GaussMask-DSSS: Enhancing Covert Spread Spectrum Communication With Gaussian Cloaking and Deep Learning-Aided Synchronization

Achieving secure communication with a low probability of detection (covertness) is critical yet challenging, particularly when employing practical digital modulations that can compromise the statistical indistinguishability assumed in theoretical models. This paper introduces a novel end-to-end framework leveraging digitally modulated covert signal modeling, obfuscation, and deep learning to attain simultaneous covertness and reliability. Firstly, we propose a novel approach to covert performance evaluation for modulated covert signals against detection. To address the deteriorated covertness considering modulation schemes, we further propose generating Gaussianized camouflage signals via a multi-stage transmitter pipeline encompassing spreading, jitter, filtering, and non-linear transformations, designed to mimic noise statistics effectively. At the receiver, a specialized deep learning architecture, CovertSyncNet, performs robust joint dynamic synchronization and symbol recovery. This receiver incorporates dedicated components to precisely estimate time-varying chip offsets and invert the complex, nonlinear distortions inherent in the camouflaged signal, enabling accurate demodulation. Extensive simulations rigorously validate our approach, demonstrating that high reliability is maintained despite the heavy camouflage. Concurrently, enhanced covertness is confirmed through metrics indicating low statistical distinguishability from Gaussian noise. This work highlights the significant potential of deep learning to bridge the gap between theory and practice, realizing communication systems that are simultaneously reliable, secure, and highly covert, even under realistic operational conditions.