AbstractMost current wireless standards rely on the conventional orthogonal frequency-division multiplexing (OFDM) technology. Yet, classical OFDM may not be sufficient to support the emerging machine-type communications (MTC), which strictly require high reliability, ultralow latency connectivity among low-cost small devices such as sensors. Motivated by this, this thesis develops and analyzes a variety of novel reliable and low-complexity multicarrier systems based on index modulation (IM) and deep learning (DL) techniques, particularly tailored to mission-critical MTC applications, such as industrial automation. By integrating the IM concept into the OFDM framework, the resulting so-called OFDM-IM scheme activates only a subset of subcarriers and carries additional data bits via the active indices. Thus, OFDM-IM achieves not only higher reliability and energy efficiency, but also a better trade-off between reliability and spectral efficiency than OFDM. In this thesis, a novel framework for analyzing the error performance of OFDM-IM in the presence of imperfect state channel information and various detection types is first presented. Based on this, two novel IM-based multicarrier schemes are proposed to enhance transmit diversity of OFDM-IM, using either repetition codes or spreading matrices. The proposed schemes significantly improve reliability over existing schemes at the cost of increased detection complexity. Therefore, for practical deployments, various low-complexity near-optimal detectors are proposed.
The current IM-based schemes suffer from higher detection complexity than OFDM, which is considered as the penalty for their performance improvement. Moreover, most of them have focused on the single-user communications only. The second part of this thesis utilizes DL with deep neural networks (DNNs) in order to address these fundamental issues. In particular, a coherent DL-aided multicarrier system whose modulation and demodulation blocks are modeled by DNNs based on an autoencoder structure is proposed, which can learn to optimize both diversity and coding gains in fading channels to be higher than those of hand-designed schemes. In addition, a novel noncoherent energy autoencoder for multicarrier systems is proposed, which can efficiently decode data without any channel estimation schemes at either the transmitter or the receiver. These schemes are then extended to multiuser scenarios, in order to accommodate any numbers of users, sub-carriers, antennas and data streams, as well as any transmission directions. The proposed learning-based schemes achieve higher reliability and flexibility at even lower complexity compared to the baseline schemes, and thus are attractive to various MTC applications, which require reliable, low-latency and low-complexity communications.
|Date of Award||Jul 2020|
|Sponsors||Engineering & Physical Sciences Research Council|
|Supervisor||Youngwook Ko (Supervisor) & Michalis Matthaiou (Supervisor)|