基于Stacked-TCN的空间混叠信号单通道盲源分离方法

doi:10.12305/j.issn.1001-506X.2021.09.32

摘要/Abstract

摘要：

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。

关键词: 欠定盲源分离, 同频干扰, 单通道, 时域卷积网络

Abstract:

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.

Key words: underdetermined blind source separation, co-frequency interference, single channel, time-domain convolutional network (TCN)

中图分类号:

TN911.7

赵孟晨, 姚秀娟, 王静, 董苏惠. 基于Stacked-TCN的空间混叠信号单通道盲源分离方法[J]. 系统工程与电子技术, 2021, 43(9): 2628-2636.

Mengchen ZHAO, Xiujuan YAO, Jing WANG, Suhui DONG. Single-channel blind source separation method of spatial aliasing signal based on Stacked-TCN[J]. Systems Engineering and Electronics, 2021, 43(9): 2628-2636.

图/表 10

图1

图2

图3

表1

表2

表3

图4

图5

表4

表5

参考文献 49

1	肖楠, 梁俊, 王伟, 等. 卫星认知无线网络中频谱感知与分配策略设计[J]. 中南大学学报(自然科学版), 2015, 46 (12): 4520- 4528. doi: 10.11817/j.issn.1672-7207.2015.12.019
	XIAO N , LIANG J , WANG W , et al. Design of spectrum sensing and allocating strategy forsatellite cognitive radio networks[J]. Journal of Central South University (Science and Technology), 2015, 46 (12): 4520- 4528. doi: 10.11817/j.issn.1672-7207.2015.12.019
2	蒋丽丽, 陈国彬, 张广泉. 认知无线电网络中基于盲源分离的频谱检测算法[J]. 中国科技论文, 2017, 12 (8): 868- 872. doi: 10.3969/j.issn.2095-2783.2017.08.005
	JIANG L L , CHEN G B , ZHANG G Q . Blind source separation-based spectrum sensing in congnitive radio networks[J]. China Science Paper, 2017, 12 (8): 868- 872. doi: 10.3969/j.issn.2095-2783.2017.08.005
3	杨柳, 张杭. 通信中的盲源分离问题及解决方案探讨[J]. 通信技术, 2014, 47 (1): 1- 6.
	YANG L , ZHANG H . Discussion of blind source separation problem andits solution in communication[J]. Communications Technology, 2014, 47 (1): 1- 6.
4	LI C , ZHU L , LUO Z . Underdetermined blind source separation of adjacent satellite interference based on sparseness[J]. China Communications, 2017, 14 (4): 140- 149. doi: 10.1109/CC.2017.7927572
5	LI Y , WANG Y , DONG Q . A novel mixing matrix estimation algorithm in instantaneous under-determined blind source separation[J]. Signal Image and Video Processing, 2020, 14 (5): 1- 8.
6	XIE Y , XIE K , XIE S . Underdetermined blind source separation for heart sound using higher-order statistics and sparse representation[J]. IEEE Access, 2019, 7, 87606- 87616. doi: 10.1109/ACCESS.2019.2925896
7	DAVIES M E , JAMES C J . Source separation using single channel ICA[J]. Signal Processing, 2007, 87 (8): 1819- 1832. doi: 10.1016/j.sigpro.2007.01.011
8	BOFILL P , ZIBULEVSKY M . Underdetermined blind source separation using sparse representations[J]. Signal Processing, 2001, 81 (11): 2353- 2362. doi: 10.1016/S0165-1684(01)00120-7
9	ISIK Y, ROUX J L, CHEN Z, et al. Single-channel multi-speaker separation using deep clustering[EB/OL]. [2020-10-05]. http://arXiv.org/abs/1607.02173.
10	YU D, KOLBAEK M, TAN Z H, et al. Permutation invariant training of deep models for speaker-independent multi-talker speech separation[C]//Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing, 2017: 241-245.
11	KOLBAEK M , YU D , TAN Z H , et al. Multi-talker speech separation with utterance-level permutation invariant training of deep recurrent neural networks[J]. IEEE Trans.on Audio Speech and Language Processing, 2017, 25 (10): 1901- 1913. doi: 10.1109/TASLP.2017.2726762
12	DO H D , TRAN S T , CHAU D T . Speech separation in the frequency domain with autoencoder[J]. Journal of Communications, 2020, 15 (11): 841- 848.
13	YOSHⅡ K, TOMIOKA R, MOCHIHASHI D, et al. Beyond NMF: time-domain audio source separation without phase reconstruction[C]//Proc. of the International Society for Music Information Retrieval, 2013: 369-374.
