Waveform-Defined Security Enhancement via Signal Generation optimization
Tongyang Xu · 2020
Traditional defence strategies of physical layer security (PLS) are highly dependent on channel environments. This work investigates a waveform-defined security (WDS) framework, which can fundamentally prevent signal interception. In the traditional WDS framework, by intentionally tuning waveform parameters to weaken feature diversity and enhance feature similarity, eavesdroppers cannot correctly identify feature-similarity dominant signals using deep learning (DL) classifiers. The imperfect signal classification would result in subsequent detection errors. This work aims to optimize the framework by further complicating signal classification using a newly proposed signal generation architecture. Results show that the new signal generator can cut distinguishable signal features. In this case, classification accuracy at eavesdroppers is reduced by up to 53% leading to an enhanced WDS framework. Meanwhile, legitimate users maintain performance reliability regardless of signal generation architectures.