A MODULATION RECOGNITION ALGORITHM BASED ON ATTENTION MECHANISM AND GATED DENSE CONVOLUTIONAL NETWORKS

Automatic modulation recognition (AMR) is an important part of non cooperative communication system, and it is also a research difficulty in the field of communication. In order to solve the above problems, this paper proposes an attention based gated dense convolutional network (AGDCN) modulation recognition algorithm by combining JP2dense convolutional network (DenseNet), gated recurrent unit (GRU) and attention mechanism. The algorithm extractedJP the spatial and temporal features of the signal, and combined them to solve the problem of low recognition rate. Attention mechanism was added to the network to adaptively adjust the weight of GRU training process and effectively strengthen the learning of key features. Experiments show that the proposed AGDCN model outperforms other mainstream neural network algorithms. Specifically, when SNR exceeds 2 dB, the recognition rate of 11 modulation types can reach 90%.