A DNN-Based Channel Estimation Method for Spectral Efficient Frequency Division Multiplexing Systems

A channel estimation method based on deep neural network (DNN) for spectral efficient frequency division multiplexing (SEFDM) systems is proposed. The method employs uniform spaced orthogonal pilot symbols to achieve the channel estimation. To be specific, the received pilot signals are used as the input of the four-layer DNN in order to extract the channel features. Simulation results show that the proposed scheme can yield a smaller mean square error (MSE) and, in turn, perform better demodulation in comparison with the conventional least square (LS) method. In particular, the DNN-based method is more robust to the number of pilots, which indicates its superiority.