LED color prediction using a boosting neural network model for a visual-MIMO system

Abstract Color decision of Light-emitting diode (LED) by smartphone cameras is a challenging area in visual- multiple-input multiple-output (MIMO) systems. In this study, we use a generalized color modulation (GCM) technique for a visual-MIMO system. We propose a boosting neural network (BNN) model that can predict LED color from an LED image. To develop this learning model, we use LED image pixels as input features by resizing all LED images to 10 × 10 pixels through bicubic anti-aliasing interpolation. The model is trained in three stages: (1) select the coefficient of the activation function, (2) train each feature to build weak learners, and (3) train the weak learners to predict LED color. Then, we make a symbol decision by measuring the minimum Euclidean distance between the predicted color of the received symbol and transmitted symbol colors. We evaluate our prediction by measuring the root-mean-square error (RMSE) of our test dataset at different environmental light intensities. We also measure the average closeness accuracy and symbol error rate (SER) performance of the proposed method with respect to transmission distances and different sizes of constellation diagrams. Finally, we compare the performance of our proposed BNN model with that of a multiple-linear-regression method.