See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/232644462
Inquiry-Based Bluetooth Indoor Positioning via
RSSI Probability Distributions
Article · June 2010
DOI: 10.1109/SPACOMM.2010.18
CITATIONS
READS
30
463
6 authors, including:
Ling Pei
Jingbin Liu
59 PUBLICATIONS 493 CITATIONS
58 PUBLICATIONS 564 CITATIONS
Shanghai Jiao Tong University
SEE PROFILE
Finnish Geodetic Institute
SEE PROFILE
Heidi Kuusniemi
Yuwei Chen
100 PUBLICATIONS 808 CITATIONS
76 PUBLICATIONS 903 CITATIONS
Finnish Geospatial Research Institute
SEE PROFILE
Finnish Geodetic Institute
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Finland's EGNOS Monitoring and Performance Evaluation (FEGNOS) View project
INSURE -Information Security of Location Estimation and Navigation Applications View project
All content following this page was uploaded by Ruizhi Chen on 05 February 2014.
The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Journal of Global Positioning Systems (2010)
Vol.9, No.2 :122-130
DOI: 10.5081/jgps.9.2.122
Using Inquiry-based Bluetooth RSSI Probability Distributions for
Indoor Positioning
Ling Pei, Ruizhi Chen, Jingbin Liu, Heidi Kuusniemi, Tomi Tenhunen, Yuwei Chen
Department of Navigation and Positioning, Finnish Geodetic Institute, FGI, Masala, Finland
Abstract
Fingerprinting is a common technique for indoor
positioning using short range Radio Frequency (RF)
technologies such as Wireless Location Area Network
(WLAN) and Bluetooth (BT). It works in two phases:
The first phase is a data training phase in which a radio
map for the targeted area is generated in advance, while
the second phase is the real-time location determination
phase using the radio map. Considering the work amount
for generating the radio map, only a few samples of the
Radio Signal Strength Indicator (RSSI) are typically
collected at each reference point. The limited samples
are not able to represent the real signal distribution well
in the conventional fingerprint approach such as in an
occurrence-based solution. This paper presents a new
solution using the Weibull function for approximating
the Bluetooth signal strength distribution in the data
training phase. This approach requires only a few RSSI
samples to estimate the parameters of the Weibull
distribution. Compared to the occurrence-based solution,
the Weibull function utilizes the shape, shift, and scale
parameters to describe the distribution over the entire
RSSI domain. This study indicates that the reliability and
accuracy of the fingerprint database is improved with the
Weibull function approach. A Histogram Maximum
Likelihood position estimation based on Bayesian theory
is utilized in the positioning phase. The test results show
that the fingerprinting solution using the Weibull
probability distribution performs better than the
occurrence-based fingerprint approach.
Keywords: Bluetooth, indoor positioning, RSSI,
fingerprint, Baysian estimation
_____________________________________________
1.
Introduction
Location-based Service (LBS) is now becoming one of
the standard features in mobile devices. More and more
research concentrates on the personal navigation for both
outdoor and indoor environments. However, Global
Navigation Satellite System (GNSS) technologies are
still struggling for indoors due to the unavailability or
attenuation of the GNSS signals. There are many radio
technologies such as cellular networks, Wireless Local
Area Network (WLAN), and Bluetooth (BT) that are
now adopted for indoor positioning without modifying
neither the user terminals, nor the existing infrastructure.
Radio Signal Strength Indicator (RSSI), a standard
measure in most radio technologies, has attracted a lot of
attentions (Bahl & Padmanabhan, 2000 and Ekahau Inc.)
for being adapted as measurements in indoor positioning.
Bluetooth is a technology with low power consumption
for short-range wireless data and voice communication
(Muller, 2001). It has been utilized in the communication
and proximity market (Naya et al., 2005) for a long time.
