An AnguLoc oc: Concu current Angle of Ar Arrival Estimation on - - PowerPoint PPT Presentation

An AnguLoc

• c: Concu

current Angle of Ar Arrival Estimation

• n for
• r Indoor
• or

Lo Localiz alizatio ion wit ith UW UWB Rad adio ios

Milad Heydariaan, Hossein Dabirian, Omprakash Gnawali Networked Systems Laboratory, University of Houston Contact: milad@cs.uh.edu DCOSS 2020 June 2020

Wha What is s Indo ndoor Localization? n?

• Finding location of people, things, and places indoors
• Market size: $18.74 billion by 2025*

2

Navigation Tracking

* https://www.reportlinker.com/p05763837/Indoor-Location-based-Services-Market-Analysis-Report-By-Product-By-Technology-By-Application-By-End-Use-And-Segment-Forecasts.html

Heydariaan, Dabirian & Gnawali - University of Houston

Ultr Ultra-wi wideba deband nd (UWB WB) Radi dios

• “GPS at the scale of your living room” [Apple Inc.]
• Accurate (10 cm)
• Global market size of $58 million in 2019*
• At least 75 million units of iPhone 11 by the end of 2019**
• The UWB market is expected to grow significantly
• iPhone 12
• Android
• UWB Alliance and FiRa Consortium

NXP, Qorvo, Decawave, Bosch, Samsung, Hyundai

* https://www.absolutereports.com/global-ultra-wideband-market-15311454 ** https://www.bloomberg.com/news/articles/2019-10-14/apple-s-lower-prices-users-aging-handsets-drive-iphone-demand

Decawave DW1000 chip

3

Wireless interference will be an issue

Heydariaan, Dabirian & Gnawali - University of Houston

UW UWB In Inter erfer eren ence

• Avoiding Interference
• Time-division multiple access (TDMA)
• ALOHA
• Carrier sensing not feasible
• Mitigating Interference
• Forward Error Correction (FEC)
• Retransmissions
• Exploiting Interference
• Concurrent Transmissions

4

Not Scalable Not Efficient

A B C Destructive interference prevents packet reception

Heydariaan, Dabirian & Gnawali - University of Houston

Wi Wireless ss Interferenc nce vs. s. Sc Scalability a and E Effici ciency cy

5

Non-Scalable Inefficient Scalable and Efficient Level of Interference Low Medium High Avoiding Interference Mitigating Interference Exploiting Interference

Heydariaan, Dabirian & Gnawali - University of Houston

Se Sequential Loc Localization

• n - 1

Initiator A1 A2 An …

Request Request Request

One Request

Initiator A1 A2 An

R e q u e s t

…

6 Heydariaan, Dabirian & Gnawali - University of Houston

Sequential Responses

Se Sequential Loc Localization

• n - 2

Initiator A1 A2 An …

Response 1

Initiator A1 A2 An

R e q u e s t R e s p

• n

s e 1

…

7 Heydariaan, Dabirian & Gnawali - University of Houston

5 ms

Sequential Responses

Se Sequential Loc Localization

• n - 3

Initiator A1 A2 An …

Response 1 Response 2

Initiator A1 A2 An

R e q u e s t R e s p

• n

s e 1 R e s p

• n

s e 2

…

8 Heydariaan, Dabirian & Gnawali - University of Houston

5 ms 5 ms

Sequential Responses

Se Sequential Loc Localization

• n - 4

Initiator A1 A2 An …

Response 1 Response 2 Response n

Initiator A1 A2 An

R e q u e s t R e s p

• n

s e 1 R e s p

• n

s e 2 R e s p

• n

s e n

…

9 Heydariaan, Dabirian & Gnawali - University of Houston

5 ms 5 ms 5 ms

Con Concurr rrent Loc Localization

• n - 1

Initiator A1 A2 An …

Request Request Request

One Request

Initiator A1 A2 An

R e q u e s t

…

10 Heydariaan, Dabirian & Gnawali - University of Houston

5 ms

Con Concurr rrent Loc Localization

• n - 2

Initiator A1 A2 An …

Response 1 Response 2 Response n

Initiator A1 A2 An

R e q u e s t R e s p

• n

s e 1 R e s p

• n

s e 2 R e s p

• n

s e n

…

11 Heydariaan, Dabirian & Gnawali - University of Houston

5 ms 128 ns 128 ns

Concurrent Responses

Con Concurr rrent Packets in IEEE 802. 802.15. 15.4 4 UWB B PHY

We can only demodulate Data from the first arriving packet

12

First Path for P1 First Path for P2 1 Amplitude

We can observe combined preamble in channel impulse response (CIR)

