Performance Studies on machine learning based channel modelling for vehicular visible light communication

Car headlights and taillights, to provide cost-effective and high-data-rate interference-resilient communication, have spawned Vehicular Visible Light Communication (V2LC), a possible additional technology to radio-frequency systems. Since they can be formulated in an empirical or deterministic way, traditional optical channel models can be easily adapted to changing vehicle environments. To counter the weakness of the conventional modelling techniques, the paper examines channel modelling techniques grounded in machine learning and suited to V2LC scenarios. A few variables, such as the shape of LEDs, speed, road topology, and atmospheric perturbations, do significantly influence the received signal strength in vehicular optical channels that are highly non-linear in nature and vary over time. To solve these issues, the paper gives Machine Learning-based Vehicular Visible Light Communication Channel Modelling (ML-V2LC-CM), a hybrid learning system as a proposed framework. The regression models and ensemble learning of the framework and deep neural predictors are all integrated, which can be used to estimate the channel gain, path loss, and signal distortion. Compared to baseline empirical models, experimental tests show that ML-V2LC-CM is significantly more effective. The framework performs well and has the highest performance of 18.7% in terms of improvement in prediction, 22.4% in terms of reduction in Root Mean Square Error (RMSE), Signal-to-Noise Ratio (SNR) estimation error of less than 1.6 dB, and low prediction latency of 4-7ms. It is very robust when blocked with a degradation rate of 6-9% and gives over 92% generalization behaviour under a variety of lighting conditions and provides fusion-layer improvement of 11.3% greater than the best standalone model. The model demonstrates high consistency in channel stability estimation of more than 95% even in high rate of movement of vehicles.