No description available.

Internet-based applications and services are pervading everyday life. Moreover, the growing popularity of real-time, time-critical and mission-critical applications set new challenges to the Internet community. The requirement for reducing response time, and therefore latency control is increasingly emphasized. This thesis seeks to reduce queueing delay through active queue management. While mathematical studies and research simulations reveal that complex trade-off relationships exist among performance indices such as throughput, packet loss ratio and delay, etc., this thesis intends to find an improved active queue management algorithm which emphasizes delay control without trading much on other performance indices such as throughput and packet loss ratio. The thesis observes that in TCP/IP network, packet loss ratio is a major reflection of congestion severity or load. With a properly functioning active queue management algorithm, traffic load will in general push the feedback system to an equilibrium point in terms of packet loss ratio and throughput. On the other hand, queue length is a determinant factor on system delay performance while has only a slight influence on the equilibrium. This observation suggests the possibility of reducing delay while maintaining throughput and packet loss ratio relatively unchanged. The thesis also observes that queue length fluctuation is a reflection of both load changes and natural fluctuation in arriving bit rate. Monitoring queue length fluctuation alone cannot distinguish the difference and identify congestion status; and yet identifying this difference is crucial in finding out situations where average queue size and hence queueing delay can be properly controlled and reasonably reduced. However, many existing active queue management algorithms only monitor queue length, and their control policies are solely based on this measurement. In our studies, our novel finding is that the arriving bit rate distribution of all sources contains information which can be a better indication of congestion status and has a correlation with traffic burstiness. And this thesis develops a simple and scalable way to measure its two most important characteristics, namely the mean ii and the variance of the arriving rate distribution. The measuring mechanism is based on a Zombie List mechanism originally proposed and deployed in Stabilized RED to estimate the number of flows and identify misbehaving flows. This thesis modifies the original zombie list measuring mechanism, makes it capable of measuring additional variables. Based on these additional measurements, this thesis proposes a novel modification to the RED algorithm. It utilizes a robust adaptive mechanism to ensure that the system reaches proper equilibrium operating points in terms of packet loss ratio and queueing delay under various loads. Furthermore, it identifies different congestion status where traffic is less bursty and adapts RED parameters in order to reduce average queue size and hence queueing delay accordingly. Using ns-2 simulation platform, this thesis runs simulations of a single bottleneck link scenario which represents an important and popular application scenario such as home access network or SoHo. Simulation results indicate that there are complex trade-off relationships among throughput, packet loss ratio and delay; and in these relationships delay can be substantially reduced whereas trade-offs on throughput and packet loss ratio are negligible. Simulation results show that our proposed active queue management algorithm can identify circumstances where traffic is less bursty and actively reduce queueing delay with hardly noticeable sacrifice on throughput and packet loss ratio performances. In conclusion, our novel approach enables the application of adaptive techniques to more RED parameters including those affecting queue occupancy and hence queueing delay. The new modification to RED algorithm is a scalable approach and does not introduce additional protocol overhead. In general it brings the benefit of substantially reduced delay at the cost of limited processing overhead and negligible degradation in throughput and packet loss ratio. However, our new algorithm is only tested on responsive flows and a single bottleneck scenario. Its effectiveness on a combination of responsive and non-responsive flows as well as in more complicated network topology scenarios is left for future work.

