Fourth Moment-Assisted Deep Learning-Based Channel State Estimation for OFDM Systems

A deep learning-based channel state information (CSI) estimation method for orthogonal frequency division multiplexing systems with phase shift keying or 4-point quadrature amplitude modulation is proposed. The method uses all received values including both known pilots and unknown data symbols. Received values are raised to the fourth power, obtained 2D data are smoothed by a convolutional neural network, and then fourth power root is taken with subsequent resolving of phase ambiguity. Simulation results showed that this method outperforms baseline methods that use only pilot symbols for CSI estimation.