Modeling and simulation of wireless link quality (ETT) through principal component analysis of trace data

Abstract

Principal Component Analysis (PCA) is a powerful method in data analysis. In this paper, we employ the capabilities of PCA combined with statistical fits to trace data to develop tractable models that can be used to simulate the quality of links in wireless mesh networks using the expected transmission time (ETT) metric. We apply principal component analysis to ETT traces from a wireless mesh network to determine what features in the ETT traces are important and to extract any meaningful relationships therein. We demonstrate that PCA can be used to efficiently approximate large volumes of ETT values. In particular, the ETT trace for each link can be expressed as a combination of two basis vectors -one fairly stable and the other containing the variations in time. We also show how the extracted features can be employed to simulate ETT for a given network topology with and without known ETT trace data.