On Metric Sorting for Successive Cancellation List Decoding of Polar Codes

TL;DR

This paper improves the efficiency of metric sorter units in hardware implementations of polar code decoders by proposing two new architectures that optimize speed and size based on the structure of path metrics.

Contribution

It introduces two novel metric sorter architectures tailored for successive cancellation list decoders, enhancing speed and area efficiency based on log-likelihood ratio structures.

Findings

First sorter is 14% faster and 45% smaller for list size 32.

Second sorter reduces area by up to 36% for smaller list sizes.

Proposed architectures outperform existing solutions in synthesis results.

Abstract

We focus on the metric sorter unit of successive cancellation list decoders for polar codes, which lies on the critical path in all current hardware implementations of the decoder. We review existing metric sorter architectures and we propose two new architectures that exploit the structure of the path metrics in a log-likelihood ratio based formulation of successive cancellation list decoding. Our synthesis results show that, for the list size of $L=32$, our first proposed sorter is $14\%$ faster and $45\%$ smaller than existing sorters, while for smaller list sizes, our second sorter has a higher delay in return for up to $36\%$ reduction in the area.