Automatic Space-Time Block Code Recognition Using Convolutional Neural Network With Multi-Delay Features Fusion

Automatic signal recognition (ASR) is becoming increasingly important in spectrum identification and cognitive radio, but most existing space-time block code (STBC) recognition algorithms are traditional ones and do not account for the complementarities between different features. To overcome these deficiencies, a multi-delay features fusion scheme for ASR of STBC using a convolutional neural network (CNN) is proposed in this study. The proposed scheme tries to fuse different time-delay features of received STBC signals to obtain more discriminating features. The dilated convolution of different dilation rates is applied to realize automated multi-delay feature extraction. Then, two fusion methods, i.e., max-correlation (MC) fusion and multi-delay average (MDA) fusion, are proposed to combine the features of different time delays, and a residual block is applied to achieve better representation of signals. Finally, the simulation results demonstrate the superior performance of the proposed method. It is notable that the recognition accuracy can reach 97.4% with a signal-to-noise ratio (SNR) of -5 dB. In addition, the proposed multi-delay features fusion CNN (MDFCNN) scheme does not need a priori information, i.e., modulation type, channel coefficients, and noise power, which is well suited to non-collaborative communication.