Abstract
Summary form only given. Although there exist many methods for impulse response estimation, most of them are based on the assumption that the measurement noise n(k) is white Gaussian. However, there are cases in which n(k) is not Gaussian but a mixture of a dominant Gaussian noise and an impulsive noise. For the same signal-to-noise ratio (SNR), the performance of most existing methods when data are contaminated by this mixture of noises is worse than that when data are contaminated by a Gaussian noise. U. K. Bhargava and R. L. Kashyap (1988) proposed an approach based on Huber's function that is robust against this mixture of noises. By simulation, their method has been shown to be robust for 0 ≤ λ ≤ 0.22. This mixture noise is modeled by a Bernoulli-Gaussian model. The obtained maximum-likelihood (ML) estimator is nonlinear in nature and implemented by an iterative block component method (BCM). Results of computer simulations are given for SNRs ranging from 8.2 dB to 34.2 dB using this BCM to indicate the performance of the proposed ML estimator.