Information and Statistical Efficiency When Quantizing Noisy DC Values

TL;DR

This paper analyzes the statistical efficiency of different estimators for quantized noisy constant signals, comparing their performance and proposing conditions for optimality.

Contribution

It provides a detailed comparison of simple, moment-based, and maximum-likelihood estimators for quantized noisy signals, including efficiency expressions and performance improvements.

Findings

Arithmetic mean efficiency compared to Cramér-Rao bound.

Moment-based and maximum-likelihood estimators outperform simple averaging under certain noise conditions.

Simulation results confirm improved estimation when noise is comparable to quantization step.

Abstract

This paper considers estimation of a quantized constant in noise when using uniform and nonuniform quantizers. Estimators based on simple arithmetic averages, on sample statistical moments and on the maximum-likelihood procedure are considered. It provides expressions for the statistical efficiency of the arithmetic mean by comparing its variance to the proper Cram\'er-Rao lower bound. It is conjectured that the arithmetic mean is optimal among all estimators with an exactly known bias. Conditions under which its statistical performance are improved by the other estimation procedures when the exact bias is not known are found and analyzed. Using simulations and analysis of experimental data, it is shown that both moment-based and maximum-likelihood-based estimators provide better results, when the noise standard deviation is comparable with the quantization step and the noise model of quantization can not be applied.