Normalized LMS algorithm and data-selective strategies for adaptive graph signal estimation

Abstract

This work proposes a normalized least-mean-squares (NLMS) algorithm for online estimation of bandlimited graph signals (GS) using a reduced number of noisy measurements. As in the classical adaptive filtering framework, the resulting GS estimation technique converges faster than the least-mean-squares (LMS) algorithm while being less complex than the recursive least-squares (RLS) algorithm, both recently recast as adaptive estimation strategies for the GS framework. Detailed steady-state mean-squared error and deviation analyses are provided for the proposed NLMS algorithm, and are also employed to complement previous analyses on the LMS and RLS algorithms. Additionally, two different time-domain data-selective (DS) strategies are proposed to reduce the overall computational complexity by only performing updates when the input signal brings enough innovation. The parameter setting of the algorithms is performed based on the analysis of these DS strategies, and closed formulas are derived for an accurate evaluation of the update probability when using different adaptive algorithms. The theoretical results predicted in this work are corroborated with high accuracy by numerical simulations.

Adaptive Graph Filters in Reproducing Kernel Hilbert Spaces: Design and Performance Analysis

“…The application of linear shift-invariant filter models is widely employed in the literature, e.g., to design graph spectral filters [11], [12] and model dynamic graph signals [15], [16]. Several works deal with adaptive learning of graph filters, see, e.g., [19]- [23]. These methods were later extended to multitask graphs [24], [25].…”

Joint Forecasting and Interpolation of Time-Varying Graph Signals Using Deep Learning

“…Global methods, on the other hand, interpolate the unknown signal values at once and can provide better results by taking the entire network into account at the expensive of a higher computational burden [63], [65]. Many GSP interpolation schemes have been proposed [66]- [70].…”

Diffusion-based Virtual Graph Adjacency for Fourier Analysis of Network Signals

“…Real networks and their corresponding data come in vastly different shapes and applications, ranging from genetic interaction networks [3] and the human brain [4] to sensor networks and smart cities [5]. Graph signal processing (GSP) explores pairwise relations between elements of a network to construct tools suitable for the processing of network data [1], [2], [6]- [12]. In GSP, networks are modeled as graphs and data defined over, or generated by, elements of these networks are modeled as a graph signal-a mapping from the set of vertices into the set of complex numbers.…”