Multitask Diffusion Adaptive Step-size Normalised Least Mean Square Algorithm

This paper present the adaptive step-size normalized least mean square algorithm with multitask diffusion strategy for distributed model. The unconstrained global cost function is considered and the local cost function is derived with the mixed-node approach. Adapt-Then-Combine strategy of diffusion network is investigated on the local cost function. The mixed-node scheme is given by the orthogonal projection matrix of its coefficient, when the other connected nodes are undisturbed. Simulation is used the system identification for experiment. Results show that proposed algorithm can outperform the fast convergence with promising the mean square error scheme.