Link Quality Estimation Method For WSNs Based On Stacked Autoencoder

Wireless Sensor Networks(WSNs)are made up of a large number of nodes deployed in a monitoring area with self-organized manner.Links are the basis of the interconnection of nodes and communication.The limited resources of sensing nodes,the complexity of the monitoring environment,and the diversity of noise lead to the directionality,asymmetry,instability,and "gray zone" of communication and other space-temporal characteristics,which cause volatility of communication and affect data transmission.Effective link quality estimation is the basis of topology management and routing control.It can guarantee the transmission of data,as well as improve the throughput rate,and extend the life of the entire network.Considering the asymmetry of the link and the feature extraction ability of Stacked Autoencoder(SAE),this thesis proposes a Stacked Autoencoder-based link quality estimator(LQE-SAE).By analyzing the asymmetry of the receive signal strength indicator(RSSI),link quality indicator(LQI)and signal to noise ratio(SNR),the uplink and downlink of the parameters are selected.Zero-filling method is used to process the missing link information caused by packet loss during the period of detection cycles.Then,the SAE model is used to extract the asymmetric characteristics of the uplink and downlink for RSSI,LQI and SNR,respectively.Furthermore,the asymmetric features are fused to obtain the link features.These features are given as inputs of the support vector classification(SVC),for which the link quality level divided by PRR is taken as its label,then the link quality level can be derived.The validity of the estimator was verified by the accuracy.The performance of the estimator on five different link quality levels was analyzed by the recall rate.And the stability of the estimator was evaluated by the stability grades.This thesis collects link quality information by our link quality test bed for training the model and analysing the information in indoor corridors,groves,parking lots,and roads.Compared with the link quality estimator based on SVC,Extreme Learning Machine(ELM)and Wavelet Neural Network(WNN)in four scenarios,the results show that our estimator is superior to the SVC,ELM and WNN in accuracy.The stability experiment was carried out in other corridors by adjusting the degree of interference,and the results show that the LQE-SAE has good stability.