Alarm schemes for the elderly that are currently available are expensive. The vast majority of these schemes use a portable trigger worn by the client to transmit to a receiver unit within the client's home. This receiver unit then calls for help through the public telephone network. A large proportion of the overall cost of this type of scheme is the cost of providing each client with a receiver unit and a modem. In this thesis we look at the possibility of transmitting a high frequency alarm signal directly to a service centre from a portable transmitter worn by the client. This method should be much cheaper than a conventional telephone based system, as a receiver and modem for each client is not required. To obtain the required range, we sacrifice the speech link between the client and the service centre and use spread spectrum techniques to reduce the equivalent noise bandwidth of the system. There is a channel reserved exclusively for elderly alarm schemes, this channel in the high frequency radio band. In the high frequency radio band, the background noise is often dominated by man-made components and is therefore impulsive. The distribution of the impulsive noise in these channels is non-stationary and unknown to the system designer. An analysis of the performance of a direct sequence spread spectrum system using matched filter reception is developed. This analysis concentrates on providing a guaranteed minimum performance from a limited precision digital matched filter, regardless of the noise distribution. The performance of a non-linear clipping device preceding the matched filter is analysed and is shown to improve performance in impulsive noise. Finally, we study the design of a practical alarm scheme based on direct sequence spread spectrum.

Continuous-time (CT) SD modulators are growing increasingly popular in wideband A/D conversion. Such applications require high orders of quantisation noise shaping and multi-bit quantisers, to compensate for the resulting low oversampling ratios. These, however, add circuit complexities and excess loop delay that are detrimental to the modulator control loop. A CT SD modulator can potentially achieve a higher performance with a lower bandwidth requirement than its discrete-time (DT) counterpart. Unfortunately, CT SD modulators suffer from problems not seen in dependent jitter and non-rectangular DAC pulse shapes. Low-oversampling CT SD modulators are also sensitive to a low bandwidth in the input stage of the forward filter. These effects are often tackled by a modification ion the CT SD modulator architecture. Unfortunately, the resulting CT SD modulator control loop is often suboptimal and sensitive to changes in the modulator coefficients. In this thesis, a design-by-optimisation method is proposed to find the optimum that satisfies the constraints set by the implementation of the CT SD modulator, its feasibility and any other additional design criteria. Robustness in the final design is ensured by optimising directly on the coefficients of the CT SD modulator, and evaluating the stability and the performance of its DT equivalent model. The DT equivalent is computed using a numerical implementation of the impulse-invariance technique. This implementation can deal with both excess loop delay and non-rectangular DAC pulse shapes, numerically and in general form. This is, to the best of the author’s knowledge, the first implementation of its kind. It is shown that for a low value of loop delay, a resonator based forward filter is optimum, while, contrary to assumptions made in the literature, real poles in the forward filter are preferable for moderate values of loop delay. It is hoped that the bandwidth requirement of CT modulators can be minimised in a similar manner. A numerical curve fitting approach is also presented that models many of the nonidealities in an integrator circuit response with a low-order model transfer function.

This thesis studies the problem of soft-output equalization of the mobile radio channel in interference-limited environments, where it is often difficult to obtain an accurate statistical model of the (non-Gaussian) disturbance. The first part of the thesis proposes a new technique for single-channel MAP trellis equalization in the presence of multipath and non-Gaussian interference. The approach is based on the <i>non-parametric </i>estimation of the density function of the overall disturbance by means of <i>kernel smoothing. </i>The work considers the problem of density estimation with limited volume of data, and addresses the use of a whitening filter in the presence of coloured interference. As an application, simulation results are provided for the GSM system, showing a significant performance improvement with respect to the trellis equalizer based on the Gaussian assumption. The second part of the thesis considers the case of an antenna array receiver, and studies a simple method to derive the reliability information at the output of a deterministic decision-feedback <i>least-squares </i>space-time equalizer. Computer simulations for the Enhanced Data Rates for GSM Evolution (EDGE)/Enhanced General Packet Radio Service (EGPRS) system show that the receiver performance can be significantly improved by a soft-output calculation based on short-term statistics of the equalizer output error. The thesis also addresses the additional use of soft-decision feedback, which provides further robustness to the proposed soft-output equalizer. The study shows the relevance of probabilistic processing for robust equalization of the wireless channel in the presence of non-Gaussian interference, and emphasizes the advantages of strategies that do not rely on a statistical model of the disturbance.

