Low-Power Wireless Link Quality Prediction Based On Neural Network

A good link quality prediction mechanism is essential for low-power wireless links.Real-time and accurate link quality prediction can perceive changes in link quality in advance,quickly select better links for data transmission,and effectively improve network transmission efficiency.Therefore,it is more suitable for time-varying low-power wireless sensor networks.However,in actual wireless sensor network applications,the existing link quality prediction mechanism lacks sufficient consideration for the impact of link fluctuations.At the same time,the mapping error between the physical layer parameters and the packet reception rate is introduced to make its prediction accuracy Lower.This paper analyzes the research background and significance of wireless link quality prediction,and then researches the research status of link quality prediction at home and abroad.The main contents of this paper are as follows:(1)Using to the measured data of different link quality obtained in multiple experimental scenarios,the statistical characteristics and convergence of the three common link quality parameters of packet receiving rate,received signal strength indicator and link quality indicator are analyzed.(2)Summarizes the traditional link quality prediction model.The existing link quality prediction mechanism lacks sufficient consideration of the impact of link fluctuations.A more effective link quality prediction mechanism WNN-LQP is proposed for links with large fluctuations.The experimental results show that,compared with the similar method WNNLQE,WNN-LQP has higher prediction accuracy,especially in the case of large link fluctuations,so it is more suitable for low-power wireless links.Simultaneously,the limitations of wavelet neural network itself and the prediction error of WNN-LQP is still large,a link quality prediction method RNN-LQP with memory effect is proposed.This mechanism makes use of the short-term memory characteristics of recurrent neural networks,which can more accurately predict link quality.Compared with WNN-LQP,RNN-LQP has higher prediction accuracy and reliability.(3)Aiming at the problems of traditional link quality prediction methods,a new link quality fast prediction model is proposed.This model can avoid the errors caused by the mapping model,so as to achieve more accurate PRR prediction.Based on the model,a fast link quality prediction method WNN-FLQP based on wavelet neural network is proposed.The analysis of the prediction effect under the small time window and the large time window respectively shows that by eliminating the mapping error between physical layer parameters and PRR,WNN-FLQP can predict link quality more accurately without reducing the prediction agility.(4)In order to further verify the effectiveness of the proposed fast prediction model,and further reduce the prediction error of WNN-FLQP under different links,a fast prediction method of link quality considering the memory effect RNN-FLQP is proposed.The prediction effects under the small time window and the large time window are analyzed,respectively,which further verifies the effectiveness of the proposed fast prediction model.Compared with the existing prediction methods,the prediction effect of this method is obviously better.Finally,the correlation analysis of the predictable time is carried out.The results show that the longer the future time window is,the worse the prediction effect is.