Research On X-ray Pulsar Cyclostationary Signal Processing Algorithm

Deep space exploration is a great journey for human beings to acquire space knowledge,go deep into the universe and realize self-awareness.The discovery of pulsar provides a new direction for exploring large-scale space-time.At the same time,as a new autonomous celestial navigation method,X-ray pulsar navigation provides navigation information for spacecraft,especially for deep space navigation,by using the characteristics of X-ray pulsar's celestial distribution and cosmic lighthouse.Aiming at the key problems in the field of X-ray pulsar signal processing,using the characteristics of pulsar rotation period stability and pulsar signal cyclic stability,this dissertation effectively realized the known pulsar signals identification,candidate pulsar signals searching and pulsar phase information extraction by extracting effective time-frequency characteristics and combining with related neural networks respectively.The main research contents are as follows:The dissertation summarized the basic physical characteristics of pulsar,including its radiation mechanism,basic classifications and signal characteristics.According to the cyclostationarity of X-ray pulsar signal,the common analysis methods were given,including cyclostationarity definition,low-order and high-order information of cyclic statistics,and the corresponding time-domain and frequency-domain feature representation methods.The feature representation method of pulsar signal based on cyclostationary theory was studied.Firstly,a mathematical model was established based on the cyclostationarity of the signal,in which periodic jitter,correlation interference and Gaussian noise were considered.The signal model was divided into a definite component and a stochastic component.Study on their spectrum characteristics shows that the spectrum of useful signal is discrete,and the spectrum of random signal is continuous.The cyclic statistics was introduced to analyze the distribution characteristics of the second-order and third-order cyclic spectrum(cyclic bispectrum)of pulsar signal,and the direct estimation strategy of cyclic bispectrum was given.Compared with the traditional spectrum,it has better anti-noise and anti-interference characteristics and more abundant phase information.Finally,the X-ray Timing Explorer(RXTE)observation data was introduced and the proposed method was verified based on simulation and measured data.The feature representation method provides a modeling idea and effect guarantee for the subsequent combination with neural network.A method for the identification of known X-ray pulsars based on the bispectrum and a deep convolution neural network(DCNN)was proposed.Due to physical limitation,such as photon flux,propagation distance and detector area,the received pulsar signal tends to be weak.Hence,traditional spectrum searching techniques based on fast Fourier transform(FFT)need to accumulate observation data for a relatively long period of time to obtain a sufficient signal-to-noise(S/N)gain.In this regard,this dissertation used a highorder spectrum estimation method including non-uniform sampling,high pass filtering and autocorrelation to suppress the noise to a great extent.At the same time,combining this two-dimensional feature with Goog Le Net made full use of DCNN's advantage in twodimensional data mining so that the pulsar signal can be accurately identified.The Rossi X-ray Timing Explorer(RXTE)data from three pulsars,PSR B0531+21,PSR B0540-69 and PSR B1509-58,were selected for the experiment,and the identification task was realized with a classification accuracy greater than 90%,with observation times of only0.5 s,40 s and 15 s,respectively.Further experiments reveal that the high-pass filter and autocorrelation can effectively suppress the cosmic background and random noise and that the nonuniform sampling of the bispectrum can avoid frequency leakage.Although the time complexity(O(N2))of the algorithm is higher than those of the traditional FFT methods(O(Nlog N)),the algorithm reduces the requirement of the observation duration time.Thus,the computational complexity is comparable to that of traditional methods.A new concept of Two-Dimensional Autocorrelation Profile Map(2D-APM)was proposed,which was combined with DCNN to search for unknown X-ray pulsar signals.It is an important subject in pulsar astronomy to search for effective information from numerous unknown pulsar signals.Starting from the space X-ray pulsar signals,the 2D-APM feature modelling method was proposed utilizing epoch-folding of the autocorrelation function of X-ray signals and expanding the time-domain information of the periodic axis.Compared with the traditional profile,the model has a stronger anti-noise ability and more abundant information and characteristics of higher consistency.The new feature was investigated with double Gaussian components and it reveals that the characteristic distribution of the model is closely related to the distance between the double peaks of the profile.After that,the deep convolutional neural network named InceptionRes Net was built based on tensorflow framework.In order to overcome data imbalance and to improve network generalization ability,the generation strategy of sample set,network training scheme and hyper parameters adjustment strategy were given.After the network converges to a stable state,more than 99% of the pulsar signals were recognized and more than 99% of the interference were rejected successfully.This result verifies the high degree of agreement between the network and the feature model and shows the high potential of the proposed method in pulsar searching.A method for extracting phase information of X-ray pulsar based on transformer structure was proposed.High precision time of arrival(TOA)estimation is the guarantee of X-ray pulsar navigation.In this part,2D-APM was simplified by removing the autocorrelation part to recover the phase information,and it is thus named 2D-PM.Considering that 2D-PM has temporal relationship in the phase direction,we used the pioneering structure transformer which has been widely used in NLP in recent years to model the temporal relationship and to achieve the purpose of high-precision phase estimation.The simulation results show that the algorithm proposed in this dissertation is better than the traditional cross-correlation algorithm,and has great application potential in X-ray pulsar navigation.