LocAP: Autonomous Millimeter Accurate Mapping of WiFi Infrastructure · 2020. 3. 24. ·...
Transcript of LocAP: Autonomous Millimeter Accurate Mapping of WiFi Infrastructure · 2020. 3. 24. ·...
LocAP: Autonomous Millimeter Accurate Mapping of WiFi Infrastructure
Roshan Ayyalasomayajula, Aditya Arun, Chenfeng Wu, Shrivatsan Rajagopalan, Shreya Ganesaraman, Aravind Seetharaman, Ish Kumar Jain, Dinesh Bharadia
http://wcsng.ucsd.edu/locap/
Indoor Navigation: Applications
2
Indoor Navigation: Mapping
3
Mapping
Indoor Navigation: Mapping
4
Mapping
Indoor Navigation: Localization
5
Localization
Indoor Navigation: Localization
6
DLoc [Mobicom'20]
MonoLoco [MobiSys'18]
Chronos [NSDI'16]
ToneTrack [Mobicom'15]
SpotFi [Sigcomm'15]
ArrayTrack [NSDI'13]
EZ [Mobicom'10]
Horus [MobiSys'05]
RADAR [Infocomm'00] Localization
Indoor Navigation: Localization
7
DLoc [Mobicom'20]
MonoLoco [MobiSys'18]
Chronos [NSDI'16]
ToneTrack [Mobicom'15]
SpotFi [Sigcomm'15]
ArrayTrack [NSDI'13]
EZ [Mobicom'10]
Horus [MobiSys'05]
RADAR [Infocomm'00]
RSSI
~10 meter accuracyLocalization
Indoor Navigation: Localization
8
DLoc [Mobicom'20]
MonoLoco [MobiSys'18]
Chronos [NSDI'16]
ToneTrack [Mobicom'15]
SpotFi [Sigcomm'15]
ArrayTrack [NSDI'13]
EZ [Mobicom'10]
Horus [MobiSys'05]
RADAR [Infocomm'00]
RSSI
~10 meter accuracy
CSI
<1 meter accuracy
Localization
Indoor Navigation: What is missing?
9
Mapping Localization
Indoor Navigation: What is missing?
10
Mapping Localization
?
Indoor Navigation: What is missing?
11
Indoor Navigation: What is missing?
12
Anchor Locations are not accurately known in physical space
Contributions of LocAP: Reverse Localization
13
Contributions of LocAP: Reverse Localization
❖ First to establish requirements for Reverse Localizing the anchor points
◦ Needs millimetre accurate reverse localization
❖ Developed a novel algorithm that accurately reverse localizes the anchor points
❖ Deployed it on an autonomous system
❖ Demonstrated it in a real world that performs 50x times better than state-of-the-art
14
Contributions of LocAP: Reverse Localization
❖ First to establish requirements for Reverse Localizing the anchor points
◦ Needs millimetre accurate reverse localization
❖ Developed a novel algorithm that accurately reverse localizes the anchor points
❖ Deployed it on an autonomous system
❖ Demonstrated it in a real world that performs 50x times better than state-of-the-art
15
Primer: WiFi-based User Localization
16
Primer: WiFi-based User Localization
17
Primer: WiFi-based User Localization
18
Primer: WiFi-based User Localization
19
Primer: WiFi-based User Localization
20
Primer: WiFi-based User Localization
21
𝑑sin(𝜃)𝜃
𝑑
Primer: WiFi-based User Localization
22
𝑑sin(𝜃)𝜃
𝑑
What happens when there are errors in Anchor Attributes
Errors in Anchor Attributes mis-locates the User
23
Errors in Anchor Attributes mis-locates the User
24
Location Error
Errors in Anchor Attributes mis-locates the User
25
Location Error Orientation Error
Errors in Anchor Attributes mis-locates the User
26
Location Error Orientation Error Antenna Spacing Error
Quantifying Reverse Localization Requirements
27
0 8
10
Y (
m)
X (m)
Requirement for Deployment Orientation
