Bio-Inspired and Machine Learning-Driven Multipath Routing Protocol for MANETs Using Predictive Link Analytics

Abstract

Mobile Ad Hoc Networks (MANETs) are decentralized, infrastructure-less wireless networks in which nodes function simultaneously as end devices and routers, enabling the dynamic establishment and maintenance of communication paths as required. One of the fundamental problems in MANETs is that links are prone to failure owing to, link breakage due to energy drain, node mobility, and changing environmental conditions. While the Ad hoc On-demand Multipath Distance Vector, AOMDV protocol, offers multipath fault tolerance, it is dependent upon hop count and is therefore also subject to link failures. In order to overcome this, link quality metrics, such as Expected Transmission Count, ETX, and bio-inspired optimization algorithms, such as Ant Colony Optimization, ACO have been investigated. Recent advancements in machine learning, particularly in the realm of predictive models such as Long Short-Term Memory networks and methods of ensemble learning like Random Forests, present promising options for link quality prediction that considers historical and real-time data in a more dynamic fashion. The goal of this proposal is to combine AOMDV, ACO, ETX, LSTM, Random Forests, and Predictive Analytics into one intelligent multipath routing protocols for MANETs.