Networked systems are now playing central and important roles in many engineering applications. Although a lot of research efforts have been made to the control of such networked systems, some important questions still remain unanswered. The thesis aims to identify and address some of the theoretical challenges in the optimization and control of networked systems that are not explored by the existing literature. In particular, the three main contributions of the thesis are in the area of asynchronous distributed optimization, online optimization using coordinate descent methods, and security and privacy of networked systems. For asynchronous distributed optimization, we propose an asynchronous algorithm to minimize the sum of convex cost functions using dual decomposition and block coordinate subgradient methods. The prominent feature of our algorithm is that the communication, computation and stepsize update in the network is not coordinated. Moreover, we show that our asynchronous algorithm covers some existing results in the literature as special cases. Under assumptions weaker than those used in the literature, we are able to prove almost sure convergence of the asynchronous algorithm. A numerical example is provided to illustrate the effectiveness of the algorithm. In addition to traditional offline optimization problems, we also consider the problem of online optimization where the optimization problem may change over time. In this part, we use coordinate descent methods to solve the time-varying optimization problem online, in an adaptive fashion. Specifically, we consider three coordinate selection rules, namely, random coordinate descent, cyclic coordinate descent, and Gauss-Southwell coordinate descent, that are commonly used in the literature of coordinate descent algorithms and extend them to the online setting. In contrast to offline optimization where convergence to the optimal point is desired, we solve the time-varying problem partially at each step and use the notion of regret as the measure of performance of our online algorithms. We provide an in-depth regret analysis for online coordinate descent algorithms under different assumptions and show that they are comparable to existing regret bounds of online gradient descent in the same settings. Lastly, a time-varying quadratic problem is studied numerically to illustrate the main results. Finally, we address the security and privacy of networked systems. We design controllers for nonlinear networked control systems (NCSs) using semi-homomorphic encryption which enables control signal computation using encrypted signals directly. Thus, the security of the NCSs is further enhanced by preserving the privacy of information flowing through the network connecting the plant and the controller. By using Lyapunov based methods, we provide sufficient conditions on the encryption parameters that guarantee robust stability of the NCS in the presence of disturbances which covers the existing result on linear systems as a special case. Furthermore, we discuss the trade-offs between the required computational resources, security guarantees and the closed-loop performance. We test our controller on a numerical example to show the performance of the encryption based controller.

Structured numerical methods are developed for solving a class of constrained network optimal control problems with discrete-time dynamics arising from the path-graph interconnection of N heterogeneous sub-systems. Structured dynamics of this kind are relevant in the operation of irrigation channels, vehicle platoons, supply chains, and radial power networks and, also arise from the discretization of one-dimensional partial differential equations. The size of the optimization problem grows with increase in the number of sub-systems N and the time horizon T. To solve large instances of such problems, involving potentially millions of variables, it is imperative to exploit the special spatio-temporal structure. In this thesis, two numerical algorithms are developed to exploit the special problem structure, with a view to achieving decomposable computations and suppressing the computational effort required to solve it. To start, a structured splitting method, based on the alternating direction method of multipliers (ADMM), is developed. It is shown that the computations at each iteration involve structured matrix-vector products, which are decomposable across spatial and temporal dimension of the optimal control problem, into parallelizable sub-problems of size independent of N and T. The overall arithmetic complexity of each ADMM iteration is O(NT). Secondly, a structured preconditioned conjugate gradient (PCG) solver is developed for the Newton steps of second-order methods. The main innovation pertains to O(NT) arithmetic complexity of each PCG iteration. Again, the associated computations are decomposable across spatial and temporal dimensions of the optimal control problem, into mostly parallelizable sub-problems of size independent of N and T. Numerical results are provided for a mass-spring-damper chain and an automated irrigation channel to compare the proposed techniques against each other and existing approaches. It is observed that the ADMM approach can take a large number of iterations to converge with high accuracy. This can result in higher overall arithmetic complexity, and correspondingly higher data-exchange overhead if computations are distributed for parallel implementation, compared to an interior point method in which the proposed structured PCG solver is used to determine Newton directions.

