Estimation of Pilot-assisted OFDM Channel Based on Multi-Resolution Deep Neural Networks

To realize the reliable communication of Unmanned Aerial Vehicle (UAV) in high-speed mobile environment, this paper proposed a pilot assisted Orthogonal Frequency Division Multiplexing (OFDM) channel estimation method based on multi-resolution depth neural networks. The model adapted all the full connection layers into dense convolution layers to improve the computational efficiency and prediction performance of channel estimation. Simulation results show that compared with traditional channel estimation methods, the normalized mean square error (NMSE) of multi-resolution depth neural networks is reduced by 78.84%. In addition, the method reduces the symbol error rate (SER) and the bit error rate (BER) by about 66.93%.