A new class of rank-metric codes and their list decoding beyond the unique decoding radius

TL;DR

This paper introduces a novel class of rank-metric codes with not small constant ratio, enabling efficient list decoding beyond the traditional radius using algebraic techniques and list pruning methods.

Contribution

The paper constructs rank-metric codes with constant ratio and develops efficient list decoding algorithms beyond the half-radius, employing two-variable polynomials and list pruning strategies.

Findings

Decodable beyond the $(1-R)/2$ radius

Codes with constant ratio $ ho( ext{C})$

Deterministic and randomized list decoding algorithms

Abstract

Compared with classical block codes, efficient list decoding of rank-metric codes seems more difficult. Although the list decodability of random rank-metric codes and limits to list decodability have been completely determined, little work on efficient list decoding rank-metric codes has been done. The only known efficient list decoding of rank-metric codes $\mC$ gives decoding radius up to the Singleton bound $1-R-\Ge$ with positive rate $R$ when $\rho(\mC)$ is extremely small, i.e., $\Theta(\Ge^2)$ , where $\rho(\mC)$ denotes the ratio of the number of rows over the number of columns of $\mC$ \cite[STOC2013]{Guru2013}. It is commonly believed that list decoding of rank-metric codes $\mC$ with not small constant ratio $\rho(\mC)$ is hard. The main purpose of the present paper is to explicitly construct a class of rank-metric codes $\mC$ with not small constant ratio $\rho(\mC)$ and efficiently list decode these codes with decoding radius beyond $(1-R)/2$. Our key idea is to employ two-variable polynomials $f(x,y)$, where $f$ is linearized in variable $x$ and the variable $y$ is used to "fold" the code. In other words, rows are used to correct rank errors and columns are used to "fold" the code to enlarge decoding radius. Apart from the above algebraic technique, we have to prune down the list. The algebraic idea enables us to pin down the messages into a structured subspace of dimension linear in the number $n$ of columns. This "periodic" structure allows us to pre-encoding the messages to prune down the list. More precisely, we use subspace design introduced in \cite[STOC2013]{Guru2013} to get a deterministic algorithm with a larger constant list size and employ hierarchical subspace-evasive sets introduced in \cite[STOC2012]{Guru2012} to obtain a randomized algorithm with a smaller constant list size.