An on demand reinforcement learning approach for mobile ad hoc networks (MANETs) routing protocol

Abstract

<P>A NOVEL DYNAMIC ON DEMAND ROUTING SCHEME FOR MANETS, WHICH USES THE ON DEMAND ROUTING CAPABILITY OF AD HOC ON-DEMAND DISTANCE VECTOR (AODV) ROUTING PROTOCOL AS THE BASE POLICY FOR Q-LEARNING ROUTING PROTOCOL. DUE TO NODE MOBILITY AND POWER LIMITATION, THE NETWORK TOPOLOGY CHANGES FREQUENTLY SUCH THAT THE ROUTE FOUND BY AODV MAY NO LONGER BE OPTIMAL OR IN EXISTENCE. THE PROPOSED ON-DEMAND Q-LEARNING ROUTING PROTOCOL (OQRP) IS ABLE TO DYNAMICALLY DISCOVER AND PROVIDE MULTIPLE ALTERNATIVE ROUTES TO THE DESTINATION WITHOUT PRODUCING ADDITIONAL CONTROL MESSAGES TO AODV. IN ADDITION, IT USES Q-LEARNING TO CHOOSE THE OPTIMAL ROUTE TO THE DESTINATION DYNAMICALLY WHENEVER THE TOPOLOGY CHANGES. THESE TWO PROPERTIES ARE CURRENTLY NOT FOUND IN THE OTHER MODIFIED VERSIONS OF AODV PROTOCOL. </P><P>IN THIS THESIS, IT IS SHOWN THAT THE PROPOSED OQRP REDUCES END-TO-END DELAY, NUMBER OF ROUTE REQUEST, OVERHEAD AND INCREASES CONNECTIVITY WHEN COMPARED TO AODV AND MULTIPLE ROUTES AODV PROTOCOLS. </P>