Ranging detection algorithm for indoor UWB channels and research activities relating to a UWB-RFID localization system Dr Choi Look LAW Founding Director Positioning and Wireless Technology Centre School of Electrical and Electronic Engineering Nanyang Technological University, Singapore
Contents Introduction to UWB-RFID localization Indoor UWB channel measurements Ranging detection algorithm Results on one way ranging accuracy Active UWB-RFID localization based on TDOA
Scalable UWB-RFID Positioning System Motivation An ultra-wideband (UWB) enabled radio frequency identification (RFID) system scalable real time identification, localization, positioning and tracking of objects/nodes. scalability to thousands of nodes over area size of hundreds of meters applications such as logistics and environmental monitoring and protection positioning capability at sub-meter level using low power (<1mW) active tags and (<1 uw) passive tags. UWB-RFID Tag (x, y, z)? Figure 1: Scalable UWB-RFID Localization System Innovative Ideas Our innovative ideas are centered on UWB enabled backscatter RFID system architecture. For example, passive tags mounted on the walls will help to guide visually handicapped person to navigate in his home or a shopper to localize his position with respect to the goods that he wish to purchase Locator
Indoor UWB channel measurements Measurement Campaign Objectives Obtain a database of UWB channel profiles in various indoor environments. Test UWB ranging accuracy in indoor environment with FCC PSD mask compliance UWB signal and Model the ranging error statistically. Analyze the UWB ranging performance in indoor environment and use the analysis to facilitate the ranging parameters setting in both LOS and NLOS cases.
Measurement System Setup
Measurement Setup Antenna Tx Setup Rx Setup
Bi-conical Antenna Pattern z θ r θ = 45 ϕ = 38.7 r= 1.4cm R= 3.2cm y 3.1GHz Azimuth Plane 1.6 GHz Azimuth Plane x R Coaxial cable 3.1GHz Elevation Plane 1.6 GHz Elevation Plane
Test the Omni-directional properties of measurement setup Max-Min=.22V( = -15.65dB) D=1m Tx θ Rx Amplitude (V).1 -.1 455 454.5 454 453.5 453 452.5 Delay (ns) 452 451.5 1 12515175 75 5 25 Max-Min=2ps 3 325 2 22525275 Angle (Degree)
Measurement Environment Indoor Office Laboratory Room Open Hall Corridor
Campaign Summary Environment Sample Points Sample Spacing Maximum Distance LOS or NLOS Indoor Office 1257.2m 26m LOS and NLOS* Lab 271.2m 5m LOS Open Hall 61.5m 3m LOS Corridor 31 1m 3m LOS Total 162 * LOS Line of Sight NLOS Non Line of Sight
Indoor Office Layout with Measurement Routes 1257 measurement points in indoor environment.2 meter spacing Maximum distance is 26 meters 12 1 8 6 691 27 meters 4 2 267 NLOS 17 282 LOS NDDP DDP 27 meters
Received direct path pulse shape with Tx-Rx distance of 1m Direct Path
NLOS received waveforms with heavy blockage x 1-3 Original Waveform -Tx1L6D3 6 4 Amplitude (V) 2-2 -4 518 52 522 524 526 528 Time (ns) Crosscorrelation Function -Tx1L6D3.4.3.2.1 Amplitude -.1 -.2 -.3 -.4 518 52 522 524 526 528 Time (ns)
Ranging Algorithm Problem Statement Time of Arrival ranging systems using impulse radio UWB What is the optimum threshold setting and search window size for direct path detection How does SNR, LOS and NLOS environments affect these optimum settings
Ranging Performance Analysis The received signal r(t) is modeled as, L () α ( τ ) α ( τ ) () r t = s t + p s t + n t d d i i i i= 1.. Eq (1) Whereα d and τ d are the amplitude and propagation delay of direct path α i and τ i are the amplitude and propagation delay of i th multipath p i is the polarity of i th multipath n(t) is the WGN process
After correlated with the pulse template, the resulting waveform within [τp -δ, τp] can be expressed M R t = α R t τ + pα R t τ + R t () ( ) ( ) () c d ss d i i ss i ns i= 1.. Eq (2) Where R ss is the autocorrelation function of pulse template α M = α p and τ M =τ p Let us define: ρ d =α d /α p β d = τ p - τ d (Normalized direct path amplitude) (Time difference between Peak path and Direct path)
Conclusion on Ranging settings for LOS - According to measurement results, the direct path is not the largest path in17 profiles out of 289 profiles in LOS. - For LOS, simple strategy is enough: setting search period δ>2ns and detection threshold γ =mα p with m=.5~.6 14 12 1 β d (ns) 8 6 4 2.7.75.8.85.9.95 1 ρ d
