基于Stacked-TCN的空间混叠信号单通道盲源分离方法

针对空间互联网星地通信场景中的混叠信号分离精度不足问题, 提出了基于深度学习的堆叠时域卷积网络(stacked time-domain convolutional network, Stacked-TCN)分离方法。首先, 对混合信号提取编码特征表示。然后, 通过时域卷积网络训练得到源信号的深层特征掩模, 将每个信号源的掩模与混合信号编码特征做Hadamard乘积, 得到源信号的编码特征表示。最后, 使用1-D卷积, 对源信号特征进行解码, 得到原始波形。实验采用负的比例不变信噪比作为网络训练的损失函数, 即单通道盲源分离性能的评价指标。结果表明, Stacked-TCN方法与其他4种算法相比, 所提方法具有更好的分离精度和噪声鲁棒性。

Single-channel blind source separation method of spatial aliasing signal based on Stacked-TCN

Aiming at the problem of insufficient separation accuracy of aliased signals in space Internet satellite ground communication scene, a stacked time domain convolutional network (Stacked-TCN) separation method based on deep learning is proposed. Firstly, the coding feature representation is extracted from the mixed signal. Then, the deep feature mask of the source signal is trained through the time-domain convolution network. The mask of each signal source and the coding feature of the mixed signal are multiplied by Hadamard to obtain the coding feature representation of the source signal. Finally, 1-D convolution is used to decode the characteristics of the source signal to obtain the original waveform. In the experiment, the negative scale invariant source to noise ratio is used as the loss function of network training, that is, the evaluation index of single channel blind source separation performance. The results show that the Stacked-TCN method has better separation accuracy and noise robustness than the other four algorithms.