Signal Reconstruction Based On Generative Adversarial Networks

The signal reconstruction in a complex electromagnetic environment is an approximation to the electromagnetic environment.In addition to getting information about the electromagnetic environment,signal reconstruction is also widely used in the field of communication countermeasures.It is one of the key technologies for communication signal processing.With the rapid development of communication technology,traditional methods of signal reconstruction have shown signs of fatigue in increasingly complex electromagnetic environments,making it difficult to reconstruct signals accurately.This paper proposes a signal reconstruction method based on Generative,adversarial networks(GAN).For the scene of signal reconstruction,the existing Generative adversarial networks is improved and the quality of the generated signal is effectively improved,make the signal reconstruction based on Generative,adversarial networks has good robustness and generalization.The work of this article mainly includes the following points:First,the training of Generative adversarial networks includes multiple rounds of crosstraining.In each round,the optimization and update of the generator and discriminator neural network are calculated by the loss function,so the merits of the loss function and the quality of the generated data are closely related?In this paper,based on the application scenario of signal reconstruction,the cross-entropy-based loss function and the EM-based loss function are mainly studied for the effect of the Generative adversarial networks algorithms.Based on this,the penalty items with excessive signal fluctuations are added to improve effectively.The quality of the generated data.Second,in deep learning,the architecture of the neural network plays a crucial role in the effectiveness of the algorithm.Different application scenarios apply different neural network architectures.For the signal reconstruction problem to be dealt with in this paper,the design of the neural network framework also determines the quality of the generated data.The DCGAN-based neural network architecture is difficult to learn and simulate the sequence characteristics of communication signals effectively,resulting in poor data quality.To solve this problem,this paper cancels the application of the convolution layer in the network architecture design of the neural network,and adopts a full connection layer;In addition,the discriminator's pre-training and batch data diversity features in the generator are introduced,so that the Generative adversarial networks can generate signals more accurately and further improve the signal reconstruction performance.Thirdly,in the framework of communication signal reconstruction technology,it is difficult to measure the signal parameters and extract features in a complex electromagnetic environment for existing signal reconstruction mechanisms.This article combines the Conditions generative adversarial networks(CGAN),extracts the middle layer output of the discriminator neural network as conditional information,and completes automatic extraction of signal features through deep neural network learning;In the signal generator,the condition information is combined with Gaussian noise as input data,and the signal is automatically reconstruct through the mapping of the neural network.The signal reconstruction based on.