Transient Signal Estimation With Higher-order-statistics

Two methods or the estimation of the parameters of exponentially d roped sinusoids are introduced based on third order statistics of the observation signal. These methods may be se n as extensions of the minimum norm / principal eigenvecto s method (Kumaresan-Tufts) to higher order statistics dom ins. The strong points and limitations of the new methods are discussed, as well as sufficient conditions for existent of their solutions. The utilization of these methods in t e case of finite length signals in the I presence of addltiv Gaussian noise (white or colored) is addressed. Monte- arlo simulations demonstrate the effectiveness of the n w methods when the additive noise is colored Gaussian w"th unknown autocorrelation sequence. The case of an ens mble of data records is studied when the exponentially d roped sinusoids are assumed to have random phase.