Professional Documents
Culture Documents
(J) Indoor Localization System Using Dual-Frequency Bands and Interpolation Algorithm - 2020 - Yyf
(J) Indoor Localization System Using Dual-Frequency Bands and Interpolation Algorithm - 2020 - Yyf
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.
11184 IEEE INTERNET OF THINGS JOURNAL, VOL. 7, NO. 11, NOVEMBER 2020
such as the k-nearest neighbor (KNN), probabilistic methods, rather than using a single-frequency value as fingerprint data,
neural networks, support vector machine (SVM), and neural the difference of two frequencies, 2.4 and 5 GHz, is used
networks [12], the positioning server compares the received to define a more robust signature and eliminates the device
TD vector with the stored fingerprints. The fingerprints of the heterogeneity between the offline phase and online phase.
TD match the RPs to get the closest coordinate or coordinate On the whole, our contributions are summarized as follows.
combination, which is the last estimated position. All match- 1) At each RP, DFD-CSI not only collects the RSS value
ing algorithms are completed on the server, as shown in Fig. 1, at 2.4 GHz but also include 5 GHz, and includes the dif-
and finally, only the localization result is fed back to TD. ference between the two as the final reference data. The
In the actual positioning application process, diversified fingerprint of two frequency bands at each RP comes
devices are implemented in offline and online signal measure- from just one AP, not from different APs. The DFD
ments, which will affect the performance of indoor localization algorithm solves the issue of inconsistent fingerprints
because of the heterogeneity [18], [19]. The state of the art caused by heterogeneous devices while maintaining the
of the solution requires two or three APs to obtain rela- stability of single-frequency fingerprint localization.
tive values to get rid of heterogeneity [20], [21]. There is 2) The DFD-CSI system uses the CSI method to interpo-
a great drawback in this way, especially when the relative late the fingerprint database, which not only reduces the
distance between two APs changes, and such a situation is labor force but also has a better effect on localization
very common in an indoor environment. Meanwhile, IEEE accuracy and controllability compared with the crowd-
802.11-based WiFi operates in the 2.4- and 5-GHz spectra. sourcing approach. Based on the measured fingerprint,
Presently, almost all routers, that is, AP in indoor localization, the CSI performs segmentation interpolation on the fin-
support these two frequency bands. In the meantime, WiFi- gerprint data of each AP, ensuring that the obtained
based fingerprint indoor localization only uses one frequency interpolation data border on the real data.
band [22], [23], which is a waste of information contained in 3) In a real indoor scene, we tested the proposed system.
another frequency band. Because the DFD-CSI system is extended from the tradi-
On the other hand, RSS values collection and radio map tional WiFi-RSS fingerprinting method, the conventional
establishment are an indispensable stage in fingerprint-based algorithm can be applied to our system with only minor
indoor localization. The localization accuracy is extremely modifications. Therefore, DFD-CSI can be deployed
dependent on the density of radio map while building a high- in an actual scene where the indoor localization is
density fingerprint database is labor intensive, time consuming, demanded.
and overpriced. However, there is a fact that the RSS value Fig. 2 shows the framework of DFD-CSI, which is simi-
of the same AP in the fixed place is not stable, but is con- lar to the most traditional fingerprint-based indoor localization
stantly changing. The variation of RSS caused by the opening system except that of the DFD and CSI. The remainder of
and shutting doors, movement of people, temperature, humid- this article is organized as follows. In Section II, the algo-
ity, and electromagnetic interference in offices requires some rithm of two frequencies difference is formulated. The CSI
means to deal with it. When the fingerprint database is estab- application algorithm in fingerprint database establishment is
lished, we need to regularly maintain it ensuring the validity presented in Section III and evaluation of DFD-CSI is in
and correctness of the radio data. To avoid updating finger- Section IV. Finally, Section V concludes this article.