As widely supported by mobile devices, Bluetooth is a
potential technology to become an alternative for indoor
positioning (Simon & Robert, 2009, Anastasi et al., 2003,
Bargh & Groote, 2008, Jevring & Groote, 2008, Huang,
2005, Bruno & Delmastro, 2003, Hallberg et al., 2003,
and Pandya et al., 2003). The effective range of the radio
signal of a class 1 Bluetooth device (e.g. the Bluegiga
Access Point(AP) 3201) is up to 200 meters, while that
for the class 2 device (e.g. the Bluetooth module in a
smart phone) is about 20-30 meters according to the
specifications of Bluetooth 2.0 (Specification of the
Bluetooth System, Core Specification v2.0+EDR, 2004).
Bandara et al. (2004) developed a multi-antenna
Bluetooth AP for location estimation based on RSSIs.
The test obtained 2 meters of error in a 4.5m x 5.5m area
with four antennas. Sheng and Pollard (2006) modified
the Bluetooth standard to estimate the distance between a
reference transmitter and a mobile receiver, using RSSI
measurements and a line-of-sight radio propagation
model within a single cell. The high-density Bluetooth
infrastructure is necessary to acheive an accurate
position in the above two approaches. In order to
minimize the Bluetooth infrastructure, Damian et al.
(2008) used only one class 1 Bluetooth AP for a home
localisation system, which combined the measurements
of the link quality, RSSI, and celluar signal quality to
obtain room-level accuracy. In this paper, we present a
Bluetooth locating solution in a reduced Bluetooth
infrastructure area by using RSSI only.
Pei, et al.: Using Inquiry-based Bluetooth RSSI Probability Distributions for Indoor Positioning
123
2.
The RSSI Measurement
Bluetooth
There are two types of possible solutions for acquiring
the Bluetooth RSSI measurements: the connection-based
solution and the inquiry-based solution (Naya et al.,
2005). In the connection-based
solution, a
communication connection between an AP and a mobile
phone is needed to establish before carrying out the
RSSI measurements. The RSSI measurements can be
updated at a frequency of 1 Hz via the established
communication channel. However, APs might
continually adjust the transmission power of the
communication link to reduce the transmission errors
and save the energy. The transmission power adjustment
makes it impossible to use the RSSI measurement to
infer the distance between a mobile phone and an AP.
Nevertheless, this is not the case for the inquiry-based
solution because it retrives the RSSIs from the inquiry
response that utilizes static transmission power instead
of the adjustable one. Therefore, the RSSI measurements
of the inquiry-based solution reflect the distances
between the mobile devices and APs. After the above
analyzing, the inquiry-based solution is adopted in our
study even though the RSSI update frequency is lower
than that of the connection-based solution.
As shown in Figure 1, the components of the proposed
inquiry-based Bluetooth locating system in this paper
consist of two parts: the Bluetooth network and mobile
phones. The server connected with several APs over a
WLAN/Ethernet network is responsible for the system
kernel functions, especially positioning calculations. The
APs are synchronized by the Server when inquiring the
mobile phones in their surroundings and relay the
positions from the Server to mobile phones.
Whenever RSSI measurements are needed for
positioning, the server will send a trigger to all APs to
scan the mobile devices in their surroundings.
Mobile device might be miss-detected for three reasons:
1) The time or frequency domain between the mobile
device and AP does not overlap during the inquiring
process; 2) the mobile device is waiting to answer the
inquiry from another AP. One mobile device can only
answer one AP at a time; and 3) the inquiring process
times out without obtaining a successful measurement
for a reason e.g. that the communication between the AP
and the corresponding mobile phone is blocked. The
probability of being miss-detected for each device will
increase when the number of participating APs increases
(Peterson et al., 2006a) as shown in Table 1. The
occurrence of the miss-detected cases will decrease the
number of RSSI measurements in a certain sampling
duration.
WLAN/Ethernet
Server
Mobile Phone
Access Point
Mobile Phone
Access Point
Mobile Phone
Access Point
Mobile Phone
Figure 1: System components of the inquiry-based
Bluetooth indoor locating system.
Table 1. Rates of missed-detection
Number of
Missed-detection Rate
participating APs
After 6.4 s
1
0%
2
7.5%
3
8.3%
6
8.9%
Having completed the inquiring task, all APs will send
the RSSI measurements back to the server either for the
purpose of calculating the current positions of the mobile
devices or generating the radio map database.
3.
