Data-Aided Learning for Multi-Antenna OFDM Systems with Superimposed Pilot in Time-Varying Channel

Deep learning (DL) is now widely applied in the physical layer, with particular attention to end-to-end learning-based communication transceivers. In this paper, we focus on superimposed pilots and propose a novel data-aided end-to-end learning scheme, which is applied to orthogonal frequency division multiplexing (OFDM) systems in time and frequency selective fading environment. Specifically, we conduct multidimensional joint training including time-frequency, power, and geometric shaping of constellation. And we perform the end-to-end training on the transmitter and the proposed data-aided receiver, achieving more reliable data transmission in multi-antenna mobile scenarios.