A Fast Algorithm for Unitary ESPRIT
Peng Yingning · Systems engineering and electronics · 2003
Although unitary ESPRIT improves the parameter estimation of standard ESPRIT algorithm with a reduced computational burden, it requires the singular value decomposition (SVD) of the real-valued covariance matrix. A method to estimate signal and noise subspaces based on rational approximation of the covariance matrix without costly SVD is presented. The computational complexity is also reduced significantly by unitary transformation from the sampled data matrix to the real-valued matrix when centro-symmetric array configurations are used. Furthermore, the forward-backward averaging leads to an improved performance, especially for correlated signals. Simulations confirm that the new algorithm has a performance comparable to that of unitary ESPRIT and outperforms both standard and rationally-approximated ESPRIT algorithms.