A deep learning-based predictive channel modeling method and related apparatus

Abstract

The application discloses a kind of prediction channel modeling methods based on deep learning, comprising: collecting channel data under different frequency bands;From the channel data, extract multipath parameters;The multipath parameters are cluster identified, and each type of multipath cluster is obtained;The channel prediction network is trained by the channel data of each type of multipath cluster;By the channel prediction network trained and the channel data under known frequency band, the channel data under predicted frequency band is obtained by prediction.The method can realize high-precision frequency domain channel prediction.The application also discloses a kind of prediction channel modeling device based on deep learning, equipment and computer readable storage medium, all have the above technical effects.

Description

A deep learning-based predictive channel modeling method and related apparatus Technical Field

[0001] This application relates to the field of communication technology, and in particular to a deep learning-based predictive channel modeling method; it also relates to a deep learning-based predictive channel modeling apparatus, device, and computer-readable storage medium. Background Technology

[0002] 6G communication technology will expand upon 5G networks to encompass massive bandwidth, massive connectivity, wide coverage, and high intelligence, further deepening mobile internet, the Internet of Things, and intelligent communication. The measurement, characteristic analysis, and modeling of wireless communication channels are fundamental to the design, performance evaluation, optimization, and deployment of wireless communication systems. Predictive channel modeling includes key steps such as channel measurement, channel parameter extraction, channel characteristic analysis, modeling, and prediction. Existing technologies can only analyze channel characteristics for some known frequency bands and scenarios, failing to fully explore the complex relationships between channel characteristics and frequency bands and scenarios, and unable to predict channel characteristics for unknown scenarios and frequency bands. This makes them unsuitable for the technological vision of 6G channel modeling across all frequency bands, full coverage scenarios, and all application scenarios. Furthermore, current AI (Artificial Intelligence)-based channel parameter prediction and modeling methods typically learn from training datasets of independent channel characteristics, such as received power, delay spread, and model angle information. The AI ​​algorithms used in these methods are generally deep learning network algorithms such as feedforward neural networks, radial basis function neural networks, and convolutional neural networks. However, these AI-based channel parameter estimations of channel measurement data first increase the computational complexity of channel prediction, and the independent channel parameter index values ​​obtained after prediction cannot effectively and directly reflect the continuous channel characteristics in the time and frequency domain.

[0003] Therefore, how to achieve high-precision frequency domain channel prediction has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] The purpose of this application is to provide a deep learning-based predictive channel modeling method capable of achieving high-precision frequency domain channel prediction. Another purpose of this application is to provide a deep learning-based predictive channel modeling apparatus, device, and computer-readable storage medium.

[0005] To address the aforementioned technical problems, this application provides a deep learning-based predictive channel modeling method, comprising:

[0006] Collect channel data in different frequency bands;

[0007] Extract multipath parameters from the channel data;

[0008] Cluster identification is performed on the multipath parameters to obtain various multipath clusters;

[0009] The channel prediction network is trained using the channel data from the various multipath clusters described above;

[0010] By using the trained channel prediction network and channel data in the known frequency band, channel data in the predicted frequency band can be predicted.

[0011] Optionally, the multipath cluster includes:

[0012] Direct multipath clusters, single-hop multipath clusters, and multi-hop multipath clusters.

[0013] Optionally, the channel prediction network includes: an autoencoder and a convolutional gated recurrent unit;

[0014] Accordingly, each type of multipath cluster corresponds to a set of autoencoders and convolutional gated loop units.

[0015] Optionally, training the channel prediction network using channel data from various multipath clusters includes:

[0016] The channel prediction network is trained using channel data from various multipath clusters in the first and second frequency bands to obtain a channel prediction network for predicting channel data in the third frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

[0017] Optionally, training the channel prediction network using channel data from various multipath clusters includes:

[0018] The channel prediction network is trained using channel data from various multipath clusters in the first and third frequency bands to obtain a channel prediction network for predicting channel data in the second frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

[0019] Optionally, training the channel prediction network using channel data from various multipath clusters includes:

[0020] The channel prediction network is trained using channel data from various multipath clusters in the second and third frequency bands to obtain a channel prediction network for predicting channel data in the first frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

[0021] Optional, also includes:

[0022] Verify the accuracy of the channel prediction network;

[0023] Adjust the parameters of the channel prediction network based on the verification results.

