Professional Documents
Culture Documents
5, OCTOBER 2009
821
I. INTRODUCTION
UE to the lack of precision of conventional location systems such as Global Positioning System (GPS) in indoor
and urban environments, the research on location methods based
on other types of wireless networks is attracting more and more
attention in recent years [1]. Nowadays, the great deployment of
WLAN IEEE 802.11 wireless networks makes this technology
ideal for developing such location systems.
Different wireless location systems are based on the estimation of a mobile station (MS) position by using the values of
certain measurements taken from radio-frequency signals that
travel between the MS and the fixed stations. These measurements can be time-of-arrival (TOA), time-difference-of-arrival
(TDOA), received-signal-strength (RSS), and angle-of-arrival
(AOA) [2].
Manuscript received July 01, 2008; revised May 19, 2009. Current version
published October 21, 2009. This work was supported in part by the Regional
Ministry of Education (Consejeria de Educacion) of Castilla y Leon (Spain)
under Grant VA002A06. The associate editor coordinating the review of this
manuscript and approving it for publication was Dr. Dorota Brzezinska.
S. Mazuelas, A.Bahillo, E. Garcia, and J. Blas are with the Center
for the Development of Telecommunications (CEDETEL) Edificio Solar,
Parque Tecnologico de Boecillo, 47151 Boecillo (Valladolid), Spain (e-mail:
smazuelas@cedetel.es).
F. A. Lago, P. Fernandez, R. M. Lorenzo, and E. J. Abril are with the Department of Signal Theory and Communications and Telematic Engineering.
University of Valladolid, 47011 Valladolid, Spain (e-mail: flago@tel.uva.es).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/JSTSP.2009.2029191
822
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 3, NO. 5, OCTOBER 2009
the receiver, is known as path loss, and it is modeled to be inversely proportional to the distance between the emitter and the
receiver raised to a certain exponent. This exponent is known as
path loss exponent [15], [16], path loss factor [17], or path loss
gradient [18].
Other factors that affect RSS values are the multipath or fast
fading and the shadowing or slow fading. These two factors can
be modeled with Rayleigh or Rician and log-normal distribu, can be modeled
tions [11], [16]. Therefore, RSS values,
by means of the following expression: [17]
(1)
is the transmitted power,
and
are the transwhere
mitter and receiver gains, respectively, is the distance between
emitter and receiver, is the path loss exponent, and and
are the parameters that conform the Rayleigh/Rician and lognormal distributions, respectively.
Averaging over certain time interval we can eliminate the fast
fading term [19], [20]. Thus, taking logarithmic units in (1) and
following the derivation steps shown in [17], RSSI values can
be modeled by the following expression:
(2)
where is the distance between the transmitter and the receiver
and the term denotes a Gaussian random variable with zero
mean caused by shadowing [21], [22]. The term is a constant
which depends on several factors: averaged fast and slow fading,
, and transmitted power , therefore, in practice,
gains ,
the value of can be often known beforehand [17].
Expression (2) has been widely used in order to describe RSS
values as a function of the distance between the transmitter and
the receiver in wireless communications [15], [16]. Usual examples of the use of this expression with specific values for
and are the known propagation models of free space, OkumuraHatam, Egli, etc. [15].
To sum up, our basic assumptions are given as follows.
In any propagation environment present between the MS
and each one of the APs, the relationship between RSSI
values and the distance between AP and MS can be modeled by (2).
parameter remains constant in scenarios where
The
and
, and the transmitted power
the antenna gains,
are also constant, situation typically found in most IEEE
802.11 WLANs. Therefore, it is possible to obtain the
value of by measurements taken in a generic environment similar to the one in which we are going to carry
out the location [15], [17], [18], and this value will be
valid whereas the antenna gains and the transmitted power
remains the same as the ones present when we obtain this
value.
Even though the value of the path loss exponent depends
highly on the specific channel environment presented between each specific AP and the MS at a given instant, we
can assume that this parameter remains constant during a
short period of time [18].
