一种利用深度学习的非均匀无记忆信源恢复方法

Signal Recovery for Non-uniform Non-memory Sources Utilizing Deep Learning

针对非均匀无记忆信源这一特殊自然冗余信源的接收端符号恢复问题,基于全连接神经网络模型,设计一种将接收信号的信噪比和无记忆信源的符号分布随接收数据一起作为模型输入的神经网络译码器架构。并提出一种基于此神经网络的迭代译码算法,实现在发送符号分布未知情况下的自然冗余译码。仿真结果表明,利用自然冗余可以提升接收端的符号检测性能,即使在信源分布未知的情况下也能获得理论上最优的性能。

Aiming at the symbol recovery problem at the receiver side for the special natural redundancy sources of non-uniform non-memory sources,a neural network decoder architecture is proposed,which is based on the fully connected neural network model.The architecture incorporates the signalto-noise ratio of the received signal and the symbol distribution of the memoryless source along with the received data as inputs to the model.An iterative decoding algorithm based on this neural network model is proposed to realize the natural redundancy decoding in the case of unknown distributions of transmitted symbols.The simulation results show that the symbol detection performance at the receiver side can be improved by using natural redundancy.Moreover,the optimal performance can be theoretically obtained by the proposed algorithm,even when the source distribution is unknown.