Dilated convolutional neural network-based signal detection for optical OFDM-IM

Orthogonal frequency division multiplexing with index modulation (OFDM-IM) is an advanced signal transmission technology, offering both spectral and energy efficiency advantages. The most commonly employed deep learning (DL)-based signal detection technologies for the OFDM-IM system, which rely on fewer index combinations or low-order modulation schemes, inherently limit the capability of visible light communication (VLC) systems to achieve high data rate and flexible allocation of active subcarriers. In this paper, to address the performance degradation of conventional DL-based signal detection technologies under more index combinations or higher-order modulation schemes in VLC systems, we propose a dilated convolutional neural network for optical OFDM-IM (DCNN-OIM) system. Specifically, the proposed DCNN-OIM detector integrates a dilated convolution method into the convolutional neural network, which can effectively improve the signal detection accuracy without introducing additional computational complexity under more index combinations or higher-order modulation schemes. The simulation results demonstrate that, in comparison with the related signal detection technologies, the DCNN-OIM detector can carry out lower bit error ratio (BER) and computational overhead.