Fingerprinting with RSSIs
As mentioned above, fingerprinting with RSSIs consists
of two phases: the data training phase and the
positioning phase as shown in Figure 2. The training
phase includes the steps of obtaining a radio map for the
targeted area based on a RSSI training data set, while the
positioning phase includes the steps of finding a location
based on the fingerprints stored in the radio map.
For the data training phase, the targeted area is divided
into cells. The center of the each cell is considered as a
reference point. The coordinates of the reference points
(x n , y n ) are determined in advance.
The RSSI
measurements at each reference point from all “visible”
APs are collected and stored as fingerprints in the
database of the radio map.
Pei, et al.: Using Inquiry-based Bluetooth RSSI Probability Distributions for Indoor Positioning
124
D = [R1 , R2 ,..., Rw ]
(3)
where w is the maximum number of the reference points
in the radio map.
To speed up the computation process, a bin-based
solution is adopted. The signal strength distribution is
divided into p bins. The fingerprints for the i-th reference
point can be redefined as
P( A1 B1 | Ri ) P( A2 B1 | Ri )
P( A B | R ) P( A B | R )
1 2
i
2 2
i
Ri =
P( A1 B p | Ri ) P( A2 B p | Ri )
Figure 2: Two phases for Bluetooth positioning
During the positioning phase, the unknown coordinates
(x u , y u ) of a mobile device are estimated by matching
the snap shot of the current RSSI measurements to the
fingerprints stored in the radio map (Youssef et al., 2003
and Roos et al., 2002).
3.1 Fingerprint Database
At each reference point, the RSSI probability
distributions of all APs are stored. If we denote the
fingerprint for the i-th reference point as Ri , then, we
have
P ( A1O1 | R i ) P ( A2 O1 | R i )
P( A O | R ) P( A O | R )
1 2
i
2 2
i
Ri =
P ( A1Ov | Ri ) P( A2 Ov | Ri )
P ( Ak O1 | Ri )
P ( Ak O 2 | Ri ) (1)
P( Ak Ov | Ri )
where A stands for the AP, while O refers to the RSSI
measurement.
In the conventional fingerprinting approach, the
probability of a RSSI measurement On between the
reference point
Ri and the AP Am can be expressed as
P( Am On | Ri ) =
C On
Ni
(2)
where CO is the number of occurrences that the RSSI
n
measurement On appeared in the training data set of the
i-th reference point. Here N i is the total number of
training samples collected at the i-th reference point. The
entire fingerprint database is expressed as
P( Ak B1 | Ri )
P( Ak B 2 | Ri ) (4)
P( Ak B p | Ri )
In the conventional occurrence-based solution, at the i-th
reference point, the probability of the RSSI
measurements within the bin Bn for AP Am can be
expressed as
P( Am Bn | Ri ) =
∑
j < En
CO j
j ≥ En −1
(5)
Ni
Where E n −1 and E n are the left and right edges of bin
Bn respectively. C O j stands for the number of
[
occurrences that the value of the RSSI measurement
appeared within the range of E n −1 , E n ) . All the RSSI
measurements in the bin Bn are cumulated for counting
the occurance probability.
3.2 Modelling Fingerprints with the Weibull
Function
The bin-based solution requires a large training data set
in order to obtain a good estimate of the RSSI
probability distribution. In this paper, we introduce the
Weibull function to proximate the RSSI probability
distribution. The Weibull function is a traditional method
for modelling the signal strength of radio propagation
(Sagias & Karagiannidis, 2005). The probability density
function can be expressed as
k x − θ k −1 −( xλ−θ ) k
) e
,x ≥θ
(
f ( x) = λ λ
0,
x <θ
(6)
While the cumulate distribution function is defined as
F ( x) = 1 − e
−(
x −θ
λ
)k
(7)
Pei, et al.: Using Inquiry-based Bluetooth RSSI Probability Distributions for Indoor Positioning
125
where x is the variable of the function, k is the shape
parameter, λ is the scale parameter, and θ is the shift
parameter. When θ=0, this reduces to a 2-parameter
distribution.
