Multi-criteria optimization of wireless connectivity over sparse networks

Abstract

Opportunistic networking is at the basis of cyber-physical Mobile Networks in Proximity (MNP), through its unique perspective over mobility and the incorporation of socio-inspired networking algorithms. However, results in the field are mostly theoretical, proven to account for stricter hit rate and latency requirements in specific environments. They generally assume that two devices being in proximity automatically see one-another, an assumption which might not stand under real-world conditions (Bluetooth assumes a peering session and close-proximity, WiFi Direct implementations are different between manufacturers, etc.).Our previous studies in the area show that WiFi is still the most feasible media for opportunistic contacts. WiFi-enabled devices, with out-of-the-box networking capabilities, can connect in an ad-hoc opportunistic network, over wireless routers, and thus support a cyber-physical infrastructure for opportunistically spreading information.In this article, we propose a machine learning algorithm that aims to increase the number of contacts between mobile nodes by using a smarter WiFi access point selection heuristic. The algorithm is based on properly balancing signal strength, latency, bandwidth, and, most importantly, the number of friends predicted to connect to the respective access point. We show through simulations based on real-life tracing data-sets that our proposed solution not only increases the likelihood of opportunistic contacts, but it also evenly distributes social subgraphs of users over wireless networks while improving the overall hit rate.