Takahiro Yamamoto*1, Shigemi Ishida*1, Ryota Kimoto *1, Shigeaki Tagashira*2, Akira Fukuda *1 *1 Kyushu University, *2Kansai University Design of BLE 2-Step Separate Channel Fingerprinting

Background 2 ■ IoT systems and location-based services • Location is important information ■ Localization • Outdoor: GPS • Indoor: Manual measurement or indoor localization system ■ Localization using wireless signals • BLE is popularly used in many IoT systems

BLE Localization 3 ■ BLE (Bluetooth Low Energy) • Low power, low cost, narrow-band wireless communication technology • Frequency hopping spread spectrum (FHSS) ■ BLE localization • Uses advertising packets sent on 3 channels separated by up to 78MHz

Unstable BLE Signal Strength [Ishizuka+14] “A fundamental study on a indoor localization method using BLE signals and PDR for a smart phone – sharing results of experiments in Open Beacon Field Trial (in Japanese)”, IEICE Tech. Rep. MoNA 4 ■ Received signal strength (RSS) drastically changes time to time • Because of frequency separation of 3 advertising channels [Ishizuka+14] ε ༷ ore) ਤ 2 BLE γάφϧͷ RSSI ͱڑ཭ଌҐ݁Ռ զʑ͸ɼਤ 3 ͷΑ͏ʹ BLE σόΠεΛਖ਼ࡾ֯ܗʹ഑ஔ͠ɼத ৺ʹ RSSI ड৴༻ͷ୺຤Λ഑ஔ͢Δ͜ͱͰɼෳ਺ BLE σόΠ 45dB 10dB Each measurement suffered from RSS changes by up to 45dB → RSS cannot directly be mapped to distance Average also suffered from 10dB changes → Long-term measurement cannot overcome the unstable RSS

Related Work [Zhu+14] “RSSI based Bluetooth Low Energy indoor positioning”, IPIN [Paterna+17] “A Bluetooth Low Energy indoor positioning system with channel diversity, weighted trilateration and kalman filtering”, Sensors [Li+18] “Indoor positioning algorithm based on the improved RSSI distance model”, Sensors [Faragher+15] “Location fingerprinting with Bluetooth low energy beacons”, IEEE J. Sel. Areas Commun. [Ishida+16] “Proposal of separate channel fingerprinting using Bluetooth Low Energy”, IIAI-AAI 5 ■ Accuracy improvement in BLE-based localization • With filtering outliers [Zhu+14][Paterna+17] • Compensation using multiple BLE devices [Li+18] • Fingerprinting [Faragher+15] ■ We also proposed separate channel fingerprinting (SCF) [Ishida+16] • Fingerprinting employing channel diversity of 3 adv. channels to improve localization accuracy

Separate Channel Fingerprinting (SCF) [Ishida+16] [Ishida+16] “Proposal of separate channel fingerprinting using Bluetooth Low Energy”, IIAI-AAI 6 ■ Separately measure RSS on 3 advertising channels ■ Channel diversity is employed as location-specific feature

High Max Error Problem 7 ■ SCF suffers from high max error • Compared to unified channel fingerprinting (UCF), which is a conventional BLE fingerprinting, SCF improves mean localization error • Max error is more than 6 meters for both UCF and SCF ⇒Want to reduce max error while improving localization accuracy 0 2 4 6 0 200 400 Frequency UCF 0 2 4 6 Localization error [meters] 0 200 400 Frequency SCF mean = 0.71 mean = 0.59

Key Idea: Localization in 2-Steps 8 1. Coarse localization • Ignore channel information and averaged over RSS measured on 3 adv. channels 2. Fine-grained localization • Utilize channel-diversity in fingerprinting

9 1. Coarse localization • UCF over whole localization area 2. Fine-grained localization • SCF over an area based on the result of coarse localization 2-Step SCF: System Overview DB 1 - -52 BLE channel RSS 1 - -58 2 - -55 2 - -60 location (0,0) (0,0) (0,1) (0,1) 2 - -58 (0,1) : : : : Unified-Channel Fingerprints DB R 1 37 -52 BLE channel RSS 1 38 -58 2 37 -55 2 38 -60 location (0,0) (0,0) (0,1) (0,1) 2 39 -58 (0,1) : : : : Separate-Channel Fingerprints Estimated location in coarse localization Training location used in location estimation Training location ignored in location estimation Nearest locations selected in location estimation Estimated location

2-Step SCF: Training Phase 10 ■ Constructs separate- and unified-channel fingerprint databases • Collect a set A!,# of RSS samples of BLE beacon measured on channel ∈ {37,38,39} at location ∈ L • Separate channel fingerprint • ! = !,# 37 , !,# 38 , !,# 39 , !,$ 37 , … , !,% 39 • !,& = median A!,& • Unified channel fingerprint • 2 ! = !,#, !,$, !,', … , !,% • !,& = median ⋃(∈ '*,'+,', A!,&

2-Step SCF: Estimation Phase 11 ■ Coarse localization • Collect a set B# of RSS samples of BLE beacon measured on channel ∈ {37,38,39} at a target location • Unified channel target fingerprint / • 4 = #, $, ', … , % • & = median ⋃(∈ '*,'+,', B& • Estimate location using unified channel fingerprint 1 ! and unified channel target fingerprint /

2-Step SCF: Estimation Phase 12 ■ Fine-grained localization • Separate channel target fingerprint • = # 37 , # 38 , # 39 , $ 37 , … , % (39) • & = B& • Estimate location using separate channel target fingerprint and separate channel fingerprint ! in a limited area • The limited area is a circular area of radius R centered on the location estimated in coarse localization • Insufficient number of separate channel fingerprints in a limited area means localization failure

13 BLE beacon Reference location Target location 10 meters ■ H-shpaed corridor in our Univ • 24 Silicon Labs BLED112 beacons • Measure RSS on MacBook Pro • Collect fingerprints at 46 reference locations • Collect target fingerprints at 7 target locations Evaluation: Setup BLE Beacons

14 ■ Localization accuracy • 95th percentile of localization errors for all localization trials 0 2 4 6 8 10 Localization error [meters] 0.0 0.2 0.4 0.6 0.8 1.0 Cumulative probability UCF SCF 2S-SCF Localization Error Method Localization accuracy [meters] Max error [meters] UCF 6.76 6.83 SCF 2.60 6.95 2-Step SCF 1.00 2.05 61.5%

15 ■ Localization success rate • Increases as R increases • Saturated at 60% when R > 2.0 ■ Localizaton accuracy • Increases as R increases ■ Trade-off between: • Small localization accuracy • High localization success rate ⇒Need to determine the best R 2 4 6 8 10 Area-limit radius R [meters] 0.0 0.5 1.0 1.5 2.0 2.5 Localization accuracy [meters] Localization accuracy Localization success rate 0 10 20 30 40 50 60 Localization success rate [%] Area-Limit Radius

Summary 16 ■ BLE-based localization • Suffers from low accuracy due to freqnecy separation • Previously proposed SCF utilizing channel diversity to improve localization accuracy, which still suffers from high max error ■ 2-step SCF • Coarse location estimation w/o channel diversity • Fine-grained location estimation w/ channel diversity in a limited area • Improved localization accuracy by 61.5% while reducing max localization error

© 2020 Shigemi ISHIDA, distributed under CC BY-NC 4.0