A robust affine projection algorithm with feedback compensation of the condition number

In this paper we propose a robust affine projection algorithm (APA) which enforces the control of the regularization parameter based on the feedback of the condition number (CN) of the matrix inversion required to implement the AP algorithm. The CN-based regularization helps reducing the EMSE, and may also bound the inversion errors introduced in finite precision computations. The CN is determined via a computationally efficient method, resulting in little additional computational cost over the standard APA. Simulations show that when the input signal is highly colored, the proposed algorithm outperforms the standard isin-APA with fixed regularization, as well as other recently proposed APA variants.