on the eighth CSI and the target pilot pattern.

[0211] The twelfth processing module is used to extract image features from the multi-view image data of the receiver in the simulation environment of the second wireless scenario using the image feature extraction neural network.

[0212] The thirteenth processing module is used to reconstruct the neural network through the CSI and predict the complete tenth CSI in the simulation environment of the second wireless scene based on the image features and the ninth CSI.

[0213] The fourteenth processing module is used to obtain the generalization applicability result of the target channel prediction model based on the fourth error information and fourth similarity information between the tenth CSI and the eighth CSI.

[0214] It should be noted that the channel prediction device provided in the embodiments of the present invention is a device capable of executing the above-described channel prediction method. Therefore, all embodiments of the above-described channel prediction method are applicable to this device and can achieve the same or similar technical effects.

[0215] like Figure 7 As shown, this embodiment of the invention also provides a channel prediction device, including: a processor 701; and a memory 703 connected to the processor 701 via a bus interface 702, the memory 703 being used to store programs and data used by the processor 701 when performing operations, and the processor 701 calling and executing the programs and data stored in the memory 703.

[0216] The transceiver 704 is connected to the bus interface 702 and is used to receive and send data under the control of the processor 701. Specifically, the processor 701 is used to read the program in the memory 703 and execute the following processes:

[0217] Based on the target pilot pattern, obtain partial first channel state information (CSI) of the first wireless scenario, and obtain multi-view image data of the first wireless scenario;

[0218] Based on the first CSI, the multi-view image data, and the target channel prediction model, the complete CSI in the first wireless scenario is predicted.

[0219] The target channel prediction model includes an image feature extraction neural network, the target pilot pattern, and a CSI reconstruction neural network. The image feature extraction neural network is used to extract image features from multi-view image data. The CSI reconstruction neural network is used to predict the complete CSI based on the image features and partial CSI. The image feature extraction neural network, the target pilot pattern, and the CSI reconstruction neural network are trained based on the complete second CSI in the simulation environment of the first wireless scenario and the multi-view image data in the simulation environment of the first wireless scenario.

[0220] Optionally, the processor 701 is further configured to:

[0221] The receiver's second CSI at the target location is obtained in the simulation environment of the first wireless scenario, and the receiver's multi-view image data at the target location is obtained in the simulation environment of the first wireless scenario, wherein the target location includes a first location and a second location.

[0222] Based on the first network hyperparameters of the first neural network, the second CSI at the first position, and the multi-view image data at the first position, the first image feature extraction parameters, the first pilot pattern, and the first CSI reconstruction parameters in the second neural network are trained to obtain the second image feature extraction parameters, the second pilot pattern, and the second CSI reconstruction parameters.

[0223] Based on the second CSI at the second position, the second pilot pattern, the second image feature extraction parameters, the first network hyperparameters, the multi-view image data at the second position, and the second CSI reconstruction parameters, the third CSI at the second position is predicted.

[0224] Obtain first error information and first similarity information between the second CSI and the third CSI;

[0225] If the first error information and the first similarity information do not meet the first preset condition, the first network hyperparameters are adjusted.

[0226] The process involves training the first image feature extraction parameters, the first pilot pattern, and the first CSI reconstruction parameters in the first neural network and the second neural network based on the first network hyperparameters of the first neural network, the second CSI at the first position, and the multi-view image data at the first position, to obtain the second image feature extraction parameters, the second pilot pattern, and the second CSI reconstruction parameters. This process continues until the first error information and the first similarity information meet the first preset condition. Based on the last adjusted first network hyperparameters and the second image feature extraction parameters corresponding to the last adjusted first network hyperparameters, the image feature extraction neural network is obtained. The second pilot pattern corresponding to the last adjusted first network hyperparameters is used as the target pilot pattern. The CSI reconstruction neural network is obtained based on the first CSI reconstruction parameters in the second neural network corresponding to the last adjusted first network hyperparameters.

[0227] Optionally, the processor 701 is specifically used for:

[0228] Using the first neural network, feature extraction is performed on the multi-view image data at the first location according to the first image feature extraction parameters to obtain an environmental feature map;

[0229] Based on the first pilot pattern and the second CSI, a fourth CSI is obtained for the portion below the first position;

[0230] Based on the first CSI reconstruction parameters in the second neural network, the environmental feature map, and the fourth CSI, the complete fifth CSI in the simulation environment of the first wireless scenario is predicted;

[0231] Obtain the loss information between the fifth CSI and the second CSI;

[0232] The first image feature extraction parameters are adjusted according to the loss information, the first pilot pattern is adjusted according to the loss information and the depth probability sampling algorithm, the first CSI reconstruction parameters are adjusted according to the loss information, and the process returns to the step of using the first neural network to extract features from the multi-view image data at the first position according to the first image feature extraction parameters to obtain an environmental feature map, until the number of iterations exceeds the preset number.

