Intelligent health monitoring in 6G networks: Machine learning-enhanced VLC-based medical body sensor networks

Recent advancements in Artificial Intelligence (AI)-enabled wireless communication are accelerating the adoption of Sixth Generation (6G) technologies in critical environments such as hospitals, where traditional Radio Frequency (RF) systems face limitations due to electromagnetic interference concerns. Visible Light Communication (VLC), which utilizes existing LED-based lighting infrastructure, offers the dual advantage of high-speed data transmission and reduced electromagnetic interference (EMI). However, patient movement within clinical environments introduces considerable variability in signal reception and alters channel characteristics dynamically. This research introduces an innovative methodology that merges site-specific ray tracing with Machine Learning (ML) techniques to model VLC-based Medical Body Sensor Network (MBSN) channels across different hospital scenarios. The first contribution involves implementing a Q-learning-driven adaptive modulation algorithm capable of maintaining target symbol error rates (SER) in real time without requiring prior environmental data. The second component involves designing a Long Short-Term Memory (LSTM) model to estimate path loss and Root Mean Square (RMS) delay spread under time-varying hospital conditions. The third contribution presents a comprehensive ML-based position estimation framework that evaluates six different algorithms—Linear Regression, Support Vector Regression, K-Nearest Neighbors, Multilayer Perceptron (MLP), LSTM, and Gated Recurrent Units—for accurate patient localization. To date, this appears to be the first study that integrates ray-traced Channel Impulse Response (CIR) modeling with ML-driven analysis in medical environments. Simulation findings indicate that the Q-learning model reliably meets SER targets, with spectral efficiency (SE) performing close to optimal at near-threshold conditions. Additionally, the LSTM-based predictions reveal that in the Intensive Care Unit (ICU) scenario, the sensor at position D1 produces the largest Root Mean Square Error (RMSE) for both path loss (1.6797 dB) and RMS delay spread (1.0567 ns). Conversely, in the Family-Type Patient Room (FTPR), sensor D3 yields the highest RMSE for path loss (1.0652 dB) and delay spread (0.7657 ns), confirming strong predictive performance under realistic operational conditions. For position estimation, MLP emerges as the optimal architecture, achieving sub-meter accuracy of 58.6 cm for combined D1-D2-D3 sensor configurations in FTPR, with individual sensors yielding 63.5 cm (D1), 75.0 cm (D2), and 73.1 cm (D3), while in the more complex ICU environment, MLP maintains acceptable precision of 217.1 cm for D1-D2-D3, with paired configurations achieving 202.1 cm (D1-D2) and 216.3 cm (D1-D3), all while offering 35-48% reduction in computational requirements and 37-89% faster hyperparameter optimization compared to sequential models, making it the most practical solution for real-time patient tracking in healthcare facilities.