Fig. 5: Results of laboratory tests of the photonic lantern wavefront sensor.
Shown here are the predicted Zernike coefficients (crosses) and the true values (black lines) for a randomly selected set of 40 measurements. Red points are predictions from a model trained with 48,000 measurements, green points with 4800 measurements and blue with 480 measurements. The difference between the predicted and true values is plotted at the bottom of each panel. Each measurement consists of a combination of the first 10 Zernike terms each with a randomly chosen amplitude between approximately −0.12π and 0.12π radians applied to the SLM. Resulting combined wavefronts for each measurement have peak-to-valley amplitudes of order π radians (limited by SLM hardware). Predictions are performed by the neural network described in the text, using the 19 output intensities of the lantern. The neural network accurately predicts the Zernike terms of the wavefront injected into the lantern, with a root-mean-squared-error of 5.1 × 10−3 π radians.
