Frequency Resolved Optical Gating (FROG) is the most widespread method for ultrafast pulse measurement. Due to the bandwidth limit of photodetectors, we take an indirect measurement of the pulse , and retrieve the original pulse using a reconstruction algorithm. Among these algorithms, SVD FROG is the most common one, but it needs a full spectrogram and suffers from restrictions on the number of measurements. However, a more liberal algorithm recently applied, Ptychography FROG takes any partial measurements as long as they contain high SNR. These methods are able to reconstruct ultrafast pulses, but they all suffer from severe time consumption (tens of seconds), without having the ability to reconstruct the pulse on the fly (fractures of a second). We propose using deep neural network algorithms to reconstruct ultrashort pulses from FROG measurements, without losing any of the advantages above while increasing time performance to a video applicable level. This project aims to become an avant-garde step of deep neural network implementation in photonics.

We develop a revolutionary approach for ultrahigh-speed high-resolution microscopy. We invite excellent and highly motivated undergraduate students to join this research direction. Experimental and/or algorithmic projects are available.