Deep learning for signal clock and exposure estimation in rolling shutter optical camera communication.

Area of Science:

• Optical Camera Communication (OCC)
• Deep Learning Applications
• Signal Processing

Background:

• Rolling shutter (RS)-based optical camera communication (OCC) links face bandwidth limitations due to camera exposure time.
• Longer exposures cause intersymbol interference (ISI), corrupting received signals.
• Reducing exposure time increases bandwidth but results in dark images unsuitable for practical applications.

Purpose of the Study:

• To develop a deep learning (DL) estimator for unknown transmitter clock and camera exposure time in RS-OCC systems.
• To enable OCC receivers to operate with various cameras without prior knowledge of their internal settings.
• To mitigate the impact of exposure-related ISI on signal reception.

Main Methods:

• A DL-based estimator was trained using synthetic images generated for numerous scenarios.
• The estimator analyzes distorted images to deduce transmitter clock and camera exposure time.
• Validation was performed using over 7000 real-world images.

Main Results:

• The DL estimator achieved relative errors below 1% for transmitter clock estimation.
• Relative errors were below 2% for camera exposure time estimation.
• These low errors ensure optimal performance for subsequent equalization and decoding stages, maintaining bit error rates below the forward error correction limit.

Conclusions:

• The proposed DL estimator effectively determines critical camera and transmitter parameters in RS-OCC.
• This approach decouples receiver operations from specific camera models, facilitating cloud-based architectures.
• The estimator is crucial for reliable OCC reception under moderate exposure conditions.