This thesis presents a fully pipelined and parameterised parallel hardware implementation of a large vocabulary, user-independent and continuous speech recognition system for use in mobile applications. Algorithm acceleration is achieved by realising in hardware the most time-consuming components of the speech recognition system. By adopting a parallel solution, the necessary calculations can be completed in a sufficiently short elapsed time for embedded target systems. Sphinx 3 is identified as an appropriate speech recognition system for this work and is profiled to determine the most time-consuming parts of the code. As these parts of the code employ calculations based on floating point operations, which are not suitable for the high-performance and low-power execution on embedded systems, these calculations have been converted to scaled integer operations. It is verified using the AN4, RM1 and TIMIT speech databases that the scaled integer version of the speech recognition system can achieve a similar word error rate to the original floating point version, while taking less than 8% of the calculation time used by the original version. The scaled integer version of the speech recognition system is redesigned in VHDL for parallel implementation in electronic hardware. The designs of a calculation module and a data module are described, both of which can be configured according to the number of parallel units and the data module can be configured according to the total numbers of feature vectors and senones used in the speech representation. The hardware designs are synthesised to a range of FPGAs and the results showed that the larger Virtex7 devices are capable of holding several thousands of senones which are sufficient for most recognition tasks. Hardware designs with different numbers of parallel calculation units are simulated at both behavioural level and platform-based level and the resulting implementations are able to operate in real time. The results show that the hardware implementation, even with only one calculation unit, can perform the same calculations almost 80 times faster than does a modern embedded microprocessor, even when operating at only one fifth of the clock frequency. With larger numbers of parallel calculation units, the whole design can operate at even lower clock frequencies, saving power while maintaining a rapid calculation speed. The hardware designs are also implemented on a physical system having both an FPGA and a microprocessor board to demonstrate the operational capabilities of a full system.

Synthetic aperture techniques have been applied very successfully for many years in astronomy and radar to obtain high resolution images, an outstanding example in recent years being the use in remote sensing satellite systems. In underwater acoustics, because of the inherent problems caused by random fluctuations in the signal path, the slow velocity of the acoustic wave and the unknown movements of the transducer as it traverses the aperture, the application of the synthetic aperture technique has mainly been limited to the very useful but rather inferior non-coherent technique known as side-scan sonar. However the rapid advances that are being made in micro-chip technology and fast digital signal processing, and the development in image processing algorithms has created renewed interest in the possible application of the synthetic aperture technique to underwater acoustics. This thesis describes such a study.

The thesis describes various detection processes which are suitable for use in a synchronous serial data-transmission system operating at a transmission rate of up to 20,000 bits per second over a slowly time-varying channel, The methods of operation of different detection processes are first described, with reference to binary and quaternary baseband signals, which includes the case when V,S,B,signals are transmitted over ·telephone circuits or H,F,radio links. The results of computer simulation tests are presented, comparing the tolerances of the detection processes to additive white Gaussian noise with the tolerances of conventional linear and non-linear equalisers. Several different time invariant channels are used in the tests. It is shown that two relatively simple detection processes can achieve a considerable improvement in tolerance to noise over both linear and non-linear equalisers of optimum design, Several of the most promising detection processes and a few new detection processes are then modified to use 4-point, 16-point and 32-point Q,A,M,signals. The methods of operation of the different detection processes are then described for the signal format being considered. The results of computer simulation tests are presented comparing their tolerances to additive white Gaussian noise with those · of the linear and non-linear equalisers of optimum design. It is shown that these detection processes operate most efficiently, in terms of tolerance to noise and in the number of sequential operations, if the transmitted signal contains the smallest number of possible signal levels. Finally, several simple methods of estimating the sampled impulse response of the channel are presented. One of these channel estimation techniques gives a very low error in the estimated response, while giving a good rate of adaptation to a time-varying channel.

