Compressed sensing based loss tomography using weighted ℓ1 minimization

Abstract

Network tomography allows the measurements of end-to-end to infer network internal links characteristics such as packet loss rates and delay. In this paper, we focus on the problem of estimating links loss rates, especially locating the congested links in network. Applying concepts of compressed sensing and Maximum A-Posteriori (MAP) estimation, we propose a new loss tomography scheme. Contrary to existing works that use ℓ1 minimization, the proposed scheme adopts weighted ℓ1 minimization as the implementation of compressed sensing, whose weights can be set wisely in order to improve tomography result. We exploit the temporal correlations of link losses and determine weights using the links prior congestion probabilities. The probabilities can be uniquely identified from multiple measurements by solving boolean algebra equations. We conduct a simulation performance analysis of loss tomography, demonstrating that higher estimation accuracy can be obtained through the proposed scheme.