Zhang et al.

EURASIP Journal on Wireless Communications and Networking

(2019) 2019:51
https://doi.org/10.1186/s13638-019-1371-y

RESEARCH Open Access

A novel CSI-based fingerprinting for

localization with a single AP
Lei Zhang1,2,3, Enjie Ding1,2,3, Yanjun Hu1,2,3* and Yafeng Liu1,2,3

Abstract
WiFi-based indoor localization techniques are critical for location-based services. Among them, fingerprint-based
method gains considerable interest due to its high accuracy and low equipment requirement. One of the major
challenges faced by fingerprint-based position system is that in some places there are not enough access points
(AP) to provide features for accurate location. To address that, we propose a novel fingerprint-based system using
only a single AP. We propose a novel phase decomposition method to obtain the phase of multipath provided by
a AP and use the decomposed phase as a fingerprint after the feature exaction by principal component analysis
(PCA). Performance in the laboratory, meeting room, and corridor is investigated, and our system is also compared
with a RSSI-based and a CSI-based fingerprint localization system. As the experimental results suggest, the
minimum mean distance error is 0.6 m in the laboratory, 0.45 m in the meeting room, and 1.08 m in the corridor,
outperforming the other two systems.
Keywords: WiFi network, Indoor localization, Channel state information, Phase decomposition, MIMO-OFDM

1 Introduction scenarios as being susceptible to multipath effect [11].

In outdoor environments, GPS-based localization tech- The positioning accuracy of most systems ranges from
niques make our lives more convenient in many aspects 3 to 10 m [12]. To overcome the fluctuation of RSSI,
[1, 2]. Since the GPS signals cannot penetrate efficiently Hossain et al. proposed SSD [13] that localizes nodes
indoors to localize objects [3], some other techniques using the signal strength difference. However, RSSI is
are used to replace GPS to provide accurate location in- limited as it fails to reflect the spatial, temporal, and
formation [4, 5] in indoor environment. Among the in- environmental characteristics.
door localization techniques, the WiFi-based localization Signals transmitted in WiFi networks exploit orthog-
techniques have become one of research hotspots due to onal frequency division multiplexing (OFDM) [14],
the ubiquitous WiFi infrastructure. However, in some which are transmitted and received by multiple subcar-
occasions, there may not be enough APs to implement riers. The received signal impacted by scattering, fading,
the location methods [6–8], hence it is necessary to and power loss with distance can be represented by the
study the location method based on single AP. physical layer information known as channel state
Since the received signal strength indication (RSSI) is information (CSI). By modifying driver firmware, CSI
capable of reflecting the distance between the AP and can be obtained using Intel 5300 or Atheros 9390 cards
the node, traditional WiFi-based models use RSSI as a [15, 16]. Due to multipath, subcarriers will propagate
metric. A RSSI-based model or fingerprint can be built, along different paths, which makes the amplitude and
and examples for RSSI-based localization systems are phase of each subcarrier different from each other.
DWELT system [9] as well as Horus system [10]. It is witnessed in recent years that CSI-based tech-
Nevertheless, RSSI fluctuates significantly in some niques have been widely adopted in localization. In com-
parison with coarse-grained RSSI, CSI is a fine-grained
* Correspondence: 303363209@qq.com indicator containing amplitude and phase information
1
School of Information and Control Engineering, China University of Mining during signal propagation [17–20]. Moreover, similar to
and Technology, Xuzhou 221000, China
2
IoT Perception Mine Research Center, China University of Mining and RSSI-based techniques, model-based techniques are also
Technology, Xuzhou 221000, China used. LiFS using the subcarriers with minimal
Full list of author information is available at the end of the article

