Representing Communication and Learning in Femtocell Pilot Power Control Algorithms

Abstract

The overall goal of evolving algorithms for femtocellsis to create a continuous on-line evolution of thefemtocell pilot power control algorithm to optimisetheir coverage. Two aspects of intelligence are usedfor increasing the complexity of the input and thebehaviour, communication and learning. In this initialstudy we investigate how to evolve more complexbehaviour in decentralised control algorithms bychanging the representation of communication andlearning. The communication is addressed by allowingthe femtocell to identify its neighbours and take thevalues of its neighbours into account when makingdecisions regarding the increase or decrease of pilotpower. Learning is considered in two variants: the useof input parameters and the implementation of abuilt-in reinforcement procedure. The reinforcementallows learning during the simulation in addition tothe execution of fixed commands. The experimentscompare the new representation in the form of differentterminal symbols in a grammar. The results show thatthere are differences between the communication andlearning combinations and that the best solution usesboth communication and learning.