![]() ![]() 5: Blue: the ideal amplitude mask to transform the input pulse into the desired pulse and orange: the ANN predicted amplitude mask.įig. ![]() The shaped output spectrum with the modelled amplitude mask indeed resembles the desired spectrum, as show in Fig. 5 shows that, once trained, the ANN can predict the ideal amplitude mask for the input spectrum and goal spectrum in Fig. 4: Neural network input and output when predicting the amplitude mask.įig. 3: Neural network input and output in the training stage.įig. After training, the ANN can take a concatenation of the input spectrum and the desired spectrum and output the corresponding mask.įig. The ANN is trained with a concatenation of the input and output vectors as the ANN input and the amplitude mask as the labelled output. This creates a triad of 1D vectors: the input spectrum, the pulse shaper output spectrum and the corresponding amplitude mask. The ANN is trained by sending input spectra through the pulse shaper using an estimated amplitude mask and measuring the pulse shaper output. The ANN is used to characterize the AOPDF transfer function, F(I,A), and predict the ideal mask, A, for a given input spectrum, I, and a desired output spectrum. These three factors make determination of the ideal mask vector a non-trivial task,Īnd the calculated mask vector generally differs from the ideal mask vector by some small amount, which could be enough to significantlyĬhange the shape of the masked pulse spectrum. Spectrometer and AOPDF can have non-linear responses. Wavelength vector and the spectrometer wavelength vector, requiring that the calculated mask is shifted or re-scaled. Secondly, there is generally a slight calibration off-set between the AOPDF So the square root of the measured spectrum must be used. First, the AOPDF masks the amplitude rather than the intensity, Ideally, the mask could be generated by dividing the desired Unfortunately, predicting the necessary mask is not always straightforward. 2: Blue: spectrum of pulse without an amplitude mask and orange: desired pulse spectrum. 2, blue line), 2) calculate the mask that would generate the desired pulse spectrum and 3) measure the pulse spectrum with the amplitude mask to confirm that the masked spectrum resembles the desired spectrum.įig. Pulse spectrum without an amplitude mask applied by the AOPDF (Fig. Generally, one desires the pulse spectrum to have a specific shape, such as a super-Gaussian (Fig. 1: Acousto-optic programmable dispersive filter (AOPDF) input and output. 1, the AOPDF accepts a 1D amplitude mask vector, A, and an input pulse spectrum, I, and outputs a shaped pulse spectrum, S.įig. The AOPDF is a literal black box which shapes the spectral amplitude and phase of a light pulse as a laser beam is transmitted Spectrum using an acousto-optic programmable dispersive filter (AOPDF), such as the Dazzler by Fastlite. Machine learning tools for spectral pulse shaping in pythonĪn artificial neural network (ANN) is used to determine the ideal amplitude mask that can transform a given pulse spectrum into a desired pulse
0 Comments
Leave a Reply. |