Deep Learning Based Predictive Power Allocation for V2X Communication

As an essential technology of the fifth generation communication (5G), Vehicle-to-Everything (V2X) has attracted wide attention lately. A well-designed power allocation scheme can decrease the interference among terminals to yield a significant performance gain for V2X communication. Accurate channel state information (CSI) feedback, which is crucial to power allocation, is hard to be tracked due to fast time-varying channel in V2X scenario. This paper investigates power allocation problem under delayed CSI feedback in V2X system. We focus on maximizing sum throughput of V2X system while meeting Quality of Service (QoS) constraint requirement of each link. First of all, a power allocation scheme utilizing projective constraint analysis (PCA) is proposed to guarantee reliability of V2X network and improve system throughput, namely PCA-PA. Subsequently, we develop a Deep Neural Network (DNN) based predictive power allocation algorithm, namely DNN-PPA. This algorithm aims to address delayed CSI feedback in V2X system utilizing the normalized data obtained from PCA-PA scheme. Furthermore, a V2X communication model compliant with the 3rd Generation Partnership Project (3GPP) standard is simulated to validate the performances of PCA-PA and DNN-PPA. Simulation results illustrate that PCA-PA has superior performances compared to existing power allocation approaches and the throughput performance of DNN-PPA is impressively close to optimal solution under delayed CSI feedback.