As applications of deep learning continue to be discovered and implemented, the problem of robustness becomes increasingly important. It is well established that deep learning models have a serious vulnerability against adversarial attacks. Malicious attackers targeting learning models can generate so-called "adversarial examples'' that are able to deceive the models. These adversarial examples can be generated from real data by adding small perturbations in specific directions. This thesis focuses on the problem of explaining vulnerability (of neural networks) to adversarial examples, an open problem which has been addressed from various angles in the literature. The problem is approached geometrically, by considering adversarial examples as points which lie close to the decision boundary in a high-dimensional feature space. By invoking results from high-dimensional geometry, it is argued that adversarial robustness is impacted by high data dimensionality. Specifically, an upper bound on robustness which decreases with dimension is derived, subject to a few mathematical assumptions. To test this idea that adversarial robustness is affected by dimensionality, we perform experiments where robustness metrics are compared after training neural network classifiers on various dimension-reduced datasets. We use MNIST and two cognitive radio datasets for our experiments, and we compute the attack-based empirical robustness and attack-agnostic CLEVER score, both of which are approximations of true robustness. These experiments show correlations between adversarial robustness and dimension in certain cases.

Optimization-based approaches are developed for two problems in the management of dynamical systems. The first approach relates to the scheduling of rigid-profile inputs under constrained continuous-time dynamics. A parsimonious discretization is developed which yields a sub-optimal but feasible solution to this non-convex semi-infinite program. The second approach relates to uncertain preview in constrained receding horizon control. An approach to managing the interaction between the controller and the preview scheduler is developed to ensure recursive feasibility, with scope to balance control and scheduling objectives.

The integrity of the positioning system solution can be defined as a measure of trust one can put in the value of the estimated position. Some authors have called it ``a guarantee of safety''. Due to the safety implications, the Global Navigation Satellite Systems (GNSS) position integrity monitoring was first used in civil aviation. Integrity has become a key performance metric for Intelligent Transport Systems (ITS) given the ongoing efforts to develop more robust integrity monitoring algorithms for land-based applications in GNSS-challenged environments. However, ``a guarantee of safety'' cannot be given without any risk of misleading information associated with it. This risk exists due to the different error sources that impact the GNSS signals and needs to be up to a specified tolerable level that differs depending on the application. Any integrity monitoring algorithm proposed for ITS needs to be able to deal with problems specific to the urban environments: reduced measurement availability, satellite/user geometry and presence of large errors such as multipath. The algorithm proposed in this research is based on a robust, multi-sensor and cooperative positioning system with the Particle Filter (PF) as the underlying position estimator. This research proposes three novel integrity monitoring algorithms built on Bayesian Receiver Autonomous Integrity Monitoring (BRAIM). By using BRAIM, the need for measurement redundancy is mitigated. Although, BRAIM (with the PF as the underlying estimator) tests for different number and different magnitudes of faults through the employment of multiple particles (i.e., hypotheses), Fault Detection and Exclusion (FDE) is added to BRAIM. FDE is added to remove the impact of large biases on the a posteriori distribution used to bound the integrity risk estimate. This results in a novel FDE+BRAIM algorithm. In order to decrease the estimated integrity risk, a Spatial Feature Constraint (SFC) algorithm is implemented to constrain solutions to feasible locations within a road feature. By integrating SFC and BRAIM, the second novel algorithm is proposed: SFC+BRAIM. Lastly, a combination of both novel algorithms is proposed: the FDE+SFC+BRAIM algorithm. The performance of the proposed algorithms was evaluated for GPS only, multi-sensor and distributed cooperative data. The best performance is achieved by cooperative SFC+BRAIM. That method achieved the median probability of misleading information of 5.84*10^-9/epoch for the horizontal alarm limit of 5 m and integrity risk of 1*10^-7/epoch.

