Exploiting temporal and spatial correlation in wireless sensor networks

Abstract

Collecting data continuously from Wireless Sensor Networks (WSNs) with limited power and bandwidth is a challenging problem. Such networks have potential utility in a wide range of disciplines such as medical, industrial, environmental, and military applications. For long-term monitoring and surveillance applications, the objective is often times to cover as large an area as possible while still acquiring high-resolution information about the sensed environment. The main challenges involve energy-efficiency and scalability of the techniques used to acquire the substantial amount of data from the sensor network. Addressing these challenges, we propose a method that effectively exploits both spatial and temporal correlation inherent in sensor network data. The technique combines recent advances in Compressed Sensing (CS) theory, data compression, and a novel random access communication protocol. Sensor nodes perform in-situ temporal compression and transmit their compressed data over a random access channel to a fusion center (FC) to recover the field. If packets collide at the FC, they are simply discarded. A CS recovery algorithm, executed at the FC, allows the entire field to be recovered from the so-obtained observations. This method of spatio-temporal compression is decentralized and requires minimal feedback from the FC. Furthermore, the method does not require synchronized sensors and is robust to node failures, packet losses, and sensor noise. This approach is demonstrated on synthetic climate measurement data and seismic reflection data. Compared to a conventional time-division access without data compression, as well as to random access with CS but without temporal compression, the proposed method significantly improves energy and bandwidth efficiency.