Bounds on the performance of Raptor codes

Presented by Francisco Lázaro


Abstract: A fountain code is a rateless erasure code, i.e., a code that is potentially able to generate an endless amount of data. Fountain codes were originally proposed for reliable multicasting, where a transmitter wants to deliver data to multiple users over a (lossy) data network were data packets might be lost. In the last decade Raptor codes with Maximum Likelihood (ML) decoding, the most advanced construction of fountain codes known so far, have found their way into many standards, such as 3GPP MBMS, DVB-H and IETF.  Furthermore, they are currently being proposed for novel applications such as coded distributed computing, distributed storage and coded caching. In this talk we present generic bounds on the performance of Raptor codes under ML decoding. The bounds are shown to be tight by means of simulations, and can be used to design Raptor codes with low error floors.

Bio: Francisco Lázaro works as researcher at the Institute for Communications and Navigation from the German Aerospace Center (DLR). He holds a Telecommunication Engineering degree from the University of Zaragoza (Spain), and a PhD from the Technical University of Hamburg (Germany). His main research interests include rateless codes and random access protocols.