These basic hypotheses can be condensed in the following stateare RSSI values obtained from
ment: if
Authorized licensed use limited to: National Taiwan University. Downloaded on August 06,2010 at 03:32:43 UTC from IEEE Xplore. Restrictions apply.
MAZUELAS et al.: ROBUST INDOOR POSITIONING PROVIDED BY REAL-TIME RSSI VALUES IN UNMODIFIED WLAN NETWORKS
(3)
Due to the importance of this result in the development and
techniques presented in this paper, we are going to dedicate the
first part of Section VI to prove the suitability of expression (3)
with a large amount of measurements taken in several indoor
propagation environments.
(10)
and since
, the CRLB is
(11)
(12)
(4)
and therefore the probability density function (pdf) of
ditioned on the distance between the MS and the
pressed as
conis ex-
(13)
(5)
thus the maximum-likelihood estimator of
823
is
(6)
To determine such maximum, we differentiate the log-likelihood function. Thereby we obtain the following score function:
is short
since we are assuming that the lapse of time
enough in order to assume the distance between the AP and
the MS does not change, it is also logical to assume that
,
and, therefore
(7)
therefore
(14)
(8)
and thus
(9)
Authorized licensed use limited to: National Taiwan University. Downloaded on August 06,2010 at 03:32:43 UTC from IEEE Xplore. Restrictions apply.
(15)
824
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 3, NO. 5, OCTOBER 2009
(16)
and therefore the CRLB is
(17)
Authorized licensed use limited to: National Taiwan University. Downloaded on August 06,2010 at 03:32:43 UTC from IEEE Xplore. Restrictions apply.
MAZUELAS et al.: ROBUST INDOOR POSITIONING PROVIDED BY REAL-TIME RSSI VALUES IN UNMODIFIED WLAN NETWORKS
825
(22)
Fig. 3. Least squares solution of the six radical axes corresponding to four
circles.
(21)
where we have weighted the squares sum with the radii of each
circle in order to apply more relevance to the smaller circles
and the minus sign indicates that the sum of squares and the
compatibility are inversely related.
B. Least Squares Optimization
During a time interval
, using known APs positions
and RSSI values in (15), the expression (21) which quantifies
the compatibility of the distance estimates depends only on the
(23)
Thereby, we have transformed the problem of estimating
different path loss exponents into a nonlinear least squares
problem. This can be solved by using robust techniques like the
LevenbergMarquardt algorithm [25], [26].
C. Path Loss Exponents Constraints
The fact that all the distance estimates are compatible does
not imply that all the estimates are precise: compatibility is a
necessary but not sufficient condition to achieve precision in the
distance estimates. Therefore, in order to obtain optimum path
loss exponents in our specific problem we do not have to calculate the path loss exponents which maximize the compatibility
in a global fashion but the path loss exponents which, belonging
to a feasible set of solutions, maximize the compatibility function
(24)
In order to determine an appropriate feasible set, we can use
some heuristic reasonings. For example, it is clear that the path
loss exponents are always going to lie between a maximum and a
minimum which can be roughly known a priori. That is
.
Also, we can determine other constraints to the path loss exponents which make the set of feasible solutions more restricis the set of RSSI values obtained
tive. Thereby, if
with respect to the
APs. Let
in an instant of time
be the
APs sorted according to the average RSSI values obtained in this lapse of time, that is
Authorized licensed use limited to: National Taiwan University. Downloaded on August 06,2010 at 03:32:43 UTC from IEEE Xplore. Restrictions apply.
(25)
826
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 3, NO. 5, OCTOBER 2009
(31)
is
if
(32)
is a convex set, indeed it is a polyhedral set and therefore variants of the LevenbergMarquardt algorithm [27] can be applied
in order to resolve (32), where we can choose, as an initial guess
in the algorithm, a rough approximation of the path loss expoor the center of the polyhedron .
nents like
VI. RESULTS WITH MEASUREMENTS AND SIMULATIONS
A. Relationship Between RSS and Distance
(27)
Moreover, we can obtain other type of constraints imposing
relations between different distances. It is clear that although
,
, this fact does not imply that
.
