Independent Component Analysis using EEGLAB

I am using ICA from EEGLAB toolbox (not sure if I can use MATLAB's RICA() function as well). I am trying to use Infomax or JADER ICA to blindely seperate two sources which transmitting at a carrier froquency of 1GHz and (1GHz+100kHz). I have two recievers(antennads and 2-channel software-defined radio) that are pointing at each transmitter with 20MHz bandwidth. and signals from each transmitter plus the mutual interference. The performance of JADER and Infomax are not good and I am trying to pinpoint the reason. Is there any constraint on the type of source signals that ICA is able to recover? for example, is this true that "the pdf of the sources should be supergausian or sub-gaussian" for ICA to be able to recover them? Does it matter if my signal is not centered at the neither of the carrier frequencies? How can I figure out what is wrong and what method can help me for blind source separation?