Texture analysis is one of the key techniques of image understanding and processing with widespread applications from low level image segmentation to high level object recognition. Gaussian Markov random field (GMRF) is a particular model based texture feature extraction scheme which uses model parameters as texture features. In this thesis a novel robust texture descriptor based on GMRF is proposed specially for texture segmentation and classification. For these tasks, descriptive features are more favourable relative to the generative features. Therefore, in order to achieve more descriptive features, with the GMRFs, a localized parameter estimation technique is introduced here. The issues arising in the localized parameter estimation process, due to the associated small sample size, are addressed by applying Tikhonov regularization and an estimation window size selection criterion. The localized parameter estimation process proposed here can overcome the problem of parameter smoothing that occurs in traditional GMRF parameter estimation. Such a parameter smoothing disregards some important structural and statistical information for texture discrimination. The normalized distributions of local parameter estimates are proposed as the new texture features and are named as Local Parameter Histogram (LPH) descriptors. Two new rotation invariant texture descriptors based on LPH features are also introduced, namely Rotation Invariant LPH (RI-LPH) and Isotropic LPH (I-LPH)descriptors. The segmentation and classification results on large texture datasets demonstrate that these descriptors significantly improve the performance of traditional GMRF features and also demonstrate better performance in comparison with the state-of-the-art texture descriptors. Satisfactory natural image segmentation is also carried out based on the novel features. Furthermore, proposed features are employed in a real world medical application involving tissue recognition for emphysema, a critical lung disease causing lung tissue destruction. Features extracted from High Resolution Computed Tomography (HRCT) data are used in effective tissue recognition and pathology distribution diagnosis. Moreover, preliminary work on a Bayesian framework for integrating prior knowledge into the parameter estimation process is undertaken to introduce further improved texture features.

In the last few years, the explosion of multimedia content on the Web has made multimedia resources the “first class citizen” of the Web. While these resources are easily stored and shared, it is becoming more difficult to find specific video/audio content, especially to identify, link, navigate, search and share the content inside multimedia resources. The concept of media fragment refers to the deep linking into multimedia resources, but making annotations to media fragments and linking them to other resources on the Web have yet to be adopted. The Linked Data principles offer guidelines for publishing Linked Data on the Web, so that data can be better connected to each other and explored by machines. Publishing media fragments and annotations as Linked Data will enable the media fragments to be transparently integrated into current Web content. This thesis takes the Linked Data approach to realise the interlinking of media fragments to other resources on the Web and demonstrate how the Linked Data can help improve the indexing of media fragments. This thesis firstly identifies the gap between media fragments and Linked Data, and the major requirements that need to be fulfilled to bridge that gap based on the current situation of presenting and sharing multimedia data on the Web. Then, by extending the Linked Data principles, this thesis proposes Interlinking Media Fragment Principles as the basic rationale and best practice of applying Linked Data principles to media fragments. To further automate the media fragments publishing process, a core RDF model and a media fragment enriching framework are designed to link media fragments into the Linked Open Data Cloud via annotations and visualise media fragments on the Web pages. A couple of examples are implemented to demonstrate the use of interlinked media fragments, including the case to enrich YouTube videos with named entities and using media fragments for video classifications. The Media Fragment Indexing Framework is proposed to solve the fundamental problem of media fragments indexing for search engines and, as an example, Twitter is adopted as the source for media fragment annotations. The thesis concludes that applying Linked Data principles to media fragments will bring semantics to media fragments, which will improve the multimedia indexing on a fine-grained level and new research areas can be explored based on the interlinked media fragments.

