Publication Search Results

Now showing 1 - 10 of 25

  • (2006) Chan, Victor Chi-Wai

  • (2006) Gupta, Amit Kumar
    The Key algorithms of JPEG2000, the new image compression standard, have high computational complexity and thus present challenges for efficient implementation. This has led to research on the hardware optimization of JPEG2000 for its efficient realization. Luckily, in the last century the growth in Microelectronics allows us to realize dedicated ASIC solutions as well as hardware/software FPGA based solutions for complex algorithms such as JPEG2000. But an efficient implementation within hard constraints of area and throughput, demands investigations of key dependencies within the JPEG2000 system. This work presents algorithms and VLSI architectures to realize a high performance JPEG2000 compression system. The embedded block coding algorithm which lies at the heart of a JPEG2000 compression system is a main contributor to enhanced JPEG2000 complexity. This work first concentrates on algorithms to realize low-cost high throughput Block Coder (BC) system. For this purpose concurrent symbol processing capable Bit Plane Coder architecture is presented. Further optimal 2 sub-bank memory and an efficient buffer architectures are designed to keep the hardware cost low. The proposed overall BC system presents the highest Figure Of Merit (FOM) in terms of throughput versus hardware cost in comparison to existing BC solutions. Further, this work also investigates the challenges involved in the efficient integration of the BC with the overall JPEG2000 system. A novel low-cost distortion estimation approach with near-optimal performance is proposed which is necessary for accurate rate-control performance of JPEG2000. Additionally low bandwidth data storage and transfer techniques are proposed for efficient transfer of subband samples to the BC. Simulation results show that the proposed techniques have approximately 4 times less bandwidth than existing architectures. In addition, an efficient high throughput block decoder architecture based on the proposed selective sample-skipping algorithm is presented. The proposed architectures are designed and analyzed on both ASIC and FPGA platforms. Thus, the proposed algorithms, architectures and efficient BC integration strategies are useful for realizing a dedicated ASIC JPEG2000 system as well as a hardware/software FPGA based JPEG2000 solution. Overall this work presents algorithms and architectures to realize a high performance JPEG2000 system without imposing any restrictions in terms of coding modes or block size for the BC system.

  • (2006) Chan, Francis Chun Ngai
    The objective of the thesis is concerned on the problem of detecting superpositional signals in a wireless channel. In many wireless systems, an observed signal is commonly represented as a linear combination of the transmitted signal with the interfering signals dispersed in space and time. These systems are generally known as the interference-limited systems. The mathematical model of these systems is generally referred as a superpositional model. A distinguished characteristic of signal transmission in a time-varying wireless channel is that the channel process is not known a priori. Reliable signal reception inherently requires exploiting the structure of the interfering signals under channel uncertainty. Our goal is to design computational efficient receivers for various interference-limited systems by using advanced statistical signal processing techniques. The thesis consists of four main parts. Firstly, we have proposed a novel Multi-Input Multi-Output (MIMO) signal detector, known as the neighbourhood exploring detector (NED). According to the maximum likelihood principle, the space time MIMO detection problem is equivalent to a NP-hard combinatorial optimization problem. The proposed detector is a sub-optimal maximum likelihood detector which eliminates exhaustive multidimensional searches. Secondly, we address the problem of signal synchronization for Global Positioning System (GPS) in a multipath environment. The problem of multipath mitigation constitutes a joint estimation of the unknown amplitudes, phases and time delays of the linearly combined signals. The complexity of the nonlinear joint estimation problem increases exponentially with the number of signals. We have proposed two robust GPS code acquisition systems with low complexities. Thirdly, we deal with the problem of multipath mitigation in the spatial domain. A GPS receiver integrated with the Inertial Navigation System (INS) and a multiple antenna array is considered. We have designed a software based GPS receiver which effectively estimates the directions of arrival and the time of arrival of the linearly combined signals. Finally, the problem of communications with unknown channel state information is investigated. Conventionally, the information theoretical communication problem and the channel estimation problem are decoupled. However the training sequence, which facilitates the estimation of the channel, reduces the throughput of the channel. We have analytically derived the optimal length of the training sequence which maximizes the mutual information in a block fading channel.