Distribution of NLOS Direct Path Amplitude and Time of Arrival Probability Density 2 1.5 1.5 Curve Fitting Exp. Results Probability Density 5 x 17 4 3 2 1 Curve Fitting Exp. Results f ρ d.2.4.6.8 1 ρ d 1 d 1 = exp 2 πq( μ/ σρ) σρρd ( ρ ρ ) d where ( ) Q x 2 x 1 x = exp dx 2π 2 ;.2.4.6.8 1 β x 1-7 d (second) (( ln ρ ) ) 2 d μ 2σ 2 ρ βd βd f β ( β ) exp d d βd = η η.. Eq (3).. Eq (4)
- Evaluate the performance by large error probability ( Estimated arrival time of direct path true arrival time of direct path > T c /2) - The large error probability is related to three events H1 = { βd > δ} H { β δ} { α n γ} = + < 2 d d ns { γ } { α γ} Zmax sup{ Rns t } H = Z > + n 3 max d ns Where, () =, t [τp δ, τp ]and δ β d. n ns =R ns (τ p ), - Since three events are exclusive, the large error probability is Lgr ( γδ, ) = ( ) + ( ) + ( ) P P H P H P H 1 2 3. Eq (5)
Ignoring the intermediate derivation process, the final equation will be, δ PLgr = 1 Pexp 1 Ψ m, + 1 1 P Γ m, PΨ m, η ( ( κ )) ( ) ( κ) ( κ) ( ) ( m κ ) ( m ) δ 2δ 2 ηp Ω, exp exp η Ω( m, κ) 2 η Ω, κ....eq (6) Where m = γ α is normalized threshold p κ α p = is signal-to-noise (SNR) ratio σ ns
Comparison of simulation and analytical results, (Search window size=1ns) P Lgr 1.9.8.7.6.5.4.3 Analy. (κ=15db) Analy. (κ=2db).2 Analy. (κ=3db) Sim. (κ=15db) Sim. (κ=2db) Sim. (κ=3db).2.4 m.6.8 1
25 Adaptive ranging parameters Setting For NLOS, if channel parameters are given, numerical search may be performed with Eq(6) to obtained the optimum setting.8 δ opt (ns) 2 15 1 m opt.6.4 5 1 2 3 4 κ (db) 1.2 1 2 3 4 κ (db) 1-1 P min 1-2 1-3 1 2 3 4 κ (db)
Performance curves for various SNR and search windows size κ=1db κ=15db κ=2db κ=3db
Conclusion on Ranging settings for NLOS For NLOS, If channel parameters are not available, a two-state threshold settings method is proposed: (1). δ is predefined and fixed. A worst-case false alarm rate P fls is predefined m = δλ 1 2ln ln 1 fls κ ( P ) λ is a parameter related to the RMS bandwidth of pulse template (2). If the calculated m for a particular κ is larger than 1, the largest path is taken as the direct path and the earliest path searching path does not initialized.
Performance of optimum setting by numerical searching versus performance of two-state setting strategy with δ=5ns.9.8.7.6 Optimum P fls =1-1 P fls =1-2 P fls =1-3 P fls =1-6 P Lgr.5.4.3.2 1 15 2 25 3 35 κ (db)
Ranging Error Performance - Comparison of Coherent (CLEAN) and Non- coherent (Energy detection) LOS Indoor Office Lab NLOS Open Hall Corridor By Non-Coherent Detection Mean (m).18 4.336.11.31.2 STD. (m).15 1.22.14.19.15 Max (m).67 93.94 7.8 6 By Coherent Detection Mean (m).15 2.4.1.15.11 STD. (m).1 3.784.13.12.8 Max (m) 1 38.55.8.49.31
Distance Error(m) Distance Error(m) 2 15 1 2 15 1 5-5 5 1 15 2 25 3 5-5 Ranging Error Performance Actual Distance(m) -1 5 1 15 2 25 3 Actual Distance(m ) Probability of ε < Abscissa Probability of ε < Abscissa 1.8.6.4.2 Non-coherent Detection Coherent Detection.2.4.6.8.8 ε (m).6.4.2 - CDF of ranging errors ε ε = 5m LOS NLOS.5 1 1.5 2 ε (m) Non-coherent Detection Coherent Detection
Active UWB-RFID Localization UWB-RFID Locator/Reader Active UWB-RFID: TDOA computation in central controller Time synchronization among locators are through a hard wire UWB-RFID Active Tag 1 Central Location Controller
.8 Active UWB-RFID Localization using TDOA.8.2 1.5.6 6.6.8.45.4 5.8 4.45.4.35.35.2 5 5 5.3.3.3.3.2.3.3.3.2.2.3.3.2 5.2.5 1 5.45.32.4.45.4.35.45.4.4.45.6.8 2 5.8.6.2.6.8-2 -4.8.6.6.6.8 5.2-6 Lower bound of positioning error 5 1.5 5.2-6 -4-2 2 4 6 x(m) Measurement Environment 1 8 6 4 2-2 -4-6 -8-8 -6-4 -2 2 4 6 8 y(m).3.3.3.2.2.3 5 5 Positioning error between UWB measured locations and actual locations of 121 points Most locations positioning error < 1cm.2.3.5 x(m) 6 4 2-2 -4-6 estimated locations (blue dots) vs. actual locations (red star) -6-4 -2 2 4 6 x(m) y(m) y(m)
End of Presentation Thank you for your attendance
Acknowledgement (i) Mr Xu Chi Research Engineer and Part time PhD student (ii) Mr Fang Chao Research Engineer and Part time PhD student (iii) Mr Hu Sanming International attachment PhD student from SEU, Nanjing China (iv) Ms Xu Jun Full time PhD student (v) Mr Zhou Yuan Full time PhD student (AGS scholar) (vi) Ms Jiang Jisu Full time PhD student (vii) Ms Thida Than Research Engineer