Model-Data Dual-Driven Adaptive Prior-Enhanced Intelligent Turbo Decoding Algorithm
Ruobing Li, Lizhe Liu, Yang Shuo, Yong Li, Bin Wang · 2025
To address the inadequate bit error rate (BER) performance of traditional Turbo decoding algorithms, we propose an adaptive prior-enhanced intelligent Turbo decoding algorithm based on Model-Data Dual-Driven.By deeply unfolding the iterative decoding process of the Max-Log-MAP algorithm, we construct the Adaptive Prior-Enhanced Turbo Decoding Network (APE-TurboNet) model. Within this framework, we introduce the Adaptive Prior-Enhanced Module (APEMod), which employs dual-driven deep learning principles. This module utilizes learnable weights to linearly adjust extrinsic information. It further incorporates fully connected layers to enhance parameter differentiability and improve the capabilities of nonlinear feature extraction. Additionally, a learnable mixing coefficient is employed to effectively combine both linear computation results and nonlinear extraction outcomes, leading to a more accurate estimation of extrinsic information. We also propose a novel hybrid loss function to optimize the training process. Simulation results demonstrate that the proposed algorithm significantly outperforms conventional Turbo decoding algorithms in terms of BER performance.