Noname manuscript No.

(will be inserted by the editor)

Implementation of the Generic Equation (25) on the

Journal Paper ”Use of Flip Ambiguity Probabilities in
Robust Sensor Network Localization”

Anushiya A Kannan · Barış Fidan ·

Guoqiang Mao

Received: date / Accepted: date

This note describes the implementation methodology of Equation (25) of the Jounal
Paper ”Use of Flip Ambiguity Probabilities in Robust Sensor Network Localization”.
Sample values of each of dAD , dBD and dCD are taken to span the distance range
[0, R] at a step size of R/N . Integration is implemented using the MATLAB function
trapz. In choice of N there is a trade-off between accuracy and computational cost. In
our case, N was chosen to be 100.

1 function NumericalCalculationOfFlip(xy,R,N,sigma)
2 %xy − true coordinates of A,B,C and D
3 %transmission range R
4 %Number of steps N
5 %Standard deviation of Gaussian noise sigma
6
7 % create distance matrix containing possible values
8 % x axis representing measured distance dAD,
9 % y axis representing measured distance dBD, and
10 % z axis representing measured distance dCD
11 Delta=R/N;
12 d=Delta:Delta:R;
13 for i=1:N,

The work was supported by National ICT Australia-NICTA. NICTA is funded by the Aus-
tralian Government as represented by the Department of Broadband, Communications and the
Digital Economy and the Australian Research Council through the ICT Centre of Excellence
program.

Anushiya A Kannan
School of Electrical and Information Engineering, University of Sydney, NSW 2006, Australia.
Tel.: +61-2-88122431
Fax: +61-2-88122431
E-mail: [email protected]
Barış Fidan
University of Waterloo, Ontario, Canada
E-mail: [email protected]
Guoqiang Mao
The University of Sydney and National ICT Australia
E-mail: [email protected]
2

14 dAD(:,:,i)=ones(N,1)*d;
15 dBD(:,:,i)=d'*ones(1,N);
16 dCD(:,:,i)=i*Delta*ones(N,N);
17 end
18
19 %Create Identity Matrices
20 I zeta AB=Calculate I zeta XY(dAD,dBD,xy,R);
21
22 xy changedOrder=[xy(2,:);xy(3,:);xy(1,:);xy(4,:)];
23 I zeta BC=Calculate I zeta XY(dBD,dCD,xy changedOrder,R);
24
25 xy changedOrder=[xy(3,:);xy(1,:);xy(2,:);xy(4,:)];
26 I zeta AC=Calculate I zeta XY(dAD,dCD,xy changedOrder,R)
27 % Define normalised Gaussian transfer functions
28 measured dCD=shiftdim((dCD(1,1,:)),2)';
29 CD=norm(xy(3,:)−xy(4,:));
30 TF=1/(sigma*sqrt(2*pi))*exp(−(dCD−CD).ˆ2/(2*sigmaˆ2));
31 NF=cdf('normal',R,CD,sigma)−cdf('normal',0,CD,sigma);
32 YCD=TF/NF; clear TF NF
33
34 measured dBD=dBD(:,1,1)';
35 BD=norm(xy(2,:)−xy(4,:));
36 TF=1/(sigma*sqrt(2*pi))*exp(−(dBD(:,:,1)−BD).ˆ2/(2*sigmaˆ2));
37 NF=cdf('normal',R,BD,sigma)−cdf('normal',0,BD,sigma);
38 YBD=TF/NF;
39
40 measured dAD=dAD(1,:,1);
41 AD=norm(xy(1,:)−xy(4,:));
42 TF=1/(sigma*sqrt(2*pi))*exp(−(measured dAD−AD).ˆ2/(2*sigmaˆ2));
43 NF=cdf('normal',R,AD,sigma)−cdf('normal',0,AD,sigma);
44 YAD=TF/NF;
45
46 % Calculate the analytical probability P(zeta AB BC AC AD/ABCD)
47 Y=YCD.* I AB.* I BC.* I AC;
48 for i=1:N,
49 for j=1:N,
50 YC(i,j)=trapz(measured dCD,shiftdim((Y(i,j,:)),2)');
51 end
52 end
53 Y=YBD.*YC;
54 for i=1:N,
55 YB(1,i)=trapz(measured dBD,Y(:,i)');
56 end
57 Y=YAD.*YB;
58 P zeta AB BC AC ABCD=trapz(measured dAD,Y);
59 end

0.1 Function Calculating IζXY

In order to implement Equation (25), it is necessary to calculate the three dimensional

indicator function IζXY which is set when there is a flipped realization with respect to
line XY .
This function uses two other functions to calculate the corresponding λC and to
AB
create an indicator function δZD which is set to one when RD 1
⊂ HC and reset
AB
to zero when RD1 ⊂ HC . These functions are explained in Subsections 0.2 and 0.3
respectively.
 3

This function ignores the negligible flips (See Section 6 of Journal Paper) by not
calculating λC for the cases of non-intersecting circles C(p(A), dAD ), C(p(B), dBD ) and
setting them to be zero.
The matlab code of this function is as follows.

