Deep Learning based Nonlinear Equalization for DCO-OFDM Systems

In this paper, we propose a deep learning (DL) based nonlinear post equalization (NPE) scheme to deal with the severe channel impairments of direct current biased optical orthogonal frequency division multiplexing (DCO-OFDM). The deep neural network is employed in the NPE to learn the nonlinear channel characteristics of visible light communication (VLC), and then implicitly transform the distorted symbols into the desired bit stream directly. In addition, the traditional solutions and algorithms in terms of channel estimation and symbol de-mapping are also incorporated into the network architecture to accelerate the convergence. The learning and generalization ability of the proposed scheme are investigated by simulations. Results demonstrate that the proposed scheme can efficiently address the nonlinear impairments of VLC channel and then recover the original symbols with comparable performance, which demonstrates the prospect and validity of DL methodology in channel equalization.