Implementation of Extended Kalman Filter for Battery State of Charge Estimation on an Embedded Platform

State of Charge (SoC) estimation is essential for battery-powered embedded systems. Recent studies show that the Extended Kalman Filter (EKF) with a second-order RC battery model offers a trade-off between computational complexity and estimation accuracy, making it suitable for resource-constrained platforms like embedded systems. This paper focuses on optimizing EKF implementation to minimize single-run execution time and resource usage by exploring techniques such as loop unrolling, function inlining, the lookup table method, and data access strategies. EKF implementations were tested on an STM32-based embedded platform, and their performance was evaluated based on execution time and memory footprint. Results show that function inlining, combined with a lookup table for polynomial calculations and loop unrolling for matrix operations, offer the best execution time. However, using for loops for matrix multiplication without function inlining results in the most efficient memory usage. Results presented in this paper provide a good groundwork for further improvements of battery SoC algorithms, aiming to enhance energy efficiency and extend overall system operation time.