GAN network for cross-domain channel estimation in the 5G communication system

• Main Author: Quan, Yeming
• Other Authors: Teh Kah Chan
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Electrical and electronic engineering::Wireless communication systems
• Online Access: https://hdl.handle.net/10356/149449
• Institution: Nanyang Technological University

Advanced signal processing algorithms and sophisticated device generation processes in wireless communication systems enable block-based wireless communication systems with good communication performance. However, as the individual modules are designed for different purposes and situations, it is easy to achieve optimal performance on a particular module, but difficult to optimize the whole system. Recently, due to the successful applications of deep learning in other fields, researchers have considered its application to communication systems in order to improve systems performance. In this report, random functions and channel models are used to generate training and test data. Deep learning-based autoencoders and decoders are used as replacements for the signal processing modules in conventional wireless communication systems. The channel distribution is modeled using conditional generative adversarial networks. Generative adversarial networks consist of two parts, a generator, and a discriminator. It aims to generate images similar to the target images by random sampling from the latent space. Using the real channel output as training data, we can model the channel distribution and build an end-to-end communication system with a structure of autoencoder - conditional generative adversarial network - decoder, which is based on deep neural networks. Simulation results show that the deep learning-based structure can achieve global optimality through an end-to-end loss function. Comparison with the theoretical BER curve also shows this to be a promising communication architecture. Keywords: Deep learning, deep neural network, autoencoder, conditional generative adversarial network.