Channel Metamodeling for Explainable Data-Driven Channel Model

Machine learning can produce accurate data-driven channel models, but their black-box nature makes it harder to explain the models and to understand underlying channel characteristics. In this letter, we propose a channel metamodeling approach for such a black-box data-driven channel model. Our approach enables us to express the data-driven channel model in terms of transparent mathematical expressions based on symbolic function approximation methods. Through experiments with synthetic and real datasets, we demonstrate that our approach produces a channel metamodel of the data-driven channel model for each dataset that is highly accurate and allows us to easily explain the data-driven channel model and to understand the underlying channel characteristics.