DSpace@nitr >
National Institue of Technology- Rourkela >
Thesis (Doctor of Philosophy) >

Please use this identifier to cite or link to this item: http://hdl.handle.net/2080/690

Full metadata record

DC FieldValueLanguage
contributor.authorDas, Susmita-
contributor.authorSatapathy, J K (Guide)-
date.accessioned2008-05-23T10:05:38Z-
date.available2008-05-23T10:05:38Z-
date.issued2004-
identifier.citationAdaptive Equalisation of Communication Channels Using ANN Techniques, Thesis submitted in partial fulfillment of the requirements for the award of the Doctor of Philosophy in Electrical Engineering, Submitted to National Institute of Technology, Rourkelaen
identifier.urihttp://hdl.handle.net/2080/690-
descriptionCopyright for the thesis belongs to National Institute of Technology Rourkelaen
description.abstractChannel equalisation is a process of compensating the disruptive effects caused mainly by Inter Symbol Interference in a band-limited channel and plays a vital role for enabling higher data rate in digital communication. The development of new training algorithms, structures and the selection of the design parameters for equalisers are active fields of research which are exploiting the benefits of different signal processing techniques. Designing efficient equalisers based on low structural complexity, is also an area of much interest keeping in view of real-time implementation issue. However, it has been widely reported that optimal performance can only be realised using nonlinear equalisers. As Artificial Neural Networks are inherently nonlinear processing elements and possess capabilities of universal approximation and pattern classification, these are well suited for developing high performance adaptive equalisers. This proposed work has significantly contributed to the development of novel equaliser structures with reduced structural complexity in the neural network paradigm based on both the feedforward neural network (FNN) and the recurrent neural network (RNN) topologies. Various innovative techniques like hierarchical knowledge reinforcement, genetic evolutionary concept, transform domain based approach and sigmoid slope tuning using fuzzy logic approach have been incorporated into an FNN framework to design highly efficient equaliser structures. Subsequently, novel hybrid configurations using cascaded modules of RNN and FNN have also been proposed in this thesis work. Further, suitable modifications in the Back-Propagation and Real-Time- Recurrent-Learning algorithms have been incorporated to update the connection weights of the proposed structures. Significant performance improvement over the conventional FNN and RNN based equalisers, faster adaptation and ease of implementation in realtime applications are the major advantages of the proposed neural equalisers. Exhaustive simulation studies carried out on various linear and nonlinear channels verify the efficacy of the proposed neural equalisers. Further, all the proposed FNN based equalisers are of decision feedback type as inclusion of this technique significantly improves the performance alongwith a considerable reduction in the structural complexity. A proper selection of feedforward order, decision delay and feedback order is a challenging task in such equalisers as these key design parameters play a crucial role for an impressive performance. A detailed study of various factors influencing the bit error rate performance of optimal Bayesian equaliser has been undertaken in the present work. This study has given an insight for proposing of a novel approach in the parameter selection issue, which can eliminate the use of cumbersome procedure of determining these design parameters from graphical analysis. Thus a major breakthrough has been achieved in successfully evaluating these parameters of the equaliser structure. The new methodology and its logical interpretation that led to the development of some empirical relationships have emerged as a powerful tool for selecting the key design parameters directly from the channel characteristics.en
format.extent5463678 bytes-
format.mimetypeapplication/pdf-
language.isoen-
publisherNational Institute of Technology, Rourkelaen
subjectAdaptive Equalisationen
subjectInter Symbol Interferenceen
subjectCommunication Channelen
subjectDecision Feedback Equaliseren
subjectNonlinear Equalisersen
subjectFeedforward Neural Networken
titleAdaptive Equalisation of Communication Channels Using ANN Techniquesen
typeThesisen
Appears in Collections:Thesis (Doctor of Philosophy)

Files in This Item:

File Description SizeFormat
Final-thesis-susmita dada.pdf5335KbAdobe PDFView/Open

Show simple item record

All items in DSpace are protected by copyright, with all rights reserved.

 

Powered by DSpace Feedback