Ultrashort Chirped Pulse Amplification in Fiber Based on Deep Learning

Hao Sui; Hongna Zhu; yan Zhang; Bin Luo; Xihua Zou

doi:10.3788/AOS221454

Acta Optica Sinica, Volume. 43, Issue 3, 0319001(2023)

Ultrashort Chirped Pulse Amplification in Fiber Based on Deep Learning

Hao Sui¹, Hongna Zhu^1、*, yan Zhang², Bin Luo³, and Xihua Zou³

Author Affiliations

¹School of Physical Science and Technology, Southwest Jiaotong University, Chengdu 610031, Sichuan, China

²National Key Laboratory of Science and Technology on Blind Signal Processing, Chengdu 610041, Sichuan, China

³School of Information Science & Technology, Southwest Jiaotong University, Chengdu 610031, Sichuan, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Ultrashort pulse chirp amplification process. (a) Signal pulse propagation; (b) pump pulse propagation

Download full size

Fig. 2. Normalized intensity of signal pulse at different propagation distance. (a) 100 m; (b) 200 m; (c) 300 m; (d) 400 m

Download full size

Fig. 3. Predicting ultrashort pulse chirp amplification based on deep learning. (a) Deep convolutional network structure; (b) convolutional block structure

Download full size

Fig. 4. Normalized error on testing set for different $C_{s}$

Download full size

Fig. 5. Signal pulse propagation process when $C_{s}$ =-9. (a) Predicted result of network; (b) theoretical result; (c) difference between predicted result and theoretical result

Download full size

Fig. 6. Normalized error on testing sets in cases of different initial FWHM and $C_{s}$ , different initial pulse power and $C_{s}$ , and different initial FWHM, power, and $C_{s}$

Download full size

Fig. 7. Signal pulse propagation when FWHM, peak power, and $C_{s}$ of initial pulse are 8.8 ps, 0.02 mW, and -4, respectively. (a) Predicted result; (b) theoretical result; (c) difference between predicted result and theoretical result

Download full size

Table 1. Typical fiber parameters used in simulation
View table
Table 1. Typical fiber parameters used in simulation
Nonlinear coefficient /（km· W） Fiber loss coefficient /km
Dispersion slope /
（ps·nm^-2·km^-1）
Zero dispersion wavelength /nm Pump wavelength /nm
Signal
wavelength /nm
15 0.05 0.02 1550 1553 1531

Table 2. Datasets with different signal pulse parameters
View table
Table 2. Datasets with different signal pulse parameters
Case Full width at half-maximum /ps Peak power /mW Chirp Total sample number for training Total sample number for testing
1 10 0.01 -10-10 161 40
2 8-12 0.01 -5-5 871 200
3 10 0.01-0.05 -5-5 871 200
4 8-12 0.01-0.05 -5-5 2041 500

Table 3. NRMSE on testing set under different cases
View table
Table 3. NRMSE on testing set under different cases
Epoch NRMSE
Case 1 Case 2 Case 3 Case 4
2000 0.1349 0.1343 0.1797 0.2010
6000 0.0435 0.0585 0.0774 0.0920
10000 0.0261 0.0400 0.0497 0.0584

Table 4. Normalized computing time on testing set for different cases
View table
Table 4. Normalized computing time on testing set for different cases
Method Normalized computing time
Case 1 Case 2 Case 3 Case 4
SSF（split-step Fourier method） 1.01 4.79 4.93 11.76
Proposed deep learning method 1 1.14 1.15 1.45

Tools

Get Citation

Copy Citation Text

Hao Sui, Hongna Zhu, yan Zhang, Bin Luo, Xihua Zou. Ultrashort Chirped Pulse Amplification in Fiber Based on Deep Learning[J]. Acta Optica Sinica, 2023, 43(3): 0319001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites