University Archives

Accurate classification of radio frequency (RF) signal and modulation types remains a significant challenge in complex, real-world wireless environments. This study presents an experimental, quantitative evaluation of a deep learning approach for joint RF signal and modulation classification. A multi-task convolutional neural network (CNN) model is proposed to simultaneously identify signal type and modulation scheme. The model was trained and evaluated using three datasets generated by Andro Computational Solutions, LLC including RadComDynamic, RadComAWGN, and RadComOTA2.45GHz, filtered to signal-to-noise ratios (SNR) ranging from 0-18 dB. The proposed model achieved a modulation type classification accuracy of 93.8% and a signal type classification accuracy of 96.5% on all three datasets, maintaining strong performance across varying SNRs. Additionally, a proof-of-concept real-time classification pipeline was implemented to evaluate the feasibility of deploying the model for live RF signal identification. The results indicate strong generalization and robustness, supporting the practical deployment of deep learning-based multi-task models for RF signal classification in complex environments.