On the error in reconstructing a non-bandlimited function by means of the bandpass sampling theorem