Generalized covariance for non-Gaussian signal processing and GC-MUSIC under Alpha-stable distributed noise

Abstract

Direction of arrival (DOA) estimation is one of the most important techniques applied in many practical engineering applications. Multiple signal classification (MUSIC) has gained increasing attention due to its high resolution in space. Many methods have been studied to handle the noise of Alpha-stable distribution within the framework of non-Gaussian signal processing. Inspired by a state-of-the-art concept, bounded nonlinear covariance (BNC), a more generalized concept, named generalized covariance (GC), is proposed. A series of existing concepts based on the fractional lower-order moment (FLOM), the correntropy, and the BNC are unified in the name of GC. Then, the convergence of GC under Alpha-stable distributed random variables is addressed. Furthermore, four different types of nonlinear functions are introduced for GC and BNC to handle impulsive noise, including sigmoid functions, score functions, FLOM mapping, Gaussian-like functions. These curves with different parameters are also exhibited in detail to illustrate their capabilities to suppress outliers. Also, GC-MUSIC is proposed, and its performances are compared with other 6 MUSIC-like algorithms in the presence of heavy-tailed impulsive noise. Besides, Cramer-Rao bound (CRB) of root mean squared error is also deduced and exhibited. Through Monte-Carlo simulations, the superiority of GC-MUSIC and BNC-MUSIC under Alpha-stable distributed noise is demonstrated.