Compressive Sensing-based Data Loss And Reconstruction In Wireless Sensor Networks

Many basic scientifc works require the analysis of environment data, which areobtained from wireless sensor networks deployed on target areas (indoor, forest andsea). Generally, the complexity of data is a signifcant requirement and of utmost im-portance as well as the accuracy because these works heavily rely on the completesensory data. However, data loss in wireless sensor networks is common, thus datareconstruction is required. In the data gathering process, data reconstruction is a basicoperation.Because of general existence, several works have studied the missing value prob-lem in varieties of applications with targeted methods proposed. Yet a wireless sensornetwork has its special characteristics due to noise, collision, unreliable link, and unex-pected damage, i.e., high loss rate and special loss patterns, which lead to the fact thatexisting interpolation methods fail to provide satisfactory performance. In particular,their accuracy is low on the data reconstruction.To address this problem, this paper proposes a novel approach called ESTI-CSbased on compressive sensing to reconstruct the massive missing data.Firstly, we analyze the real sensory data from Intel Indoor, GreenOrbs, and OceanSense projects. They all1) share special loss patterns such as element random loss,block random loss, element frequent loss in row and successive elements loss in rowand2) exhibit the features of low-rank structure, spatial similarity, temporal stabilityand multi-attribute correlation.Secondly, motivated by these observations, we then develop an environmentalspace time improved compressive sensing (ESTI-CS) algorithm with a multi-attributeassistant (MAA) component to optimize the missing data estimation. Finally, the extensive simulation results on real sensory data sets show that theproposed approach signifcantly outperforms existing solutions in terms of reconstruc-tion accuracy, whatever the data loss follows random pattern or the special patterns.