IoUT-Oriented an Efficient CNN Model for Modulation Schemes Recognition in Optical Wireless Communication Systems

The rapid advancement of the Internet of Underwater Things (IoUT) necessitates robust, high-capacity communication systems that can operate efficiently in the challenging conditions of underwater environments. Optical Wireless Communication (OWC) systems, leveraging the advantages of high data rates and low latency, offer a compelling solution for IoUT. However, accurate modulation recognition in these systems remains a significant challenge due to the variable nature of underwater channels. This paper explores the application of Convolutional Neural Networks (CNNs) for modulation recognition in the OWC systems, focusing specifically on 64-QAM (Quadrature Amplitude Modulation) and 32-PSK (Phase Shift Keying). A CNN model-based approach is proposed to automatically extract and classify modulation features from received signals, demonstrating superior performance compared to traditional recognition methods. The model is applied to a dataset of 626 simulated images, categorized into two modulation types: 64QAM and 32PSK. Keras and TensoFlow frameworks are used to implement the model, the CNN undergoes hyperparameter tuning and data augmentation to optimize accuracy. The model’s performance is assessed using a confusion matrix, along with precision-recall (PR) and receiver operating characteristic (ROC) curves. The experimental results show that the CNN achieves high accuracy in recognizing modulation types, with a testing accuracy of 100% and a testing loss rate of $1.82\times 10^{-6}$ . Additionally, the model records a Precision, Recall, F1-score, and area under the ROC of 100%. The experiments reveal that the CNN model achieves high accuracy in differentiating between 64-QAM and 32-PSK under varying underwater conditions, highlighting its potential for enhancing IoUT communication reliability.