Optimization of List-Based Reed-Solomon Decoding via Deep Learning
Year: 2026
Type: conference-paper
Abstract: This paper presents a deep learning-based optimization of a list-decoding algorithm for a concatenated Hamming-Reed-Solomon code transmitted over an AWGN channel with BPSK modulation. The proposed method reformulates early termination of the erasure-pattern list as a weighted regression problem, addressing class imbalance through an asymmetric Weighted Mean Squared Error (WMSE) loss function that penalizes index underestimation more heavily than overestimation. Simulation results confirm that the neural network predictor reduces the average number of decoding iterations across the evaluated SNR range while maintaining error-correction performance comparable to the baseline list decoder.
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7170154791
