Distributed self-tuning of sensor networks

Abstract

This work is motivated by the need for an ad hoc sensor network to autonomously optimise its performance for given task objectives and constraints. Arguing that communication is the main bottleneck for distributed computation in a sensor network we formulate two approaches for optimisation of computing rates. The first is a team problem for maximising the minimum communication throughput of sensors and the second is a game problem in which cost for each sensor is a measure of its communication time with its neighbours. We investigate adaptive algorithms using which sensors can tune to the optimal channel attempt rates in a distributed fashion. For the team problem, the adaptive scheme is a stochastic gradient algorithm derived from the augmented Lagrangian formulation of the optimisation problem. The game formulation not only leads to an explicit characterisation of the Nash equilibrium but also to a simple iterative scheme by which sensors can learn the equilibrium attempt probabilities using only the estimates of transmission and reception times from their local measurements. Our approach is promising and should be seen as a step towards developing optimally self-organising architectures for sensor networks.