Wireless signal generation method based on generative adversarial network
Through the generative adversarial network decoupling of the distance characteristics and environmental characteristics of wireless signals, a more realistic wireless signal is generated, which solves the applicable scope and practicality of signal generation in the prior art, and realizes a closer relationship between signal and environmental characteristics. Reduces data acquisition costs.
Patent Information
- Application Number
- CN202411303015.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-18
- Publication Date
- 2025-01-07
AI Technical Summary
The existing wireless signal generation technology is based on simplified mathematical model, resulting in a large difference between the generated signal and the real signal, limited scope of application, and low practicality.
Using a generative adversarial network-based method, the VAE generation network, the reverse network and the discriminator interactive training is used to decouple the distance characteristics and environmental characteristics in the wireless observation signal, generate more realistic wireless signals, and use synthesis method and conversion method to generate targets. Signal.
It improves the close relationship between the signal and the spatial environment characteristics, and the generated signals are more practical, expands the scope of application of the method, and reduces the data acquisition cost for wireless system development and evaluation.
Smart Images
Figure CN119276383A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of signal generation, and in particular to a wireless signal generation method based on a generative adversarial network. Background Art
[0002] Existing wireless signal generation methods are divided into two categories: one is model-driven method and the other is data-driven method. The model-driven method is based on the laws of radio physical propagation, builds a spatial mathematical model of the radio propagation environment, simulates the generation process of radio signals in a given environment, and thus generates the required wireless signal data. This method has strong interpretability, can build explicit mathematical models for specific variables, can be flexibly expanded, and can build wireless transmission models for different environments according to needs.
[0003] The data-driven method is based on the wireless signal data collected in the real environment. It uses a neural network to learn the wireless propagation knowledge in the current environment, build a non-explicit model of the relevant variables in the current environment, and generate wireless signal data under the relevant variables. This method can capture and express the complex characteristics and nonlinear relationships in the current environment, thereby generating more realistic wireless signal data in the current environment. This method requires the collection of a certain amount of real data in advance and the annotation of relevant variables.
[0004] Existing wireless signal generation technologies mostly adopt model-driven technology routes, which simulate the generation process of wireless signals under given environmental factors by constructing a signal physical propagation model. However, this type of technology is based on the matching degree between the constructed model and the actual propagation model. As mentioned above, multiple factors are coupled with each other, the relationship is complex, and the mathematical model is relatively complex. Most of them approximate certain characteristics or properties of wireless signals by simplifying or assuming them. In essence, the high-dimensional information of wireless signals is projected into low-dimensional space for approximate reconstruction. Only some specific features or characteristics of the generated signal are similar to the real signal, and there are large differences between them and the actual real signal. The scope of application is limited and the practicality is low. Summary of the invention
[0005] In view of the above situation, the main purpose of the present invention is to propose a method and system for long-tail recommendation of users with different activity levels using NMF to solve the above technical problems.
[0006] The present invention provides a method for generating a wireless signal based on a generative adversarial network, the method comprising the following steps:
[0007] Step 1: Input the wireless observation signal into the VAE generation network, learn the network parameters and decouple the distance features and environmental features in the wireless observation signal, and use the encoder to generate distance feature coding and environmental feature coding;
[0008] By learning network parameters and decoupling distance feature encoding and environment feature encoding, the decoder is used to generate the reconstructed signal;
[0009] Decouple the environment feature encoding and distance feature encoding through the classifier to obtain the environment label and distance label respectively;
[0010] Update the parameters of the generative adversarial learning network model through the loss function of the VAE generated network;
[0011] Step 2: Based on the distance label and the environment label, the generated signal is obtained through the inverse signature generator and decoder in the inverse network; the parameters of the generative adversarial learning network model are updated through the loss function of the inverse network;
[0012] Step 3: Input the wireless observation signal, the reconstructed signal and the generated signal into the discriminator to generate a classification result, judge the integrity of the signal by the classification result, and use the loss function of the discriminator to update the parameters of the generative adversarial learning network model;
[0013] Repeating steps 1 to 3 in an iterative manner until stopping after a specified number of times, outputting updated parameters of the generative adversarial learning network model, and constructing a distribution through the updated parameters of the generative adversarial learning network model;
[0014] Step 4: Input the distance label and environment label into the signal generator, and generate the target signal through the synthesis method; input the distance label, environment label and target domain environment label into the signal generator, and generate the target signal through the conversion method.
