Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224314528

Modified Taylor Series Expansion Based Positioning Algorithms

Conference Paper · June 2008


DOI: 10.1109/VETECS.2008.582 · Source: IEEE Xplore

CITATIONS READS
5 261

3 authors:

Kegen Yu Y. Jay Guo


UNSW Sydney University of Technology Sydney
73 PUBLICATIONS   1,622 CITATIONS    410 PUBLICATIONS   4,624 CITATIONS   

SEE PROFILE SEE PROFILE

Ian James Oppermann


SIRCA Technology
128 PUBLICATIONS   2,702 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Geodetic Time Series Analysis View project

Algorithms and Stochastic models for Positioning, Tracking and Navigation View project

All content following this page was uploaded by Ian James Oppermann on 23 May 2014.

The user has requested enhancement of the downloaded file.


Modified Taylor Series Expansion Based
Positioning Algorithms
Kegen Yu and Y. Jay Guo Ian Oppermann
Wireless Technologies Laboratory Centre for Wireless Communications
CSIRO ICT Centre, Australia University of Oulu, Finland

Abstract—In this paper, we propose a modified two stage until a pre-specified criterion is satisfied. This method usually
Taylor series (TS) method for position estimation in a 3-D requires an initial position estimate close to the actual position
environment when either the time-difference-of-arrival (TDOA) and the convergence may not be satisfactory.
or the distance measurements are available. It is aimed to
improve the convergence performance of the traditional Taylor In this paper we focus on the TS method that provides a
series method. Simulation results demonstrate that the modified good tradeoff between complexity and accuracy. Based on the
TS method can improve the position estimation convergence observation that the traditional TS method performs poorly in
considerably. terms of convergence in some circumstances, we are motivated
Index Terms—Modified Taylor-series method, iterative position to propose a modified two stage Taylor series method to
estimation, TDOA, round-trip-time.
improve the performance of the original TS method. This is
I. I NTRODUCTION achieved by exploiting the second-order information which is
included in the third term in the series. The modified TS
In radio positioning1 , position estimation algorithms may be algorithm is derived when either TDOA measurements or
broken into two broad categories: iterative and non-iterative range measurements are provided.
methods for both cellular systems and ad hoc networks. The remainder of the paper is organized as follows. Sec-
A variety of non-iterative algorithms have been developed tion II derives the modified Taylor series method when
for position estimation. The most straightforward one is the the TDOA measurements are available, whereas Section III
direct method [1–3] which directly solves a set of simul- presents the modified Taylor series method based on the range
taneous equations with four anchors for 3-D positioning measurements. Section IV evaluate the proposed modified TS
based on time-difference-of-arrival (TDOA) measurements. method through simulation, and conclusions are drawn in
This method, however, may not effectively exploit extra Section V.
measurements to improve position accuracy. The spherical-
interpolation method [4] and related approaches [5, 6] were II. M ODIFIED TS A LGORITHM BASED ON TDOA
developed to exploit extra measurements. To approach optimal M EASUREMENTS
estimation, the two-stage maximum likelihood approach was
Without loss of generality, let us consider positioning in
considered [7] and the linear-correction least square (LS)
a wireless sensor network (WSN)2 . In a WSN, there are
approach was considered in [8]. When using range mea-
two types of nodes: anchor nodes and lower cost ordinary
surements, the standard least-squares approach is commonly
sensor nodes3 . Anchor nodes have higher complexity and
exploited [9–11]
computational capability and the positions of the anchor nodes
Two iterative methods are often employed for position
are assumed to be known in this paper. Ordinary sensor
estimation. One is the Taylor series method [7, 12–16] and the
nodes are devices with lower complexity, which may not be
other is the optimization based method [17–19]. In the latter
capable of performing complex tasks; but they are assumed
method, an objective/cost function is first defined and then
to be able to carry out time-of-arrival (TOA) estimation. For
a minimization algorithm is applied to solve the positioning
convenience, an anchor node is denoted by an anchor while
problem. Both unconstrained and constrained minimization
an ordinary sensor node is denoted by a sensor.
algorithms can be employed [20, 21]. The Quasi-Newton DFP
Let us first consider the case where the anchors are perfectly
(Davidon-Fletcher-Powell) algorithm was exploited in [22–
synchronized such as by sharing the same global time refer-
24]. A Gauss-Newton type Levenberg-Marquardt method [25]
ence, through cable connections, or by other means. A signal
was studied in [26]. Convex optimization was considered in
is transmitted from the sensors and the neighboring anchors
[27] and the simplex method [28] was employed in [29]. In the
detect the incoming signal and estimate the TOA.
TS method, a set of nonlinear equations is linearizied based on
a Taylor expansion. The set of linearized equations is solved to 2 The results are also suited to cellular networks in which the position
produce a new approximate position and the process continues information of base stations is known, whereas that of mobile stations is
to be determined.
1 Several parameters of the radio signal have been commonly exploited for 3 Alternatively, anchor nodes may be termed ”master nodes and sensor nodes
positioning, including time-of-arrival (TOA) (and thus TDOA and round-trip- may be termed ”slave nodes. In some cases, anchor nodes are fixed and sensor
time (RTT)), received signal strength (RSS), and angle of arrival (AOA). nodes are mobile.

