Machine Learning Applications to Short Packet Communications

  • Ahlam Al-Shukaili

Student thesis: Phd

Abstract

Recent machine-type communications represent a significant paradigm shift that will revolutionize the design of wireless communication systems. This shift is driven by the promise of ultra-reliable low-latency communications (URLLC) introduced by 5G and 6G technology. Unlike the primary focus of conventional systems, which is on achieving high transmission rates, URLLC aims to support extremely low latency and high reliability in data transmissions. Thus, Short Packet Communication (SPC) is being introduced as a key enabler for URLLC. This thesis aims to enhance the performance of SPC using machine learning (ML) algorithms. Initially, we study the performance analysis of SPC and develop an accurate evaluation of the packet error probability in the presence of interference. Subsequently, we investigate the performance of sparse recovery algorithms within the context of SPC. Specifically, we propose two algorithms, Compressive Sampling Matching Pursuit (CoSaMP) and Stagewise Matching Pursuit (St-OMP), for sparse recovery. Also, we present a general form of the Symbol Error Rate (SER) utilizing pairwise error probability. Further, we investigate the potential application of ML techniques in SPC. We apply supervised learning, namely Support Vector Machine (SVM) and K-Nearest Neighbours (KNN), and compare them with the application of unsupervised learning, Expectation Maximization (EM), to SPC. To mitigate packet overhead, we employ the Label Assisted Transmission (LAT) method. Additionally, we utilize Silhouette Analysis to determine the optimal clustering number. Finally, we successfully use a supervised learning approach to recover the spreading factor in the Long-Range (LoRa) system using SVM and KNN. The applied algorithms showed significant improvements in the performance of SPC compared to the baseline schemes. Specifically, SVM and KNN algorithms show promising results in signal classification with different signal representations.
Date of Award31 Dec 2023
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorHujun Yin (Supervisor) & Khairi Hamdi (Supervisor)

Cite this

'