[0015] The present invention also proposes a wireless signal generation system based on a generative adversarial network, the system comprising:
[0016] VAE generates a network module for:
[0017] The wireless observation signal is input into the VAE generation network, and the distance feature and environment feature coding are generated by using the encoder by learning the network parameters and decoupling the distance feature and environment feature in the wireless observation signal;
[0018] By learning network parameters and decoupling distance feature encoding and environment feature encoding, the decoder is used to generate the reconstructed signal;
[0019] Decouple the environment feature encoding and distance feature encoding through the classifier to obtain the environment label and distance label respectively;
[0020] Update the parameters of the generative adversarial learning network model through the loss function of the VAE generated network;
[0021] Reverse Network Module for:
[0022] Based on the distance label and the environment label, the generated signal is obtained through the inverse signature generator and decoder in the inverse network; the parameters of the generative adversarial learning network model are updated through the loss function of the inverse network;
[0023] Discriminator module, which is used to:
[0024] The wireless observation signal, the reconstructed signal and the generated signal are input into the discriminator to generate the classification result, the soundness of the signal is judged by the classification result, and the loss function of the discriminator is used to update the parameters of the generative adversarial learning network model;
[0025] Signal generation module for:
[0026] The distance label and the environment label are input into the signal generator, and the target signal is generated by the synthesis method; the distance label, the environment label and the target domain environment label are input into the signal generator, and the target signal is generated by the conversion method.
[0027] Compared with the prior art, the present invention has the following beneficial effects:
[0028] 1. The present invention constructs wireless signal generation as a hierarchical latent variable estimation problem, which can be applied to both the synthesis method and the conversion method, fully exploiting the coupling relationship between spatial environmental characteristics and signals. Compared with traditional data-driven methods, it has better flexibility and interpretability, and expands the scope of application of the method.
[0029] 2. The present invention fully explores the coupling relationship between signals and labels through interactive adversarial training between the generating network and the inverse network, and introduces a discriminator to distinguish and evaluate the authenticity of the observed signal and the reconstructed signal, thereby enhancing the closeness between the generated signal and the label features. Compared with the traditional learning framework, the relationship between the generated wireless signal and the real space environment features is closer, which is relatively more realistic and practical.
[0030] 3. The present invention is used to generate wireless signals tightly coupled with spatial environment characteristics, which can effectively support relevant directional data enhancement technology and applications, and can effectively reduce the data collection cost of relevant wireless system development and evaluation.
[0031] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description or learned through embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 A flowchart of a wireless signal generation method based on a generative adversarial network proposed by the present invention;
[0033] Figure 2 This is a system framework diagram of a wireless signal generation system based on a generative adversarial network proposed in the present invention. DETAILED DESCRIPTION
[0034] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.
[0035] These and other aspects of the embodiments of the present invention will be apparent with reference to the following description and accompanying drawings. In these descriptions and accompanying drawings, some specific implementations of the embodiments of the present invention are specifically disclosed to represent some ways of implementing the principles of the embodiments of the present invention, but it should be understood that the scope of the embodiments of the present invention is not limited thereto.
[0036] First, we define the distance latent variable and the environment latent variable. The posterior probability distribution can be expressed as p(z d , z e |x)、p(z d |d) and p(z e |k). However, the posterior probability distribution p(z d , z e |x)、p(z d |d) and p(z e |k) is difficult to estimate, so we construct an approximate posterior probability distribution to construct q(z d , z e |x;φ) corresponds to p(z d , z e |x), build Corresponding to p(z d |d), build Corresponding to p(z e |k), and is differentiable everywhere.