The parameters of the Weibull function can be estimated
with a limited number of RSSI sample measurements
(e.g. 20). The function parameters λ , k , θ can be
calculated with (Papoulis, 2002):
(
)
k = δ / ln(2), 1.5 ≤ k ≤ 2.5
δ <2
2 × (k + 0.15)
λ = δ × (k + 0.15) 2 ≤ δ ≤ 3.5
3.5 × (k + 0.15)
δ > 3.5
θ = O − λ × Γ(1 + 1 k )
1 n
O = ∑ Oi
n i =0
δ
(9)
(11)
(12)
Γ is
O is the mean value of the RSSI measurement
set Oi ,
function. The term ( k + 0.15) is an approximation of
is the standard deviation.
the gamma
the expression 1 / Γ(`1 + 2 / k ) − Γ (1 + 1 / k ) when
2
1.5 ≤ k ≤ 2.5 .
For each possible RSSI measurement in this study, the
distribution probability can be expressed as
P( x) = F ( x + 0.5) − F ( x − 0.5)
(13)
Because the RSSI measurements are rounded to an
integer. The probability for each bin in the fingerprint
database can be generated as
P( Am Bn | Ri ) = ∫
x+w
x
Using a Weibull function based fingerprint database, we
can calculate the probability for any arbitrary RSSI
measurement. Considering the computation cost, we still
adopt the bin-based solution in this paper by pregenerating the fingerprint database using Weibull
functions derived from limited samples.
3.3 Positioning with Bayesian Histogram Maximum
Likelihood algorithm
The Bayesian theorem and Histogram Maximum
Likelihood algorithm are used for positioning (Youssef
et al., 2003 and Roos et al., 2002).
Given
(10)
1 n
δ=
(Oi − O) 2
∑
n i =0
where
(8)
radio map can be reduced in this case because it just
requires storing three parameters for each vector
between an AP and a reference point.
f ( x)dx = F ( x + w) − F ( x) (14)
where w is the width of the bin, x is the RSSI value at
the left edge of bin.
In theory, the radio map can be represented by a set of
Weibull functions. Each Weibull function has three
parameters representing the probability distribution of
the RSSI measurements between an AP Am and a
mobile phone at a reference point Ri . The size of the
vector O
find
= {O1 , O2 ...Ok } from APs, the problem is to
the
the
RSSI
l
location
probability P (l | O)
Bayesian theorem
being
measurement
with
the
conditional
maximized. Using the
arg max l [ P (l | O )] = arg max l [
P (O | l ) P (l )
]
(15)
P(O )
where P (O) is constant for all l , therefore, the Equation
(15) can be reduced as
arg max l [ P(l | O)] = arg max l [ P(O | l ) P(l )]
(16)
We assume that the mobile device has equal probability
to access each reference point, so P (l ) can be
considered as constant in this case, Equation (16) can be
simplified as
arg max l [ P (l | O)] = arg max l [ P (O | l )]
(17)
Now it becomes a problem of finding the maximum
conditional probability of
P(O | l ) = ∏ P(On | l )
k
(18)
n =1
where the conditional probability P(On | l ) is derived
from the RSSI distribution pre-stored in the fingerprint
database. If the RSSI measurement
On belongs
to the
bin B j , Equation (18) can be expressed as
P(O | l ) =
∏ P( A
k
m =1
m
B j | Ri )
(19)
Pei, et al.: Using Inquiry-based Bluetooth RSSI Probability Distributions for Indoor Positioning
126
while taking Equation (5) and (14) into account.
Therefore, the problem becomes to find the maximum
∏ P( A
0.12
Weibull-based,11589 samples
Occurrence-based,11589 samples
k
m =1
m
B j | Ri ) in the fingerprint database.
0.1
Results and Discussions
In order to evaluate the preformance of the solution
proposed in this paper, two test cases have been carried
out. The first case is a static test with a long session of
collecting 11589 RSSI samples, while the second case is
a dynamic test conducted inside the official building of
the Finnish Geodetic Institute. The objectives of the first
test case are to
• determinate if the shapes of the Weibull functions
derived from a limited RSSI samples can
approximate the reference shape derived from the
long session of 11589 RSSI measurements, and
• compare the positioning performance (in static case)
of the Weibull-based solution to that of the
conventional occurrence-based solution.