[0024] To address the aforementioned technical problems, this application also provides a deep learning-based predictive channel modeling apparatus, comprising:

[0025] The acquisition module is used to acquire channel data in different frequency bands;

[0026] Extraction module, used to extract multipath parameters from the channel data;

[0027] The identification module is used to perform cluster identification on the multipath parameters to obtain various multipath clusters;

[0028] The training module is used to train the channel prediction network using channel data from various types of multipath clusters;

[0029] The prediction module is used to predict channel data in the predicted frequency band by using the trained channel prediction network and channel data in the known frequency band.

[0030] To address the aforementioned technical problems, this application also provides a deep learning-based predictive channel modeling device, comprising:

[0031] Memory, used to store computer programs;

[0032] A processor for executing the computer program to implement the steps of the deep learning-based predictive channel modeling method as described in any of the preceding claims.

[0033] To address the aforementioned technical problems, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the deep learning-based predictive channel modeling method described in any of the preceding claims.

[0034] The deep learning-based predictive channel modeling method provided in this application includes: collecting channel data in different frequency bands; extracting multipath parameters from the channel data; performing cluster identification on the multipath parameters to obtain various multipath clusters; training a channel prediction network using the channel data of the various multipath clusters; and predicting channel data in the predicted frequency band using the trained channel prediction network and the channel data in the known frequency band.

[0035] As can be seen, the deep learning-based predictive channel modeling method provided in this application performs multi-band channel measurement, extracts multipath parameters and identifies multipath clusters, fully considers the channel characteristics of multipath clusters, and can achieve high-precision frequency domain channel prediction. It breaks through the technical bottleneck that traditional channel prediction models are only applicable to channel simulation and characteristic analysis of known frequency bands and known scenarios.

[0036] The deep learning-based predictive channel modeling apparatus, device, and computer-readable storage medium provided in this application also have the aforementioned technical effects. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the prior art and embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 is a flowchart illustrating a deep learning-based predictive channel modeling method provided in an embodiment of this application.

[0039] Figure 2 is a schematic diagram of a propagation path provided in an embodiment of this application;

[0040] Figure 3 is a schematic diagram of a channel prediction network based on a convolutional gated recurrent unit and an autoencoder provided in an embodiment of this application;

[0041] Figure 4 is a schematic diagram of a convolution gated loop unit provided in an embodiment of this application;

[0042] Figure 5 is a schematic diagram of a predictive channel modeling provided in an embodiment of this application;

[0043] Figure 6 is a schematic diagram of another predictive channel modeling provided in an embodiment of this application;

[0044] Figure 7 is a schematic diagram of another predictive channel modeling provided in an embodiment of this application;

[0045] Figure 8 is a schematic diagram of a deep learning-based predictive channel modeling device provided in an embodiment of this application;

[0046] Figure 9 is a schematic diagram of a deep learning-based predictive channel modeling device provided in an embodiment of this application. Detailed Implementation

[0047] The core of this application is to provide a deep learning-based predictive channel modeling method that can achieve high-precision frequency domain channel prediction. Another core aspect of this application is to provide a deep learning-based predictive channel modeling apparatus, device, and computer-readable storage medium.

[0048] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0049] Please refer to Figure 1, which is a flowchart illustrating a deep learning-based predictive channel modeling method provided in an embodiment of this application. As shown in Figure 1, the method includes:

[0050] S101: Collect channel data under different frequency bands;

[0051] Channel data at different frequency bands can be measured in typical wireless communication scenarios. For example, in a typical urban setting, a high-precision channel detector can be used to measure channel data at multiple locations in the 0.7 GHz, 2.3 GHz, and 3.7 GHz bands. In actual cross-band measurement activities, the antennas and communication bandwidths of the transmitting and receiving ends can be configured differently. For example, the transmitting end can use a single antenna, the receiving end can use a 4×4 array receiving antenna, and the communication bandwidth can be 100 MHz.