[0037] Assume the following conditions:
[0038] (1)q(z d , z e |x;φ)=q(z d |x;φ d )q(z e |x;φ e ), φ={φ d ,φ e};
[0039] (2) Distribution p(x|z) d , z e ;θ), as well as Both come from parameters θ and A family of parametric distributions, where
[0040] (3) Distribution parameters
[0041] Assuming our given dataset It contains N groups of samples, which can be divided into parameters θ, φ and Make an estimate:
[0042]
[0043] Among them, L ELBO,D Denotes the lower bound of evidence, D KL represents KL divergence;
[0044] For distribution parameters The estimation is done by optimizing the following problem:
[0045]
[0046] Among them, L KL，D Represents the KL divergence between the approximate distribution and the true distribution, and its definition is:
[0047]
[0048] Through the learning framework, training is completed and parameter estimation is obtained, which is then brought in to obtain approximate expressions of all distribution functions, thereby achieving signal generation.
[0049] Then, assuming:
[0050]
[0051] Among them, ε e I and ε d I is the covariance matrix, ε e With ε d All represent arbitrarily given minimum values.
[0052] Then, binary variables, i.e., classification results, are introduced to represent the corresponding hierarchical latent variables:
[0053]
[0054] Distance and environment are high-level latent variables, distance features and environment features are low-level latent variables, and the classification result is the introduced discriminant latent variable;
[0055] Defining the parameter positions corresponding to the classification results, the empirical objective function of the discriminator can be calculated as follows:
[0056]
[0057] Among them, L ML，D represents the empirical objective function of the discriminator;
[0058] The empirical objective function of the evidence lower bound of the discriminator is expressed as:
[0059]
[0060] Among them, L ELBO，D An empirical objective function representing the lower bound of the evidence;
[0061] Therefore, the latent variable short leg problem can be solved by the following optimization problem:
[0062] First pass Calculate the updated parameter ω * ;
[0063] Then, the updated parameter ω * Bring in L ELBO，D and L KL，D Solve to get the updated parameter φ * ,θ * , The calculation formula is as follows:
[0064]
[0065] The above optimization problem can be solved by gradient descent.
[0066] See also Figure 1 The embodiment of the present invention provides a method for generating a wireless signal based on a generative adversarial network, the method comprising the following steps:
[0067] Step 1: Input the wireless observation signal into the VAE generation network, learn the network parameters and decouple the distance features and environmental features in the wireless observation signal, and use the encoder to generate distance feature coding and environmental feature coding;
[0068] By learning network parameters and decoupling distance feature encoding and environment feature encoding, the decoder is used to generate the reconstructed signal;
[0069] Decouple the environment feature encoding and distance feature encoding through the classifier to obtain the environment label and distance label respectively;
[0070] Update the parameters of the generative adversarial learning network model through the loss function of the VAE generated network;
[0071] In step 1, the wireless observation signal is input into the VAE generation network. By learning the network parameters and decoupling the distance features and environmental features in the wireless observation signal, the encoder is used to generate the distance feature encoding and the environmental feature encoding. The corresponding process has the following relationship:
[0072] g(·;φ):x→z d , z e ;
[0073] Among them, x represents the wireless observation signal, φ represents the network parameter, z d Represents the distance feature encoding, z e represents the environmental feature encoding, and g represents the decoder;
[0074] By learning the network parameters and decoupling the distance feature encoding and the environment feature encoding, the decoder is used to generate the reconstructed signal. The corresponding process has the following relationship:
[0075]
[0076] in, represents the reconstructed signal, θ represents the network parameters;
[0077] By decoupling the environment feature coding and distance feature coding through the classifier, the environment label and distance label are obtained respectively. The relationship between the corresponding processes is:
[0078]
[0079] Among them, f represents the classifier, φ e With φ d All represent network parameters, k represents environment label, and d represents distance label;
[0080] In step 1, the parameters of the generative adversarial learning network model are updated through the loss function of the VAE generation network. The loss function of the VAE generation network is constructed by the lower bound of empirical evidence, and the relationship is:
[0081]
[0082] Where D represents the data set, ω and Both represent network parameters, E pD(x) represents the expectation of the probability distribution of the wireless observation signal, E q(z|x；φ) represents the expectation of the approximate posterior probability distribution of the latent variable given the network parameters and the wireless observation signal, E pD(x，d) represents the expectation of the joint probability distribution of the wireless observation signal and the distance tag, E pD(x，k) represents the expectation of the joint probability distribution of the wireless observation signal and the environment label, represents the approximate posterior probability distribution of the environment label given the network parameters about the environment, the wireless observation signal and the environment network parameters, represents the network parameters about the environment, represents the network parameters regarding distance, represents the estimated distance of the estimator, h represents the discriminator network learning function, L VAE Represents the loss function of the VAE generator network.
