QLA-MAODV: A Q-learning adaptive multicast routing protocol for mobile ad-hoc networks

Abstract

Mobile Ad-hoc Networks face challenges in achieving efficient multicasting due to dynamic topology changes and unreliable links. Existing multicast approaches either suffer from low packet delivery ratio or high overhead. These approaches rely on simple metrics like hop count to find the optimal path to the destination. Once the path is selected, all packets are sent over the same path as long as it remains available. However, a path that is deemed optimal at a specific instance of time may not retain its optimality at a subsequent moment due to node mobility. Moreover, using a metric like hop count that does not consider link quality can lead to poor packet delivery ratio, as it can favor an unreliable path over a reliable one just because it is the shortest. To tackle these concerns, a Q-Learning Adaptive Multicast Ad-hoc On-Demand Distance Vector routing protocol is proposed. It is an adaptive and bandwidth-efficient solution that utilizes link reliability as a routing metric instead of hop count, aiming to build a more stable multicast tree. By leveraging Q-learning principles, the proposed protocol continuously updates path costs to detect any deterioration. Additionally, the protocol dynamically explores the network using periodic group hello messages, enabling the identification of alternative paths and proactively switches to them if path costs deteriorate. Simulations conducted in Network Simulator 3 demonstrate the superiority of the proposed protocol over the traditional Multicast Ad-hoc On-Demand Distance Vector protocol. Furthermore, it outperforms a modified version, called Multicast Ad-hoc On-Demand Distance Vector-Route Reliability, that uses link reliability as a metric, demonstrating enhanced packet delivery ratio and reduced multicast-related overhead.