基于多任务学习的联合调制识别和码元速率估计

To address the high computational complexity caused by sequential modulation recognition and symbol rate estimation in wireless communications, a deep learning-based multi-task neural network model is proposed, to complete two tasks at the same time in a single computation. The model employs a hard parameter sharing architecture to improve the convolutional long short-term memory neural network (CLDNN) model, replacing the long short-term memory network with a temporal convolutional network, which utilizes dilated causal convolutions to enhance temporal feature extraction and incorporating channel and spatial attention mechanisms to emphasize critical information. Experimental results demonstrate that the proposed model outperforms the comparative models in both tasks. Compared with CLDNN model, the modulation recognition accuracy of the proposed model is increased by 3. 81%, the symbol rate estimation error is reduced by 3. 29%, and the floating point operations and parameter count are reduced by 77. 00% and 67. 83%, respectively. Moreover, compared with single-task networks, the computational load is reduced by approximately 50% without compromising performance.