Real-Time 3-D Localization of Jamming Drones Using Covariance Matrix-Based Vision Transformer

초록

In this paper, we propose a method for 3D localization of jamming drones based on a Covariance Matrix-Based Vision Transformer (CM-ViT), enabling real-time estimation. Since the emitted waveform of jamming signals is unknown, covariance matrices obtained from multiple array antenna systems are used as input rather than using raw IQ data. The proposed network estimates the 3D signal source coordinates (x, y, z), accounting for the arbitrary spatial locations of jamming drones. By utilizing the attention mechanism of the Vision Transformer (ViT), the proposed approach more effectively exploits global correlation patterns in covariance matrices than conventional deep learning methods, leading to improved localization performance. Furthermore, to alleviate excessive training dataset requirements, the received signals at the array antennas in multipath environments are modeled using a Rician fading channel. To evaluate the performance of CM-ViT, it is compared with MUSIC, Capon, and a convolutional neural network (CNN)-based method in terms of average localization error and runtime across varying signal-to-noise ratio (SNR) levels. The localization error is reduced by approximately 46% compared to beamforming-based approaches and by 19% compared to the CNN-based method, while CM-ViT achieves a reduced estimation time of 68 & micro;s, enabling real-time estimation. These results demonstrate that the proposed method is suitable for real-time localization of high-mobility airborne targets and achieves higher localization accuracy.

키워드