System modelling using neural networks in the presence of noise

In this paper, we have designed a multi layer neural network (MLNN) to estimate an unknown model of a non-linear signal generator in both cases of noise-free and noisy environments. The signal-to-noise ratio (SNR) has taken different values and simulation program has been executed for each case of SNR. We have obtained almost the accurate models for the noise-free examples. And robustness of the neural network model against noise has been examined through different values of SNR. We have obtained nearly good models using considerably large noise power. The model parameters such as the model size and the learning rate of the learning algorithm have been minimized in each case. Superiority of neural network models have been demonstrated by comparing the model performance in each case with that of the linear finite-impulse-response filter (FIR) model for that case.