The confined nature of surface plasmons (SPs) often imposes challenges on their experimental detection and makes specific near-field probes necessary. While various SP detection methods have been developed in the optical domain, only a few examples of SP imaging have been reported in the terahertz range. In this thesis, specific problems of current terahertz near-field detection systems have been addressed which has led to the development of two new SP imaging methods. In the first method, SP imaging is demonstrated using the integrated subwavelength aperture near-field probe. The photoconductive antenna inside the probe is sensitive to the SP electric-field despite the orthogonal spatial orientation between the antenna and the SP polarisation. This enables SP imaging directly on a metallic surface employing a photoconductive antenna. This unexpected sensitivity has been applied to SP imaging in two examples: first, the SP propagation has been imaged on a resonant THz bow-tie antenna and second, the SP excitation by a strongly focused terahertz beam directly on the metallic probe surface has been investigated. The second method presents an electro-optic micro-resonator for SP imaging. A micro-resonator structure has the potential to provide a better sensitivity and spatial resolution, as well as a lower level of invasiveness compared to bulk crystals, which are commonly used in terahertz near-field systems. The micro-resonator design is explained in detail and the impact of the micro-resonator geometry on the probe performance is discussed. This micro-resonator has then been fabricated and embedded into an electro-optic detection system. This detection system has been fully characterised with the focus on two functional units which are essential for its performance: a tapered parallel plate waveguide for broadband terahertz transmission and the balanced detector for noise reduction. The overall performance of the detection system has been evaluated for its use as a terahertz near-field microscope.

No description available.

The successful development of general-purpose humanoid robots, in contrast to traditional pre-programmed problem solving machines, has opened a new research area of how a robot could be programmed by an end-user, not engineers, to suit individual needs. In this respect, Robot Learning from Demonstration has been actively studied, aiming to enable robots learn various tasks from human users. Although much effort has been put, there are many challenges still remaining until the goal is realized. One of the important challenges is the automatic learning of task representations and reuse of the learned tasks, where each task can be expressed as a series of primitive action components. To deal with such challenges, syntactic approaches to task learning and related issues are investigated. Firstly, efficient goal-oriented task representation methods using stochastic context-free grammars are studied, which enable robots to understand the human's intended actions even in the presence of both observation errors and human execution errors. By exploiting the task knowledge, it is demonstrated that the robot can correctly identify unexpected, out-of-context actions and perform the intended actions under reasonable amount of noise. Taking a step further, the automatic learning of these task representations from human demonstrations are studied. It is demonstrated throughout the experiments that the robot is able to learn critical task structures and generalize them. This is essential for understanding more complex tasks sharing the same underlying structures. Following these studies, the unsupervised discovery of the optimal number of primitive action detectors required to represent a task is studied. Through a diverse set of real-world and simulated experiments that include learning object-related games, postural sequence tasks of dance and surveillance tasks, this thesis demonstrates the effectiveness of syntactic approaches for robot learning from demonstrations.

The field of energy harvesting comprises all methods to produce energy locally and from surrounding sources, e.g. solar illumination, thermal gradients, vibration, radio frequency, etc. The focus of this thesis is on inertial power generation from host motion, in particular for low frequency and random excitation sources such as the human body. Under such excitation, the kinetic energy available to be converted into electrical energy is small and conversion efficiency is of utmost importance. Broadband harvesting based on frequency tuning or on non-linear vibrations is a possible strategy to overcome this challenge. The technique of frequency up-conversion, where the low frequency excitation is converted to a higher frequency that is optimal for the operation of the transducer is especially promising. Regardless of the source excitation, energy is converted more efficiently. After a general introduction to the research area, two different prototypes based on this latter principle and using piezoelectric bending beams as transducers are presented, one linear design and one rotational. Especially for human motion, the advantages of rotational designs are discussed. Furthermore, magnetic coupling is used to prevent impact on the brittle piezoceramic material when actuating. A mathematical model, combining the magnetic interaction forces and the constitutive mechanical and electrical equations for the piezoelectric bending beam is introduced and the results are provided. Theoretical findings are supported by experimental measurements and the calculation model is validated. The outcome is the successful demonstration of a rotational energy harvester, tested on a custom made shaking set-up and in the real world when worn on the upper arm during running.

