neural network for function approximation

Hello all, I'been working with EMG signals from four different muscles. Each signal is a vector of 60K data. I used wavelet packet transform to extract the features from these signals using a moving window, I got 31 features x 300 windows.

What can I do to select the best features? Once I have selected the features, how can I manage to create a network to fit a function, in this case the target is a signal also recorded during the data acquisition and is a 60k vector.

Thank you.