[0233] The first image feature extraction parameters after the last adjustment are used as the second image feature extraction parameters, the first pilot pattern after the last adjustment are used as the second pilot pattern, and the first CSI reconstruction parameters after the last adjustment are used as the second CSI reconstruction parameters.

[0234] Optionally, the processor 701 is specifically used for:

[0235] Using the second neural network and the proximal gradient iteration algorithm, the fifth CSI is predicted based on the first CSI reconstruction parameters, the environmental feature map, and the fourth CSI.

[0236] Optionally, the first preset condition includes:

[0237] The first error information is less than the second error information corresponding to the first channel prediction model, and the first similarity information is greater than the second similarity information corresponding to the first channel prediction model.

[0238] The first channel prediction model includes at least one of the following:

[0239] The first sub-channel prediction model includes: a randomly generated second pilot pattern and a CSI reconstruction neural network;

[0240] The second sub-channel prediction model includes: a randomly generated second pilot pattern, the image feature extraction neural network, and the CSI reconstruction neural network;

[0241] The third sub-channel prediction model includes the target pilot pattern and the CSI reconstruction neural network.

[0242] Optionally, the processor 701 is further configured to:

[0243] Obtain environmental information of the first wireless scenario;

[0244] Using modeling tools, a simulation environment for the first wireless scenario is constructed based on the environmental information.

[0245] Optionally, the processor 701 is specifically used for:

[0246] Using a ray tracing channel simulation tool, the second CSI of the receiver at the target location in the simulation environment is generated.

[0247] Optionally, the processor 701 is specifically used for:

[0248] Using an autonomous driving simulation platform, multi-view image data of the receiver at the target location is obtained in the simulation environment of the first wireless scenario.

[0249] Optionally, the target location may further include a third location;

[0250] The processor 701 is also used for:

[0251] Based on the second CSI at the third position and the target pilot pattern, the sixth CSI at the third position is obtained;

[0252] The image feature extraction neural network is used to extract image features from the multi-view image data at the third position;

[0253] By using the CSI reconstruction neural network, the complete seventh CSI at the third position is predicted based on the image features and the sixth CSI;

[0254] The accuracy of the target channel prediction model is obtained based on the third error information and the third similarity information between the seventh CSI and the second CSI at the third position.

[0255] Optionally, the processor 701 is further configured to:

[0256] Acquire complete eighth CSI data and multi-view image data of the receiver in the simulation environment of the second wireless scenario;

[0257] Based on the eighth CSI and the target pilot pattern, a portion of the ninth CSI in the simulation environment of the second wireless scenario is obtained;

[0258] The image feature extraction neural network is used to extract image features from the multi-view image data of the receiver in the simulation environment of the second wireless scenario;

[0259] By using the CSI reconstruction neural network, based on the image features and the ninth CSI, the complete tenth CSI in the simulation environment of the second wireless scenario is predicted.

[0260] Based on the fourth error information and fourth similarity information between the tenth CSI and the eighth CSI, the generalization applicability result of the target channel prediction model is obtained.

[0261] Among them, Figure 7 In this context, the bus architecture may include any number of interconnected buses and bridges, specifically linking various circuits together, represented by one or more processors (processor 701) and memory (memory 703). The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides a user interface 705. A transceiver 704 may be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. Processor 701 is responsible for managing the bus architecture and general processing, and memory 703 may store data used by processor 701 during operation.

[0262] An embodiment of the present invention provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the steps in the channel prediction method described above and achieve the same technical effect. To avoid repetition, further details are omitted here.

[0263] In this embodiment of the invention, the module can be implemented in software so that it can be executed by various types of processors. For example, an identified executable code module may include one or more physical or logical blocks of computer instructions, which may be constructed as objects, procedures, or functions. Nevertheless, the executable code of the identified module does not need to be physically located together, but may include different instructions stored in different bits, which, when logically combined, constitute the module and achieve the module's intended purpose.

[0264] In practice, an executable code module can be a single instruction or many instructions, and can even be distributed across multiple different code segments, different programs, and across multiple memory devices. Similarly, operational data can be identified within the module and can be implemented in any suitable form and organized within any suitable type of data structure. This operational data can be collected as a single dataset or distributed across different locations (including different storage devices), and can exist, at least in part, solely as electronic signals within the system or network.

