Recent Advances in Resource Allocation via Sparsity-Aware Orthogonal Initialization of Deep Neural Networks in Free Space Optical Communications: A Systematic Review

Abstract

Free Space Optical (FSO) communication has emerged as a promising technology for high-speed wireless communication due to its large bandwidth, license-free spectrum, and immunity to electromagnetic interference. However, FSO systems are highly susceptible to atmospheric turbulence, misalignment, and signal fading, which significantly degrade communication reliability and performance. Recent advancements in Artificial Intelligence, particularly deep learning techniques, have introduced innovative solutions for optimizing resource allocation and improving system robustness. This study presents a systematic review of resource allocation strategies in FSO communication systems, emphasizing sparsity-aware orthogonal initialization in deep neural networks. Deep learning models, including convolutional neural networks and graph neural networks, have been successfully applied to mitigate turbulence effects and optimize power allocation and channel estimation. For instance, model-free deep learning approaches enable efficient resource allocation without requiring explicit system models, thereby enhancing adaptability and performance in dynamic environments. Additionally, convolutional neural networks have demonstrated significant improvements in reducing bit error rates under atmospheric disturbances. Furthermore, sparsity-aware and orthogonal initialization techniques enhance training stability, convergence speed, and generalization capability of deep neural networks, making them suitable for complex FSO scenarios. This review highlights recent advancements, identifies key challenges such as computational complexity and environmental variability, and discusses future research directions for developing efficient, intelligent, and adaptive FSO communication systems.