Prediction of deep learning for spectrum-pulse width on petawatt laser

Abstract

The spectrum is a crucial parameter to a petawatt laser which is adopting the chirped pulse amplification technique. In such complex systems with high gain and wide spectrum bandwidth, the shape of the spectrum is crucial to the final output pulse width. In daily operation, the width of the compressed pulse will have some abnormal fluctuation, and the shape of the spectrum before compressed is also changed at the same shot. It will mislead the power and intensity estimation in laser-matter interaction experiments. So far, no theory has been able to analyze the relationship between spectrum and pulse width completely. Because it is hard to describe the fluctuation of the compressed pulse width which the online measure spectral phase in the high power laser system is difficult. In this paper, we first found and analyzed the relation between spectral variation and pulse width in the petawatt laser. With the support of existing data, we establish an end-toend deep learning model to map the petawatt laser’s spectrum before the compressor to the compressed pulse width. The deep learning scheme which based on Bayesian Neural Network (BNN) can provide an estimate of uncertainty as a function of pulse width to improve the accuracy of the model. After 20000 iterations, the Mean Square Error (MSE) is reduced to 0.08 in the validation test. Under the experiment, the model realizes an effective predict of the compressed pulse width. With the help of deep learning, we can get more information on the spectrum rather than the center wavelength and spectrum width to predict the compressed pulse width. It should be emphasized that this method will help to avoid unstable pulse output caused by an abnormal spectrum and to improve the operating efficiency of the petawatt laser system.