Recent years have seen the rise to prominence of a powerful new computational paradigm - the so-called artificial neural network. Loosely based on the microstructure of the central nervous system, neural networks are massively parallel arrangements of simple processing elements (<i>neurons</i>) which communicate with each other through variable strength connections (<i>synapses</i>). The simplicity of such a description belies the complexity of calculations which neural networks are able to perform. Allied to this, the emergent properties of noise resistance, fault tolerance, and large data bandwidths (all arising from the parallel architecture) mean that neural networks, when appropriately implemented, represent a powerful tool for solving many problems which require the processing of real-world data. A computer vision task (viz. the classification of regions in images of segmented natural scenes) is presented, as a problem in which large numbers of data need to be processed quickly and accurately, whilst, in certain circumstances, being disambiguated. Of the classifiers tried, the neural network (a multi-layer perceptron) was found to provide the best overall solution, to the task of distinguishing between regions which were 'roads', and those which were 'not roads'. In order that best use might be made of the parallel processing abilities of neural networks, a variety of special purpose hardware implementations are discussed, before two different analogue VLSI designs are presented, complete with characterisation and test results. The latter of these chips (the EPSILON device) is used as the basis for a practical neuro-computing system. The results of experimentation with different applications are presented. Comparisons with computer simulations demonstrate the accuracy of the chips, and their ability to support learning algorithms, thereby proving the viability of the use of pulsed analogue VLSI techniques for the implementation of artificial neural networks.

Nonstationary time series arise in many different disciplines, and there are many different reasons for wishing to study them. The particular interest in this thesis is in modelling the time series so as to obtain certain parameters of interest from it. Whatever the reason for studying such a time series and whatever the method chosen, in order to accommodate the nonstationarity of the series it is important to use an adaptive algorithm whose parameters are permitted to vary with time. The first aim of this thesis will be to examine existing adaptive algorithms, highlighting their strengths and weaknesses to determine which, if any, offers the best way forward towards developing new algorithms. Following this, rather than consider a specific class of algorithm a generic algorithm which contains the properties of more than one class of algorithm will be examined. To facilitate the development of this algorithm hyperparameters and hypermodels will be introduced. Results of simulations run to test the algorithms performance will be given. The second aim of this thesis will be to develop a new algorithm, the fast adaptive forward backward least squares algorithm. This algorithm incorporates a 'forgetting factor' to enable the tracking of nonstationary signals. Simulations will be performed which show that the algorithm can outperform the unwindowed version in the presence of a nonstationary signal. Stabilization techniques will be introduced which will prevent the algorithm exhibiting numerical instabilities to which this type of algorithm are prone. Simulations results will be presented to give guidelines for the choice of values of feedback gains which are to be used to prevent the exhibition of instability. Finally the advantages and limitations of both the new and existing algorithms will be summarized and suggested areas of future research outlined.

