Machine Learning Based Channel Estimation for 5G NR-V2V Communications: Sparse Bayesian Learning and Gaussian Progress Regression

In intelligent transportation systems (ITS), 5G NR-V2V communication is adopted to expand the connectivity and coverage of ITS applications, where the data needs to be processed directly in the on-board units (OBUs) deployed in the vehicle to meet the ultra-reliable and low latency communication (URLLC). However, the rapid variation of channel brings great challenges to channel estimation of V2V communication. In recent years, the channel estimation based on deep learning can learn the nonlinear characteristics of the channel, while requires a large amount of training data, which cannot guarantee the low latency data processing. Addressing this challenge, machine learning algorithm including sparse Bayesian learning (SBL) and Gaussian process regression (GPR), are proposed to accurately estimate V2V channel, so as to ensure the low complexity of real-time computation and greatly improve the reliability of system transmission. Our algorithm consists of two stages: 1) For pilot symbols, the channel estimation is converted into sparse basis coefficients recovery problem by basis expansion model (BEM), and then a sparse prior-based SBL (SP-SBL) algorithm is proposed to solve the problem. 2) The estimated pilot is used as the training set of GPR, and the Bessel function reflecting channel correlation is designed as the kernel function to accurately track the channel. Finally, the simulation results show that the proposed algorithm is superior to the deep learning based channel estimation algorithm, which ensures the URLLC of NR-V2V data transmission.