The estimation of evolutionary spectrum by square-root filtering algorithm

This paper deals with the spectrum estimation of a kind of nonstationary stochastic process, the oscillating process. For time-varying model parameters of evolutionary spectrum, which has been supposed to follow sample-wise random walk model, we adopt recursive algorithm to estimate them. The square-root filtering and its fast implementation is proposed in this paper. The algorithm has advantages of better numerical stability and reduced dynamic range over Kalman algorithms. The computational cost is approximately equivalent to the conventional Kalman algorithm and more efficient than other square-root formulations.