scientific article; zbMATH DE number 15832
[For the entire collection see Zbl 0708.00019.] The authors deal with the ``true nature of the persistence of excitation (PE) condition and its implications for robustness of adaptive systems. The standard PE (or PE for LMS) has been considered as: the full rank summed outer product condition on the input/regressor sequence, which implies exponential asymptotic convergence of the parameter estimates to their desired values in the ideal setting. The authors show that this ``definition of PE is inadequate for some algorithms. They emphasize through an example the dangers of analysing the adaptive element separately from the rest of the system, especially when the adaptive terms are embedded in a feedback loop. They show examples where there is no exponential stability to an average equilibrium, despite a PE for LMS excitation sequence. They also show via an example a lack of boundedness of the parameter estimates in the presence of small disturbances. Further, that for specific examples, parameter convergence is generically impossible and the existence, the value assumed, and in some cases the stability of equilibrium points are dependent on the input sequences, sometimes leading to a bursting phenomenon. The authors identify the following issues that should be considered in the reevaluation of the term PE: It is imprecise to say ``this signal is PE without reference to a particular algorithm and a particular setting in which the algorithm will be applied. PE is a local, and not a global concept. PE should be defined with respect to a particular equilibrium set or trajectory, which may itself depend on the input sequence, the stepsize, or other features of the algorithm. The `` equilibrium set may only be an ``average stationary set. The authors finally propose the following ``concept behind PE: PE is a condition on the features of an algorithm and its inputs/regressor sequence which causes a particular set to have a region of attraction about the set and the algorithm should be ``robust to all forms of suitably small perturbation. It has been known for some time that problems can arise in adaptive settings, where the overall system lacks stability, although parts of the system have been shown to possess adequate performance. This is related to the problem of assuming the validity of the separation principle, where it really does not apply. In adaptive settings, the overall system must be considered. This paper shows through simple, effective and interesting examples, how the current conception of the PE condition, one of the more relevant concepts in the adaptive literature, must be reevaluated.
- Persistent excitation in adaptive systems
- Persistence of excitation conditions and the convergence of adaptive schemes
- Persistency of excitation criteria for linear, multivariable, time- varying systems
- Convergence properties of adaptive systems and the definition of exponential stability
- Persistency of excitation in possible unstable continuous-time systems, identification, and indirect adaptive control
- Adaptive systems, lack of persistency of excitation and bursting phenomena
- Persistency of excitation criteria for linear, multivariable, time- varying systems
- Robust adaptive regulation without persistent excitation
- Persistence of excitation conditions and the convergence of adaptive schemes
- scientific article; zbMATH DE number 3982484 (Why is no real title available?)
- Internal persistency of excitation in indirect adaptive control
- Persistent excitation in adaptive systems
- Persistency of excitation in possible unstable continuous-time systems, identification, and indirect adaptive control
- A result on exponential tracking error convergence and persistent excitation
- Comments on "An exponentially stable adaptive friction compensator"
- Convergence properties of adaptive systems and the definition of exponential stability
- On key properties of the Lion's and Kreisselmeier's adaptation algorithms
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3972484)