print databases one by one at every RP, several crowdsourcing
approaches have been carried out in which general users con-
tributing new-fangled RSS can participate in the site survey II. D UAL -F REQUENCY D IFFERENCE
activity [14]–[17]. The automatic updating fingerprint using Considering the hardware differences of MDs (e.g., smart-
crowdsourced data however causes other numerous issues, phones, sweeping robot, tablets, and smart watch), manifold
such as power consumption, information redundancy, device MDs may get different measurement data, even for the exact
diversity, privacy protection, and participation willingness. same fixed AP. To illustrate this, we conduct an experiment and
These disadvantages have to be faced with practical appli- collect RSS samples using two mobile phones Mi Mix2s and
cation, and it is difficult to solve them all on the basis of Meizu M2, and two laptops Thunderobot G150T and Hasee
crowdsourcing. Therefore, another approach different from G8-CT7NT, respectively. The measured value in 22 different
crowdsourcing is needed to reoptimize the establishment of locations from one AP (TP-LINK TL-WDR5620) was col-
indoor fingerprint databases. lected 30 times severally and the average was taken as the
In view of the above two issues, a dual-frequency differ- RSS value to remove randomness. Fig. 3 shows the location
ence (DFD) and cubic spline interpolation (CSI) fingerprinting of 22 points and the typical linear shift following the signals
indoor localization system (DFD-CSI) is proposed in this arti- of the four devices.
cle. In the offline phase, DFD-CSI performs corresponding CSI It can be seen from Fig. 2 that no matter 2.4 or 5 GHz, RSS
operations on the fingerprint data of each AP based on half, values collected by different devices in each frequency band
one-quarter, or even fewer fingerprint data, and then constructs are discrepant. Such inconsistencies would cause the degra-
the full fingerprint database using all of the interpolated data. dation of the positioning system performance. Consequently,
Taking advantage of the accomplished interpolation database, device heterogeneity has been identified as a cause for affect-
the system not only reduces the amount of fingerprint collec- ing the localization accuracy in [24]–[27]. In these strategies,
tion but also ensures the localization accuracy. In the process, one method is based on AP-sequence, it has nevertheless
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.
YANG: INDOOR LOCALIZATION SYSTEM USING DUAL-FREQUENCY BANDS AND INTERPOLATION ALGORITHM 11185
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.
11186 IEEE INTERNET OF THINGS JOURNAL, VOL. 7, NO. 11, NOVEMBER 2020
N APs that were heard in the offline phase are also heard
during the online phase. It should be noted that NTD may not
be the same as Ni , because we have N over all available APs
in the offline database that might not be heard from each
location and neither in the online phase. For the APs that
are in the database but not heard by the device, there is one
solution: set −100 dBm that is equal to the sensitivity of the
device.
L5 f52 d
is the transmitted carrier’s frequency in WiFi, and L is the = 10 lg − 10 lg (β2.4 − β5 )
system loss factor. The effects of heterogeneous devices are P5
dB 2
L2.4 f2.4 d0
mainly concentrated on three parameters PAP , GAP , and GMD . + (X2.4 − X5 ). (7)
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.
YANG: INDOOR LOCALIZATION SYSTEM USING DUAL-FREQUENCY BANDS AND INTERPOLATION ALGORITHM 11187
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.
11188 IEEE INTERNET OF THINGS JOURNAL, VOL. 7, NO. 11, NOVEMBER 2020
expression is given by
D d 2 s(l) 2
s∗ (l) = arg min V(k, s(k)) + α dl (10)
s 0 dl2
0≤k≤L
where k is a natural number and k ∈ [0, D], and the char-
acteristics of which shows that it is a cost function, α is the
regularization factor that controls the smoothness of the solu-
tion curve. As can be seen from (10), it consists of two distinct
sections, the argument term and integral term. The meaning
of the argument term is to make the solution of the equation
sufficiently close to the minimum value of the potential func-
tion V(l, f ). It is not continuous in the process of fingerprint Fig. 7. Schematic of spline representation by the segment between knot
database interpolation while we interpolate each RSS value at points.
some discrete distances l = k. The second term is an integral
expression to implement a smoothness constraint which makes Algorithm 1 Proposed CSI Model
the final curve solution more inclined to a smooth curve with Input: Fingerprint data F Mi obtained from coarse-grained measurement,
a low average curvature. As a result, the transition between RP including 2.4GHz, 5GHz and their difference data (as shown in Eq.
