Author Affiliations
1Key Laboratory of All Optical Network and Advanced Telecommunication Network, Ministry of Education, Institute of Lightwave Technology, Beijing Jiaotong University, Beijing 100044, China2School of Information and Communication Engineering, Communication University of China, Beijing 100024, Chinashow less
Fig. 1. Evolution of optical pulse in optical fiber. (a) Input and output power in time domain; (b) relative intensity of input and output power spectral density in frequency domain; (c) pulse evolution in time domain; (d) pulse evolution in frequency domain
Fig. 2. Schematic of detecting fiber parameters using neural network
Fig. 3. Schematic of optical fiber parameter measurement using different methods. (a) Complete signal measurement method; (b) power spectrum measurement method
Fig. 4. Output spectra corresponding to input pulses with different peak powers. (a) 600 W; (b) 800 W; (c) 1000 W
Fig. 5. Structure diagram of convolutional neural network
Fig. 6. Measurement results of optical fiber parameters using all information in decision tree algorithm. (a) Measurement results of ; (b) measurement results of D
Fig. 7. Measurement results of optical fiber parameters using all information in K-nearest neighbor algorithm. (a) Measurement results of ; (b) measurement results of D
Fig. 8. Measurement results of optical fiber parameters using all information in random forest algorithm. (a) Measurement results of ; (b) measurement results of D
Fig. 9. Measurement results of optical fiber parameters using all information in FCN algorithm. (a) Measurement results of ;(b) measurement results of D
Fig. 10. Measurement results of optical fiber parameters using all information in CNN algorithm. (a) Measurement results of ;(b) measurement results of D
Fig. 11. Measurement results of optical fiber parameters using power spectrum in decision tree algorithm. (a) Measurement results of ; (b) measurement results of D
Fig. 12. Measurement results of optical fiber parameters using power spectrum in K-nearest neighbor algorithm. (a) Measurement results of ; (b) measurement results of D
Fig. 13. Measurement results of optical fiber parameters using power spectrum in random forest algorithm. (a) Measurement results of ; (b) measurement results of D
Fig. 14. Measurement results of optical fiber parameters using power spectrum in FCN algorithm. (a) Measurement results of ;(b) measurement results of D
Fig. 15. Measurement results of optical fiber parameters using power spectrum in CNN algorithm. (a) Measurement results of ;(b) measurement results of D
Algorithms of machine learning | Loss of full information /% | Loss of power spectrum information /% |
---|
Decision tree algorithm | 0. 7405 | 0. 6594 | Random forest algorithm | 0. 2705 | 0. 1918 | K-nearest neighbor algorithm | 1.2716 | 0. 3289 | FCN algorithm | 0.8544 | 0. 2910 | CNN algorithm | 1.5244 | 0. 6045 |
|
Table 1. Loss of five machine learning algorithms in test set