Research

Current research areas:

• Covert communication and security

• Reinforcement learning for decoding LDPC codes

• Change-point detection

1. Covert communication and security

Consider a situation when Alice (sender) is speaking with Bob (legitimate receiver) while an intruder Willie (illegitimate receiver)  is snooping over this conversation (see figure). Since Bob is a legitimate receiver, he knows the parameters associated with the communication mechanism used by Alice. Willie (being an illegitimate receiver) doesn't know these transmission parameters but wants to identify it. While Willie wants to identify these parameters correctly, Alice wants design a communication mechanism such that Willie will make an error with high probability. This problem of studying the tradeoff between the aims of Alice and Willie is called as the problem of “covert communication”. Such a setup has applications in military surveillance where one has access to  adversary’s data but doesn't know the associated parameters.

In this project, the aim is to study this problem of covert communication algorithmically (provide an algorithm for Willie’s parameter estimation) and theoretically (find information theoretic limits for Alice’s communication mechanism). We will also explore the possibilities of using machine learning for this parameter estimation and compare the performance of the ML based solution versus theoretical solution.

The parameter estimation problem here can be formulated as a signal detection problem and various machine learning based solutions can be provided to it. Designing neural networks for this is also of interest.

2. Reinforcement learning for decoding LDPC codes

Low-density parity-check (LDPC) codes have been used popularly in 5G and will be used in next generation wireless systems as well. These codes are typically decoded via belief propagation (BP) decoding. BP decoding is an iterative algorithm where in each iteration, messages are passed between a set of nodes (here nodes corresponds to the nodes in graphical representation of LDPC codes). It has been observed that scheduling of these messages play a crucial role in the convergence of BP decoding algorithm. 

In this project, we will explore the possibilities of using reinforcement learning for this scheduling.

3. Change-point detection