Using Complex Neural Networks on Noise Cancelling

Abstract

We present a method to reduce noise on signals applying complex auto-associative neural networks. Experimental results using the sine, triangular and sawtooth signals are performed to validate our results. This method is based on the use of complex neural networks learning and is capable of eliminating or reducing noise on learned signals. First, a training set consisting of signal values without noise at different phases is used for learning. Second, The signal is processed first using the fast Fourier transform (FFT) to obtain the frequency components. Third, three different complex neural networks are trained using thetransformed clean signal until a minimum error criterion is satisfied. Finally, noisy transformed signals are applied to the complex neural network for simulation purposes. The neural networks' output signal is inverse transformed and compared with the clean signal for each case to verify the difference between them.