Advances in silicon technology over the last decade have led to increased integration of analogue and digital functional blocks onto the same single chip. In such a mixed signal environment, the analogue circuits must use the same process technology as their digital neighbours. With reducing transistor sizes, the impact of process variations on analogue design has become prominent and can lead to circuit performance falling below specification and hence reducing the yield. This thesis explores the methodology and algorithms for an analogue integrated circuit automation tool that optimizes performance and yield. The trade-offs between performance and yield are analysed using a combination of an evolutionary algorithm and Monte Carlo simulation. Through the integration of yield parameter into the optimisation process, the trade off between the performance functions can be better treated that able to produce a higher yield. The results obtained from the performance and variation exploration are modelled behaviourally using a Verilog-A language. The model has been verified with transistor level simulation and a silicon prototype. For a large analogue system, the circuit is commonly broken down into its constituent sub-blocks, a process known as hierarchical design. The use of hierarchical-based design and optimisation simplifies the design task and accelerates the design flow by encouraging design reuse. A new approach for system level yield optimisation using a hierarchical-based design is proposed and developed. The approach combines Multi-Objective Bottom Up (MUBU) modelling technique to model the circuit performance and variation and Top Down Constraint Design (TDCD) technique for the complete system level design. The proposed method has been used to design a 7th order low pass filter and a charge pump phase locked loop system. The results have been verified with transistor level simulations and suggest that an accurate system level performance and yield prediction can be achieved with the proposed methodology.

Scanning near-field optical microscopy (SNOM) is a system that can image beyond the conventional diffraction limit. It does this by collecting the information contained within evanescent fields. This unique ability to image using evanescent fields also enables SNOM to directly measure the electric field distribution in waveguides, where light is guided by total internal reflection. When SNOM is used with a spectrally resolving detector, local temporal phenomena can be detected by analysing spectral interference in the spectra collected by the probe. This spectrally resolving configuration was used to directly measure inter-modal group velocity difference in a multimode ridge waveguide and, using the modes’ spatial profiles to experimentally determine the mode amplitude coefficient ratio. Such an ability to provide measurements on the local dispersion characteristics and relative modal amplitudes of guided light establishes SNOM as a route for investigating the conversion of current single mode photonic devices into multimode devices. The spectrally resolving SNOM system was also used to investigate the sources of temporal delays created by a quasi disordered scattering sample, which was based on John H. Conway’s pinwheel tiling. Whilst the measurements do not create a complete picture of the scattering phenomena in this work, suggestions for improvement are offered with the aim establishing spectrally resolving SNOM systems as tools for mapping localised temporal phenomena in disordered scattering systems.

There are currently available many approaches aimed at tracking objects moving in sequences of images. These approaches can suffer in occlusion and noise, and often require initialisation. These factors can be handled by techniques that extract objects from image sequences, especially when phrased in terms of evidence gathering. As yet, the newer approaches to arbitrary shape extraction avoid discretisation affects but do not include motion. The moving-object evidence gathering approach has yet to include arbitrary shapes and can require high order description for complex motions. Since the template approach is proven for arbitrary shapes, we re-deploy it for moving arbitrary shapes, but in a way aimed to avoid discretisation problems. As the template approach has already been seen to reduce computational demand in the extraction of arbitrary shapes, we further deploy it to describe the motion of moving arbitrary shapes. As with the shape templates, we use Fourier descriptors for the motion templates, yielding an integrated framework for the representation of shape and motion. This prior specification of motion avoids the need to use an expensive parametric model to capture data that is already known. Furthermore, as the complexity of motion increases, a parametric model would require increasingly more parameters, leading to a rapid and catastrophic increase in computational requirements, whilst the cost and complexity of the motion template model is unchanged. The new approach combining moving arbitrary shape description with motion templates permits us to achieve the objective of low dimensionality extraction of arbitrarily moving arbitrary shapes with performance advantage as reflected by the results this new technique can achieve.

