Streaming-Codes for Multicast over Burst Erasure Channels

TL;DR

This paper analyzes the capacity limits of real-time streaming over burst-erasure channels with two receivers, providing full characterization in large-delay regimes and partial results in low-delay regimes, using novel coding and converse techniques.

Contribution

It introduces new capacity characterizations for streaming over burst-erasure channels, including novel code constructions and converse proofs for different delay regimes.

Findings

Capacity fully characterized in large-delay regime.

Partial capacity characterization in low-delay regime.

Novel coding and converse techniques introduced.

Abstract

We study the capacity limits of real-time streaming over burst-erasure channels. A stream of source packets must be sequentially encoded and the resulting channel packets must be transmitted over a two-receiver burst-erasure broadcast channel. The source packets must be sequentially reconstructed at each receiver with a possibly different reconstruction deadline. We study the associated capacity as a function of burst-lengths and delays at the two receivers. We establish that the operation of the system can be divided into two main regimes: a low-delay regime and a large-delay regime. We fully characterize the capacity in the large delay regime. The key to this characterization is an inherent slackness in the delay of one of the receivers. At every point in this regime we can reduce the delay of at-least one of the users until a certain critical value and thus it suffices to obtain code constructions for certain critical delays. We partially characterize the capacity in the low-delay regime. Our capacity results involve code constructions and converse techniques that appear to be novel. We also provide a rigorous information theoretic converse theorem in the point-to-point setting which was studied by Martinian in an earlier work.