Abstract
Vehicle-to-Vehicle (V2V) communications play a vital role in intelligent transportation. Especially in the 6G environments, the accuracy and efficiency of channel estimation techniques for V2V communication are crucial for realizing reliable autonomous driving and traffic systems. Although the convolutional neural network (CNN) has exhibited notable effectiveness in channel estimation for wireless communication systems, there are still severe open challenges in achieving desirable performance and computation complexity. To fill the gap, a novel deep learning-based channel estimation network (CEN) for multi-scene V2V channel estimation is proposed in this paper. Firstly, a novel bidirectional long short-term memory (Bi-LSTM) framework is introduced for V2V channel estimation. Then, the fully connected neural network (FCNN) network is used for the output dimensionality reduction. Finally, the temporal averaging (TA) processing is designed for eliminating the noise. Simulation results show that the proposed channel estimation scheme is superior to traditional channel estimation algorithms with desirable performance and lower computational load in urban environments.