Channel modeling method and device, computer equipment and readable storage medium

Abstract

The invention relates to a channel modeling method and device, computer equipment and a readable storage medium. The method comprises the following steps: acquiring a sending signal sample and a real signal corresponding to the sending signal sample; simulating a first sending signal sample in the sending signal samples transmitted through the communication channel through a generator model in the generative adversarial network model to obtain a first generation signal; training a discriminator model in a generative adversarial network model based on the first sending signal sample, the first generation signal and a real signal corresponding to the first sending signal sample; simulating a second sending signal sample in the sending signal samples transmitted through the communication channel through the generator model to obtain a second generation signal; and performing signal source classification on the second generation signal through the trained discriminator model to obtain a sample classification result, and updating the generator model based on the sample classification result to obtain a trained generator model. By adopting the method, the communication channel can be accurately simulated.

Description

Technical Field

[0001] This application relates to the field of laser communication technology, and in particular to a channel modeling method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] Free-space optical (FSO) communication, as a core technology for optical wireless transmission, has become an important solution for 5G (5th Generation Mobile Communication Technology) backhaul and inter-satellite links due to its advantages such as GHz (gigahertz) bandwidth, Tbps (terabits per second) transmission rate, deployment flexibility, and inherent resistance to electromagnetic interference. Currently, the global FSO market is expanding rapidly, and its technological maturity is accelerating from the laboratory to large-scale commercial application. However, in FSO communication systems, factors such as atmospheric turbulence in the transmission link can cause signal distortion and a decrease in transmission quality. Traditional technologies have limited capabilities in characterizing complex channel effects, resulting in low accuracy in simulating actual communication channels. Summary of the Invention

[0003] Therefore, it is necessary to provide a channel modeling method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can accurately simulate communication channels to address the aforementioned technical problems.

[0004] Firstly, this application provides a channel modeling method, including:

[0005] Acquire the transmitted signal sample and the corresponding real signal. The real signal is the signal obtained at the receiving end after the transmitted signal sample is sent from the transmitting end and transmitted through the communication channel.

[0006] By using the generator model in the generative adversarial network model, the first transmitted signal sample in the transmitted signal sample is simulated through the communication channel to obtain the first generated signal;

[0007] Based on the first transmitted signal sample, the first generated signal and the real signal corresponding to the first transmitted signal sample, the discriminator model in the generative adversarial network model is trained to obtain the trained discriminator model.

[0008] By using a generator model, the second transmitted signal sample in the transmitted signal sample is simulated through the communication channel to obtain the second generated signal;

[0009] The trained discriminator model is used to classify the source of the second generated signal to obtain the sample classification result. The generator model is then updated based on the sample classification result to obtain the trained generator model, which is used to simulate the communication channel.

[0010] In some embodiments, training a discriminator model in a generative adversarial network (GAN) model based on a first transmitted signal sample, a first generated signal, and a real signal corresponding to the first transmitted signal sample to obtain a trained discriminator model includes: inputting the first transmitted signal sample and the first generated signal into the discriminator model in the GAN model to classify the signal source and obtain a first classification result; inputting the first transmitted signal sample and the real signal corresponding to the first transmitted signal sample into the discriminator model to classify the signal source and obtain a second classification result; and updating the discriminator model based on the first classification result and the second classification result to obtain a trained discriminator model.

[0011] In some embodiments, the trained discriminator model is used to classify the signal source of the second generated signal to obtain a sample classification result, and the generator model is updated based on the sample classification result to obtain a trained generator model. This includes: inputting the second generated signal and the second transmitted signal sample into the trained discriminator model to classify the signal source to obtain a sample classification result; determining the generator loss based on the sample classification result; and updating the generator model based on the generator loss to obtain a trained generator model.

[0012] In some embodiments, the first classification result includes a first classification distribution, the second classification result includes a second classification distribution, and the sample classification result includes a third classification distribution; updating the discriminator model based on the first and second classification results to obtain a trained discriminator model includes: determining the discriminator loss based on the distance between the first and second classification distributions and the gradient penalty term; updating the discriminator model according to the discriminator loss to obtain a trained discriminator model; determining the generator loss based on the sample classification results includes: determining the generator loss based on the third classification distribution.

[0013] In some embodiments, the transmitted signal sample includes a first transmitted signal sample and a second transmitted signal sample; by simulating the transmission of the first transmitted signal sample through a communication channel using a generator model in a generative adversarial network model, a first generated signal is obtained, including: inputting the first transmitted signal sample and a first random noise signal into the generator model in the generative adversarial network model to simulate transmission through a communication channel, and obtaining the first generated signal output by the generator model; by simulating the transmission of the second transmitted signal sample through a communication channel using the generator model, a second generated signal is obtained, including: inputting the second transmitted signal sample and the second random noise signal into the generator model to simulate transmission through a communication channel, and obtaining the second generated signal output by the generator model.

[0014] In some embodiments, updating the generator model based on the sample classification results to obtain the trained generator model includes: updating the generator model based on the sample classification results to obtain the updated generator model; if the training termination condition is not met, returning to the step of obtaining the transmitted signal sample and the real signal corresponding to the transmitted signal sample, so as to perform the next training on the updated generator model until the training termination condition is met to obtain the trained generator model.

[0015] In some embodiments, the channel modeling method further includes: acquiring a target transmitted signal; inputting the target transmitted signal into a trained generator model so that the trained generator model can simulate the transmission of the target transmitted signal through the communication channel, and outputting a simulated received signal corresponding to the target transmitted signal.

[0016] Secondly, this application also provides a channel modeling apparatus, comprising:

[0017] The signal sample acquisition module is used to acquire the transmitted signal sample and the corresponding real signal. The real signal is the signal obtained at the receiving end after the transmitted signal sample is sent from the transmitting end and transmitted through the communication channel.

[0018] The first signal generation module is used to simulate the first transmitted signal sample in the transmitted signal sample transmitted through the communication channel by using the generator model in the generative adversarial network model to obtain the first generated signal.

