Privacy, including the right to privacy of correspondence, is a human right. Privacy-enhancing technologies, such as the Tor anonymity network, help maintain this right. The increasing use of Tor from mobile devices raises new challenges for the continued effectiveness of this low-latency anonymity network. Mobile Tor users may access the Internet from a range of wireless networks and service providers. Whenever a wireless network hands-off a mobile device’s connection from one access point to another, its external Internet Protocol (IP) address changes, and the connection to the Tor network is dropped. Every dropped connection requires the Tor circuit to be rebuilt. The time required to rebuild the circuit negatively impacts client performance. This research is the first to highlight this negative impact and to investigate the likely extent of the impact for typical usage scenarios and mobility models. The increased network churn caused by circuit rebuilding also negatively impacts anonymity. A novel metric (q-factor) is proposed here to measure the trade-off between anonymity and performance over the duration of a communication session. Two new solutions to the problems of managing mobility in a low-latency anonymity network are proposed in this thesis. The first solution relies on adaptive client throttling, based on a Kaplan-Meier estimator of the likelihood of a mobile network hand-off. The second solution relies on the use of a static bridge relay (mBridge) that acts as a persistent ‘home’ for a mobile Tor connection, so avoiding the need to recreate the Tor circuit whenever the mobile device is handed-off. The effectiveness of these solutions has been measured using the new q-factor metric. Both solutions provide better performance for mobile Tor clients than the standard Tor client implementation, although some performance reduction by comparison with static Tor clients remains. The bridge relay solution (mBridge) has been shown to offer better performance than client throttling, but is more vulnerable to certain types of attack. A strength of both solutions is that changes are restricted to client devices, the existing algorithms and protocols of the interior Tor network are unaffected.

Third party damage from activities such as work carried out by contractors’ poses risks to gas pipelines. Within the UK, emergency plans are drawn up in an attempt to mitigate the significant consequences of any pipeline failure. The Control of Major Accident Hazards 1999 and the Pipeline Safety Regulations 1996 place legislative requirements on UK gas infrastructure providers, to regularly test emergency plans with simulation exercises. The exercises are intended to support the preparation of responders for dealing with incidents of failure. Software simulation is not currently utilised to facilitate the testing of emergency response plans in the UK gas pipeline industry. This project serves to evaluate the user acceptance of a software simulation prototype to enable the testing of emergency response planning strategies in the UK gas industry. Current emergency planning legislation and strategies applied to satisfy legislation within the UK gas industry are reviewed. The adoption and application of software simulation for the development of applied skill in other industries is examined, to determine the potential for use in testing emergency response planning for gas incidents. The Technology Acceptance Model (TAM) is the theoretical framework that underpins the study of user acceptance, of a software simulation prototype designed for running exercises to test emergency response plans. A case study evaluation of the user acceptance of the prototype, by representatives experienced in testing emergency response planning strategies in the gas industry, is presented. The participants in this case study are drawn from the Police, Fire and Rescue Service, Local Authority and Gas infrastructure provider, that perform a range of job roles operating at Operational, Tactical and Strategical levels. The research findings demonstrate that the participants perceive software simulation of emergency response planning processes for gas incidents to be beneficial. The TAM claims that if users perceive a system to be useful they are likely to adopt that system. Furthermore if users don’t perceive a system to be easy to use, according to the TAM, they will still adopt it after the correct training has been provided. Users would be most likely to adopt and use the software to facilitate emergency response planning exercises, if the correct training is provided. Software simulation offers great potential for the testing of emergency plans, it provides a controlled environment where decisions and responses can be audited and mistakes can be made without serious consequence. Software simulation has been shown to enhance, rather than replace, existing emergency response planning processes.

