Modulation Recognition of Aliasing Electromagnetic Signal Based on Improved Capsule Network

As the fifth-generation (5G) /the sixth-generation (6G) communication technologies evolve rapidly, the electromagnetic spectrum has become increasingly congested, resulting in signal aliasing interference. This poses a significant challenge for signal modulation identification. To effectively address this issue, we present a deep learning model that combines the attention based mechanism of Capsule Network (CapsNet) with Long and Short-term Memory Networks(ACLNet). Firstly, we transform the aliasing signal recognition problem into a multi-label classification task by adjusting the activation and loss functions of the output layer. Secondly, we incorporate an attention mechanism that enables ACLNet to automatically concentrate on the most relevant information for the task, thereby enhancing the model's performance and generalization. Furthermore, by leveraging the structural benefits of CapsNet and Long and Short-term Memory Networks(LSTM), ACLNet is better equipped to capture the signal's time-frequency characteristics and hierarchical structure. Finally, we adopt a dynamic threshold to determine output labels to adapt to the variations in SNR. The experimental results demonstrate that our proposed ACLNet exhibits significant recognition performance advantages in multi-signal aliasing scenarios.