© The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made.
Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 2 of 14

interference by multipath is designed by Wang et al. [21] 2 Related work

for accurate localization of fingerprint. The combination Increasing demand of location-based service (LBS) draws
of the amplitude with spatial diversity is used by Wu et a great deal of attention to WiFi-based localization in in-
al. [22] for indoor location. Furthermore, to achieve door environments. We catalog this into two classes,
higher accuracy by obtaining high data dimension, FIFS device-free localization and device-based localization.
[23] and CSI-MIMO [20] systems, both aggregate ampli-
tude and phase information of all subcarriers of all an- 2.1 Device-free localization
tennas for indoor location fingerprinting. Angle of As it is not convenient for a subject to carry a device,
arrival (AOA) estimation is also applied in CSI-based Jie et al. [25] implement a localization system with
localization such as ArrayTrack [24] whose key method non-equipped entity. Some other systems [26–30] have
is that make the number of antennas greater than the been proposed which constitute fingerprint-based or
number of paths, and then deduces AOAs using MUSIC model-based solutions. Model-based algorithms, which
algorithm. By increasing the number of APs or the num- do not require any laborious effort to build and main-
ber of antennas, CSI-MIMO and ArrayTrack achieve tain a radio map, estimate the distance from the object
high positioning accuracy, respectively. In practice, most to an AP by using statistical models. LiFS [21] achieves
of today’s WiFi networks deployment is primarily for the accuracy of about 1 m in indoor environments by
communications rather than for positioning. Thus, there using selected subcarriers of CSI to build a model. Due
may be insufficient APs to provide CSIs to generate a to the changes of direction, reflecting, or scattering sig-
fingerprint, and increasing the number of antennas re- nals, it is hard to obtain an accurate relationship be-
sults in additional hardware costs which may not be tween the signal and the location. Fingerprint-based
available in practical applications as well. algorithms, however, need no prior knowledge of the
In the present study, we try to advance CSI-based in- relationship between the distance and signals, gain
door localization to achieve centimeter-level localization much attention in the localization systems. General
resolution only by using a single commercial AP under fingerprint-based technique has two phases named the
the phase of multipath offered by CSI of subcarriers offline and localization phase. Offline phase is to collect
using multiple input multiple output (MIMO) to derive signals for generating fingerprints from every spot of an
the location signatures. interested area to build a radio map. During the
The main contributions of this paper are as follows: localization phase, observed fingerprint is matched
against the radio-map by using matching algorithms.
1. We design a method to obtain phases of multipath by Zhou et al. [31] establish the nonlinear relationship be-
decomposing construction CSI matrix of a single AP. tween CSI fingerprints during the offline phase and es-
2. We propose to use phase information of multipath timate locations through SVM regression. Unlike
for indoor fingerprinting. To the best of our fingerprinting localization, MaTrack [32] uses
knowledge, this is the first work to leverage phase Dynamic-MUSIC method to identify the angles for
information of a single AP for indoor fingerprinting. localization through detecting the subtle reflection sig-
3. We analyze the effectiveness of the proposed location nals from the human body.
signature using deterministic k-nearest neighbor
(KNN), probabilistic Bayes algorithms, and support 2.2 Device-based localization
vector machine (SVM) in the experimental scenarios. Due to security or commercial reasons, in some scenar-
4. We investigate impact factors on the localization ios, we not only need the position but also the identity
accuracy of our system such as the number of information of a person in which the device-free
decomposition path, the size of training and localization system is not suitable. Device-based
estimation samples, the number of APs, and the localization system can provide identity information as
size of the cell. human carrying specialized equipment. Compared to
5. We evaluate the performance of our system with using RSSI of WiFi for localization such as Horus [10]
RSSI-based and CSI-based localization system. and RADAR [33], CSI-based localization systems pro-
vide higher accuracy. PILA [18] designs an algorithm to
The rest of this study is structured as follows. The re- identify direct path from multipath and uses MUSIC al-
lated work is briefly presented is in Section 2. Section 3 in- gorithm to jointly estimate the AOA and time of arrival
troduces preliminary study including CSI. In Section 4, (TOA) of each path. PhaseFi [34] implements a scheme
our proposed localization system is detailed. The evalu- in which the calibrated phase information is extracted
ation results of our system are discussed in Section 5. The for fingerprinting in the offline stage and a deep learn-
conclusion is drawn and the future research is directed in ing approach is used for location estimation in the
the final Section. localization stage. As the bandwidth of WiFi is limited,
Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 3 of 14

