The communications phenomena at the end of the 20th century were the Internet and mobile telephony. Now, entering the new millennium, an effective combination of the two should become a similarly everyday experience. Current limitations include scarce, exorbitantly priced bandwidth and considerable power consumption at higher data rates. Relaying systems use several shorter communications links instead of the conventional point-to-point transmission. This can allow for a lower power requirement and, due to the shorter broadcast range, bandwidth re-use may be more efficiently exploited. Code division multiple access (CDMA) is emerging as one of the most common methods for multi user access. Combining CDMA with time division duplexing (TDD) provides a system that supports asymmetric communications and relaying cost-effectively. The capacity of CDMA may be reduced by interference from other users, hence it is important that the routing of relays is performed to minimise interference at receivers. This thesis analyses relaying within the context of TDD-CDMA systems. Such a system was included in the initial draft of the European 3G specifications as opportunity driven multiple access (ODMA). Results are presented which demonstrate that ODMA allows for a more flexible capacity coverage trade-off than non-relaying systems. An investigation into the interference characteristics of ODMA shows that most interference occurs close to the base station (BS). Hence it is possible that in-cell routing to avoid the BS may increase capacity. As a result, a novel hybrid network topology is presented. ODMA uses path loss as a metric for routing. This technique does not avoid interference, and hence ODMA shows no capacity increase with the hybrid network. Consequently, a novel interference based routing algorithm and admission control are developed. When at least half the network is engaged in in-cell transmission, the interference based system allows for a higher capacity than a conventional cellular system. In an attempt to reduce transmitted power, a novel congestion based routing algorithm is introduced. This system is shown to have lower power requirement than any other analysed system and, when more than 2 hops are allowed, the highest capacity. The allocation of time slots affects system performance through co-channel interference. To attempt to minimise this, a novel dynamic channel allocation (DCA) algorithm is developed based on the congestion routing algorithm. By combining the global minimisation of system congestion in both time slots and routing, the DCA further increases throughput. Implementing congestion routed relaying, especially with DCA, in any TDD-CDMA system with in-cell calls can show significant performance improvements over conventional cellular systems.

Allocating resources to variable bitrate (VBR) teletraffic sources is not a trivial task because the impact of such sources on a buffered switch is difficult to predict. This problem has repercussions for call admission control (CAC) on asynchronous transfer mode (ATM) networks. In this thesis we report on investigations into the nature of several types of VBR teletraffic. The purpose of these investigations is to identify parameters of the traffic that may assist in the development of CAC algorithms. As such we concentrate on the correlation structure and marginal distribution; the two aspects of a teletraffic source that affect its behaviour through a buffered switch. The investigations into the correlation structure consider whether VBR video is selfsimilar or non-stationary. This question is significant as the exponent of self-similarity has been identified as being useful for characterising VBR teletraffic. Although results are inconclusive with regards to the original question, they do show that self-similar models are best able to capture the video data's behaviour. The investigations into the marginal distributions are in two parts. The first considers applying a structured Markovian model to ATM data and demonstrates how model parameters can be estimated from measurable properties of teletraffic data. This has implications for parametric CAC. The second part considers the use of stable distributions in teletraffic characterisation and modelling. We show that several teletraffic datasets are heavy tailed and then develop a framework for the estimation of stable distribution parameters. We finish by considering the effective bandwidths of stable distributions and models and by considering the effect of stable parameters on model behaviour. This is done in an attempt to develop a CAC algorithm based on the paradigms of self-similarity and stable distributions.

