Comparative Study of Distributed Estimation Precision by Average Consensus Weight Models

Abstract

Distributed algorithms for an aggregate function estimation are an important complement of many real-life applications based on wireless sensor networks. Achieving a high precision of an estimation in a shorter time can optimize the overall energy consumption. Therefore, the choice of a proper distributed algorithm is an important part of an application design. In this study, we focus our attention on the average consensus algorithm and evaluate six weight models appropriate for the implementation into real-life applications. Our aim is to find the most suitable model in terms of the estimation precision in various phases of the algorithm. We examine the deviation of the least precise estimate over iterations for a Gaussian, a Uniform and a Bernoulli distribution of the initial states in strongly and weakly connected networks with a randomly generated topology. We examine which model is the most and the least precise in various phases. Based on these findings, we determine the most suitable model for real-life applications.Distributed algorithms for an aggregate function estimation are an important complement of many real-life applications based on wireless sensor networks. Achieving a high precision of an estimation in a shorter time can optimize the overall energy consumption. Therefore, the choice of a proper distributed algorithm is an important part of an application design. In this study, we focus our attention on the average consensus algorithm and evaluate six weight models appropriate for the implementation into real-life applications. Our aim is to find the most suitable model in terms of the estimation precision in various phases of the algorithm. We examine the deviation of the least precise estimate over iterations for a Gaussian, a Uniform and a Bernoulli distribution of the initial states in strongly and weakly connected networks with a randomly generated topology. We examine which model is the most and the least precise in various phases. Based on these findings, we determine the most suitable model for real-life applications.