[0052] S102: Extract multipath parameters from the channel data;

[0053] Based on the acquired channel data, multipath parameters are extracted using a high-precision channel parameter estimation algorithm. Multipath parameters mainly include power, time delay, horizontal angle of arrival (Angle of Arrival) or horizontal angle of departure (Angle of Departure), and pitch angle of arrival (Angle of Arrival) or pitch angle of departure (Angle of Departure). The high-precision channel parameter estimation algorithm can be either the spatial alternation generalized maximization algorithm or the Richard expectation-maximization algorithm. Alternatively, a Bartlett filter can be used to obtain the angular power spectral density.

[0054] S103: Perform cluster identification on the multipath parameters to obtain various multipath clusters;

[0055] The extracted multipath parameters are used to identify clusters, resulting in various types of multipath clusters. The multipath cluster identification algorithm can be either the K-power-means algorithm or a standard K-means algorithm, with the K-power-means algorithm being the preferred choice.

[0056] Additionally, in some embodiments, the multipath cluster includes:

[0057] Direct multipath clusters, single-hop multipath clusters, and multi-hop multipath clusters.

[0058] As shown in Reference 2, based on the propagation path of the multipath cluster, the types of multipath clusters can include line-of-sight cluster (LoSC), single bounce reflection cluster (SC), and multiple bounce reflection cluster (MC).

[0059] S104: Train the channel prediction network using the channel data of the various multipath clusters;

[0060] A channel prediction network is established and trained using channel data from various multipath clusters.

[0061] In some embodiments, the channel prediction network includes: an autoencoder and a convolutional gated recurrent unit;

[0062] Accordingly, each type of multipath cluster corresponds to a set of autoencoders and convolutional gated loop units.

[0063] In this embodiment, the channel prediction network includes an autoencoder and a convolution-gated recurrent unit (Conv-GRU). Specifically, this embodiment establishes a channel prediction network based on a convolution-gated recurrent unit (Conv-GRU) network and an autoencoder. The overall structure of the channel prediction network is shown in Figure 3. This channel prediction network includes an encoder and Conv-GRU units. Each type of multipath cluster corresponds to a set of encoders and Conv-GRU units. The loss function can be any (Mean absolute error, MAE), and the activation function is the rectified linear unit (ReLU).

[0064] Referring to Figure 4, the Conv-GRU unit includes two Conv-GRU subnetworks, which transmit the predicted channel matrix laterally and input the frequency point f vertically. c1,i and f c2,i Frequency point f c1,i and f c2,i They belong to the measured frequency band f respectively c1 and predicted frequency band f c2 The outputs of the three Conv-GRU units are summed and then input into the decoder to output the final prediction result.

[0065] In Figure 4, z t This indicates the update gate function, r tσ represents the reset gate function, and σ represents the sigmoid function.

[0066] The parameter configuration for the channel prediction network is shown in Table 1.

[0067] Table 1

[0068] Parameter values: Encoder and Decoder layers: 3, 3; Auto-encoder hidden layer dimensions: 10, 24, 5, 12, 128; Auto-encoder embedding dimensions: 90×90×4, 45×45×16, 15×15×144; Maximum number of iterations: 500; Loss function: MAE coefficient; Regularization coefficient: 0.2; Conv-GRU layers: 1; Conv-GRU node count: 2; Convolutional kernel size: 3×3; Padding: SAME; Streide: 1; Batch dimensions: 32. surface

[0069] As shown in Figure 3, when the channel prediction network is trained using the angular power spectral density of various multipath clusters, the angular power spectral density of each multipath cluster is input into a corresponding set of encoders and Conv-GRU units. The outputs of the three sets of Conv-GRU units are summed and then input into the decoder, which outputs the final prediction result H(f). c2 f c2 The predicted frequency band is represented by H(M). The angular power spectral density of a direct multipath cluster is expressed as H(M). LoSC ,f c1 The angular power spectral density of a single-hop reflection multipath cluster is expressed as H(M). SC ,f c1 The angular power spectral density of a multihop reflection multipath cluster is expressed as H(M). MC ,f c1 f c1 This indicates the frequency band that has been measured.

