Large Scale Optimization of Analog Circuits with Microwave Applications

Abstract

<p>This thesis addresses itself to computer oriented techniques for large scale optimization of analog circuits. New techniques for simulation and sensitivity analysis are described and are used to improve the performance of circuit optimization. A powerful automatic decomposition technique is developed directly enabling a normal optimizer to solve large circuit problems. Our theory is applied to the design of microwave circuits.</p> <p>The status of large scale circuit optimization and the state-of-the-art of microwave CAD are reviewed. The necessity of circuit oriented optimization techniques is demonstrated by formulating design, modelling, diagnosis and tuning into optimization problems.</p> <p>A comprehensive treatment of large change sensitivity computation for linearized circuits using generalized Householder formulus is presented. A technique for circuit response updating via a minimum order reduced system is developed. By avoiding re-analysis of the complete circuit, our method is responsible for efficient simulation of large circuits when a subset of the circuit parameters is frequently perturbed.</p> <p>An elegant theory for simulation and exact sensitivity analysis of branched cascaded networks is described. Our approach explicitly takes the circuit-structure into consideration and does not deteriorate as the overall network becomes large. The practicality of the theory is illustrated by efficient optimization of microwave multiplexers consisting of multi-cavity filters distributed along a waveguide manifold. Examples of optimizing 12- and 16-channel multiplexers are provided.</p> <p>A novel and general automatic decomposition technique for large scale optimization of microwave circuits is presented. The partitioning approach proposed by Kondoh for FET modelling problems is verified. The application of our technique is demonstrated by the large scale optimization of a 16-channel multiplexer involving 399 nonlinear functions and 240 variables.</p>