row1
Abstract
In this paper, we present a Deep Neural Network (DNN)–based framework that employs Radio Frequency (RF) hologram tensors to locate multiple Ultra-High Frequency (UHF) passive Radio-Frequency Identification (RFID) tags. The RF hologram tensor exhibits a strong relationship between observation and spatial location, helping improve robustness in dynamic environments and with varying equipment. Since RFID data is often marred by noise, we implement two types of deep neural network architectures to clean up the RF hologram tensor. Leveraging the spatial relationship between tags, the deep networks effectively mitigate fake peaks in the hologram tensors caused by multipath propagation and phase wrapping.
In contrast to fingerprinting-based localization systems that use deep networks as classifiers, our framework treats localization as a regression problem, preserving the ambiguity between fingerprints. We also present an intuitive peak-finding algorithm to estimate locations using the sanitized hologram tensors. The framework is implemented with commodity RFID devices, and its superior performance is validated through extensive experiments.
Summary
This paper details MulTLoc, a real-time Neural Network system made of two types of deep network architectures, for the purpose of identifying and localizing multiple passive RFID tags at once.
When RFID signals emitted from UHF passive tags reach an RF holographic tensor, the readings are fed into a model which attempts to find their location through signal peaks. The model will sometimes be wrong, and detect fake peaks, so a separate neural network accurately sorts the real peaks from the fake ones. The algorithm for intuitively finding the peaks is included in the paper.
All equipment used in the experimentation of this paper are commodity products. The components required to test the system include a Zebra AN720, a Zebra FX 9600 Reader, a Microsoft Kinect Sensor V2, and UPM Raflatac Frog 3D Tags.