From the introduction of wireless communications in the 1980s, the wireless communication market has grown explosively and produced demand for more capacity. Since the direct sequence code division multiple access (DS-CDMA) technique has more capacity than the current time division multiple access technique, it has been adopted for third generation mobile communications systems and it serves as the platform for the work in this thesis. At the downlink of the DS-CDMA systems, the multiple access interference (MAI) will be introduced due to the non-zero cross correlation between the spreading codes with arbitrary time shifts. Thus, the system's capability is limited by the amount of interference instead of the background noise. To achieve full capacity utilization, it is crucial to suppress the interference in the system. Linear multiuser detection (MUD) and symbol level equalisers are two common methods to suppress the MAI at the downlink. However, these methods have considerable computational complexity and more importantly, they are not suitable for most cases of the downlink where the long scrambling codes are used. Recently, chip level equalisers have been proposed to partially restore the orthogonality of the spreading codes by inverting the channel transfer function prior to the despreading. The MAI can then be suppressed by the conventional correlator after the chip level equalisers. Due to the time varying channel in the downlink, adaptive implementation should be adopted at the mobile terminal. The length of the adaptive filter is an important factor which affects all the aspects of its performance measures such as the convergence rate, computational complexity and MSE (mean square error) performance. However, till now, little research work related to this important parameter has been done. In this thesis, the relation between the MSE performance and the length of the adaptive filters based on several different adaptive algorithms is firstly given. Then the influence of the value of the threshold parameters on the performance of dynamic length algorithm is analyzed. Based on this analysis, a new type of dynamic length algorithms is proposed and its implementation issues both under the static channel environment and the time varying channel environment are also considered. Then, the relationship between the MSE performance and the tracking performance with the length of the chip level equalisers is given for both the chip rate implementation and symbol rate implementation scenario. The dynamic length algorithm proposed is used at the downlink of CDMA systems. Both the chip rate implementation and symbol rate implementation are considered. The performance of this new type of the chip level equalisers outperforms the corresponding fixed length chip level equalisers with a marginal increase in computational complexity.

Reconfiguration in an Integrated Circuit (IC) design has become increasingly important in the recent years. Some of the driving factors behind this trend are reduction in transistor size, ever changing standards, very high IC mask costs and short time to market. The programmable hardware design however suffers from performance degradation due to the added flexibility contrary to the end user demand for very high speed and low power electronics. Domain specific reconfigurable architectures provide a powerful solution to the problem by carefully tailoring the domain of the reconfiguration for the increased performance. This research work focused on investigating such low power reconfigurable VLSI architectures for forward error correction (FEC) to be deployed in a unified communication platform. The viterbi and turbo decoding are very well known techniques for FEC decoding and are essential components in many current and up coming standards such as WCDMA, WLAN, GSM, CDMA2000, ADSL and 3GPP. This thesis presents a reconfigurable unified implementation with a unified state machine control for combined turbo-viterbi decoder array. The amount of flexibility in the reconfigurable design is carefully tailored to meet the performance constraints imposed by these standards. Work on reconfigurable viterbi decoder provided the new novel reconfigurable trace back methodology, new segmentation and memory management techniques along with an open trellis structure that can support multiple standards. The work on reconfigurable turbo array generated novel implementation technique for low power input metrics management and reconfiguration, low power branch metrics generation, a new matrix normalization scheme and a completely flexible open trellis low power reconfigurable design. Turbo decoder design is combined with a novel low power implementation methodology for 3GPP internal interleaver. The interleaver implementation gives significant reduction in storage requirement for interleaved patterns and hence much improved power performance.

An adaptive array is a signal processor used in conjunction with a set of antennae to provide a versatile form of spatial filtering. The processor combines spatial samples of a propagating field with a variable set of weights, typically chosen to reject interfering signals and noise. In radar, the spatial filtering capability of the array facilitates cancellation of hostile jamming signals and aids in the suppression of clutter. In many applications, the practical usefulness of an adaptive array is limited by the complexity associated with computing the adaptive weights. In a partially adaptive beamformer only a subset of the available degrees of freedom are used adaptively, where adaptive degree of freedom denotes the number of unconstrained or free weights that must be computed. The principal benefits associated with reducing the number of adaptive degrees of freedom are reduced computational burden and improved adaptive convergence rate. The computational cost of adaptive algorithms is generally either directly proportional to the number of adaptive weights or to the square or cube of the number of adaptive weights. In radar it is often mandatory that the number of adaptive weights be reduced with large antenna arrays because of the algorithms computational requirement. The number of data vectors needed for the adaptive weights to converge to their optimal values is also proportional to the number of adaptive weights. Thus, in some applications, adaptive response requirements dictate reductions in the number of adaptive weights. Both of these aspects are investigated in this thesis. The primary disadvantage of reducing the number of adaptive weights is a degradation in the steady-state interference cancellation capability. This degradation is a function of which adaptive degrees of freedom are utilised and is the motivation for the partially adaptive design techniques detailed in this thesis. A new technique for selecting adaptive degrees of freedom is proposed. This algorithm sequentially selects adaptive weights based on an output mean square error criterion. It is demonstrated through simulation that for a given partially adaptive dimension this approach leads to improved steady-state performance, in mean square error terms, over popular eigenstructure approaches. Additionally, the adaptive structure which results from this design method is computationally efficient, yielding a reduction of around 80% in the number of both complex multiplications and additions.

