Tracking moving sources using subspace-based adaptive linear methods

Several works reported in the literature show that the subspace-based linear methods are computationally much more interesting than the eigendecomposition-based techniques and only slightly less accurate from the statistical point of view. They therefore have a clear potential for real time applications. Here we retain the basic ideas behind this class of methods and formulate the subspace tracking problem as a classical adaptive least squares (LS) one. Solving this adaptive LS problem results in subspace tracking algorithms of computational complexity linearly proportional to the sample vector dimension. We suggest a possible implementation for tracking the direction-of-arrival (DOA) of slowly moving sources using the LS approach. The problem of estimating crossing targets is also discussed and we propose an efficient strategy to deal with it.