Implementation aspects of various higher-order statistics estimators

The purpose of this paper is twofold. First, a modified robust maximum likelihood higher-order statistics estimator is derived based on the assumption that the obtained estimates (moments or cumulants) follow a generalized Gaussian distribution (GGD) of parameter α. By determining the parameter α which corresponds to the lowest bias and variance of estimation the most appropriate GGD is determined as a function of data length. Then the efficiency of various higher-order statistics (HOS) estimators is assessed in terms of VLSI implementation merits such as data throughout delay, cost, and round-off noise performance. The obtained results provide guidelines in choosing the appropriate estimator in HOS applications.