Neural network based Interference cancellers occupy a prominent place in the class of multi user detectors for CDMA.The present work focuses on the effect of pre-processing of the data fed to the neural networks in those receivers. The effect on bit error rate due to pre-processing is presented, with respect to a variation of signal to noise ratios and the number of parallel interference canceller stages. The work considered, R a y l e i g h multipath fading channel as the medium and DPSK as baseband modulation.