JultikaUniversity of Oulu repository

Abstract

To enable autonomous driving in intelligent transportation systems, vehicular communication is one of the promising approaches to ensure safe, efficient, and comfortable travel. However, to this end, there is a huge amount of application data that needs to be exchanged and processed which makes satisfying the critical requirement in vehicular communication, i.e., low latency and ultra-reliability, challenging. In particular, the processing is executed at the vehicle user equipment (VUE) locally. To alleviate the VUE’s computation capability limitations, mobile edge computing (MEC), which pushes the computational and storage resources from the network core towards the edge, has been incorporated with vehicular communication recently. To ensure low latency and high reliability, jointly allocating resources for communication and computation is a challenging problem in highly dynamics and dense environments such as urban areas. Motivated by these critical issues, we aim to minimize the higher-order statistics of the end-to-end (E2E) delay while jointly allocating the communication and computation resources in a vehicular edge computing scenario. A novel risk-sensitive distributed learning algorithm is proposed with minimum knowledge and no information exchange among VUEs, where each VUE learns the best decision policy to achieve low latency and high reliability. Compared with the average-based approach, simulation results show that our proposed approach has the better network-wide standard deviation of E2E delay and comparable average E2E delay performance.