14	FÉVOTTE C , IDIER J . Algorithms for nonnegative matrix factorization with the β-divergence[J]. Neural computation, 2011, 23 (9): 2421- 2456. doi: 10.1162/NECO_a_00168
15	何继爱, 宋宇霄. Kalman滤波下的多信号单通道盲源分离[J]. 信号处理, 2018, 34 (7): 843- 851.
	HE J A , SONG Y X . Blind source separation of the multi-signal single channelbased on Kalman filtering[J]. Signal Processing, 2018, 34 (7): 843- 851.
16	何继爱, 刘琳芝, 李英堂. LCL-FRESH滤波器实现单通道盲源分离[J]. 信号处理, 2014, 30 (2): 236- 242. doi: 10.3969/j.issn.1003-0530.2014.02.015
	HE J A , LIU L Z , LI Y T . Single-channel blind source separation achieved by the LCL-FRESH filter[J]. Signal Processing, 2014, 30 (2): 236- 242. doi: 10.3969/j.issn.1003-0530.2014.02.015
17	宋阳, 王翔, 马忠正, 等. CFE条件下基于循环维纳滤波的单通道分离算法[J]. 中国电子科学研究院学报, 2014, 9 (2): 186- 193. doi: 10.3969/j.issn.1673-5692.2014.02.014
	SONG Y , WANG X , MA Z Z , et al. Single channel signals separation algorithm based oncyclostationary filtering under CFE correction[J]. Journal of China Academy of Electronics and Information Technology, 2014, 9 (2): 186- 193. doi: 10.3969/j.issn.1673-5692.2014.02.014
18	LUO Y , MESGARANI N . Conv-TasNet: surpassing ideal time-frequency magnitude masking for speech separation[J]. IEEE Trans.on Audio, Speech, and Language Processing, 2019, 27 (8): 1256- 1266. doi: 10.1109/TASLP.2019.2915167
19	CHOLLET F. Xception: deep learning with depthwise separable convolutions[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 1251-1258.
20	LUO Y, CHEN Z, YOSHIOKA T. Dual-path RNN: efficient long sequence modeling for time-domain single-channel speech separation[C]//Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing, 2020: 46-50.
21	SHI Z, LIU R, HAN J. La furca: iterative context-aware end-to-end monaural speech separation based on dual-path deep parallel inter-intra bil-STM with attention[EB/OL]. [2020-10-05]. http://arXiv.org/abs/2001.08998.
22	SUBAKAN C, RAVANELLI M, CORNELL S, et al. Attention is all you need in speech separation[EB/OL]. [2020-10-05]. http://arXiv.org/abs/2010.13154.
23	HENNEQUIN R , KHLIF A , VOITURET F , et al. Spleeter: a fast and efficient music source separation tool with pre-trained models[J]. Journal of Open Source Software, 2020, 5 (50): 2154. doi: 10.21105/joss.02154
24	HAN C, LUO Y, LI C, et al. Continuous speech separation using speaker inventory for long multi-talker recording[EB/OL]. [2020-10-05]. http://arXiv.org/abs/2012.09727.
25	FAN C, TAO J, LIU B, et al. Deep attention fusion feature for speech separation with end-to-end post-filter method[EB/OL]. [2020-10-05]. http://arXiv.org/abs/2003.07544.
26	LIU Y, DELFARAH M, WANG D L. Deep CASA for talker-independent monaural speech separation[C]//Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing, 2020: 6354-6358.
27	NGUYEN V N, SADEGHI M, RICCI E, et al. Deep variational generative models for audio-visual speech separation[EB/OL]. [2020-10-05]. http://arXiv.org/abs/2008.07191.
28	SHI J, XU J, FUJITA Y, et al. Speaker-conditional chain model for speech separation and extraction[EB/OL]. [2020-10-05]. http://arXiv.org/abs/2006.14149.
29	WANG D L , CHEN J . Supervised speech separation based on deep learning: an overview[J]. IEEE Trans.on Audio, Speech, and Language Processing, 2018, 26 (10): 1702- 1726. doi: 10.1109/TASLP.2018.2842159
30	ZHAO J, GAO S, SHINOZAKI T. Time-domain target-speaker speech separation with waveform-based speaker embedding[C]//Proc. of the Interspeech, 2020: 1436-1440.
31	LUO Y, HAN C, MESGARANI N. Ultra-lightweight speech separation via group communication[EB/OL]. [2020-10-05]. http://arXiv.org/abs/2011.08397.
32	ZHANG L, SHI Z, HAN J, et al. FurcaNeXt: End-to-end monaural speech separation with dynamic gated dilated temporal convolutional networks[C]//Proc. of the International Conference on Multimedia Modeling, 2020: 653-665.
