Abstract
The fifth-generation wireless communication system (5G) is expected to support a wide range of newly emerging applications alongside today’s current cellular mobile broadband services. One key use of 5G will be Ultra-reliable and Low-Latency Communications (URLLC), which guarantee the rapid delivery of short packets (up to 1 ms) at a success probability rate of 99.999\%. The challenging reliability and latency requirements of URLLC cannot be delivered with the existing cellular networks, as they require significant modification of the air interface. The first part of this thesis provides a fresh and in-depth examination of URLLC, by presenting a literature review on the proposed state-of-the-art technologies to enable URLLC, i.e. from the perspective of the Physical (PHY) and Medium Access Control (MAC) layers. The study emphasizes the latency aspect, discusses key challenges that arise when striving to meet URLLC requirements, and suggests potential related directions for future researchers.
The main aim of the thesis is to minimize the latency associated with the presence of the Hybrid Automatic Repeat reQuest (HARQ) feedback scheme. Thus, we initially investigated a supervised learning method to provide early HARQ (E-HARQ) feedback on the decodability status of the coded-received signal ahead of the process of completing decoding. This strategy permitted the transmitter to react faster thereby minimizing the signal round-trip time (RTT). The simulation results demonstrate the capacity of the proposed mechanism to speed up the feedback, enhancing the prediction accuracy by 12\% with the introduction of a new feature derived from channel state information.
In the third chapter, we aim to further expedite early HARQ feedback by decoupling the feedback from signal decoding. To this end, we exploited recent advancements in the field of recurrent neural networks (RNN) by proposing a novel deep learning-based algorithm (Deep-HARQ), that employs a Long Short-Term Memory (LSTM) to estimate the decodability of the coded-received in-phase and quadrature (I/Q) signals, prior to accomplishing the majority of the complex reception tasks. As demonstrated, the simulation results revealed a more rapid estimation response, with enhanced accuracy.
Lastly, we study a Swift HARQ protocol that can speedily respond when the link requires multiple HARQ attempts due to the poor radio propagation conditions. The strategy also relies on machine learning techniques to estimate early the decodability of the packet within the maximum allowable retransmitting attempts. This can allow the transmitter to react faster by dropping the not decodable packet, i.e., including the whole retransmission attempts, or activate the repetition mode where some HARQ feedback can be omitted. Numerical analysis and simulation achieve a delay reduction reach to more than half the latency of the conventional HARQ when the link requires multiple HARQ retransmission. Throughout this study, all the proposed solutions have been evaluated using a realistic dataset involving collecting training and validation samples from a waveform compatible with 3rd Generation Partnership Project (3GPP) 5G NR Release 15 standards.