I found the following assertion in a neural networks FAQ:
Radial networks typically have only one hidden layer, but it can be useful to include a linear layer for dimensionality reduction or oblique rotation before the RBF layer
But I could not find any "formal" reference (published works) showing it. I found some papers describing EBFNN's, but they implement full covariance matrices on the RBF units. I could not find anything about this approach with an extra linear layer before the RBF layer. The theory is ok for me, it makes sense and works. What I need is any published work with this information. Is there one?