Extending the subspace method for blind identification

The Moulines subspace procedure (Moulines et al., 1995), for blind identification leads to an inconsistent estimation when the order of the channel is unknown or when it is close to being unidentifiable. The non-robustness of the subspace approach to order over-estimation is considered a major limitation to its use in practical situations. This is usually the strongest argument in favour of linear prediction-based methods for channel identification. In this contribution, we propose a simple extension to the subspace approach which is robust to both over-estimation and lack of channel disparity.