Browsing by Author "Poyraz, Salih"

Now showing 1 - 5 of 5

Citation - WoS: 1
Citation - Scopus: 7
An Identification Method Using ESPRIT and PCA for Radar Targets in Resonance Region
(IEEE, 2013) Salih Poyraz; Mustafa Secmen; Poyraz, Salih; Secmen, Mustafa
This paper aims at the development of a fast and sufficiently aspect- independent identification method with the processing of wideband scattered signals from radar targets. In the given method estimation of signal parameters by rotational invariance techniques (ESPRIT) is mainly used for the purpose of classification of targets in resonance region. Multiple reference vectors acquired by using this technique are reduced to one vector for each target with principal component analysis (PCA) and this sole vector is used as feature vector of the given target. The identification phase is employed according to the highest correlation coefficients between ESPRIT vectors of test signals and feature vectors. The proposed method is applied to small-scale airplane targets modeled with wires and about 88 percent identification rate is obtained even for SNR level of -5 dB. The results for other similar dimension reduction techniques are also demonstrated and PCA is found to give better accuracy rates.
Location Independent Radar Target Classification Method with Strategy Specific Late-Time Intervals
(IEEE, 2014) Salih Poyraz; Mustafa Secmen; Poyraz, Salih; Secmen, Mustafa; A Kurowska; J Misiurewicz
This work expounds a target classification method in the resonance scattering region having reduced target's distance aspect angle and noise dependencies. In the given method crucial optimum late-time intervals of the scattered signals are determined by using time-frequency representations. The time instants belonging to maximum and mean power values in time-frequency distributions are used which are independent from targets' positions. Then the feature vectors are formed for each target by using the given time-frequency distributions over these selected late-time regions at several different reference aspects and they are eventually used for the classification in test stage. In this study two different strategies having target-specific and signal-specific late-time intervals are designed. The simulations are carried out with lossless dielectric spheres being challenging targets in terms of scattering mechanism. The performances of designed strategies as well as other similar methods in the literature are compared for different popular time-frequency representations. It is found the strategy with target-specific late-time intervals combined with the Wigner-Ville distribution have better results such that it gives more than 70 percent accuracy for the noisy signals of SNR = 5 dB.
Citation - Scopus: 2
The Comparison of JADE Based DOA Estimation Methods for Unknown Noncoherent Source Groups Containing Coherent Signals with Frequency Matching
(IEEE, 2014) Ahmad Aminu; Mustafa Secmen; Salih Poyraz; Aminu, Ahmad; Secmen, Mustafa; Poyraz, Salih; A Bikonis; K Nyka
In this study all parameters (including the frequency of each signal group) of direction-of-arrival (DOA) problem are aimed to be extracted in the presence of unknown noncoherent source groups which are consisting of coherent signals. The antenna elements used in the fading analysis and application are isotropic linear and uniformly distributed and all parameters of the complete signal are assumed to be unknown except the number of coherent signal in each noncoherent group. To obtain the desired parameters (number of noncoherent groups arrival angles fading coefficients frequencies) it is proposed a four-step approach. First the number of noncoherent signal groups is determined by the minimum description length (MDL). Then effective steering vectors are estimated using the joint approximate diagonalization of eigenmatrices (JADE) algorithm. In the third step by using these steering vectors some popular high-resolution DOA methods such as the modified forward backward linear prediction (MFBLP) estimation of signal parameters via rotational invariance techniques (ESPRIT) and root multiple signal classification (MUSIC) algorithms are realized to calculate the angle and fading coefficient of each coherent signal. Finally a frequency matching is realized to assign the possible frequency to each group. The proposed approach is summarized and simulation results comparing the performance of high-resolution methods are demonstrated. According to the results MFBLP is found to have better accuracy performances.
The Performance Comparison of Wigner-Based Radar Target Classification Methods for Resonance Region Targets
(IEEE, 2014) Salih Poyraz; Mustafa Secmen; Poyraz, Salih; Secmen, Mustafa
This study includes the performance comparison of the target classification methods based on Wigner distribution in the resonance scattering region where the dimensions of the target close to wavelengths. In the suggested method important optimum late-time intervals of the scattered signals are theoretically determined by using the Wigner energy maps of the signals. Then targets' feature vectors are determined for each target by using the Wigner distributions over the selected late-time region at several different reference aspects and Principal Component Analysis (PCA). These vectors are used for classification in test stage. When it is compared with the methods in literature the suggested method has important features such as not being required to find pole numbers and values with high sensitivity and being independent from aspect angles. In this study tests are realized with lossless dielectric spheres which are geometrically simple but complex targets in terms of scattering mechanism and the method containing target-specific optimum late-time intervals is especially found as more successful.
Yüksek çözünürlük ve zaman-frekans dağılım tekniklerini kullanan bir elektromanyetik hedef sınıflandırma yönteminin tasarımı ve gerçekleştirilmesi
(2015) Poyraz, Salih; Seçmen, Mustafa
In this thesis, it is aimed to develop a fast and sufficient target classification method, which is independent from aspect angle and polarization. By processing the scattered radar signals, a method having low decision time and high accuracy rates even at high noise levels, are expected. In thesis, it is assumed that a certain number of targets to be classified are in resonance scattering region. Besides, it is assumed that predetermined and moderate number of reference signals corresponding to different angle/polarization cases is available for each target. In the first part of the suggested method, by using high resolution techniques such as ESPRIT, Min-Norm, MUSIC, some feature vectors are obtained in real frequency domain. However, the proposed method requires the vectors as the number of reference signals for each target and here the problem is that a serious increament of decision time. For this reason, in the second stage of suggested method, it is aimed to decrease the number of feature vectors to one by applying dimension reduction technique or high resolution techniques for multiple signals to these vectors. In this way, decision speed can be faster and system memory may be used more sufficiently. In the test stage, the scattered signal concerning to any angle/polarization case of a target is processed with the same high resolution technique and a test vector is formed. Finally, classification beetween test vector and feature vectors is done by the help of highest correlation coefficients. Thus, a radar target classification method independent from angle and polarization as possible under high noise situations is constituted.In the suggested project, after some comparisons and trials signal processing methods are chosen for best effort as ESPRIT and principal component analysis (PCA). In second stage of study, for the given method, essential optimum late-time intervals of the scattered signals are determined by using time-frequency representations. The time instants, independent from targets positions, are applied which are belong to maximum and mean power values in time-frequency distributions. Then, the feature vectors are formed for each target by using the given time-frequency distributions over these selected late-time regions at several different reference aspects and they are eventually used for the classification in test stage. In this thesis, two different strategies are created as 'target-specific' and 'signal-specific' late-time intervals. The lossless dielectric spheres are used for the simulations. The performances of designed strategies as well as other similar methods in the literature are compared for various well-known time-frequency representations.