Shortened Polarization Kernels

TL;DR

This paper introduces a method for shortening large polarization kernels to achieve high error exponents, improving the performance of polar codes with various lengths.

Contribution

A novel shortening algorithm for large polarization kernels that optimizes error exponents and is applicable to kernels up to size 32.

Findings

Shortened kernels outperform traditional polar codes in simulations.

The method effectively shortens kernels from sizes 16 and 32 to 9-31.

Numerical results show improved error performance with the proposed approach.

Abstract

A shortening method for large polarization kernels is presented, which results in shortened kernels with the highest error exponent if applied to kernels of size up to 32. It uses lower and upper bounds on partial distances for quick elimination of unsuitable shortening patterns. The proposed algorithm is applied to some kernels of sizes 16 and 32 to obtain shortened kernels of sizes from 9 to 31. These kernels are used in mixed-kernel polar codes of various lengths. Numerical results demonstrate the advantage of polar codes with shortened large kernels compared with shortened and punctured Arikan polar codes, and polar codes with small kernels.