Blind Recognition of Convolutional Codes: A Matrix Transformation-Aided Deep Learning Approach

This paper focuses on the blind recognition of convolutional codes, a significant research problem in cognitive radios and signal interception. The existing methods based on deep learning (DL) usually directly take the received sequence as the input of a network, of which the recognition accuracy for high-rate convolutional codes is often poor. An identification framework called MT-CNN, combining matrix transformation with convolutional neural networks (CNN), is proposed in this paper. We offer a novel matrix transformation algorithm of which the result can highlight the differences between different encoders. Our proposed MT-CNN method adopts a feature fusion strategy, employing the codeword matrix and its feature map obtained through matrix transformation as the network's input. Simulations show that the proposed approach could provide significant improvements compared to existing methods, especially for the convolutional codes with a high rate.