Abstract
Given a set of output measurements x{k) of a linear time-invariant nonminimum-phase ARM A (p, q) system H(z), the well-known exhaustive search method (ESM) and Giannakis and Mendel's spectrally equivalent minimum-phase (MP) HMP(z)-all-pass (AP) Hap (z) decomposition based methods include two steps for estimating the coefficients of H(z) = HMP(z). HMP(z). In the first step, Hmp(z) is obtained by a correlation-based spectral estimation method. In the second step, the ESM determines H(z) from HMP (z) by cumulant matching, and the MP-AP decomposition-based methods identify HAP (z) from the cumulants of the preprocessed data A(k), which is the output of the inverse filter 1/HMp (z) with input x(k). In this correspondence, we propose a new method only for the second step, which determines the optimum HMP (z) from HMl, (z) by maximizing the absolute value of a single Mth-order (M ≥ 3) cumulant of the output signal of an all-pass filter H'AP (z) with input d(k). The optimum HMP(z) = l/HMP(z). Some simulation results are provided to support the proposed method. © 1993 IEEE