Conjugate gradient projection subspace tracking

In this correspondence, we develop a new subspace tracking algorithm called the conjugate gradient projection subspace tracker (CGPST). The algorithm is based on a recently introduced RLS-like subspace cost function, which we recursively minimize using conjugate gradient iterations. Subspace averaging concepts are used to produce an O(r²m) algorithm that updates an r-dimensional subspace of C^m. The algorithm is parallelizable, rapidly convergent, numerically stable, and computationally efficient. Simulation studies test the algorithm´s performance and show it to compare favorably with other subspace trackers