Date of Award
Electrical and Computer Engineering
Engineering, Electronics and Electrical.
CC BY-NC-ND 4.0
The past few years have visualized tremendous research interests along with associated breakthroughs in Optical Fiber Communications (OFCs) and Artificial Neural Networks (ANNs). The first one has resulted in the elimination of the transmission-speed bottleneck of the pre-mid-1980 era, and hence cost-effective deployment of optical fibers in both short- and long-haul telecommunication networks. And the second has led to the development of massively parallel interconnection structure with simple, i.e., of the add-compare-select type, processing elements for solving large-scale, multi-criterion decision making problems in both on-line and off-line fashions at very high speed. To utilize effectively the vast fiber bandwidth, researchers have proposed the Wavelength Division Multiplexing (WDM) technique which reduces the impact of speed mis-match between nodal processing and link transmission. Further enhancements in performance can be achieved when neurally arbitrated multi-connected regular topology--constructed using WDM--is used as the operational connection structure of any arbitrary physical topology. This dissertation deals with the development of new and efficient techniques for evaluating architecture and packet routing strategy, and enhancing performance using ANNs, of the Manhattan Street Network (MSN) which is a special kind of two-connected regular mesh topology. MSN uses uni-directional links with adjacent channels carrying packets in opposite directions. The end nodes of every row and column are directly connected using wrap-around links. A new analytical technique for evaluating the architecture of arbitrarily large MSNs in a traffic-distribution-specific manner has been developed. The validity of this technique has been verified through computer simulation. A novel simple analytical technique for evaluating deflection routing in the MSNs has also been developed. The advantage of this technique lies in the fact that although it computes the Mean Packet Transfer Time (MPTT) by simply adding the Mean Inter-Node Distance (MIND), deflection and waiting penalties, it offers a reasonable degree of accuracy as ratified by simulation. Finally, it is shown that when Grossberg's ANN along with a fuzzy logic based pre-processor is used for packet route arbitration, congestion-free operation of a network can be easily achieved.Dept. of Electrical and Computer Engineering. Paper copy at Leddy Library: Theses & Major Papers - Basement, West Bldg. / Call Number: Thesis1992 .K458. Source: Dissertation Abstracts International, Volume: 54-05, Section: B, page: 2659. Advisers: Majid Ahmadi; Malayappan Shridhar. Thesis (Ph.D.)--University of Windsor (Canada), 1992.
Khasnabish, Bhumip., "High-performance supra-high-speed packet-switched networking using neural arbiters." (1992). Electronic Theses and Dissertations. 2305.