I want to model and estimate parameters of an aircraft by AI methodology and the input signal has great effect and role in simulation and data collecting.
A common signal which is used for exciting the system dynamics in the case of a conventional aircraft is a frequency sweep (in Matlab, it is known as chirp signal).
Alternatively, I employed a set of random pulses (with random pulse widths) as the input signals to identify aircraft dynamics using NARX neural networks.
You can try the Matlab sample dataset. The most common one is the Iris flower dataset which can be in access through Matlab help documents. You can use it as an input of ANN for the purpose of classification.
"The best input signal" can be understood as numbers - they certainly should be standardized. The way of standardization can influence the accuracy of predictions (or classifications). Another meaning of the input signal is a set of features selected as input. You chose this, which you suspect, they influence the output. A good practice is to check the linear correlation of each feature with each other and with the output. If there are inputs that are strongly correlated, you can exclude one of them. You can also remove one by one, every time checking the error (If lowered the feature stays excluded). The choice can be done also through principal component analysis (concerning the input only) or by any tool called "explainable" e.g. drop-out loss. As the result of these, you will get a ranking of inputs sorted by the strength of their influence on the output.
The standardization methods I described in one of my articles (standardization on mean=0 and std.dev=1 is omitted there, but it is popular) as well as feature importance ranking is described in another one.