Modulation Recognition Method for Wireless Signals Based on Joint Neural Networks

In the realms of Internet of Things (IoT), satellite communication, and related scenarios, automatic modulation recognition is crucial for accurate signal demodulation. In complex communication environments, accurately identifying diverse modulation types is a challenging task. This paper introduces an automatic modulation recognition approach leveraging a joint neural network framework. The method integrates a flow-based collaborative training module for signal enhancement, a deep learning mechanism for feature extraction, and a two-dimensional sparse weighting mechanism. This method enhances the input signal through enhancement processing and strengthens attention to different dimensional features via a weighting mechanism, thereby suppressing irrelevant features with lower weights. The network architecture is optimized in terms of layer depth and connectivity to enhance modulation identification accuracy and model stability under non-ideal conditions. Experimental evaluations conducted on the RML2016.10a dataset across varying SNR demonstrate the method’s robustness in low SNR environments and its effective recognition performance for high-order modulated signals compared to baseline models.