Predistortion of OFDM signals for VLC systems using phosphor-converted LEDs

In Visible Light Communications (VLC) systems, the Light-Emitting Diode (LED) is the dominant source of non-linearity and memory effects, which are originated on phenomena that take place in both the electrical and optical domains. The impact that these LED non-idealities have on the received data symbols becomes even more notable with OFDM waveforms due to their high Peak-to-Average power Ratio (PAPR) of these signals. One simple way to address this problem consists in selecting a suitable Input Back-Off (IBO) value, forcing the LED to work in its linear region. However, such an approach limits the VLC system coverage, as it reduces the dynamic range of the OFDM signal that modulates the intensity of the optical wireless link. To provide a balance between these two conflicting goals, the use of digital predistortion can be considered instead, in order to compensate nonlinear distortion and memory effects that are added in the VLC transmitter. For this purpose, this paper studies the sources of nonlinearity and memory in phosphor-converted (PC)-LEDs in both electrical and optical domains. After that, different approaches are presented to model these effects in the PC-LED, namely the Wiener-Hammerstein, memory polynomial, and Convolutional Neural Network (CNN) models. Finally, the performance of each of these approaches for digital predistortion are experimentally evaluated in a software-defined VLC demonstrator, observing a notable improvement on the Error Vector Magnitude (EVM) when compared to the case in which no compensation is performed in transmission.