L-band InAs/InP quantum dash laser spatial OAM light modes classification under smoke environment: An image processing enhanced deep learning approach

In this work, we experimentally generated Laguerre Gaussian (LG) and its multiplexed form (Mux-LG) in the 1610 nm regime of the optical communication band employing InAs/InP quantum dash laser diode. Later, we investigated the detection of these spatial light modes encoding schemes under smoke channel conditions employing convolutional neural network (CNN) and uNET deep learning algorithms in conjunction with multiple received orbital angular momentum (OAM) modes images as input for the first time. We studied OAM modes classification and visibility estimation and reported identification accuracies of > 92% and > 96%, respectively, with uNET even for a challenging visibility range of 0-50 m. In general, exploiting the similarity of temporally successive images resulted in better performance of deep learning networks than just a single input image. Lastly, we propose a simple yet powerful image processing technique as a pre-processing stage for the received mode patterns for visibility estimation via deep learning regression and showed an improvement of similar to 4 m in root mean square error (RMSE).