The objective of the second test case is to evaluate the
positioning performances of the Weibull-based solution
in a dynamic scenario.
4.1 Static Test
In order to establish a reference for comparison, we
conducted a long-term measurement campaign. It lasted
for 20 hours and 11589 RSSI samples were collected.
Considering that the occurrence-based probability
distribution derived from 11589 RSSI samples is close to
the real RSSI probability distribution, we utilized it as
the benchmark distribution for the purpose of
comparison.
0.06
0.04
0.02
0
55
60
65
70
75
RSSI
80
85
90
95
Figure 3: Weibull-based (k=2.5, λ=10.275, θ=61) vs.
occurrence-based probability distribution with 11589
samples
In Figure 4, the blue dash line stands for the probability
distribution derived from a Weibull-based solution using
20 RSSI samples randomly selected from the large data.
The green dash line is the probability distribution
derived from the occurrence-based solution for the same
data set of 20 RSSI measurement samples, while the red
solid line is the benchmark distribution.
0.35
Weibull-based,20 samples
Occurrence-based,20 samples
Occurrence-based,11589 samples
0.3
0.25
Probability
4.
Probability
0.08
0.2
0.15
0.1
By using Equations (8)-(12), the parameters of the
Weibull function derived from 11589 RSSI samples
were calculated as follows: shape k=2.5, scale λ=10.275
and shift θ=61. By using Equation (13), we got the
Weibull-based probability distribution as the blue line
shown in Figure 3. The red solid line is the benchmark
distribution. The shapes of the two lines are similar.
From our experience, it is scarcely bearable for a person
collecting samples at one reference point for more than
two minutes. About 20 RSSI samples can be obtained
over a two-minute sampling duration. Therefore, we
selected 20 samples as the limited sampling case for
comparison.
0.05
0
55
60
65
70
75
RSSI
80
85
90
95
Figure 4: Weibull-based (k=2.5, λ=9.275, θ=61) vs.
occurrence-based probability distributions with 20
samples
It is obvious that the shape of the Weibull function
derived from 20 RSSI samples is similar to that of
benchmark distribution. By comparing the probabilities
estimated with the conventional occurrence-based
solution for the case of 20 samples to that estimated with
the Weibull function, it is obvious that probabilities
estimated with the Weibull function are closer to those
derived from the benchmark distribution. For example,
the true probability for the RSSI measurements values of
68 and 69 should be close to 0.1 based on benchmark
distribution. These values are zero while they are
Pei, et al.: Using Inquiry-based Bluetooth RSSI Probability Distributions for Indoor Positioning
127
estimated with the conventional occurrence-based
approach, and about 0.11 if they are estimated with the
Weibull function.
0.18
Comparing to the benchmark distribution, Figure 5
shows the probability distributions derived from the
Weibull-based solution with 11589 samples (red line),
Weibull-based solution with 20 samples (blue line), and
occurrence-based solution with 20 samples (green line).
Table 2 presents the numerical statistics of the
probability differences.
0.12
0.16
Probability density
0.14
0.1
0.08
0.06
0.04
0.02
0.25
0.15
0.1
0.05
0
55
60
65
70
75
RSSI
80
85
90
60
65
70
75
RSSI
80
85
90
95
Figure 6: Probability densities estimated with Weibull
functions for all sessions of 20 RSSI measurement
samples. The red line is the benchmark distribution.
95
Figure 5: Comparison of the probability distributions.
It can be seen from the results that the Weibull-based
probability distributions estimated from 11589 samples
and that from 20 RSSI measurement samples have
similar shapes. The probability distributions estimated
with the Weibull-based solution are significantly better
than that obtained from the conventional occurrencebased approach.
In order to reduce the computation time, the Weibull
functions are “digitized”. Using Equation (14), the
probability densities shown in Figure 6 are cumulated as
the bin-based probability in each bin as shown in Figure
7. In our study, the bin edge x is defined as [-55 -60 -65 70 -75 -80 -85 -90 -95]. The width of the bin w is -5. All
the RSSI values lager than -55 belong to the 1st-bin. The
minimum possible RSSI value is -95. Thus, there are
nine bins designed in our study.