System identification deals with the problem of building mathematical models of dynamical systems based on observed data. As data has become the main source of information in many settings nowadays and models of dynamical systems are used in most fields of science and engineering, system identification is of great practical relevance. However, the obtained model is of little use without a statement about the uncertainty assigned with the model. Finite-sample system identification (FSID) methods provide guaranteed confidence regions for the unknown model parameter of dynamical systems under mild statistical assumptions} for a finite number of data points. In this thesis, two FSID methods, the Leave-out Sign-dominant Correlation Region (LSCR) and Sign-Perturbed Sums (SPS) methods are extended to Errors-In-Variables (EIV) systems. In EIV systems the measurements of both the input and the output are corrupted by noise. The LSCR and SPS methods require a sequence which is independent corresponding to the true parameter but correlated when the model parameter is different from the true system parameter to construct the confidence region. In standard systems where the noise-free input is available, the data generating system is noise-invertible in the sense that the noise signals can be recovered from the measured signals given the true system. The recovered noise signal can play the role of the required sequence in the LSCR and SPS methods. However, EIV systems are not noise-invertible. Hence, standard FSID methods are not applicable. In this thesis, building on the LSCR and SPS methods three different approaches to the construction of the confidence regions for EIV systems have been developed. In the first approaches, all exiting signals are assumed to be i.i.d. Gaussian processes. An appropriate correlation sequence required for both LSCR and SPS is computed by a Kalman filter and accordingly, a state-space form of the EIV system where both input and output are regarded as outputs is utilized. The constructed confidence regions include the true parameter with a user-chosen probability, and parameter values different from the true ones will be left out of the confidence region as the number of data points increases. The theoretical results are illustrated in simulation examples. In the second approach, the input and the input noise are i.i.d. Gaussian process while there are no restrictive assumptions on the output noise. By utilizing an alternative regression model and swapping the role of the input and the prediction error in the LSCR and SPS methods, confidence regions are constructed which include the true model parameter with a guaranteed user-chosen probability. It is shown that the confidence regions are asymptotically included in an epsilon-neighborhood of the true parameter. The shape and size of the confidence regions are investigated, and an ellipsoidal approximation which can be computed at low computational cost is proposed for the SPS confidence region. The methods and their properties are illustrated in numerical experiments. Finally, the importance of constructing a confidence region for a module in a network of dynamical systems is the motivation behind the third approach. In dynamic networks, all measurements are usually contaminated by noise, and an EIV model is a natural representation of a module in the network. In a dynamic network, the input of a module is not usually an independent sequence since it is typically the output of another dynamical system. The third approach in this thesis constructs confidence regions for EIV systems without making any assumption on the true input signal. The approach is used to construct a confidence region for a single module in a simple cascade network by incorporating additional data and taking advantage of the cascade structure. This is done without estimating other modules in the network. The method is illustrated in numerical experiments.

