Numeric issues in test software correctness

Abstract

Test system designers are comfortable with the concepts of precision and accuracy with regard to measurements achieved with modern instrumentation. In a well-designed test system, great care is taken to ensure accurate measurements, with rigorous attention to instrument specifications and calibration. However, measurement values are subjected to representation and manipulation as limited precision floating-point numbers by test software. This paper investigates some of the issues related to floating point representation of measurement values, as well as the consequences of algorithm selection. To illustrate, we consider the test case of standard deviation calculations as used in the testing of infrared focal plane arrays. We consider the concept of using statistically-based techniques for selection of an appropriate algorithm based on measurement values, and offer guidelines for the proper expression and manipulation of measurement values within popular test software programming frameworks.