A blocked MCC estimator for group sparse system identification

The maximum correntropy criterion (MCC) algorithm and its kernel based variants exhibit robust system identification against impulsive noise. However, MCC based methods have limited performance when processing sparse especially block sparse system. This study proposes a mixed norm constraint in the objective function and derives block sparse MCC (BSMCC) method and extends BSMCC method to an optimal version by choosing the adequate block size. In details, a mixed ℓ1,0-norm constraint is integrated into the cost function of the varied kernel MCC method. Then its iterative optimization is derived via finding the stochastic gradient direction. The optimal block size is selected by estimation error comparisons among 3 types of parallel computing for iterations in a setting interval. Furthermore, the performance analysis such as parameter setting and computational cost of BSMCC method are provided. Finally, the simulations and echo cancellations are conducted to confirm the superior performance of the proposed BSMCC and OBSMCC methods.