Data Collection And Recovery Via Compressed Sensing In Wireless Sensor Networks

As the core technology of the Internet of Things,wireless sensor network(WSN)has received more and more attention because of its application flexibility and information-aware effectiveness.It is difficult to meet transmission and processing of high-density mass of information as hardware resources and energy are limited in WSN,which becomes the major technical problem that restricts the large-scale application of WSN.In recent years,compression sensing(CS)theory has gained extensive attention and research.Compressed sensing combines the sampling and compression processes to discard redundant information in sparse or compressible signals,thereby reducing the signal sampling frequency and saving storage and transmission costs.The emergence of compression sensing theory provides a new technology solution for mass data acquisition,transmission,storage and node endurance of wireless sensor networks,and accelerates the development of Internet of Things.In this paper,based on the analysis of the data characteristics of wireless sensor networks,we use the key technology of compression sensing theory as the main research content and dedicated to applied CS theory to WSN.This paper has launched a number of research work:1.Environmental data mining in WSN: Sparse signals in nature are rare,but most signals can be sparsely represented on a domain.At the same time,the environmental data(temperature,light,humidity)collected by the sensing node have strong temporal correlation and spatial correlation.The monitoring data of three WSNs that have been put into use are chosen as the research object.The simulation results show that the selected environment data has strong sparse characteristics and spatio-temporal correlation properties.2.Measurement matrix: Based on the Restricted Isometry Property(RIP),the Gaussian random matrix,the Bernoulli random matrix,the partial orthogonal matrix,the Toeplitz matrix and the loop matrix are studied.Based on this,the sparse two-dimensional matrix and sparse random matrix are introduced,which are easy to implement in hardware and have low storage space requirements.The simulation results show that the recovery error of the Partial orthogonal matrix,the loop measurement matrix and the Toeplitz matrix reduce in turn.The Bernoulli random matrix,the Gaussian random matrix,the sparse two-dimensional matrix and sparse random matrix have similar recovery performance and are better than other measurement matrix.These measurement matrices can realize reconstruction of the signal with high accuracy when the number of measurements M satisfies certain conditions.3.WSN topology optimization and data aggregation scheme: When the WSN scale is large,it will increase the amount of computation,storage and transmission of data in the network,and seriously affect the performance of the network.In this paper,the topology of WSN is optimized,that is,the network is segmented by the diffusion wavelet,and the data is aggregated independently in each subnet.The data is transmitted directly to sink receiver by the selected central node.The simulation results confirm the feasibility of this scheme.At the same time,a new scheme of data aggregation in subnet is proposed based on compression sensing theory.4.Reconstruction algorithm: Two kinds of traditional reconstruction algorithms based on compression sensing are studied: convex relaxation algorithm and greedy algorithm.A novel reconstruction algorithm named BCS-STGR(BCS-STGR)based on spatiotemporal correlation is proposed.This algorithm combines the topology optimization scheme of WSN,Data aggregation scheme,and the temporal and spatial correlation characteristics of environment data in WSN.It has superior performance in energy utilization and reconstruction precision.Typically,the normalized mean absolute error of our recovery scheme is less than 5%.Furthermore,the energy consumption is reduced more than 50% against plain CS.