Abstract
Optical wireless communications (OWC) has recently gained a lot of interest among industrial and academic communities. The main inhibitor factor of this resurgence of interest is the fact that radio-frequency (RF) spectrum is already so densely occupied to handle the increasingly high demand, and hence, exploring higher frequency spectrum, including the optical range, would be a relief. Another reason behind such an interest resides in the relatively simple deployment of OWC systems. However, before a real deployment of OWC systems, there is a persistent need to establish its fundamental performance limits (e.g. capacity, secrecy capacity, and capacity region) and extract design guidelines for building efficient, reliable, and secure OWC systems. Indeed, due to different propagation channels and different transmit constraints, RF communications and OWC are fundamentally quite different. For instance, the popular intensity modulation and direct detection (IM-DD), which is a favorable scheme for OWC due to its simplicity, has some subtle differences in comparison with RF systems manifested in the nonnegativity of the transmit signal, in addition to constraints on the peak- and average-intensity of the signal. These, in turn, make the fundamental performance limits and the optimal transmission schemes for OWC based on IM-DD different from those for RF systems.
Since the fundamental performance limits of OWC play a vital role in extracting guidelines and communication protocols for designing reliable and secure systems, this dissertation addresses those limits in an OWC setting. Particularly, this dissertation presents novel contributions to the understanding of the fundamental limits of multiuser OWC with and without secrecy constraints. When a secrecy constraint is imposed, this dissertation provides analytical results on the characterization of the optimal transmission schemes for secure and reliable OWC when input-dependent Gaussian noise and Poisson noise models are considered. Additionally, an asymptotic analysis of the secrecy capacity (the fundamental performance limit for secure communications) is presented. Furthermore, a two-user optical multiple access channel model, which depicts a multiuser OWC scenario without secrecy constraints, is proposed and the optimal multiuser transmission schemes that achieve the capacity region (fundamental performance limit of this multiuser scenario) are developed. Moreover, the capacity region of the considered optical multiple access channel is explicitly characterized in a closed-form expression in the regime where the peak- and average-intensity constraints are vanishingly small. After establishing the fundamental performance limits of OWC, powerful machine learning techniques, such as deep learning, are employed for the implementation of OWC systems. In particular, a simple and cost-effective learning-based system with (near-)optimal performance is proposed and is implemented by merely taking off-the-shelf deep learning models, applying them to an OWC design problem, and tuning them based on the easily generated training data.