Adaptive Normalized LMP Estimation for Graph Signal Processing

Abstract

We propose an adaptive normalized least mean pth power (GNLMP) algorithm for graph signal processing (GSP), which estimates sampled graph signals under impulsive noise. Compared to the recently introduced adaptive GSP least mean pth power (GLMP) algorithm, the GNLMP algorithm reduces the number of iterations to converge to a steady graph signal. Different from adaptive GSP normalized least mean square (GNLMS) algorithm, the GNLMP algorithm has the ability to reconstruct a graph signal that is corrupted by non-Gaussian noise with heavy-tailed characteristics. Simulations show the performance of the GNLMP algorithm in estimating steady-state and time-varying graph signals, utilizing spectral properties such as bandlimitedness and sampling, faster and more robust in comparison to GLMP and GNLMS.