GAN-Based Fractional Doppler Channel Estimation for OTFS Modulation

Orthogonal Time Frequency Space (OTFS) modulation is a novel technique that enables efficient and robust communication in time-varying wireless channels, demonstrating unique properties such as resistance to inter-symbol interference and effective exploitation of multipath diversity. This paper proposes a fractional Doppler channel estimation scheme based on Generative Adversarial Networks (GAN) for OTFS modulation in high-mobility air-ground communication scenarios. Specifically, we improve the structure of the Conditional GAN (cGAN) generator by using the initial channel response matrix obtained through the least squares (LS) algorithm as the initial input, while the received signal serves as the conditional input. This structure compensates for missing channel features during training and enhances model accuracy. Simulation results demonstrate that the proposed scheme outperforms conventional LMMSE and OMP algorithms in channel estimation performance.