On parameter convergence in adaptive control (Q795780)

The intuition suggests that if a parameter vector converges, the plant loop is asymptotically time-invariant. If the input r has spectral lines at frequencies \(\nu_ 1,...,\nu_ N,\) we expect \(y_ p\) will also. Since \(y_ p\to y_ M,\) the asymptotic closed loop plant transfer function matches the model transfer function at the N distinct frequencies. If N is large, the asymptotic closed loop transfer function must be precisely the model transfer function; that is the parameter error vector converges to zero. It is the purpose of this paper to make this intuitive argument formal. The result is known in case of the uniform asymptotic and (equivalently) the exponential stability of the schemes under a sufficient richness of a certain vector of signals w(t) appearing inside the time-varying feedback loop. However, at present it is impossible to say whether a given reference input results in a sufficient richness of w(t). In this paper, it is shown that when the reference input has as many spectral lines as there are unknown parameters, then the adaptive control error and parameter error vector converge to zero exponentially. The adaptive systems of \textit{K. S. Narenda}, \textit{Y. Lin} and \textit{L. S. Valavani} [IEEE Trans. Autom. Control AC-25, 440-448 (1980; Zbl 0467.93049)] are considered. At first, the main result is proved for the relative degree 1 case. Next, the extensions to the higher relative degree cases are discussed. The proof is as follows. At first, the theorem that the sufficient richness of w(t) is equivalent to that of \(w_ M(t)\) is proved. Next, the theorem that if r(t) has special lines at \(\nu_ 1,\nu_ 2,...,\nu_{2n}, w_ M\) is sufficiently rich is proved.