Performance of Parallel Decentralized Sensor Network with Decision Feedback

Abstract

In most previous studies, distributed sensor networks are typically assumed to be memory less. In this paper, we consider a distributed sensor network with feedback at both the sensor level as well as at the fusion center. Specifically, we analyze (1) a local feedback system (LFS), where the most recent local decision is fed back to its corresponding local sensor; (2) a local and global feedback system 1 (LGFS1) where the most recent local decision is fed back to its corresponding local sensor and the most recent global decision is fed back to the fusion center, and (3) a local and global feedback system 2 (LGFS2), where the most recent global decision is fed back to all the local sensors and to the fusion center in addition to the local decision being fed back to its corresponding local sensor. For all the cases, we derive the decision rule and compare the global probability of error using simulations. We show that in an error-free channel, any form of feedback improves GPE performance relative to no feedback system. However, feeding the global decision back to local sensors not only drains resources but also provides the worst performance among feedback schemes