| PhD Seminar

Name of the Speaker: Mr. Vavilapalli Satya Venkata Sandeep (EE22D004)
Guide: Dr. Arun Pachai Kannu
Venue: ESB-244 (Seminar Hall)
Date/Time: 25th March 2025 (Tuesday), 12:00 PM
Title: Study of Sparse Regression Codes (SPARCs) for multi user communication

Abstract :

Sparse regression codes (SPARCs) are a class of error control codes where the transmitted codewords are generated by the sparse superposition of the columns from a dictionary matrix. Decoding at the receiver employs efficient sparse recovery algorithms, which can be broadly categorized into convex optimization, greedy methods, and message passing algorithms. Recent studies have shown that SPARCs can achieve capacity in an AWGN channel using approximate message passing (AMP) decoding.

In our work, we examine SPARCs in multi-user, fading, and non-coherent(where there is no knowledge of the channel state information (CSI)) channel settings. We propose the maximum likelihood matching pursuit (MLMP) decoder, which is a greedy and iterative based decoder designed to identify one active column contributing to the codeword in each iteration based on partial ML metric by considering other undetected columns as noise. We compare the block error rate (BLER) performance of our MLMP decoder against other sparse recovery algorithms, specifically orthogonal matching pursuit (OMP) and compressive sampling matching pursuit (CoSaMP). Our MLMP decoder has been shown to outperform these alternatives at both lower and higher code rates. Additionally, we introduced two enhancements to MLMP: MLMP with replacement (MLMP-R) and parallel MLMP. We also developed modified versions of OMP and CoSaMP, which showed better performance than the original versions of these algorithms.

We further explore multi-user communication under both symmetric (where the channel fading variances for users are the same) and asymmetric (where the channel fading variances differ) channel conditions. We observed a multi-user diversity gain at higher code rates, even in the absence of channel state information (CSI), in symmetric channel conditions. Furthermore, we investigated rate adaptation in asymmetric conditions and found that, for higher code rates, the optimal resource allocation among users is to allocate more resources to the user with the best channel conditions. At lower code rates, any resource partitioning tends to yield similar performance.

For future work, we plan to study AMP in the context of our multi-user non-coherent fading channel scenario. We also intend to investigate SPARCs in a massive MIMO environment and apply learned methods for sparse recovery.