Video transcoding is an essential tool to promote inter-operability between different video communication systems. This thesis presents two novel video transcoders, both operating on bitstreams of the cur- rent H.264/AVC standard. The first transcoder converts H.264/AVC bitstreams to a Wavelet Scalable Video Codec (W-SVC), while the second targets the emerging High Efficiency Video Coding (HEVC). Scalable Video Coding (SVC) enables low complexity adaptation of compressed video, providing an efficient solution for content delivery through heterogeneous networks. The transcoder proposed here aims at exploiting the advantages offered by SVC technology when dealing with conventional coders and legacy video, efficiently reusing information found in the H.264/AVC bitstream to achieve a high rate-distortion performance at a low complexity cost. Its main features include new mode mapping algorithms that exploit the W-SVC larger macroblock sizes, and a new state-of-the-art motion vector composition algorithm that is able to tackle different coding configurations in the H.264/AVC bitstream, including IPP or IBBP with multiple reference frames. The emerging video coding standard, HEVC, is currently approaching the final stage of development prior to standardization. This thesis proposes and evaluates several transcoding algorithms for the HEVC codec. In particular, a transcoder based on a new method that is capable of complexity scalability, trading off rate-distortion performance for complexity reduction, is proposed. Furthermore, other transcoding solutions are explored, based on a novel content-based modeling approach, in which the transcoder adapts its parameters based on the contents of the sequence being encoded. Finally, the application of this research is not constrained to these transcoders, as many of the techniques developed aim to contribute to advance the research on this field, and have the potential to be incorporated in different video transcoding architectures.

Music understanding is a process closely related to the knowledge and experience of the listener. The amount of knowledge required is relative to the complexity of the task in hand. This dissertation is concerned with the problem of automatically decomposing musical signals into a score-like representation. It proposes that, as with humans, an automatic system requires knowledge about the signal and its expected behaviour to correctly analyse music. The proposed system uses the blackboard architecture to combine the use of knowledge with data provided by the bottom-up processing of the signal's information. Methods are proposed for the estimation of pitches, onset times and durations of notes in simple polyphonic music. A method for onset detection is presented. It provides an alternative to conventional energy-based algorithms by using phase information. Statistical analysis is used to create a detection function that evaluates the expected behaviour of the signal regarding onsets. Two methods for multi-pitch estimation are introduced. The first concentrates on the grouping of harmonic information in the frequency-domain. Its performance and limitations emphasise the case for the use of high-level knowledge. This knowledge, in the form of the individual waveforms of a single instrument, is used in the second proposed approach. The method is based on a time-domain linear additive model and it presents an alternative to common frequency-domain approaches. Results are presented and discussed for all methods, showing that, if reliably generated, the use of knowledge can significantly improve the quality of the analysis.

When this research was started, TD-SCDMA (Time Division Synchronous Code Division Multiple Access) was still in the research/ development phase, but now, at the time of writing this thesis, it is in commercial use in 10 large cities in China including Beijing and Shang Hai. In all of these cities HSDPA is enabled. The roll-out of the commercial deployment is progressing fast with installations in another 28 cities being underway now. However, during the pre-commercial TD-SCDM trail in China, which started from year 2006, some interference problems have been noticed especially in the network planning and initialization phases. Interference is always an issue in any network and the goal of the work reported in this thesis is to improve network coverage and capacity in the presence of interference. Based on an analysis of TD-SCDMA issues and how network interference arises, this thesis proposes two enhancements to the network in addition to the standard N-frequency technique. These are (i) the introduction of the concentric circle cell concept and (ii) the addition of a relay network that makes use of other users at the cell boundary. This overall approach not only optimizes the resilience to interference but increases the network coverage without adding more Node Bs. Based on the cell planning parameters from the research, TD-SCDMA HSDPA services in dense urban area and non-HSDPA services in rural areas were simulated to investigate the network performance impact after introducing the relay network into a TD-SCDMA network. The results for HSDPA applications show significant improvement in the TDSCDMA relay network both for network capacity and network interference aspects compared to standard TD-SCDMA networks. The results for non- HSDPA service show that although the network capacity has not changed after adding in the relay network (due to the code limitation in TD-SCDMA), the TD-SCDMA relay network has better interference performance and greater coverage.

