The deep integration of artificial intelligence (AI) and communication technology is the typical feature of the 6G network.On the one hand, AI injects new vitality into the development of the 6G network, which can effectively use the data generated by the historical operation of the network.It enables the network to be self-maintained and selfoptimized, and accelerates the process of network intelligence.On the other hand, the rich scenarios and IoT devices of the 6G network provide a large number of application fields and massive data for AI.These can enable the better deployment of AI, fully demonstrate the performance advantages of AI, and provide high-quality services for users.However, in practice, it is difficult to give full play to the performance advantages of AI due to the difficulty of sample collection, high cost of the collection, and lack of universality which caused by the complexity of the environment.Therefore, academia and industry introduce generative adversarial network (GAN) into the design of wireless networks.The powerful feature learning and feature expression ability of GAN can generate a large number of generated samples, which realizes the expansion of the wireless database.The introduction of GAN can effectively improve the generalization ability of AI models for wireless networks.Owing to the excellent performance of GAN, the generative model represented by GAN has attracted increased attention in the field of wireless networks, and rapidly became the new research hotspot of 6G networks.Firstly, the principle of GAN and its different versions of improved derived models were summarized.Then, the framework, advantages and disadvantages of each model were analyzed.Secondly, the research and application status of these models in wireless networks were reviewed.Finally, the research trends of GAN were proposed for the 6G network requirements, which provided some valuable exploration for future research.