Conventional pulse-code and differential pulse-code modulators for encoding video signals are difficult to realise economically. To alleviate this problem, a technique which divides the modulators into two stages is proposed. The first stage is a two-bit instantaneously adaptive delta modulator operating at a high clock rate and using low-precision components. Two-bit signals conveying polarity and magnitude information are produced by this delta modulator and used as the input to the second stage, a code converter. The code converter transforms, digitally, delta modulated signals into Pulse Code Modulation (PCM) or Differential Pulse Code Modulation (DPCM) signals. The resolution of the final PCM or DPCM encoder depends on the performance of the delta modulator used as the input stage. For that reason, the performance of the two-bit Instantaneously Adaptive Delta Modulation (2BIADM) encoder is evaluated. This evaluation is made in two steps. First, a semi-empirical anaysis of the High Information Delta Modulation (HIDM) is made, because the 2BIADM system is derived from the HIDM. Then the performance of the 2BIADM is derived considering the HIDM as a reference. For the HIDM and 2BIADM modulators operating at the same sampling frequency, the 2BIADM presents an improvement in peak signal-to-noise ratio (SNR) Of 6 to 8 dB. Expressions are established to enable SNR to be calculated for the HIDM as a function of the encoding parameters. The expressions also apply to Constant Factor Delta Modulation, and represent the only known method of estimating numerically the SNR for instantaneously adaptive delta modulators. The 2BIADM was tested, built and operated at a low sampling rate. This gave an insight into the operation of the proposed system, and complemented the computer simuLation analyses. The principles for the code conversion from the 2BIADM to PCM or DPCM are fully discussed. The 2BIADM does not impose restrictions on the values that the coefficients of the digital low-pass filter required in the code converter can assume. For low bandwidth expansion rates, it was verified that a 2BIADM-to-PCM conversion filter with 5 stages performs better than a HIDM-to-PCM conversion with a filter having 256 stages (both encoders operating at the same word-rate). A generalization of the 2-bit encoder to a N-bit adaptive DPCM system is outlined.

In protocol tunnelling, one application protocol is encapsulated within another carrier protocol in an unusual way to circumvent firewall policy. Application-layer tunnels are a significant security and resource abuse threat for networks because those applications which are restricted by firewalls such as high data-rate games, peer-to-peer file sharing, video and audio streaming, and chat are carried through via allowed protocols like HTTP, HTTPS and the firewall security policy is thwarted. Protocols such as HTTP and HTTPS are indispensable today for any network which has to be connected to the Internet; hence these become a high value target for running restricted applications via tunnelling. The identification of the actual application running across a network is important for network management, optimization, security and abuse prevention. The existing techniques for identification of applications running across the network, for example port number based identification, and packet data analysis techniques are not always successful, especially for applications which use encrypted tunnels. This work describes a statistical approach to detect applications which are running using application layer tunnels. Previous work has shown the packet size distribution to be an effective metric for detecting most network applications, both UDP and TCP based applications. In this work it is shown how packet stream statistics including packet size distributions can be used to differentiate and identify networked tunnelled applications successfully. Tunnelled applications are identifiable using the traffic statistical parameters. Traffic trace files of the applications were captured, statistical parameters were derived from the trace files, and then these parameters were used for training machine learning algorithms. The trained machine learning algorithm is then able to classify the other packet trace data as belonging to an application. Five different machine learning algorithms have been applied, and their performance accuracy is discussed. The entropy distance based Nearest Neighbour machine learning algorithm and the Euclidean Distance based Nearest Neighbour classifier had better results than others. This method of identification of tunnelled applications can be complimentary to other network security systems such as firewalls and Intrusion Detection Systems.

This thesis forms part of a joint universities project in which it is required to design and build a digital modem for the transmission of speech or data over a 900MHz land mobile radio channel. The main objectives being to try to maximize the bandwidth efficiency and attain near-optimum system performance. The theoretical modem design is presented here. The other parts of the system, that is the error control coding, speech coding, RF design qnd the actual hardware implementation are described elsewhere. All the systems described here have been designed to satisfy the overall system requirements. In particular, it must be possible to build this modem with existing technology without undue equipment complexity. All aspects of the digital'modem design are addressed, namely the choices of modulation scheme, pulse shaping filtering, packet structure and timing and synchronization methods suitable for the transmission of a digital signal over the fading radio channel. The important problems of channel estimation and data detection are examined in more detail. The first system described is one in which only one digital signal is transmitted in a narrowband channel. In the second system a novel technique of transmitting two signals in the same frequency band from two different mobiles to a single base station is described, which makes use of the fact that these two transmission paths are fading independently. The third system describes a method for transmitting back to these mobiles from the base station, again in the same frequency band. Although these systems have been designed specifically for use over 900MHz cellular land mobile radio channels, the ,techniques described are directly applicable to digital signal transmission over any flat fading channel.

This research is germane to the area of on-body antennas and the characterisation of antennas in close proximity to biological matter. The ranges of frequencies discussed are currently popular for mobile communications, namely 0.9GHz to 6GHz with spot frequencies of GSM900, GSM1800 and WiFi2.5GHz. Particular attention is given to the elimination of errors in measurement. This is achieved by the characterisation of an anechoic chamber; a study of the effects of cables; a study of the interaction of surface currents and the human body; a study of tissue simulating liquid; the design of a simple body phantom; the characterisation of the on-body channel for human males in wet and dry clothing and a comparison of perturbation on antennas close to humans and a phantom.