Digital visual communication has been increasingly adopted as an efficient new medium in a variety of different fields; multi-media computers, digital televisions, telecommunications, etc. Exchange of visual information between remote sites requires that digital video is encoded by compressing the amount of data and transmitting it through specified network connections. The compression and transmission of digital video is an amalgamation of statistical data coding processes, which aims at efficient exchange of visual information without technical barriers due to different standards, services, media, etc. It is associated with a series of different disciplines of digital signal processing, each of which can be applied independently. It includes a few different technical principles; distortionrate theory, prediction techniques and control theory. The MPEG (Moving Picture Experts Group) video compression standard is based on this paradigm, thus, it contains a variety of different coding parameters which may result in different performance depending on their values. It specifies the bit stream syntax and the decoding process as its normative parts. The encoder details remain nonnormative and are configured by a specific design. This means that the MPEG video encoder has a great deal of flexibility in the aspects of performance and implementation. This thesis deals with control techniques for the data rate of compressed video, which determine the encoding efficiency and video quality. The video rate control is achieved by adjusting quantisation step size depending on the occupancy of a transmission buffer memory which stores the compressed video data for a specific period of time. Conventional video rate control techniques have generally been based either on linear predictive or on control theoretic models. However, this thesis takes a different view on digital video and MPEG video coding, and focuses on the non-stationary and nonlinear nature of realistic moving pictures. Furthermore, considering the MPEG encoding structure involved in the different disciplines, A series of improvements for video rate control are proposed, each of which enhances the performance of the MPEG encoder. A nonlinear quantisation control technique is investigated, which controls the buffer occupancy with the quantiser using a series of nonlinear functions. Linear and nonlinear feed-forward networks are also employed to control the quantiser. The linear combiner is used as a linear estimator and a radial basis function network as a nonlinear one. Finally, fuzzy rulebased control is applied to exploit the advantages of the nonlinear control technique which is able to provide linguistic judgement in the control mechanism. All these techniques are employed according to two global approaches (feedforward and feedback) applied to the rate control. The performance evaluation is carried out in terms of controllability over bit rate variation and video quality, by conducting a series of simulations.

With the emergence of nonlinear dynamical systems analysis over recent years it has become clear that conventional time domain and frequency domain approaches to speech synthesis may be far from optimal. Using state space reconstructions of the time domain speech signal it is, at least in theory, possible to investigate a number of invariant geometrical measures for the underlying system which give a more thorough understanding of the dynamics of the system and therefore the form that any model should take. This thesis introduces a number of nonlinear dynamical analysis tools which are then applied to a database of vowels to extract the underlying invariant geometrical properties. The results of this analysis are then applied, using ideas taken from nonlinear dynamics, to the problem of speech synthesis and a novel synthesis technique is described and demonstrated. The tools used for the analysis are time delay embedding, singular value decomposition, correlation dimension, local singular value analysis, Lyapunov spectra and short term prediction properties. Although there have been many papers written about these tools, and algorithms proposed, there are currently no generally accepted techniques, especially for the calculation of Lyapunov spectra in the presence of noise and data length limitations. This thesis introduces all of the above tools and looks in detail at Lyapunov exponents and two major novel modifications are proposed that are demonstrated to be more robust than conventional techniques. The novel robust techniques are applied to a large database of vowel sounds showing that the vowels tested show evidence of nonlinear, low-dimensional, non-chaotic behaviour. It is particularly the evidence of non-chaotic behaviour that is of importance from a synthesis point of view and is used in the final section of the thesis which introduces a novel synthesis technique. The synthesis technique, which is based on ideas taken from nonlinear dynamics theory is detailed and demonstrated showing that it is capable of high quality natural sounding speech.

