基于前馈神经网络的分组码译码方案

在构造出分组码格图的基础上，利用一种基于前馈神经网络的多输入最小值选择网络实现分组码分组码的软判决及硬判决译码。计算结果表明，前馈神经网络总能找到全局最优解，从而使该译码算法的性能同于最大似然译码。由于该前馈网络的计算时延非常短，且基于它的译码器与传统译码器相比硬件实现简单，从而使译码的复杂性降低，时延减小。

Block codes decoding scheme based on the feedforward neural network

On the basis of the trellis diagram of linear block codes, a new neural network decoding method of linear block codes is presented, which uses the feedforward network to determine the minimum one of several inputs. The simulation result shows that this network can always find the global optimum, thus making the performance of the decoding method approach that of the ideal maximum likelihood decoding. Because the feedforward neural network has a very short delay and the decoder based on it can be easily implemented by hardware, compared with traditional decoding methods, this decoding method can greatly reduce its decoding complexity and delay.