[0019] The discriminator training module is used to train the discriminator model in the generative adversarial network model based on the first transmitted signal sample, the first generated signal and the real signal corresponding to the first transmitted signal sample, and to obtain the trained discriminator model.

[0020] The second signal generation module is used to simulate the second transmitted signal sample in the transmitted signal sample through the communication channel by using a generator model to obtain the second generated signal;

[0021] The generator training module is used to classify the source of the second generated signal using the trained discriminator model, obtain the sample classification result, and update the generator model based on the sample classification result to obtain the trained generator model. The trained generator model is used to simulate the communication channel.

[0022] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the method provided in the first aspect above.

[0023] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method provided in the first aspect above.

[0024] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the method provided in the first aspect above.

[0025] The aforementioned channel modeling method, apparatus, computer equipment, computer-readable storage medium, and computer program product, through a generator model in a generative adversarial network (GAN) model, simulate the transmission of a first transmitted signal sample from a transmitted signal sample through a communication channel to obtain a first generated signal. Based on the first transmitted signal sample, the first generated signal, and the corresponding real signal, a discriminator model in the GAN model is trained to obtain a trained discriminator model. The generator model then simulates the transmission of a second transmitted signal sample from a transmitted signal sample through the communication channel to obtain a second generated signal. The trained discriminator model classifies the signal source of the second generated signal to obtain a sample classification result, and updates the generator model based on the sample classification result to obtain a trained generator model for simulating the communication channel. The generator model and discriminator model used to simulate the communication channel are obtained through adversarial training, enabling the generator model to learn the deterministic characteristics and random perturbations of the communication channel, thereby accurately simulating the actual communication channel. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1This is a diagram illustrating the application environment of the channel modeling method in one embodiment;

[0028] Figure 2 This is a flowchart illustrating a channel modeling method in one embodiment;

[0029] Figure 3 This is a schematic diagram of the discriminator model training process in one embodiment;

[0030] Figure 4 This is a schematic diagram of the structure of an FSO communication system based on JEM in one embodiment;

[0031] Figure 5 This is a schematic diagram of the GAN channel model structure in one embodiment;

[0032] Figure 6(a) is a schematic diagram of the distribution of GAN output data under weak turbulence in one embodiment;

[0033] Figure 6(b) is a schematic diagram of the GAN output data distribution under medium-strong turbulence in one embodiment;

[0034] Figure 7(a) is a schematic diagram of the fitting results of weak turbulence under 2km conditions in one embodiment;

[0035] Figure 7(b) is a schematic diagram of the fitting results of medium-strong turbulence under 2km conditions in one embodiment;

[0036] Figure 7(c) is a schematic diagram of the fitting results of weak turbulence under 5km conditions in one embodiment;

[0037] Figure 7(d) is a schematic diagram of the fitting results of medium-strong turbulence under 5km conditions in one embodiment;

[0038] Figure 8 This is a structural block diagram of a channel modeling apparatus in one embodiment;

[0039] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0041] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0042] The channel modeling method provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed in the cloud or on another network server. Terminal 102 can transmit transmitted signal samples and corresponding real signals to server 104. The real signals are the signals obtained at the receiving end after the transmitted signal samples are sent from the sending end and transmitted through the communication channel. The transmitted signal samples may include a first transmitted signal sample and a second transmitted signal sample, and the corresponding real signals may include the real signals corresponding to the first transmitted signal sample and the real signals corresponding to the second transmitted signal sample. After obtaining the transmitted signal samples and their corresponding real signals, server 104 can simulate the transmission of the first transmitted signal sample through the communication channel using a generator model in a generative adversarial network (GAN) model to obtain a first generated signal. Server 104 can then train a discriminator model in the GAN model based on the first transmitted signal sample, the first generated signal, and the corresponding real signal to obtain the trained discriminator model. Server 104 can simulate the second transmitted signal sample in the transmitted signal sample through the communication channel using a generator model to obtain a second generated signal. The trained discriminator model is then used to classify the signal source of the second generated signal to obtain a sample classification result. Based on the sample classification result, the generator model is updated to obtain a trained generator model, which is then used to simulate the communication channel.

[0043] Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection equipment. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0044] In one exemplary embodiment, such as Figure 2 As shown, a channel modeling method is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking the server in the example, the explanation includes the following steps 202 to 210. Wherein:

[0045] Step 202: Obtain the transmitted signal sample and the corresponding real signal. The real signal is the signal obtained at the receiving end after the transmitted signal sample is sent from the transmitting end and transmitted through the communication channel.

[0046] The transmitted signal sample can be determined from the transmitted signals of the communication channel and used as sample data to train the channel simulation model. The transmitted signal can be transmitted through the communication channel. For example, the transmitting end can input the transmitted signal at the first end of the communication channel, and after transmission through the communication channel, the receiving end can obtain the received signal at the second end of the communication channel. The real signal corresponds to the transmitted signal sample; the real signal can be the signal received at the receiving end after the transmitted signal sample is transmitted from the transmitting end through the communication channel. That is, the transmitted signal sample is the signal sent by the transmitting end before transmission through the communication channel, and the real signal is the signal received by the receiving end after transmission through the communication channel. A communication channel is the channel through which signals are transmitted in a communication system. Its core is the physical or logical path connecting the transmitting end and the receiving end, enabling information transmission. A communication channel can be a visible physical line or an invisible wireless frequency band. The type of communication channel can be determined by the transmission medium; different media correspond to different channel characteristics, such as transmission speed and anti-interference capability.

[0047] For example, the communication channel being modeled can include various types, such as wired channels that transmit data via physical cables like network cables, fiber optic cables, and telephone lines, and wireless channels that transmit data via electromagnetic waves like cellular channels and Bluetooth channels. For instance, the communication channel can include a free-space optical (FSO) channel within the wireless channel. The server can obtain transmitted signal samples and their corresponding real signals. The real signals are the signals obtained after the transmitted signal samples have been transmitted through the communication channel. Based on the transmitted signal samples and the corresponding real signals, the communication channel can be modeled.

