From neural to wavelet network

Wavelet transform by means of a neural network is considered as a multivariate function approximation where the neural network is structured in a multi-input multi-output form. By means of this, the hierarchical wavelet decomposition is shaped as a parallel decomposition. That is, the input to the network is a block of discrete data and the output is a block of the wavelet transform, all resolution levels being computed in parallel. This approach is especially of concern for time varying systems where FFT techniques are not applicable and systems where the time-frequency approach plays an important role; real time systems for instance.