Radio Map Efficient Building Method Using Tensor
Completion for WLAN Indoor Positioning System
Lin Ma
1,2
, Wan Zhao
1,2
, Yubin Xu
1,2
, Cheng Li
3
1
Communication Research Center, Harbin Institute of Technology, Harbin, China
2
Key Laboratory of Police Wireless Digital Communication, China Ministry of Public Security, Harbin, China
3
Electrical and Computer Engineering, Faculty of Engineering, Memorial University, St. John’s, NL, Canada
malin@hit.edu.cn
Abstract—Wireless local area network (WLAN) fingerprint-
based indoor positioning system has a wide application prospect
because it supplies good positioning performance without the
requirement of additional hardware installations. However, huge
labor and time are needed to establish the fingerprint database
called radio map. To address this problem, we propose an
efficient radio map building method using tensor completion. The
cost of the proposed method is reduced by lessening the number
of reference points (RPs) to be measured. Due to the strong
correlation between data, estimating received signal strength
(RSS) at unmeasured reference points can be formulated as a low
rank tensor completion problem. And the issue can be
transformed into a convex optimization problem by substituting
rank operation with trace norm operation. We solve the problem
by employing the alternating direction method of multipliers
(ADMM) algorithm. The experiment results indicate that the
proposed method can not only reduce the effort of radio map
building remarkably, but also achieve high positioning accuracy.
Keywords—Radio map; Low rank tensor completion(LRTC);
Efficient building; RSS;WLAN; Indoor positioning
I.
I
NTRODUCTION
Nowadays, location based service (LBS) receives extensive
attention with the rapid development of smart device and the
Internet [1]. Although global position system (GPS) has been
widely used to provide outdoor location information, it cannot
meet the demand for high accuracy in many indoor
environments because GPS signals are seriously blocked by
roofs and walls of buildings. Thus, positioning methods based
on Bluetooth [2], radio-frequency identification (RFID) [3],
ultrasound [4], ultra-wideband [5], vision [6], and wireless
local area network (WLAN) [7] are investigated in order to
establish high-precision indoor positioning system. Among
these approaches, WLAN indoor positioning method have
obvious advantages because it makes full use of the widely
available access points (APs) based on IEEE 802.11 instead of
additional hardware installations. Hence, it is the most
prominent option for the future indoor positioning and
navigation service. Actually, there are two types of WLAN
indoor positioning system. They base on geometric
measurement technique and fingerprint technique respectively.
Geometric measurement techniques such as angle of arrival
(AOA), time of arrival (TOA) and time difference of arrival
(TDOA) are easily influenced by multipath fading and non-
line-of-sight propagation. Fingerprint technique, by contrast,
has merits of complexity and accuracy.
Generally, fingerprint-based WLAN indoor positioning
system estimates user position by matching received signal
strength (RSS) obtained from two phases called offline phase
and online phase [8]. In offline phase, a fingerprint database
named radio map, which records the mapping between known
position space (coordinates of RPs) and signal space (RSS
values from all APs) is established. In online phase, each user
terminal senses RSS and employs mapping recorded in radio
map to estimate its position in real time. Usually, considerable
reference points are set when building radio map since the
accuracy of positioning depends on the density of reference
points. However, the cost of achieving a high-precision
positioning system through that way is a huge labor and time
consumption, which hinders the spread of fingerprint-based
WLAN indoor positioning system. Therefore, it is worth
researching on how to build radio map in an efficient way.
So far, a lot of researches have been done to address the
problem above. These studies can be mainly divided into two
categories. One is to build propagation model, and the other is
to do interpolation. In [9,10], propagation models are proposed
to estimate the signal patterns at each virtual reference point.
However, these models are too simple to predict detailed signal
fading and propagation under complex indoor electromagnetic
environment, which causes low positioning accuracy. On the
other hand, various spatial interpolation methods like newton
interpolation [11], linear interpolation [12], inverse distance
weighting interpolation [13], radial basis function interpolation
[14], kriging interpolation [15], and cubic spline interpolation
[16] are introduced to build radio map. These methods reduced
workload by decreasing the number of measured reference
points and estimating RSS values at unmeasured reference
points by interpolation. Although the positioning accuracy of
interpolation methods, especially cubic spline interpolation
approach in [16], are higher than propagation model method,
interpolation methods are still limited by only using the local
information to speculate signal values at unknown reference
points without making use of global information. In addition,
the result of spatial interpolation is affected by the permutation
of reference points. In other words, different interpolation
results will be obtained if reference points are sorted by
distance from x-axis, distance from y-axis, distance from
origin, or even the order in which data are reported. Hence, in
order to obtain optimal interpolation result, reference points
should be arrayed in the sequence with the strongest spatial
correlation. However, accurate spatial correlation is unknown
before interpolation, which causes interpolation result is
uncertain.
978-1-5386-3180-5/18/$31.00 ©2018 IEEE