If the start time is not an integer multiple of the sample time for a zero order hold block in simulink, the output until the first sample is zero. When this is passed back to a system with faster samples (or even a scope), the results look very strange!
Why is it not possible for the zero order hold to a) have an 'initial conditions' parameter? b) take its first sample when the simulation starts?
The workaround is to start the simulation from t=0, but it would be nice if this wan't necessary.