The changing requirements for powertrain control systems, driven by legislation and competition, have influenced the evolution of electronic circuits in the past and continue to do so. Those factors and that evolution have now progressed to the point where some traditional engineering solutions that have been successful in the past may no longer be optimum. Perhaps there now is a better way to shape the underlying integrated circuit hardware for powertrain systems, specifically engine and transmission control. This paper presents reasons for adopting single-precision floatingpoint arithmetic (1-bit sign, 8-bit exponent and 23-bit mantissa) in conjunction with a Fast Fourier Transform (FFT) library, rather than using the double-precision arithmetic, to replace a portion of the map-table based control systems. It attempts to provide an unbiased benchmark study that systematically compares the accuracy and hardware cost of both approaches, using the Mat lab as a reference. The benchmark data suggest that a reasonable single-precision floating-point DSP library implementation can offer a robust and efficient system solution for a typical powertrain application that meets current economic constraints. Although certain computations can best be done on double-precision floating-point arithmetic, such cases are rare. Consequently, using a fast single-precision floating-point arithmetic solution supplemented with double-precision floating-point arithmetic is the optimum compromise for future powertrain control designs.
展开▼
