Parametric spectral estimation methods of clutter profile for adaptive radar detection and classification

Eraslan, Berna
Identification of unwanted echoes in a received radar signal is crucial in order to improve the radar detection performance. In the scope of thesis, currently proposed parametric spectrum estimation techniques, such as MUSIC, ESPRIT and Burg, are evaluated in order to estimate moments of clutter components in received radar echo. Since none of these methods has the ability of estimating Doppler spread and adequate accuracy, Stochastic Maximum Likelihood (SML) method is implemented, working with the best performing optimization and line search method. Since SML estimation accuracy is highly initial point dependent and computationally expensive, a novel estimation technique (Turbo) is proposed which works recursively. Proposed TurbomethodoutperformedthemethodssuggestedinliteraturewithitshighDoppler resolution,accuracy and low computational cost. Moreover,Turbo performance is optimized by utilizing Burg estimates for initial point selection. After designing nearly optimal estimator, estimated parameters is used to design the detection filter which maximizestheNormalizedSINRatitsoutputevenwithasmallnumberofsecondary data. Finally, for clutter classification, a problem specific Neural Network architecture is designed. The proposed Neural Network performance is also evaluated with estimates of novel Turbo method.
Citation Formats
B. Eraslan, “Parametric spectral estimation methods of clutter profile for adaptive radar detection and classification,” Thesis (M.S.) -- Graduate School of Natural and Applied Sciences. Electrical and Electronics Engineering., Middle East Technical University, 2019.