A comparison of MLP and RBF neural networks architectures for electromagnetic field prediction in indoor environments

In this paper two different neural network architectures are investigated for enough accurate field strength prediction in the complex indoor environment. The investigation includes multilayer perceptron (MLP) and radial basis function (RBF) neural networks. It has been already shown for neural networks as powerful tool in RF propagation prediction. Standard empirical or deterministic field prediction methods are difficult applicable in the case of complex indoor environments, so the neural networks can be the reasonable choice. The neural network models are trained with measured values of the field strength at arbitrary points. The backpropagation training algorithm (Levenberg-Marquardt with Bayesian regularization) is compared with particle swarm optimization (PSO) algorithm used in neural network training. After careful tuning training algorithm parameters the results showed smaller RMS errors for the PSO training case compared with backpropagation algorithm. Also, the better results are abstained by the RBF network architecture.