Deep learning for joint adaptations of transmission rate and payload length in vehicular networks

Abstract

Recently, vehicular networks have emerged to facilitate intelligent transportation systems (ITS). They enable vehicles to communicate with each other in order to provide various services such as traffic safety, autonomous driving, and entertainments. The vehicle-to-vehicle (V2V) communication channel is doubly selective, where the channel changes within the transmission bandwidth and the frame duration. This necessitates robust algorithms to provide reliable V2V communications. In this paper, we propose a scheme that provides joint adaptive modulation, coding and payload length selection (AMCPLS) for V2V communications. Our AMCPLS scheme selects both the modulation and coding scheme (MCS) and the payload length of transmission frames for V2V communication links, according to the V2V channel condition. Our aim is to achieve both reliability and spectrum efficiency. Our proposed AMCPLS scheme improves the V2V effective throughput performance while satisfying a predefined frame error rate (FER). Furthermore, we present a deep learning approach that exploits deep convolutional neural networks (DCNN) for implementing the proposed AMCPLS. Simulation results reveal that the proposed DCNN-based AMCPLS approach outperforms other competing machine learning algorithms such as k-nearest neighbors (k-NN) and support vector machines (SVM) in terms of FER, effective throughput, and prediction time.