Evaluating the gap between compressive sensing and distributed source coding in WSN

Abstract

Despite the large body of theoretical research available on compression algorithms for wireless sensor networks (WSNs), only recently have researchers started to design and analyze practical distributed compression techniques. Also, approaches belonging to different fields such as signal processing (e.g., discrete Fourier transforms and compressive sensing) or information theory (e.g., distributed source coding) and networking are seldom evaluated against one another. In the present contribution, we consider practical lossy compression schemes that rely on different techniques, such as the exploitation of the temporal and spatial dynamics of the signal as well as recent algorithms based on Compressive Sensing (CS). These techniques are adapted so as to be efficiently applied, within the same data collection framework, to a distributed WSN. Hence, we carry out a comparative performance analysis of these schemes, assessing their performance in terms of reconstruction error vs energy requirements. From this, several interesting observations are derived, which allow the identification of the best performing algorithm(s) as a function of the spatio-temporal characteristics of the signal. For CS, we assess the impact of the node selection scheme (scheduling) and gauge its performance gap with respect to an idealized CS scheme where the signal covariance matrix is perfectly known at the reconstruction point. We finally identify areas that need improvement, especially for the enhancement of CS-based compression.