Leveraging multi-task learning to model task relationships in composite VLC waveform demodulation.

Area of Science:

• Electrical Engineering
• Computer Science
• Optical Communications

Background:

• Multi-task learning (MTL) is increasingly used in visible light communication (VLC) systems for demodulation.
• Existing research often treats modulation streams independently, limiting performance in composite waveforms.

Purpose of the Study:

• To introduce a novel modulation-relationship-driven framework for enhanced demodulation in VLC.
• To quantify the empirical relationships between different modulation tasks within a composite waveform.
• To enable modulation-aware physical layer (PHY) waveform design for resource-constrained VLC receivers.

Main Methods:

• A cross-stitch (CS) architecture was employed for joint classification of beacon phase shift keying (BePSK) and beacon position modulation (BePM).
• A learnable alpha-matrix was used to quantify inter-modulation coupling.
• A normalized coupling score was calculated to measure task dependency.

Main Results:

• The framework successfully jointly classifies BePSK and BePM within composite waveforms.
• A normalized coupling score of 0.71 was achieved, indicating significant task dependency.
• Feature sharing was transformed into a quantifiable design axis.

Conclusions:

• The proposed framework offers an interpretable measure of task dependency in MTL for VLC.
• Results facilitate modulation-aware PHY waveform design for efficient VLC systems.
• This work contributes to the development of explainable, NextGen VLC systems.