Presentation made by Alireza Ghods and given by Dr. Stefano Severi at CCP Workshop co-located with IEEE Intelligent Vehicles Conference, 19th June 2016 Gothenburg (Sweden)
(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7
Localization in V2X Communication Networks
1. Localization in V2X Communication
Networks
Alireza Ghods, Stefano Severi, Giuseppe Abreu
s.severi@jacobs-university.de
School of Engineering & Science - Jacobs University Bremen (GERMANY)
June 19, 2016
6. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
Model and Some Notation
A network of with N vehicles in η-dimensional space
[θ1, . . . , θnT
, anT+1 , . . . , aN ]
dij θi − θj = θi − θj, θi − θj
First nT vehicles (targets) have unknown positions
K = N − nT of the remaining vehicles (anchors) in the
periphery have estimated positions (subject to errors)
Anchor location errors described by covariance matrix Σk
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 6/21
7. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
Ranging Model
For each j-th hop:
˜dj ∼ (dj, σ2
j )
σ2
j σ2
0 ·
dj
d0
α
where α ≥ 0 is pathloss factor and σ2
0 is the ranging
variance at a reference distance d0.
For a complete multihop path:
¯dk
nk
˜dj,
¯σ2
k
nk
σ2
j .
where nk is number of hops
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 7/21
8. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
Fundamental Error Limit
The FIM and the MSE
The covariance matrix associated with the location
estimate of a single target ˆθ is
Ωθ E (ˆθ − θ)(ˆθ − θ)T
The Cramér-Rao lower bound (CRLB) relates Ωθ to the
Fisher Information Matrix
Ωθ F−1
θ
Fθ ∝ N( ¯dk, σk)
Anchor uncertainty not considered!
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 8/21
9. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
Constructing the FIM
Standard: element-wise derivative of log-likelihood function
Alternative: sum of products of information vectors
Fθ =
k∈K
ukuT
k
where k is the anchor’s index and the information vector
is
uk =
∂ ak − θ
∂θ
Fk =
1
¯dk
[(xak
− xθ), (yak
− yθ)]T
Fk
Fk =
1
¯σ2
k
1 +
α2 σ2
0
2 dα
0
( ak − θ )α−2
in which Fk is the information intensity
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 9/21
10. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
FIM with Anchor Uncertainty
Augmented Parameter Vector
Augmented parameter vector θ
Θ = θT
, aT
1, aT
2, · · · , aT
K
T
Hence
ΩΘ E ( ˆΘ − Θ)( ˆΘ − Θ)T
ΩΘ F−1
Θ
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 10/21
11. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
FIM with Anchor Uncertainty
Augmented Information Vectors
The FIM of Θ can be approximated by (Bayesian rule)
FΘ ≈ FM + FΣ,
where FM accounts for the multi hop ranging, while FΣ
accounts for anchor uncertainty
The approximation holds whenever θ − ak tr(Σk), ∀ k
The extended information vector is then
vk
∂ ak − θ
∂Θ
=
1
√
Fk
uT
k, 01×η·(k−1), −uT
k, 01×η·(K−k)
T
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 11/21
12. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
Decomposing the Augmented FIM
The Multihop Component
The multi hop component of FΘ becomes
FM =
K
k=1
vkvT
k =
A BT
B C
,
where
A
K
k=1
ukuT
k
BT
−u1uT
1, · · · , −uKuT
K
C diag u1uT
1, · · · , uKuT
K
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 12/21
13. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
Decomposing the Augmented FIM
Adding the Anchor Uncertainty Component
The anchor uncertainty component FΘ is
FΣ
0η×η 0η×ηK
0Kη×η Σ−1
where Σ diag (Σ1, · · · , ΣK).
Finally
FΘ ≈
A BT
B C + Σ−1 ,
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 13/21
14. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
Relevant Minor: Schur Complement
Taking η × η Schur complement of FΘ
F∗
θ = A − BT
Σ−1
+ C
−1
B,
=
K
k=1
ukuT
k −
K
k=1
ukuT
k Σ−1
k + ukuT
k
−1
ukuT
k,
=
K
k=1
uk 1 − uT
k Σ−1
k + ukuT
k
−1
uk uT
k,
=
K
k=1
uk 1 − uT
k Σk −
ΣkukuT
kΣk
1 + uT
kΣkuk
uk uT
k,
=
K
k=1
uk 1 − uT
kΣkuk +
uT
kΣkukuT
kΣkuk
1 + uT
kΣkuk
uT
k,
where we used the Sherman-Morrison formula
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 14/21
15. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
Relevant Minor: Schur Complement
Simplifying further...
F∗
θ =
K
k=1
uk 1 − uT
kΣkuk +
uT
kΣkukuT
kΣkuk
1 + uT
kΣkuk
uT
k,
=
K
k=1
uk 1 − νk +
ν2
k
1 + νk
uT
k,
=
K
k=1
1
1 + νk
ukuT
k,
where νk uT
kΣkuk
Anchor uncertainty appears as a
reduction of information intensity
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 15/21
16. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
Some Results ...
One-dimensional and two-dimensional scenarios
considered
Road: 500 meters long, 10 wide
Only vehicles at borders can self-localize via GPS
Neighborhood set: dij ≤ 70 meters
How well GPS location estimates propagate through the
network
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 16/21
17. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
0 50 100 150 200 250 300 350 400 450 500
0
0.5
1
1.5
2
2.5
3
3.5
Monodimensional Scenario
Performance for different GPS errors, SNR = 5dB
ErrorStandarDeviationε
Road Length [m]
GPS Σ = 0.9
GPS Σ = 0.5
No GPS Error
Anchor Vehicles
Selected Targets
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 17/21
18. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
0 50 100 150 200 250 300 350 400 450 500
0
1
2
3
4
5
6
Monodimensional Scenario
Performance for different SNR
ErrorStandarDeviationε
Road Length [m]
SNR = 0 dB
SNR = 5 dB
SNR = 10 dB
Anchor Vehicles
Selected Targets
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 18/21
19. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
0 50 100 150 200 250 300 350 400 450 500
0
1
2
3
4
5
6
7
8
9
10
Bidimensional Scenario
Error Bounds on x-Dimension for Selected Targets with SNR = 5 dB
RoadWidth[m]
Road Length [m]
Anchors Vehicles
Target Vehicles
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 19/21
20. Typical Dense
Urban
Environment
Cooperative
Network
Localization
Model and Notation
Ranging Model
FIM Formulation
Anchor Uncertainty
Results
Why the Huge Errors in Y-Axis
In 2D the covariance matrix is
Ω∗
θ =
σ2
x σxy
σxy σ2
y
From that, error ellipsis with diameters
λx
1
2
σ2
x + σ2
y − (σ2
x − σ2
y )2 + 4σ2
xy
λy
1
2
σ2
x + σ2
y + (σ2
x − σ2
y )2 + 4σ2
xy
A numerical example:
θ =
464.0172
7.1399
Xa =
0 500.0000
2.5000 2.5000
F =
1.1425 0.0010
0.0010 0.0021
Ω =
0.8757 −0.4230
−0.4230 479.9479
CCP Workshop 2016 Localization in V2X Communication Networks June 19, 2016 20/21