Heydariaan, Dabirian & Gnawali - University of Houston

Preamble 1 SFD 1 PHR 1 Data 1 Preamble 2 SFD 2 PHR 2 Data 2 Preamble 1 + Preamble 2 SFD 1 PHR 1 Data 1 P1 P2 P1,2

+

=

P1,2 = P1 + P2

TX TX Scheduling Uncertainty

• In concurrent localization protocols: !"# = !%# + '"#
• Difference between precision of !"# and !%#
• Causes inaccuracy in ToA estimation
• Causes up to 2.4 m of localization error in DW1000
• State-of-the-art concurrent TDoA solutions
• Wired correction: deployment issues
• Wireless correction: additional packets, antenna delay calibration, 1-cycle lag
• Our solution (AnguLoc): Concurrent AoA

13

Initiator Ri

R e q u e s t R e s p

• n

s e i

'"# !%# !"#

Heydariaan, Dabirian & Gnawali - University of Houston

Con Concurr rrency-ba based ed Localization n Solut utions ns

14

Related Work Feasibility Study Solution for TX Scheduling Uncertainty Accuracy Localization Method TREK1000 (Sequential)

• 0.30 m

TWR1 Corbalán [EWSN’18] Concurrent TWR ! ~ 2 m TWR Corbalán [IPSN’19] Chorus Concurrent TDoA2 ! ~ 1.2 m TDoA Großwindhager [IPSN’19] SnapLoc Concurrent TDoA Wired/Wireless Correction ~ 1.2 m (without correction)5 TDoA Heydariaan [DCOSS’19] ! ! ~ 2 m TWR Heydariaan [DCOSS’20] AnguLoc Concurrent AoA3 Immune Against TX Scheduling Uncertainty 0.67 m ADoA4

1 TWR: Two-Way Ranging 2 TDoA: Time Difference of Arrival 3 AoA: Angle of Arrival 4 ADoA: Angle Difference of Arrival Heydariaan, Dabirian & Gnawali - University of Houston 5 Authors said they achieved better results with wired/wireless corrections

θ

• θ

θ1 θ2 θ1 θ2

Con Concurr rrent An Angle of

• f Arri

Arrival Loc Localization

• n –

Ch Challenges and Opport

• rtunities
• Opportunities
• Concurrent AoA is more accurate than concurrent TDoA

Concurrent AoA is not affected by TX scheduling uncertainty

• Self-localization is highly scalable

An unlimited number of tags

• Challenges
• Front-back ambiguity of angle measurements
• Unknown tag tilting

15 Heydariaan, Dabirian & Gnawali - University of Houston

Con Contri ribution

• ns
• Feasibility of concurrent AoA
• Angle difference of arrival algorithm overcomes
• Front-back ambiguity of angle measurements
• Unknown tag tilting
• Increasing accuracy of concurrency-based localization

Heydariaan, Dabirian & Gnawali - University of Houston 16

An Angle of Ar Arrival Using Phase Difference of Ar Arrival

17

θ Transmitter p d Receiver A Receiver B Xtal

Angle of Arrival with two receivers running

• n the same crystal oscillator (Xtal).

Heydariaan, Dabirian & Gnawali - University of Houston

! = # sin ' ( = 2* + , = 2* ( ! = +

• !

' = sin./ ,( 2*# We calculate , by calculating phase for first path in CIR

An Angle of Ar Arrival Hardware

18 Heydariaan, Dabirian & Gnawali - University of Houston

Decawave PDoA node (DWM1002)

Con Concurr rrent An Angle of

• f Arri

Arrival

• AoA is ! = sin&' ()

*+,

• - is the wavelength
• . is the distance between

antennas

• / is the difference in phase

between two antennas calculated at each responder’s first path

19

First Responder First Path Second Responder First Path We can calculate phase information by detecting first path of each responder

