Vehicular visible light communication channel modeling and performance analysis

Highly automated driving and autonomous vehicles rely on numerous sensors, technologies, and sophisticated software to ensure road safety with limited or no human intervention. Connected vehicle technologies, enabling the exchange of information between vehicles, infrastructure, and other road users, are provisioned to be among the key enablers of autonomous driving. Current radio frequency (RF) based IEEE 802.11p and cellular-vehicle-to-everything (C-V2X) technologies are mainly adopted to support basic safety applications for connected vehicles. However, these technologies suffer from the scarcity of the RF spectrum, together with the network congestion and security related concerns, where they lack to address advanced driving use cases for automated driving. Therefore, Vehicular Visible Light Communications (V-VLC) is recently proposed as a complementary vehicular communication scheme, which aims to provide secure directional communications in the license-free visible light spectrum by using vehicle light emitting diode (LED) lights. However, the performance of V-VLC systems mainly depends on the V-VLC channel characteristics. In this thesis, comprehensive understanding and modeling of the V-VLC channel with respect to varying environmental conditions and link geometries is investigated along with the practical V-VLC system performance evaluations. First, we provide an up-to-date literature review regarding the V-VLC channel modeling. Consequently, we highlight that measurement based comprehensive VVLC channel modeling is missing in the literature. Second, we propose a channel frequency response (CFR) based channel modeling with channel sounding, measurement data processing, and characterization steps. We demonstrate that nearby vehicle existence increases the channel gain, while root-mean-square (RMS) delay spread increases with the longer inter-vehicular link distance. We demonstrate that with the increasing link distance, multi-path components become more distinguishable since the optical receiver captures more reflection components from the road surface and nearby objects. Third, we investigate the various environment and link scenarios for V-VLC channel and noise modeling. We conduct indoor V-VLC channel characterization to evaluate the nearby vehicle, building columns, floor slabs, and wall effects on vehicle-to-vehicle (V2V) V-VLC signal propagation. We demonstrate that the occupied lane and receiver inclination angle play a crucial role in the propagation of V-VLC signals, where reflections increase received signal strength (RSS) beyond 25 m for next lane scenarios and nearby vehicle reflections can be considered negligible for 30 ° tilted receiver for the same lane scenarios. We further characterize the V2V V-VLC channel under turbulence conditions by using seasonal and daily atmospheric refractive index measurements. We statistically characterize and classify the turbulent V-VLC channels of different weather conditions. Moreover, the nonline-of-sight (NLoS) V-VLC channel due to vehicle reflections is characterized by the introduction of an empirical NLoS V-VLC path loss model and the vehicle surface dependent channel impulse response (CIR) model. We model the V-VLC channel noise by identifying the time correlation properties of different ambient light conditions, where the proposed models enable V-VLC channel noise synthesis to simulate V-VLC noise behavior. Fourth, we propose a data-driven Machine Learning (ML) based channel modeling, which incorporates multiple channel input variables to predict path loss and CFR. The proposed model employs, multi-layer-perceptron based neural network (MLP-NN), radial basis function (RBF), and random forest machine learning algorithms with data pre-processing and hyper-parameter selection steps adopted for V-VLC channel measurement data. This model provides an adaptable and generic methodology to a variety of V-VLC scenarios since the models can be expanded with additional channel input and output variables. Moreover, the developed models can be used for system simulations with various channel inputs. Fifth, real-world V-VLC system implementation with channel measurement and modeling based system performance evaluations are proposed. We implement an IEEE 802.15.7 standard compliant, Software-Defined Radio (SDR) based V-VLC system, to evaluate different physical layer (PHY) modes and inter-vehicular distances in an indoor parking garage environment. The developed scheme has the ability to evaluate the performance of evolving visible light communications (VLC) standards and algorithms for various channel conditions. We further propose a cooperative multiple input multiple output (MIMO) orthogonal frequency division multiplexing (OFDM) V-VLC scheme and provide performance evaluation by using channel measurement data from the production vehicle's brake lights. Finally, we evaluate the performance of ML aided V-VLC channel path loss prediction scheme with the comparison of the traditional fitting based V-VLC path loss model and IEEE 802.11p path loss prediction models. The proposed V-VLC model demonstrates 2.34 dB better path loss prediction performance than the fitting based model. In summary, this thesis provides V-VLC channel modeling methodology and practical V-VLC channel models which are useful for system design, simulations, and performance evaluations. Furthermore, it validates the practicality of the proposed models and methodologies while identifying future research directions to support novel V-VLC applications.