0.8
0.7
0.6
0.5
Probability
0.2
Probability difference
0
55
Weibull-based,11589 samples
Weibull-based,20 samples
Occurrence-based,20 samples
0.4
0.3
Table 2. Statistics of the probability differences
WeibullWeibullOccurrencebased
based
based
(20 samples) (20 samples)
(11589
samples)
0.0105
0.0099
0.0275
Mean
0.0136
0.0122
0.0471
Std
0.0502
0.0431
0.2490
Max
For a more detailed investigation, as shown in Figure 6,
the large data set of 11589 RSSI measurements is
divided into hundreds of sessions that contain 20
samples each (blue lines in Figure 6). The Weibull
function for each session is derived and compared with
the benchmark distribution (red line in Figure 6).
0.2
0.1
0
1
2
3
4
5
BIN number
6
7
8
9
Figure 7: Bin-based probabilities estimated with Weibull
functions (blue line) and that estimated with the
benchmark distribution (red line).
It is not difficult to see that the shapes of most Weibull
functions derived from 20 RSSI samples are close to that
of the benchmark distribution (the red line in Figure 7).
Table 3 gives the statistics for the differences between
the probabilities deriving from the Weibull functions and
that from the benchmark distribution. According to the
statistics, the Weibull functions derived from 20 RSSI
measurement samples effectively approximate the
probability distributions. The largest difference of mean
value between two probabilities is 0.37 appearing in the
4th-bin. Probability distributions represented with the
Pei, et al.: Using Inquiry-based Bluetooth RSSI Probability Distributions for Indoor Positioning
128
Weibull functions obtained from 20 RSSI samples are
similar to the benchmark distribution. The maximum
difference standard deviation is less than 0.0758.
4.2 Dynamic Indoor Positioning
The dynamic indoor test cases were carried out at the
Finnish Geodetic Institute (FGI) with only three
Bluetooth APs (red points in Figure 8) mounted inside
the office building. The distance between two adjacent
APs is about 20 meters. From our field test results, most
mobile phones such as Nokia N8, N95, N95 8G,
Navigator 6710, Xpress 5800, and HTC Desire can be
scanned by the AP in a range of 30 meters without
blockage. The length of each corridor is more than 40
meters.
We used a NovAtel SPAN GPS/IMU reference system
with 1 Hz output as the reference (green line in Figure 8).
The Nokia N95 8G phone was used as the user terminal
in the test cases. In order to initialize the SPAN system,
the test started from the outside of the building for
unobstructed GPS availability. Having initiated the
SPAN system, a user who held the Bluetooth-enabled
handset (Nokia N95 8G) entered into the building and
walked along the corridor. Finally, the user got out of the
building from another exit as shown in Figure 8. The
purple-circled line in Figure 8 stands for the Bluetooth
positioning solutions.
Figure 8: Test route
For comparison, the same location determination
algorithms are applied for the WLAN positioning
solutions (black-pointed line in Figure 8) using the same
mobile device. There are 8 WLAN APs installed in the
same test environment. As shown in Figure 9, the
horizontal error is 5.1 meters for Bluetooth-based
solutions, while that for the WLAN positioning solution
is 2.2 meters. It is easy to understand that the Bluetoothbased solution has a lower positioning accuracy
compared to the WLAN solution because the number of
APs for the Bluetooth-based solutions is much less than
that of the WLAN positioning solution.
12
BT-only
WLAN-only
10
Horizontal error [m]
The static positioning test is intended to evaluate the
locating accuracy and stability over time. In this study,
two sets of overnight static tests were carried out in two
days at the same reference point, one lasted for about 20
hours, while the other lasted for about 24 hours. The test
data sets are applied for position estimation using the
occurrence-based and Weibull-based
fingerprint
databases respectively. The occurrence-based fingprint
database is generated by using Equation (5), while the
Weibull-based solution is derived from Equation (14).
The test results are presented in Table 4. We can see that
the Weibull-based solution performs significantly better
than the occurrence-based solution. The accuracy of the
Weibull-based solution in the 20 hours test case is 1.43
meters better than that derived from the occurrencebased solution for the same data set. In 24 hours test
case, the error of Weibull-based solution is 1.88 meters
lower than that of the occurrence-based solution.