However, it is logical that this distance difference exists if the
difference in the average signal strength is great enough. Therefore,
(28)
moreover, this difference in the distances depends on how great
the difference in the average signal strength is. Therefore, if
thus
, where is a value which decreases when the difference between the average RSS increases.
it occurs that
In the case that
(29)
In this section, we are going to show the suitability of expressions (2) and (3). In order to do that we carried out a large campaign of measurements on the second floor of the Higher Technical School of Telecommunications, University of Valladolid
(Spain), shown in Fig. 9.
As for APs, we used eight identical wireless broadband
routers with two antennas each, in diversity mode, which is
typically found on most IEEE 802.11 WLAN routers. APs were
configured to send a beacon frame every 10 ms to constant
power. Diversity circuitry determines which antenna has better
reception and switches it on in a fraction of a second while
it turns off the other antenna. Therefore, both antennas are
never active at the same time. APs have omnidirectional rubber
duck antennas mounted. These are vertically polarized with a
symmetrical 360 radiation pattern in the horizontal plane and
with a vertical beamwidth of approximately 75 . As for MS,
we used a WLAN cardbus adapter with a vertically polarized
omnidirectional external antenna, also found on most IEEE
802.11 WLAN adapters.
Authorized licensed use limited to: National Taiwan University. Downloaded on August 06,2010 at 03:32:43 UTC from IEEE Xplore. Restrictions apply.
MAZUELAS et al.: ROBUST INDOOR POSITIONING PROVIDED BY REAL-TIME RSSI VALUES IN UNMODIFIED WLAN NETWORKS
827
(33)
Fig. 6. Simulation Scenario.
and
are the height of the AP and the MS, rewhere
spectively, which are located on the same floor. Therefore, we
can assume that the antenna gains of the devices remain constant in all indoor environments.
In order to obtain a value for that we can use in the subsequent location process, we carried out several measurements in
a specific indoor environment taking place in a corridor of the
School. In Fig. 4, we can see the relation between distance and
dBm and
.
these RSSI values, getting
In order to show the reliability of (2) and the suitability of
the assumptions made in Section III and expressed in (3), we
performed eight groups of measurements in four types of environments, varying the distance between the transmitter and the
receiver. In Table I, we can see the values obtained for the path
loss exponents, , in environments with a different number of
Authorized licensed use limited to: National Taiwan University. Downloaded on August 06,2010 at 03:32:43 UTC from IEEE Xplore. Restrictions apply.
828
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 3, NO. 5, OCTOBER 2009
Fig. 9. Trilaterations obtained on the second floor of the Higher Technical School of Telecommunications Engineering.
TABLE I
VALUES FOR (2) TAKING
dBm WITH REAL MEASUREMENTS
IN SEVERAL ENVIRONMENTS
= 052
,
,
,
and
, where
are the six
different path loss exponents that characterize the propagation
channel from the six APs sorted by their proximity to the MS.
Also the standard deviation of shadow fading was simulated as
a random value between 2.85 dBm and 3.45 dBm. We repeated
this simulation 5000 times randomly varying the position of the
MS between the circles shown in Fig. 6.
In Fig. 7, we can see the histogram of the 30 000 errors occurred in the path loss exponents estimation by using the method
dB,
,
introduced in Section V, where we use
, and
as the parameters used to obtain the
path loss exponents constraints. As we can see, the method introduced is very accurate, achieving a mean error of 0.2003 and
a standard deviation of 0.2052.
C. Estimation of Distances From RSSI Values
In this section, we show the accuracy in distance estimation
obtained by using the estimated path loss exponents. In order
to do that, in the simulation scenario explained before, we estimated distances between the MS and different APs by means
of (15). To show the suitability of the estimated path loss exponents, we compare in Fig. 8 the errors that occurred in the
estimated path loss exponents with:
the CRLB of errors shown in Section IV;
the errors in the ideal case of knowing the real path loss
exponents (RPLE);
,
,
the errors using the values
, and
which are the middle
values of the random values used to simulate the path loss
exponents, (MPLE);
Authorized licensed use limited to: National Taiwan University. Downloaded on August 06,2010 at 03:32:43 UTC from IEEE Xplore. Restrictions apply.
