How to tell if c++ S-function is being linearized?

Hello,

I have a non-linear c++ s-function in which I need to perform additional calculations to update various parameters that are not persistent during linear analysis runs to obtain an accurate result. This s-function is being linearized by MATLAB's numerical perturbation methods. In the time domain I do not need to perform these calculations again. Is there a flag or parameter somewhere that can be read by an s-function which shows if the current model is being linearized?

Currently I've hobbled together the s-function reading a variable I'm setting in my workspace for linear analysis runs, but this isn't an ideal solution to maintain portability across different applications.

I am running on r2014b.

Thanks, Seth

EDIT: For now I've decided to look through MATLAB's dbstack and look for the 'linearize' function somewhere in that stack. This does not seem ideal but works for the time being.

[Alain Kuchta]

Can you give more detail about why the behavior of the S-function needs to be different when the block is being linearized as opposed to be executed as part of a simulation?

It is possible to supply a custom linearization for a block, which would remove the need for Simulink to linearize the block via numerical perturbation. For more information refer to the following documentation:

http://www.mathworks.com/help/slcontrol/ug/augment-the-linearization-of-a-block.html

[Seth Hillery]

Alain,

The most clear example for my case is the s-function has several output angles which are 'wrapped' values between (-pi, pi]. If I am linearizing very near to the pi/-pi boundary condition the resulting linear solution is incorrect due to the discrete jump in output value when perturbing the system. If I am linearizing I need to 'unwrap' the output such that the result is continuous in this situation producing the correct linearized result, but in the normal simulation run I need the result to be wrapped.

There is one further example which is most likely due to how my code is structured. During the normal simulation I have certain properties which are updated during the derivative function and then retrieved in the output function without re-calculating these properties, which improved run time by a significant amount. In the normal simulation run this seems to have minimal to no impact on the results, but when linearized this creates some inconsistencies, so I have to update these properties on each output calculation when being linearized. I'm not sure the exactly how MATLAB performs the perturbation on the s-function (with respect to order of operations and internal memory states) so it's hard to tell if this is the best solution but the results are correct when re-calculating these properties.

I am familiar with the exact linearization specification option, but unfortunately the exact linearization for my system is not explicitly known. This case is a relatively large system with >100 inputs and >200 states, with the generic capability of supporting N states depending on the required model fidelity. I have external linearization code which I am comparing the MATLAB linearization to for result verification, but this solution is also performing a numerical perturbation to obtain the linearized system, so even if I specified an exact linearization then I would still need to perform a perturbation with the external code for each linear run about the current operating point. Using MATLAB's built-in linearization capability simplifies the process.

Seth

[Alain Kuchta]

Hi Seth,

Thanks for providing a detailed description. I think using a block parameter to indicate when the block should behave in normal mode vs. linearization mode is the best solution available. I'm not aware of any way to detect that a block is being executed as part of linearization of a model.