978-1-4244-1645-5/08/$25.00 ©2008 IEEE 2656


In the Cartesian system, the range (distance) between anchor Applying the weighted least-square estimator produces
i and the sensor of interest is given by  −1 T −1
 δ = AT R−1 A A R D, (9)
di,0 = (x − xi )2 + (y − yi )2 + (z − zi )2 = c(ti − t0 ), where R may be replaced by an identity matrix if it is un-
i = 1, 2, ..., N (1) known. Given an initial position guess pv , we can compute δ
where N anchors are assumed available, (x, y, z) are the based on (9). Then, the position estimate is updated according
unknown position coordinates of the sensor, and (xi , yi , zi ) to
are the known coordinates of anchor i. c is the speed of light, p(k+1)
v = p(k)
v + δ.
ti is the signal TOA at anchor i, and t0 is the transmit time where k indexes the iteration number. The position estimate
at the sensor. is refined continually until δ is sufficiently small. This is the
Suppose that the transmit time t0 is not available at the Taylor series method [12].
anchors or it is not accurate enough to be employed. Therefore, Now let us keep the second order terms in the Taylor series
we make use of the TDOA measurements. Subtracting (1) for of fi (p). Then (6) becomes
i = 1 from (1) for i = 2, 3, ... N produces
1
di,0 − d1,0 = c(ti − t1 ), i = 2, 3, ..., N. (2) fi,v + (ai,1 ai,2 ai,3 )δ + δ T Hi δ ≈ dˆi,1 ,
2
Define i = 1, 2, ..., N − 1, (10)
T
p = [x, y, z] , where
(3)  
fi (p) = di+1,0 − d1,0 , i = 1, 2, ..., N − 1, ∂ 2 fi (p) ∂ 2 fi (p) ∂ 2 fi (p)
 ∂x2 ∂x∂y ∂x∂z 
and let t̂i be the TOA estimate at anchor i. Then, Hi = 

∂ 2 fi (p)
∂x∂y
∂ 2 fi (p)
∂y 2
∂ 2 fi (p)
∂y∂z

 . (11)
∂ 2 fi (p) ∂ 2 fi (p) ∂ 2 fi (p)
fi (p) = dˆi+1,1 + i+1,1 , i = 1, 2, ..., N − 1, ∂x∂z ∂y∂z ∂z 2 p=pv

where dˆi,1 = c(t̂i − t̂1 ) and i,1 is the corresponding range Here,
difference estimation error with the covariance matrix denoted   2 
∂ 2 fi (p)  1 xi+1 − xv
by R. Let the initial position estimate be = 1−
∂x2 p=pv dˆi+1 dˆi+1
pv = [xv , yv , zv ]T , (4)  2 
1 x1 − xv
and the estimation error be − 1− ,
dˆ1 dˆ1
   
