Independent Component Analysis for Radio Network Prediction Enhancement

We propose a method to enhance the quality and precision of prediction results using measurements in the context of radio network modelling. The proposed method involves the use of an Independent Component Analysis (ICA) block and a MultiLayer Perceptron (MLP) Artificial Neural Network (ANN). The role of the ICA block is to make the variables at the input of the ANN statistically independent so that it can perform its learning and prediction on individual one-dimensional distributions. The application of the proposed method to a third generation cellular radio network prediction tool has shown that without ICA, ANN training has a poor performance. We have also shown that, in the proposed scheme, ICA performs better than Principle Component Analysis (PCA). This enhancement method can advantageously be used to calibrate prediction results according to measurements.