Robust spectral estimation based on ARMA model excited by a t-distribution process

This paper will focus on proposing a new objective function of autoregressive moving average (ARMA)-model-based spectral estimation. The objective function is derived by assuming that the obtained residual signal is identically and independently distributed. The probability density function is assumed to be t-distribution. Small portions of residual signals with large amplitude is given a small weighting factor and large portions of residual signals with small amplitude is assiggned a large weighting factor. By doing so, the effect of large amplitude error signal is suppressed and the adaptation step is less affected. The simulation results for image enhancement show that when the input is impulsive noise contaminated images, the obtained processed image by using t-distribution with small a degrees of freedom is much better than that when large a is applied.