and interpolation point will be more stable and unobtrusive. 8)
Output: Fingerprint database F Mi−CSI after interpolation
It is worth noting that although the RP and RSS values are
1 Obtain the initial fingerprint database F Mi
discrete, the final solution s(l) of (10) is a continuous snake 2 for i = {1, 2, . . . , N − 1} do
within the range of [0, D]. It has been proven in previous 3 for f i and f i+1 are two adjacent fingerprint data in F Mi do
studies that s(l) is a cubic spline which has knot points at 4 Calculate the cubic spline expression s(l) between f i and f i+1
according to Eq. 11-15
integers [32]. 5 Interpolate fingerprint data f i−CSI according to s(l)
In order to achieve the snake expansion numerically, we 6 end for
need to construct the following spline polynomial between two 7 end for
8 return F Mi−CSI = F Mi f 1−CSI · · · f i−CSI · · · f (N−1)−CSI
adjacent interpolation points, RPi and RPi+1 :
s∗i (l) = ai (l − ki )3 + bi (l − ki )2 + ci (l − ki )2 + di (11)
where ai , bi , ci , and di are the cubic spline coefficients which IV. E XPERIMENT R ESULTS AND A NALYSIS
is our new smoothness parameter instead of α.
A. Testbed Setup
Assuming a total of m RPs in one direction can hear the
signal of an AP, then we consider the simplified calculation Our experimental site is a portion of the first floor in the
problem between two RPs (knot points) Qizhen Building located in the Northwestern Polytechnical
University (NWPU). It is a comprehensive office building,
s∗i (li ) = fi (12) which includes a lobby, corridor, report room, office, etc. The
s∗ (li+1 ) = fi+1 (13) different rooms are separated by concrete walls and the entire
∗i
si (li+1 ) = s∗i+1 (li+1 ) (14) building structure is relatively complicated. The entire site
∗ is an irregular pattern with a longest length of 86.4 m and
si (li+1 ) = s∗i+1 (li+1 ) (15)
a total area of over 4000 m2 . Fig. 8 illustrates the layout of
where (12) and (13) are based on the equality of data at the the testbed, in which nine H3C dual-frequency WAP722 WiFi
knot points, and (14) and (15) are based on the continuity of routers were installed as APs in this field test. All APs are uni-
differential. Through the above formula, we can calculate the formly installed on the ceiling in advance, and they have the
values of the four parameters ai , bi , ci , and di , thereby esti- same height to the floor. We used four types of equipment to
mating the expression of s∗i (l) as shown in Fig. 7. According collect and test RSS values in the experimental environment,
to Algorithm 1, all snake splines can be calculated using knot and they included Mi Mix2s (phone), Meizu M2 (phone),
points and any interpolated RSS values can be calculated at Thunderobot G150T (laptop), and Hasee G8-CT7NT (laptop).
any distance in the interval [0, D]. According to the expression The collecting program ran on all equipment in order to collect
in (8), three types of fingerprints are interpolated, respectively, RSS fingerprints from multiple APs.
to complete the upgrade of the whole fingerprint database. By leveraging the corresponding program, the MD sends
The interpolation algorithm described above is executed by the collected MAC address of each AP and its corresponding
the maintainer of the indoor localization system. Compared RSS value to the server, and multiple RP points are collected
with the localization system using crowdsourced data, this to form a fingerprint database on the central server side. In the
method will only transmit the user’s effective localization proposed algorithm, MDs are only responsible for collecting
information during the execution process and does not need fingerprint data and uploading the data to the server, the DFD-
to consume too much power for uploading crowdsourced CSI calculation is performed at the server, and finally, the
information. The consequence of this processing is that the server feeds back the localization result to the MD.
system will not obtain the user’s real-time localization without There are 290 RPs throughout the test site as shown in
permission, thereby protecting the user’s privacy. Fig. 8, and the distance between each two is 3 m. On each
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.
YANG: INDOOR LOCALIZATION SYSTEM USING DUAL-FREQUENCY BANDS AND INTERPOLATION ALGORITHM 11189
TABLE I
M AJOR F INGERPRINT DATABASE S YMBOLS
(a)
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.
11190 IEEE INTERNET OF THINGS JOURNAL, VOL. 7, NO. 11, NOVEMBER 2020
Fig. 9. Cumulative localization error distribution of homogeneous devices: (a) Mi Mix2s; (b) Meizu M2; (c) Thunderobot G150T; and (d) Hasee G8-CT7NT.
