Denoising and error correction in wireless sensor networks

Measurements of wireless sensor networks (WSNs) are often polluted by random measurement noises and corrupted by unpredictable sensory reading errors. For a typical field monitoring scenario, this paper considers to correct sensory reading errors and recover the monitoring field, subject to measurement noises. The key factor to enable successful de-noising and error correction is that the monitoring field can often be represented by a sparse signal vector; signal sparsity makes sensory readings of WSNs to be redundant, which offers inherent fault tolerance against measurement noises and sensory reading errors. Specifically, this paper focuses on two approaches: one is the l regularized least squares (LRLS) approach which was proposed to handle noises in statistical signal processing, and the other is the cross-and-bouquet (CAB) approach which was proposed to correct errors in computer vision. Discussion of their relationship reveals that the CAB approach is robust to measurement noises, while the two approaches have similar performance when sensory reading errors are dense. Extensive simulation results validate the effectiveness of the two approaches.