Reset-free iterative identification based on the finite-dimensional signal subspace

This paper proposes a new approach for the iterative identification of continuous-time systems, which is based on the projection type ILC (iterative learning control) concepts. Unlike any other ILC methods, this paper gives a framework to perform ILC without resetting the initial condition at each iteration, which can be achieved by introducing the dynamics into the system representation in the finite-dimensional signal subspace. Therefore, it is not necessary to wait for the equilibrium state patiently or reset the system forcibly. Furthermore, a class of gain decreasing filters are introduced, which plays a crucial role in effective parameter convergence in the presence of heavy noise. Combination of these results gives us the estimates which converge to the true system parameters against measurement noise. A numerical example is given to demonstrate its effectiveness.