Lightweight signal recognition based on hybrid model in wireless networks

Signal recognition is a key technology in wireless networks, with broad applications in both military and civilian fields. Accurately recognizing the modulation scheme of an incoming unknown signal can significantly enhance the performance of communication systems. As global digitization and intelligence advance, the rapid development of wireless communication imposes higher standards for signal recognition: (1) Accurate and efficient recognition of various modulation modes, and (2) Lightweight recognition compatible with intelligent hardware. To meet these demands, we have designed a hybrid signal recognition model based on a convolutional neural network and a gated recurrent unit (CnGr). By integrating spatial and temporal modules, we enhance the multi-dimensional extraction of the original signal, significantly improving recognition accuracy. Additionally, we propose a lightweight signal recognition method that combines pruning and depthwise separable convolution. This approach effectively reduces the network size while maintaining recognition accuracy, facilitating deployment and implementation on edge devices. Extensive experiments demonstrate that our proposed method significantly improves recognition accuracy and reduces the model size without compromising performance.