The performance of modulation classification in single radio is sensitive to signal-to-noise ratio(SNR).A hybrid maximum likelihood(ML) modulation classification algorithm using multiple radios was proposed.All received baseband signals from different radios working on synchronous mode were fused at a fusion center to make the global classification decision by using the hybrid ML.Due to the spatial diversity,the performance of modulation classification in low SNR regimes was improved.The joint likelihood function contained multiple dimensional unknown parameters.In order to alleviate the computational complexity associated with the ML estimates of the unknown parameters,the expectation-maximization(EM) algorithm was adopted, in which the constellation symbols were represented unobserved data.The proposed algorithm completed the classification of BPSK,QPSK,8PSK and 16QAM,as well as the unknown parameters estimation. Compared to the algorithm based on moments, the unknown parameters estimate based on EM estimation provided superior performance in precision.The simulation results showed that when the number of radios is four and SNR of signal is more than -2 dB, the average probability of correct classification is more than 95%.