28
0 5 10 15 20
Anchor Orientation Error (degree)
0
0.5
1
1.5
2
2.5
3
Use
r Localiz
ation E
rror
(m)
Requirement for Deployment Orientation
29
0 5 10 15 20
Anchor Orientation Error (degree)
0
0.5
1
1.5
2
2.5
3
Use
r Localiz
ation E
rror
(m)
Requirement for Deployment Orientation
30
0 5 10 15 20
Anchor Orientation Error (degree)
0
0.5
1
1.5
2
2.5
3
Requirement: 7o
Target: 50cm
Use
r Localiz
ation E
rror
(m)
Stringent Requirements for Reverse Localization
31
❖ Requirement of median error →
Location Error Orientation Error Antenna Spacing Error
Stringent Requirements for Reverse Localization
32
❖ Requirement of median error →
< 25 cm < 7° < 5 mm
Location Error Orientation Error Antenna Spacing ErrorMillimeter requirements make manual mapping impossible
LocAP Deploys an Autonomous System
33
X
Y
LocAP Deploys an Autonomous System
34
X
Y
LocAP Deploys an Autonomous System
35
X
Y
Anchor Localization
36
X
Y
Anchor Localization
❖ Locating Anchor is equivalent to Locating any one of the Antenna – First Antenna
❖ Existing User Localization algorithms applied in reverse
❖ 100s of bot locations to triangulate the First Antenna
37
X
Y
Anchor Localization
❖ Locating Anchor is equivalent to Locating any one of the Antenna – First Antenna
❖ Existing User Localization algorithms applied in reverse
❖ 100s of bot locations to triangulate the First Antenna
38
X
Y
Anchor Localization
❖ Locating Anchor is equivalent to Locating any one of the Antenna – First Antenna
❖ Existing User Localization algorithms applied in reverse
❖ 100s of bot locations to triangulate the First Antenna
39
X
Y
𝛽1
Anchor Localization
❖ Locating Anchor is equivalent to Locating any one of the Antenna – First Antenna
❖ Existing User Localization algorithms applied in reverse
❖ 100s of bot locations to triangulate the First Antenna
40
X
Y
𝛽1
Anchor Localization: Handling Bot Errors
❖ A few bot locations can be erroneous
❖ We propose a novel Visual Sensors based Confidence Metric
❖ We neglect Locations with less Confidence
41
XX
Y
𝛽1𝛽2 𝛽3
Anchor Localization: Handling Bot Errors
❖ A few bot locations can be erroneous
❖ We propose a novel Visual Sensors based Confidence Metric
❖ We neglect Locations with less Confidence
42
XX
Y
𝛽1𝛽2 𝛽3
Achieved cm-accurate anchor locations
Can we locate each antenna in a similar way?
43
0
0.04
0.08
0.12
0.16
0.2
4 8 16 32 64
Med
ian
Lo
caliz
atio
n E
rro
r (m
)
Number of bot locations
Can we locate each antenna in a similar way?
44
0
0.04
0.08
0.12
0.16
0.2
4 8 16 32 64
Med
ian
Lo
caliz
atio
n E
rro
r (m
)
Number of bot locations
Can we locate each antenna in a similar way?
45
Localization error saturates to 3cm
Does not meet mm-accurate requirement for antenna separation
How to bridge this gap?
46
How to bridge this gap?
❖ Maximum Bandwidth of WiFi is 160MHz = ~2 meter resolution
❖ First key insight of LocAP is that the carrier frequency fc of these signals is 5GHz ~mm accurate resolution
47
How to bridge this gap?
❖ Maximum Bandwidth of WiFi is 160MHz = ~2 meter resolution
❖ First key insight of LocAP is that the carrier frequency fc of these signals is 5GHz ~mm accurate resolution
48
How to bridge this gap?