Wireless sensor networks (WSNs) have a diverse range of industrial, scientific and medical applications where the sensor nodes are of low cost, standard with respect to hardware architecture, processing abilities and communicate using low-power narrow-band radios. Position information of the sensing nodes within those applications is often a requirement in order to make use of the data recorded by the sensors themselves. On deployment, sensing nodes normally have no prior knowledge of their position and thus a localization mechanism is often a requirement. The process of localizing a 'blind' device consists of ranging estimates or angle measurements to a set of references with a prior knowledge of their position relative to a co-ordinate system and the position computation of the blind device in relation to the fixed references. This research focuses on the process of ranging to enable two-dimensional localization of sensing nodes within WSNs. Alternative ranging methods for the specified application field have not demonstrated their ability to meet the resolution and accuracy (resolution 0.3 m with accuracy better than ± 1.0 m line-of-sight) required. A novel radio frequency (RF) time-of-flight (TOF) ranging system is presented in this work to mitigate those problems. The system has been prototyped using a TI CC2431 development platform with ranging and data packet transfer performed on a single channel in the 2.4 GHz ISM frequency band. The frequency difference between the two transceivers involved with ranging is used to obtain sub-clock TOF phase offset measurement in order to achieve high resolution TOF measurements. Performance results have been obtained for the line-of-sight (LOS), non-line-of-sight (NLOS) and indoor conditions. Accuracy is typically better than 7.0m RMS for the LOS condition over 250.0m and 15.8m RMS for the NLOS condition over 120.0m using a sample average of one-hundred two-way ranging transactions. Indoors accuracy is measured to 1.7m RMS using a 1000 sample average over 8.0m. Corresponding results are also presented for the algorithms suitability for localizing sensor nodes in two-dimensions. Ranging performance is bound by the signal-to-noise ratio (SNR), signal bandwidth, synchronization and frequency difference between devices. This ranging algorithm demonstrates a novel method where resolution and accuracy are improved time dependent in comparison to frequency dependent methods using narrow-band RF.

This thesis presents my investigation into in-band pumped erbium doped fibre amplifiers (EDFAs) and their performance under high power continuous wave (cw) operation and high energy low repetition rate pulsed operation. In addition, Q-switched erbium doped fibre lasers were investigated and used as the seed laser for a high energy low repetition rate EDFA system. Furthermore, the power scaling of all-fibre frequency doubled fibre lasers based on periodically poled silica fibre (PPSF) was also investigated. In Q-switched fibre lasers, the multiple-peak phenomenon (MPP) is an undesirable effect in which the Q-switched pulse develops sub-structure or even breaks into multiple sub pulses. I demonstrated that the MPP can be eliminated by increasing the acousto-optic modulator rise time. An experimentally validated numerical model was also used to explain the origin of MPP. Next, I showed that the interplay between MPP and modulation instability (MI) changes the detail of the spectral evolution of the Q-switched pulses. The in-band EDFAs were investigated using 1535 nm pump fibre lasers. For cw operation, a highly efficient (~ 80%), high power (18.45 W) in-band, core pumped erbium/ytterbium co-doped fibre laser was demonstrated. Using a fitted simulation model, I showed that the significantly sub-quantum limit conversion efficiency of in-band pumped EDFAs observed experimentally can be explained by concentration quenching. I then numerically studied and experimentally validated the optimum pumping configuration for power scaling of in-band, cladding pumped EDFAs. My simulation results indicate that a ~ 77% power conversion efficiency with high output power should be possible through cladding pumping of current commercially available pure erbium doped active fibres providing the loss experienced by the cladding guided 1535 nm pump due to the coating absorption can be reduced to an acceptable level by better coating material choice. The power conversion efficiency has the potential to exceed 90% if concentration quenching of erbium ions can be reduced via improvements in fibre design and fabrication. For low repetition rate pulsed operation, I demonstrated and compared high-energy, in-band pumped EDFAs operating at 1562.5 nm under both a core pumping scheme (CRS) and a cladding pumping scheme (CLS). The CRS/CLS sources generated smooth, single-peak pulses with maximum pulse energies of ~1.53/1.50 mJ, and corresponding pulse widths of ~176/182 ns respectively, with an M^2 of ~1.6 in both cases. However, the conversion efficiency for the CLS was >1.5 times higher than the equivalent CRS variant operating at the same pulse energy due to the lower pump intensity in the CLS that mitigates the detrimental effects of concentration quenching. With a longer fibre length in a CLS implementation a pulse energy of ~2.6 mJ was demonstrated with a corresponding M^2 of ~4.2. Using numerical simulations I explained that the saturation of pulse energy observed in my experiments was due to saturation of the pump absorption. For the frequency doubling work, the fundamental pump source of the PPSF was a master oscillator power amplifier seeded with a tuneable external cavity laser. During the high power operation, the heat deposition along the PPSF shifted the optimal quasi-phase matched wavelength to a longer wavelength. This shift must be compensated to achieve optimal performance of the PPSF under test and was achieved in my experiment by tuning the central wavelength of the pump source. At the end of the high power experiment, the PPSF samples degraded to ~40% of their pristine PPSF normalised efficiencies. The glass property of the PPSF had also been changed by the high power exposure. A high power all-fibre frequency doubled laser was demonstrated with 1.13 W of second harmonic average power with ~27% internal conversion efficiency.

