Noise Variance Estimation Through Joint Analysis of Intrinsic Dimension and Differential Entropy

Intrinsic dimensional analysis and differential entropy estimation are closely related to the point that they can be jointly estimated as two complementary outputs of a single approach. In this brief letter, it is shown that, given a noisy signal, intrinsic dimension and differential entropy, jointly analyzed through visual approach, can be used to estimate noise variance directly from noisy signal samples. Preliminary results corroborating this idea are presented through experiments with chaotic (Lorenz chaotic model) and almost periodic (human vocalization) signals, including comparisons with results from methods in the concerned literature.