Improved Syndrome-based Neural Decoder for Linear Block Codes

TL;DR

This work improves on previous works in terms of allowing full decoding of the message rather than codewords, allowing thus the application to non-systematic codes, and proving that the single-message training property is still viable.

Abstract

In this work, we investigate the problem of neural-based error correction decoding, and more specifically, the new so-called syndrome-based decoding technique introduced to tackle scalability in the training phase for larger code sizes. We improve on previous works in terms of allowing full decoding of the message rather than codewords, allowing thus the application to non-systematic codes, and proving that the single-message training property is still viable. The suggested system is implemented and tested on polar codes of sizes (64,32) and (128,64), and a BCH of size (63,51), leading to a significant improvement in both Bit Error Rate (BER) and Frame Error Rate (FER), with gains between 0.3dB and 1dB for the implemented codes in the high Signal-to-Noise Ratio (SNR) regime.

Figures (5)

• Figure 1: General system model.
• Figure 2: SBND architecture.
• Figure 3: A RNN implementation of the message bit-flip estimator of Figure \ref{['fig:sbd_architecture']}. $\boldsymbol{h}_{i,t}$ represents the state of the $i$th GRU cell $\boldsymbol{g}_i$ at the time step $t$.
• Figure 4: Error rate studies for two polar codes of block lengths $64$ and $128$ and code rate $1/2$. The continuous lines and the dotted lines represent the $(64,32)$ and $(128,64)$ polar codes, respectively.
• Figure 5: Error rate studies for a ($63,51$) BCH code. Continuous lines represent message-to-message error rates and dotted lines depict codeword-to-codeword error rates. Codeword and message BER are the same for Choukroun et alChoukroun_2022.

Theorems & Definitions (2)

• proof
• proof