Improved strategy for adaptive rank estimation with sperical subspace trackers

We present an improved adaptive rank detection algorithm for on-line estimation and tracking of the signal subspace dimension in applications of spherical subspace trackers. The proposed algorithm uses different adaptive thresholds for the rank increase (up) and decrease (down) tests as well as a special set of fast tracking eigenvalue estimates in the rank decrease test, which can be obtained at little extra cost. It is based on an original investigation of the detection per-formance for the up and down tests that takes into account the exponential nature of the eigenvalue update in spherical subspace trackers. Through computer experiments in multi-user detection, it is shown that with the proposed algorithm, the time required to detect a rank decrease is significantly less than with existing methods