Performance Evaluation of Various Training Algorithms for ANN Equalization in Visible Light Communications with an Organic LED

Abstract

This paper evaluates the effect of training algorithms in an artificial neural network (ANN) equalizer for a feedforward multi-layer perceptron configuration in visible light communication systems using a low bandwidth organic light source. We test the scaled conjugate-gradient, conjugate-gradient backpropagation and Levenberg-Marquardt back propagation (LM) algorithms with 5, 10, 20, 30, and 40 neurons. We show that, LM offers superior bit error rate performance in comparison to other training algorithms based on the mean square error. The training methods can be selected based on the trade-off between complexity and performance.