Description:

Sensor networks can be used for detection tasks. If the sensor nodes are battery driven, they are subject to severe power constraints. Wireless communication is using most of the power of such sensor nodes.

Detection tasks (detecting a certain state of the environment) can be modelled using statistics. In the past, various distributed detection problems have been formulated and studied. However, in most of these settings, all the nodes are constantly communicating, and people usually assume that there is a special sink node, the fusion center. In ad hoc sensor networks, however, we would prefer to have nodes that only communicate when necessary, without using a fusion center.

Message passing algorithms are very simple algorithms in networks, where one or several messages are passed from node to node while being modified. In [1], a message-passing algorithm is used to implement a distributed learning scheme.

The aim of this project is to apply the ideas of message-passing to the problem of ad-hoc detection.

Objective:

Find schemes for detecting events in an ad-hoc manner using message-passing techniques.

Prerequisites:

• Solid background in probability and statistics
• Hypothesis testing / detection
• linear algebra

References:
[1] J.B. Predd, S.R. Kulkarni, and H.V. Poor, “Distributed Kernel Regression: An Algorithm for Training Collaboratively”, arxiv.org, 2006

Contact:

Etienne Perron (LICOS) * Email: etienne.perron@epfl.ch * Office: INR-033 * Tel: 36457

Supervisor: Prof. Suhas Diggavi

back to master semester projects