16QAM-OFDM VLC System Based on Frequency Domain Precompensation and DNN Post-Equalization

Visible light communication (VLC) technology utilizes LEDs as light sources, offering several benefits, such as the use of license-free spectrum, green, and safe. However, conventional LED devices face challenges due to their restricted bandwidth and nonlinearity under different current levels. These issues lead to signal distortion and intersymbol interference (ISI), which are particularly problematic in high-order modulation schemes. In this work, we propose a low-complexity scheme combining S₂₁ precompensation and shallow hidden layer deep neural networks (DNNs)-based post-equalization in a VLC system. At the transmitter, S₂₁ precompensation is used to suppress the influence of high-frequency power fading. The post-equalization at the receiver side uses the least square (LS)-DNN cascaded equalization algorithm, where LS performs linear equalization and DNN performs nonlinear equalization. Since the S₂₁ precompensation and LS algorithm have completed partial equalization, it can provide a good initial state for DNN training, thus speeding up the convergence of DNN. To better capture the dependency between symbols, we preprocess the dislocation filling before the data is sent to the input layer of DNN. In the experiment, a VLC system with the 3-dB bandwidth of 9 MHz is used. With S₂₁ precompensation, its 3-dB bandwidth is expanded to about 43 MHz. The two hidden layer DNN is used to carry nonlinear equalization. If an 80-Mb/s 16QAM OFDM signal is transmitted in a 3-m VLC system, the scheme proposed can expand the available peak-to-peak voltage V_pp range by 30% compared with precompensation LS algorithm and achieve transmission over 9 m under 7% hard-decision forward error correction (HD-FEC) BER threshold of 3.8e-3. The comparison with LSTM-DNN-based scheme shows that the proposed scheme can improve the performance of VLC system in terms of BER and EVM, while the training complexity is 2-order of magnitude lower.