By Topic

Online supervised incremental learning of link quality estimates in wireless networks

Sign In

Cookies must be enabled to login.After enabling cookies , please use refresh or reload or ctrl+f5 on the browser for the login options.

Formats Non-Member Member
$33 $13
Learn how you can qualify for the best price for this item!
Become an IEEE Member or Subscribe to
IEEE Xplore for exclusive pricing!
close button

puzzle piece

IEEE membership options for an individual and IEEE Xplore subscriptions for an organization offer the most affordable access to essential journal articles, conference papers, standards, eBooks, and eLearning courses.

Learn more about:

IEEE membership

IEEE Xplore subscriptions

6 Author(s)
Gianni A. Di Caro ; Dalle Molle Inst. for Artificial Intell. (IDSIA), Manno-Lugano, Switzerland ; Michal Kudelski ; Eduardo Feo Flushing ; Jawad Nagi
more authors

We address the problem of link quality estimation in wireless networks and propose a distributed online protocol based on supervised incremental learning. We first identify a set of easily measurable network features that jointly determine the quality of a wireless link. These features summarize the local network configuration which is associated to the link, and include signal strengths, topology, and local traffic characteristics of the two end-points of the link and of their neighbors. At every node and for every wireless link, the protocol passively gathers measurements to quantify the current value of the network features and to assess the related link quality value according to a selected metric (the packet reception ratio, in our case). A node uses these measurements as labeled training samples for the incremental and supervised learning of the regression mapping from a local network configuration to a link quality estimate. The learned regression model can then be used by network protocols to derive in real-time robust estimates of link qualities after measuring the current local configuration. Nodes can also cooperate by exchanging training samples, speeding up in this way the overall learning process. This results particularly useful when the local network configurations are continually changing because of mobility and/or varying traffic patterns. We validate the protocol both in simulation, considering mobile ad hoc networks, and on a real sensor network testbed of 139 nodes. We also study the application of the prediction model in the context of routing, showing its efficacy improving the performance of the OLSR ad-hoc routing protocol.

Published in:

Ad Hoc Networking Workshop (MED-HOC-NET), 2013 12th Annual Mediterranean

Date of Conference:

24-26 June 2013