Low Complexity Feature-Based Modulation Classifier and Its Non-Asymptotic Analysis

In this paper, we propose a reduced-complexity modulation classifier using multi-cycle features extracted from the Spectral Correlation Function (SCF) in order to distinguish among QAM, BPSK, MSK and AM modulation schemes. We analytically derive SCF statistics of the noise and signal features used for classification for finite number of samples, and use Chebyshev inequality to upper bound the minimum number of spectral averages required to attain a predetermined correct classification probability. Both theoretical and simulation results show that the proposed classifier requires on the order of 50 spectral averages to achieve a correct classification probability of 0.9 at SNR = 5 dB. The algorithm and corresponding analysis presented in this paper can be extended to classify other modulation schemes.