Wireless communication receiver design based on machine learning

Abstract

In digital wireless communications systems, equalizers are needed to reduce the effect of inter-symbol interference (ISI) due to multipath fading channel. Recent works have demonstrated that machine learning approaches are suitable to solve different tasks in mobile communication systems. In this final year project, we study how to apply machine learning to wireless communication receiver design, and the task we focus on is equalization. Two types of artificial neural network (ANN), i.e., Long Short-Term Memory (LSTM) and Gated Recurring Units (GRU), are considered in our study. Extensive analysis has been done to find the optimal structure of the two ANNs, as well as the optimal setting of the training parameters. The performance of the two ANNs has been tested under different scenarios, which includes different modulation types (4QAM, 16QAM and 64QAM), channel types (time-invariant and time-varying), and waveforms (single-carrier and multicarrier, i.e., OFDM). In addition, the performance of the ANNs are compared with some well-known conventional equalization techniques, i.e., decision feedback equalization (DFE) in single-carrier, and least square (LS) or minimum mean square error (MMSE) channel estimation plus single-step equalization in OFDM.