Lightweight CNN-based eye diagrams denoising autoencoder for 16-QAM IsOWC system

A lightweight convolutional neural network (CNN)-based denoising autoencoder (DAE) model is proposed to efficiently suppress noise degradations from a 16-quadrature amplitude modulated (16-QAM) inter-satellite optical wireless communication (IsOWC) system. In such a system, the quality of the received optical signals often shown by the eye diagrams (EDs) is usually limited due to the presence of various dominant noise sources. These consist of pointing error due to the noncoincidence of the optical beams, amplified spontaneous emission noise from the optical amplifiers, and the relative intensity noise of laser sources. The cumulative impact of all these impairments distorts the ED and thus deteriorates the data integrity and induces a high bit error rate (BER). The model is trained on an extensive dataset of 52,500 noisy EDs with different noise variances in the range of 0.05 to 0.25 with a step size of 0.05, which can enable the model to adapt itself and generalize for various conditions. Moreover, a different hyperparameter tuning is rigorously investigated over the training model. It achieves a peak signal-to-noise ratio of 41.720 dB, a structural similarity index of 0.9884, a lower BER of −4.84, and an image reconstruction error of 03×10−6. These results provide significant conversion in ED clarity that improves both signal quality and data fidelity during the reconstruction. Hence, this lightweight DAE-based CNN can be directly integrated into the IsOWC system and potentially introduces a high-speed, high-performance, and highly efficient adaptivity-dynamical spectrum quantity to facilitate the satellite internet technology with global broadband capability.