How to randomize simulation seed in simulink?

I have a simulink simulation which contains a custom matlab function to generate random numbers based on the pearsrnd.

function y = skewedPRNG(mu, var, skew)
    kurt = skew^2+3;
    r = pearsrnd(mu,var,skew,kurt,1,1);
y = r;

However, since the simulink seed is fixed, I'm getting the same random signal. How can I change the simulation's seed randomly so that I will get a different signal every time I run the simulation?

Note, I need to generate the random numbers according to a skewed distribution, so using the available random number blocks is not helpful.

Edit: the simulation is Desktop Real-Time. I neglected to mention, sorry.

Edit 2: Desktop Real-Time Kernel Mode, sorry again.

Answer 1

[Paul]

Hi Amr,

When I first saw the question, I was quite surprised. To my uderstanding, the call to pearsrnd in the Matlab Function block should be executed in Matlab, and therefore would need to be declared extrinsic. It isn't, but the simulation still compiles and runs to yield, as you noted, the same output every time. I can't explain this result and, frankly, I'm surprised that there isn't an error when the simulation is being built. I'm very curious as to what is happening. I'm even more curious that rng('shuffle') works, even if the simualation runs slowly.

If you ever do get an explanation as to how all of that works, please post back.

Having said that, if you do declare pearsrnd as extrinsic, like this

function y = skewedPRNG(mu, var, skew)
coder.extrinsic('pearsrnd');
y = 0; % specifies size and type of y
kurt = skew^2 + 3;
r = pearsrnd(mu,var,skew,kurt,1,1);
y = r;
end

then the random numbers generated from pearsrnd are derived from the base Matlab random number generator and the results will change from one run to the next. If you want repeatability, you can control the seed in the usual way, typically with a call to rng, before the simulation executes

rng(100)

If you go down this path, you'd have to set the base Matlab seed prior to each simulation execution, either manually if using the play button or the sim command from the command line, or as a command in the script or function if using the sim command in one of those, or in a callback function.

Of course, this seed will affect everything in the simulation that uses the base Matlab random numbers.

Also, after the simulation runs the Matlab random number generator will be in a different state than when the simulation started. If that's a problem you may want to consider saving the random number generator state before the simulation executes and then resetting it after the simulation ends.

As far as I know, Stats toolbox functions always use the default random number stream, so we can't create a new randstream object that would be used just for the simulation, and thereby not "corrupt" the default stream.

I'm afraid I don't know anything about "Desktop Real-Time" and so don't know how that part of the Question might impact the solution.

[Amr Okasha]

Hi Paul,

Thank you for your answer. I tested the solution and indeed it works for regular simulations, and even for Desktop Real-time Connected I/O simulations. I even combined it with the Model callbacks StartFcn found in Model Properties by adding rng('shuffle') there. Together with the extrinsic function, it generates random results without the need to call rng in matlab itself

Unfortunately, I need to run the simulation in Kernel Mode, which generates stand-alone code. I'm afraid coder.extrinsic does not compile there as it gives the following error,

"The extrinsic function 'pearsrnd' is not available for standalone code generation. It must be eliminated for stand-alone code to be generated. It could not be eliminated because its outputs appear to influence the calling function. Fix this error by not using 'pearsrnd' or by ensuring that its outputs are unused."

[Stefanie Schwarz]

Hi Paul,

As long as a MATLAB function is capable of generating C code, it can be used in a MATLAB Function block without the use of coder.extrinsic. That's the case for pearsrnd (see Extended Capabilities section at the bottom).

Running a simulation in Desktop Real-Time Kernel Mode means that we need to generate C code from the entire model to compile it and then execute it in a real-time kernel installed on the host computer. This workflow unfortunately won't allow extrinsic functions. But your comment will still be very useful for "regular" Simulink users! Thanks for that!

[Paul]

Hi Stefanie,

Thanks for pointing out that Extended Capabilities section.

So for the original case, where coder.extrinsic is NOT used, what is happening? What random stream is pearsrnd using in that case and how does the user control the seeding of that stream?

I'm asking about "regular" Simulink use.

Answer 2

[Bhavana Ravirala]

To generate different random signal for every simulation, you can execute the following command:

rng('shuffle');

This results in a different sequence of random numbers after each call to rng. So, you can find the following function useful for your workflow.

function y = skewedPRNG(mu, var, skew) 
     rng('shuffle'); 
     kurt = skew^2+3; 
     r = pearsrnd(mu,var,skew,kurt,1,1); 
y = r; 

Please refer to the below documentations for more information.

• Control random number generator - MATLAB rng (mathworks.com)
• Generate Random Numbers That Are Different - MATLAB & Simulink (mathworks.com)

Hope this helps!

[Amr Okasha]

Adding

rng('shuffle'); 

slows down the simulation immensly unfortunately.

I'm hoping to use this for a desktop real-time application. It does not build in a desktop real-time environment.

I'm looking for something that will randomize the rng at the start of every run for the entire simulation, not just a block or function.