δ = pv − p = [δx , δy , δz ]T . (5) ∂ 2 fi (p)  1 x1 − xv y1 − y v
=
Expanding fi (p) in Taylor series and retaining the first two ∂x∂y p=pv dˆ1 dˆ1 dˆ1
  
terms produce 1 xi+1 − xv yi+1 − yv
− ,
fi,v + ai,1 δx + ai,2 δy + ai,3 δz ≈ dˆi+1,1 , dˆi+1 dˆi+1 dˆi+1
   
i = 1, 2, ..., N − 1, (6) ∂ 2 fi (p)  1 x1 − xv z1 − z v
=
∂x∂z p=pv dˆ1 dˆ1 dˆ1
where the estimation and linearization errors are dropped, and   
1 xi+1 − xv zi+1 − zv
fi,v = fi (pv ), − ,
 dˆi+1 dˆi+1 dˆi+1
∂fi (p)  x1 − xv xi+1 − xv   2 
ai,1 = = − , ∂ 2 fi (p)  1 yi+1 − yv
∂x p=pv dˆ1 dˆi+1 = 1−
 ∂y 2 p=pv dˆi+1 dˆi+1
dˆi = (xv − xi )2 + (yv − yi )2 + (zv − zi )2 ,  2 
 (7) 1 y1 − y v
∂fi (p)  y1 − yv yi+1 − yv − 1− ,
ai 2 = = − , dˆ1 dˆ1
∂y p=pv ˆ
d1 dˆi+1 
   
∂fi (p)  z 1 − zv zi+1 − zv ∂ 2 fi (p)  1 y 1 − yv z1 − z v
ai,3 = = − =
∂z p=pv ˆ
d1 dˆi+1
. ∂y∂z p=pv dˆ1 dˆ1 dˆ1
  
1 yi+1 − yv zi+1 − zv
Equation (6) can be rewritten in a compact form as − ,
dˆi+1 dˆi+1 dˆi+1
Aδ ≈ D, (8)   2 
∂ 2 fi (p)  1 zi+1 − zv
= 1−
where ∂z 2 p=pv dˆi+1 dˆi+1
[A]i,j = ai,j , 1 ≤ i ≤ N − 1, 1 ≤ j ≤ 3,  2 
 T 1 z1 − z v
− 1− .
D = dˆ2,1 − f1,v , dˆ3,1 − f2,v , . . . , dˆN,1 − fN −1,v , dˆ1 dˆ1

2657
It would be impractical to directly handle (10) due to its non- stage can be derived to be the same expression as given by
linearity. Instead, we propose to retain the linearity as in (6) (14) but with
1 T T
and exploit the second order information in the third term.
More specifically, we first make use of (9) to produce δ̃. Then, A1 = A + H1 δ̃, HT2 δ̃, . . . , HTN δ̃ ,
we approximate the second-order term in (10) by 2
where A and D are given by (17) and H is given by (11),
1 T 1
δ Hi δ ≈ δ̃ T Hi δ. (12) whose elements are
2 2
  2 
The RHS of (12) would usually be a good approximation to ∂ 2 fi (p)  1 xi − xv
= 1− ,
the LHS provided that there is no dramatic difference between ∂x2 p=pv dˆ1 dˆ1
δ̃ and δ. Substituting (12) in (10) yields    
∂ 2 fi (p)  1 xi − xv yi − y v
  = − ,
1 T ∂x∂y p=pv dˆi dˆi dˆi
fi,v + (ai,1 ai,2 ai,3 ) + δ̃ Hi δ ≈ dˆi,1 ,    
2 ∂ 2 fi (p)  1 xi − xv zi − z v
= − ,
i = 1, 2, ..., N − 1. (13) ∂x∂z p=pv dˆi dˆi dˆi
  2  (18)
∂ 2 fi (p)  1 yi − y v
From (13), it is clear that the linearity is retained and the = − 1 − ,
second order information is also taken into account. Similarly ∂y 2 p=pv dˆ1 dˆ1
   
