Generative Adversarial Network for Wireless Communication: Principle, Application, and Trends

Abstract

Generative adversarial network (GAN) has attracted wide attention because of its remarkable ability to learn highdimensional and complex data distributions based on game theory and machine learning. In this paper, we focus on the role that GANs play in wireless communications, which has been well researched and developed in recent years. With the gradual increase in communication rates and the complicacy of communication scenarios, some urgent issues have emerged, such as complex channel generation, high-dimensional channel estimation, and insufficient real-world signal acquisition. The powerful nonlinear fitting property of GAN can break through the bottleneck of traditional communication techniques, whose applications can be categorized into data generation, performance optimization, and data classification according to different purposes and network structures. For each category, a universal network structure summarizing the characteristics and targets is investigated, and the performance of GANs in communication is also analyzed with simulation examples. Moreover, the open research issues are discussed at the end to provide some directions for future study.