An approach for AI-based filter design by means of neural networks

Abstract

In this work a Neural Network Approach (NNA) for the design of high-Q active inductors (AIs) based active RF filters is presented. The proposed approach consists on the extraction of the S-parameters of an active device properly biased at different bias points by exploiting a neural network which allows to automatically determine the best solution for the design. In this way it is possible to define easily the best network configuration and filter order at the early design stage. As an example of application, a passband filter has been conceived following the proposed approach; it has been optimised for operating at 960 MHz with a 3 dB bandwidth of about 5 MHz and a very high slope factor.