TABLE II
C ROSS E XPERIMENT S ETTING IN THE H ETEROGENEOUS S CENARIO at 65% on average of these four heterogeneous conditions,
while RSS, SSD, and AP-sequence were 35%, 52%, and
55.5%, respectively. It can be seen from the above analysis
that the positioning accuracy of the DFD algorithm is sig-
nificantly better than the other three algorithms when using
heterogeneous devices. More importantly, Fig. 10 indicates
that although different kinds of devices may generate different
case of homogeneous devices due to the existing factors of localization results, the proposed DFD algorithm outperforms
RSS(2.4,ij) and RSS(5,ij) in (8). preceding robust positioning methods, whether the device is
In the next experiment, we analyze the situation of heteroge- heterogeneous or not.
neous devices. There are totally four fingerprinting databases The mixed use of heterogeneous devices will not affect the
and four devices in our experimental setting. In order to accuracy of indoor localization following the DFD algorithm.
deal with heterogeneous devices, we use different databases To demonstrate this advantage, we conducted experiments with
and devices to verify the performance status of RSS, SSD, the heterogeneous devices in Table II, there are four cases
AP-sequence, and DFD based on cross-testing methods. The in the detection of heterogeneous devices here. It is enough,
specific cross correspondences of database and device are however, to illustrate the convenience that the DFD algo-
shown in Table II, for instance, Hasee G8-CT7NT is the rithm brings to the localization with various and numerous
TD while using the fingerprinting database established by Mi heterogeneous devices at the same time in real life.
Mix2s.
The average localization error of the four heterogeneous
devices using the DFD algorithm is 3.7 m. As can be seen C. Efficiency Comparison Between CSI and Crowdsourcing
from Fig. 10, the positioning accuracy of different algorithms In order to directly verify the effectiveness of the CSI algo-
when facing heterogeneous devices is quite obvious. Similarly, rithm itself, we first set up the fingerprint database according
based on the localization error of 4 m, DFD achieved it to the RPs shown in Fig. 8. Similar to the above experiment,
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.
YANG: INDOOR LOCALIZATION SYSTEM USING DUAL-FREQUENCY BANDS AND INTERPOLATION ALGORITHM 11191
Fig. 10. Cumulative localization error distribution of heterogeneous devices: (a) Mi Mix2s; (b) Meizu M2; (c) Thunderobot G150T; and (d) Hasee G8-CT7NT.
we used two heterogeneous (Mi Mix2s and Meizu M2) devices FMeizu−CSI , respectively, with a 27.5% improvement in posi-
to collect fingerprint data of 290 RPs. After 50 days, on the tioning accuracy. From the whole point of view, whether it
basis of retaining the initial fingerprint databases FMi and is heterogeneous or homogeneous devices, our proposed CSI
FMeizu , we performed CSI algorithm processing on the two algorithm has shown obvious results in improving position-
fingerprint databases and upgraded to obtain the other two ing accuracy. Although the effect of localization has been
corresponding fingerprint databases FMi−CSI and FMeizu−CSI . reduced due to changes in the fingerprint database after
In the online phase, we took Mi Mix2s as TD and used the 50 days, the improvement consequence is still obvious after
DFD algorithm to test localization errors of the four finger- the corresponding interpolation is performed.
print databases, respectively. The whole 50 TPs are divided In a struggle to reduce the fingerprinting calibration efforts,
into two groups, TP1–TP25 as a group and TP26–TP50 as several research have been implemented on crowdsourcing
another, then performed localization accuracy tests on the two approaches in which ordinary users can participate in the fin-
groups of TPs. gerprint collection activity. Ordinary users, people walking
It can be seen from Fig. 11(a) that the mean localization indoors and needing location information, collect RSS data on
errors calibrated by FMi are larger than FMi−CSI , of which the path they move forward and contribute the fingerprint to
TP1–TP25 are tested using the initial fingerprint database FMi , the database. The following experiment is to verify and com-
and TP26–TP50 are tested using the interpolated fingerprint pare the localization performance of our proposed CSI method
database FMi−CSI . The mean positioning error has dropped and crowdsourcing method. Indoor experiments are conducted
from 4.8 to 3.6 m after interpolation, which means that the on the ground floor of the Qizhen building at NWPU as
positioning accuracy has increased by 25%. Fig. 11(a) shows described above.
