Radar Target Detection And Classification Based On The Joint Distribution

This paper is aiming at the radar target detection and classification in strong clutteror noise. In the modern warfare, the stealthy aircrafts/warships, low altitude penetrationweapons, and electronic countermeasures have been widely used. Therefore, the radarshould have the abilities to detect and classify the target in strong clutter or noise. In thisdissertation, we study the joint distribution (including time-frequency distribution andtime-frequency rate distribution) of the radar echoes and the features extraction from thejoint distribution. Then apply the results to detect the weak target in sea clutter and themaneuvering range-spread target. Also they are used to classify the helicopter byestimating the micro-Doppler parameters of the main rotor.The main content of this dissertation is summarized as follows.The first part focuses on the weak target detection in the sea clutter. The derivationand the construction of the sea waves are analyzed, indicating that the sea clutter isnon-stationary. Based on the difference between the sea clutter and the target echo inthe time-frequency domain, we propose an efficient method for detecting slow-movingweak targets based on time-frequency iteration decomposition. This method consists ofthree stages: first, we present a fast signal synthesis method (FSSM) based on theeigenvalue decomposition. The FSSM can synthesize a signal faster and moreaccurately from the Wigner distribution (WD). And then, we present a signal iterationdecomposition method (IDM) from the masked WD (MWD) and the FSSM. By theIDM, the small component of a signal can be obtained, even when it is very close to alarge component in the time-frequency plane. Finally, based on the IDM and twocriterions, the detection method is proposed. The proposed method is evaluated byX-band sea echoes with a weak simulated target or a real target. Results demonstratethat it not only detects the slow-moving weak target but shows its instantaneous state.The second part focuses on the maneuvering range-spread target in strong whiteGaussian noise. Using the mixer output received by the high resolution radar (HRR), wepropose three types of methods to detect a range-spread target.1) Based on thetime-frequency decomposition of the cross S-method (CSM) of two adjacent mixeroutputs, a range-spread target detection method is proposed. This method consists ofthree steps. First, we propose a signal synthesis method (SSM) based on the singular value decomposition. The SSM synthesizes two signals in their normalized forms fromtheir cross Wigner distribution (CWD) and concentrates their energy on two singularvalues. Second, we derive the CSM of two adjacent mixer outputs. Third, wedecompose the CSM of two adjacent mixer outputs by the SSM, thereby obtainingsingular values. The concentration of the singular values is used to detect therange-spread target. The proposed method is evaluated by the raw radar data withoutrange migration correction. Results show that it outperforms the conventional methods.In addition, we prove that the proposed method has the constant false-alarm rate (CFAR)property.2) Based on the matched signal of one mixer output, we propose two rangespread target detector. At the beginning, we derive the matched signal from the mixeroutput directly. The matched signal is a sinusoidal signal and shares the same frequencyas the largest component of the target signal. Then we define the matched ambiguityfunction (MAF) and the modified matched filter (MMF). Based on the concentration ofthe MAF and the MMF at zero Doppler or frequency, we propose two range-spreadtarget detectors: MAF-D and MMF-D. The two detectors use a single mixer output andthus have the ability to detect the target with a high translational velocity and rotationalvelocity. They are evaluated by the recorded radar data. Results show that theyoutperform the conventional detectors and are robust against the target gesture.3) Usingthe frequency rate (FR) function of one mixer output, we propose a range spread targetdetector. From the cubic phase function (CPF), we define a FR function and discuss theFR range of a discrete LFM signal. From the concentration at zero FR of the FRfunction of a mixer output, we derive the range-spread target detector. The detector hasthe ability to detect the target with a high velocity. Finally, experimental results arepresented by the recorded radar data, which show that the proposed detectoroutperforms the detectors using the high resolution range profile (HRRP).The third part focuses on the parameters estimation of the high-order polynomialphase signal. In this part, we propose a high resolution time-frequency raterepresentation (TFRR), which is a potential way to detect a target. The analyticalformula of the TFRR is presented. And the TFRR is shown to have a narrowerfrequency rate (FR) support than the cubic phase function (CPF). Consequently, theTFRR can be used to analyze the signal with close components in the time-frequencyrate (TFR) domain. Due to the bilinear transform, the TFRR suffers the cross term whenthe instantaneous frequency (IF) functions of the components are cross or very close. Tosuppress the cross term, we propose the smoothed TFRR (STFRR) by introducing anFR window to the TFRR. In addition, the application of the STFRR in analyzing a noisy high-order polynomial phase signal is given, which indicates the potential in targetdetection.The fourth part focuses on the helicopter classification in high pulse repetitionfrequency (PRF) radar. After analyzing the micro-Doppler of main rotor, we proposetwo types of micro Doppler parameter estimation methods due to whether the pulseaccumulative time is longer than the time interval between two successive flashes. Thenthe estimated parameters are used to classify the helicopter.1) We propose twohelicopter classification methods when the pulse accumulative time is longer than thetime interval between two successive flashes: The first one is based on the matchedfilter (MF), including the time MF (TMF) and the time-frequency MF (TFMF). Thesecond one derives from time-frequency masks (TFMs). Simulation results demonstratethat both methods have the ability to classify helicopters with the same micro-Dopplerparameters. Also, they are robust against errors of the estimated parameters and thegesture of the helicopter.2) We propose a method to estimate the blade rotational rateand radius when the pulse accumulative time is shorter than the time interval betweentwo successive flashes. Error function is constructed between a sinusoid and themicro-Doppler signal extracted from the time-frequency distribution of the echo.Solving the error function in criterion MMSE, we can obtain the micro-Dopplerparameters, i.e. rotational rate and radius. The validity and accuracy of the proposedmethod are evaluated via both synthetic and experimental data.