Adapting a normalized gradient subspace algorithm to real-valued data model [articol]

Abstract

A new gradient approach to adaptive subspace-based frequency estimation of multiple real valued sine waves is considered in this paper. The new approach proposed here combines the normalized gradient subspace tracking technique based on Oja learning rule - NOOja (for the signal subspace update) with the ESPRIT-like frequency estimation of real-valued sinusoids (for frequency values retrieval). Consequently, a new adaptive subspace-tracking algorithm for frequency estimation is proposed. The method proposed brings a significant reduction in arithmetical complexity at the same level of accuracy. The algorithm is tested in numerical simulations and compared to complex-valued NOja method.