Heydariaan, Dabirian & Gnawali - University of Houston

Con Concurr rrent Se Self-Loc Localization

• n P

Prot

• toc
• col
• l

20 Heydariaan, Dabirian & Gnawali - University of Houston

AREF

T2

A2 A1 A3 A4

T1 T4 T3 T5 T6

Con Concurr rrent Se Self-Loc Localization

• n P

Prot

• toc
• col
• l

21

• 1. AREF broadcasts SYNC

Heydariaan, Dabirian & Gnawali - University of Houston

AREF

T2

A2 A1 A3 A4

T1 T4 T3 T5 T6

Con Concurr rrent Se Self-Loc Localization

• n P

Prot

• toc
• col
• l

22

• 1. AREF broadcasts SYNC
• 2. Ai’s reply concurrently

Heydariaan, Dabirian & Gnawali - University of Houston

AREF

T2

A2 A1 A3 A4

T1 T4 T3 T5 T6

Con Concurr rrent Se Self-Loc Localization

• n P

Prot

• toc
• col
• l

23

• 1. AREF broadcasts SYNC
• 2. Ai’s reply concurrently

Heydariaan, Dabirian & Gnawali - University of Houston

AREF

T2

A2 A1 A3 A4

T1 T4 T3 T5 T6

Con Concurr rrent Se Self-Loc Localization

• n P

Prot

• toc
• col
• l

24

• 1. AREF broadcasts SYNC
• 2. Ai’s reply concurrently

Heydariaan, Dabirian & Gnawali - University of Houston

AREF

T2

A2 A1 A3 A4

T1 T4 T3 T5 T6

Con Concurr rrent Se Self-Loc Localization

• n P

Prot

• toc
• col
• l

25

• 1. AREF broadcasts SYNC
• 2. Ai’s reply concurrently

Heydariaan, Dabirian & Gnawali - University of Houston

AREF

T2

A2 A1 A3 A4

T1 T4 T3 T5 T6

Con Concurr rrent Se Self-Loc Localization

• n P

Prot

• toc
• col
• l

26

• 1. AREF broadcasts SYNC
• 2. Ai’s reply concurrently
• 3. Tags (T1 through T6)

receive all replies and measure AoA concurrently

Heydariaan, Dabirian & Gnawali - University of Houston

AREF

T2

A2 A1 A3 A4

T1 T4 T3 T5 T6

Ev Evaluation of Concurrent Ao AoA

• Sequential AoA as baseline
• Ideally should have similar accuracy

27

Sequential AoA Concurrent AoA Experimental Setup == ? Results Results

Heydariaan, Dabirian & Gnawali - University of Houston

Con Concurr rrent Ao AoA Expe Experimental Setup up

28

R1 R2 PDoA Node R2 R1 PDoA Node R2 R2

We change angle of one responder when the

• ther one is static. Dipole antenna performance

expected to degrade near extreme angles. We change distance of one responder when the other one is static. C-AoA performance expected to degrade at longer distances.

Heydariaan, Dabirian & Gnawali - University of Houston

Pe Performance of Concurrent Ao AoA fo for 2 anchors

29

Concurrency does not significantly affect AoA estimation in different angles (100 measurements per data point) Concurrency only affects AoA estimation in longer distances (100 measurements per data point)

Heydariaan, Dabirian & Gnawali - University of Houston

Pe Performance of Concurrent Ao AoA fo for 3+ anchors

30

Concurrency does not significantly affect AoA estimation when adding more anchors showing 100 measurements per box plot. Number of receiver tags are still unlimited.

Heydariaan, Dabirian & Gnawali - University of Houston

2D 2D Se Self-Loc Localization

• n w

with A Angle D Difference ce of

• f A

Arri rrival

31 Heydariaan, Dabirian & Gnawali - University of Houston

A1 A2 A3

2D 2D Se Self-Loc Localization

• n w

with A Angle D Difference ce of

• f A

Arri rrival

32 Heydariaan, Dabirian & Gnawali - University of Houston

A3 θ1,2 A1 A2 Angle difference between 1 pair of anchors. Possible locations are shown with 2 curves. Infinite possible locations.

2D 2D Se Self-Loc Localization

• n w

with A Angle D Difference ce of

• f A

Arri rrival

33 Heydariaan, Dabirian & Gnawali - University of Houston

θ1,2 A2 A3 A1 θ1,3 Angle difference between 2 pairs of anchors. Possible locations are shown with 4 curves. 2 possible locations.

2D 2D Se Self-Loc Localization

• n w

with A Angle D Difference ce of

• f A

Arri rrival

34 Heydariaan, Dabirian & Gnawali - University of Houston

θ1,2 A1 θ1,3 Angle difference between 3 pairs of anchors. Possible locations are shown with 6 curves. 1 possible location. θ2,3 A2 A3

2D 2D Se Self-Loc Localization

• n w

with A Angle D Difference ce of

• f A

Arri rrival

35 Heydariaan, Dabirian & Gnawali - University of Houston

θ1,2 A1 θ1,3 Angle difference between 3 pairs of anchors. Possible locations are shown with 6 curves. 1 possible location. θ2,3 A2 A3

& ', ( = *

+,-

(/-,+ − 1 /-,+)3 1 4 = argmin

;,<

&(', ()

AD ADoA Ch Challenges – Fr Front-ba back Ambi bigui guity

Heydariaan, Dabirian & Gnawali - University of Houston 36

θ1 θ2 θ1 θ2

!",$ = !" + !$ !",$ = !" − !$

AD ADoA Ch Challenges – Unknown Ti Tilting

Heydariaan, Dabirian & Gnawali - University of Houston 37

6 cases for tags inside the room in a 4-anchor setting. We run 6 optimizations and choose the answer with the least residual error.

