Architecture-Driven Fixed-Point Scaling for AUTOSAR-based ECUs: A Production-Feasible Architecture for Centralizing Scaling Semantics

Floating-point arithmetic is widely used in automotive embedded software to scale Controller Area Network signals and calibration parameters with fractional factors such as 0.1. However, floating-point operations, even on microcontrollers equipped with floating-point units, can increase execution time and CPU load. In AUTOSAR architectures, converting floating-point scaling to fixed-point is not trivial because scaling semantics must be integrated consistently across components, yet AUTOSAR platform toolchains offer only limited automation at the Application Data Type level. Although CompuMethod definitions can express scaling, integration typically remains manual and distributed across application software components, reducing consistency and reusability. This study presents an architecture-driven methodology that formalizes fixed-point scaling as a centralized architectural service, realized through a parser-driven fixed-point macro generation pipeline. Standardized CAN DBC and calibration metadata are parsed to automatically generate integer-only macros for raw-to-physical and physical-to-raw transformations. The generated macros are integrated into dedicated AUTOSAR-compliant Scaling Service software components, consolidating scaling logic and improving reliability and maintainability. The approach requires no changes to toolchains, compiler settings, or hardware, enabling direct deployment in AUTOSAR-based software. The methodology was applied to a production-grade Integrated Charging Control Unit targeting Electric Vehicle Communication Controller software. Evaluation included cycle-accurate profiling, edge-based timing, isolated CPU load calculation, and average current measurement. Results show a 98.84% reduction in floating-point operations and a 92.67% reduction in conversion-related source lines. Task execution time decreased by 16.13%, CPU load decreased by 6.88%, and average current consumption showed a repeatable 0.81% reduction. These results demonstrate that the proposed methodology improves execution efficiency and is applicable to production AUTOSAR-based ECUs.