Connections between the least-squares and the subspace approaches to blind channel estimation

In this correspondence, we study the connections between the least-squares and the subspace approaches to blind channel estimation. By examining the properties and connections of the so-called multichannel filtering and data selection transforms, we establish a relationship between the identification equations used in the two approaches. Next, it is shown that the least-squares and subspace estimators are identical for the case when there are two subchannels. In general, the two algorithms are different in their utilization of the noise subspace.