A new generative AI based for modelling free space optical links

Deep learning (DL) networks are being extensively utilized for modeling wireless fading channels. A different multipath channel is created when optical transmission is done wirelessly in free space optical (FSO) systems. The promising nature of FSO communication systems are restricted by the presence of atmospheric turbulence and pointing errors, the mathematical modeling of which is typically complex. In order to utilize the generalization capability of the DL systems, the recently introduced generative adversarial networks (GANs) can be invoked to learn the statistical nature of atmospheric turbulence and pointing error. Therefore, in the present research work, the new DL network namely DGG-GAN has been implemented to follow the varying channel conditions of the FSO systems. Since the GAN systems are made up of two DL networks (generator and discriminator) which compete with each other in the duration of training the GAN, the samples generated by the network can follow the real data closely. The DGG-GAN has been trained with real data set (following the double generalized Gamma (DGG) distribution) offline and then evaluated online to measure performance measures such as outage probability, ergodic capacity and bit error rate of the FSO systems. The trained DGG-GAN DL model has been tested to give mean square error (MSE) of \(MSE=-64.1585\) $MSE=-64.1585$ dB and root mean square error (RMSE) of \(RMSE=-11.7925\) $RMSE=-11.7925$ dB, validating the accuracy of the model. The trained model is utilized to evaluate the FSO link performance where all strengths of turbulences have been captured. The generalization capability of the trained model has been established in the simulation results.