Efficient medium access control protocols using reinforcement learning for collision mitigation in wireless sensor networks

Energy efficiency is of utmost importance for the resource-constrained wireless sensor networks (WSN). Rightly so, there are numerous methods/techniques/algorithms proposed in the literature to improve the energy efficiency from the hardware and software standpoints. This thesis focusses on reducing energy wastage by mitigating collisions in WSN using intelligent communication protocols. The main objective of the thesis is to improve the efficiency of the MAC protocols through dynamic frame size adaptation and optimal channel access strategy. The first protocol which is the reinforcement learning based dynamic frame size adaptation (Smart-DFSA) proposed to address the dynamic frame size adaptation issue faced by the Aloha protocol in the radio frequency identification (RFID) networks. The protocol needs to work with partial or incomplete information about the tag population in each interrogation round and dynamically adapt the frame size to maximize the throughput. The reinforcement learning technique is used to select an optimal frame size for current round based on the collision information from the previous frame. The efficiency of the protocol in improving throughput, delay, and protocol overheads are evaluated by comparing it with the current industry standard protocols, namely Electronic Product Code Class 1 Generation 2 (EPCC1G2) and its variants, in sparse and dense tag populations. The second protocol, namely Bandit-MAC solves the collision in Aloha protocol by slowly replacing the blind transmission strategy with an efficient medium access strategy similar to the timedivision medium access (TDMA) method. In this regard, the reinforcement learning technique is used to learn a time slot accessing schedule in a distributed manner which provides a unique slot for each node in the network. The proposed protocol is evaluated using extensive simulations by comparing it with relevant protocols from the literature in terms of various performance metrics including throughput and delay. The results show that Smart-DFSA performs equal or better than EPC-C1G2 protocol in delay, throughput and time system efficiency when simulated for sparse and dense environments while requiring one order of magnitude lesser message exchanges between the reader and the tags. Energy efficiency is achieved by reducing the control message overhead which consumes more energy than the computation. As for the Bandit-MAC, the results show clearly that the proposed protocol is superior to the base protocol (frame slotted Aloha) by 39.3% in throughput, 97.3% in delay and 36.7% in packet delivery ratio performances while being close to the performance of a TDMA protocol which is the LMAC protocol.