[0048] Step 204: By using the generator model in the generative adversarial network model, the first transmitted signal sample in the transmitted signal sample is simulated through the communication channel to obtain the first generated signal.

[0049] Generative Adversarial Networks (GANs) are deep learning models that learn data distributions through adversarial interactions between a generator model and a discriminator model, ultimately generating new data highly similar to real data. The generator model in a GAN can simulate a communication channel; for example, it can output a corresponding generated signal based on the input signal. The generated signal can simulate the signal obtained after transmitting the input signal through the communication channel. The first generated signal is the signal obtained by the generator model simulating the transmission of a first sample of the transmitted signal through the communication channel.

[0050] Optionally, the server can obtain a generative adversarial network (GAN) model, which may include a generator model. The generator model can be used to simulate a communication channel; that is, the original signal input to the generator model can correspond to the transmitted signal of the communication channel, and the generated signal output by the generator model can correspond to the received signal of the communication channel. Thus, a signal can be input into the generator model to simulate the output of the input signal through the communication channel, resulting in the transmitted signal. The server can use the generator model to simulate the transmission of a first transmitted signal sample through the communication channel and obtain a first generated signal output by the generator model. This first generated signal can simulate the signal obtained after the first transmitted signal sample has been transmitted through the communication channel.

[0051] Step 206: Based on the first transmitted signal sample, the first generated signal, and the real signal corresponding to the first transmitted signal sample, train the discriminator model in the generative adversarial network model to obtain the trained discriminator model.

[0052] The generative adversarial network model includes a discriminator model, which can be used to classify the source of the input signal to determine whether the input signal belongs to the generated signal output by the generator model or to the real signal.

[0053] For example, the server can determine the real signal corresponding to the first transmitted signal sample, and train a discriminator model in the generative adversarial network model based on the first transmitted signal sample, the first generated signal, and the real signal corresponding to the first transmitted signal sample, to obtain the trained discriminator model. In some embodiments, the server can construct signal pairs based on the first transmitted signal sample, the first generated signal, and the real signal corresponding to the first transmitted signal sample, respectively, to train the discriminator model in the generative adversarial network model using the constructed signal pairs. For example, the server can construct two signal pairs: (first transmitted signal sample, first generated signal) and (first transmitted signal sample, real signal corresponding to the first transmitted signal sample), and train the discriminator model based on these two signal pairs to obtain the trained discriminator model.

[0054] Step 208: Using the generator model, simulate the transmission of the second transmitted signal sample from the transmitted signal sample through the communication channel to obtain the second generated signal.

[0055] The second generated signal is the signal obtained by the generator model simulating the transmission of a second transmitted signal sample through a communication channel. For example, the server can simulate the transmission of a second transmitted signal sample through a communication channel using the generator model, and obtain the second generated signal output by the generator model. The second generated signal can simulate the signal obtained after the second transmitted signal sample is transmitted through the communication channel. In some embodiments, the server can input the second transmitted signal sample into the generator model and obtain the second generated signal output by the generator model. The second generated signal can be used to simulate the received signal obtained after the second transmitted signal sample is transmitted through the communication channel.

[0056] Step 210: Using the trained discriminator model, classify the signal source of the second generated signal to obtain the sample classification result, and update the generator model based on the sample classification result to obtain the trained generator model. The trained generator model is used to simulate the communication channel.

[0057] The sample classification result is obtained by classifying the source of the second generated signal using the trained discriminator model. Optionally, the server can also classify the source of the second generated signal using the trained discriminator model. For example, the server can input the second generated signal into the trained discriminator model, which will then output the corresponding sample classification result. The sample classification result can be used to characterize whether the trained discriminator model determines whether the second generated signal belongs to the generated signal output by the generator model or to the real signal. The server can update the generator model using the sample classification result. For example, it can determine the generator loss based on the sample classification result and update the model parameters of the generator model based on the generator loss to obtain the trained generator model. The trained generator model can be used to simulate a communication channel.

[0058] In some embodiments, the server can update the generator model based on the sample classification results to obtain the updated generator model. When the training termination condition is met, such as when the number of training iterations reaches a threshold or the loss converges, the training can be terminated, and the server can obtain the trained generator model based on the updated generator model. If the training termination condition is not met, the server can perform the next training iteration based on the updated generator model until the training termination condition is met, thus obtaining the trained generator model.

[0059] In the aforementioned channel modeling method, a generator model within a generative adversarial network (GAN) simulates the transmission of a first transmitted signal sample through a communication channel, resulting in a first generated signal. Based on the first transmitted signal sample, the first generated signal, and the corresponding real signal, a discriminator model within the GAN is trained, yielding a trained discriminator model. The generator model then simulates the transmission of a second transmitted signal sample through the communication channel, resulting in a second generated signal. The trained discriminator model classifies the signal source of the second generated signal, obtaining a sample classification result. Based on this result, the generator model is updated, resulting in a trained generator model used to simulate the communication channel. The generator model and discriminator model are trained adversarially, enabling the generator model to learn the deterministic characteristics and random perturbations of the communication channel, thus accurately simulating the actual communication channel.

[0060] In one exemplary embodiment, such as Figure 3 As shown, the discriminator model training process, namely, training the discriminator model in the generative adversarial network model based on the first transmitted signal sample, the first generated signal, and the real signal corresponding to the first transmitted signal sample, to obtain the trained discriminator model, includes steps 302 to 306. Wherein:

[0061] Step 302: Input the first transmitted signal sample and the first generated signal into the discriminator model in the generative adversarial network model to classify the signal source and obtain the first classification result.

[0062] The first classification result is the classification result obtained by the discriminator model for classifying the signal source of the first generated signal. It is used to characterize whether the discriminator model determines whether the first generated signal belongs to the generated signal output by the generator model or to the real signal.

