Convolutional Neural Network Modulation Recognition by Incorporating Attention Mechanisms

In order to solve the drawbacks of the traditional manual extraction modulation recognition technique with the inability to accurately establish the signal feature description and high complexity, this paper proposes an automatic modulation recognition method by Convolutional neural networks and attention mechanism (CNN-AM). The CNN-AM model first incorporates channel preprocessing and pruning operations to reduce computational complexity. Meanwhile, CNN-AM uses a convolutional neural network incorporating the attention mechanism for modulation signal feature extraction and dynamically adjusts the network weights so that the network can adaptively focus on the signal features that are most effective for modulation type recognition and improve the accuracy of modulation recognition. The experimental results show that CNN-AM has lower complexity and higher accuracy compared to 1DCNN-PF and Conv1D_LSTM, which verifies the effectiveness and superiority of the fused attention mechanism in automatic modulation recognition technology.