An energy-efficient diffusion strategy over adaptive networks

We propose an energy-efficient diffusion strategy where each node performs the combination step in the diffusion LMS strategy only at every L time slots. This proposed strategy reduces the inter-node communication overhead by L-fold, but yields a steady-state network mean-square deviation (MSD) higher than the traditional diffusion LMS strategy. To improve the steady-state network MSD performance, we further propose a step-size switching mechanism that switches to a decaying step-size in the adaptation step of the diffusion LMS strategy as the MSD converges. To ensure adaptability of the new strategy, we also incorporate a parameter change detection method to switch the step-size back to a constant value if the underlying parameter varies. Simulation results suggest that the proposed method can effectively reduce the network energy consumption without sacrificing steady-state network MSD performance.