Extension of the CNN-Based Demodulation Method for Image Sensor-Based Visible Light Communication Considering Real Image Parameters

This paper proposes a convolutional neural network (CNN)-based demodulation method to enhance the performance of visible light communication (VLC) between digital signage and mobile terminals such as smartphones. Unlike conventional methods, the proposed approach employs a sliding window mechanism to enable flexible demodulation of data signals of arbitrary size by scanning a compact CNN trained to demodulate <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$3 \times 3$</tex-math></inline-formula> data signal cells. The model also incorporates spatial context from surrounding cells to improve robustness against inter-symbol interference. To ensure adaptability to real-world conditions, the CNN is trained using simulated received images that reproduce degradation effects—such as noise, blur, and displacement—extracted from actual captured images. The proposed method is evaluated in an indoor experimental setup using an OLED display and a USB camera, replicating a practical communication scenario between signage and an image sensor. Communication experiments were conducted using 24 monochromatic background colors from the Macbeth Color Chart with varying signal intensities applied to the Cb component in the YCbCr color space. The results show that the proposed method significantly outperforms the conventional threshold-based demodulation approach, particularly under low signal intensity conditions, thereby demonstrating its effectiveness for practical applications.