Our research addresses significant challenges in optical wireless communications in space, which are adversely affected by atmospheric turbulence, light attenuation, and detector noise, leading to degraded communication reliability. To mitigate these issues, we have developed a neural network-based channel estimator optimized across a wide range of signal-to-noise ratio levels. The proposed estimator achieves performance comparable to the minimum mean square error estimator while maintaining reduced computational complexity. Additionally, we introduced a novel autoencoder framework incorporating advanced features such as layer normalization and multiple decoders. These enhancements improve receiver learning capabilities and bit error rate performance under both perfect and imperfect channel state information conditions.

The autoencoder framework is designed to handle multiple code rates across diverse fading channels, making it a scalable and adaptable solution for dynamic optical communication environments. Furthermore, as the Poisson channel is the most accurate model for optical communication, our work addresses the non-differentiability of Poisson optical communication channels by integrating the covariance matrix adaptation evolution strategy with autoencoders, achieving near-optimal bit error rate performance without relying on Gaussian approximations. We also propose enhanced autoencoder designs for medium access control and transport layer settings, utilizing advanced techniques such as formulation layers to balance computational efficiency and performance.

The proposed solutions have been evaluated using a system tool kit simulator for a downlink optical communication channel connecting a geostationary satellite to a ground station. The results demonstrate that the neural network-based channel estimator consistently outperforms state-of-the-art learning-based frameworks and achieves parity with minimum mean square error estimators. Similarly, the autoencoder framework surpasses benchmark methods and popular convolutional coding techniques under both perfect and imperfect channel state information conditions with various code rates. Together, these contributions represent a significant advancement in the design of low-complexity, high-performance communication systems for space optical communications.

Event Host: Abdelrahman ELFikky, Ph.D. Candidate, Electrical & Computer Engineering 

Advisor: Zouheir Rezki