Automated human action recognition plays a critical role in the development of human-machine communication, by aiming for a more natural interaction between artificial intelligence and the human society. Recent developments in technology have permitted a shift from a traditional human action recognition performed in a well-constrained laboratory environment to realistic unconstrained scenarios. This advancement has given rise to new problems and challenges still not addressed by the available methods. Thus, the aim of this thesis is to study innovative approaches that address the challenging problems of human action recognition from video captured in unconstrained scenarios. To this end, novel action representations, feature selection methods, fusion strategies and classification approaches are formulated. More specifically, a novel interest points based action representation is firstly introduced, this representation seeks to describe actions as clouds of interest points accumulated at different temporal scales. The idea behind this method consists of extracting holistic features from the point clouds and explicitly and globally describing the spatial and temporal action dynamic. Since the proposed clouds of points representation exploits alternative and complementary information compared to the conventional interest points-based methods, a more solid representation is then obtained by fusing the two representations, adopting a Multiple Kernel Learning strategy. The validity of the proposed approach in recognising action from a well-known benchmark dataset is demonstrated as well as the superior performance achieved by fusing representations. Since the proposed method appears limited by the presence of a dynamic background and fast camera movements, a novel trajectory-based representation is formulated. Different from interest points, trajectories can simultaneously retain motion and appearance information even in noisy and crowded scenarios. Additionally, they can handle drastic camera movements and a robust region of interest estimation. An equally important contribution is the proposed collaborative feature selection performed to remove redundant and noisy components. In particular, a novel feature selection method based on Multi-Class Delta Latent Dirichlet Allocation (MC-DLDA) is introduced. Crucial, to enrich the final action representation, the trajectory representation is adaptively fused with a conventional interest point representation. The proposed approach is extensively validated on different datasets, and the reported performances are comparable with the best state-of-the-art. The obtained results also confirm the fundamental contribution of both collaborative feature selection and adaptive fusion. Finally, the problem of realistic human action classification in very ambiguous scenarios is taken into account. In these circumstances, standard feature selection methods and multi-class classifiers appear inadequate due to: sparse training set, high intra-class variation and inter-class similarity. Thus, both the feature selection and classification problems need to be redesigned. The proposed idea is to iteratively decompose the classification task in subtasks and select the optimal feature set and classifier in accordance with the subtask context. To this end, a cascaded feature selection and action classification approach is introduced. The proposed cascade aims to classify actions by exploiting as much information as possible, and at the same time trying to simplify the multi-class classification in a cascade of binary separations. Specifically, instead of separating multiple action classes simultaneously, the overall task is automatically divided into easier binary sub-tasks. Experiments have been carried out using challenging public datasets; the obtained results demonstrate that with identical action representation, the cascaded classifier significantly outperforms standard multi-class classifiers.

This research looks at the design of an open transport protocol that supports a range of services including multimedia over low data-rate networks. Low data-rate multimedia applications require a system that provides quality of service (QoS) assurance and flexibility. One promising field is the area of content-based coding. Content-based systems use an array of protocols to select the optimum set of coding algorithms. A content-based transport protocol integrates a content-based application to a transmission network. General transport protocols form a bottleneck in low data-rate multimedia communicationbsy limiting throughpuot r by not maintainingt iming requirementsT. his work presents an original model of a transport protocol that eliminates the bottleneck by introducing a flexible yet efficient algorithm that uses an open approach to flexibility and holistic architectureto promoteQ oS.T he flexibility andt ransparenccyo mesi n the form of a fixed syntaxt hat providesa seto f transportp rotocols emanticsT. he mediaQ oSi s maintained by defining a generic descriptor. Overall, the structure of the protocol is based on a single adaptablea lgorithm that supportsa pplication independencen, etwork independencea nd quality of service. The transportp rotocol was evaluatedth rougha set of assessmentos:f f-line; off-line for a specific application; and on-line for a specific application. Application contexts used MPEG-4 test material where the on-line assessmenuts eda modified MPEG-4 pl; yer. The performanceo f the QoSc ontrolledt ransportp rotocoli s often bettert hano thers chemews hen appropriateQ oS controlledm anagemenatl gorithmsa re selectedT. his is shownf irst for an off-line assessmenwt here the performancei s compared between the QoS controlled multiplexer,a n emulatedM PEG-4F lexMux multiplexers chemea, ndt he targetr equirements. The performanceis also shownt o be better in a real environmentw hen the QoS controlled multiplexeri s comparedw ith the real MPEG-4F lexMux scheme.