[0070] Using convolutionally gated recurrent units (GRUs) enables small-scale fading feature extraction from multipath clusters, while using autoencoders enables large-scale fading feature extraction, thus achieving high-precision channel prediction. Compared to general long short-term memory (LSTM) networks, GRU-based prediction networks, and LSTM-based prediction networks, the channel prediction network based on autoencoders and GRUs used in this embodiment exhibits the smallest prediction error and higher accuracy.

[0071] In some embodiments, training the channel prediction network using channel data from various multipath clusters includes:

[0072] The channel prediction network is trained using channel data from various multipath clusters in the first and second frequency bands to obtain a channel prediction network for predicting channel data in the third frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

[0073] In this embodiment, the channel data used to train the channel prediction network are low-frequency and mid-frequency channel data, resulting in a channel prediction network for predicting high-frequency channel data. For example, channel data at 0.7 GHz and 2.3 GHz are used as training data to predict channel data at 3.7 GHz.

[0074] In some embodiments, training the channel prediction network using channel data from various multipath clusters includes:

[0075] The channel prediction network is trained using channel data from various multipath clusters in the first and third frequency bands to obtain a channel prediction network for predicting channel data in the second frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

[0076] In this embodiment, the channel data used to train the channel prediction network are low-frequency and high-frequency channel data, resulting in a channel prediction network used to predict mid-frequency channel data. For example, channel data at 0.7 GHz and 3.7 GHz are used as training data to predict channel data at 2.3 GHz.

[0077] In some embodiments, training the channel prediction network using channel data from various multipath clusters includes:

[0078] The channel prediction network is trained using channel data from various multipath clusters in the second and third frequency bands to obtain a channel prediction network for predicting channel data in the first frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

[0079] In this embodiment, the channel data used to train the channel prediction network are mid-frequency and high-frequency channel data, resulting in a channel prediction network used to predict low-frequency channel data. For example, channel data at 2.3 GHz and 3.7 GHz are used as training data to predict channel data at 0.7 GHz.

[0080] S105: Using the trained channel prediction network and the channel data under the known frequency band, predict the channel data under the predicted frequency band.

[0081] When the channel prediction network is trained using mid- and low-frequency channel data, extrapolation prediction can be achieved through the channel prediction network trained on the training data to predict high-frequency channel data.

[0082] For example, as shown in Figure 5, channel data at 0.7 GHz and 2.3 GHz are used as training data to predict channel data at 3.7 GHz.

[0083] When the channel prediction network is trained using mid- and high-frequency channel data, extrapolation prediction can be achieved through the channel prediction network trained on the training data to predict low-frequency channel data.

[0084] For example, as shown in Figure 6, channel data at 2.3 GHz and 3.7 GHz are used as training data to predict channel data at 0.7 GHz.

[0085] When training a channel prediction network using low- and high-frequency channel data as training data, the channel prediction network using the training data can achieve inference prediction to predict mid-frequency channel data.

[0086] For example, as shown in Figure 7, channel data at 0.7 GHz and 3.7 GHz are used as training data to predict channel data at 2.3 GHz.

[0087] Furthermore, in some embodiments, it also includes:

[0088] Verify the accuracy of the channel prediction network and adjust the parameters of the channel prediction network based on the verification results.

[0089] The root mean square error (RMSE) and mean absolute percentage error (MAS) can be used for accuracy verification. The accuracy of the prediction is evaluated by calculating the RMSE and MAS of the measured channel data and the predicted channel data, and then the parameters shown in Table 1 are adjusted according to the accuracy.