[0083] Step 2: Based on the distance label and the environment label, the generated signal is obtained through the inverse signature generator and decoder in the inverse network; the parameters of the generative adversarial learning network model are updated through the loss function of the inverse network;
[0084] In step 2, based on the distance label and the environment label, the generated signal is obtained through the inverse signature generator and decoder in the inverse network. The relationship between the corresponding process is:
[0085]
[0086] Among them, z′ d represents the distance feature code generated by the inverse feature code generator, z′ e represents the distance feature code generated by the inverse feature code generator, Indicates the generation of a signal;
[0087] The parameters of the generative adversarial learning network model are updated through the loss function of the inverse network. The loss function expression of the inverse network is:
[0088]
[0089] in, represents the network parameters, represents the distance latent variable estimation, L inv Represents the loss function of the inverse network.
[0090] Step 3: Input the wireless observation signal, the reconstructed signal and the generated signal into the discriminator to generate a classification result, judge the integrity of the signal by the classification result, and use the loss function of the discriminator to update the parameters of the generative adversarial learning network model;
[0091] Repeating steps 1 to 3 in an iterative manner until stopping after a specified number of times, outputting updated parameters of the generative adversarial learning network model, and constructing a distribution through the updated parameters of the generative adversarial learning network model;
[0092] In step 3, the loss function of the discriminator is used to update the parameters of the generative adversarial learning network model. The loss function expression of the discriminator is:
[0093]
[0094] Among them, L disrepresents the loss function of the discriminator, ξ represents the classification result, q(ξ|x, z; ω) represents the approximate posterior probability distribution of the classification result given the wireless observation signal, hidden variables and network parameters, E p(z) represents the expectation of the probability distribution of the latent variable, E p(x|z；θ) represents the expectation of the probability distribution of the wireless observation model given the dependent variables and network parameters;
[0095] Output updated parameters through the updated generative adversarial learning network, and use the updated parameters to construct the distribution p(x|z d , z e θ * ), q(z d , z e |x;φ * )and where φ * ,θ * , and ω * These are updated parameters.
[0096] Step 4: Input the distance label and the environment label into the signal generator, and generate the target signal through the synthesis method; input the distance label, the environment label and the target domain environment label into the signal generator, and generate the target signal through the conversion method;
[0097] In step 4, the distance label and the environment label are input into the signal generator, and the target signal is generated by synthesis. The specific steps are as follows:
[0098] Using distance labels and environment labels, through distribution and Estimate the distance feature coding and the environment feature coding respectively;
[0099] Using distance feature coding and environmental feature coding, based on the distribution p(x|z d , z e θ * ) generating a target signal;
[0100] The distance label, environment label and target domain environment label are input into the signal generator, and the target signal is generated by the conversion method. The specific steps are as follows:
[0101] Generate acquisition signals using target domain environment labels and distance labels;
[0102] Based on the collected signal, by distributing q(z d , z e |x;φ * ) Generate distance feature encoding;
[0103] Using environment tags, distributed Generate environmental feature codes;
[0104] Based on distance feature coding and environmental feature coding, Generate the target signal.
[0105] See also Figure 2 The embodiment of the present invention further provides a wireless signal generation system based on a generative adversarial network, the system comprising:
[0106] VAE generates a network module for:
[0107] The wireless observation signal is input into the VAE generation network, and the distance feature and environment feature coding are generated by using the encoder by learning the network parameters and decoupling the distance feature and environment feature in the wireless observation signal;
[0108] By learning network parameters and decoupling distance feature encoding and environment feature encoding, the decoder is used to generate the reconstructed signal;
[0109] Decouple the environment feature encoding and distance feature encoding through the classifier to obtain the environment label and distance label respectively;
[0110] Update the parameters of the generative adversarial learning network model through the loss function of the VAE generated network;
[0111] Reverse Network Module for:
[0112] Based on the distance label and the environment label, the generated signal is obtained through the inverse signature generator and decoder in the inverse network; the parameters of the generative adversarial learning network model are updated through the loss function of the inverse network;
[0113] Discriminator module, which is used to:
[0114] The wireless observation signal, the reconstructed signal and the generated signal are input into the discriminator to generate the classification result, the soundness of the signal is judged by the classification result, and the loss function of the discriminator is used to update the parameters of the generative adversarial learning network model;
[0115] Signal generation module for:
[0116] The distance label and the environment label are input into the signal generator, and the target signal is generated by the synthesis method; the distance label, the environment label and the target domain environment label are input into the signal generator, and the target signal is generated by the conversion method.
