Blind identification of space–time block codes based on deep learning

Deep Learning (DL) has important applications to both commercial and military communications, such as software-defined radio, cognitive radio and spectrum surveillance. While DL has been intensively studied for modulation recognition, there are very few investigations for blind identification of Space-Time Block Codes (STBCs). This paper proposes a Residual Network (RN)-based model for identifying 6 kinds of STBC signals with a single receiving antenna, including the same length of coding matrix. In our work, we use the frequency-domain correlation function of a single time delay as the training data of DL model. Then, we explore the suitable RN structure for blind identification of STBCs. Finally, we compare the RN model with convolutional neural network and traditional method, and test the performance of RN model. Simulation results show that our RN-based model provides good performance with low sensitivity to decay of the dataset, such as sample length and data size. At the same time, better identification accuracy can be achieved under the condition of different modulation types and channel fading parameters at low Signal to Noise Ratio (SNR).