Pilot-Aided Deep Learning Based Phase Estimation for OFDM Systems with Wiener Phase Noise

Orthogonal frequency-division multiplexing (OFDM) and high-order modulations have a wide range of applications in coherent optical communications. However, the existence of phase noise will greatly affect the system performance. To overcome this issue, this paper proposes a pilot-aided deep learning (PADL)-based phase estimation scheme, since deep learning has been a hot trend to be applied on digital signal processing in communications. To be specific, preliminary phase noise is first estimated by the pilot-aided (PA) method, and then the estimates are fed into the neural network to train for more accurate estimation. Simulation results show that the mean square error performance of the proposed method is much better than the conventional PA method. Especially for high-order modulations (M>64), the bit error rate of the PADL-based receiver is smaller than that with even double pilots used for phase estimation, which verifies a stronger robustness in our design using limited spectrum resources.