This thesis describes the research carried out to assess the applicability of speech waveform perturbation analysis to the problem of automatic speaker verification. It also describes the development of the technique into an operational system. The techniques of waveform perturbation analysis have been studied in the past with a view to their use as potential indicators of vocal fold dysfunction. In essence, they quantify the natural cycle-to-cycle fluctuations of vocal fold vibrations, also known as pitch periods, and are thus related to the perceived hoarseness and roughness of the voice. A major aim of this thesis is to determine if such features offer a viable dimension in speaker space along which discrimination can take place. The field of speaker verification is reviewed, and a number of previously exploited techniques are described. The likely mechanism for the production of speech waveform perturbation, separated into 'jitter' of period durations and 'shimmer' of peak amplitude values, are examined, and the existing techniques of quantification reviewed. In order to ensure accurate cycle-to-cycle measurement of period durations in real time, a new pitch determination algorithm, based on multi-feature investigation of the waveform peaks, has been developed. Its accuracy was assessed using both previously used pitch determination algorithms, and a laryngograph device. It was found to offer high accuracy pitch synchronous period estimates. The quantification of perturbation was carried out using the technique of median smoothing. This technique approximates removal of the more gradual changes in pitch period due to intonation. Measures of residual irregularity are then combined with additional long term intonational measures, to form a 10-dimensional profile of the speaker. Using an all male population of 72 speakers, a verification accuracy of 87% was achieved. The system was trained on approximately 60 seconds of continuous speech (from each speaker), and tested on 10 second utterances. A number of approaches to allow accurate classifiction of talkers are discussed, and their relative merits investigated experimentally. The effects of time spacing the training and testing data are studied. Also, the effectiveness of extending the speaker profile to include long term, spectrum combinations of features. The techniques of feature selection, used to limit the effects of finite training data, are also explored and extended. Results are presented for a study which employed professionally trained mimics in order to assess the effectiveness of the system under the most stringent conditions. The distribution of error rates within a population (i.e. the existence of particularly inconsistent speakers, or 'goats') is also studied, with a view to minimising their detrimental effects on system efficacy. Finally, this thesis describes the translation of the above system into an operational near real-time system, employing a digital signal processing microprocessor device.

In this thesis receiver architectures for an unknown, time-varying Rayleigh fading channel are investigated. This includes fast fading scenarios, where the channel impulse response (CIR) can change significantly between two adjacent samples. Channel estimation based on the minimum mean squared error criterion (MMSE) applied to smoothing and linear prediction is considered. One of the key objectives in this thesis is the analysis of error propagation effects due to decision feedback. The studied receiver architectures are divided into two main parts: one-shot receivers which detect the received symbol on a symbol-by-symbol basis, and sequence detectors which jointly estimate and detect the entire received signal sequence. Considering one-shot receivers, a decision directed receiver is studied using differential modulation (DPSK). The receiver can significantly improve the fast fading performance of conventional DPSK, through linear predictive channel estimation. It is demonstrated through simulation that the performance of the decision directed receiver is better than that of an idealised reference receiver where channel estimation is not corrupted by decision feedback errors (e.g. by means of employing a pilot signal). Furthermore, a receiver employing coherent modulation is considered. The necessary phase reference is provided by time multiplexed pilot symbols. A receiver which exclusively uses these pilot symbols for channel estimation is the pilot aided receiver. The performance for slow fading is excellent, whereas the performance degrades as the Doppler frequency increases. The degradation is proportional to the spacing of the pilots. The performance of both the decision directed and the pilot aided receiver can be significantly improved by employing a second stage channel estimation filter, using a smoothing type estimation filter.