Study On Link Quality Prediction Mechanism For Wireless Sensor Networks

The nodes are often deployed in harsh field environment which influences thewireless wave used in the communication among nodes greatly. This will cause the lossof data. Then the retransmission mechanism is generally used which can improve thesuccess rate of data transmission to a certain extent but cause the waste of energy. So, ifthe upper-layer protocol can select a good quality link to transmit data, the transmissionrate of data will be improved and the energy will be saved by reducing theretransmission of data. Therefore, it is necessary to study the mechanism of the linkquality prediction in WSNs. This project comes from the National Nature ScienceFoundation: monitoring and studying the WSNs’ own health status used in the eventdetection.This paper firstly studies the link characteristic of WSNs through the experiment.Then two link quality prediction methods in WSNs are proposed，which are LQP-ES (aLink Quality Prediction method for WSNs based on Exponential Smoothing) andLQP-BP (a Link Quality Prediction method for WSNs based on BP artificial neural).LQP-ES method mainly includes the computational model of PRR value, thebuilding of model based on exponential smoothing and the routing metric of link quality.Combining mean RSSI with mean LQI to calculate the computational model of PRR isproposed according to the characteristics of mean RSSI, mean LQI and PRR.Exponential smoothing prediction algorithm is used to build the prediction model. ETXmethod is used to measure the routing of the link quality.LQP-BP method mainly includes the analysis and classification of the link and thedesign of BP artificial neural network. With mean RSSI as the standard, the link isdivided into sudden link and stable link. And different links use different methods tocollect samples. According to the design principles of BP artificial neural network, theprediction model used for times series prediction is designed.The experiment results show that both prediction methods can be used in theprediction of the link quality in WSNs. The calculation process of the LQP-ESprediction method is completed on the common nodes which have limited resources (forexample TelosB, Mica2, MicaZ et al) has less calculation. But the calculation process ofthe LQP-BP prediction method is completed on the high performance nodes which are embedded the Linux operation system (for example Imote2), has higher predictionaccuracy and more calculation compared with the former prediction method.