Distributed Estimation with Sparsely Accessible Information

Mahdi Shamsi, Farokh Marvasti

Session: Frames, Riesz bases, and related topics (Jorge Antezana)

Keywords: Distributed Estimation, DLMS, Sparse Mask, Signal recovery over network.

TL;DR: Proposing a thresholding-based algorithm with optimal combination weights to enhance Diffusion LMS performance under partial observability, achieving significant accuracy gains.

Abstract:

This paper addresses sparse observability constraints in Diffusion Least Mean Squares (DLMS) and proposes a framework for analyzing combination strategies. A thresholding-based algorithm is introduced to identify the sparse support vector under incomplete information. The method effectively handles sparse observations in both time and transform domains, achieving a 30–40 dB improvement in Mean Square Deviation (MSD) over conventional DLMS.

Submission Number: 57