基于全卷积神经网络的机车信号降噪

邢玉龙; 王剑; 赵会兵; 朱林富

doi:10.3969/j.issn.0258-2724.20191111

基于全卷积神经网络的机车信号降噪

doi: 10.3969/j.issn.0258-2724.20191111

邢玉龙^1,2,,
王剑^1,2, ,,
赵会兵³,
朱林富^4,5

基金项目: 国家自然科学基金重大项目（61490705）

详细信息

作者简介:
邢玉龙（1992—），男，博士研究生，研究方向为交通信息工程及控制，E-mail：xingyulong@bjtu.edu.cn

通讯作者:
王剑（1978—），男，教授，研究方向为轨道交通自动化与控制，E-mail：wangj@bjtu.edu.cn

中图分类号: U284
计量
- 文章访问数: 502
- HTML全文浏览量: 207
- PDF下载量: 26
- 被引次数: 0
出版历程
- 收稿日期: 2019-11-18
- 修回日期: 2020-05-18
- 网络出版日期: 2021-01-11
- 刊出日期: 2021-04-15

Cab Signal Denoising Process Based on Fully Convolutional Networks

XING Yulong^{1,2
,},
WANG Jian^{1,2
, ,},
ZHAO Huibing³,
ZHU Linfu^4,5

摘要

摘要: 机车信号从钢轨提取轨道电路信号作为行车凭证，其译码输出性能对列控系统的可靠性和安全性有直接影响. 但列车运行过程中，机车信号不可避免地混入大量噪声和干扰，译码前需要降噪以提高准确性. 为此，提出一种基于全卷积神经网络（fully convolutionalnetworks, FCN）的机车信号降噪方法，该方法利用基于原始波形“端到端”处理方式的FCN，直接从时域对机车信号进行降噪处理，以提高信噪比（signal-to-noise ratio，SNR）；并利用仿真和实测数据对本方法进行了实验. 结果表明：相较于传统基于频谱的滤波方法，本方法对带内干扰有更显著的效果，采用FCN能使机车信号信噪比提高8～14 dB，可有效降低带内噪声.
- 铁路信号 /
- 列控系统 /
- 机车信号 /
- 信号处理 /
- 神经网络
Abstract: Since cab signals extract information from track circuits as the running token, its decoding performance has a direct impact on the reliability and security of train operation control system. However, as it is inevitable that a lot of noise and interference will mix into the cab signal during operation, it is necessary to denoise before decoding in order to improve demodulation accuracy. To this end, a raw waveform-based fully convolutional network (FCN) for denoising is proposed in an end-to-end manner, which denoises the cab signal in time domain directly and improves the signal-to-noise ratio (SNR). This proposed network is validated through simulation and measured data. The experimental results show that compared with the traditional spectrum-based denoising methods, this method has a more significant effect on in-band interference; FCN can improve the SNR of cab signals by 8～14 dB and effectively reduce the in-band interference.
- railway signal /
- train operation control system /
- cab signal /
- signal processing /
- neural networks

HTML全文

图 1 全连接层结构

Figure 1. Fully connected layer structure

下载: 全尺寸图片幻灯片

图 2 局部连接

Figure 2. Local connection

下载: 全尺寸图片幻灯片

图 3 FCN降噪模型整体结构

Figure 3. Overall structure of denoising model based on FCN

下载: 全尺寸图片幻灯片

图 4 神经网络结构

Figure 4. Network structure

下载: 全尺寸图片幻灯片

图 5 载频1.7 kHz机车信号降噪前、后频谱

Figure 5. Spectrum of cab signal with 1.7 kHz carrier frequency befor and after denoising

下载: 全尺寸图片幻灯片

图 6 一起信号误译示例

Figure 6. Example of signal decoding error

下载: 全尺寸图片幻灯片

图 7 误译信号降噪前、后频谱

Figure 7. Spectrum of decoding-error signal befor and after denoising

下载: 全尺寸图片幻灯片

表 1 信息特征参数

Table 1. Information characteristic parameters Hz

参数		指标
${f_{\rm{c}}}$	上行	1998.7，2001.4，2598.7，2601.4
${f_{\rm{c}}}$	下行	1698.7，1701.4，2298.7，2301.4
${f_{\rm{d}}}$		${\rm{10} }{\rm{.3 + 1} }{\rm{.1} } n{\simfont\text{，} }n = 0{\simfont\text{～} } 17$
$\Delta f$		$ \pm 11$

下载: 导出CSV

表 2 样本的噪声特性

Table 2. Noise characteristics of samples

噪声类型	相关参数		样本/个	样本长度/点
噪声类型	频率/Hz	幅度/V	样本/个	样本长度/点
单频噪声	1650～2650	0～2	25000	2048
谐波干扰	1650、1700、···、2650	0～1	25000	2048
带内干扰	${f_{\rm{c}}} - {f_{\rm{d}}}{\simfont\text{～} } {f_{\rm{c}}} + {f_{\rm{d}}}$	0～2	25000	2048
白噪声		0～10	25000	2048