[0063] For example, the server can input a first transmitted signal sample and a first generated signal into a discriminator model within a generative adversarial network (GAN) model. The discriminator model then classifies the source of the first generated signal based on the first transmitted signal sample and the first generated signal, thereby determining whether the first generated signal belongs to the generated signal output by the generator model or to a real signal, and outputting a first classification result. In some embodiments, the discriminator model may include sequentially connected layers such as an input layer, at least one convolutional layer, and a discriminator layer, allowing for signal source classification through these layers.

[0064] Step 304: Input the first transmitted signal sample and the real signal corresponding to the first transmitted signal sample into the discriminator model to classify the signal source and obtain the second classification result.

[0065] The second classification result is the classification result obtained by the discriminator model in classifying the signal source of the real signal corresponding to the first transmitted signal sample. It is used to characterize whether the discriminator model determines whether the real signal corresponding to the first transmitted signal sample belongs to the generated signal output by the generator model or belongs to the real signal.

[0066] For example, the server can input the first transmitted signal sample and the real signal corresponding to the first transmitted signal sample into the discriminator model in the generative adversarial network model. The discriminator model can then classify the source of the real signal corresponding to the first transmitted signal sample based on the first transmitted signal sample and the real signal corresponding to the first transmitted signal sample. Thus, the discriminator model can determine whether the real signal corresponding to the first transmitted signal sample belongs to the generated signal output by the generator model or belongs to the real signal, and output a second classification result.

[0067] Step 306: Based on the first classification result and the second classification result, update the discriminator model to obtain the trained discriminator model.

[0068] Optionally, the server can update the discriminator model based on the first and second classification results to obtain the trained discriminator model. For example, the server can determine the discriminator loss based on the first and second classification results, and update the model parameters of the discriminator model using the discriminator loss to obtain the trained discriminator model.

[0069] In this embodiment, the server can use the discriminator model to classify the signal source of the real signal corresponding to the first generated signal and the first transmitted signal sample, and update the discriminator model based on the classification results, thereby ensuring that the discriminator model can accurately identify the real signal and the generated signal output by the generator model.

[0070] In an exemplary embodiment, the trained discriminator model is used to classify the signal source of the second generated signal to obtain a sample classification result, and the generator model is updated based on the sample classification result to obtain a trained generator model. This includes: inputting the second generated signal and the second transmitted signal sample into the trained discriminator model to classify the signal source to obtain a sample classification result; determining the generator loss based on the sample classification result; and updating the generator model based on the generator loss to obtain a trained generator model.

[0071] The sample classification result is obtained by the trained discriminator model in classifying the source of the second generated signal. It is used to characterize whether the trained discriminator model determines that the second generated signal belongs to the generated signal output by the generator model or to the real signal. The generator loss is used to update the model parameters of the generator model, and the generator loss can be determined based on the sample classification result output by the trained discriminator model.

[0072] Optionally, the server can input the second generated signal and the second transmitted signal samples into the trained discriminator model, so that the trained discriminator can classify the signal source of the second generated signal and output the corresponding sample classification result. The server can determine the generator loss of the generator model based on the sample classification result, and update the generator model based on the generator loss to obtain the trained generator model.

[0073] In this embodiment, the server classifies the source of the second generated signal using the trained discriminator, determines the generator loss based on the obtained sample classification results, and updates the generator model according to the generator loss. The trained discriminator can be used to guide the training of the generator model, ensuring the training effect of the generator model, thereby enabling the generator model to accurately simulate the actual communication channel.

[0074] In an exemplary embodiment, the first classification result includes a first classification distribution, the second classification result includes a second classification distribution, and the sample classification result includes a third classification distribution; based on the first and second classification results, the discriminator model is updated to obtain the trained discriminator model, including: determining the discriminator loss based on the distance between the first and second classification distributions and the gradient penalty term; updating the discriminator model according to the discriminator loss to obtain the trained discriminator model; determining the generator loss based on the sample classification results includes: determining the generator loss based on the third classification distribution.

[0075] The classification distribution is the distribution obtained by the discriminator model in classifying the source of the input signal. This distribution characterizes the discriminator model's classification of the input signal, such as determining whether the input signal is a generated signal output by the generator model or a real signal. The distance between the first and second classification distributions characterizes the difference between them, and may include the Wasserstein distance. The gradient penalty term adjusts the gradient of the discriminator loss to ensure that the discriminator loss satisfies 1-Lipschitz continuity, meaning the L2 norm of the discriminator loss gradient does not exceed 1, thus ensuring the effectiveness of the discriminator loss.

[0076] Optionally, the classification result may specifically include a classification distribution, such as a first classification result including a first classification distribution, a second classification result including a second classification distribution, and a sample classification result including a third classification distribution. For training the discriminator model, the server can determine the distance between the first and second classification distributions, such as the Wasserstein distance between them. The server can construct the discriminator loss based on the distance between the first and second classification distributions and the gradient penalty term. Minimizing the discriminator loss is equivalent to maximizing the difference between the first and second classification distributions, enabling the discriminator to accurately classify the signal source for both real and generated signals. The server can update the model parameters of the discriminator model based on the discriminator loss to obtain the trained discriminator model. For training the generator model, the server can determine the generator loss based on the third classification distribution. In some embodiments, the goal of training the generator model is to minimize the Wasserstein distance between the real and generated distributions, so that the classification distribution corresponding to the generated signal output by the generator model approximates the classification distribution corresponding to the real signal.

[0077] In this embodiment, the classification result may include a classification distribution. The server can determine the discriminator loss based on the distance between the classification distributions and the gradient penalty term, and determine the generator loss based on the classification distribution. This can ensure the reliability of the discriminator loss and the generator loss, avoid mode collapse, effectively stabilize the model training process, and improve the model training effect.

