Information Processing in Sensor Networks

In this paper we investigate the capability of large-scale sensor networks to measure and transport a two-dimensional field. We consider a data-gathering wireless sensor network in which densely deployed sensors take periodic samples of the sensed field, and then scalar quantize, encode and transmit them to a single receiver/central controller where snapshot images of the sensed field are reconstructed. The quality of the reconstructed field is limited by the ability of the encoder to compress the data to a rate less than the single-receiver transport capacity of the network. Subject to a constraint on the quality of the reconstructed field, we are interested in how fast data can be collected (or equivalently how closely in time these snapshots can be taken) due to the limitation just mentioned. As the sensor density increases to infinity, more sensors send data to the central controller. However, the data is more correlated, and the encoder can do more compression. The question is: Can the encoder compress sufficiently to meet the limit imposed by the transport capacity? Alternatively, how long does it take to transport one snapshot? We show that as the density increases to infinity, the total number of bits required to attain a given quality also increases to infinity under any compression scheme. At the same time, the single-receiver transport capacity of the network remains constant as the density increases. We therefore conclude that for the given scenario, even though the correlation between sensor data increases as the density increases, any data compression scheme is insufficient to transport the required amount of data for the given quality. Equivalently, the amount of time it takes to transport one snapshot goes to infinity.