Compared to the occurrence-based solution, the Weibullbased solution improves the accuracy by 25.91% and
32.53% respectively for two long-term static positioning
test cases.
Mean horizontal error [m]
BT: 5.1
WLAN: 2.2
8
6
4
2
0
600
650
700
Time since beginning of test [s]
Figure 9: WLAN and Bluetooth locating errors
750
Pei, et al.: Using Inquiry-based Bluetooth RSSI Probability Distributions for Indoor Positioning
129
Table 3. The statistics of the difference between the Weibull-based probability distribution using 20 samples and the
occurrrence-based probability distribution using total measurements
BIN number
Mean
Std
Max
1
2
3
4
5
6
7
8
9
0
0
0
-0.0001
0.0009
0.0135
-0.1299
0.0684
0.2293
0.0217
0.0690
0.3723
0.1607
0.0758
0.3657
0.0194
0.0500
0.3002
0.0192
0.0082
0.1259
0.0136
0.0005
0.0137
0.0006
0
0.0006
Table 4. Static locating test
Database
Time
Error
5.
Weibull-based
20 h
4.09 m
Conclusions and discussion
Bluetooth as an existing wireless infrastructure has been
widely utilized in personal area network communication.
The proximity approaches based on Bluetooth have also
been investigated in recent years. To pursue a practical
Bluetooth locating solution with sufficient accuracy in a
wider area, this study enlightens an inquiry-based
Bluetooth indoor locating approach via RSSI probability
distributions.
The test result shows that RSSI probabilistic approach is
a reasonable way for Bluetooth locating. Since the
Weibull function is utilized for approximating the
probability distribution of Bluetooth signal strength, the
reliability and accuracy of the fingerprint database is
improved significantly. It reduces the amount of work
needed for generating the fingerprint database.
6.
Future work
The following aspects will be considered to improve the
locating performance in the related future research
efforts: firstly, the Weibull-based fingerprint database
will be optimized; secondly, without a timely update,
more intelligent position estimation algorithms are
needed for better location prediction; and finally, more
Bluetooth features such as link quality and celluar signal
quality will be studied.
Acknowledgements
This work is supported by the project of iSPACE
(indoor/outdoor Seamless Positioning and Applications
for City Ecosystem) funded by the Finnish Funding
Agency for Technology and Innovation (TEKES), Nokia,
Fastrax, Bluegiga and Indagon. This work is also a part of
the project INOSENCE (INdoor Outdoor Seamless
Navigation for SEnsing Human Behavior) funded by the
Academy of Finland.
Occurrence-based
24 h
3.90 m
20 h
5.52 m
24 h
5.78 m
References
Anastasi G., Bandelloni R., Conti M., Delmastro F.,
Gregori E., and Mainetto G. (2003), Experimenting
an indoor Bluetooth-based positioning service,
Proceedings of the 23rd International Conference on
Distributed Computing Systems Workshops, April
2003, pp. 480–483.
Bahl P. and Padmanabhan V. N.(2000), RADAR: An InBuilding RF-Based User Location and Tracking
System, Proceedings of IEEE Infocom 2000, pp.
775–784, March 2000.
Bandara U., Hasegawa, M., Inoue, M., Morikawa,
H.,
Aoyama,
T.
(2004),
Design
and
implementation of a Bluetooth signal strength
based location sensing system, IEEE Radio and
Wireless Conference, pp.319 – 322.Atlanta USA,
September, 2004.
Bargh M. and Groote R. (2008), Indoor localization
based on response rate of bluetooth inquiries,
Proceedings of the first ACM international workshop
on Mobile entity localization and tracking in GPSless environments ,September 2008.
Bruno R. and Delmastro F. (2003), Design and analysis
of a Bluetooth-based indoor localization system,
Personal Wireless Communications, pp.711–725.
Damian K., Sean M. and Terry D. (2008), A BluetoothBased Minimum Infrastructure Home Localisation
System. In Proceedings of 5th IEEE International
Symposium on Wireless Communication Systems,
October 2008, Reykjavik, Iceland, pp:638 – 642.
