Regression Model Based Consensus for Clock Synchronization of Wireless Sensor Network

Abstract

Consensus is a process of reaching a state agreement throughout a distributed network under the condition of only exchanging information between nearby network nodes. In a finite dimensional state space averaging is an effective and scalable method to achieve consensus. In this work the principle of averaging is extended to a model-based framework for the purpose of synchronising the parameters of a linear model associated with each node. Each node updates the model parameters upon receiving a set of linear regression results sent by the neighboring nodes. It is shown that this formulation of regression-based averaging scheme is capable of synchronising all model parameters throughout the network. The regression-based model consensus strategy is then applied to the clock synchronisation problem of a wireless sensor network (WSN), where both clock drift and offset uncertainties are considered. Simulation verifies that the clock readings throughout a connected network can be synchronised by the regression-based synchronisation strategy.