The recent decade has witnessed the rapid growth of data traffic driven by various bandwidth-rich applications. Accordingly, both short-reach and long-haul fiber based optical networks are in great demand. For the long-haul transports, coherent detection is dominant due to its superior performance. Although the hardware structure of coherent systems possesses large footprint and the corresponding DSP algorithms are complicated, the cost is amortised by the high capacity and long transmission distance. While for short- to medium-reach transports such as intra- and inter- data center connections and metropolitan networks, cost is one primary concern. As such, direct detection has attracted extensive research interests due to its simple structure and low cost. To support short- and medium-reach optical transports in a cost-effective manner, field recovery is a promising solution since it enables the chromatic dispersion (CD) compensation. Given the cost of the transmission link, direct detection with the recovery of optical field has attracted extensive attention. For direct detection systems, the signal-signal beat interference (SSBI) induced by the square-law detection is a major limiting factor of obtaining the replica of information-bearing signal. As such, various algorithms dealing with SSBI have been proposed in the recent years. In this thesis, the optical field recovery of directly detected single sideband (SSB) and double sideband (DSB) signals has been studied and proposed. For SSB signals, without inserting a frequency gap to accommodate SSBI, Kramers-Kronig (KK) and iterative cancellation (IC) receivers enable the high spectral efficiency. The appropriate modulation formats fitting for both KK and IC receivers have been analysed. As KK and IC receivers are designed for the transmission links consisting of several spans of fiber, CD impacts on the performance of KK and IC receivers are investigated. Results show that the single-carrier modulation format is the better fit for KK receivers, while OFDM signals outperform single-carrier signals for IC receivers. Due to accumulated CD impacts after transmission, the peak-to-average power ratio (PAPR) of the single-carrier signals increases, which is more likely to violate the minimum phase condition of KK receivers compared to the back-to-back (btb) condition. Accordingly, the KK receiver requires a higher CSPR after transmission, while the optimal CSPR for the IC receiver remains the same as the btb case. The first-order polarization mode dispersion (PMD) impacts are also investigated, and it is demonstrated that PMD is not a major limiting factor for the KK receiver. For the field recovery of DSB signals, the direct detection scheme called carrier-assisted differential detection (CADD) has been theoretically analysed and experimentally demonstrated. The algorithm of recovering DSB signal field using CADD receiver has been elaborated, and the design guideline of CADD receiver including the joint optimization of several key parameters is given via simulations. Besides, the first-time experimental demonstration of the CADD receiver has been conducted. Experimental results show that the required receiver bandwidth is reduced by 41% compared with SSB based direct detection schemes. From the perspective of practical implementation, the IQ imbalance impacts of the CADD scheme have been analysed, and the tolerance of amplitude and phase mismatch is given. Lastly, to alleviate the requirement of high CSPR, several DSP algorithms have been proposed. For the SSB direct detection scheme, both enhanced SSBI mitigation and virtual CSPR enhancement schemes can effectively reduce the CSPR by 2 to 3 dB. For the DSB signal based CADD receiver, a simple but effective power loading scheme is proposed to enhance the performance of low-frequency subcarriers, and hence predominantly reduce the required high CSPR.

Wireless communication systems with low-resolution quantization are envisioned to be a major part in future wireless communication networks because of their potential to improve the energy efficiency of the network. In this thesis, we present a comprehensive and rigorous analytical investigation on the performance impact of using low-resolution phase quantization at the receiver of a wireless communication system, when compared to traditional high-resolution systems. To that end, we consider three different system setups; a point-to-point wireless communication system with coherent detection, a point-to-point wireless communication system with non-coherent detection and a multi-antenna system with coherent detection. We study the optimum detectors and draw fundamental insights on the error probability performance of low-resolution quantization systems in the presence of fading and noise. Firstly, we focus on coherent detection with M-ary phase shift keying (M-PSK) modulation and, derive the optimum maximum likelihood (ML) detector for a single-input single-output (SISO) system. Utilizing the structure of the derived detector, a general average symbol error probability (SEP) expression for M-PSK modulation with n-bit quantization is obtained when the wireless channel is subject to Nakagami-m fading. We show that a transceiver architecture with n-bit quantization is asymptotically optimum in terms of communication reliability if n is greater than or equal to log_2(M +1). The coherent detection techniques discussed above require channel state information (CSI) to be available at the receiver. Due to the non-linear nature of quantization, channel estimation has been one of the major challenges associated with low-resolution quantization based systems. Taking these into account, next we focus on non-coherent detection by adopting the differential quadrature phase shift keying (DQPSK) modulation scheme to differentially encode the transmit data. At the receiver side, we employ non-coherent detection that does not require instantaneous CSI. With DQPSK modulation, the ML detector is derived using which, a general average SEP expression with n-bit quantization is obtained when the wireless channel is subject to Rayleigh fading. It is shown that a transceiver architecture with n-bit quantization is asymptotically optimum in terms of communication reliability if n is greater than or equal to 4. That is, the decay exponent for the average SEP is the same and equal to 1 with infinite-bit and n-bit quantizers for n is greater than or equal to 4. Therefore, when the channel knowledge is not available at the receiver, the quantizer has to use one additional bit to achieve optimum communication robustness. Finally, we extend our investigation to low-resolution quantization based multi-antenna wireless communication systems equipped with one transmit antenna and N receive antennas. We derive the ML detector and then propose three sub-optimum detection rules based on selection combining which have less computational complexity compared to the ML detector. We also note that the simple sub-optimum detector that selects the path with the channel that locates the rotated constellation point furthest away from the decision boundary is asymptotically optimum in terms of communication reliability if n is greater than or equal to 3. An extensive simulation study is performed to illustrate the accuracy of the derived results.

