ALGORITHM EXTENSION OF CUBIC PHASE FUNCTION FOR ESTIMATING QUADRATIC FM SIGNAL Pu Wang, Jianyu Yang School of Electronic Engineering Univ. of Elec. Sci. and Tech. of China 610054 Chengdu, P. R. China {pwang, jyyang }@uestc.edu.cn Igor Djurovi´ c Electrical Engineering Department University of Montenegro 81000 Podgorica, Montenegro igordj @cg.ac.yu ABSTRACT In this paper, an extended algorithm for parameter estimation of quadratic FM signal is derived by exploring the time di- versity in the cubic phase (CP) function. The performance of the proposed algorithm is analyzed in terms of estimate bias and variance, and compared with other methods. Although the proposed algorithm employs a fourth-order nonlinearity which results in higher threshold SNR, it provides a number of advantages, such as low mean-square error (MSE) for the estimates at high SNR and simply extension for multicompo- nent signals. Extension to cubic FM signal is also discussed. The theoretical analysis is verified by the simulation results. Index TermsParameter estimation, FM signal, statisti- cal signal processing. 1. INTRODUCTION In the signal processing literature, considerable attention has been paid to parameter estimation of the frequency-modulated (FM) signal from noisy observations. The FM signal can be found in a number of applications such as radar, sonar, geo- physics, and biomedicine [1], [2]. This paper focuses on the quadratic FM signal and also discusses the cubic FM signal. The most accurate way for analyzing the quadratic FM signal is the maximum likelihood (ML) estimation. It yields optimal results but requires a three-dimensional maximiza- tion, and thus it is computational exhausting. Moreover, if the objective function is not convex, the maximization is easy to converge to local maxima. To avoid the multidimensional search, the suboptimal approaches such as the high-order am- biguity function (HAF) [1], [3], integrated general ambiguity function (IGAF) [4], and product HAF (PHAF) [5], are pro- posed. Recently, a bilinear transform — the CP function was presented in [2] and [6]. For a quadratic FM signal defined as s(n)= Ae (n) =Ae j(a0+a1n+a2n 2 +a3n 3 ) , (N 1) 2 n (N 1) 2 , (1) where A, φ(n), and {a i } 3 i=0 are the amplitude, phase and phase coefficients respectively, and N is odd, the CP func- tion is presented as CP(n, Ω) = + 0 s(n + τ )s(n τ )e -jΩτ 2 dτ. (2) Substituting s(n) in (2) with (1), the result is s(n + τ )s(n τ )= A 2 e j2[φ(n)+(a2+3a3n)τ 2 ] . (3) From (2) and (3), the CP function will peak at 2(a 2 + 3a 3 n), which is the instantaneous frequency rate (IFR) of (1) [2], [6]. Once the IFR is obtained, the parameters, a 2 and a 3 , can be estimated by selecting two different time positions and solving the resulting equations set. In this paper, we explore the time diversity in the CP function by using two special time positions, which are symmetric with respect to origin, i.e., one is n and another n. Although this extension results in fourth-order nonlinearities, it offers the following advantages: low mean-square error (MSE) at high SNR for estimat- ing a 3 ; simple extension to multicomponent signals. This paper is organized as follows. In Section 2, we present the algorithm for estimating the quadratic FM signal in two steps. Section 3 derives the statistical results for the estimate using the first-order permutation analysis. In Section 4, ex- tension to multicomponent case is considered. The simula- tion results are provided to validate the theoretical results in Section 5. Section 6 focuses on further development. Finally, conclusions are drawn in Section 7. 2. THE PROPOSED ALGORITHM Motivated by the CP function, we further exploit the time di- versity on the basis of the CP function, which forms a fourth- order nonlinear estimator — the Generalized Cubic Phase (GCP) III  1125 1424407281/07/$20.00 ©2007 IEEE ICASSP 2007