Citation

Rahman, Tanzeel Ur and Li, Guijun and Khan, Adnan Daud and Ali, Farman and Iqbal, Saeed and Kamal, Shahid and Ouyang, Zhengbiao (2026) Hybrid Adversarial Contrastive Framework for Robust Optical Free-Space Communication Under Atmospheric Turbulence and Domain Shift. Engineered Science. ISSN 2576988X

Text es2290.pdf - Published Version Restricted to Repository staff only Download (1MB)

Abstract

Optical Free Space Communication (OFSC) is gaining prominence in consumer electronics, enabling high-speed home networks, portable satellite links, and disaster recovery systems. However, atmospheric turbulence and domain shifts degrade signal quality, increase bit error rates (BER), and compromise the accuracy of pseudo-labels in semi-supervised learning (SSL). To address these challenges, we propose a Hybrid Adversarial Contrastive Framework (HACF) tailored for OFSC in consumer settings. HACF integrates turbulence-aware pseudo-label refinement, utilizing ensemble uncertainty quantification and confidence filtering, as well as adversarial consistency enforcement and cross-domain feature alignment via Maximum Mean Discrepancy (MMD) and Cross-Domain Contrastive Loss (CDCL). The framework also employs uncertainty-driven SSL for label refinement. Experimental results show that HACF outperforms conventional and adversarial-only methods, achieving lower BER across varying turbulence levels, faster convergence, improved domain alignment, and suppression of unreliable pseudolabels. This work advances robust, adaptive OFSC solutions for low-label, high-noise consumer environments.

Item Type:	Article
Uncontrolled Keywords:	Optical free-space communication
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5101-6720 Telecommunication. Including telegraphy, telephone, radio, radar, television
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	04 Aug 2026 07:10
Last Modified:	04 Aug 2026 07:10
URII:	http://shdl.mmu.edu.my/id/eprint/16501

Downloads

Edit (login required)