[0117] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0118] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0119] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.
Claims
1. A wireless signal generation method based on a generative adversarial network, characterized in that: The method is implemented based on a generative model, which is composed of a VAE generative network, a reverse network, a discriminator, and a signal generator. The VAE generative network and the reverse network share a decoder. The method includes the following steps: Step 1: Input the wireless observation signal into the VAE generation network, learn the network parameters and decouple the distance features and environmental features in the wireless observation signal, and use the encoder to generate distance feature coding and environmental feature coding; By learning network parameters and decoupling distance feature encoding and environment feature encoding, the decoder is used to generate the reconstructed signal; Decouple the environment feature encoding and distance feature encoding through the classifier to obtain the environment label and distance label respectively; Update the parameters of the generative adversarial learning network model through the loss function of the VAE generated network; Step 2: Based on the distance label and the environment label, the generated signal is obtained through the inverse signature generator and decoder in the inverse network; the parameters of the generative adversarial learning network model are updated through the loss function of the inverse network; Step 3: Input the wireless observation signal, the reconstructed signal and the generated signal into the discriminator to generate a classification result, judge the integrity of the signal by the classification result, and use the loss function of the discriminator to update the parameters of the generative adversarial learning network model; Repeating steps 1 to 3 in an iterative manner until stopping after a specified number of times, outputting updated parameters of the generative adversarial learning network model, and constructing a distribution through the updated parameters of the generative adversarial learning network model; Step 4: Input the distance label and environment label into the signal generator, and generate the target signal through the synthesis method; input the distance label, environment label and target domain environment label into the signal generator, and generate the target signal through the conversion method.
2. The method for generating wireless signals based on a generative adversarial network according to claim 1, characterized in that: In step 1, the wireless observation signal is input into the VAE generation network, and the distance feature and the environment feature in the wireless observation signal are decoupled by learning the network parameters, and the encoder is used to generate the distance feature encoding and the environment feature encoding. The relationship between the corresponding process is: g(·;φ):x→z d ,z e ; Among them, x represents the wireless observation signal, φ represents the network parameter, z d Represents the distance feature encoding, z e represents the environmental feature encoding, and g(·) represents the decoder.
3. The method for generating wireless signals based on a generative adversarial network according to claim 2, characterized in that: In step 1, by learning the network parameters and decoupling the distance feature coding and the environment feature coding, the decoder is used to generate the reconstructed signal. The relationship between the corresponding process is: in, represents the reconstructed signal, and θ represents the network parameters.
4. The method for generating wireless signals based on a generative adversarial network according to claim 3, characterized in that: In step 1, the environment feature coding and the distance feature coding are decoupled by the classifier to obtain the environment label and the distance label respectively. The relationship between the corresponding processes is: f(·;φ e ):z e →k f(·;φ d ):z d →d; Among them, f represents the classifier, φ e With φ d They all represent network parameters, k represents the environment label, and d represents the distance label.
5. The method for generating wireless signals based on a generative adversarial network according to claim 4, characterized in that: In step 1, the parameters of the generative adversarial learning network model are updated by the loss function of the VAE generated network. The loss function of the VAE generated network is constructed by the lower bound of empirical evidence, and the relationship is: Where D represents the data set, ω and Both represent network parameters, E pD(x) represents the expectation of the probability distribution of the wireless observation signal, E q(z|x；φ) represents the expectation of the approximate posterior probability distribution of the latent variable given the network parameters and the wireless observation signal, E pD(x，d) represents the expectation of the joint probability distribution of the wireless observation signal and the distance tag, E pD(x，k) represents the expectation of the joint probability distribution of the wireless observation signal and the environment label, represents the approximate posterior probability distribution of the environment label given the network parameters about the environment, the wireless observation signal and the environment network parameters, represents the network parameters about the environment, represents the network parameters regarding distance, represents the estimated distance of the estimator, h represents the discriminator network learning function, and L represents the loss function of the VAE generation network.
