Data-Driven Machine Learning Analysis of Pulse Compression in Praseodymium-Doped Tapered Photonic-Crystal Fiber

We present a realistic numerical model based on data-driven machine-learning (ML) techniques for generating high-energy ultrashort pulses at 4.5 μm. To achieve this objective, we propose a nonlinear amplifying loop mirror (NALM) configuration employing a self-similarly designed Praseodymium doped As2S3(Pr3+: As2S3) tapered photonic crystal fiber (TPCF) compressor, which enables simultaneous pulse amplification and compression. Owing to the increasing complexity of Pr3+: As2S3TPCF design, arising from the large parameter space associated with both fiber geometry and pulse characteristics, conventional numerical optimization becomes computationally expensive. Therefore, ML techniques are employed to optimize the TPCF structure and the input pulse parameters. Using the optimized design, we demonstrate that a 5 ps pulse at 4.5 μm can be compressed to 282.6 fs after propagation through 1.65 m of fiber, corresponding to a maximum compression factor of 17.68, with a pedestal energy of 2.217%.