Wi-Fi Channel Quality Prediction with Liquid Neural Networks

The behavior of the wireless spectrum is not entirely random, and often it exhibits patterns that can be predicted to some degree. This enables industrial applications and networks to anticipate variations in link quality and proactively take appropriate countermeasures, in such a way to meet the typical requirements about reliability and determinism of such contexts. To determine to what extent this can be done, four real Wi-Fi channels were periodically sampled and the resulting datasets were used to train and test the relatively new liquid neural network machine learning models, specifically designed for time-series prediction. In particular, two main representatives of this class of models were examined, namely, the liquid time-constant (LTC) and the closed-form continuous-time (CFC) models. Results show appreciable improvements for the CFC model compared to the state-of-the-art approaches and good stability regarding generalization under variable channel conditions.