This page is a copy of research/systems_and_control/signalproc/topics/sys_id (Wed, 31 Aug 2022 15:09:08)
Fundamental problems in systems identification
A brief description of the research problems under current focus is as follows.
- Partial least squares (PLS)
- Study of the statistical properties of PLS approach to parameter estimation of linear regressions
- Cramér-Rao bounds under parameter constraints
- Derivation of such bounds and applying them to problems in blind channel identification/equalisation.
- Signal separation by using system identification tools
- Separating dynamic mixtures of two signals (e.g. two speech signals) by means of system identification and spectral analysis techniques.
- Identification of poorly excited systems
- Using subspace-based techniques to prevent ill-conditioning when the input to the system under study is poorly exciting, and application to stereophonic echo cancellation.
- Biased regression
- New biased estimators (based, among other things, on cross validation) with better performance in ill conditioned problems than the commonly used unbiased methods.
- Cramér-Rao bounds for general state-space models
- Algorithms are developped for computing such bounds from discrete-time data of stochastic state-space models given either in continous or discrete time.
- Identification of systems in closed loop
- Instrumental variable algorithms are analysed as an alternative to some bias-compensated least squares estimators.
Selected references:
[[OPUS style=numbered;pubids=29657,29695,29724]]