Simultaneous input & state estimation, singular filtering and stability

Abstract

Input estimation is a signal processing technique associated with deconvolution of measured signals after filtering through a known dynamic system. Kitanidis and others extended this to the simultaneous estimation of the input signal and the state of the intervening system. This is normally posed as a special least-squares estimation problem with unbiasedness. The approach has application in signal analysis and in control. Despite the connection to optimal estimation, the standard algorithms are not necessarily stable, leading to a number of recent papers which present sufficient conditions for stability. In this paper we complete these stability results in two ways in the time-invariant case: for the square case, where the number of measurements equals the number of unknown inputs, we establish exactly the location of the algorithm poles; for the non-square case, we show that the best sufficient conditions are also necessary. We then draw on our previous results interpreting these algorithms, when stable, as singular Kalman filters to advocate a direct, guaranteed stable implementation via Kalman filtering. This has the advantage of clarity and flexibility in addition to stability. En route, we decipher the existing algorithms in terms of system inversion and successive singular filtering. The stability results are extended to the time-varying case directly to recover the earlier sufficient conditions for stability via the Riccati difference equation.