Correct modeling of light emitting diode (LED) nonlinearity is vital for evaluating the performance of advanced modulation formats and nonlinear distortion mitigation methods in visible light communication systems. In this paper, we review the existing simplified nonlinear LED models: static polynomial, Wiener and Hammerstein and provide strong experimental evidence against their even coarse correctness. Instead, we propose an evenly uncomplicated model of LED, which is based on rigorous factorization of the Volterra 2nd order kernel stemming from the recombination rate equations of the LED. We further experimentally validate this model in four different LEDs by: (i) finding that linear transmitter pre-emphasis, against the results of the communications theory in a linear, bandlimited channel, can at very high input powers improve the transmission perfromance compared to linear equalization at the receiver, which is explained in our model by a reduction of the nonlinear distortion by the pre-emphasis; (ii) by comparing the measured 2nd order quadratic kernels of LEDs in regular form and with the constraints of our model. Finally, we propose a two-parameter calibration method of our model, which requires measurement of the LED bandwidth and 2nd order harmonic distortion at a single frequency. Excellent convergence of the signal transmitted in the calibrated model to the rate equation model and to the measurement is observed.