[0265] When a module can be implemented using software, considering the current level of hardware technology, modules that can be implemented in software can be implemented using hardware circuits by those skilled in the art to achieve the corresponding functions, without considering cost. These hardware circuits include conventional very-large-scale integrated circuits (VLSI) or gate arrays, as well as existing semiconductors such as logic chips and transistors, or other discrete components. Modules can also be implemented using programmable hardware devices, such as field-programmable gate arrays, programmable array logic, and programmable logic devices.

[0266] The exemplary embodiments described above are with reference to the accompanying drawings. Many different forms and embodiments are feasible without departing from the spirit and teachings of the invention. Therefore, the invention should not be construed as limiting the exemplary embodiments set forth herein. Rather, these exemplary embodiments are provided to make the invention complete and convey the scope of the invention to those skilled in the art. In these drawings, component dimensions and relative dimensions may be exaggerated for clarity. The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. As used herein, unless clearly indicated otherwise, the singular forms “a,” “an,” and “the” are intended to include all such forms. It will be further understood that the terms “comprising” and / or “including”, when used in this specification, indicate the presence of the stated features, integers, steps, operations, components, and / or elements, but do not exclude the presence or addition of one or more other features, integers, steps, operations, components, and / or groups thereof. Unless otherwise indicated, when stated, a range of values ​​includes the upper and lower limits of the range and any subranges in between.

[0267] A specific embodiment of the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above-described functionality. Figure 1 The various processes of the method embodiments shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0268] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method of channel prediction, characterized by, The method comprises: obtaining a first channel state information (CSI) of a part in a first wireless scene according to a target pilot pattern, and obtaining multi-view image data of the first wireless scene; predicting a complete CSI of the first wireless scene according to the first CSI, the multi-view image data and a target channel prediction model; wherein the target channel prediction model comprises an image feature extraction neural network, the target pilot pattern and a CSI reconstruction neural network, the image feature extraction neural network is used to extract image features of the multi-view image data, and the CSI reconstruction neural network is used to predict the complete CSI according to the image features and the partial CSI; wherein the method further comprises: obtaining a complete second CSI of the first wireless scene in a simulation environment of a receiver at a target position, and obtaining multi-view image data of the receiver at the target position in the simulation environment of the first wireless scene, wherein the target position comprises a first position and a second position; training first image feature extraction parameters, a first pilot pattern and first CSI reconstruction parameters in a second neural network according to first network hyperparameters of a first neural network, the second CSI at the first position and the multi-view image data at the first position, to obtain second image feature extraction parameters, a second pilot pattern and second CSI reconstruction parameters in the second neural network; predicting a third CSI at the second position according to the second CSI at the second position, the second pilot pattern, the second image feature extraction parameters, the first network hyperparameters, multi-view image data at the second position and the second CSI reconstruction parameters in the second neural network; obtaining first error information and first similarity information between the second CSI and the third CSI; adjusting the first network hyperparameters if the first error information and the first similarity information do not satisfy a first preset condition; returning to the step of training the first image feature extraction parameters, the first pilot pattern and the first CSI reconstruction parameters in the second neural network according to the first network hyperparameters of the first neural network, the second CSI at the first position and the multi-view image data at the first position, until the first error information and the first similarity information satisfy the first preset condition, obtaining the image feature extraction neural network according to the first network hyperparameters after the last adjustment and the second image feature extraction parameters corresponding to the first network hyperparameters after the last adjustment, obtaining the target pilot pattern as the second pilot pattern corresponding to the first network hyperparameters after the last adjustment, and obtaining the CSI reconstruction neural network according to the second CSI reconstruction parameters in the second neural network corresponding to the first network hyperparameters after the last adjustment.

2. The method of claim 1, wherein, The first image feature extraction parameter, the first pilot pattern and the first CSI reconstruction parameter in the second neural network are trained according to the first network hyperparameter of the first neural network, the second CSI at the first position and the multi-view image data at the first position, to obtain a second image feature extraction parameter, a second pilot pattern and a second CSI reconstruction parameter in the second neural network, including: The first image feature extraction parameter, the first pilot pattern and the first CSI reconstruction parameter in the second neural network are trained according to the first network hyperparameter of the first neural network, the second CSI at the first position and the multi-view image data at the first position, to obtain a second image feature extraction parameter, a second pilot pattern and a second CSI reconstruction parameter in the second neural network, including: The environment feature map is obtained by performing feature extraction on the multi-view image data at the first position according to the first image feature extraction parameter by using the first neural network; The fourth CSI of part of the first position is obtained according to the first pilot pattern and the second CSI; The complete fifth CSI in the simulation environment of the first wireless scene is predicted according to the first CSI reconstruction parameter in the second neural network, the environment feature map and the fourth CSI; Loss information between the fifth CSI and the second CSI is obtained; The first image feature extraction parameter is adjusted according to the loss information, the first pilot pattern is adjusted according to the loss information and a deep probability sampling algorithm, the first CSI reconstruction parameter in the second neural network is adjusted according to the loss information, and the step of obtaining the environment feature map by performing feature extraction on the multi-view image data at the first position according to the first image feature extraction parameter by using the first neural network is returned until the number of iterations exceeds a preset number of times; 3. The method of claim 2, wherein, The last adjusted first image feature extraction parameter is taken as the second image feature extraction parameter, the last adjusted first pilot pattern is taken as the second pilot pattern, and the last adjusted first CSI reconstruction parameter in the second neural network is taken as the second CSI reconstruction parameter in the second neural network. The complete fifth CSI in the simulation environment of the first wireless scene is predicted according to the first CSI reconstruction parameter in the second neural network, the environment feature map and the fourth CSI, including:

