Connected Graph Multitask Diffusion LMS Via Orthonormal Codes in Ad-Hoc Networks

In this work, a multitask diffusion least mean square (MDLMS) algorithm is developed via orthonormal codes in ad-hoc networks. Unlike the existing MDLMS approaches, where the adaptive combiner matrix is altered and becomes a disconnected graph, the newly proposed MDLMS approach preserves the combining matrix as a connected graph based on the orthonormal codes. The connected graph property allows nodes located in a similar cluster to exchange their knowledge by node cooperation to nearby and faraway nodes. In the simulations, the performance of the newly proposed MDLMS and the existing adaptive combiner methods are similar; however, the newly proposed MDLMS method posses a unique feature which doesn’t alter the connected graph property over ad-hoc networks.