Input-distribution-aware parallel decoding of block codes

TL;DR

This paper introduces input-distribution-aware decoding techniques that adapt parallel decoding efforts based on input distribution, significantly reducing runtime complexity while maintaining decoding performance across various codes and algorithms.

Contribution

It proposes M-IDA and MD-IDA methods that leverage input distribution sampling for low-cost, effective IDA decoding applicable to multiple decoding algorithms.

Findings

Achieves runtime reductions of 17% to 67% with minimal error degradation.

Demonstrates effectiveness across BCH codes with Chase and ORBGRAND algorithms.

Maintains decoding performance comparable to original IDA decoding.

Abstract

Many channel decoders rely on parallel decoding attempts to achieve good performance with acceptable latency. However, most of the time fewer attempts than the foreseen maximum are sufficient for successful decoding. Input-distribution-aware (IDA) decoding allows to determine the parallelism of polar code list decoders by observing the distribution of channel information. In this work, IDA decoding is proven to be effective with different codes and decoding algorithms as well. Two techniques, M-IDA and MD-IDA, are proposed: they exploit the sampling of the input distribution inherent to particular decoding algorithms to perform low-cost IDA decoding. Simulation results on the decoding of BCH codes via the Chase and ORBGRAND algorithms show that they perform at least as well as the original IDA decoding, allowing to reduce run-time complexity down to 17% and 67\% with minimal error correction degradation.