The thesis addresses several critical challenges in the implementation of Model Predictive Control (MPC) for online settings, with a focus on the numerical strategies employed in solving the inherent optimisation problem at the centre of MPC. First, an MPC-specific early termination condition is considered for the family of interior-point solvers. The proposed condition allows the computational efforts associated with solving a class of MPC problems to be reduced without compromising the stability properties of the closed-loop system. Second, it is assumed that an optimisation algorithm has already been selected, and the design of a suboptimal MPC algorithm without terminal conditions is required. The proposed design approach considers the MPC problem horizon length and an acceptable suboptimality degree to minimise the algorithmic complexity associated with finding a solution. To this end, the stabilising properties of the feasible suboptimal solutions (with an appropriately defined measure of suboptimality with direct links with the closed-loop performance of the system) are utilised, along with the ability to estimate the algorithmic complexity of the process of obtaining such solutions. Through numerical simulations, it is shown that the smallest stabilising prediction horizon is not necessarily the optimal choice, and the complexity can be further reduced using a larger horizon length. This is shown to be consistent with the predictions obtained from the developed framework. Third, the case where the constraint-respecting stabilising control law is to be constructed using a set of precomputed (sub)optimal control laws. A framework for approximating the optimal control law with a special family of barycentric functions and the corresponding stability certification method is proposed. The proposed stability certificate is less conservative than the state-of-the-art approaches, which results in the method to require fewer precomputed control laws. The proposed methodology demonstrates sub-exponential growth of the number of approximation sub-regions, and potentially allows for Approximate Explicit MPC to be applied to a broader range of systems. Finally, a novel family of algorithms for solving finite-time optimal control problems with state and input constraints is proposed. The aforementioned family, termed interior-point DDP algorithms (IPDDP), are a product of combining the interior-point and differential dynamic programming (DDP) ideas. The interior-point DDP algorithms are of linear complexity in the problem's size and can either handle infeasible solution guesses or preserve the feasibility at all times. The IPDDP method is shown to have a local quadratic convergence without appealing to any convexity properties of the associated problem. Once these three main contributions of the thesis are completed, further potential research directions and extensions are outlined as avenues for future work.

