Learning from the Syndrome

TL;DR

This paper introduces the syndrome loss, a new training method for neural decoders that improves error correction performance without extra inference costs, adaptable to changing channels.

Contribution

The syndrome loss is a novel loss function that relaxes the syndrome constraint, enhancing neural decoder training for error correction.

Findings

Lower frame error rates across various short block codes.

No additional inference cost during deployment.

Effective for online adaptation to channel changes.

Abstract

In this paper, we introduce the syndrome loss, an alternative loss function for neural error-correcting decoders based on a relaxation of the syndrome. The syndrome loss penalizes the decoder for producing outputs that do not correspond to valid codewords. We show that training with the syndrome loss yields decoders with consistently lower frame error rate for a number of short block codes, at little additional cost during training and no additional cost during inference. The proposed method does not depend on knowledge of the transmitted codeword, making it a promising tool for online adaptation to changing channel conditions.