Improved Analog Filter Design by Random Search

The mainstream procedural analog filter design is a step by step process that restricts the designer to existing well-formed transfer functions and implementation topologies. At each step, approximations to the requirements are allowed to accumulate resulting in implementations failing to meet the original specifications. This paper presents a generalized analog circuit optimizer for linear systems, such as active filters. A random descent search is compared with other methods and shown to be an efficient method to stochastically search the design space and find available component values that minimize the cost function. A computationally efficient simulation technique mixing symbolic and numerical computation is employed that is one order of magnitude faster than solely numeric approaches for repeated evaluations of the same circuit. The optimizer is shown to produce improved designs of typical circuit topologies without prior information. The improved performance is found for single operational amplifier circuits by using unconventional higher order filter topologies. Monte Carlo bounds checking is optionally included in the optimizer to account for expected component variations, producing designs remaining within their original specifications despite nonideal components.