Blind recognition algorithm of modulation mode and space-time block code via convolutional neural network

Concentrating on the joint recognition problem of modulation and space-time coding, algorithm for blind identification of modulation and space-time coding based upon fourth-order cyclic cumulant and feature parameter extraction in correlation matrix is proposed in this paper. The algorithm first uses the fourth-order cyclic cumulant to identify different modulation modes (4PSK, 8PSK, 16QAM, 32QAM, 64QAM). Different space-time block codes (Alamouti, OSTBC3, OSTBC4, NOSTBC2, and NOSTBC4) are identified by extracting feature parameters in the correlation matrix. The five recognized modulation modes and five space-time codes are sent to convolutional neural network, then 25 combination modes of modulation and space-time coding are recognized by using the Dropout layer, Gaussian noise layer, Flatten layer, dense layer, and fully connected layer successively by applying the Softmax activation function. The research results indicate that the emanated algorithm may effectively identify 5 modulation modes and 5 space-time codes, and can identify 25 combination modes of modulation and space-time coding. Under the condition that SNR is greater than or equal to 0 dB, the joint recognition rate of modulation and space-time coding can reach 100%.