[0078] In an exemplary embodiment, the transmitted signal sample includes a first transmitted signal sample and a second transmitted signal sample. A first generated signal is obtained by simulating the transmission of the first transmitted signal sample through a communication channel using a generator model in a generative adversarial network (GAN) model. This includes: inputting the first transmitted signal sample and a first random noise signal into the generator model in the GAN model to simulate transmission through the communication channel, thereby obtaining the first generated signal output by the generator model. A second generated signal is obtained by simulating the transmission of the second transmitted signal sample through a communication channel using the generator model. This includes: inputting the second transmitted signal sample and the second random noise signal into the generator model to simulate transmission through the communication channel, thereby obtaining the second generated signal output by the generator model.

[0079] The transmitted signal samples may include a first transmitted signal sample and a second transmitted signal sample, which can be different transmitted signals. Random noise signal refers to interference signals whose amplitude, phase, and other parameters change irregularly over time. The first random noise signal and the second random noise signal can be different random noise signals. In some embodiments, the generator model may include sequentially connected layers such as an input layer, at least one convolutional layer, and an output layer to simulate a communication channel.

[0080] Optionally, for the first transmitted signal sample, the server can input the first transmitted signal sample and the first random noise signal into the generator model in the generative adversarial network model to simulate the transmission of the first transmitted signal sample through a communication channel and obtain a first generated signal output by the generator model. The first generated signal is used to characterize the received signal obtained after the first transmitted signal sample is transmitted through the communication channel, so that the generated signal output by the generator model can simulate the received signal obtained after the transmitted signal is transmitted through the communication channel. For the second transmitted signal sample, the server can input the second transmitted signal sample and the second random noise signal into the generator model to simulate the transmission of the second transmitted signal sample through the communication channel and obtain a second generated signal output by the generator model.

[0081] In this embodiment, the server can simulate transmission for the first transmitted signal sample and the second transmitted signal sample using a generator model, so that the generator model can learn the deterministic characteristics and random disturbances of the communication channel, thereby enabling it to accurately simulate the actual communication channel.

[0082] In an exemplary embodiment, updating the generator model based on the sample classification results to obtain the trained generator model includes: updating the generator model based on the sample classification results to obtain the updated generator model; if the training termination condition is not met, returning to the step of obtaining the transmitted signal sample and the real signal corresponding to the transmitted signal sample, so as to perform the next training on the updated generator model until the training termination condition is met to obtain the trained generator model.

[0083] Optionally, the server can update the generator model parameters based on the sample classification results to obtain an updated generator model, which can then be used as the generator model after this training. The server can determine whether the training termination conditions are met, such as whether the training iteration threshold has been reached or whether the loss has converged. If the training termination conditions are not met, the server can return to the step of obtaining the transmitted signal sample and the corresponding real signal to perform the next training iteration on the updated generator model. For example, the server can return to the step of obtaining the transmitted signal sample and the corresponding real signal to obtain the next transmitted signal sample for the next training iteration on the updated generator model, until the training termination conditions are met, at which point the training can be terminated, thus obtaining the trained generator model. In some embodiments, if the training termination conditions are determined to be met, the server can terminate the training and obtain the trained generator model based on the updated generator model.

[0084] In this embodiment, after updating the generator model based on the sample classification results and obtaining the updated generator model, if the training termination condition is not met, the server can perform the next training on the updated generator model until the training termination condition is met and the trained generator model is obtained. The training termination condition can be used to control the training of the generator model, which can ensure the training effect of the generator model and enable the generator model to accurately simulate the actual communication channel.

[0085] In an exemplary embodiment, the channel modeling method further includes: acquiring a target transmission signal; inputting the target transmission signal into a trained generator model so that the trained generator model can simulate the transmission of the target transmission signal through the communication channel, and outputting a simulated received signal corresponding to the target transmission signal.

[0086] The target transmitted signal can be a transmitted signal that needs to be simulated for transmission. The simulated received signal can be obtained by the trained generator model after simulating the transmission of the target transmitted signal through the communication channel, so as to simulate the received signal obtained after the target transmitted signal is transmitted through the communication channel.

[0087] For example, the server can obtain the target transmission signal that needs to be simulated to be transmitted through the communication channel, and input the target transmission signal into the trained generator model so that the trained generator model can simulate the transmission of the target transmission signal through the communication channel and output the simulated received signal corresponding to the target transmission signal.

[0088] In this embodiment, the server can input the target transmitted signal into the trained generator model, and the trained generator model can output the simulated received signal corresponding to the target transmitted signal to simulate the received signal obtained after the target transmitted signal is transmitted through the communication channel. Thus, the actual communication channel can be accurately simulated through the generator model.

[0089] This application also provides an application scenario in which the above-described channel modeling method is applied. Specifically, the communication channel being modeled in this application scenario is an FSO channel, and the channel modeling method is applied in this scenario as follows:

[0090] Free-space optical communication (FSO), as a core technology for optical wireless transmission, has become an important solution for scenarios such as 5G backhaul and inter-satellite links due to its advantages of GHz-level bandwidth, Tbps-level transmission rates, deployment flexibility, and inherent resistance to electromagnetic interference. Currently, the global FSO market is expanding rapidly, and its technological maturity is accelerating from the laboratory to large-scale commercial application. However, despite its advantages in many aspects, FSO communication systems still face numerous challenges in practical applications. One of the most severe challenges currently facing FSO communication is that factors such as atmospheric turbulence in the transmission link can cause signal distortion and a decrease in transmission quality. Therefore, research on the transmission link is helpful in addressing the challenge of unstable reliability in FSO communication.

[0091] Joint Encoding-Modulation (JEM) is an innovative communication technology based on deep learning that optimizes communication systems by jointly encoding source, channel, and modulation. Compared to traditional separate architectures, JEM converts the source into discrete-time analog symbols, resulting in stronger stability and noise robustness, significantly improving system reliability and efficiency. Currently, semantic communication, joint source-channel coding and modulation, and discrete-time analog transmission demonstrate that this approach performs well in both wireless and optical communication, showing advantages in improving fiber optic communication performance and handling complex free-space optical communication scenarios (such as atmospheric turbulence).

