High-Performance IEEE 802.15.4 MAC Protocol Based On Reinforcement Learning

With the rapid development of modern information technology,the importance of wireless sensor networks based on the IEEE 802.15.4 standard,which is one of the fundamental operating protocols for the Internet of Things,is increasing.However,since the nodes in the IEEE 802.15.4 standard contend for the channel using binary exponential backoff mechanisms,the valuable channel resources are largely wasted in the competition when the network has a large number of nodes.In addition,Reinforcement Learning(RL)is gradually being applied to the field of communications.After performance analysis on the Medium Access Control(MAC)layer based on the IEEE 802.15.4 standard,this thesis aims to employ RL algorithms to the wireless sensor networks in order to improve overall channel usage and transmission throughput of wireless sensor networks.Firstly,this thesis provided an overview of the IEEE 802.15.4 MAC protocol.It mainly explained the types of devices,topologies of networks,Carrier Sense Multiple Access/Collision Avoidance(CSMA/CA)algorithm under beacon-enabled mode or non-beacon enable mode and various types of data transfer.Secondly,in a star topology network under saturated traffic condition,a Markov model is established and numerically analyzed for the IEEE 802.15.4 slotted CSMA/CA algorithm in beacon-enabled model.And the performance of the saturated networks with or without ACK is compared,accordingly.Thirdly,the channel utilization and‘effective channel utilization' of the star network based on the IEEE 802.15.4 MAC protocol with ACK mechanism are further investigated.This thesis further proposed that 'effective channel utilization' can capture accurate throughput performance of network compared with conventional channel utilization rate.Fourthly,the influence of network's parameters,such as the number of nodes,the length of data packet and backoff window size,is further analyzed.And we proposed that the 'effective channel utilization' can be improved by changing the backoff window size.When the network's environment is relatively stable,there exists an optimal backoff widow size to achieve high throughput.Finally,this thesis proposed an adaptive backoff window selection algorithm based on RL algorithm.That is,the network coordinator adopts the RL algorithm and can adaptively select backoff window size according to the environment of the network,and dictates the length of the backoff window size of all end devices,which can maximize the 'effective channel utilization' while ensuring that all the end devices have fair access to the channel.