Milojković Jovan, Brkić Srdjan, Ivaniš Predrag, Vasić Bane
School of Electrical Engineering, University of Belgrade, 11000 Belgrade, Serbia.
Tannera Technologies LLC, Veljka Dugosevica 54, 11000 Belgrade, Serbia.
Entropy (Basel). 2025 Jan 9;27(1):49. doi: 10.3390/e27010049.
In this paper, we introduce a novel gradient descent bit-flipping algorithm with a finite state machine (GDBF-wSM) for iterative decoding of low-density parity-check (LDPC) codes. The algorithm utilizes a finite state machine to update variable node potentials-for each variable node, the corresponding finite state machine adjusts the update value based on whether the node was a candidate for flipping in previous iterations. We also present a learnable framework that can optimize decoder parameters using a database of uncorrectable error patterns. The performance of the proposed algorithm is illustrated for various regular LDPC codes, both in a binary symmetric channel (BSC) and the channel with additive white Gaussian noise (AWGN). The numerical results indicate a performance improvement when comparing our algorithm to previously proposed GDBF-based approaches.
在本文中，我们介绍了一种用于低密度奇偶校验（LDPC）码迭代译码的、带有有限状态机的新型梯度下降比特翻转算法（GDBF-wSM）。该算法利用有限状态机来更新变量节点势——对于每个变量节点，相应的有限状态机根据该节点在先前迭代中是否为翻转候选节点来调整更新值。我们还提出了一个可学习框架，该框架可以使用不可纠正错误模式数据库来优化译码器参数。针对各种正则LDPC码，在二进制对称信道（BSC）和加性高斯白噪声（AWGN）信道中展示了所提算法的性能。数值结果表明，将我们的算法与先前提出的基于GDBF的方法相比，性能有所提升。