Ekahau.Inc., http://www.ekahau.com/, visited on 4 April
2010.
Pei, et al.: Using Inquiry-based Bluetooth RSSI Probability Distributions for Indoor Positioning
130
Hallberg J., Nilsson M., and Synnes K. (2003),
Positioning with Bluetooth, Proceedings of the 10th
International Conference on Telecommunications,vol.
2(23), pp. 954–958 ,2003.
Sheng Z. and Pollard, J.K. (2006), Position
measurement using Bluetooth, IEEE Transactions
on Consumer Electronics, Vol 52(2), pp. 555-558.
Huang A. (2005), The use of Bluetooth in Linux and
location aware computing, Master of Science
dissertation
Simon H. and Robert H. (2009), Bluetooth Tracking
without Discoverability, LoCA 2009: The 4th
International Symposium on Location and Context
Awareness, May 2009.
Jevring M., Groote R., and Hesselman C. (2008),
Dynamic optimization of Bluetooth networks for
indoor localization, First International Workshop on
Automated and Autonomous Sensor Networks, 2008.
Specification of the Bluetooth System, (2004), Core
Specification v2.0 + EDR, Bluetooth SIG,
http://www.bluetooth.org/, visited on 18 August
2010.
Muller N. (2001), Bluetooth Demystified, McGraw-Hill,
New York.
Youssef M., Agrawala A., and Shankar A. U. (2003),
Wlan location determina- tion via clustering and
probability distributions, Proceedings of the First
IEEE International Conference on Pervasive
Computing and Communications, pp:143-150. IEEE
Computer Society, Texas, USA, March 2003.
Naya F., Noma H., Ohmura R., and Kogure K. (2005),
Bluetooth-based indoor proximity sensing for
nursing context awareness, Proceedings of the 9th
IEEE International Symposium on Wearable
Computers, pp. 212–213, September 2005.
Pandya D., Jain R., and Lupu E. (2003), Indoor location
estimation using multiple wireless technologies, the
14th IEEE Proceedings on Personal, Indoor and
Mobile Radio Communications, vol. 3, pp. 2208–
2212, August 2003.
Papoulis A. (2002), Probability, Random Variables And
Stochastic Processes, McGraw-Hill Education (India)
Pvt Ltd, 2002.
Peterson B., BaldwiR. n , and Raines R. (2006a),
Bluetooth Discovery Time with Multiple Inquirers,
Proceedings of the 39th Annual Hawaii International
Conference on System Sciences, pp.232.1,January
2006.
Peterson B., Baldwin R., and Kharoufeh J. (2006b),
Bluetooth Inquiry Time Characterization and
Selection, IEEE Transactions on Mobile Computing,
vol 5 (9), pp.1173-1187,September 2006.
Roos T., Myllymaki P., Tirri H., Misikangas P., and
Sievanen J. (2002), A probabilistic approach to
WLAN user location estimation,
International
Journal of Wireless Information Networks, Vol 9(3),
pp.155-164, July 2002.
Sagias C. and Karagiannidis K. (2005), Gaussian class
multivariate Weibull distributions: theory and
applications in fading channels, Institute of
Electrical and Electronics Engineers. Transactions on
Information Theory, Vol 51 (10), pp. 3608–3619,
2005.
Zaruba G. and Gupta V. (2004), Simplified Bluetooth
Device Discovery Analysis and Simulation,
Proceedings of the 37th Hawaii International
Conference on System Sciences, Hawaii, USA, Jan
2004.
Biography
Dr. Ling Pei (born 1977, Email:ling.pei@fgi.fi) is a
specialist research scientist of the Navigation and
Positioning Department at Finnish Geodetic Institute
(FGI). He received his Master in mechanical engineering
from Jiangxi Agricultural University in 2003 and Ph.D.
degree in test measurement technology and instruments
from the Southeast University, China, in 2007. Since
2007, he has joined the Finnish Geodetic Institute (FGI),
where his research interests include indoor/outdoor
seamless positioning, ubiquitous computing, wireless
positioning, mobile computing, context-aware and
location-based services.