The main motivation of this thesis is to design low-complexity high efficiency noncoherent fusion rules for the parallel triple-layer wireless sensor networks (WSNs) based on frequency-hopping Mary frequency shift keying (FH/MFSK) techniques, which are hence referred to as the FH/MFSK WSNs. The FH/MFSKWSNs may be employed to monitor single or multiple source events (SEs)with each SE having multiple states. In the FH/MFSKWSNs, local decisions made by local sensor nodes (LSNs) are transmitted to a fusion center (FC) with the aid of FH/MFSK techniques. At the FC, various noncoherent fusion rules may be suggested for final detection (classification) of the SEs’ states. Specifically, in the context of the FH/MFSK WSNs monitoring single M-ary SE, three noncoherent fusion rules are considered for fusion detection, which include the benchmark equal gain combining (EGC), and the proposed erasure-supported EGC (ES-EGC) as well as the optimum posterior fusion rules. Our studies demonstrate that the ES-EGC fusion rule may significantly outperform the EGC fusion rule, in the cases when the LSNs’ detection is unreliable and when the channel signal-to-noise ratio (SNR) is relative high. For the FH/MFSKWSNs monitoring multiple SEs, six noncoherent fusion rules are investigated, which include the EGC, ES-EGC, EGC assisted N-order IIC (EGC-NIIC), ES-EGC assisted N-order IIC (ES-EGC-NIIC), EGC assisted r-order IIC (EGC-rIIC) and the ES-EGC assisted r-order IIC (ES-EGC-rIIC). The complexity, characteristics as well as detection performance of these fusion rules are investigated. Our studies show that the ES-EGC related fusion rules are highly efficient fusion rules, which have similar complexity as the corresponding EGC related fusion rules, but usually achieve better detection performance than the EGC related fusion rules. Although the ES-EGC is a single-user fusion rule, it is however capable of mitigating the multiple event interference (MEI) generated by multiple SEs. Furthermore, in some of the considered fusion rules, the embedded parameters may be optimized for the FH/MFSK WSNs to achieve the best detection performance. As soft-sensing is often more reliable than hard-sensing, in this thesis, the FH/MFSK WSNs with the LSNs using soft-sensing are investigated associated with the EGC and ES-EGC fusion rules. Our studies reveal that the ES-EGC becomes highly efficient, when the sensing at LSNs is not very reliable. Furthermore, as one of the applications, our FH/MFSK WSN is applied for cognitive spectrum sensing of a primary radio (PR) system constituted by the interleaved frequencydivision multiple access (IFDMA) scheme, which supports multiple uplink users. Associated with our cognitive spectrum sensing system, three types of energy detection based sensing schemes are addressed, and four synchronization scenarios are considered to embrace the synchronization between the received PR IFDMA signals and the sampling operations at cognitive spectrum sensing nodes (CRSNs). The performance of the FH/MFSK WSN assisted spectrum sensing system with EGC or ES-EGC fusion rule is investigated. Our studies show that the proposed spectrum sensing system constitutes one highly reliable spectrum sensing scheme, which is capable of exploiting the space diversity provided by CRSNs and the frequency diversity provided by the IFDMA systems. Finally, the thesis summarises our discoveries and provides discussion on the possible future research issues.

