Diffusion LMS algorithm with multi-combination for distributed estimation over networks

We propose diffusion least-mean-square (LMS) algorithms utilizing multi-combination steps. Unlike conventional diffusion LMS, we allow each node in the network to use information from multi-hop neighbors for improving the approximation accuracy of a global cost function. The resulting distributed algorithms consist of adaptation and multi-combination steps. Through the multi-combination, each node can use information from non-adjacent nodes at each time instant, resulting in enhanced performance. Performance analysis gives stability condition and quantifies the steady-state behaviors. The simulation result indicates that the proposed algorithm outperforms the conventional diffusion LMS algorithm.