Distributed Compression for Partially Cooperating Sensors and Gaussian Relevant Signals

Abstract:
This paper considers the well known Chief Executive Officer (CEO) scenario where each sensor in a network observes noisy versions of the same signal of interest. These measurements have to be locally compressed in order to forward them over capacity limited links to a common receiver. To improve the performance in an information theoretic measure the non-cooperative CEO scenario is extended allowing partial cooperation among sensors. Therefore, each sensor not only forwards information to the common receiver, but also to a neighboring sensor via capacity limited links. This extension of the non-cooperative CEO scenario changes the statistical dependencies of the involved random variables. A heuristic approach based on the algorithmic solution of the non-cooperative CEO problem is considered, which optimizes the quantizer of all sensors in a greedy manner. Simulations demonstrate that exchanging instantaneous side-information during runtime can increase the overall performance of the CEO problem in terms of information theoretic measures. However, there still remains a gap to a fully cooperative CEO scenario, where each sensor has access to the measurements of all sensors.