Comprehensive evaluation of cumulant-based and Kolmogorov-Smirnov-based modulation classifiers

Kolmogorov-Smirnov (K-S) test is a simple and robust method to distinguish distribution, which makes the final decixsion by comparing the differences between the empirical cdf of the samples of interest and the cdf of a hypothesized distribution. K-S-based modulation classifiers can be applied to different channels to distinguish multiple candidate modulation formats. In this paper, we evaluate the classification probability of the K-S classifiers comprehensively in different scenarios. Extensive simulation results that compared with the cumulant-based methods show that the K-S classifiers can achieve better classification performance, especially when distinguishing higher order modulation formats.