[0092] By jointly training and optimizing the encoder-decoder network and FSO channel end-to-end, a high-performance JEM model can be obtained, improving the robustness of the FSO communication system. However, current integrated communication solutions are mainly based on Additive White Gaussian Noise (AWGN) channel training, which is an ideal model that differs significantly from real-world FSO communication scenarios. Furthermore, traditional FSO channels based on statistical distribution models are developed for specific loss characteristics and often use simplistic assumptions, limiting their ability to characterize complex channel effects and resulting in insufficient accuracy in real-world channels. Therefore, designing a channel model for integrated FSO communication is beneficial for further improving the performance of FSO communication systems.

[0093] In data-driven deep learning channel modeling algorithms, the completeness of training data and the ability to represent channels directly affect model performance. This application constructs an integrated FSO communication system to ensure the acquisition of complete training data. The structure of the JEM-based FSO communication system is as follows: Figure 4 As shown, an IM (Integrated Modular) / DD (Distributed Development) architecture is adopted, including core components such as a JEM encoder, modulator, arbitrary waveform generator, FSO channel, and optical receiver. At the transmitting end, the signal source is first encoded and modulated using the JEM model. The resulting discrete analog symbols are converted into a specific channel waveform by the arbitrary waveform generator, and then loaded onto an optical carrier generated by a laser with a wavelength of 1550 nm by the modulator. Subsequently, an optical amplifier boosts the power of the modulated optical signal. During FSO channel transmission, the optical signal is affected by atmospheric attenuation, turbulence, and other factors. At the receiving end, the optical receiver converts the received optical signal into an electrical signal, a low-pass filter removes signal noise, and finally, a digital oscilloscope acquires, stores, and displays the processed signal, which is then output from the output module. For the integrated FSO communication system, the input signal to the modulator is not the binary symbol used in traditional communication systems, but rather a discrete analog symbol x generated by the JEM network. The electrical signal output by the receiver is defined as the target data y for supervised learning. The two form a complete end-to-end training sample pair (x, y). By collecting multidimensional datasets with different turbulence intensities, transmission distances, and transmission powers in the system, data-driven deep learning FSO channel modeling under different conditions is achieved.

[0094] Free-Signal (FSO) channels exhibit significant strong nonlinearity, non-Gaussian distribution, and time-varying dynamic characteristics. They are also affected by factors such as atmospheric turbulence, resulting in high costs for acquiring measured data and a scarcity of samples in extreme scenarios. These characteristics present core requirements for channel modeling techniques: "no pre-defined distribution, adaptability to dynamic changes, and efficient data utilization." Generative Adversarial Networks (GANs), with their adversarial learning mechanism without pre-defined distribution, excellent small-sample data augmentation capabilities, and scalable temporal and conditional generation architecture, perfectly align with the core requirements of FSO channel modeling. They can specifically address the pain points of traditional modeling methods, providing an efficient solution for FSO channel modeling.

[0095] Based on this, this application proposes a channel modeling method for integrated free-space optical communication. Based on a data-driven approach, it collects data from the transmitter and receiver of an integrated FSO communication system and uses a generative adversarial network (GAN) model to train the generator and discriminator models to achieve efficient simulation of FSO channel characteristics. This scheme has the advantages of closely resembling the real channel and being differentiable, and can be used for end-to-end optimization of integrated FSO communication systems. Figure 5 The diagram shows the structure of the GAN channel model proposed in this application, which consists of an adversarial training framework composed of a generator model (generator network) and a discriminator model (discriminator network). The generator model is responsible for learning the implicit representation of the channel transmission characteristics, while the discriminator model establishes the discrimination criteria for signal authenticity. The two models achieve parameter optimization through a mini-max game.

[0096] Specifically, the generator model adopts a deep one-dimensional convolutional architecture, which is mathematically expressed as follows:

[0097] (1)

[0098] in, This is for sending signals (which are the signal samples during training). The input is a uniformly distributed random noise vector (i.e., a random noise signal). The network input is concatenated along its feature dimensions and transformed using tensors before entering the cascaded convolutional module. Specifically, the input layer will... and The input tensor is concatenated along the feature dimensions and transposed to obtain an input tensor of dimensions (2, 128). The number of channels in the convolutional layers are 512, 256, 128, and 128, respectively, and the activation function is LReLU (Leaky Rectified Linear Unit). (The hyperparameter, used to control the slope when the input is negative) is 0.2, and the convolutional layers use a residual connection design; the output layer transposes the signal and outputs a 128-dimensional analog received signal. This structure achieves progressive mapping of the feature space through multi-level dilated convolutions, and the final layer uses deconvolution to compensate for the resolution loss caused by downsampling, ensuring the output signal... With the actual received signal Maintain dimensional consistency.

[0099] The discriminator model is constructed as a binary classification convolutional neural network. The identification criteria are as follows:

[0100] (2)

[0101] in, This indicates concatenation along the channel dimension. It is the Sigmoid activation function. A one-dimensional convolutional layer is repeatedly stacked N times. The input layer will emit signals. With the signal to be identified The input is concatenated into a dual-channel input; the feature extraction layer is a 5-level convolutional module with channels of (128, 256, 512, 128, 2), all using... The activation function is LReLU with a value of 0.2; the discrimination layer outputs a scalar discrimination probability after global average pooling.

[0102] The model training employs an alternating optimization strategy, with the loss function defined as follows:

[0103] (3)

[0104] (4)

[0105] in, From real datasets, Input to the generator model, For expectation calculation; For discriminator loss; This is the generator loss. For the discriminator model to be designed for real signals The discrimination probability (i.e., the classification result); The generator model is based on the input signal and random noise The output generated signal, This represents the discrimination probability of the discriminator model for the generated signal.

[0106] Discriminator model updates need to be fixed (Generator model) parameters, and minimize the binary cross-entropy loss:

[0107] (5)

[0108] Among them, real samples Generate samples .