6. The method for generating wireless signals based on a generative adversarial network according to claim 5, characterized in that: In step 2, based on the distance label and the environment label, the generated signal is obtained through the inverse signature generator and decoder in the inverse network. The relationship between the corresponding process is: Among them, z′ d represents the distance feature code generated by the inverse feature code generator, z′ e represents the distance feature code generated by the inverse feature code generator, Indicates that a signal is generated.
7. The method for generating wireless signals based on a generative adversarial network according to claim 6, characterized in that: In step 2, the parameters of the generative adversarial learning network model are updated by the loss function of the reverse network, and the loss function expression of the reverse network is: in, represents the network parameters, represents the distance latent variable estimation, L inv Represents the loss function of the inverse network.
8. The method for generating wireless signals based on a generative adversarial network according to claim 7, characterized in that: In step 3, the loss function of the discriminator is used to update the parameters of the generative adversarial learning network model. The loss function expression of the discriminator is: Among them, L dis represents the loss function of the discriminator, ξ represents the classification result, q(ζ|x, z; ω) represents the approximate posterior probability distribution of the classification result given the wireless observation signal, hidden variables and network parameters, E p(z) represents the expectation of the probability distribution of the latent variable, E p(x|z；θ) represents the expectation of the probability distribution of the wireless observation model given the dependent variables and network parameters; Output updated parameters through the updated generative adversarial learning network, and use the updated parameters to construct the distribution p(x|z d , z e θ * ), q(z d , z e |x;φ * )and Among them, φ*, θ*, and ω* are updated parameters.
9. The method for generating wireless signals based on a generative adversarial network according to claim 8, characterized in that: In step 4, the distance tag and the environment tag are input into the signal generator, and the target signal is generated by the synthesis method. The specific steps are as follows: Using distance labels and environment labels, through distribution and Estimate the distance feature coding and the environment feature coding respectively; Using distance feature coding and environmental feature coding, based on the distribution p(x|z d , z e θ * ) generating a target signal; The distance label, environment label and target domain environment label are input into the signal generator, and the target signal is generated by the conversion method. The specific steps are as follows: Generate acquisition signals using target domain environment labels and distance labels; Based on the collected signal, by distributing q(z d , z e |x;φ * ) Generate distance feature encoding; Using environment tags, distributed Generate environmental feature codes; Based on distance feature coding and environmental feature coding, Generate the target signal.
10. A wireless signal generation system based on a generative adversarial network, characterized in that: The system applies a wireless signal generation method based on a generative adversarial network as described in any one of claims 1 to 9 above, and the system comprises: VAE generates a network module for: The wireless observation signal is input into the VAE generation network, and the distance feature and environment feature coding are generated by using the encoder by learning the network parameters and decoupling the distance feature and environment feature in the wireless observation signal; By learning network parameters and decoupling distance feature encoding and environment feature encoding, the decoder is used to generate the reconstructed signal; Decouple the environment feature encoding and distance feature encoding through the classifier to obtain the environment label and distance label respectively; Update the parameters of the generative adversarial learning network model through the loss function of the VAE generated network; Reverse Network Module for: Based on the distance label and the environment label, the generated signal is obtained through the inverse signature generator and decoder in the inverse network; the parameters of the generative adversarial learning network model are updated through the loss function of the inverse network; Discriminator module, which is used to: The wireless observation signal, the reconstructed signal and the generated signal are input into the discriminator to generate the classification result, the soundness of the signal is judged by the classification result, and the loss function of the discriminator is used to update the parameters of the generative adversarial learning network model; Signal generation module for: The distance label and the environment label are input into the signal generator, and the target signal is generated by the synthesis method; the distance label, the environment label and the target domain environment label are input into the signal generator, and the target signal is generated by the conversion method.