In recent years, the growth of smart phone industry drives its users to demand better quality multimedia services while achieving faster internet speeds. With a higher number of subscribers and more data hungry applications, there is a need to improve the current 3G HSDPA data rate until the 4G network is fully implemented globally. Therefore, the aim of this thesis is to improve the downlink data rate of the HSDPA system throughput through both transmitter and receiver optimisations while balancing the complexities of the optimisation algorithms. Initially, improving system throughput through power allocation strategies is investigated in this thesis. Current schemes proposed in the literature that use Equal SINR Equal Rate (ESER) allocation can improve the system throughput for single rate allocation but require a rather high complexity. In this thesis, a System Value (SV) criterion is proposed to decrease the traditional ESER allocation method while achieving similar system throughput performance. To further increase the system throughput, a Successive Interference Cancellation (SIC) receiver is proposed which not only improves the data rate but also has a reduced complexity. The proposed SIC receiver can be used in conjunction with both ESER allocation schemes as well as equal energy allocation schemes. Another transmitter optimisation strategy is to improve the signature sequences. A signature sequence design that minimises correlations between the sequences while reducing the Inter-Symbol Interference (ISI) and noise is proposed. To reduce the amount of channel state information required, a signature sequence selection method which removes the signature sequences that are heavily affected by interference is also proposed. The proposed optimisation strategies have been verified through MATLAB simulations. Moreover, practical measurements using the National Instruments (NI)-PXIe testbed have been carried out for the proposed SIC algorithm which further confirm its effectiveness in a real world environment.

This thesis investigates learning-based resource allocation techniques for future wireless networks (FWNs). Motivated by recent technological developments, two types of FWNs are studied: energy harvesting (EH) wireless sensor networks (WSNs), and high-capacity cellular networks (HC-CNs) with caching capabilities. In an EH-WSN, each node is powered by a rechargeable battery and harvests energy from the environment. First, a multi-access throughput optimisation problem is studied, when the access point schedules EH sensor nodes without the knowledge of their battery states. A low-complexity policy is shown to be optimal in certain cases, and a scheduling algorithm, which takes into account the random processes governing the energy arrivals in the system, is proposed, and compared to an upper bound. Second, a point-to-point communication system with an EH transmitter is considered. Since the characteristics of the environment in which the sensor will be deployed are not known in advance, we assume no a priori knowledge of the random processes governing the system, and propose a learning theoretic optimisation for the system operation. The performance of the proposed algorithm is compared to that of two upper bounds, obtained by providing more information to the transmitter about the random processes governing the system. We then turn our attention to content-level selective offloading to an infostation terminal in an HC-CN. The infostation, which stores high data-rate content in its cache memory, allows cellular users in the vicinity to directly download the stored content through a broadband connection, reducing the latency and the load on the cellular network. The goal of the infostation cache controller is to store the most popular content such that the maximum amount of traffic is o oaded to the infostation. The optimal cache content management problem when content popularity is unknown is studied, and a number of algorithms to learn the content popularity pro le are proposed. The performances of these algorithms are compared to that of an informed upper bound, obtained when the content popularity pro le is known.

In this thesis, we consider algorithms and systems which dynamically guide evacuees towards exits during an emergency to minimise building evacuation time. We observe that the "shortest safe path" routing approach is inadequate when congestion is a predominant factor, and therefore focus on systems which manage congestion. We first implement a "Reactive" metric which compares paths based on real-time transit times. We find that regular route corrections must be issued to address the constant changes in path delays, and that routes oscillate. We also implement a model-based "Proactive" metric which forecasts the increase in future congestion that results from every routing decision, allowing the routing algorithm to operate offline. We combine both metrics with the Cognitive Packet Network (CPN), a distributed self-aware routing algorithm which uses neural networks to efficiently explore the building graph. We also present the first thorough sensitivity analysis on CPN's parameters, and use this to tune CPN for optimal performance. We then compare the proactive and reactive approaches through simulation and find both approaches reduce building evacuation times -- especially when evacuees are not evenly distributed in the building. We also find major differences between the Proactive and Reactive approach, in terms of stability, flexibility, sensory requirements, etc. Finally, we consider guiding evacuees using dynamic exit signs, whose pointing direction can be controlled. Dynamic signs can readily be used with Reactive routing, but since Proactive routing issues routes on an individual basis, one display is required for each evacuee. This is incompatible with dynamic signs; therefore we propose a novel algorithm which controls the dynamic signs according to the Proactive algorithm's output. We simulate both systems, compare their performance, and review their practical limitations. For both approaches, we find that updating the sign's display more often improves performance, but this may reduce evacuee compliance and make the system inefficient in real-life conditions.

