Regresar

Expanding the Coverage of Multihop V2V with DCNNs and Q-Learning

Abstract:

One of the most critical challenge in a vehicle-to-vehicle (V2V) scenario is the transmission safety messages (BSMs) e.g., geographical location, braking information, speed, the status of the turn signal, and direction of travel. The protocol adopted to transmit BSMs in V2V is refered as Dedicated Short-Range Communications (DSRC). The limited communication range of DSRC have shown that is necessary to employ a multi-hop communication strategy to reach as many target vehicles as possible. In this paper, we overcome the coverage limitation of multi-hop connectivity in V2V networks and propose a methodology consisting of two machine learning (ML) tasks. First, two deep convolutional neural networks (DCNN) are created and tuned to segment terrestrial imagery into different environments. The multi-environments are anticipated to have different propagation models. The second part uses a Q-learning algorithm to find the optimal multi-hop path with the lowest propagation loss, based on the results of the environment segmentation. The optimal multi-hop link is simulated and compared with a direct link transmission, showing that our proposal can extend the coverage of multi-hop wireless links by transmitting the BSMs via the optimum path.