For acoustic radiation analysis ANSYS Acoustic Guide (R15) suggests use 0,2*lambda of the main frequency but I was told to consider 1 lambda. ANSYS suggestion would surely decrease the size of the whole model.
My research supervisor G. A. Askaryan taught me many years ago that the good scientific result impresses scientific community if you managed to improve physical effect at least twice, and it is better if three times or five times, and it is even better much. Dear Fabio, I answered your question?
Thank you so much Alexander! So, I would like to ask you if you could be a little more specific in your answer. What I understood based on what you say is that it would be better if I use the bigger model to ensure that my results are reliable, isn't it? Thank you again.
When you asked about big model, I remembered at once my grandmother who taught me: "Sasha, be careful with big subjects therefore bypass the tram in front, and a horse behind. That it well kicked you and you grew wiser".
The effect of the domain size can be seen in low range of the spectrum, as long waves don't have the space to develop! I have studied wave propagation in soil in structural problems not in acoustics but the theory is the same if you are using FEM (BEM would work without problem in infinite domain, it is oriented to this kind of problems but linear problems, generally).
The domain boundary can be based on: Absorbing layer, Perfectly Matched Layers or Infinite Element among others. The domain size will depend on the formulation used by the infinite elements (take a look to the theory manual).
The general rule is to model at less the excitacion and the domain part where non-linearities could be expected (very general). Some authors talk about 3*(lambda)max others 1*(lambda)max.
I have reach good results using absorbing boundary layer based on Lysmer-Kuhlemeyer theory using 0.5*(lambda)max (don't forget that is in structural analysis!)
If your model is too big, think of using explicit solver, if avalaible!