1: \begin{abstract}

2: The method to design exponentially stable adaptive observers is proposed for linear time-invariant not necessarily observable, but detectable systems with unknown parameters. Unlike existing adaptive solutions, the system state-space matrices {\it{\textbf{A, B}}} are not restricted to be represented in the observer canonical form to implement the observer. The original system description is used instead, and, consequently, the original state vector is obtained. The class of systems for which the method is applicable is identified via three assumptions related to: \linebreak ({\it{\textbf{i}}}) the identifiability of the unique parameters of {\it{\textbf{A}}} and {\it{\textbf{B}}} from the characteristic polynomials of output transfer function and the fact that we know mappings ({\it{\textbf{ii}}}) from such polynomials parameters to unique parameters of {\it{\textbf{A}}} and {\it{\textbf{B}}} and ({\it{\textbf{iii}}}) from them to Luenberger correction gain. In case they are met and the regressor is finitely exciting, the proposed adaptive observer, which is based on the known GPEBO and DREM procedures, ensures exponential convergence of both system parameters and states estimates to their true values. Detailed analysis for stability and convergence has been provided along with simulation results to validate the developed theory.

3: \end{abstract}

4: