Out-of-Distribution Detection for Multiple Signal Sources in Connected Vehicles

Abstract

Vehicular communications, essential for intelligent transport systems, utilize multi-carrier, single-carrier, and radar signals to share frequency spectrum, hardware, and signal processing resources, thereby ensuring robust performance in dynamic environments. Detecting out-of-distribution (OOD) signals that deviate from expected patterns, poses significant challenges to system reliability and safety. This paper proposes a deep learning (DL)-based method for OOD signal detection. We first introduce a sequence-input-based autoencoder that processes received signals' in-phase and quadrature components. By applying the Mahalanobis distance in the autoencoder's latent space, we obtain an modified loss. Subsequently, using Youden's J statistic, we determine an optimal threshold, enhancing the detection accuracy for OOD. Experimental results demonstrate that our method outperforms conventional DL models in detecting OOD signals, offering a trade-off of increased computational complexity for enhanced detection accuracy.