This paper is concerned with off-line reference shaping for closed loop systems with input and state constraints. We propose a method to generate reference signals which improve the tracking performance at the nominal while the constraints are not violated in the presence of uncertainties. As for the robust constraint fulfillment, the reference signal is represented as a combination of impulse and step signals. This combination is not unique and produces a new degree of freedom. This extra-freedom enables us to exploit the information on both impulse and step responses to reduce the conservativeness. Furthermore, the slack variables are introduced appropriately to reduce the number of constraint conditions. The effectiveness of the method is demonstrated by simulation.