This research focuses on the investigation of a fast simulation method for broadband telecommunication networks, such as ATM networks and IP networks. As a result of this research, a hybrid simulation model is proposed, which combines the analytical modelling and event-driven simulation modelling to speeding up the overall simulation. The division between foreground and background traffic and the way of dealing with these different types of traffic to achieve improvement in simulation time is the major contribution reported in this thesis. Background traffic is present to ensure that proper buffering behaviour is included during the course of the simulation experiments, but only the foreground traffic of interest is simulated, unlike traditional simulation techniques. Foreground and background traffic are dealt with in a different way. To avoid the need for extra events on the event list, and the processing overhead, associated with the background traffic, the novel technique investigated in this research is to remove the background traffic completely, adjusting the service time of the queues for the background traffic to compensate (in most cases, the service time for the foreground traffic will increase). By removing the background traffic from the event-driven simulator the number of cell processing events dealt with is reduced drastically. Validation of this approach shows that, overall, the method works well, but the simulation using this method does have some differences compared with experimental results on a testbed. The reason for this is mainly because of the assumptions behind the analytical model that make the modelling tractable. Hence, the analytical model needs to be adjusted. This is done by having a neural network trained to learn the relationship between the input traffic parameters and the output difference between the proposed model and the testbed. Following this training, simulations can be run using the output of the neural network to adjust the analytical model for those particular traffic conditions. The approach is applied to cell scale and burst scale queueing to simulate an ATM switch, and it is also used to simulate an IP router. In all the applications, the method ensures a fast simulation as well as an accurate result.

A real dynamic plant is used to compare, test and assess the present theoretical techniques of adaptive, learning or intelligent control under practical criteria. Work of this nature has yet to be carried out if "intelligent control" is to have a place in everyday practice. The project follows a natural pattern of development, the construction of computer programmes being an important part of it. First, a. real plant - a model steam engine - and its electronic interface with a general purpose digital computer are designed and built as part of the project. A rough mathematical model of the plant is then obtained through identification tests. Second, conventional control of the plant is effected using digital techniques and the above mentioned mathematical model, and the results are saved to compare with and evaluate the results of "intelligent control". Third, a few well-known adaptive or learning control algorithms are investigated and implemented. These tests bring out certain practical problems not encountered or not given due consideration in theoretical or simulation studies. Alternatively, these problems materialise because assumptions made on paper are not readily available in practice. The most important of these problematic. assumptions are those relating to computational time and storage, convergence of the adaptive or learning algorithm and the training of the controller. The human operator as a distinct candidate for the trainer is also considered and the problems therein are discussed. Finally, the notion of fuzzy sets and logic is viewed from the control point and a controller using this approach is developed and implemented. The operational advantages and the results obtained, albeit preliminary, demonstrate the potential power of this notion and provide the grounds for further work in this area.

Resource Management is a crucial aspect in the next generation cellular networks since the use of W-CDMA technology gives an inherent flexibility in managing the system capacity. The concept of a “Service Level Agreement” (SLA) also plays a very important role as it is the means to guarantee the quality of service provided to the customers in response to the level of service to which they have subscribed. Hence there is a need to introduce effective SLA-based policies as part of the radio resource management. This work proposes the application of intelligent agents in SLA-based control in resource management, especially when congestion occurs. The work demonstrates the ability of intelligent agents in improving and maintaining the quality of service to meet the required SLA as the congestion occurs. A particularly novel aspect of this work is the use of learning (here Case Based Reasoning) to predict the control strategies to be imposed. As the system environment changes, the most suitable policy will be implemented. When congestion occurs, the system either proposes the solution by recalling from experience (if the event is similar to what has been previously solved) or recalculates the solution from its knowledge (if the event is new). With this approach, the system performance will be monitored at all times and a suitable policy can be immediately applied as the system environment changes, resulting in maintaining the system quality of service.