DNN and LS Based Channel Estimation in OTFS System.

Orthogonal time frequency space (OTFS) system transmits information in the delay-Doppler domain, and it adapts well to high-speed mobile environments and can realize channel estimation with less overhead. Channel estimation in the OTFS system is necessary for subsequent channel equalization and information recovery process, aiming to eliminate the effects, such as inter-symbol interference (ISI) and noise influence. The paper proposes an effective channel estimation for the OTFS system using deep neural network (DNN) and the least square (LS) algorithm, considering the accurate and low latency channel estimation requirement in the Internet of Vehicles, where LS algorithm is first used to perform a rough time-frequency domain channel estimation, and then the DNN model is taken to optimize the rough channel estimation. Simulation results show that the proposed method achieves more accurate channel estimation performance with low complexity and better flexibility, compared with the typical LS and linear minimum mean square error (LMMSE) scheme.