Research On Life Prediction Technology Of Radio Frequency Oscillator Considering Degeneration And Measurement Uncertainties

In order to solve the security risks caused by the development of radio frequency(RF)technology,this paper summarizes domestic and foreign scientific research achievements on methods of lifetime predicting of RF circuits.Crystal oscillator is chosen as the object of this study.After analyzing the characteristics of the RF crystal,this paper extracts the phase noise as a reflection of the external effects in the process of the aging of the RF oscillator.Elaboration thereform begins from the prospective of acceleration effect and phase noise power-law principle,combining with a series of uncertainty analysis to obtain remaining useful life.According to the uncertainty analysis,the origin and influence of the errors in life prediction are considered and the experimental results are presented in the form of probability interval and point estimation respectively.This research mainly contains into three parts:1.The extraction of life characteristic of RF crystal oscillator.This section shows the key degrading parameters of the RF crystal and the change subjected to acceleration.2.The method of life prediction.The prediction sequence of acceleration sensitivity is transformed into adjusted cosine similarity value and conducted by the optimal pruning extreme learning machine based on M-estimation(M-OPELM).As for the phase noise prediction method,Lesson theory of the phase noise of oscillator is constructed as a linear sum of the negative power components at different frequency offsets.Then the least-squares-estimation is employed to find out the degradation law of coefficient of negative power components changed along with operation time.The coefficient data is put into the exponential degradation model with uncertainty drift,applying the Bayesian framework to iterately update the model parameters at each monitoring time and obtain the next state.3.The analysis of uncertainty.In this paper,the sources of uncertainties generated from data acquisition to result prediction are introduced and reflected in the two forms of methods.A method for reducing uncertainty is proposed by correcting the occasional error of extrapolated prediction,conducting based on similarity between historical samples and field samples.This appoach reflects the process of uncertainty spreading in predicting.It is proved that the two kinds of characteristic quantities combined with uncertainty analysis obtain good results in the prediction of the residual life of RF circuit,and presents certain significance in application.