Real-time simultaneous estimation and decomposition of random signals (Q1389901)

The aim of the paper is to derive a more efficient computational algorithm for on-line real-time simultaneous optimal estimation and multiresolutional decomposition of random signals. The proposed algorithm performs the estimation and decomposition simultaneously, using discrete wavelet transform implemented by an octave-band filter bank. Although the algorithm is developed based on the standard Kalman filtering scheme, the nature of blockwise filtering results in a smoothing-equivalent effect. Two special features are pointed out: First, instead of a two-step approach, the algorithm determines in one step the estimated signal such that the resulting signal possesses the desired decomposition. Second, the entire signal processing is performed on-line, which is real-time in the sense that as every two-point measurements flow in, a set of estimates of the signal flows out in the decomposition form. Numerical simulations are studied on a one-dimensional Brownian walk and confirm the better behaviour of the proposed algorithm.