Prediction method of output response probability for complicated acoustic systems using information criterion

In general, the evaluation of complicated stochastic systems is difficult only from the viewpoint of conventional structural approach based on their physical internal mechanisms. In this case, we adopt very often the regression analysis model between input and output signals from the viewpoint of functional approach. In the previous paper, an extended regression analysis method was proposed by considering linear and non-linear correlation information on these signals. In this paper, we first introduce an entropy criterion, i.e., the minimization of conditional entropy of the output stochastic signal conditioned by the input stochastic signal, as an evaluation criterion for the system identification. Next, a practical method of regression analysis matched to the prediction of output response probability of complicated actual stochastic systems is proposed by applying this entropy criterion to the above extended regression analysis method. The effectiveness of the proposed method is experimentally confirmed by applying it to an actual acoustic system.