This research aims to develop a realistic solution to enhance the efficiency of a transport logistics operation. The case study in this research is one of the largest agricultural suppliers in Northern Thailand. The cost of logistics in Thailand is relatively high compared to other countries, i.e. 11% of Gross Domestic Product (GDP) in 2007, and is particularly high in agricultural sector. The focus of the study is to enhance and improve transportation activities which typically account for the largest cost in logistics. The research is entitled ‘RFID enabled constraint based scheduling for transport logistics ’ The dissertation studies two important research components: 1) the data acquisition using Radio Frequency Identification Technology (RFID) for monitoring vehicles in a depot and 2) the scheduling by solving Constraint Satisfaction Optimisation Problem (CSOP) using Constraint Programming (CP). The scheduling problem of the research is to compose and schedule a fleet in which both private and subcontracting (outsourcing) vehicles are available, but to minimise the use of subcontractors. Several contributions from this study can be identified at each stage of the study ranging from extensively reviewing the literature, field studies, developing the RFID prototype system for vehicle tracking, modelling and solving the defined scheduling problems using Constraint Programming, developing a RFID-CP based real time scheduling, and validating the proposed methods. A number of validations are also carried out throughout the research. For instance, laboratory based experiments were conducted to measure the performance of the developed RFID tracking system in different configurations. Scenario tests were used to test the correctness of the proposed CP-based scheduling system, and structure interviews were used to collect feedbacks on the developed prototype from the case study company.

Active integrated antenna (MA) is a very popular topic of research during recent decades. This is mostly due to its advantages, such as compact size, multiple functions and low cost, etc. The MA system can be regarded as an active microwave circuit which the output or input port is free space instead of a conventional 50-ohm interface. The major drawbacks of the conventional MA include narrow bandwidth, low efficiency, etc. An experimental investigation on broadband slot-coupled antenna is carried out, which results an impedance bandwidth of 50 % is achieved by both a ring slot- coupled and square ring slot-coupled patch antenna. An improved design technique for broadband class-E power amplifier (PA) design, based on the theoretical analysis done by Mader [2], is introduced to calculate the circuit parameters. The technique is applied to a RF microwave class-E power amplifier design (PA) that results a bandwidth of 12 % power added efficiency (PAE) greater than 60 % is achieved. The aim of this work is to design broadband high efficiency linearly polarized (LP) and circularly polarized (CP) MA and arrays that will be useful for mobile communication system. The MA does not need conventional matching network between the amplifier and the antenna, because the antenna serves as both a harmonics-tuning network and a radiator. A novel high efficiency broadband LP MA is demonstrated using a ring slot-coupled patch antenna with a class-E PA. It exhibits a PAE over 50 % within a 14.6 % bandwidth. For the first time, a high efficiency broadband CP MA is designed using a class-E PA integrated with a broadband CP antenna. The CP AIA achieves a PAE over 50 % within a 14 % bandwidth. The axial ratio of the CP MA is below 3 dB over a 9 % bandwidth. For further improve the performance, a novel L-shaped slot-coupled broadband CP MA is employed in a 2x2 array. The array consists of four sequentially rotated broadband CP antenna elements with an element spacing of half a free space wavelength. The antenna was designed to operate in the 3G band around 2 GHz. A bandwidth of 22.7 % PAE greater than 50 % is achieved together with a peak PAE of 71.35%. A bandwidth of 27 % axial ratio below 3 dB is resulted.

