MA-RTI: Design and Evaluation of a Real-World Multipath-Assisted Device-Free Localization System

Device-free localization (DFL) systems exploit changes in the radio frequency channel by measuring, for example, the channel impulse response (CIR), to detect and localize obstacles within a target area. However, due to a lack of well-defined interfaces, missing modularization, as well as complex system configuration, it is difficult to deploy DFL systems outside of laboratory setups. This paper focused on the system view and the challenges that come with setting up a DFL system in an indoor environment. We propose MA-RTI, a modular DFL system that is easy to set up, and which utilizes a multipath-assisted (MA) radio-tomographic imaging (RTI) algorithm. To achieve a modular DFL system, we proposed and implemented an architectural model for DFL systems. For minimizing the configuration overhead, we applied a 3D spatial model, that helps in placing the sensors and calculating the required calibration parameters. Therefore, we configured the system solely with idle measurements and a 3D spatial model. We deployed such a DFL system and evaluated it in a real-world office environment with four sensor nodes. The radio technology was ultra-wideband (UWB) and the corresponding signal measurements were CIRs. The DFL system operated with CIRs that provided a sub-nanosecond time-domain resolution. After pre-processing, the update rate was approximately 46 Hz and it provided a localization accuracy of 1.0 m in 50% of all cases and 1.8 m in 80% of all cases. MA fingerprinting approaches lead to higher localization accuracy, but require a labor-intensive training phase.

Automated summary

Device-free localization (DFL) systems utilize changes in the radio frequency channel to detect and localize obstacles in a target area. This paper proposes a modular DFL system called MA-RTI, which utilizes a multipath-assisted radio-tomographic imaging algorithm. The system is configured using a 3D spatial model and idle measurements, and achieves a localization accuracy of 1.0m in 50% of cases and 1.8m in 80% of cases when deployed in a real-world office environment with four sensor nodes.