结合深度循环神经网络的单通道盲分离

Single-channel Blind Separation Combined withDeep Recurrent Neural Network

单通道信号盲分离技术一直是现代信号处理领域的研究热点。基于逐幸存路径处理(Per-Survivor Processing,PSP)的盲分离算法,能从混合信号中估计出2路发送的符号序列,缺点是计算复杂度很高。从降低算法复杂度的角度,提出了一种基于双向循环神经网络的分离方法,经过端到端的学习,可以直接从2路同频调制信号单通道混合信号中分离出2路源信号。对于超过网络输入长度的序列,提出了前后额外截取的策略,解决了每块数据在开始和结束时产生高错误率的问题。在不降低分离性能的情况下,该方案的计算复杂度低于PSP算法。

Single-channel signal blind separation has been a hot research topic in the field of modern signal processing.The blind separation algorithm based on Per-Survivor Processing(PSP)can estimate two channels of transmitted symbol sequences from mixed signals,which has the disadvantage of high computational complexity.In order to reduce the complexity of the algorithm,a separation method based on a bidirectional recurrent neural network is proposed,which can directly separate two channels of source signals from a single-channel mixed co-frequency modulated signal after end-to-end learning.For the signal sequences that exceed the input length of the network,a strategy for additional pre-and post-interception of the signal is proposed to solve the problem of high error rates at the beginning and end of each block of data.Without degrading the separation performance,the computational complexity of the proposed method is lower than that of the PSP algorithm.