RF-Chord: Towards Deployable RFID Localization System for Logistics Network

Bo Liang1,4*, Purui Wang3*, Renjie Zhao5, Heyu Guo1, Pengyu Zhang4, Junchen Guo4, Shunmin Zhu6,4, Hongqiang Harry Liu4, Xinyu Zhang5, Chenren Xu1,2,7
*Co-primary student authors
1Peking University   2Zhongguancun Laboratory   3MIT   4Alibaba Group   5UC San Diego   6Tsinghua University
7Key Laboratory of High Confidence Software Technologies, Ministry of Education (PKU)
USENIX NSDI 2023

RF-Chord localizes 180+ RFID tags within 1 second at up to 6 m range, enabling automated inventory management in logistics networks with 0% miss-reading rate.

RFID-based logistics scenario

In a typical logistics scenario, packages are discharged from trucks, scanned at an inventory gate, and sorted for warehouse check-in. The RFID-based inventory gate should meet reliability, throughput, and range requirements simultaneously.

Abstract

RFID localization is considered the key enabler of automating the process of inventory tracking and management for high-performance logistics networks. A practical and deployable RFID localization system needs to meet reliability, throughput, and range requirements. This paper presents RF-Chord, the first RFID localization system that simultaneously meets all three requirements. RF-Chord features a multisine-constructed wideband design that can process RF signals with a 200 MHz bandwidth in real-time to facilitate one-shot localization at scale. In addition, multiple SINR enhancement techniques are designed for range extension. On top of that, a kernel-layer near-field localization framework and a multipath-suppression algorithm are proposed to reduce the 99th long-tail errors. Our empirical results show that RF-Chord can localize up to 180 tags 6 m away from a reader within 1 second and with 99th long-tail error of 0.786 m, achieving a 0% miss reading rate and ~0.01% cross-reading rate in the warehouse and fresh food delivery store deployment.

System Overview

RF-Chord embraces any ISM-band reader as the tag activator. An active sniffer reader emits a low-power (-15 dBm) wideband multisine waveform to pick up tag responses over a wide frequency band. The RF frontend and FPGA hardware receive tag responses from 8 antennas and 16 frequencies simultaneously. After one-shot tag inventory, RF-Chord obtains information from both frequency and spatial domains for robust localization in multipath-rich environments. A kernel-layer-based near-field localization algorithm then suppresses the multipath effect and estimates the tag location.

RF-Chord system overview

Key Designs

One-shot Wideband with Multisine Wave

RF-Chord constructs a wideband backscatter signal using a multisine waveform. Instead of frequency hopping (which takes seconds per tag), RF-Chord transmits multiple narrowband carriers simultaneously across 200 MHz. Digital channelization on the receiver separates the combined wideband signal into parallel narrowband channels, reducing the data rate by 50x without information loss. This enables one-shot channel measurement across 8 antennas and 16 frequencies in ~5 ms.

Single-tone backscatter model

Single-tone backscatter

Multisine excitation signals

Multisine excitation signals

Analog channelization

Analog channelization

Digital channelization

Digital channelization

SINR Enhancement for Long Range

To follow FCC regulations, the sniffer's emission power must be below -13.3 dBm — 50 dB weaker than commercial readers. RF-Chord achieves 6 m range through multiple SINR enhancement techniques: (1) narrowband channelization reduces thermal noise by 29 dB; (2) high-resolution digital channelization and low crest factor waveform design handle self-interference; (3) full packet matching exploits integration gain across the entire tag response; (4) accurate clock offset mitigation and channel diversity decoding further boost SNR.

Kernel-Layer Near-Field Localization

Even with 200 MHz bandwidth, multipath remains the primary source of long-tail localization errors (distance resolution of 0.75 m is insufficient for all indoor scenarios). RF-Chord proposes a kernel-layer framework: the kernel function characterizes location estimation from a single channel, and layer functions coherently combine multiple channels into a final estimate. This framework supports choosing different kernel and layer functions for various deployment scenarios — for example, incorporating prior knowledge about the region of interest to enhance the direct path and suppress multipath.

Kernel-layer localization framework

Results

RF-Chord is evaluated at 384 locations with over 20k tag responses. Key results:

  • Throughput: Localizes up to 180 tags/s (1000x faster than previous sniffer-based wideband systems)
  • Range: Operates at up to 6 m with no obvious throughput or reliability loss
  • Reliability: 99th-percentile localization error of 0.786 m
  • Deployment: 0% miss-reading rate and ~0.01% cross-reading rate in real-world logistics

Localization Performance

CDF by bandwidth

Bandwidth impact

CDF by antenna number

Antenna number impact

CDF by algorithm

Algorithm impact

Increasing bandwidth from 50 MHz to 200 MHz reduces the 99th-percentile error from 2.398 m to 0.786 m. More antennas consistently improve long-tail performance. The multipath-suppression algorithm further reduces long-tail errors by incorporating prior information, while median performance remains stable at ~0.14 m.

Real-World Deployment

RF-Chord is deployed in a warehouse dock door and a fresh food delivery store scanning gate. Over 10,000 tags are attached to various items. RF-Chord achieves 100% tag reading accuracy (0% miss-reading) with only 0.0025%–0.0154% cross-reading rate, significantly outperforming commercial solutions like Impinj xSpan (~6% miss-reading, ~2% cross-reading).

Warehouse dock door deployment

Warehouse dock door

Fresh food delivery store deployment

Fresh food delivery store

Hardware

RF-Chord hardware board

Custom PCB with ADRV9009 RF transceiver and XCKU060 FPGA

Experimental setup

Evaluation setup with 8-antenna array

Open Source

We open-sourced RF-Chord's hardware design, evaluation dataset, and example localization scripts:

  • Hardware design (Google Drive) — Custom PCB schematics and antenna array design
  • Evaluation dataset (Google Drive) — ~20k wideband RFID channel measurements at 384 locations

The dataset is collected in the Geek Lab at Peking University, in an office environment with multiple metal reflectors. The evaluation area is 6 x 3.2 m, divided into 20 cm grids. Five tags are mounted on a guide rail at each measurement location. Example localization scripts (hologram_localization_script.m and DPE.m) are provided to reproduce the basic hologram and direct path enhancement algorithms from the paper.

BibTeX

@inproceedings{liang2023rfchord,
  title={{RF-Chord}: Towards Deployable {RFID} Localization System for Logistics Network},
  author={Liang, Bo and Wang, Purui and Zhao, Renjie and Guo, Heyu and Zhang, Pengyu and Guo, Junchen and Zhu, Shunmin and Liu, Hongqiang Harry and Zhang, Xinyu and Xu, Chenren},
  booktitle={20th USENIX Symposium on Networked Systems Design and Implementation (NSDI 23)},
  pages={1783--1799},
  year={2023}
}