| dc.description.abstract | Visible Light Communication (VLC) is an emerging wireless technology that employs
light-emitting diodes (LEDs) or lasers to transmit data over the visible light spectrum.
Due to its inherent advantages, such as license-free spectrum, high data rates, low
power consumption, and enhanced security, VLC has attracted significant attention for a
broad range of applications, including indoor wireless networking, vehicular communications,
underwater communications, smart lighting, and indoor positioning systems. Additionally,
VLC is immune to electromagnetic interference, making it particularly suitable
for environments where radio-frequency systems are undesirable or restricted.
Despite these advantages, VLC systems face several challenges, including limited coverage
range, susceptibility to interference from ambient light sources, and performance
degradation under mobility. One of the most critical challenges arises from noise, which
may originate from the inherent properties of light and hardware components. In particular,
signal-dependent shot noise (SDSN) and relative intensity noise (RIN) significantly
degrade the accuracy of channel estimation and localization in practical VLC systems.
In the first part of this thesis, we focus on channel estimation in the presence of SDSN,
proposing a neural-network-augmented estimation framework that integrates with traditional
estimators such as least squares (LS) and maximum likelihood estimation (MLE).We
develop a complete mathematical framework enabling analytical mean-square-error (MSE)
derivations and fair benchmarking. Simulations show that, under SDSN, the proposed
method consistently outperforms LS while remaining competitive when SDSN is absent.
In the second part, we extend to visible light positioning (VLP) under SDSN. For a SISO VLC system, we study range estimation using the extended Kalman filter (EKF),
MLE, and nonlinear least squares (NLS), and establish Bayesian Cramér–Rao lower bounds
(BCRLBs). Monte Carlo results confirm the analysis and demonstrate EKF’s superior
accuracy.
Finally, we address 2D localization and tracking for dynamic targets in a multipleinput
single-output (MISO) VLC configuration. Accounting for both SDSN and RIN, we
derive a closed-form 2D-BCRLB and show that EKF’s recursive updates yield superior
real-time tracking compared to static measurement-based methods. We further show that
increasing the number of light sources improves spatial diversity, and that RIN has a more
pronounced adverse effect on tracking accuracy than SDSN. Overall, this thesis provides a
noise-aware framework that bridges traditional estimation with machine learning for robust
VLC channel estimation and localization. | en_US |
