Censored regression system identification based on the least mean M-estimate algorithm

Abstract

Classical adaptive algorithms have good convergence performance in linear regression system identification. However, they will face performance degradation while dealing with censored data since only incomplete information can be obtained. In this paper, the least mean M-estimate algorithm for censored regression (CR-LMM) is proposed for the robust parameter estimation. To compensate for the bias caused by censored observation, the probit regression model is employed to derive the estimated error for constructing the M-estimate cost function. The cost function can expel the adverse impact of the impulsive noise, and it is solved by the unconstrained optimization method. Computer simulations in the impulsive environment are carried out to demonstrate that the proposed CR-LMM algorithm exhibits better convergence performance than the existing algorithms in censored regression system identification scenarios.