Abstract:

Aiming at the problems of the traditional space-frequency block code (SFBC) recognition method, such as the difficulty of extracting features manually, low recognition accuracy under low signal to noise ratio (SNR) and not suitable for non-cooperative comm unication, a SFBC automatic recognition method based on cross-correlation feature map and extended dense convolutional network is proposed. Firstly, the cross-correlation function in the frequency domain of the receiving end is computed and the dimensional transformation to obtain the two-dimensional cross-correlation characteristic graph is carried out. Then, the obtained feature map is preprocessed to enlarge the effective region of convolution kernel perception and remove the image redundancy information. Finally, the extended dense convolutional network is constructed to automatically extract the preprocessing image features and realize the SFBC classification and recognition. Simulation results show that when the SNR is -8 dB, the recognition accuracy of the SFBC signal of the proposed method reaches 96.1%. Compared with the traditional algorithm, the proposed method has better anti-low SNR and feature self-extraction ability, which verifies the effectiveness of deep learning method in the field of SFBC recognition, and lays a foundation for the subsequent research in this field.

Key words: non-cooperative communication, space frequency block code (SFBC), cross-correlation feature map, image preprocessing, deep learning, dilated dense convolutional networks (DDenseNet)