Iterated Cognition Method for Prior Model Parameters of Bayesian Estimator

Smart use of prior information is one of effective approaches to improve the performance of Bayesian estimator. At the design stage of Bayesian estimator, the prior model parameters must be specified, but these parameters may not be identical with parameters of environment at the applicant stage. The mismatched prior model can result to the performance degradation of Bayesian estimator. In this paper, a general framework of prior model parameters cognition based on the estimator performance is given at first. Base on the framework, for a Bayesian estimator of DC signal in WGN, the estimation performance is analyzed, and an iterated cognition method of prior model parameters is proposed. The computer simulation is used to analyze the sensitivity and robustness of the estimator under the mismatched prior model condition, and the iterated cognition procedure under different conditions. The computer simulation results show that, the feedback from the estimation performance to the prior model parameters is obtained with the cognitive method proposed in this paper, and the prior model can be matched with the current environment model after the repeated interactions between the estimator and environment.