Pe Performance of AD ADoA-ba based ed Algorithm hm

38

CDF of localization error, simulated for different noise levels. Sub-meter accuracy for noise level below 5° CDF of localization error, comparing AnguLoc with CTDoA in 2 experiments Static with 3000 points: 44.33% improvement Mobile with 200 points: 21.46% improvement

Heydariaan, Dabirian & Gnawali - University of Houston

Con Conclusion

• ns and Discussion
• n
• Scalability
• Anchors: Up to 5 nodes. We can make multiple groups of concurrent nodes
• Tags: Unlimited
• Efficiency: At least 4 times faster than than sequential AoA
• Accuracy (compared to CTDoA)
• 44.33% improvement for static nodes
• 21.46% improvement for mobile nodes
• Limitations
• Larger errors on the sides of dipole antennas
• Larger errors in longer distances (lower SNR)

39

Contact: milad@cs.uh.edu

Heydariaan, Dabirian & Gnawali - University of Houston

Back ckup Slides

40 Heydariaan, Dabirian & Gnawali - University of Houston

Be Benefits of

• f UWB

B Ra Radios

• s
• Accurate (10 cm)
• Long range (290 m)
• Low power consumption
• Potential for use in indoor and outdoor applications
• 3D localization
• Construction zone safety
• Mars exploration

41 Heydariaan, Dabirian & Gnawali - University of Houston

UW UWB Us Use e Cas ases es

42 https://www.firaconsortium.org/discover/use-cases Heydariaan, Dabirian & Gnawali - University of Houston

Loc Localization

• n T

Tech chnol

• log
• gies

43 https://www.decawave.com/technology1/ Heydariaan, Dabirian & Gnawali - University of Houston

ALOHA and TD TDMA in UWB-ba based ed Localization

• ALOHA

4 tags up to 4 packets per second each (total of 16 per second)

• TDMA

8 tags with 10 packets per second each (total of 80 per second) 7 anchors with total of 12 packets per second

44 Heydariaan, Dabirian & Gnawali - University of Houston

Con Concurr rrency Window and Cl Cloc

• ck Dri

rift

45

Longer response delay moves the responder peak due to larger clock skew. 23 ms additional delay causes clock skew of 40 ns.

Response Delay = 800 μs

First path For R1 First path For R2

Response Delay = 23800 μs

First path For R1 First path For R2

Δt < 1016 ns

Concurrency window. The time window for multiple responses to arrive concurrently. For DW1000, theoretically 1016 ns.

Large clock skew can break concurrency

Heydariaan, Dabirian & Gnawali - University of Houston

Ti Time-ba based ed Ao AoA Es Estimation n Usi sing ng DW1 W1000

• Path difference: ! = # × sin (
• Goal: Precision of 5°
• With one radio using CIR
• Resolution of 1001.6 !/ or 0.30048 3
• Antenna sparation: # = 4.54467

89: ;° = 3.497 3

• With two radios using ToA
• Resolution of 15.6 !/ or 0.00469 3 -> # = 4.446>?

89: ;° = 0.053 3

• Precision of 333.3 !/ or 0.1 3 -> # =

4.@ 89: ;° = 1.147 3

46 Heydariaan, Dabirian & Gnawali - University of Houston

Re Response Position Modulation

47

Concurrent responses separated by 128 ns Initiator Ri

R e q u e s t R e s p

• n

s e i

!"# + !% &'# &"#

• &"# = &'# + !"# + !%
• !% = 128× - − 1
• - = /012 -1

Heydariaan, Dabirian & Gnawali - University of Houston

Se Search ch a and Su Subtract ct ( (SS) SS) A Algori

• rithm
• 1. Divide CIR into multiple chunks of 128 ns
• 2. Upsample each chunk using FFT with upsampling factor of L=30
• 3. Normalize upsampled CIR chunk
• 4. Cross-correlate each chunk with a signal template and output the

index with maximum correlation

• 5. Consider the index as a peak if value exceeds a noise threshold of

η = 12×&'()*+

48 Heydariaan, Dabirian & Gnawali - University of Houston

Con Concurr rrent Ao AoA Al Algorithm

49 Heydariaan, Dabirian & Gnawali - University of Houston

Loc Localization

• n w

with A Angle D Difference ce of

• f A

Arri rrival

50 Heydariaan, Dabirian & Gnawali - University of Houston

!",$ = cos)* +," ⋅ +,$ +," ⋅ +,$ = cos)* . − ." . − .$ + 1 − 1" 1 − 1$ . − ." 2 + 1 − 1" 2 . − .$

2 + 1 − 1$ 2

3 ., 1 = 4

$5"

(!",$ − 7 !",$)2 9 + = argmin

@,A

3(., 1)