An iterative procedure for joint Bayesian spectrum and parameter estimation of harmonic signals

We propose an iterative algorithm for obtaining parameter estimates of harmonic signals in white Gaussian noise, which are then used for Bayesian spectrum estimation. The algorithm yields the current individual MMSE parameter estimates of each sinusoid by considering the parameters of the remaining sinusoids fixed at their current estimated values. This approach requires the evaluation of several one-dimensional integrals as opposed to the single multidimensional integral which results if all the parameters are estimated jointly. The Bayesian spectrum estimation is also simplified to a procedure that requires one-dimensional integration. The relative accuracy of our scheme compared to the ML and MUSIC methods is demonstrated by computer simulations.