Implementation and Testing of Numerical Analysis Techniques in Avionics Applications.

摘要

Errors due to finite wordlength are unavoidable when aircraft signal processing operations such as flight control, navigation, and fire control are implemented on a digital computer. To reduce these errors to tolerable levels, longer word-lengths can sometimes be employed. The effects of some of the errors, such as those due to arithmetic series truncation, machine roundoff, and quantization of system coefficients, can be lessened somewhat by appropriate numerical analysis techniques. An n-bit simulator which runs on Control Data Corporation (CDC) 6600/CYBER 74 computer systems was modified and then used to evaluate the accuracy of a flight navigation routine coded in FORTRAN. The routines were executed without the simulator to obtain results used for benchmarking. The n-bit simulator was employed to simulate the numerical characteristics of the AN/AYK-15A digital processor. Error plots were constructed which show the maximum errors occurring within small plotting intervals plotted against each individual input value. These plots were used to aid visually in analyzing the error characteristics of the avionics routine as it would be implemented on the AN/AYK-15A. A critical analysis of the error plots obtained showed that routines which are coded using single-precision floating-point arithmetic are prone to errors which exceed the error bounds specified for the routines. This occurs even though range reductions in the trigonometric function approximations are accomplished using extended precision. (Author)