Adaptive Noise Subspace Estimation Algorithm with an Optimal Diagonal-Matrix Step-Size

Abstract

In this paper, we propose a new optimal diagonal-matrix step-size for the fast data projection method (FDPM) algorithm. The proposed step-sizes control the decoupled subspace vectors individually as compared to conventional methods where all the subspace vectors are multiplied by the same step-size value (scalar case). Simulation results show that FDPM with this optimal diagonal-matrix step-size outperforms the original algorithm as it offers faster convergence rate, smaller steady state error and smaller orthogonality error simultaneously. The proposed method can easily be applied to other subspace algorithms as well.