the localization error results of homogeneous devices, while We set up the fingerprint database according to the RPs
in the display results of heterogeneous devices, Fig. 11(b), shown in Fig. 8. The initial fingerprint database has 290 RPs,
CSI plays the same role in optimizing results. The mean the fingerprint data collection time at each RP is however
localization errors are 5.1 m using FMeizu and 3.7 m using reduced to 1 s for comparison, and the mobile phone only
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.
11192 IEEE INTERNET OF THINGS JOURNAL, VOL. 7, NO. 11, NOVEMBER 2020
Fig. 11. Localization error of Mi Mix2s at 50 TPs. (a) Using database FMi Fig. 13. Performance comparison between different amount CTs and
from TP1 to TP25 and using FMi−CSI from TP26 to TP50. (b) Using database F Mi−CSI localization systems with altered APs.
FMeizu from TP1 to TP25 and using FMeizu−CSI from TP26 to TP50.
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.
YANG: INDOOR LOCALIZATION SYSTEM USING DUAL-FREQUENCY BANDS AND INTERPOLATION ALGORITHM 11193
where the TD is the homogeneous device Mi Mix2s. Due [5] K. Witrisal et al., “High-accuracy localization for assisted living: 5G
to the rough process of establishing the fingerprint database systems will turn multipath channels from foe to friend,” IEEE Signal
Process. Mag., vol. 33, no. 2, pp. 59–70, Mar. 2016.
in this experiment (1 sample instead of 40 samples at each [6] M. V. Moreno-Cano, M. A. Zamora-Izquierdo, J. Santa, and
RP), the whole positioning accuracy is decreased compared A. F. Skarmeta, “An indoor localization system based on artificial
with the previous experiment. Since the result of F Mi−CSI is neural networks and particle filters applied to intelligent buildings,”
Neurocomputing, vol. 122, pp. 116–125, Dec. 2013.
obtained by interpolation on the latest fingerprint database F Mi , [7] P. E. Lopez-de-Teruel, F. J. Garcia, O. Canovas, R. Gonzalez, and
it is not affected by altered APs, and the localization accuracy J. A. Carrasco, “Human behavior monitoring using a passive indoor
error is always maintained at 4.3 m. Its localization result is positioning system: A case study in a SME,” Procedia Comput. Sci.,
vol. 110, pp. 182–189, Jul. 2017.
worse than LAAFU just when there is no altered AP or only [8] D. Gu and K.-S. Chen, “Design and performance evaluation of wiimote-
one. However, when there is more than one altered AP in the based two-dimensional indoor localization systems for indoor mobile
fingerprint database, the LAAFU method cumulatively accu- robot control,” Measurement, vol. 66, pp. 95–108, Apr. 2015.
[9] G. E. Martín et al., “Infrared sensor system for mobile-robot positioning
mulating seven CTs in turn has a greater localization error in intelligent spaces,” Sensors, vol. 11, no. 5, pp. 5416–5438, 2011.
than CSI, and the error tends to become larger as the num- [10] L. Barkhuus and A. K. Dey, “Location-based services for mobile tele-
ber of altered APs increasing. Obviously, although the time phony: A study of users’ privacy concerns,” in Proc. 9th IFIP TC Int.
Conf. Human–Comput. Interact., vol. 3, 2003, pp. 702–712.
spent on the seven CTs is basically the same as the time to [11] N. Alsindi, R. Raulefs, and C. Teolis, Geolocation Techniques:
set up a coarse-grained fingerprint database F Mi , the position- Principles and Applications. New York, NY, USA: Springer, 2012,
ing effect is not as effective as using F Mi−CSI . When there are pp. 34–47.
[12] H. Liu, H. Darabi, P. Banerjee, and J. Liu, “Survey of wireless indoor
eight variable APs, the minimum mean localization error of positioning techniques and systems,” IEEE Trans. Syst., Man, Cybern.
the 50 TPs obtained by the crowdsourcing method is 5.5 m, C, Appl. Rev., vol. 37, no. 6, pp. 1067–1080, Nov. 2007.
which is 28% larger than the 4.3 m of the CSI algorithm. [13] P. Bahl and V. N. Padmanabhan, “RADAR: An in-building RF-based
user location and tracking system,” in Proc. IEEE Conf. Comput.