This thesis describes the application of a perceptually based spectral subtraction algorithm for the enhancement of non-stationary noise corrupted speech. Through examination of speech enhancement techniques, explanations are given for the choice of magnitude spectral subtraction and how the human auditory system can be modelled for frequency domain speech enhancement. It is discovered, that the cochlea provides the mechanical speech enhancement in the auditory system, through the use of masking. Frequency masking is used in spectral subtraction, to improve the algorithm execution time, and to shape the enhancement process making it sound natural to the ear. A new technique for estimation of background noise is presented, which operates during speech sections as well as pauses. This uses two microphones placed on opposite ends of the cellular handset. Using these, the algorithm determines whether the signal is speech, or noise, by examining the current and next frames presented to each microphone. This allows operation in non-stationary conditions, as the estimation is calculated for each frame, and a speech pause is not required for updating. A voting decision process decides the presence of speech or noise which determines which microphone the estimation is calculated from. The importance of an accurate noise estimate is highlighted with a new technique to reduce the effect of musical noise artifacts in the processed speech. This is a classic drawback of spectral subtraction techniques, and it is shown, that the trade off between noise reduction and speech distortion can be extended by this process. A new method for dealing with musical noise is described, which uses a combination of energy and variance examination of the spectrogram to segregate potential musical noise from desired speech sections. By examination of the spectrogram points surrounding musical noise sections, perceptually relevant values replace the corruption leading to cleaner enhanced speech. Any perceptual speech system requires accurate estimates of the clean speech masking thresholds, to prevent noisy sections being passed through the enhancement untouched. In this thesis, a method for the calculation of the estimated clean speech masking thresholds is derived. Classically, this requires an estimation of the clean speech before the thresholds can be derived, but this results in inaccuracy due to the presence of musical noise and spectral nulls. The new algorithm examines the thresholds produced by the corrupted speech, and the background noise, and from these determines the relationship between the two, to produce an estimate of the clean thresholds, with no operation performed on the actual speech signal. A discrepancy is found between the results for male and female speech, which, by examination of the perceptual process, is shown to be due to the different formant positions in male and female speech. Following the development of these parts, the entire enhancement algorithm is tested on a range of noise scenarios, using male and female speech. The results show, that the proposed algorithm is able to provide adequate performance in terms of noise reduction and speech quality.

Speech synthesis technology plays an important role in many aspects of man–machine interaction, particularly in telephony applications. In order to be widely accepted, the synthesised speech quality should be as human–like as possible. This thesis investigates novel techniques for the speech signal generation stage in a speech synthesiser, based on concepts from nonlinear dynamical theory. It focuses on natural–sounding synthesis for voiced speech, coupled with the ability to generate the sound at the required pitch. The one–dimensional voiced speech time–domain signals are embedded into an appropriate higher dimensional space, using Takens’ method of delays. These reconstructed state space representations have approximately the same dynamical properties as the original speech generating system and are thus effective models. A new technique for marking epoch points in voiced speech that operates in the state space domain is proposed. Using the fact that one revolution of the state space representation is equal to one pitch period, pitch synchronous points can be found using a Poincar´e map. Evidently the epoch pulses are pitch synchronous and therefore can be marked. The same state space representation is also used in a locally–linear speech synthesiser. This models the nonlinear dynamics of the speech signal by a series of local approximations, using the original signal as a template. The synthesised speech is natural–sounding because, rather than simply copying the original data, the technique makes use of the local dynamics to create a new, unique signal trajectory. Pitch modification within this synthesis structure is also investigated, with an attempt made to exploit the ˇ Silnikov–type orbit of voiced speech state space reconstructions. However, this technique is found to be incompatible with the locally–linear modelling technique, leaving the pitch modification issue unresolved. A different modelling strategy, using a radial basis function neural network to model the state space dynamics, is then considered. This produces a parametric model of the speech sound. Synthesised speech is obtained by connecting a delayed version of the network output back to the input via a global feedback loop. The network then synthesises speech in a free–running manner. Stability of the output is ensured by using regularisation theory when learning the weights. Complexity is also kept to a minimum because the network centres are fixed on a data–independent hyper–lattice, so only the linear–in–the–parameters weights need to be learnt for each vowel realisation. Pitch modification is again investigated, based around the idea of interpolating the weight vector between different realisations of the same vowel, but at differing pitch values. However modelling the inter–pitch weight vector variations is very difficult, indicating that further study of pitch modification techniques is required before a complete nonlinear synthesiser can be implemented.