MAZUELAS et al.: ROBUST INDOOR POSITIONING PROVIDED BY REAL-TIME RSSI VALUES IN UNMODIFIED WLAN NETWORKS
the errors with the two-slope path loss model recommended by the IEEE 802.11 standard [22] for this
for distances shorter
environment, that is, using
than 20 m and
for longer distances, (SPLE).
Also, in Fig. 8 we can appreciate the great accuracy obtained
by the method presented (diamond marks), since it achieved
a mean error of 9.0861 m, that is a mean error of approximately 20% of the distance between APs. Notice that all the
other methods are impractical, since in real scenarios we do not
know the exact real values for path loss exponents nor even a
very good approximation of them (MPLE), unless we perform
a wide set of previous measurements. Even so, the method presented outperforms methods using approximate values for the
path loss exponents, a remarkable fact, taking into account that
the method presented does not use any calibration. Additionally,
in Fig. 8, we can see that the values of the expression for the
CRLB derived in Section IV are very close to the errors using
exact real values for path loss exponents, a fact that is a very
logical result.
D. Trilateration Results
In order to show the feasibility and suitability of the methods
proposed, we tested the techniques explained before in the indoor environment of the second floor of the Higher Technical
School of Telecommunications Engineering.
As MS we used a laptop with an IEEE 802.11 cardbus
adapter. We implemented an ANSI C program which obtains
the RSSI values corresponding to the beacon probes sent every
10 ms by eight APs located on the second floor of the School
(see Fig. 9). At each instant, this program uses the real-time
RSSI values from the four most powerful APs in order to
estimate the path loss exponents and hence the actual distances
to these four APs, by using the method presented with the
same parameters explained in Section VI-C. To quantify the
effectiveness of the method presented, we have also estimated
the different distances by using a constant path loss exponent
of 2, which is the value recommended by the standard for
distances lower than 20 m [12].
To illustrate the improvement brought by the proposed
method, we trilaterate in order to compare the MS position
when using the estimated distances by the proposed method
and when using the estimated distances from constant path loss
exponents. We walked with the laptop along the route shown
in blue in Fig. 9 taking measurements in 121 points. Thereby,
in red we can see the trilateration positions obtained by using
the method proposed and, in purple, some positions obtained
by using constant path loss exponents, where 50.4% of the
estimated positions obtained by this method are not shown,
since they lie outside this map. Therefore, as we can see in
Fig. 9, the method proposed achieves a high level of accuracy,
and we want to point out that this result was achieved only
through trilateration without any tracking technique.
Moreover, in Fig. 10, we can see the cumulative probability
of errors obtained with the method proposed and by using constant path loss exponents. The mean of errors committed with
the method proposed was 3.97 m with a standard deviation of
1.18 m. Obviously, the positioning accuracy can be improved
by using some tracking technique such as Kalman filters, but in
this section we want to show the feasibility of the methods proposed and the reliability of the path loss exponents and distance
estimates obtained from the methods presented.
829
VII. CONCLUSION
A novel indoor location method is presented in this paper.
This novel method is based on trilateration through RSSI values
obtained in real time. The path loss exponents which characterize the propagation channel between the MS and each AP
are dynamically estimated from RSSI values by assessing and
maximizing the compatibility of the distances estimates. Once
the path loss exponent is estimated, the distance between the
MS and each AP can be obtained and, therefore, MS position is
achieved by trilateration.
With this indoor location method, neither a previous calibration stage nor any radio-map information are needed. Therefore, this method is robust in dynamic environments as well
as having a low complexity, since it uses only real-time RSSI
values, without needing any change to the WLAN IEEE 802.11
network.
Results using both simulations and measurements are shown
in order to prove the reliability and suitability of the method
proposed. A mean error slightly lower than 4 m is obtained in
an unmodified WLAN network deployment, without using any
other tracking technique.
REFERENCES
[1] A. Kpper, Location-Based Services: Fundamentals and Operation.
