Information-Theoretic Capacity and Error Exponents of Stationary Point Processes under Random Additive Displacements

TL;DR

This paper investigates the capacity and error exponents for stationary point processes under random displacements in high dimensions, revealing a sharp threshold for successful decoding based on entropy and process characteristics.

Contribution

It establishes a sharp threshold for decoding success in high dimensions and derives explicit error exponents using large deviations theory for Poisson and Matérn processes.

Findings

Decoding success probability sharply transitions based on entropy and process parameters.

Explicit error exponent formulas are derived for certain point processes.

The results connect point process displacements to Shannon's additive noise channel error analysis.

Abstract

This paper studies the Shannon regime for the random displacement of stationary point processes. Let each point of some initial stationary point process in $\R^n$ give rise to one daughter point, the location of which is obtained by adding a random vector to the coordinates of the mother point, with all displacement vectors independently and identically distributed for all points. The decoding problem is then the following one: the whole mother point process is known as well as the coordinates of some daughter point; the displacements are only known through their law; can one find the mother of this daughter point? The Shannon regime is that where the dimension $n$ tends to infinity and where the logarithm of the intensity of the point process is proportional to $n$. We show that this problem exhibits a sharp threshold: if the sum of the proportionality factor and of the differential entropy rate of the noise is positive, then the probability of finding the right mother point tends to 0 with $n$ for all point processes and decoding strategies. If this sum is negative, there exist mother point processes, for instance Poisson, and decoding strategies, for instance maximum likelihood, for which the probability of finding the right mother tends to 1 with $n$. We then use large deviations theory to show that in the latter case, if the entropy spectrum of the noise satisfies a large deviation principle, then the error probability goes exponentially fast to 0 with an exponent that is given in closed form in terms of the rate function of the noise entropy spectrum. This is done for two classes of mother point processes: Poisson and Mat\'ern. The practical interest to information theory comes from the explicit connection that we also establish between this problem and the estimation of error exponents in Shannon's additive noise channel with power constraints on the codewords.