Leveraging Data-Driven Deep Neural Network for IEEE 802.11bd Receiver Design

IEEE 802.11bd is especially designed for the next generation vehicle-to-everything (NGV) communication. However, due to vehicle’s high mobility, the performance of IEEE 802.11bd receiver can be severely deteriorated by the fast channel fading, multi-path, carrier frequency offset (CFO), and symbol timing offset (STO) impairments, and these impose great challenges to the conventional receiver design. Moreover, since the conventional receiver is built on a pre-assumed model of the signal impairments, this limits its performance improvement. To overcome this, in this paper, we explore the special frame structure (subcarrier number, pilot, and midamble) of the IEEE 802.11bd as well as the strong non-linear fitting capability of the deep neural network (DNN), and design a data-driven DNN assisted IEEE 802.11bd receiver. We optimize our symbol recovery DNN structure to provide strong symbol recovery performance given the fast channel fading, multi-path, CFO, and STO impairments. To further improve the generalization performance of our receiver, we propose a training data variety augmentation method and optimize the training data and reference data structure in DNN training and reference stages. We verify the performance of our proposed receiver in rural line-of-sight (LOS) channel, highway LOS channel, and urban LOS channel, and simulation results show that our proposed receiver can significantly reduce the symbol error rate compared with the conventional receiver, and can preserve a competitive generalization performance.