空间耦合LDPC码的深度迭代译码算法设计

空间耦合LDPC码因其在次最优迭代译码算法下能够达到最大后验概率译码性能而引起广泛关注。但其优异的阈值特性需要在码长很长时才能实现，当采用传统的迭代译码算法时，实现的复杂度将以指数增加，无法应用。为了有效降低传统迭代译码算法的复杂度，本文结合深度学习技术提出了一种空间耦合LDPC码的深度迭代译码算法。通过在消息传递过程中引入权重系数并采用深度神经网络对其进行训练获取权重系数，以此优化消息的可靠性度量值，从而提升译码收敛速度，降低译码复杂度。仿真结果表明，所提出的深度迭代译码算法可以以较少的迭代次数达到与传统迭代译码算法在较多迭代次数下的译码性能。

Deep Iterative Decoding Algorithm for Spatially Coupled LDPC Codes

Spatially coupled LDPC (SC-LDPC) codes attract much attention since they can achieve the maximum a posterior (MAP) decoding performance when using the sub-optimal iterative decoding algorithm. However, the excellent performances can be realized with large blocklength. When using the traditional iterative decoding algorithm, the complexity will be increased exponentially, which makes it impractical. To reduce the complexity, this paper presents one deep iterative decoding algorithm for SC-LDPC codes combined with deep learning technique. Assign weights to edges during message passing and train them using deep neural network, which can optimize the reliability values of the edges. Further, the decoding convergence can be accelerated and the decoding complexity can be reduced. Simulation results show that the proposed deep interative decoding algorithm with few iterations can achieve the comparable decoding performance of traditional decoding algorithm with many iterations.