Performance Enhancement in Mobile Opportunistic Networks Using Machine Learning

Abstract

Mobile Opportunistic Networks (OppNets) are a type of ad-hoc networks char- newlineacterised by intermittent connectivity and the opportunistic use of network resources. newlineThese networks, which rely on regular encounters between mobile devices to establish newlinecommunication, emerge as crucial solutions, particularly when infrastructure-based newlineapproaches encounter challenges. Our work starts with a feasibility study of the ex- newlineisting 4G-LTE and 5G infrastructures to align with user requirements and highlight newlinethe necessity of OppNets as an optimal solution. The novel routing protocols based newlineon Machine Learning (ML) algorithms, namely Q-learning, Double Q-learning, and newlineDeep Reinforcement Learning (DRL), are proposed to enhance OppNets performance. newlineConcurrently, we present a DRL-based offloading solution in Opportunistic Mobile newlineEdge Computing (OMEC). newlineWe begin by analysing the surge in wireless connections and the demand for newlinehigh-speed internet services. We analyse the period from 2019 to 2029 by consid- newlineering Smart City scenarios to evaluate the feasibility of existing 4G-LTE and 5G newlinenetworks. The exploration includes coverage, capacity, and cost considerations. The newlinestudy demonstrates the limitations of 4G-LTE while highlighting the promising capa- newlinebilities of 5G but with increased infrastructure costs. The insights derived from the newlinecoverage, capacity, and cost analysis show that 4G-LTE and 5G are not cost-effective newlinefor sensor data transmission in smart cities. Our findings show that Ahmedabad and newlineGandhinagar generate approximately 2702 terabytes and 704 terabytes of sensor data newlinedaily for various applications. Collecting such massive data using the 4G-LTE and newline5G networks requires significant cost. The study shows OppNets as a promising so- newlinelution for data collection, recognising the limitations of traditional data transmission newlinemethods. The findings highlight the complexities of managing large volumes of sensor newlinedata and also emphasise the potential benefits of OppNets in facilitating efficient data newlinecollection, especially in densely