Unsupervised Equalization over Multitask Adaptive Networks

Adaptive Networks (ANs) are groups of agents that collaborate to perform data-driven tasks such as optimization and inference. This work evaluates distributed filter estimation for unsupervised equalization over ANs using adaptive signal processing algorithms: the Constant Modulus Algorithm and Single Channel Independent Component Analysis (SCICA). Equalization is assumed to be performed using a two-filter cascade in a multitask setting, where agents experience different but correlated channels. The simulation results show the advantage of the collaborative algorithm over the standalone implementation. Additionally, a strategy for reducing data exchange while maintaining acceptable network performance is proposed for the distributed SCICA.