Channel-based blind multichannel identification without order overesttimation problems.

We investigate a new approach for blind identification of multiple FIR channels that appears to be robust to the channel length overestimation. The linear prediction approach proposed in [1] constitutes a robust approach since it provides a consistent channel estimate if the channel order is overestimated. However, we focus here on methods that are parameterized by the channel directly such as the (signal) subspace tting technique. To make the optimization criterion in these approaches well de ned, a constraint on the channel coecients has to be added. Typically, the unit norm constraint is used. It is the use of this constraint that leads to order overestimation problems. In our approach we replace this constraint by a unit norm for only the rst vector coecient of the vector channel. Our simulations demonstrate that the channel estimate obtained in this way is robust to order overestimation. Furthermore, if the exact quantities are used in the optimization criterion, the proposed channel estimate is the correct channel (up to the usual scaling factor) even if the order is overestimated. Hence, our channel estimate is consistent even with order overestimation.