A robust block adaptive filter for colored additive noise

In adaptive filter applications with colored input signals, fast convergence adaptive algorithms, such as least square (LS) type adaptation algorithms, are useful. When we apply an LS type adaptation algorithm to the adaptive noise cancellation, the estimated weights of the adaptive filter cannot converge to desired weights with sufficient accuracy because the reference signal is corrupted by additive noise. In order to reduce the degradation of the estimated weights, some robust adaptive filters have been proposed. However, the improvement of the estimated weights depends on statistics of the input and reference signals. In this report, we propose a new robust block adaptive filter which improves the accuracy of the estimated weights regardless of statistics of the input and reference signals. The improvement is done by averaging the adaptive weight vectors