With the spectra of digital modulated signal and its non-linear transform being analyzed

two new modulation feature parameters are proposed and descriptions of another two features are modified.Then a novel identification scheme is obtained

capable of blind identification of many commonly used digital modulations in AWGN

using a neural-network-ensemble-classifier.The proposed spectrum-shape parameter is anti-noise

while the new spectrum-line one is robust against modulation parameters

and the performance of neural network ensemble is much better than that of a single network classifier.Simulation results show that the recognition rate can reach 94% when

SNR

is above 5dB.