Commun. (INFOCOM), 2000, pp. 775–784.
V. C ONCLUSION [14] Y. Zhuang, Z. Syed, Y. Li, and N. El-Sheimy, “Evaluation of two WiFi
positioning systems based on autonomous crowdsourcing of handheld
In this article, we proposed the indoor localization system devices for indoor navigation,” IEEE Trans. Mobile Comput., vol. 15,
DFD-CSI to solve the device heterogeneity and fingerprint no. 8, pp. 1982–1995, Aug. 2016.
database updating issues. In view of the actuality that commer- [15] S.-H. Jung, B.-C. Moon, and D. Han, “Unsupervised learning for crowd-
sourced indoor localization in wireless networks,” IEEE Trans. Mobile
cial routers generally have dual bands routing functions in the Comput., vol. 15, no. 11, pp. 2892–2906, Nov. 2016.
2.4- and 5-GHz bands, we first use the DFD algorithm to col- [16] C. Wu, Z. Yang, and C. Xiao, “Automatic radio map adaptation for
lect fingerprint data and establish a dual-frequency database. indoor localization using smartphones,” IEEE Trans. Mobile Comput.,
vol. 17, no. 3, pp. 517–528, Mar. 2018.
It achieved a 4-m accuracy of 70.5% on average of four het- [17] R. Gao et al., “Multi-story indoor floor plan reconstruction via mobile
erogeneous devices using different fingerprint databases, while crowdsensing,” IEEE Trans. Mobile Comput., vol. 15, no. 6,
the other three methods, RSS, SSD, and AP-sequence, were pp. 1427–1442, Jun. 2016.
[18] P. Sharma, D. Chakraborty, N. Banerjee, D. Banerjee, S. K. Agarwal,
68.5%, 67.5%, and 67%, respectively. In addition, under the and S. Mittal, “KARMA: Improving WiFi-based indoor localization
status of almost equal time consumption, the localization effect with dynamic causality calibration,” in Proc. IEEE SECON, Jun. 2014,
obtained by using the coarse fingerprint database with the pp. 90–98.
[19] H. Suining, C. S.-H. Gary, Y. Lei, and L. Ning, “SLAC: Calibration-
CSI algorithm updating mode is significantly better than the free pedometer-fingerprint fusion for indoor localization,” IEEE Trans.
crowdsourcing with the LAAFU updating mode. The average Mobile Comput., vol. 17, no. 5, pp. 1176–1189, May 2018.
localization error of 50 TPs is always 4.3 m regardless of the [20] A. M. Hossain, Y. Jin, W.-S. Soh, and H. N. Van, “SSD: A robust
RF location fingerprint addressing mobile devices’ heterogeneity,” IEEE
APs with or without changes when the CSI algorithm is used. Trans. Mobile Comput., vol. 12, no. 1, pp. 65–77, Jan. 2013.
Our future work will mainly focus on using WiFi-6.0 for [21] W. Cheng, K. Tan, V. Omwando, J. Zhu, and P. Mohapatra, “RSS-ratio
high-speed indoor localization and continue to strengthen the for enhancing performance of RSS-based applications,” in Proc. IEEE
Conf. Comput. Commun. (INFOCOM), Apr. 2013, pp. 3075–3083.
usability of dual-frequency bands relationships. We will ana- [22] C. Oestges et al., “Experimental characterization and modeling of
lyze the component of fingerprint data composed of 2.4 and outdoor-to-indoor and indoor-to-indoor distributed channels,” IEEE
5 GHz, and their differences, to select an appropriate finger- Trans. Veh. Technol., vol. 59, no. 5, pp. 2253–2265, Jun. 2010.
