Low Latency Radio Access in 3GPP Local Area Data Networks for V2X: Stochastic Optimization and Learning

The next generation vehicular applications substantially shifting the paradigm of human activity have been projected to empower intelligent transportation systems. Targeting at supporting vehicle-to-everything connections, conventional mobile network architectures mandatorily requiring data routing through the core network, however, induce unacceptable costs both in end-to-end latency and backhaul resource consumption. The technical merit of moving computation and storage resources along with mobile vehicles consequently renders the mobile edge computing (MEC) a promising remedy to relieve the burden at the core network. To practice MEC, 3GPP has launched the normative works of a new paradigm known as the local area data network (LADN). Through performing in-network cache to store popular information at LADNs, a vehicle locating within the service area of an LADN is able to access particular location-based wireless application and information. Avoiding data routing through the core network, LADNs, however, encounter two critical challenges in downlink radio access to induce additional latency issues: 1) resource starvation at fronthaul links and 2) discrimination of quality-of-service requirements of vehicles with distinct capabilities. To tackle these challenges, through formulating the Lyapunov function, a stochastic optimization maximizing the utilization of fronthaul resources while stabilizing the queue (and thus latency) of each vehicle is proposed to address the resource starvation. Subsequently, a reinforcement learning-based multiarmed bandit algorithm is further proposed to achieve optimum harmonization of feedback-based and feedbackless transmissions, so as to strikes the tradeoff among energy efficiency, latency, and reliability. The performance evaluation results full demonstrate the effectiveness of the proposed design, to serve urgent needs in the deployment of LADNs.