I. Introduction
The basic idea of the conventional subspace projection method is that the most significant terms of the spectrum of the data matrix are dominated by the wall clutter. Using singular value decomposition, responses from the wall can be mitigated by excluding some dominant singular vectors [2].