下载: 导出CSV

表 3 本文算法RMSE结果与其他算法对比

Table 3. RMSE comparison with various algorithms V

噪声类型	去噪前	FCN	CNN	带通滤波	EMD	稀疏分解
单频噪声	0.593	0.244	0.430	0.493	0.386	0.317
谐波干扰	1.287	0.118	0.384	0.547	0.498	0.409
白噪声	1.199	0.264	0.321	0.654	0.332	0.300
带内干扰	0.859	0.236	0.287	0.719	0.603	0.580
全体噪声	0.945	0.169	0.198	0.686	0.554	0.526

下载: 导出CSV

表 4 本文算法SNR结果与其他算法对比

Table 4. SNR comparison with various algorithms dB

噪声类型	去噪前	FCN	CNN	带通滤波	EMD	稀疏分解
单频噪声	3.197	8.874	4.394	3.338	5.496	6.009
谐波干扰	−2.756	14.214	5.331	2.235	4.477	4.848
白噪声	−2.930	9.734	9.172	0.832	8.672	9.037
带内干扰	0.203	11.943	11.218	2.493	5.492	5.984
全体噪声	0.030	13.012	12.164	3.138	4.360	4.896

下载: 导出CSV

参考文献(17)

邱宽民. JT1-CZ2000型机车信号车载系统[M]. 北京: 中国铁道出版社, 2010: 1-12.

ZHAO Linhai, LI Zhankui, LIU Weining. The compensation capacitors fault detection method of jointless track circuit based on DBWT and WR[C]//IEEE International Conference on Intelligent Computing & Intelligent Systems. Shanghai: IEEE, 2009: 875-879.

剌博. 基于EMD降噪的轨道移频信号检测算法研究[D]. 西安: 西安理工大学, 2014.

轩春霞,王小敏,杨扬,等. 基于稀疏分解的轨道移频信号降噪算法研究[J]. 计算机测量与控制,2014,22(9): 2870-2874. doi: 10.3969/j.issn.1671-4598.2014.09.048

XUAN Chunxia, WANG Xiaomin, YANG Yang, et al. Denoising algorithm for track circuit frequency:shift signal based on sparse decomposition[J]. Computer Measurement & Control, 2014, 22(9): 2870-2874. doi: 10.3969/j.issn.1671-4598.2014.09.048

NAIK D C, MURTHY A S, NUTHAKKI R. Modified magnitude spectral subtraction methods for speech enhancement[C]//2017 International Conference on Electrical, Electronics, Communication, Computer, and Optimization Techniques（ICEECCOT）. Mysuru: [s.n.], 2017: 274-279.

EPHRAIM Y, MALAH D. Speech enhancement using a minimum-mean square error short-time spectral amplitude estimator[J]. IEEE Transactions on Acoustics, Speech, and Signal Processing, 1984, 32(6): 1109-1121.

SCALART P, FILHO J V. Speech enhancement based on a priori signal to noise estimation[C]//IEEE International Conference on Acoustics. Atlanta: IEEE, 1996: 629-632.

IBARROLA F J, DI PERSIA L, SPIES R D. A Bayesian approach to convolutive nonnegative matrix factorization for blind speech dereverberation[J]. Signal Processing, 2018, 151(4): 89-98.

HOU J C, WANG S S, LAI Y H, et al. Audio-visual speech enhancement using multimodal deep convolutional neural networks[J]. IEEE Transactions on Emerging Topics in Computational Intelligence, 2018, 2(2): 117-128.

PALAZ D, COLLOBERT R, DOSS M M. Estimating phoneme class conditional probabilities from raw speech signal using convolutional neural networks[C]//14th Annual Conference of the International Speech Communication Association. Lyon: Interspeech, 2013: 1765-1769.

OORD A V D, DIELEMAN S, ZEN H, et al. WaveNet: a generative model for raw audio[J]. Computer Science, 2016, 1: 1-15.

FU S W, WANG T W, TSAO Y, et al. End-to-end waveform utterance enhancement for direct evaluation metrics optimization by fully convolutional neural networks[J]. IEEE/ACM Transactions on Audio Speech & Language Processing, 2018, 26(9): 1570-1584.

赵自信. ZPW—2000A无绝缘移频自动闭塞系统的技术综述[J]. 铁路通信信号工程技术, 2003, 2003（增刊1）: 12-19.

ZHAO Zixin. A review of ZPW-2000 automatic block with jointless frequency-shift system[J]. Railway Signalling & Communication Engineering, 2003, 2003(S1): 12-19.

中华人民共和国铁道部. ZPW-2000轨道电路技术条件: TB/T 3206—2008[S]. 北京: 中国铁道出版社, 2008

LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine, 2015, 39(4): 640-651.

THAKKAR V, TEWARY S, CHAKRABORTY C.Batch normalization in convolutional neural networks: a comparative study with CIFAR-10 data[C]//2018 Fifth International Conference on Emerging Applications of Information Technology (EAIT). Kolkata: [s.n.], 2018: 1-5.

ZHANG Yudong, PAN Chichun, SUN Junding, et al. Multiple sclerosis identification by convolutional neural network with dropout and parametric ReLU[J]. Journal of Computational Science, 2018, 28(9): 1-10.

施引文献

附加材料(0)

访问统计