In recent years, the evolution of mobile communication has projected a tremendous growth in the capacity demand of the cellular communication network. Hence, telecommunication service operators have been researching different methods to accommodate such enormous demand growth of data communications. One such approach was to deploy additional macrocells with advanced wireless technology to cater to the bandwidth demand. This approach was not a cost-optimal one due to limited signal spectrum, inter-site distances among cells, risk of higher electromagnetic radiation propagation. Hence, a heterogeneous network deployment is to encounter the increased capacity need would be a more robust solution. Such systems deploy small-cell Base Transceiver Station (BTS) with a smaller coverage radius, alongside the traditional macrocell BTSs, to counter the capacity need and related issues. The planning of such a Small-cell Network (SCN) requires extensive forms of studies, and the purpose would be to focus on specific aspects of network planning to influence the outcome of such tasks directly. These cellular wireless networks connect with a backhaul infrastructure to offer a cost-effective, high capacity, robust, energy-efficient and future-proof connectivity between these “small cells” and the core network. This thesis presents relevant research studies performed to optimise the deployment of wireless small-cell networks. Firstly, using a novel network planning algorithm, a network of small-cells is planned for different 4G carrier frequencies. This framework also maintained the Maximum Allowable Path Loss (MAPL) level for the transmitted signal from SC. The framework incorporated geographical terrain factors of ground elevation and slope values, locations and fixed coverage area formation for the selected small-cells. An energy and cost-effective optimised backhaul architecture, based on the Gigabit Passive Optical Network (GPON) technology, leveraging an existing optical fibre network resources is separately planned and dimensioned to connect with the planned small-cell network approach mentioned above. Next, the two different SCN and GPON planning methods are combined under one optimisation framework to construct a simplified network planning method applied to any cellular technology or GPON type utilised. Finally, a network capacity analysis is done, concerning the data consumption by devices, based on the population density over the case study area and the assigned 5G Small-cell (SC) carrier frequency data rate. Based on that information and other known constraints and parameters, a corresponding optimisation framework will be developed. This framework would utilise the concept of cellular frequency spectrum refarming to share the frequency spectrum of wireless signals. In turn, this allowed various types of cellular networks from different generations to function in the same wireless frequency spectrum. In summary, the technical research contribution presented in this thesis describes multiple approaches to plan a wireless small-cell network. The research also dimensions an appropriate optical backhaul network, for different cellular and optical network characteristics, within the premises of a heterogeneous telecommunications network. Additionally, we discussed some future research directions evolving from our work, alongside concluding remarks.

The increasing penetration of residential photovoltaic (PV) systems is reducing net demand leading to displacement of synchronous generation, with serious implications on the provision of Primary Frequency Response (PFR) following a contingency. Furthermore, distribution networks require management of excessive reverse power flows caused by residential PV system to avoid voltage or asset utilisation violations. To prevent distribution network problems exports limit are often imposed, but at the sacrifice of total power exported. Through pre-curtailment of maximum generation, or re-distribution of power through dynamic optimal export limits, it is possible for residential PV systems to create a power reserve for PFR. Furthermore, time-varying net demand from residential PV will lead to many changing operating states, with implications on the oscillatory performance of synchronous generators still online. In this context, this thesis investigates and proposes methodologies to determine the role of residential PV systems in the provision of PFR and the effect of a time-varying net demand on small signal stability. To achieve this, however, several of the corresponding challenges need to be understood. Firstly, the effects at the system-level from an increase in PV penetration need to be understood. It is required to model how synchronous generators change power output to in response to a change in net demand. The dispatch of PFR for the synchronous generators must also be considered. Secondly, any pre-curtailment of a PV system for PFR, will alter the net demand and potential PFR requirements. Thirdly, residential PV systems are connected to the power system through distribution networks. Distribution networks require management to prevent network issues related to high penetrations of residential PV systems, which influences net demand. This requires modelling and understanding how distribution networks operate and are restricted by their physical limitations, along with how they are managed. This all has an impact on the net demand at the system level which needs to be considered. Finally, the time-varying nature of a power system with high penetrations of PV (and the displacement it causes) presents a challenge in assessing small signal stability, whilst also being unable to relate the performance of specific constant remaining modes of oscillation throughout the day. This thesis addresses the aforementioned challenges as follows: A unit commitment (UC) is utilised to model the behaviour of generators in time, enabling modelling of changes in power output in response to residential PV, determining which generators are forced offline, as well as the distribution of PFR among the synchronous generators. The UC is modified to pre-curtail power of residential PV systems for PFR, accounting for the change in net demand from pre-curtailment in the supply of PFR. Using a modified IEEE-9 bus system, the findings highlight that residential PV systems providing PFR can prevent the inefficient and costly operation of synchronous generators (providing PFR) at low power outputs. The need for representing distribution networks to assess the role of residential PV systems providing PFR is demonstrated using a realistic Australian MV-LV residential feeder (from the primary substation to individual customers). Export limits are imposed to prevent steady-state distribution problems. The findings highlight that if distribution network constraints are not considered, the level of synchronous generator displacement may be significantly over-estimated, with corresponding knock on affects for PFR requirements. The application of optimal dynamic export limits beyond managing steady-state issues in distribution networks are applied for providing PFR. A method to translate these optimal dynamic export limits to enable a reserve via droop settings for PFR is proposed. It was found that there is a significant PFR reserve available across a power system if optimal dynamic export limits are used. This PFR reserve from residential PV systems can help reduce system costs with synchronous generators no longer operating at low power just to provide PFR. The small signal stability of the system is assessed considering a time-varying net demand and corresponding response of synchronous generators, by integrating a UC with a small signal stability study. Furthermore, a method is presented whereby oscillatory modes that remain despite displacement can be tracked. Results showed that oscillatory modes can change their damping behaviour significantly in time, with oscillatory modes changing criticality (which are the least damped). This is significant given that an approach not considering a time-varying net demand may miss these findings which may lead to improper damping.