This thesis presents the research carried out into the effects of load and source harmonic terminations on the efficiency of a class-F power amplifier (PA). A demonstration on the direct integration of the class-F PA and an H-shaped patch antenna using an active integrated antenna (AIA) approach is also presented. To obtain a high efficiency PA, it is necessary to ensure that the power dissipated in the active device is minimised and this is achieved by ensuring that the overlapping area between the drain voltage and current waveforms in the time domain is minimised. To minimise this overlapping area, optimum source and load harmonic impedances for the fundamental frequency, and second and third harmonics, are obtained using a novel application of the simulated load/source-pull method. New forms of harmonic matching networks were designed to ensure that the active device is terminated by the optimum impedances at the gate and drain for maximum efficiency. Three PAs were designed, one operating at 0.9 GHz and the other two at 2.45 GHz. For the 0.9 GHz PA the load matching network was designed to obtain the required optimum impedances at the fundamental frequency, and second and third harmonics. As the effect of gate capacitance is small at this frequency, the source matching network was designed to obtain a conjugate match at the fundamental frequency only. At the higher frequency of 2.45 GHz, the gate capacitance has a larger effect on the efficiency of the PA and hence two designs were investigated and compared. In the first PA design the load and source matching networks were designed to obtain optimum impedances at the fundamental frequency and second harmonic. For the second PA design, these networks were designed to obtain optimum impedances at the fundamental frequency, and second and third harmonics. For these three PAs the simulated drain voltage/current waveforms, return loss, stability factor, power gain, output power and power added efficiency (PAE) are presented. The practical results are compared with those obtained by simulation. Each PA produced a PAE of greater than 70% and good agreement was obtained between the simulated and measured results. The PAE obtained in these works is comparable to that reported in published papers. Based on this research five papers have been published in journals and conferences. An H-shaped microstrip patch antenna is used in the active integrated antenna (AIA) design. The antenna must not only act as a radiator and a harmonic suppresser but also as an optimum load for the PA so that it can be connected directly to the active device in order to obtain maximum efficiency. An extensive study on this antenna was carried out. The formulas for the first four mode frequencies were derived using odd and even mode analysis while a new and simpler formula for the fourth mode frequency was obtained. A systematic design approach to obtain the dimensions of the antenna is presented for an antenna operating at a given fundamental mode frequency. For matching, a new explicit matrix input impedance formula for the H-shaped antenna has been obtained using segmentation method. Using this formula, the location of the probe feed could be adjusted to obtain the required impedance at the pre-assigned frequency. MathCAD programming is used to implement the calculations in the design of two antennas. Good agreement between the predicted, simulated and measured results is obtained for the resonant mode frequencies, input impedance and return loss. Based on this research two papers have been published in journals.

This study investigates the use of error control code, discrete wavelet transform (DWT) and artificial neural network (ANN) to improve the link performance of an indoor optical wireless communication in a physical channel. The key constraints that barricade the realization of unlimited bandwidth in optical wavelengths are the eye-safety issue, the ambient light interference and the multipath induced intersymbol interference (ISI). Eye-safety limits the maximum average transmitted optical power. The rational solution is to use power efficient modulation techniques. Further reduction in transmitted power can be achieved using error control coding. A mathematical analysis of retransmission scheme is investigated for variable length modulation techniques and verified using computer simulations. Though the retransmission scheme is simple to implement, the shortfall in terms of reduced throughput will limit higher code gain. Due to practical limitation, the block code cannot be applied to the variable length modulation techniques and hence the convolutional code is the only possible option. The upper bound for slot error probability of the convolutional coded dual header pulse interval modulation (DH-PIM) and digital pulse interval modulation (DPIM) schemes are calculated and verified using simulations. The power penalty due to fluorescent light interference (FL I) is very high in indoor optical channel making the optical link practically infeasible. A denoising method based on a DWT to remove the FLI from the received signal is devised. The received signal is first decomposed into different DWT levels; the FLI is then removed from the signal before reconstructing the signal. A significant reduction in the power penalty is observed using DWT. Comparative study of DWT based denoising scheme with that of the high pass filter (HPF) show that DWT not only can match the best performance obtain using a HPF, but also offers a reduced complexity and design simplicity. The high power penalty due to multipath induced ISI makes a diffuse optical link practically infeasible at higher data rates. An ANN based linear and DF architectures are investigated to compensation the ISI. Unlike the unequalized cases, the equalized schemes don‘t show infinite power penalty and a significant performance improvement is observed for all modulation schemes. The comparative studies substantiate that ANN based equalizers match the performance of the traditional equalizers for all channel conditions with a reduced training data sequence. The study of the combined effect of the FLI and ISI shows that DWT-ANN based receiver perform equally well in the present of both interference. Adaptive decoding of error control code can offer flexibility of selecting the best possible encoder in a given environment. A suboptimal 'soft' sliding block convolutional decoder based on the ANN and a 1/2 rate convolutional code with a constraint length is investigated. Results show that the ANN decoder can match the performance of optimal Viterbi decoder for hard decision decoding but with slightly inferior performance compared to soft decision decoding. This provides a foundation for further investigation of the ANN decoder for convolutional code with higher constraint length values. Finally, the proposed DWT-ANN receiver is practically realized in digital signal processing (DSP) board. The output from the DSP board is compared with the computer simulations and found that the difference is marginal. However, the difference in results doesn‘t affect the overall error probability and identical error probability is obtained for DSP output and computer simulations.