New York: Wiley, 2005.
[2] F. Gustafsson and F. Gunnarsson, Mobile positioning using wireless
networks: Possibilities and fundamental limitations based on available
wireless network measurements, IEEE Signal Process. Mag., vol. 22,
no. 4, pp. 4153, Jul. 2005.
[3] X. Chai and Q. Yang, Reducing the calibration effort for probabilistic
indoor location estimation, IEEE Tran. Mobile Comput., vol. 6, no. 6,
pp. 649662, Jun. 2007.
[4] S. A. Golden and S. S. Bateman, Sensor measurements for Wi-Fi location with emphasis on time-of-arrival ranging, IEEE Trans. Moblile
Comput., vol. 6, no. 10, pp. 11851198, Oct. 2007.
[5] P. Bahl, V. N. Padmanabhan, and A. Balachandran, Enhancements
to the RADAR User Location and Tracking System, Research Tech.
Rep. Microsoft, 2000, MSR-TR-2000-12.
[6] T. Roos, P. Myllymaki, H. Tirri, P. Misikangas, and J. Sievanen, A
probabilistic approach to WLAN user location estimation, Int. J. Wireless Inf. Netw., vol. 9, no. 3, pp. 155164, Jul. 2002.
[7] Z. Xiang, S. Song, J. Chen, H. Wang, J. Huang, and X. Gao, A wireless
LAN-based indoor positioning technology, IBM J. Res. Dev ., vol. 48,
no. 5/6, pp. 617626, 2004.
[8] M. Youssef and A. Agrawala, The horus WLAN location determination system, in Proc. 3rd Int. Conf. Mobile Syst., Applicat., Services
(MobiSys05), 2005, pp. 205218.
Authorized licensed use limited to: National Taiwan University. Downloaded on August 06,2010 at 03:32:43 UTC from IEEE Xplore. Restrictions apply.
830
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 3, NO. 5, OCTOBER 2009
Mr. Mazuelas received the young scientists prize for the best communication in the Union Radio-Scientifique Internationale (URSI) XXII Symposium
(Spain).
Ruben M. Lorenzo received the Telecommunication Engineer and Ph.D. degrees from the University
of Valladolid, Valladolid, Spain, in 1996 and 1999,
respectively.
From 1996 to 2000, he was a Junior Lecturer
at the University of Valladolid with the Optical
Communications Group. Since 2000, he has been
a Lecturer. He is currently the Head of the Faculty
of Telecommunication Engineering at the University of Valladolid and Research Director of the
Center for the Development of Telecommunications
(CEDETEL). His main research topics include communication systems and
networks, electromagnetic characterization, and radiolocation.
Patricia Fernandez received the Telecommunication Engineer degree from Universidad Politecnica
de Cataluna, Barcelona, Spain, in 1997 and the Ph.D.
degree in 2004 from the University of Valladolid,
Valladolid, Spain.
Since 1999, she has been a Junior Lecturer at the
University of Valladolid. Her current research interests are communication systems and networks, electromagnetic characterization, and radiolocation.
Dr. Fernandez is the author of more than 40 papers
in international journals and conferences.
Authorized licensed use limited to: National Taiwan University. Downloaded on August 06,2010 at 03:32:43 UTC from IEEE Xplore. Restrictions apply.
MAZUELAS et al.: ROBUST INDOOR POSITIONING PROVIDED BY REAL-TIME RSSI VALUES IN UNMODIFIED WLAN NETWORKS
Juan Blas received the Ph.D. degree in Telecommunication Engineering from the University of
Valladolid, Valladolid, Spain, in 2008.
He is an Associate Lecturer with the University
of Valladolid. In 2006, he joined the Center for the
Development of Telecommunications (CEDETEL),
University of Valladolid, as a Research Engineer.
His main research interests include propagation
models, interactions between electromagnetic fields
and living matter, and antennas.
831
All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately.
Authorized licensed use limited to: National Taiwan University. Downloaded on August 06,2010 at 03:32:43 UTC from IEEE Xplore. Restrictions apply.