A geometric approach for separating post non-linear mixtures

A geometric method for separating PNL mixtures, for the case of 2 sources and 2 sensors, has been presented. The main idea is to find compensating nonlinearities to transform the scatter plot of observations to a parallelogram. It then results in a linear mixture which can be separated by any linear source separation algorithm. An indirect result of the paper is another separability proof of PNL mixtures of bounded sources for 2 sources and 2 sensors.