Logo image
Machine learning assisted indoor visible light communication system
Thesis   Open access

Machine learning assisted indoor visible light communication system

Kopano Victor Menu
M.Eng., University of Johannesburg
2024
Handle:
https://hdl.handle.net/10210/519993

Abstract

Visible light communication (VLC) has emerged as a new and promising technology in wireless communications. Its ability to provide illumination, energy transfer, and data transmission through light-emitting diodes (LEDs) makes it a valuable innovation for next-generation communication systems. Among its multiple advantages are high data transfer speeds, reduced interference with radio frequency signals, improved security in the indoor environment, low cost and complexity, and lower hardware investment. However, despite these benefits, VLC systems are hindered by nonlinear channels, which have a substantial impact on data transmission accuracy and rate. In this thesis, by leveraging their capacity to solve nonlinear problems, machine learning (ML) algorithms are integrated to the VLC system to address the challenges posed by VLC channel nonlinearity and predict the received message. The work focuses on an indoor VLC system with multiple inputs and outputs. Key aspects of the system, such as LED placement, received power distribution, signal-to-noise ratio, and channel impulse response due to LED configuration are analyzed. We explored the application of various ML classification algorithms, including Random Forest (RF), Decision Trees (DT), Support Vector Machine (SVM), and Na¨ıve Bayes (NB) to enhance the indoor VLC system performance. Each algorithm’s performance is evaluated. DT, RF, SVM, AND NB produced average precision accuracy of 76.47%, 74.01%, 77.21%, and 77.21%, and average cross-validation of 79.36%, 82.96%, 78.58%, and 77.82%, highlighting the potential of ML in mitigating nonlinear challenges faced by VLC systems and demonstrating major advancements while identifying areas for further improvement.
pdf
K_V_Menu_Final_Submission_Final Draft_Rev26.17 MBDownloadView
Open Access

Metrics

1 Record Views

Details

Logo image