Wireless local area networks (WLANs) have continually evolved during the last few decades to meet the ever-growing user demands. However, popular radio frequency technologies such as Wi-Fi are now experiencing a spectrum crunch due to a multitude of bandwidth hungry applications and limited bandwidth available in the sub-6 GHz bands. Therefore, a number of complementary technologies such as 60 GHz Wi-Fi, visible light communication and optical wireless communication have emerged to build high capacity WLANs in indoor spaces. Amongst these emerging WLAN technologies, optical wireless communication, operating in the infrared range, is becoming popular as it has access to virtually unlimited bandwidth compared to radio frequency technologies. With this huge spectrum resources, it is quite straightforward to establish wireless links over 10 Gbps with optical wireless communication. In addition to that, optical wireless communication has several advantages like not causing interference to existing WLANs, high security, and simple transceiver designs. Though the physical layer of optical wireless communication is being developed fast and brings unprecedented capabilities to WLAN landscape, upper layer protocols and architectures that are essential in harnessing the benefits of physical layer to provide multi-gigabit communication have received minimal attention so far. Therefore, this thesis explores the upper layer protocols, algorithms and architectures for optical wireless networks in homogeneous and heterogeneous settings. To this end, we first evaluate the suitability of the contention-based MAC protocol of Wi-Fi standard for optical wireless networks. The inefficiencies of the contention-based MAC protocol are highly pronounced at the higher data rates of optical wireless networks. Therefore, we introduce an improved version of the Wi-Fi MAC protocol with novel dynamic contention window tuning mechanism that can operate at multi-gigabit data rates. Second, due to the lack of availability of a simulation platform to evaluate the performance of optical wireless communication networks, we develop a simulation module for optical wireless networks in the Network Simulator-3 (ns-3) project. The proposed module can deploy optical wireless networks of different architectures and layouts, apply different scheduling algorithms, and channel models. To the best of our knowledge, this is the first multi-gigabit optical wireless network simulation module. Third, we explored novel network architectures for optical wireless networks considering the massive capacity, increased number of access points and smaller cells. Subsequently, we proposed the FLOWN (full-duplex split-plane optical wireless network) architecture for optical wireless networks. The FLOWN architecture is later generalised to all the upcoming WLANs such as Wi-Fi 6, 60 GHz Wi-Fi, and visible light communication to support homogeneous or heterogeneous WLAN deployments. It features a centralised pool of hardware and software resources, a high-capacity distribution network and advanced capabilities like full-duplex and split-plane operation. Further, delay-sensitive users can only receive guaranteed quality-of-service under contention-free MAC protocols. Therefore, most of the upcoming WLAN MAC protocols deploy hybrid versions of contention-based and contention-free MAC protocols to reap the advantages of both types. Hence, we finally introduce a contention-free MAC protocol for optical wireless networks with adaptable parameters that can be tuned to the traffic requirements of the current users. Overall, our work reported in this thesis provide simulation platform for optical wireless networks and also insight into design strategies that can be used to realise centralised multi-gigabit network architectures and MAC protocols.