The rise of intelligent transportation systems has transformed how vehicles communicate, letting them share real-time data with each other and roadside infrastructure to boost safety, smooth out traffic, and make urban driving more enjoyable. But as more vehicles connect in crowded cities, the pressure on limited wireless spectrum grows intense, exposing the weaknesses of old-school orthogonal multiple access (OMA) methods. OMA sticks to a rigid one-user-per-channel setup that just can’t keep up with lots of vehicles, wasting resources and often dropping the ball on Quality of Service (QoS). Enter non-orthogonal multiple access (NOMA) a game-changer that lets multiple users share a channel by smartly juggling power levels and using successive interference cancellation to sort out the signals. It’s a big leap for efficiency, but rolling it out in vehicular networks isn’t easy. Things like pairing users, assigning channels, controlling power, and flipping between NOMA and OMA modes get tricky, especially with cars zooming around and signals shifting fast.

This thesis tackles those challenges with an adaptive framework that pulls together RSU-vehicle user (VU) association, channel assignment, NOMA?OMA switching, and power optimization into one cohesive package. It blends solid math-based optimization with Reinforcement Learning (RL) to handle the chaos of urban networks on the fly. Starting with a mathematical model, we cut total transmit power by 27% and boosted user association fairness by 18% measured with Jain’s fairness index leaving baselines like random assignment and utility-based methods in the dust. Then, RL steps in, teaching the system to adapt as things change: user numbers, channel strength, QoS needs. The RL agent learns on its own how to pair users, tweak power, and switch modes, aiming for top-notch energy savings, throughput, and fairness without being tied to rigid rules. Plus, we’ve baked in low-complexity tricks like binary relaxation and problem splitting to keep it fast enough for real-time use.

We put it through its paces with thorough MATLAB simulations, testing everything from static setups to busy scenes with 20 vehicles and five RSUs juggling multiple channels. The results shine: it’s scalable, tough, and efficient, especially when vehicles swarm urban intersections and the network keeps shifting. This work merges optimization smarts with RL’s adaptability, delivering an energy-efficient, fair, and scalable fix for NOMA-powered vehicular networks. It also sets the stage for digging deeper into distributed setups, complex traffic patterns, and even real-world trials pushing us closer to rock-solid communication for tomorrow’s urban roads.