Sandeep Pradhan's
research interests are in the area of - distributed representation and
transmission of information in sensor networks, reliability in
multiuser communication systems such as wireless networks,
multiple description source coding, multiuser information theory and coding
The prevalent trend in information theory has been to obtain
performance limits of communication using Shannon's random coding
arguments and its multi-terminal extensions, and in coding theory, the
goal has been to achieve these limits using algebraic-structured codes
with fast encoding and decoding algorithms. It turns out that
algebraic structure can also be used to weed out bad codes from an
ensemble. In this work, we build on this observation, and look at the
distributed source coding problem. We present a new approach to this
problem and an achievable rate-distortion region (an inner bound to
the performance limit) based on ``good'' random nested codes built
over abstract abelian groups. We demonstrate rate gains for this
problem over traditional coding schemes using random unstructured
codes. For certain sources and distortion functions, the new
rate region is strictly bigger than the Berger-Tung rate region,
which has been the best known achievable rate region for this problem till
now. Further, there is no known unstructured random coding scheme
that achieves these rate gains.
Recent talks:
[Talk given at
AAECC in 2007]
[Talk given at
ITA in 2009]
[Talk given at
UC Berkeley in 2009]
[Thesis defense of
Dinesh Krithivasan in
2010]
Feedback increases the capacity of multi-user channels even when the
channels are discrete memoryless. We have developed a novel approach
to this problem using a graph-based model for correlated information.
The key insight is the following. Before the communication starts, the
information of the transmitters are independent. As more channel uses
are expended, the posterior correlation (conditioned on the channel
output) among the transmitter information increases. This correlation
can be exploited to combat interference and channel noise more
effectively in subsequent stages.
Recent talks: [Talk given in ISIT 2009]
Distributed source coding (compression)
deals with separate compression (source coding) of correlated
information sources and the goal is to reduce the redundancy in the
system. Application areas include sensor networks, digital enhancement of
transmissions. The research effort involves the construction of
efficient encoder and decoder based on structured channel codes to
attain the performance bounds guaranteed by Wyner and Ziv.
Recent talks: [Talk given at
CISS in 2000] [Talk given at
AAECC in 2007]
[Tutorial given at
ISIT in 2008]
Shannon showed that separate source coding (efficient information
representation) and channel coding (efficient information
communication) can achieve the optimal end-to-end performance in an
information transmission system invloving a transmitter and a receiver.
Unfortunately, this separation is suboptimal in a multiuser setting
(more than two terminals). In this research effort, we consider a
multi-partite graph-based discrete interface for multiuser
information transmission problems. Using this approach we can envisage
a scenario where many multiterminal sources can be mapped into a
graph, and many graphs can be transmissible reliably over a
multiterminal channel. In other words, rather than smearing the source
coding component and channel coding component into a
joint-source-channel-coding block, the proposed approach will enhance
them to work with "correlated messages" or graphs, thus retaining the
Shannon-style modular approach to multiuser information transmission
problems. We are interested in evaluating the performance limits of
such interfaces in a Shannon-theoretic framework.
In the source coding part, we are interested in the smallest
exponent of the size of the graphs that can reliably represent a set of
correlated sources, and in the channel coding part, we are interested
in the largest exponent of the size of the graphs that can be reliably
communicated across a multiuser channel. This has applications in
sensor networks.
Recent talks: [Talk given at Washington
University in 2007]
[Thesis Defense of Suhan Choi in 2006] [Talk given at ITA in 2007]
In this problem, the goal is to produce an efficient representation
(compression) of a source of information into an index set such that
the decoder having access to this can reconstruct the source. The
caveat, however, is that the reconstruction has to be realtime and the
decoder at the time of reconstructing any source sample has access to
the past source samples or past side information samples.
The former is called noiseless feedforward, and the latter noisy feedforward.
Side information is any extra information that the decoder has access
to that is correlated to the original source. This problem can be
considered as a functional dual to the problem of channel coding
(communication) with feedback. This has applications in sensor
networks, economics, control theory and sequential learning theory.
In this research effort, we provide information processing strategies
and their performance limits. Our goal is also to develop
constructive and practical methods for achieving these limits.
Recent talks: [Talk at ITA by Ramji
in 2007]
[Thesis defense of Ramji Venkataramanan in 2007]
For a multiuser channel (such as a multiple access channel), if the
vector of rates of the users is inside the capacity region, then the
probability of decoding error of each user
can be driven to zero by increasing the
block-length. Multiuser reliability profile of a multiuser
communication system is the set of rates of decay of probability of
errors of the users in the system as a function of block-length. This
is very useful especially for wireless communication system. In
essence reliability can be thought of as another resource (like power,
bandwidth) that can be allocated among the users in a wireless
network. Using this concept, we can have heterogeneous quality of
service guarantees in such networks directly in the physical layer.
The goal of this research effort is to develop efficient strategies
for realizing optimum trade-off of reliability among the users in a
network.
Recent talks: [Thesis Defense of Lihua Weng
in 2005]
Multiple description coding deals with efficient encoding of information
signals for transmission across packet networks.
Applications include video streaming over packet networks.
In this research effort we provide new achievable rate regions
for multiple descriptions involving more than two channels.
We have first formulated the notion of MDS (maximum distance separable)
source-channel erasure codes and obtained its rate region information
theoretically and then used it in multiple descriptions.
This research also covers efficient construction of encoders based on
quantizers and trellis-based codes.
Recent talks: [Talk given at DIMACS in
2004]