[0090] In summary, the deep learning-based predictive channel modeling method provided in this application includes: collecting channel data in different frequency bands; extracting multipath parameters from the channel data; performing cluster identification on the multipath parameters to obtain various multipath clusters; training a channel prediction network using the channel data of the various multipath clusters; and predicting channel data in the predicted frequency band using the trained channel prediction network and channel data in a known frequency band. This method performs multi-frequency band channel measurement, extracts multipath parameters, and performs multipath cluster identification, fully considering the channel characteristics of multipath clusters. It can achieve high-precision frequency domain channel prediction, breaking through the technical bottleneck that traditional channel prediction models are only applicable to channel simulation and characteristic analysis in known frequency bands and known scenarios.

[0091] This application also provides a deep learning-based predictive channel modeling apparatus, which can be referred to in conjunction with the method described above. Please refer to Figure 8, which is a schematic diagram of a deep learning-based predictive channel modeling apparatus provided in an embodiment of this application. As shown in Figure 8, the apparatus includes:

[0092] Acquisition module 10 is used to acquire channel data in different frequency bands;

[0093] Extraction module 20 is used to extract multipath parameters from the channel data;

[0094] The identification module 30 is used to perform cluster identification on the multipath parameters to obtain various multipath clusters;

[0095] Training module 40 is used to train a channel prediction network using channel data from various types of multipath clusters;

[0096] The prediction module 50 is used to predict channel data in the predicted frequency band by using the trained channel prediction network and channel data in the known frequency band.

[0097] Based on the above embodiments, as a specific implementation method, the multipath cluster includes:

[0098] Direct multipath clusters, single-hop multipath clusters, and multi-hop multipath clusters.

[0099] Based on the above embodiments, as a specific implementation method, the channel prediction network includes: an autoencoder and a convolutional gated recurrent unit;

[0100] Accordingly, each type of multipath cluster corresponds to a set of autoencoders and convolutional gated loop units.

[0101] Based on the above embodiments, as a specific implementation method, the training module 40 is specifically used for:

[0102] The channel prediction network is trained using channel data from various multipath clusters in the first and second frequency bands to obtain a channel prediction network for predicting channel data in the third frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

[0103] Based on the above embodiments, as a specific implementation method, the training module 40 is specifically used for:

[0104] The channel prediction network is trained using channel data from various multipath clusters in the first and third frequency bands to obtain a channel prediction network for predicting channel data in the second frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

[0105] Based on the above embodiments, as a specific implementation method, the training module 40 is specifically used for:

[0106] The channel prediction network is trained using channel data from various multipath clusters in the second and third frequency bands to obtain a channel prediction network for predicting channel data in the first frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

[0107] Based on the above embodiments, as a specific implementation method, it further includes:

[0108] A verification module is used to verify the accuracy of the channel prediction network;

[0109] The parameter tuning module is used to adjust the parameters of the channel prediction network based on the verification results.

[0110] The deep learning-based predictive channel modeling device provided in this application performs multi-band channel measurements, extracts multipath parameters, and identifies multipath clusters. It fully considers the channel characteristics of multipath clusters and can achieve high-precision frequency domain channel prediction. This breaks through the technical bottleneck that traditional channel prediction models are only applicable to channel simulation and characteristic analysis of known frequency bands and known scenarios.

[0111] This application also provides a deep learning-based predictive channel modeling device, as shown in Figure 9, which includes a memory 1 and a processor 2.

[0112] Memory 1 is used to store computer programs;

[0113] Processor 2 is used to execute computer programs to perform the following steps:

[0114] Collect channel data in different frequency bands; extract multipath parameters from the channel data; perform cluster identification on the multipath parameters to obtain various multipath clusters; train a channel prediction network using the channel data of the various multipath clusters; and predict the channel data in the predicted frequency band using the trained channel prediction network and the channel data in the known frequency band.

[0115] For a description of the equipment provided in this application, please refer to the above method embodiments; further details will not be provided here.

