Distribution of Tail Symbols in DST for Markov Sources - Archive ouverte HAL Access content directly
Preprints, Working Papers, ... Year :

Distribution of Tail Symbols in DST for Markov Sources

Distribution des symboles finaux dans un arbre de recherche avec des sources de Markov

(1, 2) , (3)
1
2
3

Abstract

Lempel-Ziv'78 is one of the most popular data compression algorithm on words. Over the last decades we uncover its fascinating behavior and understand better many of its beautiful properties. Among others, in 1995 by settling the Ziv conjecture we proved that for memoryless source (i.e., when a sequence is generated by a source without memory) the number of LZ'78 phrases satisfies the Central Limit Theorem (CLT). Since then the quest commenced to extend it to Markov sources, however, despite several attempts this problem is still open. In this conference paper, we revisit the issue and focus on a much simpler, but not trivial problem that may lead to the resolution of the LZ'78 dilemma. We consider the associated Digital Search Tree (DST) version of the problem in which the DST is built over a fixed number of Markov generated sequences. In such a model we shall count the number of of the so called "tail symbol", that is, the symbol that follows the last inserted symbol. Our goal here is to analyze this new quantity under Markovian assumption since it plays crucial role in the analysis of the original LZ'78 problem. We establish the mean, the variance, and the central limit theorem for the number of tail symbols. We accomplish it by applying techniques of analytic combinatorics on words also known as analytic pattern matching.
Fichier principal
Vignette du fichier
dst-tail.pdf (164.47 Ko) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-01966158 , version 1 (27-12-2018)

Identifiers

  • HAL Id : hal-01966158 , version 1

Cite

Philippe Jacquet, Wojciech Szpankowski. Distribution of Tail Symbols in DST for Markov Sources. 2018. ⟨hal-01966158⟩

Collections

INRIA INRIA2
48 View
55 Download

Share

Gmail Facebook Twitter LinkedIn More