Deep Learning based Channel Prediction for OFDM Systems under Double-Selective Fading Channels

With the development of wireless communication and internet of vehicles (IoV), a growing number of wireless high-speed scenarios have emerged. High mobility will introduce large Doppler shift to the channel, resulting in fast time-selectivity, and multi-path transmission will lead to frequency-selectivity. In such a double-selective fading channel, in order to accurately recover the transmitted symbols, lots of pilot symbols are required for channel estimation, resulting in bandwidth wastage. In this paper, we design a novel deep learning (DL) based channel prediction network that combines the benefits of fully-connected deep neural network (FC-DNN), convolutional neural network (CNN) and long short-term memory (LSTM) to reduce the demand of pilot symbols in orthogonal frequency-division multiplexing (OFDM) systems. In particular, the three networks are deployed to perform noise reduction, interpolation and prediction, respectively. In addition, we propose a data aided decision feedback scheme in prediction to guarantee the prediction performance. Simulation results demonstrate that the proposed prediction network can achieve better performance than existing methods.