Add discrete intermediate variable to Mean block to clarify that the block output is a continuous-time variable

[beutlich]

Note, that Mean.y is marked as discrete output.

Not sure if these blocks rather belong to Modelica.Blocks.Discrete.

[HansOlsson]

I don't understand the reason for these changes.

Having "discrete" on an output is pretty useless in Modelica; as it doesn't propagate to the input (when connecting the block to another block).
Not using the base-class "SISO" causes code duplication (especially since there is no DiscreteSISO) - and there are some more complicated consequences of that in Dymola.

[beutlich]

Well, I started with Mean, which is a discrete block by design, when using the inheritance from DiscreteBlock. Additionally, SystemModeler complained about output y not being made discrete, which is the reason why I did not extend from DiscreteSISO (in Interfaces), but from DiscreteBlock.

Then, I noticed that RectifiedMean, RootMeanSquare, Harmonic all use the Mean block as component by passing the sampling frequency. For that reason I changed the icon from SISO to Discrete for these three blocks, too.

[HansOlsson]

Well, I started with Mean, which is a discrete block by design, when using the inheritance from DiscreteBlock. Additionally, SystemModeler complained about output y not being made discrete, which is the reason why I did not extend from DiscreteSISO (in Interfaces), but from DiscreteBlock.

The reason y cannot be discrete in DiscreteSISO is that the "discrete" definition in Modelica isn't propagated and thus Mean may have a discrete output but not RectifiedMean (as currently implemented; Clocked signals are better in that way).

And you cannot detect it when connecting to the output so, I don't see why making the output "discrete" is important. I understand that there has been some issues related to that in SystemModeler - but that is how Modelica is defined. On the other hand not using inheritance will cause problems in Dymola (even though it should work the same way in Modelica).

However, the base Interfaces.DiscreteBlock is also problematic, and complicates the code for little gain (you have to switch between classes to understand the new class, including realizing that the extra protected outputs aren't used etc).

In general my view is that if we duplicate code instead of using Interfaces-classes there is some problem with the design of the Interface-classes; and we should first fix that.

[beutlich]

OK, I see. Let's consider the two issues here. Mean and the 3 other blocks.

1. Looking at Mean I believe it is cleaner now. I dropped the discrete output y again.
2. Yes, some new base class helps to get things grouped together. Unfortunately, DiscreteSISO already is taken.

[MartinOtter]

An easy way to fix this in a clean way is:

block Mean "Calculate mean over period 1/f"
  extends Modelica.Blocks.Interfaces.SISO;
  parameter Modelica.SIunits.Frequency f(start=50) "Base frequency";
  parameter Real x0=0 "Start value of integrator state";
  parameter Boolean yGreaterOrEqualZero=false
    "=true, if output y is guaranteed to be >= 0 for the exact solution"
    annotation (Evaluate=true, Dialog(tab="Advanced"));
protected 
  parameter Modelica.SIunits.Time t0(fixed=false) "Start time of simulation";
  Real x "Integrator state";
  Real y_last;
initial equation 
  t0 = time;
  x = x0;
  y_last = 0;
equation 
  der(x) = u;
  when sample(t0 + 1/f, 1/f) then
    y_last = if not yGreaterOrEqualZero then f*pre(x) else max(0.0, f*pre(x));
    reinit(x, 0);
  end when;
  y = y_last
end Mean;

The output is then clearly a continuous-time variable.

[beutlich]

Probably with discrete Real y_last;, right?

[MartinOtter]

Probably with discrete Real y_last;, right?

Yes

[beutlich]

Now SystemModeler no longer complains. Looks better and cleaner now indeed. Thanks.

[HansOlsson]

Ok. That seems like a much simpler solution.
Just a tiny change: Change the initial equation to "y_last=0" instead of "y=0".

[HansOlsson]

Ok. I still don't see why it is necessary, but at least it doesn't break anything.

[beutlich]

Two reasons for the change:

1. SystemModeler raises a warning that a variable is not declared discrete. (We could argue: "tool-issue" - so what or try to fix it in a common sense.)
2. Like I initially did, the previous implementation of Mean could be rewritten to extend from DiscreteBlock or DiscreteSISO, which somehow implies theat Mean is a discrete block, i.e., a block with a discrete output variable. Now, with the intermediate discrete variable y_last, it is clarified that output y is a continuous-time variable.