부산대학교 도서관

Radio direction finding, traditionally used for localizing radio signal sources, has been adapted for Bluetooth to enable indoor localization of wireless devices. This adaptation is particularly relevant for achieving accurate indoor localization within Internet-of-Things (IoT) networks, especially in battery-powered and resource-limited embedded systems. However, the intricacies of implementing direction of arrival (DOA) methods in such systems, notably those lacking a floating-point unit (FPU), present significant computational challenges. This article addresses these challenges by introducing an innovative fixed-point DOA method, rooted in the estimation of signal parameters via rotation invariance techniques (ESPRIT). Diverging from traditional complex eigenvalue decomposition, our approach employs a simpler power method for DOA estimation and phase offset compensation, utilizing a straightforward trigonometric equation. It also integrates an improved carrier frequency estimator, also based on ESPRIT, which is tens of times more accurate than the conventional method of averaging phase differences. We conducted bare-metal level experiments on an nRF52840 system on chip to evaluate execution time, memory footprint, angle accuracy, and energy consumption. The fixed-point implementation demonstrated an execution time of 2.3 ms and an energy consumption of just 0.348 nWh. These figures represent a 5.9-fold increase in energy efficiency and a 4.4-fold improvement in speed compared to the conventional software-based floating-point approach while maintaining an angle accuracy ranging from nearly 2° to under 0.5°, depending on the signal-to-noise ratio. However, in IoT devices equipped with an FPU, the hardware-based floating-point technique still edges out, being 0.8 ms faster and slightly more energy efficient at 0.319 nWh.