Ant learning with distributed geographical localization of knowledge for adaptive routing control in ad-hoc networks

Michał Adam Kudelski

Abstract

The problem of adaptive routing control in ad-hoc networks is addressed in this dissertation from the perspective of ant learning mechanisms. The biggest challenge for a learning mechanism is the high rate of topology changes in the network. Consequently, the knowledge of connections in the network needs to be continuously acquired from the environment. Ant routing algorithms provide such functionality, yet the knowledge gathered by ant agents is strictly linked to individual nodes and has to be reorganized each time the network topology changes. As a remedy, we go back to the roots of ant algorithms and we introduce an innovative knowledge management scheme, namely the distributed geographical localization of knowledge. In the proposed approach, the geographic information is utilized to connect the knowledge gathered by ant agents with geographical locations rather than with individual nodes, in a similar manner as for biological ants a pheromone is located on the ground. Locations are represented by geographical cells that provide a generalized model of connections in ad-hoc networks. The knowledge is distributed among the nodes and it flows between the nodes as they change their locations. An extensive experimental study proved that the generalized model of connections defined on the locations level is more robust to dynamic topology changes than the classical model defined on the nodes level. Consequently, the adaptation capabilities of the underlying learning mechanism are increased and the overall routing performance of the network is improved, both in comparison to the mechanisms that do not use learning and to the learning mechanisms without geographical cells. The improvement has been analyzed in two different communication environments, namely in typical telecommunication ad-hoc networks and in ad-hoc networks of mobile robots.
Diploma typeDoctor of Philosophy
Author Michał Adam Kudelski (FEIT / AK)
Michał Adam Kudelski,,
- The Institute of Control and Computation Engineering
Title in EnglishAnt learning with distributed geographical localization of knowledge for adaptive routing control in ad-hoc networks
Languageen angielski
Certifying UnitFaculty of Electronics and Information Technology (FEIT)
Disciplineinformation science / (technology domain) / (technological sciences)
Defense Date21-12-2010
End date21-12-2010
Supervisor Andrzej Pacut (FEIT / AK)
Andrzej Pacut,,
- The Institute of Control and Computation Engineering

Internal reviewers Jarosław Arabas (FEIT / PE)
Jarosław Arabas,,
- The Institute of Electronic Systems
External reviewers Marek Amanowicz
Marek Amanowicz,,
-
Pages153
Keywords in Englishant algorithms, adaptive routing, ad-hoc networks, ant routing, knowledge management
Abstract in EnglishThe problem of adaptive routing control in ad-hoc networks is addressed in this dissertation from the perspective of ant learning mechanisms. The biggest challenge for a learning mechanism is the high rate of topology changes in the network. Consequently, the knowledge of connections in the network needs to be continuously acquired from the environment. Ant routing algorithms provide such functionality, yet the knowledge gathered by ant agents is strictly linked to individual nodes and has to be reorganized each time the network topology changes. As a remedy, we go back to the roots of ant algorithms and we introduce an innovative knowledge management scheme, namely the distributed geographical localization of knowledge. In the proposed approach, the geographic information is utilized to connect the knowledge gathered by ant agents with geographical locations rather than with individual nodes, in a similar manner as for biological ants a pheromone is located on the ground. Locations are represented by geographical cells that provide a generalized model of connections in ad-hoc networks. The knowledge is distributed among the nodes and it flows between the nodes as they change their locations. An extensive experimental study proved that the generalized model of connections defined on the locations level is more robust to dynamic topology changes than the classical model defined on the nodes level. Consequently, the adaptation capabilities of the underlying learning mechanism are increased and the overall routing performance of the network is improved, both in comparison to the mechanisms that do not use learning and to the learning mechanisms without geographical cells. The improvement has been analyzed in two different communication environments, namely in typical telecommunication ad-hoc networks and in ad-hoc networks of mobile robots.
Thesis file
Kudelski.pdf 2.17 MB

Get link to the record

Back
Confirmation
Are you sure?