In line with technological developments, there is almost no limit to collect data of high dimension in various fields including bioinformatics. In most cases, these high dimensional datasets contain many irrelevant or noisy features which need to be filtered out to find a small but biologically meaningful set of attributes. Although there have been various attempts to select predictive feature sets from high dimensional data in classification and clustering, there have only been limited attempts to do this for regression problems. Since supervised feature selection methods tend to identify noisy features in addition to discriminative variables, unsupervised feature selection methods (USFSMs) are generally regarded as more unbiased approaches. The aim of this thesis is, therefore, to provide (i) a comprehensive overview of feature selection methods for regression problems where feature selection methods are shown along with their types, references, sources, and code repositories (ii) a taxonomy of feature selection methods for regression problems to assist researchers to select appropriate feature selection methods for their research (iii) a deep learning based unsupervised feature selection framework, DFSFR (iv) a K-means based unsupervised feature selection method, KBFS. To the best of our knowledge, DFSFR is the first deep learning based method to be designed particularly for regression tasks. In addition, a hybrid USFSM, DKBFS, is proposed which combines KBFS and DFSFR to select discriminative features from very high dimensional data. The proposed frameworks are compared with the state-of-the-art USFSMs, including Multi Cluster Feature Selection (MCFS), Embedded Unsupervised Feature Selection (EUFS), Infinite Feature Selection (InFS), Spectral Regression Feature Selection (SPFS), Laplacian Score Feature Selection (LapFS), and Term Variance Feature Selection (TV) along with the entire feature sets as well as the methods used in previous studies. To evaluate the effectiveness of proposed methods, four different case studies are considered: (i) a low dimensional RV144 vaccine dataset; (ii) three different high dimensional peptide binding affinity datasets; (iii) a very high dimensional GSE44763 dataset; (iv) a very high dimensional GSE40279 dataset. Experimental results from these data sets are used to validate the effectiveness of the proposed methods. Compared to state-of-the-art feature selection methods, the proposed methods achieve improvements in prediction accuracy of as much as 9% for the RV144 Vaccine dataset, 75% for the peptide binding affinity datasets, 3% for the GSE44763 dataset, and 55% for the GSE40279 dataset.

Many animals, especially flying insects are experts on reacting to approaching predators. For robots, the ability to avoiding collisions is also crucial. In locusts, a visual neuron called the Lobula Giant Movement Detector (LGMD) has been identified to be responsible for evoking collision avoidance behaviours. It has been modelled for collision avoidance on large robots or vehicles whose computational power are abundant. For micro robots, however, the limited computational capabilities on-board prevent the LGMD model to be accomplished on the robot by its own. Therefore in earlier researches, those micro robots serve only as image grabbers and motion actuators, leaving majority of the model processed on a host device connected. The unavoidable communication and consequent latency have become the bottlenecks that restrains the employment of this promising collision avoidance model in multi-agent research fields such as swarm robotics. This research focuses on the embedded modelling and realization of this bio-inspired collision sensitive model ELGMD. By carefully considering the required on-board resource, a novel micro robot Colias IV is designed to meet the requirements. Featured with the sufficient computing power, various of sensing modalities including a tiny camera, the modularized design and other specialities, this robot has become an advantageous platform to perform embedded vision tasks. The bio-inspired neural model Embedded-LGMD (ELGMD) is realized on the micro robot that can run autonomously without any off-board guidance. Optimization on the structure and timing has guaranteed its computational efficiency. The performance of the ELGMD and the effectiveness on triggering the robot's collision avoidance behaviour are tested via systematic experiments. To achieve more precise interactive behaviours with other kinds of moving obstacles, a compound motion detection system is realized within the robot to detect various of motion patterns by integrating several neural models at a higher level, in which those LGMD-like neural models are accomplished by an unified ELGMD model with minimum reconfiguration. Experiments have been conducted to validate the improved ELGMD model and the compound motion detection system. Results of this research have demonstrated the design goals of all the proposed modules, including the hardware platform, the bio-inspired model and the compound motion detection system, indicating the practicability of implementing these bio-inspired visual motion sensing systems for further robotic studies.

