Predictive Beamforming for OTFS-Enabled URLLC in High-Mobility Vehicular Networks

Abstract

Ultra reliable low latency communication (URLLC) in vehicular networks is pivotal for meeting the stringent requirements of transportation safety. However, achieving it is very challenging due to the high mobility of vehicles and the complex propagation environment. Orthogonal time frequency space (OTFS) modulation addresses these issues by modulating symbols into the delay-Doppler (DD) domain, which can leverage full time-frequency diversity by spreading each DD domain symbol across the entire time-frequency plane. In this paper, we propose a novel OTFS-enabled ultra reliable low latency vehicular network architecture for downlink transmission. To achieve low latency, we adopt frequency division duplex (FDD) mode to transmit data frames as soon as they arrive to minimize scheduling delays. Furthermore, to enhance the received signal strength at the receiver, beamforming is applied at the transmitter. Due to the channel state information (CSI) feedback delay in FDD systems, we design a deep learning algorithm, the DD-domain Convolutional Transformer (DDCT), for predictive beamforming based on historical DD-domain CSI. In DDCT, a convolutional neural network extracts spatial features from the DD domain, and a transformer captures their temporal correlations. Extensive simulation results demonstrate the effectiveness of the proposed vehicular network architecture and the superiority of the deep learning algorithm for predictive beamforming.