A cluster of affine projection (AP) algorithms has the characteristics of a fast convergence rate under coloured noise. Furthermore, robust mixed-norm algorithms (RMNA) have been confirmed to possess outstanding steady-state properties. Simultaneously, combined step-size methods have the competence to ameliorate the tracking competence of AP-type algorithms. This paper proposes a novel combined step-size affine projection mixed-norm algorithm with a generalized maximum correntropy criterion for robust filtering named CSS-APRMNGMC. The proposed CSS-APRMNGMC integrates the advantages of mix-norm and generalized maximum correntropy criterion and applies new combined step-size methods to ameliorate tracking ability. The superior performance is verified in identification of system, which contains non-gaussian impulsive interferences.
A new algorithm called Combined Step Sizes Affine Projection Mixed-norm Algorithm(CSS-APRMNA) is proposed to solve the problem of choosing between convergence rate and steady-state misadjustment of Affine Projection Mixed-norm Algorithm(APRMNA). By using a combination of large and small step sizes, the fast convergence rate of large step size and the low steady-state misadjustment of small step size are simultaneously achieved. At the same time, a modified sigmoidal activation function with parameters is used to accommodate different step-size combinations. The mixed-norm of the system error is used as a criterion for variable mixing parameter updating to synthesize the effect of non-Gaussian impulse interference and the sensitivity of the updating. The CSS-APRMNA algorithm shows good performance in simulations of system identification that contain impulse interference.
23 May 2024 in IEEE Transactions on Circuits and Systems II: Express Briefs
In this brief, a new adaptive filtering algorithm called affine projection robust outlier detection algorithm (APRODA) that combines affine projection (AP) method and outlier detection is proposed. The convergence speed of traditional AP-type algorithms is easily affected by impulsive noise. APRODA enhances impulsive noise resistance by removing the outliers generated by impulsive noise. Also, a new step-size adjustment method is proposed. This method greatly reduces the computational complexity of the traditional step-size adjustment method by switching the step-size through the detection of steady state and mutation. Combined with APRODA, switched-APRODA is proposed. Several simulation experiments show that the proposed algorithms are significantly preferable to other AP-type algorithms in terms of convergence speed and tracking ability for system identification.