The brain holds key information regarding the information processing capability of individuals and recent advances in sensor devices and technology have attracted researchers to question the working of this complex organ. It is not only the elusiveness of the brain that has drawn recent research attention but also the claim of doctors that brain function is key in neurological disorders. Disorders like Autism and Attention Deficit Hyperactivity Disorder (ADHD) not to mention other forms of neurobiological diseases have been attributed to disproportionate and disrupted connectivity in the brain. It is envisaged that more accurate and thorough understanding such connectivity can pave the way for medical research of diseases such as these which are deeply rooted to neural level information exchange deficits. The main objective of this work is to develop an effective means to quantitatively characterise functional connectivity in the brain. Phase synchronisation is reported as the key manifestation of the underlying mechanism of information coupling between different brain regions. This work, therefore first the phase relationships between Electroencephalogram (EEG) signals have been investigated to understand the synchronisation pattern underlying them during the execution of a task. The pursuit to characterise time evolving phase synchrony leads to the identification of the existence of discrete states with quasi-stable phase topography call synchrostates in EEG datasets from range of subjects. These states exhibited switching patterns which were characteristic to the stimuli provided during a cognitive task, specifically in this case face perception tasks. The switching of these states were modelled in a probabilistic framework using a finite Markov model and the stability of the states are represented by the self-transition probabilities. The degree of phase synchronisation during the existence of each state is then translated into functional connectivity maps and complex network graph measures were applied on it to obtain a set of metrics that quantify the characteristics of such connections formed within the brain. These quantitative brain connectivity measures were used as features to solve a classification problem between autistic and typical children which resulted in an accuracy of 94.7%. The connectivity parameters were then used to characterise behavioural trait scores of anxious children by developing a regression model correlating these to the standardised behavioural scores calculated from questionnaires. Traits like sadness, state anxiety and anger could be modelled effectively using the metrics reported in this study. This work lays the foundation for further exploration of these quantitative measures for characterising a variety of neurodegenerative diseases and hence may result in a new type of diagnostic process to aid the existing tools available to the clinicians.

The technical breakthrough of Turbo Codes (TCs) initiated two decades of exciting developments leading to a suite of near-capacity techniques. It has been widely recognized that exchanging extrinsic information between the channel decoders and the modulated signal detectors assists communications systems in approaching their best possible performance potential that is predicted by the channel capacity. Nonetheless, in line with Moor’s Law, as researchers inch closer and closer to the channel capacity, the complexity of the resultant communications systems is also significantly increased. In fact, soft-decision-aided signal detection conceived for Single-Input Single-Output (SISO), Single-Input Multiple-Output (SIMO) and Multiple-Input Multiple-Output (MIMO) schemes typically contribute a substantial fraction of the total complexity, especially when multiple received samples have to be jointly detected in order to combat the deleterious effect of channel fading. Against this background, in this treatise, we firstly propose a reduced-complexity design for the classic soft-decision-aided PSK/QAM detectors, and then these reduced-complexity design guidelines are applied to a variety of communications systems spanning from coherent to non coherent, from uncoded to coded, and also from SISO to MIMO systems. Our aim is to reduce the computational complexity as much as possible, especially for complex near-capacity communications systems, while mitigating any performance loss imposed by our reduced-complexity design. First of all, we commence from the family of basic coherent SISO/SIMO systems, where both uncoded and coded PSK/QAM schemes are considered. The channel coding assisted near capacity systems design principles are introduced based on EXtrinsic Information Transfer (EXIT) charts. Furthermore, we observe that the Max-Log-MAP algorithm invoked for soft-decision-aided PSK/QAM detection aims for finding the maximum probabilities, which is similar to the action of hard-decision-aided detection of uncoded MPSK/QAM schemes. Therefore, we propose to link each a priori LLR to a reduced-size fraction of the channel’s output signal constellations, so that the Max-Log-MAP algorithm may be operated at a reduced complexity. Moreover, the corresponding reduced-complexity Approx-Log-MAP algorithm is also conceived by compensating for the Max-Log-MAP algorithm’s widely-used Jacobian approximation relying on a lookup table. Our performance results