33	KAVALEROV I, WISDOM S, ERDOGAN H, et al. Universal sound separation[C]//Proc. of the IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, 2019: 175-179.
34	PRÉTET L, HENNEQUIN R, ROYO-LETELIER J, et al. Singing voice separation: a study on training data[C]//Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing, 2019: 506-510.
35	LUO Y, HAN C, MESGARANI N, et al. FaSNet: low-latency adaptive beamforming for multi-microphone audio processing[C]//Proc. of the IEEE Automatic Speech Recognition and Understanding Workshop, 2019: 260-267.
36	LLUÍS F, PONS J, SERRA X. End-to-end music source separation: is it possible in the waveform domain?[EB/OL]. [2020-10-05]. http://arXiv.org/abs/1810.12187.
37	BAHMANINEZHAD F, WU J, GU R, et al. A comprehensive study of speech separation: spectrogram vs waveform separation[EB/OL]. [2020-10-05]. http://arXiv.org/abs/1905.07497.
38	XU C, RAO W, XIAO X, et al. Single channel speech separation with constrained utterance level permutation invariant training using grid LSTM[C]//Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing, 2018: 6-10.
39	O'SHEA T J, CORGAN J, CLANCY T C. Convolutional radio modulation recognition networks[C]//Proc. of the International Conference on Engineering Applications of Neural Networks, 2016: 213-226.
40	XIONG R, YANG Y, HE D, et al. On layer normalization in the transformer architecture[C]//Proc. of the International Conference on Machine Learning, 2020: 10524-10533.
41	KHAN Z Y , NIU Z . CNN with depthwise separable convolutions and combined kernels for rating prediction[J]. Expert Systems with Applications, 2020, 170, 114528.
42	ISIK Y, ROUX J L, CHEN Z, et al. Single-channel multi-speaker separation using deep clustering[EB/OL]. [2020-10-05]. http://arXiv.org/abs/1607.02173.
43	王少波, 郭英, 眭萍, 等. 欠定条件下同步组网跳频信号盲源分离方法[J]. 计算机工程, 2020, 46 (10): 166- 172, 181.
	WANG S B , GUO Y , SUI P , et al. Under-determined blind source separation method for synchronous network frequency-hopping signals[J]. Computer Engineering, 2020, 46 (10): 166- 172, 181.
44	赵知劲, 黄艳波. 基于小波包分解的单通道盲源分离算法[J]. 通信技术, 2017, 50 (3): 425- 429. doi: 10.3969/j.issn.1002-0802.2017.03.007
	ZHAO Z J , HUANG Y B . Single-channel blind-source separation algorithm based on wavelet packetdecomposition[J]. Communications Technology, 2017, 50 (3): 425- 429. doi: 10.3969/j.issn.1002-0802.2017.03.007
45	郭一鸣, 杨勇, 张冬玲, 等. 基于SIC的单通道同频混合信号低复杂度盲分离算法[J]. 信号处理, 2015, 31 (7): 860- 866. doi: 10.3969/j.issn.1003-0530.2015.07.014
	GUO Y M , YANG Y , ZHANG D L , et al. Blind separation algorithm of single-channel co-frequencysignals with low complexity based on SIC[J]. Signal Processing, 2015, 31 (7): 860- 866. doi: 10.3969/j.issn.1003-0530.2015.07.014
46	张建中, 文树梁, 谭澄, 等. 盲源分离联合阻塞矩阵抗雷达主瓣干扰研究[J]. 现代防御技术, 2018, 46 (1): 135- 140. doi: 10.3969/j.issn.1009-086x.2018.01.022
	ZHANG J Z , WEN S L , TAN C , et al. Radar main-lobe jamming suppression based on the method of BSS-BM union[J]. Modern Defense Technology, 2018, 46 (1): 135- 140. doi: 10.3969/j.issn.1009-086x.2018.01.022
47	WENINGER F, ROUX J L, HERSHEY J R, et al. Discriminative NMF and its application to single-channel source separation[C]//Proc. of the 15th Annual Conference of the International Speech Communication Association, 2014: 865-869.
48	LUO Y, MESGARANI N. Tasnet: time-domain audio separation network for real-time, single-channel speech separation[C]//Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing, 2018: 696-700.
49	STOLLER D, EWERT S, DIXON S. Wave-u-net: a multi-scale neural network for end-to-end audio source separation[EB/OL]. [2020-10-05]. http://arXiv.org/abs/1806.03185.

