Practical Evaluation of Low-complexity Medium Access Control Protocols for Wireless Sensor Networks

This thesis studies the potential of a novel approach to ensure more efficient and intelligent assignment of capacity through medium access control (MAC) in practical wireless sensor networks (WSNs), whereby Reinforcement Learning (RL) is employed as an intelligent transmission strategy. RL is applied to framed slotted-ALOHA to provide perfect scheduling. The system converges to a steady state of a unique transmission slot assigned per node in single-hop and multi-hop communication if there is sufficient number of slots available in the network, thereby achieving the optimum performance. The stability of the system against possible changes in the environment and changing channel conditions is studied. A Markov model is provided to represent the learning behaviour, which is also used to predict how the system loses its operation after convergence. Novel schemes are proposed to protect the lifetime of the system when the environment and channel conditions are insufficient to maintain the operation of the system. Taking real sensor platform architectures into consideration, the practicality of MAC protocols for WSNs must be considered based on hardware limitations/constraints. Therefore, the performance of the schemes developed is demonstrated through extensive simulations and evaluations in various test-beds. Practical evaluations show that RL-based schemes provide a high level of flexibility for hardware implementation