Index Coding, Error Correcting Index Codes And Matroids
Abstract
In the problem of index coding, there is a central source sending messages to a
set of receivers which demand messages over a noiseless broadcast channel. Each
receiver knows a subset of messages which is referred to as side-information. A
backward channel exists between each of the receivers and the source. Receivers
use the reverse channel to communicate the side-information back to the source.
The source uses this side-information to develop encoding schemes so that all the
receivers are satisfied with minimum number of transmissions.
In this work, we consider few problems related to index coding. First, we
consider the informed source coding problem in which the receiver communicates
only the cardinality of the side-information available to it and not the messages
that constitute the side-information. We use l-th Near Maximum Distance Sep-
arable (Near-MDS) codes to construct encoding schemes for the informed source
coding problem. The advantage of using l-th Near-MDS codes is the reduction of
required field size. For certain class of informed source coding problems, we show
that the codes constructed from l`-th Near-MDS codes are optimal under some
field size restrictions.
Links between matroid theory, network coding and index coding problems have
been an active area of research. The generalized index coding problem is a general
version of the conventional index coding problem in which both the source and
receivers possess linear functions of messages. We establish connections between
representation of discrete polymatroids and generalized index coding. We show
that the generalized index coding problem has a vector linear solution if and only
if there exists a representable discrete polymatroid satisfying certain conditions.
Further, from a discrete polymatroid a generalized index coding problem is con-
structed. It is shown that if the generalized index coding problem has a vector linear solution over the binary eld then an associated discrete polymatroid is
representable over the binary eld. For generalized index coding problems con-
structed from matroids, we show that a scalar linear solution to the constructed
problem exists if and only if the matroid is binary representable.
We also consider the index coding with restricted information (ICRI) problem
in which the source must ensure that each receiver will not be able to decode a
certain subset of messages. Interference alignment techniques are used to obtain
results on the ICRI problem. Necessary conditions for the existence of a solution
to the ICRI problem is found. A technique to obtain solutions to the ICRI problem
using contractions corresponding to restrictions is obtained.
Index coding over noisy channels is also considered. Error correcting index
codes are encoding schemes which enable every receiver to correct up to a certain
number of errors in the received message. We allow each receiver to have a di erent
error correcting capability and introduce di erential error correcting index codes to
achieve this. A set of necessary and su cient conditions for a matrix to correspond
to a vector linear di erential index code is found. We establish the link between
vector linear di erential error correcting index codes and discrete polymatroids.
It is shown that a vector linear di erential error correcting index code exists if
and only if there exists a representable discrete polymatroid satisfying certain
conditions.
Lastly, index coding problem over binary symmetric channels is studied. It
is observed that the probability of error at each receiver depends on the index
code used. A condition on the optimal index codes which minimize the maximum
error probability among all the receivers is derived. For a special class of index
coding problems, an algorithm to identify an optimal index code which gives the
best performance in terms of minimal maximum error probability across all the
receivers is provided.