Modulation Classification Based On Cumulants And Goodness Of Fit Test

Automatic signal modulation classification is an important link between signal detection and demodulation,playing an important role in both civil and military communications.Combining with domestic and foreign literature,the classification of inter-class signal modulation has achieved good results,but there are still a series of problems in the intra-class classification of high-order signals.Traditional methods for intra-class classification of high-order signals mainly use high-order cumulants,because high-order cumulants can effectively suppress the influence of Gaussian noise.However,the traditional high-order cumulants method has the following problems:1)The computational complexity of the algorithm is high.With the increase of modulation order,it is necessary to find higher order cumulative values to complete signal classification.2)The robustness of all kinds of channels is poor.3)More signal symbols are needed.Based on the above problems,,this pape'r proposes a signal modulation classification method based on high-order cumulants and goodness-of-fit test.Combining the advantages of the two methods,including the high-order cumulants can effectively suppress the influence of Gaussian noise and the goodness-of-fit test has low computational complexity as well as is robust to various channel impairments.In the algorithm,first,the high-order cumulants are used to classify the signals between classes,and then the goodness-of-fit test is used to classify the signals within classes.The goodness-of-fit test is based on Kolmogorov-Smirnov(K-S)test in statistics,using the mean square integral of the difference between the empirical cumulative distribution functions(ECDFs)from received signals and cumulative distribution functions(CDFs)of the signal under different candidate modulation formats instead of the minimum of the maximum difference between the ECDFs of the received signal and the CDFs of the signals in different candidate modulation formats as the output decision to avoid the influence of extreme values.Besides,In order to avoid the default,such as it is less sensitive at heads and tails of the distributions,more weights are given to heads and tails in new algorithm.Massive simulation results show that compared with the K-S classifiers and the traditional high-order cumulant-based classifiers,the new classifiers we proposed show better classification performance at different SNRs with less signal samples for M-QAM and M-PSK modulations in various channels,including AWGN channel,the flat-fading channel,and the channel with unknown phase as well as frequency offsets,and is also more robust to various channel impairments.