![](https://csdnimg.cn/release/download_crawler_static/15770906/bg3.jpg)
Z. Li et al.: NB Indoor Positioning System by Fusing Time and RSS via Ensemble Learning
in a limited targeted subarea of these IoT networks to
support location based services for third-party providers
of positioning services.
• Improvement of the Positioning Accuracy: We evalu-
ate our positioning system in two testing scenarios. The
experimental results reveal that the system achieves a
mean positioning accuracy of 1.61m, which outperforms
the traditional RSSI-based positioning by 36.1%.
In the remainder of the paper, related work is reviewed in
Section II. Section III discusses the problems and challenges
for indoor positioning with narrow-band signals. Our main
contributions of the proposed passive indoor positioning sys-
tem for narrow-band signals by fusing RSS and DTDOA via
an enhanced pattern matching algorithm are introduced in
Section IV. Section V describes the experiments and evalua-
tion results. Finally, Section VI concludes the paper.
II. RELATED WORK
In this section, we summarize some previous research on
indoor positioning, especially for time-based positioning and
fingerprinting approaches which are related to our work.
A. TIME-BASED POSITIONING
Time-based positioning often relies on calculating the prop-
agation distances from the target to Anchor Nodes (ANs)
based on the propagation time of radio signals, which is
referred to as ranging. The accuracy of time-based ranging
is often challenged by time synchronization among ANs
and the accuracy of timestamps. Due to the high propa-
gation speed of radio signals, time-based ranging requires
strict time synchronization among ANs to achieve high accu-
racy. In our previous work [34], we show that the measured
GPS synchronization offsets vary over time depending on
the reception conditions of the GPS signals. The maximal
measured synchronization offset is up to 171ns resulting in
a positioning error of 51m. The accuracy of timestamps is
highly dependent on signal bandwidth. For example, due to
its wide bandwidth, UWB signals can achieve nanosecond
accuracy for sample-based timestamps [22]–[24]. For WiFi
signals with a bandwidth of 20MHz, Makki et al. [35] inves-
tigate how to locate a WiFi target based on time-based rang-
ing. They only demonstrate that time-based ranging using
WiFi signals is able to achieve a median accuracy of 1m
in an environment with Line Of Sight (LOS) connection
and little multipath propagation. However, in an environment
with severe influence of multipath propagation, the posi-
tioning accuracy significantly deteriorates. In our previous
work [27], we investigate the feasibility of time-based posi-
tioning for narrow-band signals, i.e., IEEE 802.15.4 with a
bandwidth of 2MHz. We propose a physical layer times-
tamp based on symbol timing recovery for narrow-band sig-
nals which achieves nanosecond resolution. We also find
that complex indoor propagation significantly limits rang-
ing accuracy. The error sources include multipath prop-
agation and shadowing. In narrow-band systems, signals
from different propagation paths overlap in one symbol
duration, making the detection of the direct propagation path
nontrivial [27], [35].
B. FINGERPRINTING
Due to severe multipath effects, RSSI-based fingerprinting
is widely adopted in indoor environments. RADAR [15]
presents an early fingerprinting system comprised of offline
and online phases. In the offline phase, RSSIs at multiple
ANs are measured to build a RSSI map. In the online phase,
the KNN algorithm is used to match the position of a target
device to the RSSI map. HORUS [16] proposes to take a
probabilistic algorithm in the online phase to improve the
positioning accuracy.
Recently, some fine-grained information at the physical
layer rather than RSSI has been investigated to improve the
accuracy of fingerprinting. In [36], CIR (Channel Impulse
Response) is analyzed with 200MHz bandwidth and neural
network training is used to improve the positioning accuracy.
FIFS [18] is designed to use CSI for fingerprinting with
WiFi signals. With these fine-grained channel information,
the performance has impressively improved compared to
RSSI. However, CSI approaches for fingerprinting only work
for Orthogonal Frequency Division Multiplexing (OFDM)
systems, which benefit from frequency diversity. Single car-
rier systems with narrow bandwidth cannot be supported,
e.g., ZigBee with Direct Sequence Spread Spectrum (DSSS).
Besides radio signals, some ambient physical informa-
tion has recently been proposed to improve the position-
ing accuracy of fingerprinting in the domain of ubiquitous
positioning, such as magnetic field [17], [28], indoor visible
light [29]–[31] and images of landmarks [32], [33]. Obtain-
ing these physical information requires to install additional
sensors for indoor positioning, which is normally not avail-
able for the nodes in IoT networks.
III. PROBLEM STATEMENT
In this work, we aim to provide an accurate positioning sys-
tem for IoT networks using narrow-band signals. We investi-
gate to improve positioning accuracy from two aspects. First,
we aim to investigate some location-related fine-grained
information obtained from physical layer rather than RSSI
for indoor positioning. Second, we focus on designing a fin-
gerprinting solution with efficient fusion and pattern match-
ing algorithms to exploit these fine-grained physical layer
information.
A. LOCATION-RELATED INFORMATION
FOR INDOOR POSITIONING
1) FINE-GRAINED RSS FINGERPRINTS
Received signal strength is the most often used location-
related information for indoor positioning, which is attenu-
ated with propagation distance but influenced by multipath
propagation. The most common type of RSS can be easily
got from the measured radio signals in the hardwares of
radio receivers at a per packet level, which is referred to
9938 VOLUME 6, 2018