This thesis studies two substructural simple type theories, extending the "separation" and "number-of-uses" readings of the basic substructural simply typed lambda-calculus with exchange. The first calculus, lambda_sep, extends the alpha lambda-calculus of O'Hearn and Pym by directly considering the representation of separation in a type system. We define type contexts with separation relations and introduce new type constructors of separated products and separated functions. We describe the basic metatheory of the calculus, including a sound and complete type-checking algorithm. We then give new categorical structure for interpreting the type judgements, and prove that it coherently, soundly and completely interprets the type theory. To show how the structure models separation we extend Day's construction of closed symmetric monoidal structure on functor categories to our categorical structure, and describe two instances dealing with the global and local separation. The second system, lambda_inplc, is a re-presentation of substructural calculus for in-place update with linear and non-linear values, based on Wadler's Linear typed system with non-linear types and Hofmann's LFPL. We identify some problems with the metatheory of the calculus, in particular the failure of the substitution rule to hold due to the call-by-value interpretation inherent in the type rules. To resolve this issue, we turn to categorical models of call-by-value computation, namely Moggi's Computational Monads and Power and Robinson's Freyd-Categories. We extend both of these to include additional information about the current state of the computation, defining Parameterised Freyd-categories and Parameterised Strong Monads. These definitions are equivalent in the closed case. We prove that by adding a commutativity condition they are a sound class of models for lambda_inplc. To obtain a complete class of models for lambda_inplc we refine the structure to better match the syntax. We also give a direct syntactic presentation of Parameterised Freyd-categories and prove that it is soundly and completely modelled by the syntax. We give a concrete model based on Day's construction, demonstrating how the categorical structure can be used to model call-by-value computation with in-place update and bounded heaps.

The growing demand for capacity in wireless communications is the driving force behind improving established networks and the deployment of a new worldwide mobile standard. Capacity calculations show that the direct sequence code division multiple access (DS-CDMA) technique has more capacity than the time division multiple access technique. Therefore, most 3rd generation mobile systems will incorporate some sort of DS-CDMA. In this thesis DS-CDMA receiver structures are investigated from the view point of pattern recognition which leads to new DS-CDMA receiver structures. It is known that the optimum DS-CDMA receiver has a nonlinear structure with prohibitive complexity for practical implementation. It is also known that the currently implemented receiver in 2nd generation DSCDMA mobile handsets has poor performance, because it suffers from multiuser interference. Consequently, this work focuses on sub-optimum nonlinear receivers for DS-CDMA in the downlink scenario. First, the thesis reviews DS-CDMA, established equalisers, DS-CDMA receivers and pattern recognition techniques. Then the new receivers are proposed. It is shown that DS-CDMA can be considered as a pattern recognition problem and hence, pattern recognition techniques can be exploited in order to develop DS-CDMA receivers. Another approach is to apply known equaliser structures for DS-CDMA. One proposed receiver is based on the Volterra series expansion and processes the received signal at the chip rate. Another receiver is a symbol rate radial basis function network (RBFN) receiver with reduced complexity. Subsequently, a receiver is proposed based on linear programming (LP) which is especially tailored for nonlinearly separable scenarios. The LP based receiver performance is equivalent to the known decorrelating detector in linearly separable scenarios. Finally, a hybrid receiver is proposed which combines LP and RBFN and which exploits knowledge gained from pattern recognition. This structure has lower complexity than the full RBF and good performance, and has a large potential for further improvements. Monte-Carlo simulations compare the proposed DS-CDMA receivers against established linear and nonlinear receivers. It is shown that all proposed receivers outperform the known linear receivers. The Volterra receiver’s complexity is relatively high for the performance gain achieved and might not suit practical implementation. The other receiver’s complexity was greatly reduced but it performs nearly as well as an optimum symbol by symbol detector. This thesis shows that DS-CDMA is a pattern recognition problem and that pattern recognition techniques can simplify DS-CDMA receiver structures. Knowledge is gained from the DSCDMA signal patterns which help to understand the problem of a DS-CDMA receiver. It should be noted that from the large number of known techniques, only a few pattern recognition techniques are considered in this work, and any further work should look at other techniques. Pattern recognition techniques can reduce the complexity of existing DS-CDMA receivers while maintaining performance, leading to novel receiver structures.

