Efficient Construction of Polar Codes

TL;DR

This paper introduces a universal and efficient method for constructing polar codes by using partial orders and a dimension reduction technique to significantly decrease computational complexity across various channels.

Contribution

The paper proposes a novel construction process for polar codes that incorporates partial orders and a dimension reduction method, reducing computations and making the process more universal.

Findings

Partial orders determine 50% of bit channels at N=10, R=0.5.

Dimension reduction increases determined channels to 82%.

Significant computational savings in polar code construction.

Abstract

The construction of polar codes for channels other than BECs requires sorting of all bit channels and then selecting the best $K$ of them for a block length $N=2^n$. In this paper, two types of partial orders (PO) of polar codes are incorporated in the construction process to decrease the required computations. Three sets, corresponding to the good bit channels ($\mathcal{I}$), the frozen bit channels ($\mathcal{F}$), and the undetermined bit channels ($\mathcal{U}$), are selected by applying PO relations. The POs are channel independent and are therefore universal for all binary-input discrete memoryless channels. For a given specific channel, a new process, called Dimension Reduction (DR), is proposed in this paper to further reduce the size of $\mathcal{U}$. Our studies show that for $N=10$ and the code rate $R=0.5$ (being the worst code rate), incorporating PO relations alone can determine 50% of the bit channels ($|\mathcal{I}| + |\mathcal{F}| \approx N/2$). With our proposed DR, this number of the determined bit channels goes up to 82%, which brings a significant reduction of computations in the construction of polar codes.