利用交叉偏振波濾波與基於機器學習脈衝診斷�提升雙 CPA 前端架構之中央大學一百兆瓦�雷射系統時間對比

2026/06/25

研究生：楊世帆

指導教授：白植豪 教授

Temporal Contrast Enhancement of the NCU 100-TW

Laser System through a Double-CPA Front End Incorporating Cross-Polarized Wave Filtering and Machine-Learning Pulse Diagnostic

Outline

1

• Introduction

​

• Theory
• Double CPA
• XPW process
• Machine Learning for Dispersion Control

​

• Construction of Front-End of the Laser System
• Overview of NCU 100-TW Laser
• Setup of Double CPA – (Pulse stretcher, Pulse compressor, XPW)

​

• 100-TW Laser System After Upgrade
• Ring Regenerative Amplifier
• XPW
• NCU 100-TW Beam Line

​

• Dispersion Retrieval with Machine Learning for Single-shot Autocorrelator (SSA)

​

• Conclusion and Prospects

​

2

Introduction

Laser evolution

Source:

[1] Laser Programs The First 25 Years… 1972–1997 https://www.osti.gov/servlets/purl/16710

[2] F. J. McClung, R. W. Hellwarth; Giant Optical Pulsations from Ruby. J. Appl. Phys. 1 March 1962; 33 (3): 828–829.

[3] Shinji Yamashita, Sze Yun Set, Advances and challenges of mode-locked fiber lasers, Optics Communications, Volume 578, 2025, 131406

[1]

[2]

[3]

3

Introduction

Chirped Pulse Amplification – CPA (1985)

4

Introduction

Laser evolution

5

Introduction

Definition of the temporal contrast

(Not Enough for experiment)

Double CPA

6

Theory - Double CPA

CPA 1

Nonlinear Contrast Filter

CPA 2

Ring Regenerative

Amplifier

Cross-Polarization Wave

generation (XPW)

Multi-pass

Amplifier

Effects that would affect the spectral bandwidth in CPA amplifier

Gain narrowing ↑

Dynamic gain saturation ↑

7

Theory - Double CPA

Spectral Shaping Filter

Source: Alphine Research Optics.

The depth of the dip is control by the angle between incident polarization.

​

The position of the dip is control by the incident angle.

8

Theory - Double CPA

Gain / Loss / Transmission Spectrum ↑

Transmission Spectrum of New Polarizer

9

Theory - Double CPA

Without spectral shaping filter

With spectral shaping filter

38 nm

48 nm

Output spectrum bandwidth: 38 nm

Output Energy: 1.7 mJ

Output spectrum bandwidth: 48 nm

Output Energy: 1.3 mJ

Experiment measured maximum spectrum bandwidth: 51 nm

10

Theory - XPW process

With Maxwell’s equations, the coupled-wave equation for XPW

12

Theory - XPW process

Numerical calculation of the coupled-wave equation for XPW

Initial Conditions

13

Theory - XPW process

Split-Step Fourier Method (SSFM)

14

Theory - XPW process

Compare of [001]-cut and [011]-cut

[001]-cut

[011]-cut

15

Theory - XPW process

[011]-cut crystal with different input intensity

16

Theory - XPW process

Spectrum change

When the intensity increase, self-phase modulation would dominant the spectrum broadening.

Spectrum unstable, self focusing, etc.

The XPW spectrum remain the same when the input intensity is low.

17

Theory - Dispersion Control via Machine Learning

Convolutional Neural Network (CNN)

Convolutional Layer – Extract features from the input data

Pooling Layer – Reduce the dimensionality of the data

Fully connected Layer – Combined all the features

Training: (Input Data, Output Data) → Feature Tensor (model)

18

Theory - Dispersion Control via Machine Learning

Dual-Input Convolutional Neural Network (CNN)

2^nd, 3^rd, 4^th dispersion

19

Theory - Dispersion Control via Machine Learning

Data generation

Amplitude

Amplitude from superposition of random gaussian function with noise ↑

20

Theory - Dispersion Control via Machine Learning

Data generation

21

Theory - Dispersion Control via Machine Learning

After Training

22

Construction of Laser System - Overview

23

Construction of Laser System - Overview

76 MHZ

10 HZ

10 HZ

2 HZ

0.5 mJ (1%)

2 mJ

1 mJ

200 mJ

3J

CPA 1

CPA 2

24

Construction of Laser System - Pulse stretcher

Offner-type pulse stretcher

25

Construction of Laser System - Pulse compressor

26

Construction of Laser System - Ring Regenerative Amplifier

27

Construction of Laser System - XPW

28

100-TW After Upgrade - Ring Regenerative Amplifier

Pump beam on crystal

Total energy: 4.3 (mJ)

