 we have developed an algorithm for controlling discrete-time linear dynamical systems using online convex optimization techniques. This algorithm does not require any prior knowledge about the system or its dynamics and can handle both known and unknown cost functions. It also uses a single persistently exciting input output sequence to identify the system's steady-state manifold, allowing us to control the system with noisy output feedback and without requiring full-state measurements. Furthermore, our algorithm is able to achieve sublinear regret even when the measurement noise is added as an extra constant term. Finally, we demonstrate the practicality of our approach through a detailed simulation example in thermal control. This article was authored by Marco Nonhoff and M.A. Muller.