4. The method of claim 1, wherein, The fifth CSI is predicted according to the first CSI reconstruction parameter in the second neural network, the environment feature map and the fourth CSI by using the second neural network and a proximal gradient iteration algorithm. The first preset condition includes: The first error information is less than second error information corresponding to a first channel prediction model, and the first similarity information is greater than second similarity information corresponding to the first channel prediction model; The first channel prediction model includes at least one of: A first sub-channel prediction model, the first sub-channel prediction model including a randomly generated second pilot pattern and a CSI reconstruction neural network; A second sub-channel prediction model, the second sub-channel prediction model including a randomly generated second pilot pattern, the image feature extraction neural network and a CSI reconstruction neural network; 5. The method of claim 1, wherein, A third sub-channel prediction model, the third sub-channel prediction model including the target pilot pattern and a CSI reconstruction neural network. The method further includes: obtain environment information of the first wireless scene; construct a simulation environment of the first wireless scene according to the environment information by using a modeling tool.

6. The method of claim 1, wherein, obtain the second CSI of a receiver at a target position in the simulation environment of the first wireless scene, including: generate the second CSI of the receiver at the target position in the simulation environment by using a ray tracing channel simulation tool.

7. The method of claim 1, wherein, obtain multi-view image data of the receiver at the target position in the simulation environment of the first wireless scene, including: obtain multi-view image data of the receiver at the target position in the simulation environment of the first wireless scene by using an automatic driving simulation platform.

8. The method of claim 1, wherein, The target position further includes a third position. The method further includes: obtain a sixth CSI of a part at the third position according to the second CSI at the third position and the target pilot pattern; extract image features of the multi-view image data at the third position by using the image feature extraction neural network; predict a complete seventh CSI at the third position according to the image features and the sixth CSI by using the CSI reconstruction neural network; obtain the accuracy of the target channel prediction model according to third error information and third similarity information between the seventh CSI and the second CSI at the third position.

9. The method of claim 1, wherein, The method further includes: obtain a complete eighth CSI in a simulation environment of a second wireless scene and multi-view image data of a receiver in the simulation environment of the second wireless scene; obtain a ninth CSI of a part in the simulation environment of the second wireless scene according to the eighth CSI and the target pilot pattern; extract image features of the multi-view image data of the receiver in the simulation environment of the second wireless scene by using the image feature extraction neural network; predict a complete tenth CSI in the simulation environment of the second wireless scene according to the image features and the ninth CSI by using the CSI reconstruction neural network; obtain the generalization applicability result of the target channel prediction model according to fourth error information and fourth similarity information between the tenth CSI and the eighth CSI.

10. A channel prediction device, characterized by, The device is applied to the channel prediction method in any one of claims 1 to 9, and the device includes: A first obtaining module is configured to obtain a part of first channel state information (CSI) at a first wireless scene and obtain multi-view image data of the first wireless scene according to a target pilot pattern. A first processing module is configured to predict a complete CSI at the first wireless scene according to the first CSI, the multi-view image data and a target channel prediction model. The target channel prediction model includes an image feature extraction neural network, the target pilot pattern and a CSI reconstruction neural network, the image feature extraction neural network is configured to extract image features of the multi-view image data, and the CSI reconstruction neural network is configured to predict a complete CSI according to image features and a part of CSI.

11. A channel prediction device comprising: A transceiver, a processor, a memory, and a program or instructions stored on the memory and executable on the processor; wherein the processor implements the steps in the channel prediction method according to any one of claims 1 to 9 when executing the program or instructions.

12. A readable storage medium, having stored thereon a program or instructions, characterized in that, The program or instructions implement the steps in the channel prediction method according to any one of claims 1 to 9 when executed by a processor.

13. A computer program product, characterised in that, The program or instructions implement the steps in the channel prediction method according to any one of claims 1 to 9 when executed by a processor. The program or instructions implement the steps in the channel prediction method according to any one of claims 1 to 9 when executed by a processor.

Images