Image-Based VLC Signal Demodulation Using Machine Learning

Demodulation of visible light communication (VLC) signals using intensity modulation direct detection is limited by the noise inherent in the signal. To address this issue, we propose an enhanced machine learning (ML) image-based demodulator for on-off keying (OOK) modulated VLC signals. We designed and implemented a transmitter and receiver equipped with sensors to collect real-time environmental data. The transmission distance is varied, and the received waveform is converted into images. To minimize the computational load of the demodulator, we apply bicubic interpolation and image thresholding techniques to these images. Subsequently, we developed an ML-based demodulator using MobileNetV2 and trained the model with the collected dataset. To enhance the model’s versatility and accuracy, we used data augmentation techniques. Experimental results indicate that the proposed ML-driven demodulator significantly extends the communication range and increases noise tolerance, achieving a demodulation accuracy of 97.58%.