Fast dispersion-scan retrieval: neural networks vs iterative algorithm

Area of Science:

• Optics and Photonics
• Ultrafast Laser Science
• Computational Physics

Background:

• Dispersion-scan (d-scan) is an inline technique for optical pulse characterization.
• Single-shot trace acquisition enables rapid data collection.
• Fast retrieval algorithms are crucial for real-time applications.

Purpose of the Study:

• To extend neural networks (NNs) for simultaneous spectrum and spectral phase retrieval from d-scan traces.
• To compare the speed and accuracy of NNs against optimized iterative algorithms.
• To advance near single-shot and real-time optical pulse characterization.

Main Methods:

• Implementation of advanced neural network architectures.
• Simultaneous retrieval of spectrum and spectral phase from d-scan data.
• Comparative analysis of retrieval speed and quality using a benchmark iterative algorithm.

Main Results:

• Neural networks demonstrated comparable or superior retrieval quality to iterative methods.
• NNs achieved significantly faster processing times, enabling near real-time analysis.
• The extended NNs successfully retrieved both spectral amplitude and phase information.

Conclusions:

• Neural networks provide a viable and efficient alternative for d-scan data analysis.
• The developed NN approach accelerates optical pulse characterization.
• This work paves the way for advanced real-time monitoring in ultrafast optics.