demonstrate that up to 41.6% and 72.6% complexity reductions are attained for soft-decision-aided Square 64QAM and Star 64QAM detectors, respectively, which is achieved without any performance loss. This complexity reduction is substantial, especially when the soft decision-aided signal detectors are invoked several times during turbo detection. Secondly, we proceed by conceiving reduced-complexity algorithms for the non coherently detected DPSK schemes in both uncoded and coded SISO/SIMO systems. More explicitly, the DPSK transmitter modulates the data-carrying symbols onto the phase changes between consecutive transmitted symbols, so that the Conventional Differential Detection (CDD) may recover the source information by observing the phase change between every pair of consecutive received samples. However, the CDD aided DPSK suffers from a 3 dB performance penalty compared to its coherent counterpart. Moreover, an irreducible error floor occurs, when the CDD is employed in rapidly fluctuating fading channels. In order to mitigate this problem, Multiple-Symbol Differential Detection (MSDD) may be invoked in order to improve the DPSK performance by extending the observation window length from the CDD’s Nw = 2 to Nw ≥ 2. The price paid is that the MSDD complexity grows exponentially with (Nw − 1) as a result of jointly detecting the (Nw − 1) data-carrying symbols. As a remedy, the Decision-Feedback Differential Detection (DFDD) concept may be introduced in order to detect a single symbol based on previous decisions concerning the (Nw − 2) data-carrying symbols in a MSDD window. However, the DFDD inevitably imposes a performance loss due to its inherent error propagation problem. In order to retain the optimal MSDD performance, the Multiple-Symbol Differential Sphere Detection (MSDSD) facilitates the MSDD by invoking a Sphere Decoder (SD). Against this background, we firstly propose to introduce a simple correlation operation into the hard-decision-aided MSDSD employing an arbitary number of Receive Antennas (RAs), so that the SD may visit the constellation points in a zigzag fashion for the case of uncoded DPSK SIMO systems. Furthermore, we propose a reduced-complexity Schnorr-Euchner search strategy for the soft-decision MSDSD employing an arbitrary number of RAs, so that the optimum candidate may be found by visiting a reduced-size subset of constellation points, and then the rest of the constellation points may be visited in a zig-zag fashion. Our simulation results demonstrate that up to 88.7% complexity reduction is attained for MSDSD (Nw = 4) aided D16PSK. We have also proposed the near-optimum Approx-Log-MAP algorithm conceived for soft-decision-aided SD, which has not been disseminated in the open literature at the time of writing. Furthermore, the important subject of coherent versus non coherent detection is discussed in the context of coded systems, which suggests that MSDSD aided DPSK is an eminently suitable candidate for turbo detection assisted coded systems operating at high Doppler frequencies. Following this, a range of non coherent detectors designed for non-constant modulus Differential QAM (DQAM) schemes are introduced for both uncoded and coded scenarios, where the open problem of MSDSD aided Differential QAM (DQAM) is solved. More explicitly, the MSDSD relies on the knowledge of channel correlation, which is determined both by the Doppler frequency and by the noise power. For DPSK, the transmitter’s phases may form a unitary matrix, which may be separated from the channel’s correlation matrix, so that a lower triangular matrix that is created by decomposion from the inverse of the channel’s correlation matrix may be utilized in the context of sphere decoding. However, for DQAM, the transmitted symbol-amplitudes cannot form a unitary matrix, which implies that they have to be taken into account by the channel’s correlation matrix. As a result, the symbol-amplitude-dependent channel correlation matrix only becomes known, when all the symbol-amplitudes are detected. Furthermore, the classic DFDD solutions conceived for DQAMrely on the assumption of the channel’s correlation matrix being independent of the symbol-amplitudes, which implies that these DFDD solutions are sub-optimal and they are not equivalent to the decision-feedback aided version of the optimum MSDD. To circumvent these problems, we prove that although the complete channel correlation matrix remains unknown, the associated partial channel correlation matrix may be evaluated with the aid of the SD’s previous decisions as well as by relying on a single information-dependent symbol amplitude that may be readily found by the SD. As a benefit, we are able to invoke sphere decoding for both amplitude detection and phase detection in the context of MSDD aided DQAM. Furthermore, we have also improved the classic DFDD solutions conceived for DQAMby directly deriving them from the optimum MSDD. Moreover, we offer a unified treatment of diverse non coherent detectors, including CDD,MSDD,MSDSD and DFDD for a variety of DQAM constellations that exist in the literature, including Differential Amplitude Phase Shift Keying (DAPSK), Absolute-Amplitude Differential Phase Shift Keying (ADPSK) and their twisted constellations. The reduced-complexity algorithms proposed for DPSK detection are also applied to DQAM detection in both uncoded and coded systems.