29

100-TW After Upgrade - Ring Regenerative Amplifier

Seed on crystal

30

100-TW After Upgrade - Ring Regenerative Amplifier

Loss of Regenerative Cavity

Loss = 5.2 %

Output Energy = 1.2 mJ / pulse

Without Spectral Shaping Filter

With Spectral Shaping Filter

Loss = 14.4 %

Output energy = 0.56 mJ / pulse

31

100-TW After Upgrade - Ring Regenerative Amplifier

Spectrum Bandwidth with Spectral Shaping Filter

Without Spectral Shaping Filter

With Spectral Shaping Filter

FWHM = 28.5 nm

Central Wavelength = 800.0 nm

FWHM = 51.7 nm

Central Wavelength = 796.6 nm

28.5 nm

51.7 nm

Minimum Pulse duration = 41.3 fs

Minimum Pulse duration = 30.1 fs

Avg. 100 shots

Avg. 100 shots

32

100-TW After Upgrade - Ring Regenerative Amplifier

Output Beam Profile

Before Beam Expander

After Beam Expander

Beam Diameter X: 2.4 mm

Beam Diameter Y: 2.2 mm

Beam Diameter X: 5.2 mm

Beam Diameter Y: 5.0 mm

33

100-TW After Upgrade – XPW

XPW Beam Profile

Before XPW

After XPW

Beam Diameter X: 1.7 mm

Beam Diameter Y: 1.7 mm

Beam Diameter X: 2.3 mm

Beam Diameter Y: 3.0 mm

34

100-TW After Upgrade - XPW

Single Crystal

35

100-TW After Upgrade - XPW

XPW with different input intensity

Measurement error due to the energy meter.

36

100-TW After Upgrade - XPW

XPW spectral bandwidth with different input intensity (Low intensity)

S = 0.77, FWHM = 50.6 nm

S = 1.12, FWHM = 47.4 nm

S = 1.35, FWHM = 48.1 nm

S = 1.67, FWHM = 48.7 nm

37

100-TW After Upgrade - XPW

XPW spectral bandwidth with different input intensity (High intensity)

S = 2.08, FWHM = 57.1 nm

S = 2.27, FWHM = 63.5 nm

S = 2.66, FWHM = 73.4 nm

S = 2.79, FWHM = 82.7 nm

S = 2.49, FWHM = 68.7 nm

38

100-TW After Upgrade - XPW

XPW spectral bandwidth with different input intensity

39

100-TW After Upgrade - XPW Contrast

ns-contrast with XPW

Improve 3 order

Before XPW

After XPW

40

100-TW After Upgrade - XPW Contrast

ps-contrast with XPW

Before XPW

After XPW

Improve 2 order

41

100-TW After Upgrade - XPW Contrast

Final XPW output

42

100-TW After Upgrade - 100-TW Beam Line

Spectrum and Pulse Duration Comparation

Spectrum FWHM = 31.5 nm

Pulse duration = 41 fs

Spectrum FWHM = 39.0 nm

Pulse duration = 37 fs

Source:

[1] EKSMA optics https://eksmaoptics.com/out/media/EKSMA_Optics_Thin_Film_Polarizers-56.pdf

[1]

43

100-TW After Upgrade - 100-TW Beam Line

100-TW Contrast after upgrade

(Close to the instrument measurement limitation)

44

Dispersion Retrieval with Machine Learning for SSA

Test With Fused Silica

Dispersion of Fused Silica

Transform limit add Fused silica

45

Dispersion Retrieval with Machine Learning for SSA

Machine Learning Input Spectrum

46

Dispersion Retrieval with Machine Learning for SSA

Test two Different Training Model – Different Trace Range

Model 1: -400 fs to 400 fs

0 mm ↑

1.6 mm ↑

8 mm ↑

16 mm ↑

47

Dispersion Retrieval with Machine Learning for SSA

Test two Different Training Model – Different Trace Range

Model 2: -75 fs to 75 fs

0 mm ↑

1.6 mm ↑

8 mm ↑

16 mm ↑

48

Conclusion and Prospects

Dual-Input CNN

With model 2, the prediction is close to the measurement result.

​

Problems

Resolving the absolute sign ambiguity of third-order and fourth-order dispersion within the experimental noise floor remains a limitation

Result and Conclusion

49

Conclusion and Prospects

Future Prospects - High Field Experiment

Proton Acceleration Experiment (TNSA)

Maximum Proton Energy: 3 MeV → 8 MeV (With the same conditions)

Future Publication

Optics Express - Construction of Double CPA and XPW, Enhancement of Temporal Contrast.

Thanks for listening

2

Reference