Time-Delay Twin Support Vector Regression- Based Adaptive Predistorter for LED Nonlinearity in Visible Light Communications

The bit error rate (BER) of visible light communications (VLC) based on emitting diodes (LEDs) has been limited due to the nonlinearity LED. Artificial neural networks (ANNs) predistorters were applied mitigate impacts LED nonlinearity. However, none them can efficiently approach performance linear VLC systems overfitting issue. In this paper, we for first time propose a predistorter with adaptive indirect learning architecture amplitude time-delay twin support vector regression (ATD-TSVR) in orthogonal frequency division multiplexing(OFDM)-based systems. experimental results demonstrate signal distortion caused by LEDs as major nonlinear source. Moreover, simulation show superior BER, (AM/AM) curves, Power Spectral Density(PSD) and constellation plots an 80Mbit/s OFDM system. Meanwhile, compared traditional SVR approach, CPU training ATD-TSVR be reduced more than four times. pre-distortion method herein is generally applicable broadband also proves application prospect effectiveness TSVR