Deep Learning Techniques for Next Generation Wireless Networks

Abstract

Machine learning (ML) techniques have shown promising performance in solving different communication system issues. Recently, several deep learning-based end-to-end techniques have been implemented to optimize the transmitter, the channel, and the receiver blocks in one single process, thereby replacing the conventional communications system. In this thesis, we start exploring the research for the end-to-end wireless model where we used the autoencoder (AE) as a based communication system. We studied the performance of the AE with additive white Gaussian noise (AWGN) channel and compared it with the equivalent conventional communication system. Then for deep learning (DL), we introduce a DL-based detector, termed DL Index Modulation (DLIM), for IM-MIMO-OFDM using a deep neural network (DNN) in terms of error performance. Our initial results using the proposed DLIM can detect the transmitted symbols with performance comparable to near-optimal bit error rate (BER) with shorter runtime than the existing hand-crafted detectors. Next, for the Intelligent Reflecting Surfaces (IRS), we proposed an IRS-assisted end-to-end communication system that operates over AWGN channels, where the modulation and demodulation are performed by a DNN based on an AE architecture. Simulation results show the BER performance of our AE-based scheme achieves better performance gains than the existing classical IRS baseline and the AE hand-crafted baselines. Moreover, a new probabilistic model, based on the variational autoencoders (VAE), is proposed for short-packet wireless communication systems. Using this new approach, the information messages are represented by the so-called packet hot vectors (PHV), which are inferred by the VAE latent random variables (LRVs). Numerical results show that the proposed VAE, with a DL classifier, improved symbol error rate (SER) performance in comparison to the baseline schemes.