Sensitivity analysis of state space methods in spectrum estimation

In this paper, we examine the issue of parameter sensitivity in spectrum estimation. In particular, state space models are considered and robust coordinate systems for spectrum estimation are identified. For the sinusoid problem it is shown that the ideal parameter set involves estimating a unitary matrix and simple procedures to estimate such a matrix are identified. For the damped sinusoid and the general ARMA spectrum estimation problem it is shown that the balanced coordinates are robust and reliable.