A Feedback-Enhanced Learning Approach for Routing in WSN

A new modality for sensor networks emerges when considering multiple, distributed base stations that collect data from sensors. This scenario reverses the typical multiple-source, single-sink scenario, and requires new techniques to efficiently send data from single-sources to multiple-sinks. While an offline approach with full topology information can build the optimal data forwarding tree, the challenge we address here is to optimize data forwarding with only information exchanged among one-hop neighbors. The novelty of our approach lies in the use of an iterative learning technique that explores alternative routes by locally sharing feedback regarding route fitness. This paper presents our approach as well as an evaluation showing that the learned paths lead to increases in network lifetime of up to 50% over an approach without learning.