[0109] The generator model update needs to be fixed. (Discriminator model) parameters, and maximize the discriminator's false positive probability:

[0110] (6)

[0111] The Adam (Adaptive Moment Estimation) optimizer was used when training the model. The initial learning rate is set to The batch size is 64. To avoid mode collapse, a Wasserstein distance constraint is introduced into the generator loss, and a gradient penalty strategy is used to stabilize the alternating training process. The trained generator model can be used as an FSO channel model to simulate a real FSO channel environment.

[0112] In some embodiments, the performance of FSO channel modeling is first verified under different turbulence intensities, a transmission distance of 1 km, and a transmit power of 20 dBm. Figures 6(a) and 6(b) show a comparison of the modeling results data distribution. Figure 6(a) shows the distribution of GAN output data under weak turbulence, and Figure 6(b) shows the distribution of GAN output data under moderate to strong turbulence. It can be observed that the distribution of the generated output of the modeling scheme proposed in this application is highly similar to that of the actual channel output, indicating that the model's learning of data features is relatively accurate. Further, Figures 7(a)-7(d) show schematic diagrams of the GAN model fitting effect under different transmission distances. The black line signal represents generated data, and the green line signal represents real data. Figure 7(a) shows the fitting result under weak turbulence at 2 km; Figure 7(b) shows the fitting result under moderate to strong turbulence at 2 km; Figure 7(c) shows the fitting result under weak turbulence at 5 km; and Figure 7(d) shows the fitting result under moderate to strong turbulence at 5 km. As can be seen, when the transmission distance is 2km, the waveforms of the generated data and the real data show a certain similarity under both weak turbulence (Fig. 7(a)) and moderate turbulence (Fig. 7(b)) conditions. In the weak turbulence scenario, the model can roughly outline the signal change trend; under moderate turbulence, although there are some differences in fluctuation details, the overall waveform profile is consistent. This fully demonstrates that the GAN model has a certain ability to capture the characteristics of short-distance channels. When the transmission distance increases to 5km, the waveforms of the generated data and the real data also have a high degree of fit under both weak turbulence (Fig. 7(c)) and moderate turbulence (Fig. 7(d)) ​​conditions. Even under moderate turbulence, the generated data can reproduce the fluctuation characteristics of the real data well. This reflects that the model can also accurately simulate the signal changes under long-distance weak turbulence channels.

[0113] This application provides a channel modeling method for integrated free-space optical communication. Based on a data-driven strategy, it collects channel transmission data from an FSO communication system and designs and trains a Generative Adversarial Network (GAN) to model the channel. The GAN consists of a generator model and a discriminator model forming an adversarial training framework. Parameters are optimized through a mini-max game, efficiently simulating real-world channel environments. This method closely approximates real-world channels. Specifically, relying on the GAN adversarial framework, it simultaneously learns the deterministic characteristics and random perturbations of the FSO channel without requiring additional modules. The modeling results are closer to real-world channels than traditional single-feature methods. This method eliminates reliance on empirical formulas. Through pure data-driven design, it avoids the manual empirical formulas of traditional physical modeling, directly learning feature maps autonomously from measured data. This adapts to multi-factor coupling and extreme scenarios, reducing modeling errors. Finally, this method supports end-to-end optimization. Specifically, the GAN, built on a differentiable deep learning network, serves as an intermediate module for end-to-end optimization of the FSO system. Backpropagation enables coordinated adjustment of model and system parameters, improving optimization efficiency. Furthermore, the channel modeling method provided in this application has low complexity. By simplifying the number of GAN network layers, it reduces complexity while ensuring modeling accuracy, reduces training parameters and hardware dependence, and shortens the modeling cycle. It can also perform rapid inference, adapt to the real-time simulation requirements of embedded devices, and facilitates engineering implementation.

[0114] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0115] Based on the same inventive concept, this application also provides a channel modeling apparatus for implementing the channel modeling method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more channel modeling apparatus embodiments provided below can be found in the limitations of the channel modeling method described above, and will not be repeated here.

[0116] In one exemplary embodiment, such as Figure 8 As shown, a channel modeling device 800 is provided, including: a signal sample acquisition module 802, a first signal generation module 804, a discriminator training module 806, a second signal generation module 808, and a generator training module 810, wherein:

[0117] The signal sample acquisition module 802 is used to acquire the transmitted signal sample and the real signal corresponding to the transmitted signal sample. The real signal is the signal obtained at the receiving end after the transmitted signal sample is sent from the transmitting end and transmitted through the communication channel.

[0118] The first signal generation module 804 is used to simulate the first transmitted signal sample in the transmitted signal sample through the communication channel by using the generator model in the generative adversarial network model to obtain the first generated signal.

[0119] The discriminator training module 806 is used to train the discriminator model in the generative adversarial network model based on the first transmitted signal sample, the first generated signal and the real signal corresponding to the first transmitted signal sample, and to obtain the trained discriminator model.

[0120] The second signal generation module 808 is used to simulate the second transmitted signal sample in the transmitted signal sample through the communication channel by using a generator model to obtain the second generated signal.

[0121] The generator training module 810 is used to classify the source of the second generated signal using the trained discriminator model, obtain the sample classification result, and update the generator model based on the sample classification result to obtain the trained generator model. The trained generator model is used to simulate the communication channel.

[0122] In some embodiments, the discriminator training module 806 is further configured to input the first transmitted signal sample and the first generated signal into the discriminator model in the generative adversarial network model to classify the signal source and obtain a first classification result; input the first transmitted signal sample and the real signal corresponding to the first transmitted signal sample into the discriminator model to classify the signal source and obtain a second classification result; and update the discriminator model based on the first classification result and the second classification result to obtain the trained discriminator model.

[0123] In some embodiments, the generator training module 810 is further configured to input the second generated signal and the second transmitted signal samples into the trained discriminator model to classify the signal source and obtain the sample classification result; determine the generator loss based on the sample classification result, and update the generator model according to the generator loss to obtain the trained generator model.