to (9), applying the weighted least square estimator, we have ∂ fi (p) 
2
1 y i − yv zi − z v
=− ,
 −1 T −1 ∂y∂z p=pv dˆi dˆi dˆi
δ = AT1 R−1 A1 A1 R D (14)   2 
∂ 2 fi (p)  1 zi − z v
=− 1− .
where ∂z 2 p=pv dˆ1 dˆ1
1 T T
A1 = A + H1 δ̃, HT2 δ̃, . . . , HTN −1 δ̃ .
2 IV. P ERFORMANCE E VALUATION
Then, the position estimate is continually updated until the We consider two performance indices: the root mean square
threshold of δ is crossed or the threshold of the number of error (RMSE) and the failure rate. The RMSE is defined as
iterations is crossed. Clearly, at each iteration, the weighted 
 Np Ns
least square estimation is employed twice. Therefore, this may  1  
be termed a two-stage Taylor series method. RM SE =  [∆X 2 + ∆Y 2 + ∆Z 2 ] (19)
3Np Ns i=1 j=1
III. M ODIFIED TS A LGORITHM BASED ON R ANGE
where i indexes the number of different position configurations
M EASUREMENTS
of the anchors and the sensor, j indexes the number of TOA
In some circumstances the anchors may not be accurately samples for each position configuration, and
synchronized. Therefore, instead of using the TDOA, the range
measurements are now employed. For instance, the range may ∆X = x(i) − x̂(i,j) , ∆Y = y (i) − ŷ (i,j) , ∆Z = z (i) −ẑ (i,j)
be estimated by measuring the round-trip-time (RTT) between    
Here x(i) , y (i) , z (i) and x̂(i,j) , ŷ (i,j) , ẑ (i,j) are the true
a pair of nodes, the radio signal strength, or by measuring
and estimated position coordinates of the sensor, respectively.
the time difference between a radio signal and an ultrasonic
signal. Redefine The failure rate is the percentage of the cases where there is
no solution or the solution is unreasonable. With the iterative
fi (p) = di,0 , fi,v = fi (pv ). (15) methods such as the Taylor series method and the modified
Taylor series method, the failure case includes situations where
Also redefine the algorithm does not converge to a solution, the maximum
xv − xi y v − yi z v − zi number of function evaluations/ iterations is exceeded, or the
ai,1 = , ai,2 = , ai,3 = , (16)
dˆi dˆi dˆi results are greatly beyond the monitored area.
Two different sizes of the monitored area are examined: one
where dˆi is given by (7). Then, following the same procedure, has dimensions of 70m(l) × 70m(w) × 10m(h) and the other
we can derive the weighted LS solution to the position has dimensions of 150m × 150m × 20m. The positions of the
increment at the first stage as given by (9) but with anchors and the sensor of interest are randomly generated. At
each test point, 1000 runs are conducted with new random
[A]i,j = ai,j , 1 ≤ i ≤ N, 1 ≤ j ≤ 3,
 T positions of the anchors and the sensor at each run. The
(17)
D = dˆ1 − f1,v , dˆ2 − f2,v , . . . , dˆN − fN,v , performance is then averaged. The TOA error samples are
produced by using the TOA estimation algorithm in [30].
and R is the covariance matrix of the distance estimation Let us first examine the performance of the Taylor series
errors. The solution to the position increment at the second method and the modified Taylor series method when TDOA

2658
12 5
modified−G modified−G
original−G original−G
10 modified−S 4 modified−S
Average RMSE (m)

Average RMSE (m)


original−S original−S
8 3

6 2

4 1

2 0
6 8 10 12 14 16 6 8 10 12 14 16
Average SNR (dB) Average SNR (dB)

Fig. 1. RMSE of position estimation using TDOA measurements with four Fig. 3. RMSE of position estimation using RTT measurements with five
anchor nodes. ’S’ denotes for the smaller dimensions: 70m×70m×10m while anchor nodes.
’G’ for the larger dimensions:150m×150m×20m.