❖ Maximum Bandwidth of WiFi is 160MHz = ~2 meter resolution
❖ First key insight of LocAP is that the carrier frequency fc of these signals is 5GHz ~mm accurate resolution
49
360° ≡ 6cm
6° ≡ 1mm
Am
plit
ud
e
Phase ≡ Distance
Phase Difference at fc
50
Phase Difference at fc
51
Phase Difference at fc
52
Phase Difference at fc
53
𝜃
𝑑
𝑑sin(𝜃)
Relative antenna localization can achieve millimeter accuracy
Relative Antenna Localization: LocAP
❖ LocAP locates the rest of anchor’s antenna relative to First Antenna
❖ Estimate
54
𝑑
x
y
𝑑 = Antenna SeparationΨ = Deployment Orientation
Relative Antenna Localization: LocAP
55
Ψ
𝑑
x
y
Relative Antenna Localization: LocAP
❖ Leverage phase difference across two antennas
❖ Recall, we measured direction of First Antenna from the bot (𝛽1)
56
Ψ
𝑑
x
y
Relative Antenna Localization: LocAP
❖ Leverage phase difference across two antennas
❖ Recall, we measured direction of First Antenna from the bot (𝛽1)
57
Ψ
𝑑
x
y
𝛽1
∆∅ =2𝜋
𝜆𝑑sin(𝜃1)
𝜃1
Relative Antenna Localization: LocAP
❖ Leverage phase difference across two antennas
❖ Recall, we measured direction of First Antenna from the bot (𝛽1)
58
Ψ
𝑑
x
y
𝛽1
∆∅ =2𝜋
𝜆𝑑sin(𝜃1)
𝜃1
Relative Antenna Localization: LocAP
❖ Leverage phase difference across two antennas
❖ Recall, we measured direction of First Antenna from the bot (𝛽1)
59
Ψ
𝑑
x
y
𝛽1
∆∅ =2𝜋
𝜆𝑑sin(𝜃1)
𝜃1
𝜃1 = Ψ− 𝛽1 + 90°
𝛽1
𝜃1→ Ψ
known
Relative Antenna Localization: LocAP
❖ Leverage phase difference across two antennas
❖ Recall, we measured direction of First Antenna from the bot (𝛽1)
60
Ψ
𝑑
x
y
𝛽1
∆∅ =2𝜋
𝜆𝑑sin(𝜃1)
𝜃1
𝜃1 = Ψ− 𝛽1 + 90°
𝛽1
𝜃1→ Ψ
knownOnly one equation, but two unknowns (𝑑, 𝜃1)
Leverage from 100s of bot locations
61
Ψ
𝑑
x
y
𝛽1𝜃1
- /2 - /4 0 /4 /2
-
- /2
0
/2
Ph
ase
dif
fere
nce
∆∅
Direction of bot 𝜃1 (radians)
Leverage from 100s of bot locations
62
Ψ
𝑑
x
y
𝛽1𝜃1
- /2 - /4 0 /4 /2
-
- /2
0
/2
Ph
ase
dif
fere
nce
∆∅
Direction of bot 𝜃1 (radians)
Leverage from 100s of bot locations
63
Ψ
𝑑
x
y
𝛽1𝜃1
- /2 - /4 0 /4 /2
-
- /2
0
/2
Ph
ase
dif
fere
nce
∆∅
Direction of bot 𝜃1 (radians)
=0.5𝑑/𝜆
Leverage from 100s of bot locations
64
Ψ
𝑑
x
y
𝛽1𝜃1
- /2 - /4 0 /4 /2
-
- /2
0
/2
Ph
ase
dif
fere
nce
∆∅
Direction of bot 𝜃1 (radians)
=0.5𝑑/𝜆=1𝑑/𝜆
Differential Phase Difference
65
Ψ
𝑑
x
y
𝜃1
❖ Differential Phase Difference from two nearby bot locations
∆∅1∆∅2
𝜃2
Differential Phase Difference
66
Ψ
𝑑
x
y
𝜃1
❖ Differential Phase Difference from two nearby bot locations
