Enhanced Communications Security via End-to-End Deep Adversarial Learning-driven Encoding

Deep Learning (DL) is a promising solution to address spectrum management, modulation recognition, and system design challenges facing modern communication systems. For system design, DL can provide an optimized end-to-end communications system that is reliable while preserving confidentiality in the presence of an adversary. This work presents a software-defined radio-implemented DL-driven encoding process trained and tested over the air in an office environment that ensures confidentiality between authorized transceivers by preventing an adversary from reconstructing passively intercepted messages. The results show that the DL-driven encoding process outperforms Commercial-Off-The-Shelf (COTS) radios because it learns a non-standard communications scheme informed by the current wireless channel conditions.