Introduction of data intensive multimedia and interactive services together with exponential growth of wireless applications have created a spectrum crisis. Many spectrum occupancy measurements, however, have shown that most of the allocated spectrum are used inefficiently indicating that radically new approaches are required for better utilization of spectrum. This motivates the concept of opportunistic spectrum sharing or the so-called cognitive radio technology that has great potential to improve spectrum utilization. This technology allows the secondary users to access the spectrum which is allocated to the licensed users in order to transmit their own signal without harmfully affecting the licensed users' communications. In this thesis, an optimal radio resource allocation algorithm is proposed for an OFDM based underlay cognitive radio networks. The proposed algorithm optimally allocates transmission power and OFDM subchannels to the users at the basestation in order to satisfy the quality of services and interference leakage constraints based on integer linear programming. To reduce the computational complexity, a novel recursive suboptimal algorithm is proposed based on a linear optimization framework. To exploit the spatial diversity, the proposed algorithms are extended to a MIMO-OFDM based cognitive radio network. Finally, a novel spatial multiplexing technique is developed to allocate resources in a cognitive radio network which consists of both the real time and the non-real users. Conditions required for convergence of the proposed algorithm are analytically derived. The performance of all these new algorithms are verified using MATLAB simulation results.

MIMO (Multiple Input Multiple Output) technology has been presented to significantly increase the wireless channel capacity and reliability without requiring additional radio spectrum or power. In MIMO systems, multiple antennas are mounted at both the transmitter and the receiver. When this technology is employed for a compact wireless terminal, one of the most challenging tasks is to reduce the high mutual coupling between closely placed antenna array elements. The high mutual coupling produces high correlation between antenna elements and affects the channel capacity of MIMO system. The objectives of this thesis are to design practical miniaturized structures to reduce high mutual coupling for small wireless terminals. The research is conducted in the following areas. Initially, a PIFA design and two-element PIFA array are proposed and optimized to operate at 1.9GHz. A pair of two coupled quarter-wavelength linear slits is inserted in a compact ground plane, resulting in significant reduction of the mutual coupling across antenna operating frequency band. In order to take up less space on the ground plane, instead of the linear slits, miniaturized convoluted slits are implemented between the two closely placed PIFAs. Although the convoluted slits have small area and are positioned close to the edges of the ground plane, the miniaturized convoluted slit structures achieve a reduction of mutual coupling between antenna elements and succeed in reducing the effect of the human body (head and hand) to the antennas. In order to further reduce the size of the slits etched on the compact ground plane, a novel double-layer slit-patch EBG structure is proposed. It consists of a two-layer structure including conducting patches and aperture slits placed on either side of a very thin dielectric layer. They are placed in very close proximity to each other (55μm). A two-element printed CPW-fed monopole array operating around 2.46GHz and a two-element UWB planar monopole array operating from 3GHz to 6GHz have been employed to investigate the proposed slit-patch EBG structures. The optimized double-layer slit-patch EBG structure yields a significant reduction of the mutual coupling and produces the maximum miniaturization of antenna array. Another novel convoluted slit-patch EBG structure has been presented to reduce the mutual coupling between two PIFAs operating at 1.9GHz. These results demonstrate that the slit-patch EBG structure is a feasible technology to reduce the mutual coupling between multiple antennas for compact wireless terminals.

Synthetic Aperture Radar (SAR) imaging from geosynchronous orbit has significant potential advantages over conventional low-Earth orbit (LEO) radars, but also challenges to overcome. This thesis investigates both active and passive geosynchronous SAR configurations, presenting their different features and advantages. Following a system design trade-off that involved phase uncertainties, link budget, frequency and integration time, an L band bi-static configuration with 8-hour integration time that reuses the signal from a non-cooperative transmitter has been presented as a suitable solution. Cranfield Space Research Centre looked into this configuration and proposed the GeoSAR concept, an L band bi-static SAR based on the concept by Prati et al. (1998). It flies along a circular ground track orbit, reuses the signal coming from a noncooperative transmitter in GEO and achieves a spatial resolution of about 100 m. The present research contributes to the GeoSAR concept exploring the implications due to the 8-hour integration time and providing insights about its performance and its possible fields of application. Targets such as canopies change their backscattered phase on timescales of seconds due to their motion. On longer time scales, changes in dielectric properties of targets, Earth tides and perturbations in the structure of the atmosphere contribute to generate phase fluctuations in the collected signals. These phenomena bring temporal decorrelation and cause a reduction in SAR coherent integration gain. They have to be compensated for if useful images are to be provided. A SAR azimuth simulator has been developed to study the influence of temporal decorrelation on GeoSAR point spread function. The analysis shows that ionospheric delay is the major source of decorrelation; other effects, such as tropospheric delay and Earth tides, have to be dealt with but appear to be easier to handle. Two different options for GeoSAR interferometry have been discussed. The system is well suited to differential interferometry, due to the short perpendicular baseline induced by the geometry. A GeoSAR has advantages over a Low Earth Orbit (LEO) SAR system to monitor processes with significant variability over daily or shorter timescales (e.g. soil moisture variation). This potential justifies further study of the concept.