序号	方法	信号种类	信噪比范围/dB	步长/dB
1	文献[43]	BPSK	5~30	5
2	文献[44]	BASK、BPSK	5~25	5
3	文献[45]	QPSK	12~22	2
4	文献[46]	雷达信号	10~30	0. 5
5	文献[7]	EEG、ECG	—	—
6	文献[47]	语音信号	-6~9	3
7	文献[48]	语音信号	—	—
8	文献[49]	语音信号	—	—
9	Stacked-TCN	BPSK、8PSK、16QAM、64QAM、PAM4	5~20	2.5

算法	参数	值
ICA	迭代次数	100
NMF	迭代阈值	1×10^－8
NMF	迭代次数	100
TasNet	基信号数N	128
	帧长L	64
	LSTM隐层单元数	128
	LSTM-block数X	2
Wave-U-Net	卷积核大小P	5
	block数X	5
	通道数	16-32-64-128-256
Stacked-TCN	编码器filter数N	512
	帧长L	16
	瓶颈层通道数B	128
	卷积核大小P	3
	1D-block通道数H	512
	1D-block数X	8
	Repeat数R	3

混合信号	算法
混合信号	Stacked-TCN	TasNet	Wave-U-Net	ICA	NMF
BPSK_8PSK	-22.51	-2.94	-17.22	5.36	9.92
BPSK_16QAM	-18.71	-4.06	-17.02	1.64	6.32
BPSK_64QAM	-22.83	-1.44	-19.08	6.02	8.72
BPSK_PAM4	-28.44	-2.35	-12.09	5.92	8.19
8PSK_16QAM	-13.00	-3.63	-13.55	4.87	9.82
8PSK_64QAM	-5.96	-1.98	-11.31	1.07	7.67
8PSK_PAM4	-15.64	-1.95	-18.67	2.68	6.13
16QAM_64QAM	-6.32	-2.51	-1.51	5.48	8.73
16QAM_PAM4	-21.61	-2.25	-17.31	5.32	7.79
64QAM_PAM4	-5.53	-2.36	-11.92	2.50	5.98
平均值	-16.05	-2.55	-13.97	4.09	7.93

算法	单帧时长	CPU所用时间
Stacked-TCN时域掩模法	0.032	7.90×10^－3
时-频域掩模法	0.256	1.24

算法	所用时间
NMF	0.38×10^－4
ICA	0.157 27
Wave-U-Net	0.202 59
TasNet	0.233 69
Stacked-TCN	0.334 49