Stacked autoencoder, problems

I'm using a stacked autoencoder with fine tuning to perform function approximation. Feature extracted from the first autoencoder are ok, then I use it as input of a 2nd autoenc. What happen is that the feat of the 2nd autoenc. are always identical. Indipendently on the size of thelayer and on the data sets. Why? Also, the second autoencoder is not able to reconstruct the input. Here is my code:

nfeat1=8

autoenc1 = trainAutoencoder(pattern',nfeat1,'MaxEpochs',500,'EncoderTransferFunction','logsig','DecoderTransferFunction','purelin');

feat1 = encode(autoenc1,pattern');

nfeat2=3

autoenc2 = trainAutoencoder(feat1,nfeat2,'MaxEpochs',500,'EncoderTransferFunction','logsig','DecoderTransferFunction','purelin');

feat2 = encode(autoenc2,feat1);