Learning Wireless Networks' Topologies Using Asymmetric Granger Causality

摘要

Sharing spectrum with a communicating incumbent user (IU) network requiresavoiding interference to IU receivers. But since receivers are passive when inthe receive mode and cannot be detected, the network topology can be used topredict the potential receivers of a currently active transmitter. For thispurpose, this paper proposes a method to detect the directed links between IUsof time multiplexing communication networks from their transmission start andend times. It models the response mechanism of commonly used communicationprotocols using Granger causality: the probability of an IU starting atransmission after another IU's transmission ends increases if the former is areceiver of the latter. This paper proposes a non-parametric test statistic fordetecting such behavior. To help differentiate between a response and theopportunistic access of available spectrum, the same test statistic is used toestimate the response time of each link. The causal structure of the responseis studied through a discrete time Markov chain that abstracts the IUs' mediumaccess protocol and focuses on the response time and response probability of 2IUs. Through NS-3 simulations, it is shown that the proposed algorithmoutperforms existing methods in accurately learning the topologies ofinfrastructure-based networks and that it can infer the directed data flow inad hoc networks with finer time resolution than an existing method.