How do I stop AMS from interpolating discrete signals?

[Mr Alfabet]

The problem I'm running into can be explained by the following simple example:

 

I have a SystemC module sending a speed over a signal to a TDF module that integrates that speed to get a position (with sca_ss).

 

The SystemC module will send out a speed of v=1 at t=0,1,2,3, then it will start sending out a speed of v=0 after that (t=4,5,6,.....).

 

Now the TDF module will calculate the correct position for the first few seconds:

t=0 ---> x=0

t=1 ---> x=1

t=2 ---> x=2

t=3 ---> x=3

 

but after that, the TDF module will interpolate the value for the speed between t=3 and t=4, and thus result in x=3.5 for t=4 (where v=0).

 

I realize I can circumvent this by setting very small timesteps, and setting the speed v=1 at t=3.999, but I was wondering if there is a less 'hacky' way of doing this. (Also setting small time steps will drastically slow down my simulation.)

 

I'm trying to compare the SystemC-AMS output to the output from the same model in MATLAB, which does not have this problem, and the difference is way off due to this.

[karsten]

if the input signal has been changed, you can do one small timestep using the tstep parameter of the sca_tdf::sca_ss class. Alternatively you can use the dynamic timesteps of SystemC AMS 2.0 (e.g. trigerring to an event) and perform also one small timestep after a change.

 

best regards

Karsten

[Mr Alfabet]

if the input signal has been changed, you can do one small timestep using the tstep parameter of the sca_tdf::sca_ss class. Alternatively you can use the dynamic timesteps of SystemC AMS 2.0 (e.g. trigerring to an event) and perform also one small timestep after a change.

 

best regards

Karsten

I think I need a small timestep before the change, not after. Also, the changes are not scheduled, so I can't do a small timestep before the change.

[Martin Barnasconi]

Section 2.3.2 in the SystemC AMS user's guide explains the effect you see:

 

Although the TDF model of computation processes the samples at discrete time steps, the equations of these embedded functions will be solved by considering the input samples as continuous-time signals.

 

This means that your input signal is *not* considered as discrete (sample-and-hold) values, but it will interpolate. This means that between t=3 and t=4 the area is 0.5, which counts up to the integrated value of  3.5.

 

So if you really want to integrate discrete (digital) signals, I recommend to write this in plain SystemC using discrete-event semantics.

 

(Note: If you ask me, Matlab/Simulink are not a good reference for comparison, because the execution semantics and models of computations are non-standard.)