Implementation of timed automata: An issue of semantics or modeling?

We examine to what extent implementation of timed automata can be achieved using the standard semantics and appropriate modeling, instead of introducing new semantics. We propose an implementation methodology which allows to transform a timed automaton into a program and to check whether the execution of this program on a given platform satisfies a desired property. This is done by modeling the program and the execution platform, respectively, as an untimed automaton and a collection of timed automata. We also study the problem of property preservation, in particular when moving to a "better" execution platform. We show that some subtleties arise regarding the definition of "better", in particular for digital clocks. The fundamental issue is that faster clocks result in better "sampling" and therefore can introduce more behaviors.