80
90 original−S
original−S original−G
80 original−G modified−S
60 modified−G
modified−S

Failure Rate (%)


70 modified−G
Failure Rate (%)

60 40

50
20
40

30
0
6 8 10 12 14 16
20 Average SNR (dB)
6 8 10 12 14 16
Average SNR (dB)
Fig. 4. Failure rate of position estimation using RTT measurements with
Fig. 2. Failure rate of position estimation using TDOA measurements with five anchor nodes.
four anchor nodes.

measurements and by about 30% using RTT measurements.


measurements are employed4 . Fig. 1 shows the RMSE of the This decrease plus the decrease in the failure rate would at
two methods when there are four anchors. The corresponding least partially compensate the extra computation involved in
failure rates are illustrated in Fig. 2. We can observe that the obtaining the second weighted LS solution in the modified
failure rate (including the divergence) is decreased consider- method.
ably by using the modified Taylor series method, whereas the
average RMSE is basically retained the same. The failure rate V. C ONCLUSIONS
goes from about 50% with the original method down to about A modified Taylor series approach has been proposed to
20% with the modified scheme at an SNR of 15 dB. Figs. 3 improve the convergence performance of the original TS based
and 4 demonstrate the performance comparisons of the two position estimation method. The algorithm can be applied
algorithms when the RTT measurements are employed. It can
be seen that the performance gain with the modified method
is also considerable with respect to the failure rate. original-TDOA 5.0 5.0 5.0 5.0 5.1
modified-TDOA 4.1 4.0 4.0 4.1 4.2
Table I shows the average number of iteration in the case original-RTT 6.2 6.0 6.0 6.2 6.4
of five anchors and under five different groups of TOA error modified-RTT 4.3 4.1 4.1 4.2 4.3
samples. Making use of the modified scheme, the number TABLE I
of iteration is decreased by about 20% when using TDOA AVERAGE NUMBER OF ITERATION OF THE ORIGINAL AND MODIFIED
TAYLOR SERIES METHODS WHEN FIVE ANCHORS ARE EMPLOYED . T HE
4 For 3-D position estimation using TDOA measurements, at least four DIMENSIONS OF THE AREA ARE : 150 M ×150 M ×20 M .

anchors are needed.

