Sensor technology has grown tremendously in the recent decades. Using micro- and nano-technologies, sensor devices nowadays can be implemented in extremely small sizes. Networks of inexpensive sensors can now be easily deployed to areas in harsh conditions to provide continuous monitoring for environmental, military, and other scientific applications. As most sensor networks are utilized in hardly accessible areas such as a wilderness or battlefield, it is very costly or even infeasible to keep periodic maintenance of individual sensors. Therefore, the average battery life of the sensors essentially determines the life of the network itself. While the processing power consumption of a sensor can be reduced through the advance of computing technology, power consumed for communications can only be conserved by reducing the net amount of transferred information. Thus, research on advanced signal processing techniques for reducing communications power is essential for next-generation sensor networks.
Distributed source coding (DSC) has been considered as the technology to drastically reduce energy consumption due to communications in sensor networks. However, current limitations of DSC prevent its adoption in sensor networks. The objective of this research is to advance the state-of-the-art of DSC for next-generation sensor networks by remodeling DSC as graphical inference problems. The resulting technologies are expected to lead to significant reduction of power consumption for communications and thus prolonged life span of such networks. To accomplish the goal, the researchers study DSC design for sources with more than two terminals, adaptive DSC algorithms for sources with dynamic varying correlation, and variational inference of DSC beyond the belief propagation (BP) algorithm.
