Machine Learning Assisted Ultra Reliable and Low Latency Vehicular Optical Camera Communications

Optical camera communication (OCC) has emerged as a key enabling technology for the seamless operation of future autonomous vehicles. By leveraging the supreme performance of OCC, the stringent requirements of ultra-reliable and low-latency communication (uRLLC) can be met in vehicular OCC. In this thesis, a rate maximization approach is presented to vehicular OCC that aims to optimize vehicle speed, channel code rate, and modulation order while adhering to uRLLC requirements. The reliability is modelled by satisfying a target bit error rate (BER) and latency as transmission latency. To improve transmission rate and reliability, low-density parity-check codes and adaptive modulation are adopted in this thesis. First, the rate maximization problem is formulated as an optimization problem aimed at determining vehicle speed, channel code rates, and modulation order given reliability and latency constraints. Even for a small set of modulation orders, this problem is mixed integer programming, which is NP-hard. To overcome the complexity of the NP-hard problem, the proposed optimization problem is modelled as a Markov decision process and then solved it distributively using multi-agent deep reinforcement learning (DRL). Then, the optimization problem is solved using the actor-critic DRL framework with Wolpertinger architecture. A deep deterministic policy gradient algorithm is employed to operate over continuous action spaces. The proposed model and optimization formulation are justified through numerous simulations by comparing capacity, BER, and latency. From the findings, it is clear that the multi-agent DRL framework in vehicular OCC leads to improved performance in terms of maximizing the communication rate while respecting uRLLC. This work constitutes a significant step towards addressing the challenges in vehicular OCC to respect uRLLC.