∆∅1∆∅2
𝑑∆∅
𝑑𝜃=∆∅2 − ∆∅1𝜃2 −𝜃1
𝜃2
LocAP uses Differential Phase Difference
67
Dif
fere
nti
al P
has
e D
iffe
ren
ce d∆∅/d𝜃
Direction of bot 𝜃1 (radians)
0 /4 /2 3 /4-10
-5
0
5
10
Achieves mm-accurate Antenna separation and < 7° Orientation
=1𝑑/𝜆
- /2 - /4 0 /4 /2
-
- /2
0
/2
Ph
ase
dif
fere
nce
∆∅
Direction of bot 𝜃1 (radians)
=1𝑑/𝜆
LocAP uses Differential Phase Difference
68
Dif
fere
nti
al P
has
e D
iffe
ren
ce d∆∅/d𝜃
Direction of bot 𝜃1 (radians)
0 /4 /2 3 /4-10
-5
0
5
10
Achieves mm-accurate Antenna separation and < 7° Orientation
=1𝑑/𝜆
- /2 - /4 0 /4 /2
-
- /2
0
/2
Ph
ase
dif
fere
nce
∆∅
Direction of bot 𝜃1 (radians)
=1𝑑/𝜆
LocAP uses Differential Phase Difference
69
Dif
fere
nti
al P
has
e D
iffe
ren
ce d∆∅/d𝜃
Direction of bot 𝜃1 (radians)
0 /4 /2 3 /4-10
-5
0
5
10
Achieves mm-accurate Antenna separation and < 7° Orientation
=1𝑑/𝜆
- /2 - /4 0 /4 /2
-
- /2
0
/2
Ph
ase
dif
fere
nce
∆∅
Direction of bot 𝜃1 (radians)
=1𝑑/𝜆
Evaluation Setup
❖ COTS Quantenna chipsets
❖ 8 different anchors in 2000 sq. ft.
❖ 5 different antenna separations
❖ 7 different orientations
❖ Most popular anchor spacing of both linear and rectangular arrays
70
Antenna Separation Prediction Results
71
10 -3 10 -2 10 -1 10 0
Error in Relative Antenna Separation (m)
00.10.20.30.40.50.60.70.80.9
1
CD
F
Antenna Separation Prediction Results
72
10 -3 10 -2 10 -1 10 0
Error in Relative Antenna Separation (m)
00.10.20.30.40.50.60.70.80.9
1
CD
F
LocAPSpotFi
Requirement
Antenna Separation Prediction Results
73
10 -3 10 -2 10 -1 10 0
Error in Relative Antenna Separation (m)
00.10.20.30.40.50.60.70.80.9
1
CD
F
LocAPSpotFi
Requirement
Antenna Separation Prediction Results
74
10 -3 10 -2 10 -1 10 0
Error in Relative Antenna Separation (m)
00.10.20.30.40.50.60.70.80.9
1
CD
F
LocAPSpotFi
Requirement
50x improvement
Deployment Orientation Prediction Results
75
0 10 20 30 40 50 60Orientation Error( °)
0
0.10.20.3
0.4
0.50.60.70.8
0.91
CD
F
Deployment Orientation Prediction Results
76
0 10 20 30 40 50 60Orientation Error( °)
0
0.10.20.3
0.4
0.50.60.70.8
0.91
CD
F
SpotFi
LocAP
Requirement
Deployment Orientation Prediction Results
77
0 10 20 30 40 50 60Orientation Error( °)
0
0.10.20.3
0.4
0.50.60.70.8
0.91
CD
F
SpotFi
LocAP
Requirement
8x improvement
Conclusion and Future Work
❖ First Work to define the requirements for Reverse Localization
❖ Demonstrated millimeter accurate Reverse Localization
78
http://wcsng.ucsd.edu/locap/