[0124] In some embodiments, the first classification result includes a first classification distribution, the second classification result includes a second classification distribution, and the sample classification result includes a third classification distribution; the discriminator training module 806 is further configured to determine the discriminator loss based on the distance between the first classification distribution and the second classification distribution and the gradient penalty term; update the discriminator model according to the discriminator loss to obtain the trained discriminator model; the generator training module 810 is further configured to determine the generator loss based on the third classification distribution.

[0125] In some embodiments, the transmitted signal sample includes a first transmitted signal sample and a second transmitted signal sample; the first signal generation module 804 is further configured to input the first transmitted signal sample and the first random noise signal into the generator model in the generative adversarial network model to simulate transmission through a communication channel, thereby obtaining a first generated signal output by the generator model; the second signal generation module 808 is further configured to input the second transmitted signal sample and the second random noise signal into the generator model to simulate transmission through a communication channel, thereby obtaining a second generated signal output by the generator model.

[0126] In some embodiments, the generator training module 810 is further configured to update the generator model based on the sample classification results to obtain the updated generator model; if the training termination condition is not met, return to the step of obtaining the transmitted signal sample and the real signal corresponding to the transmitted signal sample, so as to perform the next training on the updated generator model until the training termination condition is met and the trained generator model is obtained.

[0127] In some embodiments, a channel simulation module is further included, which is used to acquire the target transmitted signal; input the target transmitted signal into the trained generator model so that the trained generator model can simulate the transmission of the target transmitted signal through the communication channel and output the simulated received signal corresponding to the target transmitted signal.

[0128] Each module in the aforementioned channel modeling device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0129] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 9As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a channel modeling method.

[0130] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0131] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0132] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0133] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0134] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0135] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0136] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0137] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method of channel modeling, the method comprising: The method comprises: obtaining a sending signal sample and a real signal corresponding to the sending signal sample, wherein the real signal is a signal obtained at a receiving end after the sending signal sample is transmitted through a communication channel from a sending end; simulating transmission of a first sending signal sample in the sending signal sample through the communication channel by a generator model in a generative adversarial network model to obtain a first generated signal; training a discriminator model in the generative adversarial network model based on the first sending signal sample, the first generated signal and the real signal corresponding to the first sending signal sample to obtain a trained discriminator model; simulating transmission of a second sending signal sample in the sending signal sample through the communication channel by the generator model to obtain a second generated signal; performing signal source classification on the second generated signal by the trained discriminator model to obtain a sample classification result, and updating the generator model based on the sample classification result to obtain a trained generator model, wherein the trained generator model is used to simulate the communication channel.

2. The method of claim 1, wherein, The method comprises: inputting the first sending signal sample and the first generated signal into the discriminator model in the generative adversarial network model to perform signal source classification to obtain a first classification result; inputting the first sending signal sample and the real signal corresponding to the first sending signal sample into the discriminator model to perform signal source classification to obtain a second classification result; updating the discriminator model based on the first classification result and the second classification result to obtain a trained discriminator model.

3. The method of claim 2, wherein, The method comprises: inputting the second generated signal and the second sending signal sample into the trained discriminator model to perform signal source classification to obtain a sample classification result; determining a generator loss based on the sample classification result, and updating the generator model according to the generator loss to obtain a trained generator model.

4. The method of claim 3, wherein, The first classification result comprises a first classification distribution, the second classification result comprises a second classification distribution, and the sample classification result comprises a third classification distribution; the method comprises: determining a discriminator loss based on a distance between the first classification distribution and the second classification distribution and a gradient penalty term; updating the discriminator model according to the discriminator loss to obtain a trained discriminator model; The method comprises: determining a generator loss based on the third classification distribution.

5. The method of claim 1, wherein, The sending signal samples include a first sending signal sample and a second sending signal sample; the first sending signal sample is simulated to be transmitted through the communication channel by a generator model in a generative adversarial network model to obtain a first generated signal, including: The first sending signal sample and a first random noise signal are input into the generator model in the generative adversarial network model to simulate transmission through the communication channel, and a first generated signal output by the generator model is obtained; The second sending signal sample is simulated to be transmitted through the communication channel by the generator model to obtain a second generated signal, including: The second sending signal sample and a second random noise signal are input into the generator model to simulate transmission through the communication channel, and a second generated signal output by the generator model is obtained.

6. The method of claim 1, wherein, The generator model is updated based on the sample classification result to obtain a trained generator model, including: The generator model is updated based on the sample classification result to obtain an updated generator model; If the training end condition is not met, the steps of obtaining the sending signal sample and the real signal corresponding to the sending signal sample are returned to perform next training on the updated generator model until the training end condition is met, and a trained generator model is obtained.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: Obtaining a target sending signal; The target sending signal is input into the trained generator model to simulate transmission of the target sending signal through the communication channel by the trained generator model, and a simulated received signal corresponding to the target sending signal is output.

8. A channel modeling apparatus, characterized by, The apparatus includes: A signal sample obtaining module is configured to obtain a sending signal sample and a real signal corresponding to the sending signal sample, the real signal being a signal obtained at a receiving end after the sending signal sample is transmitted through a communication channel from a sending end; A first signal generating module is configured to simulate transmission of a first sending signal sample in the sending signal sample through the communication channel by a generator model in a generative adversarial network model to obtain a first generated signal; A discriminator training module is configured to train a discriminator model in the generative adversarial network model based on the first sending signal sample, the first generated signal, and the real signal corresponding to the first sending signal sample to obtain a trained discriminator model; A second signal generating module is configured to simulate transmission of a second sending signal sample in the sending signal sample through the communication channel by the generator model to obtain a second generated signal; A generator training module is configured to perform signal source classification on the second generated signal by the trained discriminator model to obtain a sample classification result, and update the generator model based on the sample classification result to obtain a trained generator model, the trained generator model being used to simulate the communication channel. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method in any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, which is executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images