A Shrinkage Correntropy Based Algorithm Under Impulsive Noise Environment

Maximum correntropy criterion algorithm demonstrates good capacity in impulsive interference environment. Recently, a MCC algorithm uses shrinkage technique, namely the noise free MCC (NFMCC) has been proposed, which reduces the mean square deviation (MSD) significantly. A reweighted l1-norm penalized shrinkage MCC (RL
₁
-SMCC) is derived in detail. The new algorithm considers the norm penalty to use the sparse nature in various systems. Hence, the proposed algorithm provides an excellent behavior for sparse as well as non-Gaussian systems. The proposed cost function is implemented based on the correntropy theory. What's more, the new penalty is to take advantage of the sparse characteristics. Finally, based on the shrinkage method, an alterable convergence step is obtained which enables the algorithm to converge fleetly. To evaluate the RL
₁
-SMCC algorithm, several examples for system identification in non-Gaussian environment are set up which show the validity of the proposed algorithm.