Gradient of the Mutual Information in Stochastic Systems: A Functional Approach

In this letter, using a general functional model for the stochastic systems we show that, the gradient of the mutual information with respect to any arbitrary system parameters can be computed in general. Relying on this viewpoint, the validity condition of one of the main available results on the gradient of the mutual information is investigated. Also, the result on the general additive noise system model in the literature is recovered as a special case.