Logo image
Development of software-defined radio platforms and advanced error pattern models using generative techniques for robust power line communications
Dissertation   Open access

Development of software-defined radio platforms and advanced error pattern models using generative techniques for robust power line communications

Akintunde Oluremi Iyiola
Doctor of Philosophy (PHD), University of Johannesburg
2025
Handle:
https://hdl.handle.net/10210/519944

Abstract

Narrowband Power Line Communication (NB-PLC) enables low-cost, low-power connectivity for smart grid and Internet of Things (IoT) applications by leveraging existing electrical infrastructure. However, its performance in indoor settings is hindered by impulsive noise, burst errors, and temporally correlated interference, which challenge reliable communication and robust protocol design. This thesis proposes a statistically grounded, experimentally driven approach to evaluating NB-PLC performance under bursty channel conditions. First, an uncoded, reconfigurable software-defined radio (SDR) testbed—drawing on the structure of the International Telecommunication Union – Telecommunication Standardization Sector (ITU-T) Home Networking Energy Management (G.hnem) standard—is developed to capture single- and multicarrier NB-PLC signals in realistic indoor settings, yielding empirical error traces with modulationspecific temporal characteristics. Second, empirical error traces are used to train generative models that capture the temporal correlation and burstiness of NB-PLC channels. Semi-Hidden Fritchman–Markov Models (SHFMMs) are applied to bit-level data using standard and modified Baum–Welch algorithms (BWAs), and evaluated via log-likelihood and error-free run distributions. A Block-Diagonal Semi-Hidden Fritchman-Markov Model (BD-SHFMM) variant is introduced to reduce training complexity through run-length-based estimation while preserving statistical fidelity. Third, BD-SHFMMs are integrated into a modular simulation testbed to assess forward error correction (FEC) schemes under realistic burst-error conditions. Reed–Solomon (RS), convolutional, and concatenated RS + convolutional codes are evaluated across modulation formats and noise levels. Only the concatenated scheme consistently achieves low error rates and effective error dispersion, supporting the G.hnem standard design and revealing the shortcomings of memoryless error models. Collectively, the thesis offers a reproducible, statistically grounded framework for NB-PLC performance analysis that unifies physical-layer testing, stochastic error modeling, and decoder evaluation. It contributes practical tools for burst-aware design, adaptive coding, and integration into future hybrid power line communication (PLC)–wireless systems.
pdf
Iyiola AO - 218032385 (FINAL)3.83 MBDownloadView
Open Access

Metrics

1 Record Views

Details

Logo image