PROJECT TITLE :

Deep Learning on Multimodal Sensor Data at the Wireless Edge for Vehicular Network

ABSTRACT:

Because an exhaustive search among all candidate beam pairs cannot be assuredly completed within short contact times, beam selection for millimeter-wave links in a vehicular scenario is a challenging problem. Our innovative expediting beam selection makes use of multimodal data obtained from sensors such as LiDAR, camera images, and GPS to solve this problem. Along with a study on the associated tradeoffs, we present individual modality and distributed fusion-based deep learning (F-DL) architectures that are capable of executing locally as well as at a mobile edge computing center (MEC). We also formulate and solve an optimization problem for determining the output dimensions of the aforementioned F-DL architectures. This problem takes into account the latency overheads associated with practical beam-searching, MEC processing, and sensor-to-MEC data delivery. The results of extensive evaluations performed on publicly available synthetic and home-grown real-world datasets reveal a 95% and 96% improvement in beam selection speed over traditional RF-only beam sweeping, respectively. These results were obtained by conducting the evaluations on publicly available datasets. F-DL also outperforms the techniques that are considered to be state-of-the-art by 20-22% when it comes to predicting the top-10 best beam pairs.