Abstract: Embodiments of the present disclosure relate to machine learning for channel estimation against diverse Doppler effects. In this solution, the first device receives user data in a slot from a second device via a channel between the first device and the second device, the user data includes a reference symbol. Then, the first device performs, with a NN, a first channel estimation for the channel based on information on velocity of a second device and a result of a second channel estimation for the channel. The second channel estimation is performed based on the received reference symbol. After that, the first device transmits, to the second device, an indication of a result of the first channel estimation. With this solution, channel estimation is able to be performed in diverse scenarios (e.g., diverse Doppler scenarios, high-velocity scenarios and the like) with low overheads in reference signals (e.g., 1 DMRS), thereby reducing network consumption.