Deep-Learning-Based Adaptive Symbol Decision for Visual MIMO System with Variable Channel Modeling.

Area of Science:

• * Optical Communications
• * Machine Learning
• * Image Processing

Background:

• * Visual MIMO systems using LED arrays and cameras face challenges with real-world channel distortions affecting color information.
• * Existing color-similarity-based image processing methods are limited in correcting these distortions.
• * Symbol decision in these systems is complex due to channel variability.

Purpose of the Study:

• * To develop an advanced channel modeling and deep learning-based symbol decision method for visual MIMO systems.
• * To enhance communication performance between variable-color LED arrays and cameras despite channel distortions.
• * To create a robust classifier capable of adaptively determining symbols from distorted channel data.

Main Methods:

• * A novel channel modeling approach is proposed, reflecting real-world distortions beyond traditional noise models.
• * A deep neural network is designed with integrated channel identification (2D CNN) and symbol decision (1D CNN) modules.
• * The network is trained end-to-end, learning channel identification vectors and symbol correlations simultaneously.

Main Results:

• * The proposed deep learning method demonstrates significant performance improvements in a realistic channel distortion environment.
• * Average performance gains of approximately 41.8% over Euclidean distance and 6.3% over SVM methods were observed.
• * The system achieved up to 54.8% improvement against Euclidean distance and 9.2% against SVM.

Conclusions:

• * The proposed deep learning-based symbol decision and channel modeling method effectively improves visual MIMO system performance.
• * The approach provides a robust solution for accurate symbol determination in the presence of complex channel distortions.
• * This work advances optical communication techniques by integrating advanced machine learning for enhanced reliability.