语义分割网络下的混合信号频谱分离

Spectra Separation via Semantic Segmentation Network

单通道接收机下,多个时频混合信号的分离属于非稀疏欠定信号分离问题,难以求解。针对这类非稀疏欠定信号分离问题,提出了一种基于语义分割网络、从频域实现多个指定类别信号分离的新方法。利用语义分割网络提取信号的频域分布特征,克服了单通道接收机下信号先验信息过少的问题。仿真表明,该方法具有较高的分离精度,且响应时间短,可用于单通道接收机中时频混叠信号的分离。

Separating multiple time-frequency aliasing signals via a single channel receiver is a non-sparse underdetermined signal separation(NUSS) problem,which is hard to tackle. In this paper,a new semantic segmentation network(SSN) based method is proposed to separate multiple specific time-frequency aliasing signals in frequency domain. By using a well-designed SSN,the frequency distribution features of each signal component are extracted to combat with the lack of prior information of signal sources via a single channel receiver. Simulation results show that the proposed method achieves desirable separation accuracy with short response time,which can be applied to multiple time-frequency aliasing signals’ separation for a single channel receiver.