An Approach to Double-Balanced Mixer Synthesis Using Surrogate Modeling and Genetic Algorithm

In this paper, we propose an approach to double-balanced mixer synthesis. Distinctive features of the approach are using genetic algorithm to search the solution and surrogate modeling to speed-up the circuit simulation. The synthesis is carried out according to a set of requirements for conversion gain, port-to-port isolation, and LO and RF return loss. The synthesized mixer is configured as a downconverter. Used surrogate model is based on the combination of surface approximation and adaptive space mapping. Kriging interpolation is used as an approximation technique. Search algorithm varies the mixer element geometric parameters, RF and LO baluns layout, and IF filter order. We presented the synthesis example of 25-40 GHz double-balanced mixers based on $0.25 \mu \mathrm{m}$ GaAs pHEMT process. Results shown that each run gives us a solution that meets almost all requirements. Also, we estimated the influence of the LO and RF baluns surrogate model error on the mixer characteristics. The mean relative error between the EM simulation and surrogate model results was less than 4%.