Blind identification of PAM-MIMO systems based on the distribution of output differences

In this paper, we present, a new method for the Blind Identification and Source Separation of a convolutive Multi-Input Multi-Output (MIMO) system driven by multi-level inputs. The method takes advantage of the special properties of the observation differences distribution. We show that a specific set of differences can be used to reconstruct the mixing operator. Exhaustive search of the observation symbols yields sources separation. The method can be coupled with a simple clustering technique in order to treat noisy cases. Results on both noiseless and noisy scenarios show the efficiency of our approach.