Data Compression Based On Compressed Sensing In Wireless Sensor Networks

Wireless Sensor Network (WSN) is responsible for sensing, collecting,processing and monitoring environmental data, but it is easily limited by resources.The newly emerging Compressed Sensing theory holds that sparse signals can beexactly reconstructed with high probability from a small amount of non-adaptivelinear measurement through optimization. We design a data compression methodthat depends only on the structure and content of the signal, rather than thebandwidth of the signal in order to make up for the shortage of WSN.This dissertation describes the theory of Compressed Sensing in detail,researches on traditional WSN data compression and network coding method. Wedesign a linear network coding scheme using sparse random projections withCompressed Sensing. For high throughput and simple coding characteristics oflinear network coding, we select Bernoulli/Rademacher random matrix as themeasurement matrix under compressed sensing. This matrix can fulfill thecondition of RIP because of the asymptotic normality. Thus the scheme can solvesthe problem of the address header congestion which could occur in the linearnetwork coding. We adapt data packet format to meet the requirements ofCompressed Sensing. The receiver reconstructs the original data by convexoptimization when collecting enough data packets. It solves the "All or Nothing"problem appearing in linear network coding.Simulation results show that this system not only improves the efficiency ofWSN communication by reducing packets to30%number of nodes, but alsoreduces the system energy consumption under the error requirement. Comparingwith other WSN data compression, the proposed algorithm is simple and has goodcompression effect, especially useful for limited resources like WSN withoutstringent accuracy requirements.