The main subject of this thesis is the investigation of antenna array techniques for improving the performance of the downlink of wideband code division multiple access (WCDMA) mobile communication systems. These communication systems operate in frequency division duplex (FDD) mode and the antenna arrays are employed in the base station. A number of diversity, beamforming and hybrid techniques are analysed and their bit error ratio (BER) versus signalto- noise ratio (SNR) performance is calculated as a function of the eigenvalues of the mean channel correlation matrix, where this is applicable. Also, their BER versus SNR performance is evaluated by means of computer simulations in various channel environments and using different numbers of transmit antenna elements in the base station. The simulation results of the techniques, along with other characteristics, are compared to examine the relationship among their performance in various channel environments and investigate which technique is most suitable for each channel environment. Next, a combination of the channel correlation matrix eigenvalue decomposition and space-time processing is proposed as a possible open loop approach to the downlink data signal transmission. It decomposes the channel into M components in the form of eigenvectors (M is the number of transmit antennas in the base station), and attempts to minimise the transmit power that is needed to achieve a target BER at the mobile receiver by employing the optimum number of these eigenvectors. The lower transmit power and the directional transmission by means of eigenvectors are expected to lower interference levels to non-desired users (especially to those users who are not physically close to the direction(s) of transmission). Theoretical and simulation results suggest that this approach performs better than other presented open loop techniques, while the performance gain depends on M and the channel environment. In simulations it is usually assumed that the base and mobile station have access to perfect estimates of all needed parameters (e.g. channel coecients). However, in practical systems they make use of pilot and/or feedback signals to obtain estimates of these parameters, which result in noisy estimates. The impact of the noisy estimates on the performance of various techniques is investigated by computer simulations, and the results suggest that there is typically some performance loss. The loss depends on the parameter that is estimated from pilot signals, and may be a function of M, SNR and/or the channel environment. In certain beamforming techniques the base station operates the transmit antenna array in an open loop fashion by estimating the downlink weight vector from the directional information of the uplink channel. Nevertheless, in FDD systems this results in performance loss due to the separation between the uplink and downlink carrier frequencies (`FDD gap'). This loss is quantified and the results show that it is a function of M and the FDD gap. Also, a very simple technique for compensating this loss is proposed, and results obtained after its application suggest that it eliminates most of the loss. Comparison of the proposed technique with an existing compensation technique suggests that, even though the latter is more complex than the former, it yields very little additional improvement.

Noise or interference is often assumed to be a random process. Conventional linear filtering, control or prediction techniques are used to cancel or reduce the noise. However, some noise processes have been shown to be nonlinear and deterministic. These nonlinear deterministic noise processes appear to be random when analysed with second order statistics. As nonlinear processes are widespread in nature it may be beneficial to exploit the coherence of the nonlinear deterministic noise with nonlinear filtering techniques. The nonlinear deterministic noise processes used in this thesis are generated from nonlinear difference or differential equations which are derived from real world scenarios. Analysis tools from the theory of nonlinear dynamics are used to determine an appropriate sampling rate of the nonlinear deterministic noise processes and their embedding dimensions. Nonlinear models, such as the Volterra series filter and the radial basis function network are trained to model or predict the nonlinear deterministic noise process in order to reduce the noise in a system. The nonlinear models exploit the structure and determinism and, therefore, perform better than conventional linear techniques. These nonlinear techniques are applied to cancel broadband nonlinear deterministic noise which corrupts a narrowband signal. An existing filter method is investigated and compared with standard linear techniques. A new filter method is devised to overcome the restrictions of the existing filter method. This method combines standard signal processing concepts (filterbanks and multirate sampling) with linear and nonlinear modelling techniques. It overcomes the restrictions associated with linear techniques and hence produces better performance. Other schemes for cancelling broadband noise are devised and investigated using quantisers and cascaded radial basis function networks. Finally, a scheme is devised which enables the detection of a signal of interest buried in heavy chaotic noise. Active noise control is another application where the acoustic noise may be assumed to be a nonlinear deterministic process. One of the problems in active noise control is the inversion process of the transfer function of the loudspeaker. This transfer function may be nonminimum phase. Linear controllers only perform sub-optimally in modelling the noncausal inverse transfer function. To overcome this problem in conjunction with the assumption that the acoustic noise is nonlinear and deterministic a combined linear and nonlinear controller is devised. A mathematical expression for the combined controller is derived which consists of a linear system identification part and a nonlinear prediction part. The traditional filtered-x least mean squares scheme in active noise control does not allow the implementation of a nonlinear controller. Therefore, a control scheme is devised to allow a nonlinear controller in conjunction with an adaptive block least squares algorithm. Simulations demonstrate that the combined linear and nonlinear controller outperforms the conventional linear controller.