A Study on Modulation Classification of Deeply Integrated Networks Based on Neural Engineering

Modulation classification plays a crucial role in advancing wireless communication research involving the recognition of various radio signals. Most existing neural network-based modulation classification methods rely on a single architecture and often face limitations in feature extraction from signals. In order to further improve the accuracy of modulation recognition, this paper proposes a deep hybrid network that combines convolutional neural networks, Long Short-Term Memory Networks (LSTM), and Transformers to extract signal features from different dimensions, and then uses a three-layer fully connected layer for classification. The network first extracts local features of the signal through convolutional neural networks, then uses LSTM to capture sequential patterns of signal features, and finally uses Transformer's unique multi head self-attention mechanism to capture global and long-term dependencies between local features. Experiments have shown that this method is particularly effective in analyzing time-series data and can capture complex patterns between signals well. Compared with current advanced modulation classification networks on the RML2016.10a dataset, the overall classification accuracy has reached 62.09%, improving from 3.55% to 6.46%, with accuracy improvements of 7.3% to 13% above OdB.