[23] M. Youssef and A. Agrawala, “The HORUS location determination
print when it is in the online phase. Meanwhile, more effective system,” Wireless Netw., vol. 14, no. 3, pp. 357–374, Jun. 2008.
interpolation algorithms will be explored in a large indoor [24] K. Lin et al., “Enhanced fingerprinting and trajectory prediction for IoT
environment conducted to enhance the fingerprint database localization in smart buildings,” IEEE Trans. Autom. Sci. Eng., vol. 13,
no. 3, pp. 1294–1307, Jul. 2016.
updating. [25] L.-H. Chen, E. H.-K. Wu, M.-H. Jin, and G.-H. Chen, “Homogeneous
features utilization to address the device heterogeneity problem in fin-
R EFERENCES gerprint localization,” IEEE Sensors J., vol. 14, no. 4, pp. 998–1005,
Apr. 2014.
[1] E. P. Enge and P. Misra, “Special issue on global positioning system,” [26] S.-H. Fang and C.-H. Wang, “A novel fused positioning feature for han-
Proc. IEEE, vol. 87, no. 1, pp. 3–15, Jan. 1999. dling heterogeneous hardware problem,” IEEE Trans. Commun., vol. 63,
[2] N. B. John, J. Heidemann, and D. Estrin, “GPS-less low cost outdoor no. 7, pp. 2713–2723, Jul. 2015.
positioning for very small devices,” IEEE Pers. Commun. Mag., vol. 7, [27] J. Jun et al, “Low-overhead WiFi fingerprinting,” IEEE Trans. Mobile
no. 5, pp. 28–34, Oct. 2000. Comput., vol. 17, no. 3, pp. 590–603, Mar. 2018.
[3] P. A. Karegar, “Wireless fingerprint indoor positioning using affin- [28] S. He, W. Lin, and S.-H. G. Chan, “Indoor localization and automatic fin-
ity propagation clustering methods,” Wireless Netw., vol. 3, pp. 1–9, gerprint update with altered AP signals,” IEEE Trans. Mobile Comput.,
Apr. 2017. vol. 16, no. 7, pp. 1897–1910, Jul. 2018.
[4] A. Yassin et al., “Recent advances in indoor localization: A survey on [29] E. Khorov, A. Kiryanov, A. Lyakhov, and G. Bianchi, “A tutorial on
theoretical approaches and applications,” IEEE Commun. Surveys Tuts., IEEE 802.11ax high efficiency WLANs,” IEEE Commun. Surveys Tuts.,
vol. 19, no. 2, pp. 1327–1346, 2nd Quart., 2017. vol. 21, no. 1, pp. 197–216, 1st Quart., 2019.
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.
11194 IEEE INTERNET OF THINGS JOURNAL, VOL. 7, NO. 11, NOVEMBER 2020
[30] N. Patwari, A. O. Hero, M. Perkins, N. S. Correal, and Junhua Yang received the B.Eng. degree in elec-
R. J. O’Dea, “Relative location estimation in wireless sensor tronic communication engineering from Sichuan
networks,” IEEE Trans. Signal Process., vol. 51, no. 8, pp. 2137–2148, Normal University, Chengdu, China, in 2008, the
Aug. 2003. M.Eng. degree in signal and information process-
[31] S. R. Saunders and A. Aragon-Zavala, Antennas and Propagation for ing engineering from the Chengdu University of
Wireless Communication Systems, 2nd ed. Hoboken, NJ, USA: Wiley, Information Technology, Chengdu, in 2011, and
2007, pp. 18–26. the Ph.D. degree in information and communica-
[32] P. Brigger, J. Hoeg, and M. Unser, “B-spline snakes: A flexible tool tion engineering from Northwestern Polytechnical
for parametric contour detection,” IEEE Trans. Image Process., vol. 9, University, Xi’an, China, in 2019.
no. 9, pp. 1484–1496, Sep. 2000. He is currently a Lecturer with the School of
[33] S. Shao, A. Khreishah, and I. Khalil, “Enabling real-time indoor tracking Electronic Engineering, Xi’an University of Posts
of IoT devices through visible light retroreflection,” IEEE Trans. Mobile and Telecommunications, Xi’an. His current research interests include indoor
Comput., vol. 19, no. 4, pp. 836–851, Apr. 2020. localization, wireless communication, and machine learning.
Authorized licensed use limited to: Monash University. Downloaded on January 14,2021 at 11:50:16 UTC from IEEE Xplore. Restrictions apply.