Mobile social software tools have great potential in transforming the way users communicate on the move, by augmenting their everyday environment with pertinent information from their online social networks. A fundamental aspect to the success of these tools is in developing an understanding of their emergent real-world use and also the aspirations of users; this thesis focuses on investigating one facet of this: the exchange of social media. To facilitate this investigation, three mobile social tools have been developed for use on locationaware smartphone handsets. The first is an exploratory social game, 'Gophers' that utilises task oriented gameplay, social agents and GSM cell positioning to create an engaging ecosystem in which users create and exchange geotagged social media. Supplementing this is a pair of social awareness and tagging services that integrate with a user's existing online social network; the 'ItchyFeet' service uses GPS positioning to allow the user and their social network peers to collaboratively build a landscape of socially important geotagged locations, which are used as indicators of a user's context on their Facebook profile; likewise 'MobiClouds' revisits this concept by exploring the novel concept of Bluetooth 'people tagging' to facilitate the creation of tags that are more indicative of users' social surroundings. The thesis reports on findings from formal trials of these technologies, using groups of volunteer social network users based around the city of Lincoln, UK, where the incorporation of daily diaries, interviews and automated logging precisely monitored application use. Through analysis of trial data, a guide for designers of future mobile social tools has been devised and the factors that typically influence users when creating tags are identified. The thesis makes a number of further contributions to the area. Firstly, it identifies the natural desire of users to update their status whilst mobile; a practice recently popularised by commercial 'check in' services. It also explores the overarching narratives that developed over time, which formed an integral part of the tagging process and augmented social media with a higher level meaning. Finally, it reveals how social media is affected by the tag positioning method selected and also by personal circumstances, such as the proximity of social peers.

Sentiment can be defined as a tendency to experience certain emotions in relation to a particular object or person. Sentiment may be expressed in writing, in which case determining that sentiment algorithmically is known as sentiment analysis. Sentiment analysis is often applied to Internet texts such as product reviews, websites, blogs, or tweets, where automatically determining published feeling towards a product, or service is very useful to marketers or opinion analysts. The main goal of sentiment analysis is to identify the polarity of natural language text. This thesis sets out to examine quantitatively the factors that have an effect on sentiment analysis. The factors that are commonly used in sentiment analysis are text features, sentiment lexica or resources, and the machine learning algorithms employed. The main aim of this thesis is to investigate systematically the interaction between sentiment analysis factors and machine learning algorithms in order to improve sentiment analysis performance as compared to the opinions of human assessors. A software system known as TJP was designed and developed to support this investigation. The research reported here has three main parts. Firstly, the role of data pre-processing was investigated with TJP using a combination of features together with publically available datasets. This considers the relationship and relative importance of superficial text features such as emoticons, n-grams, negations, hashtags, repeated letters, special characters, slang, and stopwords. The resulting statistical analysis suggests that a combination of all of these features achieves better accuracy with the dataset, and had a considerable effect on system performance. Secondly, the effect of human marked up training data was considered, since this is required by supervised machine learning algorithms. The results gained from TJP suggest that training data greatly augments sentiment analysis performance. However, the combination of training data and sentiment lexica seems to provide optimal performance. Nevertheless, one particular sentiment lexicon, AFINN, contributed better than others in the absence of training data, and therefore would be appropriate for unsupervised approaches to sentiment analysis. Finally, the performance of two sophisticated ensemble machine learning algorithms was investigated. Both the Arbiter Tree and Combiner Tree were chosen since neither of them has previously been used with sentiment analysis. The objective here was to demonstrate their applicability and effectiveness compared to that of the leading single machine learning algorithms, Naïve Bayes, and Support Vector Machines. The results showed that whilst either can be applied to sentiment analysis, the Arbiter Tree ensemble algorithm achieved better accuracy performance than either the Combiner Tree or any single machine learning algorithm.