A Study On Related Techniques For Modulation Recognition Of Radar Signals

As for the modulation schemes of radar signals, there are pulse width, pulse repetition interval (PRI) and frequency modulations from the view of interpulse and intentional intrapulse modulation and unintentional intrapulse modulation from the aspect of intrapulse. Determining the characters of both interpulse and intrapulse modulations will give valuable insights into the function and performance of the radar used, and will be benefit to the radar emitter recognition. In this thesis,several key operations and techniques in the process of modulation recognition such as feature extraction,parameter estimation,recognition algorithm and performance evaluation, were investigated in detail to provide the supporting information for the successive operations in the course of radar signals processing.The main contributions of the thesis are summaries as follows:Firstly,two kinds of time of arrival(TOA) estimators for sinusoid signal were proposed based on reversed correlation method and Haar wavelets transformation respectively. Coarse estimation of the start position and end position by correlation detection was performed at first and then the received signal was transformed to baseband. For the reversed correlation based TOA estimator, the start position of the signal can be effectively extracted by locating inflection point of accumulation of the reversed baseband signal with the maximum output signal to noise ratio (SNR). However,for Haar wavelet transformation based TOA estimator, the start position of the signal can be effectively extracted by locating peak magnitude of the wavelet transformation of the baseband signal. In order to evaluate the performance of the two estimators, the Cramer-Rao Lower Bound (CRLB) of TOA estimation for discrete rectangular complex sinusoid signal is derived in the absence of a prior knowledge for the signal parameters such as amplitude, phase and frequency.Secondly, two specific radar emitter recognition methods based on the maximum-likelihood (ML) criterion by means of Karhunen-Loeve (K-L) series expansion and signal decimation were presented respectively. The received signals were filtered to reduce the noise out of the bandwidth and to enhance SNR at first. Then filtered signals with additive band-limited white noises were transformed to baseband. The principal components of them were extracted by K-L series expansion and the independent sample sequences were obtained by signal decimation respectively. Specific radar emitter recognitions were performed by the equivalent decision variable inferred from the likelihood function using the principal components and the independent components respectively.Thirdly, a novel recognizer that automatically reports the complex pulse repetition interval (PRI) modulation type of a radar pulse signal was developed. This method utilized the features based on the properties of PRI sequence for each PRI modulation type both in time and frequency domain. The features are defined as zeros-crossing density, harmonic amplitude ratio and sign properties of difference of the PRI sequence, which can be easily extracted from PRI sequences with small samples size.Fourthly,a reliability test statistic used to evaluate the blind processed results of normal signals (NS)and linear frequency modulated (LFM)signals was derived based on the Chebyshev's Inequality. The correlation between the received signal and the referenced signal constructed by parameter estimation in accordance with the certain modulation identification result are derived. According the differences between the numerical characteristics of the statistic in the identification of right and wrong,the reliability test performed depending on Chebyshev inequality.Fifthly, another reliability statistic to evaluate the blind processed results of normal signals and linear frequency modulated signals was derived based on Neyman-Pearson (NP) rules. The correlation between the received signal and the referenced signal constructed by parameter estimation in accordance with the certain modulation identification result are derived. According the differences between the probability distribution parameters in the identification of right and wrong,the blind processing results reliability was verified depending on NP rules.At last, a reliability statistic to verify the blind processed results for binary phase shift keying (BPSK), NS and LFM signals was presented depending on the property of the linear regression of correlation curve which was obtained by the received signals and the reference signals constructed in accordance with the certain modulation identification results. According to the different characteristics of the modulus values of the curve between different hypotheses, the problem was boiled down to lack of fit test for linear regression.