A representation of non-uniformly sampled deterministic and random signals and their reconstruction using sample values and derivatives

TL;DR

This paper explores advanced methods for reconstructing non-uniformly sampled signals using derivatives, polynomial-exponential representations, and statistical estimation techniques, enhancing signal recovery accuracy especially under limited information.

Contribution

It introduces a novel polynomial-exponential representation for signals, extends the approach to two dimensions, and develops estimation methods based on Gaussian noise correlation for improved reconstruction.

Findings

Enhanced reconstruction accuracy with derivatives in non-uniform sampling

Development of polynomial-exponential signal representation

Effective estimation techniques using Gaussian noise correlation

Abstract

Shannon in his 1949 paper suggested the use of derivatives to increase the W*T product of the sampled signal. Use of derivatives enables improved reconstruction particularly in the case of non-uniformly sampled signals. An FM-AM representation for Lagrange/Hermite type interpolation and a reconstruction technique are discussed. The representation using a product of a polynomial and exponential of a polynomial is extensible to two dimensions. When the directly available information is inadequate, estimation of the signal and its derivative based on the correlation characteristics of Gaussian filtered noise has been studied. This requires computation of incomplete normal integrals. Reduction methods for reducing multivariate normal variables include multistage partitioning, dynamic path integral and Hermite expansion for computing the probability integrals necessary for estimating the mean of the signal and its derivative at points intermediate between zero or threshold crossings. The signals and their derivatives as measured or estimated are utilized to reconstruct the signal at a desired sampling rate.