Strategies On Wake-up And Data Admission Of Nodes In Sensor Networks With Markov Decision Process

The sensor network technology is one of the most important technologies in the 21 st century.The network is usually self-organized by many sensor nodes.A sensor node is composed of sensor unit,processing unit,communication unit and power supply.The processing unit is the core module of a node,which has the functions of equipment control,task allocation and scheduling,data processing and transmission.However,the capacity and memory of the processing unit are limited.The operation of the processing unit is also easily restricted by a node's energy finitude.Sleep mechanism can effectively save the energy of nodes.But this mechanism seriously affects the efficiency of data collection and transmission.Thus,how to design the working schedule and data admission strategies are of great significance to a node.This dissertation uses Markov Decision Process(MDP)model to study the wake-up strategies,the next-hop node selection strategy and data admission strategies of a node in sensor networks under the sleep/wake-up mechanism.The main contributions of the dissertation are as follows:(1)The dissertation studies the optimal wake-up strategies of a node in a sensor network to obtain the maximum expected discounted reward.The working process of nodes in sensor networks can be regarded as the queue service process.A deterministic wake-up strategy based on the MDP model is proposed under the condition of no error and no inference during working process in this dissertation.Specifically,with the MDP model,a dynamic stochastic mathematical model is firstly established according to the sleep/wake-up mechanism,which vividly describes the working process of a node.Then,with the proposed model,a deterministic rule for when to wake up is studied and a policy is theoretically proved to obtain the maximum expected discounted reward.Later,according to the proposed optimal policy,the energy consumption is further analyzed.The theoretical results can be set in the sensor node to obtain better performance on working schedule and energy consumption so as to reduce the complexity of the node setting algorithm.Also,a new design idea of the sensor node is provided in this dissertation.(2)The dissertation studies a random wake-up strategy with minimum energy consumption of a node in a sensor network under the constraint of the expected discounted reward.Aiming at the unstable working environment,such as the channel interference,noise and weak network connection,an energy consumption model based on the Markov model is firstly studied.Then,a decision rule on the MDP model is proposed to make sure that the expected discounted reward is not reduced after wake-up action.Furtherly,an optimal policy of mimimum energy consumption is proposed.Based on the simulation,the optimal policy proposed in this dissertation makes the node achieve obvious advantages in the performance of expected discounted reward and energy consumption.(3)The dissertation studies the next-hop selection strategy for maximizing the average decision reward of a node under the delay constraints.Firstly,a next-hop selection model is proposed on the MDP model with the index of the residual energy,location,transmission energy and transmission delay of in the sensor network.Then,some theoretical results are proved for how to select the next hop under the stochastic dynamic sleep/wake-up mechanism to obtain the maximum average decision reward.The proposed next-hop selection algorithm in the dissertation can reduce the average variance of residual energy among network nodes,prolong the network life and improve the maximum number of transmitted packets of a node.(4)The dissertation studies the optimal data admission strategies of a node to achieve the optimal expected discounted reward.Data admission is an important issue in sensor networks due to the limited transmission coverage area and the limited battery capability.By constructing a suitable MDP model,this dissertation verifies that there exists an optimal(;)admission policy of when to admit or reject an arriving data packet,in order to achieve the maximum expected discounted reward in the sleep and wake-up phase,respectively.Furthermore,a formula for how to calculate the upper bound of the values of and is given out.For the identified model,the energy consumption of the sensor node is investigated.(5)The dissertation studies the random data admission strategies with the maximum throughput under the constraint of the expected discounted reward.As the traffic in a node increase,the throughput can be assumed to first increase and then start to decrease,indicating congestion in the buffer.This suggests the need for an admission mechanism to maintain high throughput.Considering the stochastic nature of sensor nodes,in this dissertation,by constructing a suitable MDP model,a probabilistic(;)admission algorithm on how to accept data packets on sleep and wake-up phases is proposed.Furthermore,the energy consumption of the model is investigated.The simulation shows that the proposed scheme can promote the performance of the data collection,throughput and the energy consumption.Finally,this dissertation concludes its research on the key technologies of sensor networks under sleep/wake-up mechanism based on MDP model.Some of the disadvantages of current research is analyzed and some improvements which needs further research are discussed briefly.