1 function I zeta XY=Calculate I zeta XY(dXD,dYD,xy,R)

2 %True distance XY between points X and Y
3 XY=norm(xy(1,:)−xy(2,:));
4 %Create a 3D dimensional condition matrix indicating
5 %whether the circles centered at X and Y intersect.
6 CirclesIntersect=(abs(dXD−XY)≤dYD).*(dYD≤min(dXD+XY,R));
7 %Calculate Lambda C
8 Lamda C=CirclesIntersect.* Calculate lambda C(dYD,dXD,xy);
9 %create Indicator indicating whether
10 %Z and D are on same side of XY or not
11 sameSide=decide ZandDonSameSideOfXY(xy);
12 %make it a 3−dimensional matrix
13 ∆ ZD=sameSide.*ones(N,N,N);
14 I zeta XY=((∆ ZD. *Lamda C)≤dZD).*(dZD≤(min(∆ ZD *R+Lamda C,R)));
15 end

0.2 Function Calculating λC

From Proposition A.1 of the Journal paper, we know that

kp(C) − pAB
C (D)k + kp(C) − p
AB
(D)k
C
λC = (0.1)
2

In order to calculate, kp(C)−pAB

C (D)k and kp(C)−p
AB
(D)k, lets extract △p(A)p(B)p(C),
C
△p(A)p(B)pAB AB
C (D) and △p(A)p(B)pC (D) as shown in Figure 0.1. Since p(C) is a ver-

pCAB (D)

hD hC
p (C )
lD lC
p(A)
hC `hC p(B)
`

p(M ) p(N)
lC
lD hD

pCAB ( D)

Fig. 0.1 Closer look at △p(A)p(B)p(C), △p(A)p(B)pAB AB

C (D) and △p(A)p(B)pC (D) from
Figure A.1 of the Journal paper.

tex in both right angled triangles p(A)p(M )p(C) and p(B)p(M )p(C), using pythagoras
theorem lC can be calculated as

kp(A) − p(C)k2 − kp(B) − p(C)k2 + kp(A) − p(B)k2

lC = (0.2)
2kp(A) − p(B)k
4

Like wise, lD can be calculated as

2 2 2
kp(A) − pAB AB
C (D)k − kp(B) − pC (D)k + kp(A) − p(B)k
lD =
2kp(A) − p(B)k
2 2
dAD − dBD + kp(A) − p(B)k2
= (0.3)
2kp(A) − p(B)k

Since hC is the height of the △p(A)p(B)p(C) and hD is the height of the △p(A)p(B)pAB
C (D)
AB
(equivalently can consider △p(A)p(B)pC (D) as well), they can be written as

2K△p(A)p(B)p(C)
hC = (0.4)
kp(A) − p(B)k

2K△p(A)p(B)pAB (D)
C
hD = (0.5)
kp(A) − p(B)k
where K△p(A)p(B)p(C) and K△p(A)p(B)pAB (D) are the area of △p(A)p(B)p(C) and
C
△p(A)p(B)pAB
C (D) respectively and can be easily calculated using Heron’s formula.
Thus kp(C) − pAB
C (D)k and kp(C) − p
AB
(D)k can be calculated as
C

kp(C) − pAB
p
C (D)k = (lD − lC )2 + (hD − hC )2
kp(C) − pAB
p
C
(D)k = (lD − lC )2 + (hD + hC )2

and substituted in (0.1) to calculate λC .

The matlab code of this function is as follows.

1 function lambda C=Calculate lambda C(dBD,dAD,xy)

2 AB=norm(xy(1,:)−xy(2,:));
3 BC=norm(xy(2,:)−xy(3,:));
4 AC=norm(xy(1,:)−xy(3,:));
5
6 %l C=(ACˆ2−BCˆ2+ABˆ2)/(2*AB);
7 %l D=(dAD.ˆ2−dBD.ˆ2+ABˆ2)/(2*AB);
8 l=((ACˆ2−BCˆ2)−(dAD.ˆ2−dBD.ˆ2))/(2*AB);
9
10 sABC=(AB+BC+AC)/2;
11 kABC=sqrt(sABC*(sABC−AB)*(sABC−BC)*(sABC−AC));
12
13 sABD=(dAD+dBD+AB)/2;
14 kABD=sqrt(sABD.*(sABD−AB).*(sABD−dBD).*(sABD−dAD));
15
16 %h C=2*kABC/AB;
17 %h D=2*kABD./AB;
18 h D1=2*(kABC−kABD)/AB;
19 h D2=2*(kABC+kABD)/AB;
20
21 CD1=sqrt(l.ˆ2+h D1.ˆ2);
22 CD2=sqrt(l.ˆ2+h D2.ˆ2);
23
24 lambda C=(x CD1+x CD2)/2;
25
26 clear AB BC AC l sABC kABC sABD kABD h D1 h D2 CD1 CD2;
27 end
 5

0.3 Function Deciding Whether Z and D Are On Same Side Of XY

The is a simple matlab code of this function using cross product and checking the signs
to make a decision.

1 function sameSide=decide ZandDonSameSideOfXY(xy)

2 cZ = cross([xy(2,:)−xy(1,:) 0],[xy(3,:)−xy(1,:) 0]);
3 cD = cross([xy(2,:)−xy(1,:) 0],[xy(4,:)−xy(1,:) 0]);
4 if (sign(cZ(3))*sign(cD(3))==−1)
5 sameSide=0;
6 else %(sign(cZ(3))*sign(cD(3))==1)
7 sameSide=1;
8 end
9 end