2659
when either the TDOA or TOA/distance measurements are [25] D. Marquardt, “Algorithm for least-squares estimation of nonlinear
provided. Compared to the original method, the convergence parameters,” SIAM J. Appl. Math., vol. 11, pp. 431–441, 1963.
[26] K. Yu and I. Oppermann, “UWB positioning for wireless embedded
of the proposed scheme is considerably increased, which is networks,” in Proc. IEEE RAWCON, (Atlanta, USA), pp. 459–462, 2004.
demonstrated through simulations under different scenarios [27] L. Doherty, K. S. J. Pister, and L. E. Ghaoui, “Convex position
estimation in wireless sensor networks,” in IEEE Conf. Computer
R EFERENCES Communications (INFOCOM), pp. 1655–1663, 2001.
[28] J. Nelder and R. Mead, “A simplex method for function minimization,”
[1] B. T. Fang, “Simple solutions for hyperbolic and related position fixes,” Computer Journal, vol. 7, pp. 308–313, 1965.
IEEE Trans. Aerosp. Elecctron. Syst., vol. 26, pp. 748–753, Sept. 1990. [29] H. Wu, C. Wang, and N.-F. Tzeng, “Novel self-configurable positioning
[2] K. Yu and I. Oppermann, “Performance of UWB position estimation technique for multi-hop wireless networks,” IEEE Trans. Networking,
based on TOA measurements,” in Proc. Joint UWBST & IWUWBS, vol. 13, no. 3, pp. 609–621, 2005.
(Kyoto, Japan), pp. 400–404, 2004. [30] K. Yu and I. Oppermann, “Timing acquisition for IR-UWB systems,” in
[3] K. Yu, M. Hedley, I. Sharp, and Y. J. Guo, “Node positioning in ad Proc. IEEE Int. Symp. on Signal Processing and Applications (ISSPA),
hoc wireless sensor networks,” in Proc. Int. IEEE Conf. on Industrial (Sydney, Australia), pp. 287–290, Aug. 2005.
Informatics (INDIN), (Singapore), pp. 641–646, 2006.
[4] J. O. Smith and J. S. Abel, “Closed-form least squares source location
estimation from range difference measurements,” IEEE Trans. Acoustics,
Speech, and Signal Processing, vol. 35, pp. 1661–1669, Dec. 1987.
[5] B. Friedlander, “A passive localization algorithm and its accuracy
analysis,” IEEE J. Ocean. Eng., vol. 12, pp. 234–245, Jan. 1987.
[6] H. C. Schau and A. Z. Robinson, “Passive source location employing
intersecting spherical surfaces from time-of-arrival differences,” IEEE
Trans. Acoustics, Speech, and Signal Processing, vol. 35, pp. 1223–
1225, Aug. 1987.
[7] Y. T. Chan and K. C. Ho, “A simple and efficient estimator for
hyperbolic location,” IEEE Trans. Signal Processing, vol. 42, pp. 1905–
1915, Aug. 1994.
[8] Y. Huang, J. Benesty, G. W. Elko, and R. M. Mersereau, “Real-time
passive source localization: a practical linear-correction least-squares
approach,” IEEE Trans. Speech and Audio Processing, vol. 9, pp. 943–
956, Nov. 2001.
[9] D. Niculescu and B. Nath, “Ad hoc positioning system (APS),” in Proc.
IEEE Global Telecommunications Conf. (GLOBECOM), (San Antonio,
TX, USA), pp. 2926–2931, 2001.
[10] A. Savvides, C.-C. Han, and M. B. Strivastava, “Dynamic fine-grained
localization in ad-hoc networks of sensors,” in Proc. ACM SIGMOBILE,
(Rome, Italy), pp. 166–179, 2001.
[11] K. Langendoen and N. Reijers, “Distributed localization in wireless sen-
sor networks: a quantitative comparison,” Computer Networks, vol. 43,
pp. 499–518, 2003.
[12] W. H. Foy, “Position-location solutions by Taylor-series estimation,”
IEEE Trans. Aerosp. Elecctron. Syst., vol. 12, pp. 187–194, Mar. 1976.
[13] D. J. Torieri, “Statistical theory of passive location systems,” IEEE
Trans. Aerosp. Electron. Syst., vol. 20, pp. 183–198, Mar. 1984.
[14] D. E. Manolakis, “Efficient solution and performance analysis of 3-D po-
sition estimation by trilateration,” IEEE Trans. Aerosp. Elecctron. Syst.,
vol. 32, pp. 1239–1248, Oct. 1996.
[15] K. W. Cheung, H. C. So, W. K. Ma, and Y. T. Chan, “Least squares al-
gorithms for time-of-arrival-based mobile location,” IEEE Trans. Signal
Processing, vol. 52, pp. 1121–1128, Apr. 2004.
[16] K. Yu and Y. J. Guo, “NLOS error mitigation for mobile location
estimation in wireless networks,” in Proc. IEEE Vehicular Technology
Conf. (VTC), Apr. 2007.
[17] P. E. Gill, W. Murray, and M. H. Wright, Practical Optimization.
London: Academic Press, 1981.
[18] R. Fletcher, Practical Methods of Optimization. Chichester: John Wiley
& Sons, 1987.
[19] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge
University Press, 2004.
[20] H. Miao, K. Yu, and M. Juntti, “Positioning for NLOS propagation:
algorithm derivations and Cramer-Rao bounds,” IEEE Trans. Vehicular
Technology, vol. 56, pp. 2568–2580, Sept. 2007.
[21] K. Yu and Y. J. Guo, “Improved positioning algorithms for non-line-of-
sight environments,” IEEE Trans. Vehicular Technology, May 2008, to
appear.
[22] R. J. Fontana, E. Richley, and J. Barney, “Commercialization of an ultra
wideband precision asset location system,” in Proc. IEEE Conf. UWB
systems and Technologies, pp. 369–373, 2003.
[23] K. Yu, J. P. Montillet, A. Rabbachin, P. Cheong, and I. Oppermann,
“UWB location and tracking for wireless embedded networks,” Signal
Processing, vol. 86, pp. 2153–2171, Sept. 2006.
[24] K. Yu, “3-D localization error analysis in wireless networks,” IEEE
Trans. Wireless Communications, vol. 10, pp. 3473–3481, Oct. 2007.

2660
View publication stats

You might also like