Distributed constrained consensus least-mean square algorithms with adjustable constraints

This paper presents novel constrained consensus least mean square (cLMS) algorithms with adjustable constraints that can improve the learning performance of distributed estimation problems in sensor networks by exploiting the spatial diversity of the estimates. For the first algorithm, the constraint vectors are adjusted by combining the components of the estimate orthogonal to its neighbor estimates. To further speed up the convergence, the second algorithm only uses one of these orthogonal components corresponding to the maximum angle between the estimate and its neighbor estimates to compute the constraint vectors. Simulation results show that both proposed algorithms provide faster convergence than the existing cLMS and diffusion LMS algorithms.