  • (2006) Herborn, Stephen
    Users (or software agents) are served by multiple networked terminal devices, each of which may in turn have multiple network interfaces. This multi-homing at both user and device level presents new opportunities for mobility handling. Mobility may be handled by switching ongoing application data streams between devices, by utilising intermediary adaptation or connectivity enhancement services, or both. However this requires middleware support that is not provided by current systems. This thesis presents a set of integrated solutions to enable this kind of mobility handling, based on concept of Personal Networks (PN). Personal Networks (PN) consist of dynamic conglomerations of terminal and service devices tasked to facilitate the delivery of information to and from a single focal point, which may be a human user or software agent. This concept creates the potential to view mobility handling as a path selection problem, since there may be multiple valid terminal device and service proxy configurations that can successfully carry a given communication session from one PN to another PN. Depending on context, it may be necessary to switch between paths. To this end, this thesis proposes and evaluates a set of inter-dependent mechanisms to facilitate the discovery and use of different candidate end-to-end paths. The proposal comprises mechanisms for secure inter-device mobility using delegated cryptographic identifiers, autonomous service proxy selection and composition, and distributed resolution of cryptographic identifiers to lower layer addresses.

  • (2006) Sadeghi, Parastoo
    In the first part of this thesis, the information capacity of time-varying fading channels is analysed using finite-state Markov channel (FSMC) models. Both fading channel amplitude and fading channel phase are modelled as finite-state Markov processes. The effect of the number of fading channel gain partitions on the capacity is studied (from 2 to 128 partitions). It is observed that the FSMC capacity is saturated when the number of fading channel gain partitions is larger than 4 to 8 times the number of channel input levels. The rapid FSMC capacity saturation with a small number of fading channel gain partitions can be used for the design of computationally simple receivers, with a negligible loss in the capacity. Furthermore, the effect of fading channel memory order on the capacity is studied (from first- to fourth-order). It is observed that low-order FSMC models can provide higher capacity estimates for fading channels than high-order FSMC models, especially when channel states are poorly observable in the presence of channel noise. To explain the effect of memory order on the FSMC capacity, the capacities of high-order and low-order FSMC models are analytically compared. It is shown that the capacity difference is caused by two factors: 1) the channel entropy difference, and 2) the channel observability difference between the high-order and low-order FSMC models. Due to the existence of the second factor, the capacity of high-order FSMC models can be lower than the capacity of low-order FSMC models. Two sufficient conditions are proven to predict when the low-order FSMC capacity is higher or lower than the high-order FSMC capacity. In the second part of this thesis, a new implicit (blind) channel estimation method in time- varying fading channels is proposed. The information source emits bits ’0’ and ’1’ with unequal probabilities. The unbalanced source distribution is used as a priori known signal structure at the receiver for channel estimation. Compared to pilot-symbol-assisted channel estimation, the proposed channel estimation technique can achieve a superior receiver bit error rate performance, especially at low signal to noise ratio conditions.

  • (2006) Wongrujira, Krit
    With the increasing growth of communication networks, systems, and devices,technology has driven much research and development on a variety of communication protocols, applications, and smart devices. As a result, a variety of heterogeneous networks, de facto and standards have emerged. In parallel, users are also demanding to seamlessly use any device on any network infrastructure within this heterogeneous environment. To go beyond this problem, one of solutions is deploying service entities to be served in/on the network. The services can be composed of many independent service entities to redirect the communication flow combined with pipelined transformations. The service in the network can start from an application to intercept the user flow in a range of formats, originating in different access networks, and deliver them appropriately formatted for a particular end terminal based on the other end user preferences. To provide independent services to be served within the network, we have created a decentralized virtual market-place, which facilitates a place for services to be traded, discovered, and composed between providers and consumers. This approach of virtual market based on a decentralized system is leading to many advantages and challenges compared to other traditional network services. This thesis presents a novel approach towards using market management techniques to improve cooperation among traders in the community, while enhancing the community-oriented network service architecture. Without the centralized control, the decentralized virtual market-place requires the inclusion of techniques to provide better incentives. Given such incentives, rational traders will choose to behave co-operatively and contribute their resources to maximize the efficiency of the community. This will allow an application with dramatically improved utility, efficiency, and robustness and hence enable whole new domains of use. The viability of the decentralized virtual market-place is demonstrated using a prototype implementation and simulation system. The results have clearly shown that the decentralized virtual market-place can improve and overcome the major problems of most existing network service models.