65 Automated Synthesis of Microwave Filters Using Modular Neural Networks

An intelligent circuit synthesis method for automatic tuning of microwave filters is reported. The intelligent agent is a neural network used for mapping RF filter performance specifications to tunable component parameter values. The neural network acts as an efficient and accurate real-time synthesis tool which when supplied with filter performance specifications calculates the required capacitor and inductor values to produce a circuit whose response is consistent with performance specifications. To demonstrate the efficacy of the approach, a tunable LC microwave high pass filter is synthesized using a modular neural network having 7 local experts, 12 inputs and 7 outputs. The resulting design - tunable from 7GHz to 12GHz in steps of 0.2GHz — required less than 3 cpuseconds per circuit synthesis on an IBM X60 Lenovo laptop computer.