Hybrid VLC-RF Channel Estimation for GFDM Wireless Sensor Networks Using Tree-Based Regressor.

Area of Science:

• Wireless Communication Systems
• Signal Processing
• Machine Learning Applications

Background:

• Hybrid channel estimation in wireless sensor networks (WSNs) using generalized frequency division multiplexing (GFDM) over visible light communication (VLC) and radio frequency (RF) links is crucial.
• Realistic hybrid channels involve additive white Gaussian noise (AWGN) and Rayleigh fading, posing challenges for traditional estimators like MMSE and LMMSE due to their rigidity in nonlinear conditions.
• Existing methods struggle with the heterogeneous and nonlinear nature of combined VLC/RF channels, necessitating novel approaches for accurate and efficient estimation.

Purpose of the Study:

• To propose a novel tree-based regression model for hybrid channel estimation in GFDM-enabled WSNs.
• To address the limitations of traditional estimators in complex, realistic channel environments.
• To develop a data-driven solution that balances accuracy with computational efficiency for resource-constrained WSNs.

Main Methods:

• A decision tree regressor was developed and trained using a dataset of 18,000 signal samples across 36 signal-to-noise ratio (SNR) levels.
• The model was evaluated against Support Vector Machine (SVM) and Random Forest algorithms for hybrid channel estimation.
• Performance metrics included accuracy, Bit Error Rate (BER), and inference time on a test dataset.

Main Results:

• The proposed tree model achieved competitive accuracy (90.83% at 10 dB, 97.63% at 30 dB) and low BER (0.0917 at 10 dB, 0.0237 at 30 dB).
• Inference efficiency was a key advantage, with the tree model completing predictions in 45.53 seconds, significantly faster than Random Forest (140.09s) and SVM (189.35s).
• A trade-off was observed: the tree model offers substantial computational savings at the cost of slightly lower predictive performance compared to ensemble methods.

Conclusions:

• The tree-based regression model provides an efficient solution for hybrid channel estimation in WSNs, particularly for real-time and low-power applications.
• Its rapid inference time makes it suitable for latency-sensitive wireless systems where computational overhead is a critical concern.
• The model represents a practical approach for scenarios prioritizing speed and resource efficiency over marginal improvements in estimation accuracy.