Dynamic geo-based resource selection in LTE-V2V communications using vehicle trajectory prediction

Vehicle to Vehicle (V2V) communication has recently been considered in 4G and 5G cellular networks. One of the challenging issues in cellular V2V is allocating radio resources to the vehicles. Although previous work have addressed this issue, the fast varying nature of vehicular traffic and its regularities implies that the mobility of the vehicles should be more attended. To this goal, we propose an autonomous geo-based resource selection algorithm that uses deep learning to predict vehicle locations in the future and alleviate the computation and signalling overhead of the cellular infrastructure in contrast to previous geo-based resource allocation algorithms. We utilize the current and the future of vehicle densities in a formulated matching problem to find the optimum assignment of sub-resource pools to geographic areas. Simulation results of a highway with diverse density scenarios and different number of available resources show that the proposed method guarantees a considerable reduction in computation and signalling overhead while in low awareness ranges, it provides up to 10% improvement in Packet Reception Ratio (PRR) and the error rate of vehicles compared to the previous Dynamic Geo-based Resource Selection Algorithm (DGRSA). The proposed method also provides up to 15% improvement in PRR and error rate compared to the modified DGRSA, which we have changed to run with an overhead equal to the overhead of our proposed method. Furthermore, our results demonstrate up to 67% and 76% improvement in blocking rate compared to DGRSA and modified DGRSA, respectively.