A feedback-enhanced learning approach for routing in WSN

  • Egorova-Förster, Anna Facoltà di scienze informatiche, Università della Svizzera italiana, Svizzera
  • Murphy, Amy Facoltà di scienze informatiche, Università della Svizzera italiana, Svizzera

11 p.

English Much research in sensor networks focuses on optimizing traffic originating at multiple sources destined for a single, base station sink. Our work reverses this assumption, targeting scenarios where individual sensor data is sent to multiple destinations. In this case, the data path that produces the least network cost is unlikely to overlap completely with any of the optimal routes between the individual pairs of source/destination nodes. If the entire topology is known, an offline approach can likely find this minimum path. However this is an unrealistic assumption. Instead, our approach uses only local information and converges toward optimal. The novelty of our approach is a technique for actively exploring alternate data routes, sharing feedback regarding route fitness, and learning better routes. While non-optimal choices are made during the discovery phase, the resulting, learned path has lower cost than the initial path. Further, our protocol identifies multiple paths with equal cost, providing additional opportunities for saving energy by switching among alternate routes throughout the lifetime of the application. This paper describes our feedbackbased protocol, shows simulation results demonstrating its benefits and explores the future opportunities of the learning technique presented.
  • English
Computer science and technology
License undefined
  • RERO DOC 10713
  • ARK ark:/12658/srd1317895
Persistent URL

Document views: 30 File downloads:
  • ITR0603.pdf: 59