DECA: Recovering fields of physical quantities from incomplete sensory data

Although wireless sensor networks (WSNs) are powerful in monitoring physical events, the data collected from a WSN are almost always incomplete if the surveyed physical event spreads over a wide area. The reason for this incompleteness is twofold: i) insufficient network coverage and ii) data aggregation for energy saving. Whereas the existing recovery schemes only tackle the second aspect, we develop Dual-lEvel Compressed Aggregation (DECA) as a novel framework to address both aspects. Specifically, DECA allows a high fidelity recovery of a widespread event, under the situations that the WSN only sparsely covers the event area and that an in-network data aggregation is applied for traffic reduction. Exploiting both the low-rank nature of real-world events and the redundancy in sensory data, DECA combines matrix completion with a fine-tuned compressed sensing technique to conduct a dual-level reconstruction process. We demonstrate that DECA can recover a widespread event with less than 5% of the data (with respect to the dimension of the event) being collected. Performance evaluation based on both synthetic and real data sets confirms the recovery fidelity and energy efficiency of our DECA framework.