| PhD Viva

Name of the Speaker: Mr. Madhusudan Kumar Sinha (EE16D028)
Guide: Dr. Arun Pachai Kannu
Online meeting link: https://meet.google.com/mxh-eymy-zka
Date/Time: 2nd July 2024 (Tuesday), 3:00 PM
Title: Generalized Sparse Regression Codes for Short Block Lengths

Abstract :

Sparse Regression Codes (SPARC) are a class of error control codes that use sparse linear combinations of columns of a dictionary matrix as codewords. SPARC combines the sparse signal recovery framework of compressive sensing with error control coding techniques. SPARC is known to be asymptotically capacity-achieving but performs poorly in short block-length regimes. We focus on improving the performance of SPARC for short block lengths by introducing low-correlation dictionary matrices, efficient encoders, and efficient decoders. The dictionary matrices are based on Gold code sequences and mutually unbiased bases. We introduce two generalizations of the SPARC to maximize the number of bits for a given sparsity level and dictionary size. We also introduce a computationally efficient greedy decoder called match and decode (MAD) decoder and its parallel computing counterpart called parallel MAD decoder. We numerically show that generalized SPARC with PMAD performs well in short block-length regimes with error performance comparable to existing error control codes of the same length. We also observe that PMAD outperforms the AMP decoder and its variants in short block length regime. We numerically show that the resulting code outperforms the lower bound for orthogonal multiple access codes in the short block length regime when used for multi-user communication.