Situational awareness of the transient dynamics of power systems is becoming increasingly relevant as these systems undergo major changes due to the massive introduction of power electronics interfaced equipment, growing transit power flows and fluctuating (renewable) generation. These developments render the steady state assumptions used in traditional static state estimation questionable. Thus, the deployment of methods for dynamic state estimation (DSE) is gaining increasing importance for power system control and protection. A key enabler for DSE is the growing availability of phasor measurement units (PMUs). By exploiting the huge potential provided by PMUs, in the present contribution a decentralized mixed algebraic and dynamic state observation approach is introduced, which is suitable for multi machine power systems with unknown inputs and partially known parameters. The effectiveness of the proposed DSE technique is demonstrated in extensive simulations based on the New England IEEE 39 bus system as well as via real world PMU measurement data.