Compressed Sensing in Multi-Hop Large-Scale Wireless Sensor Networks Based on Routing Topology Tomography

Data acquisition from a multi‐hop large‐ scale outdoor wireless sensor network (WSN) deploy‐ ment for environmental monitoring is full of chal‐ lenges. This is because the severe resource constraints on small battery‐operated motes (e.g., bandwidth, memory, power, and computing capacity), the big data acquisition volume from the large‐scale WSN, and the highly dynamic wireless link conditions in an outdoor communication environment. We present a novel com‐ pressed sensing approach which can recover the sens‐ ing data at the sink with high fidelity when very few data packets are collected, leading to a significant re‐ duction of the network transmissions and thus an ex‐ tension of the WSN lifetime. Interplaying with the dy‐ namic WSN routing topology, the proposed approach is efficient and simple to implement on the resource‐ constrained motes without a mote’s storing of any part of the random projection matrix, as opposed to other existing compressed sensing based schemes. We pro‐ pose a systematic method via machine learning to find a suitable representation basis, for any given WSN de‐ ployment and data field, which is both sparse and in‐ coherent with the random projection matrix in the compressed sensing for data acquisition. We validate our approach and evaluate its performance using a real‐world multi‐hop WSN testbed deployment in situ. The results demonstrate that our approach signif‐ icantly outperforms existing compressed sensing ap‐ proaches by reducing data recovery errors by an order of magnitude for the entire WSN observation field, while drastically reducing wireless communication costs at the same time. Key words—Compressed sensing, big data acquisi‐ tion, wireless sensor networks, deep learning, routing topology tomography, experiments, real world deploy‐ ment, validation.