[0116] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the following steps:

[0117] Collect channel data in different frequency bands; extract multipath parameters from the channel data; perform cluster identification on the multipath parameters to obtain various multipath clusters; train a channel prediction network using the channel data of the various multipath clusters; and predict the channel data in the predicted frequency band using the trained channel prediction network and the channel data in the known frequency band.

[0118] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0119] For a description of the computer-readable storage medium provided in this application, please refer to the above method embodiments; further details will not be repeated here.

[0120] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatuses, devices, and computer-readable storage media disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant details can be found in the method section.

[0121] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0122] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0123] The foregoing has provided a detailed description of the deep learning-based predictive channel modeling method, apparatus, device, and computer-readable storage medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A deep learning-based predictive channel modeling method, characterized in that, include: Collect channel data in different frequency bands in cross-frequency band scenarios; Extract multipath parameters from the channel data; Cluster identification is performed on the multipath parameters to obtain various multipath clusters; a channel prediction network is trained using the channel data of each of the various multipath clusters; the channel data in the predicted frequency band is predicted using the trained channel prediction network and the channel data in the known frequency band; the channel prediction network includes: an autoencoder and a convolutional gated recurrent unit; correspondingly, each of the various multipath clusters corresponds to a set of the autoencoder and the convolutional gated recurrent unit; The convolutional gated recurrent unit (CRU) consists of two Conv-GRU subnetworks. The CRU propagates the prediction channel matrix laterally and takes into account the frequency points belonging to the measured frequency band and the frequency points belonging to the prediction frequency band, respectively, in the vertical direction. The outputs of the CRU are summed and then input to the decoder, which outputs the final prediction result.

2. The predictive channel modeling method according to claim 1, characterized in that, The multipath clusters include: direct multipath clusters, single-hop reflection multipath clusters, and multi-hop reflection multipath clusters.

3. The predictive channel modeling method according to claim 1, characterized in that, The step of training the channel prediction network using channel data from various multipath clusters includes: training the channel prediction network using channel data from various multipath clusters in a first frequency band and a second frequency band to obtain a channel prediction network for predicting channel data in a third frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

4. The predictive channel modeling method according to claim 1, characterized in that, The step of training the channel prediction network using channel data from various multipath clusters includes: training the channel prediction network using channel data from various multipath clusters in a first frequency band and a third frequency band to obtain a channel prediction network for predicting channel data in a second frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

5. The predictive channel modeling method according to claim 1, characterized in that, The step of training the channel prediction network using channel data from various multipath clusters includes: training the channel prediction network using channel data from various multipath clusters in the second and third frequency bands to obtain the channel prediction network for predicting channel data in the first frequency band; the second frequency band is higher than the first frequency band and lower than the third frequency band.

6. The predictive channel modeling method according to claim 1, characterized in that, Also includes: Verify the accuracy of the channel prediction network; adjust the parameters of the channel prediction network based on the verification results.

7. A predictive channel modeling device based on deep learning, characterized in that, include: The acquisition module is used to acquire channel data in different frequency bands in cross-frequency band scenarios; Extraction module, used to extract multipath parameters from the channel data; The identification module is used to perform cluster identification on the multipath parameters to obtain various multipath clusters; The training module is used to train a channel prediction network using channel data from various multipath clusters; the prediction module is used to predict channel data in a predicted frequency band using the trained channel prediction network and channel data in a known frequency band; the channel prediction network includes an autoencoder and a convolutional gated recurrent unit; correspondingly, each multipath cluster corresponds to a set of the autoencoder and the convolutional gated recurrent unit. The convolutional gated recurrent unit (CRU) consists of two Conv-GRU subnetworks. The CRU propagates the prediction channel matrix laterally and takes into account the frequency points belonging to the measured frequency band and the frequency points belonging to the prediction frequency band, respectively, in the vertical direction. The outputs of the CRU are summed and then input to the decoder, which outputs the final prediction result.

8. A predictive channel modeling device based on deep learning, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the deep learning-based predictive channel modeling method as described in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the deep learning-based predictive channel modeling method as described in any one of claims 1 to 6.

Images