Efficient multiscale stochastic realization

Few fast statistical signal processing algorithms exist for large problems involving non-stationary processes and irregular measurements. A previously introduced class of multiscale autoregressive models indexed by trees admits signal processing algorithms which can efficiently deal with problems of this type. In this paper we provide a novel and efficient algorithm for translating any second-order prior model to a multiscale autoregressive prior model so that these efficient signal processing algorithms may be applied