Silvija Kokalj, INRIA Saclay

Location :

Building P, room P104,
INRIA Saclay
Parc Orsay Université
SACLAY

Abstract :

PhD dissertation research of Silvija Kokalj-Filipovic focuses on
applications of large deviation theory to the problems of data
dissemination, storage and collection, and information
recovery/decoding in wireless sensor networks.  In this talk, we
present the coding-theory aspect of this research through an analysis
of the message-passing decoding of fountain codes which extends the
concept of message passing algorithm by allowing so called doping of
the decoder. The ergodicity of the Ideal Soliton degree distribution
of code symbols allows for a tractable model of the density evolution
and, in particular, an insightful analysis of the ripple process,
resulting in a unified model for both classical and doping-enhanced
decoding. For a given number of code symbols at the decoder, the
random-walk-based analysis of this process furnishes the decoding
delay model with a prediction on the number of required doped packets.

As an application of the above decoding technique,  the dissertation
studies data collection  in circular wireless networks where data
storage is mechanized using distributed fountain coding techniques.
The goal of the presented approach is to allow for a reduced-delay
collection by a data collector who accesses the circular network at a
random position and random time. The storage nodes within the
transmission range of the network’s relays linearly combine and store
overheard relay transmissions using random decentralized strategies. A
data collector first collects a minimum set of coded packets from a
subset of storage nodes in its proximity and, by using a
message-passing decoder, attempts recovering all source packets from
this set. Whenever the decoder stalls, a source packet which restarts
decoding is polled/doped from its original source node. The collection
delay can be evaluated using the techniques described above, based on
the predicted number of required doped packets. Ideal Soliton
distribution results in a collection delay which is convincingly
smaller compared with other distributions or other collection
techniques.

Apart from this particular aspect I will also try to give a broader
outlook of my research if time permits.

Host :

Fabrice Le Fessant