Title: A Training-Based Mutual Information Lower Bound for Large-Scale Systems

Abstract: We provide a mutual information lower bound that can be used to analyze the effect of training in models with unknown parameters. For large-scale systems, we show that this bound can be calculated using the difference between two derivatives of a conditional entropy function. The bound does not require explicit estimation of the unknown parameters. We provide a step-by-step process for computing the bound, and provide an example application. A comparison with known classical mutual information bounds is provided.