Abstract
With the aim to meet the increasing demand of data rate, user capacity and quality
of services of networks, orthogonal frequency-division multiplexing (OFDM) systems
have been widely investigated and adopted in different communication scenarios during
the past two decades, e.g., wireless local area networks (WLAN), long-term evolution
(LTE) and 5G communications. It is appealing mainly in the sense that the
inter-symbol interference (ISI) wireless channel is converted into a parallel of ISI-free
sub-channels through Fourier transform with affordable computational complexity. Despite
the tremendous success that has been achieved, this well-investigated technique
faces limit, e.g., it trades the computational complexity affordability off the achieved
detection performance. Recent advances in this research area lies in the development
of deep learning algorithm for the design and optimisation on the multi-carrier system.
This thesis investigates the deep learning for the multi-carrier signal reception technique
in various multi-carrier systems. Relying on the strong nonlinear processing capability
of deep learning algorithms, a series of DNN architectures are first proposed to address
the multiuser frequency synchronisation problem for the OFDMA uplink system. The
established DNN architectures are designed by relying on the conventional OFDMA
system model. Then, a data-driven modular neural network (MNN), termed MCMNNet
is proposed to address the coherent signal detection for various multiuser multicarrier
system. Moreover, with the aim of effectively reducing the offline training
complexity, a transfer learning approach is tailored for the deep learning algorithms
presented in this thesis.
The research in this thesis potentially offers the benefit of improved detection performance,
reduced offine training complexity to the deep learning-enabled multi-carrier
receiver design. The ultimate goal is to pave the path towards better development and
utilisation of deep learning for the wireless multi-carrier system design and optimisation.