Results of investigations having the purpose to evaluate different spaces suitable for PD signal separation, are presented in this paper. Signals recorded during tests performed on real objects where the number of PD sources were a-priori known, have been projected in different spaces and the same procedure based on OPTICS clustering algorithm, has been used to separate groups of signals due to the different sources. The separation efficiency has been first evaluated in terms of correct separation then, the density and the distance between the different groups have been used to select the most effective separation space. The criteria here considered can constitute a useful tool to evaluate the efficiency of actual and future separation methods.
