Improved Decoding Algorithms for Reed-Solomon Codes

TL;DR

This paper compares and improves decoding algorithms for Reed-Solomon codes, introducing a faster, parallelizable decoding method with linear algebra techniques that enhances error correction efficiency.

Contribution

It demonstrates that the fPGZ decoder is a special case of the BM decoder, uncovers their relationship, and proposes a new parallel decoding algorithm with O(e) complexity.

Findings

fPGZ is a special case of BM decoding.

New error value formulas improve decoding accuracy.

Proposed parallel decoding algorithm achieves O(e) complexity.

Abstract

In coding theory, Reed-Solomon codes are one of the most well-known and widely used classes of error-correcting codes. In this thesis we study and compare two major strategies known for their decoding procedure, the Peterson-Gorenstein-Zierler (PGZ) and the Berlekamp-Massey (BM) decoder, in order to improve existing decoding algorithms and propose faster new ones. In particular we study a modified version of the PGZ decoder, which we will call the fast Peterson-Gorenstein-Zierler (fPGZ) decoding algorithm. This improvement was presented in 1997 by exploiting the Hankel structure of the syndrome matrix. In this thesis we show that the fPGZ decoding algorithm can be seen as a particular case of the BM one. Indeed we prove that the intermediate outcomes obtained in the implementation of fPGZ are a subset of those of the BM decoding algorithm. In this way, we also uncover the existing relationship between the leading principal minors of syndrome matrix and the discrepancies computed by the BM algorithm. Finally, thanks to the study done on the structure of the syndrome matrix and its leading principal minors, we improve the error value computation in both the decoding strategies studied (specifically we prove new error value formulas for the fPGZ and the BM decoding algorithm) and moreover we state a new iterative formulation of the PGZ decoder well suited to a parallel implementation on integrated microchips. Thus using techniques of linear algebra we obtain a parallel decoding algorithm for Reed-Solomon codes with an O(e) computational time complexity, where e is the number of errors which occurred, although a fairly large number of elementary circuit elements is needed.