English

Polynomial Approximations of Conditional Expectations in Scalar Gaussian Channels

Information Theory 2021-02-12 v1 math.IT Probability

Abstract

We consider a channel Y=X+NY=X+N where XX is a random variable satisfying E[X]<\mathbb{E}[|X|]<\infty and NN is an independent standard normal random variable. We show that the minimum mean-square error estimator of XX from Y,Y, which is given by the conditional expectation E[XY],\mathbb{E}[X \mid Y], is a polynomial in YY if and only if it is linear or constant; these two cases correspond to XX being Gaussian or a constant, respectively. We also prove that the higher-order derivatives of yE[XY=y]y \mapsto \mathbb{E}[X \mid Y=y] are expressible as multivariate polynomials in the functions yE[(XE[XY])kY=y]y \mapsto \mathbb{E}\left[ \left( X - \mathbb{E}[X \mid Y] \right)^k \mid Y = y \right] for kN.k\in \mathbb{N}. These expressions yield bounds on the 22-norm of the derivatives of the conditional expectation. These bounds imply that, if XX has a compactly-supported density that is even and decreasing on the positive half-line, then the error in approximating the conditional expectation E[XY]\mathbb{E}[X \mid Y] by polynomials in YY of degree at most nn decays faster than any polynomial in n.n.

Keywords

Cite

@article{arxiv.2102.05970,
  title  = {Polynomial Approximations of Conditional Expectations in Scalar Gaussian Channels},
  author = {Wael Alghamdi and Flavio P. Calmon},
  journal= {arXiv preprint arXiv:2102.05970},
  year   = {2021}
}

Comments

A short version of this paper has been submitted to the 2021 IEEE International Symposium on Information Theory (ISIT)