Spatially Dependent Loss Tomography for Multihop Wireless Networks

Network tomography is an inference method for network internal characteristics from end-to-end measurements. In this paper, we study network tomography for estimating packet loss rates at nodes in wireless multihop networks with spatially dependent channels, where neighboring nodes tend to have packet loss rates of comparable magnitude. We propose a network tomography scheme based on graph Fourier transform (GET). Instead of estimating the packet loss rates directly, the proposed scheme estimates the GET of a node state vector whose components are given in terms of packet loss rates at. nodes. In spatially dependent channels, the GFT of the node state vector has a few dominant components. Based on this property, we propose a new scheme with compressed sensing that can acquire a sparse vector in a linear inverse problem. With simulation experiments, we evaluate the performance of the proposed scheme and show that it captures the spatial dependency of packet loss rates at nodes.