This thesis is in the field of Optimal Control. It addresses research questions concerning both the properties of optimal controls and also schemes for control system stabilization based on the solution of optimal control problems. The first part is concerned with the derivation of necessary conditions of optimality for two classes of optimal control problems not covered by earlier theory. The first is the class of optimal control problems with a combination of mixed control-state constraints and pure state constraints in which the dynamics are described by a differential inclusion under weaker hypotheses than have previously been considered. The second is the class of optimal control problems in which the dynamics take the form of a non-smooth differential equation with delays, and where the end-time is included in the decision variables. We shall demonstrate that these new optimality conditions lead to algorithms for solution of certain optimal control problems not amenable to earlier theory. Model Predictive Control (MPC) is an approach to control system design based on solving, at each control update time, an optimal control problem. This is the subject matter of the second part of the thesis. We derive new MPC algorithms for constrained linear and nonlinear systems which, in certain significant respect, are simpler to implement than standard schemes, and which achieve performance specification under more general conditions than has previously been demonstrated. These include stability and feasibility.

This thesis is concerned with the Robust Model Predictive Control (RMPC) of linear discrete-time systems subject to norm-bounded model-uncertainty, additive disturbances and hard constraints on the input and state. The aim is to design tractable, feedback RMPC algorithms that are based on linear matrix inequality (LMI) optimizations. The notion of feedback is very important in the RMPC control parameterization since it enables effective disturbance/uncertainty rejection and robust constraint satisfaction. However, treating the state-feedback gain as an optimization variable leads to non-convexity and nonlinearity in the RMPC scheme for norm-bounded uncertain systems. To address this problem, we propose three distinct state-feedback RMPC algorithms which are all based on (convex) LMI optimizations. In the first scheme, the aforementioned non-convexity is avoided by adopting a sequential approach based on the principles of Dynamic Programming. In particular, the feedback RMPC controller minimizes an upper-bound on the cost-to-go at each prediction step and incorporates the state/input constraints in a non-conservative manner. In the second RMPC algorithm, new results, based on slack variables, are proposed which help to obtain convexity at the expense of only minor conservatism. In the third and final approach, convexity is achieved by re-parameterizing, online, the norm-bounded uncertainty as a polytopic (additive) disturbance. All three RMPC schemes drive the uncertain-system state to a terminal invariant set which helps to establish Lyapunov stability and recursive feasibility. Low-complexity robust control invariant (LC-RCI) sets, when used as target sets, yield computational advantages for the associated RMPC schemes. A convex algorithm for the simultaneous computation of LC-RCI sets and the corresponding controller for norm-bounded uncertain systems is also presented. In this regard, two novel results to separate bilinear terms without conservatism are proposed. The results being general in nature also have application in other control areas. The computed LC-RCI sets are shown to have substantially improved volume as compared to other schemes in the literature. Finally, an output-feedback RMPC algorithm is also derived for norm-bounded uncertain systems. The proposed formulation uses a moving window of the past input/output data to generate (tight) bounds on the current state. These bounds are then used to compute an output-feedback RMPC control law using LMI optimizations. An output-feedback LC-RCI set is also designed, and serves as the terminal set in the algorithm.

The constant technology shrinking and the increasing demand for systems that operate under different power profiles with the maximum performance, have motivated the work in this thesis. Modern design tools that target FPGA devices take a conservative approach in the estimation of the maximum performance that can be achieved by a design when it is placed on a device, accounting for any variability in the fabrication process of the device. The work presented here takes a new view on the performance improvement of DSP designs by pushing them into the error-prone regime, as defined by the synthesis tools, and by investigating methodologies that reduce the impact of timing errors at the output of the system. In this work two novel error reduction techniques are proposed to address this problem. One is based on reduced-precision redundancy and the other on an error optimisation framework that uses information from a prior characterisation of the device. The first one is a generic architecture that is appended to existing arithmetic operators. The second defines the high-level parameters of the algorithm without using extra resources. Both of these methods allow to achieve graceful degradation whilst variation increases. A comparison of the new methods is laid against the existing methodologies, and conclusions drawn on the tradeoffs between their cost, in terms of resources and errors, and their benefits in terms of throughput. In some cases it is possible to double the performance of the design while still producing valid results.