First of all, you need to answer this question: In what level of accuracy the model must be? This will help to find the best answer among many alternatives.

Secondly. Regardless of the system type which is a brain (I don't know this) or other physical systems, you have to do a few tests before deciding to choose the best test signal/s. Better to apply different signal values in the range in different frequencies to see if the system behaves like linear or nonlinear. If you must have different signal values in test inputs and they behave differently, it means you are likely confronting with a nonlinear system.

Going back to my first line question:

1. Nonlinear models: For accurate modeling, you can start with the non-linear models available in Matlab for example.

2. Another option: if you have limitations to have specific signal values with specific frequencies, you also may simply apply straight forward signals, do system identification, and build different models for different pairs of amplitude-frequency and then use multiple models accordingly.

3. Linear models: For approximate models, even if the system is somehow nonlinear, you can still try the linear models. If you cannot apply those complicated signals and your system behaves like linear, it will guide you to choose the simplest possible input. For example, a rectangular medium signal in range.

I suggest reading a bit about system identification. A good place to start is

System Identification: A Frequency Domain Approach / Rik Pintelon, Johan Schoukens

Or,

System Identification: Theory for the User / Lennart Ljung

They cover basic topics on system identification, including how to generate good excitation signals.