Mobility identification and clustering in sparse mobile networks

Abstract

Non-uniform distributions of mobile nodes are the norm for a mobile network. Often, there can be concentration areas or grouping of nodes. Early work has explored these features to help message disseminations. However, a mobile network application can generate complex mixing mobility patterns that render these work less effective and efficient. In addition, many applications run with in a sparse mode, namely, the network may not be connected all the time. In this paper, we propose two entropy based metrics to identify the nodes with different mobility patterns and further use the metrics to accomplish clustering. Aiming at low-end devices which have no inputs of velocity and location, we employ neighbor information through hello messages and draw speed implication through neighbor change rates. The entropy based metrics are used in a clustering algorithm to find stable nodes as cluster heads. According to the the simulation results, two metrics, namely, speed entropy and relation entropy can be applied to distinguish active nodes from stable nodes in different group mixing configurations. The simulations also show that our new metric-based clustering algorithm generates more stable clusters.