Towards Artificial Intelligence Acoustic Wave Filter Design

Imagine a neural network massively trained with the daily activity of a microwave filter design team. This work proposes a neural network to predict the design parameters of acoustic wave (AW) ladder filters. For this purpose, a prototype architecture has been built, set and trained with the objective of learning from the analysis of data that has been discarded during existing computational-expensive gradient-based methodologies. Some results shown demonstrate that a microwave AW filter can be properly synthesized from network predictions based on an Artificial Intelligence (AI), fulfilling some given frequency specifications and achieving evenly-distributed electromechanical coupling coefficient among resonators, which is such an important thing to obtain a feasible AW filter.