Abstract— In this paper a comprehensive algorithm is presented to alleviate the undesired simultaneous effects of target maneuvering, observed glint noise distribution, and colored noise spectrum using online colored glint noise parameter estimation. The simulation results illustrate a significant reduction in the root mean square error (RMSE) produced by the proposed algorithm compared to the algorithms that do not compensate all the above effects simultaneously. Keywords—Glint noise, IMM, Kalman Filter, Kinematics, Target Tracking. I. INTRODUCTION N 1995, Daeipour and Bar-Shalom utilized the IMM algorithm to implement the glint noise model in nonmaneuvering target tracking [1]. They applied two extended Kalman filters, one matched to the dynamic system with Gaussian measurement noise, and the other matched to the same dynamic system but with a high variance Laplacian noise. Later in 1998, E. Daeipour, and et al. in [2], and K. Heydari and et al. in [3]-[4], almost concurrently, extended the algorithm used in [1] to maneuvering targets. They applied a layered IMM (LIMM) algorithm to implement the target maneuvering model as well as the glint noise model. Although they pursued the same goal, they were different in methodology. The algorithm in [2] was based upon the extension of [1] with the dynamic system state equations in Cartesian and observation equations in spherical coordinates. But, the one in [3]-[4] was developed with the dynamic system state equations and observation equations both in spherical coordinates. In both [2] and [3]-[4] the filters were split into two parts, one matched to the Gaussian component and the other matched to the Laplacian component of the glint noise. However, in the former the extended Kalman filters (EKF) were exploited to deal with the Gaussian and Laplacian noise components, but in the latter, linear Kalman filter was M. A. Masnadi-Shirazi is with the Dept of Electrical Engineering, School of Engineering, Shiraz University, Shiraz, Iran (Corresponding author phone: +98-917-1135378; e-mail: masnadi@shirazu.ac.ir). S. A. Banani is a graduate student in the Dept of Electrical Engineering, School of Engineering, Shiraz University, Shiraz, Iran (e-mail: alireza_banani@yahoo.com). used for the Gaussian noise and Masrelize filter with efficient approximate score function ([5] and [6]) was used for the Laplacian noise to filter the components of the glint noise. In all the aforementioned methods, the observation noise spectrum was assumed to be white. However, in high frequency measurement radar systems the successive samples of the measurement noise are not uncorrelated, and consequently the observed noise spectrum is not white. In 1996 Wu and Chang presented the subject of maneuvering target tracking with observed colored noise and unknown parameters [7]. The drawback of this approach was the assumption of Gaussian noise distribution rather than glint distribution. Earlier, W. R Wu in [8] had reported a maximum likelihood approach to identify the glint noise parameters from recorded data. However, he assumed the spectrum of the glint noise to be white rather than colored, and also ignored maneuvering effects of the target in noise parameter estimation. In this paper a comprehensive algorithm is developed to recursively provide an online estimate of the colored glint noise parameters and cope with simultaneous effects of the following main four factors that may degrade the optimality of the Kalman filter in target tracking. The four degrading factors are: maneuvering of the target, glint and colored characteristics of the observation noise and lack of knowledge about the noise parameters. II. PROBLEM FORMULATION A. Target Model and System Equations In radar target tracking the dynamic state equations are expressed in rectangular coordinate, while the observation equations are measured in spherical coordinate. Converting either of them to the other one will result in nonlinear equations. One approach is to convert the dynamic equations from rectangular to spherical coordinate, and use approximate linearized spherical model which encounters simultaneous solution of three complicated nonlinear differential equations [6, 8]. For instance the range channel is used to be represented here. Then the corresponding approximate spherical representation for the second-order dynamic model in nonmaneuvering mode is: On Maneuvering Target Tracking with Online Observed Colored Glint Noise Parameter Estimation M. A. Masnadi-Shirazi, and S. A. Banani I World Academy of Science, Engineering and Technology International Journal of Electronics and Communication Engineering Vol:1, No:2, 2007 164 International Scholarly and Scientific Research & Innovation 1(2) 2007 scholar.waset.org/1307-6892/15993 International Science Index, Electronics and Communication Engineering Vol:1, No:2, 2007 waset.org/Publication/15993