ToneTrack [35] increases the effective bandwidth to ob- 3.2 MIMO

tain the TOA of APs by leveraging the frequency-agile MIMO technology using multiple transmitting and re-
technique. Similar to ToneTrack, CHRONOS [36] uses ceiving antennas greatly increases the capacity and
Chinese remainder theorem to measure the time of fly spectrum utilization of the communication systems
(TOF) across different channels and then estimate dis- without increasing the bandwidth. Due to the multipath
tance between devices and AP through leveraging the effects, it is difficult to achieve high-speed data transmis-
TOF measurements. In addition to considering position- sion in cluttered indoor environments [39]. However, in
ing accuracy, positioning efficiency is also of great import- MIMO systems, multipath effects can serve as a favor-
ance to localization systems. For instance, CRIL [37] able factor, as the signal among different transmitting
accounts for dynamic environments introduces coupled antennas uses multipath propagation. Moreover, if the
RSSI and INS localization system to improve the efficiency antennas of both transmitting and receiving are inde-
of location estimation. pendent, the data transmission will be improved in these
parallel spatial channels created by the MIMO system.
Set t and r as the number of transmitting and receiving
3 Preliminaries antennas, respectively. Each transmitting and receiving
3.1 CSI antenna pair is comprised of a data-stream. The CSIs
In wireless communication systems, CSI reveals the data of all antennas are defined as follows:
channel response properties of transmission links,
which includes the combined effect of scattering, fad-
ing, and power decay with distance. In most WiFi net-
works, multiple input multiple output (MIMO) and
OFDM technologies are employed to transmit data ð3Þ
more effectively. In these networks, CSI represents the
subcarriers properties of each transmitting and receiv-
ing antennas. Halperin et al. [38] released a tool by
which both the amplitude and phase measurements of
each MIMO-OFDM subcarrier can be aggregated from where the CSIrt is a vector denoting the CSI of 30 sub-
Intel’s IWL 5300. carriers between the transmitting antenna t and receiv-
According to the bandwidth of WiFi, it is considered ing antenna r.
as a narrowband flat fading channel, and the channel
model is defined as follows:
4 Motivation and localization system design
y ¼ CSIx þ n ð1Þ
In this section, we investigate the challenges posed by
CSI phase measurements to improve the location esti-
where y and x denote the received and transmitted signal
mation. In accordance with (2), CSI is the superpos-
vectors, respectively, CSI denotes the channel frequency
ition of multipath signals. If there is only a direct
response, and n is the additive white Gaussian noise.
path, phase can serve as a fingerprint for localization.
In 802.11n standard, 56 or 114 subcarriers are used for
However, in practical indoor environments, indoor
data transmission in the OFDM system. Each antenna of
propagation is dominated by severe multipath as
the IWL 5300 exports 30 out of all subcarriers and the
shown in Fig. 1, and raw phase information is not
index of subcarriers varies with bandwidth and grouping
suitable for generating fingerprints. To use phase as a
Ng. The CSIk denotes the channel frequency response of
fingerprint, we design a method to obtain the phase
the subcarrier k shown as follows:
from multipath.
The architecture of the system is designed with four
X
L
CSI k ¼ αl e− j2πð f 0 þkΔf Þτl ð2Þ key constituent blocks, as shown in Fig. 2. We will
l¼1 describe each block in detail in the following
subsections.
where αl and τl denote the signal magnitude and the
TOF of path l, respectively. f0 is the central frequency,
4.1 Overview
Δf is the frequency interval of adjacent subcarriers, and
The system is built on a commercial off-the-self WiFi in-
the 30 subcarriers are written as follows:
frastructure without additional hardware. It has two
phases, training and estimation. The overview position-
CSI ¼ ½CSI 1 ; CSI 2 ; …CSI 30 ; ð3Þ ing process is as follows:
Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 4 of 14

Fig. 1 Signal propagation

4.1.1 Training phase 4.2 Outlier removal

In training phase, a mobile device is placed at a known lo- As the outliers appear during our collection due to the
cation for the collection of CSI data. After outlier removal, indoor environmental noise, which is capable of affecting
a phase sanitation block is used to calibrate phase, and the localization system performance. Consequently, we
then the phase information of the multipath is obtained implement Pauta Criterion [40] to remove these biased
through a phase decomposition block, which is presented measurements, as shown in the following:
later in this section. A unique fingerprint of one location
is generated using the feature extraction method based on  
V m ¼ X m −X  > 3S ð4Þ
dimensionality reduction. This work is repeated on each
location until we store the whole fingerprint coupled with vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the coordinates in the fingerprint database. uX n  2
u
u X i −X
t
i¼1
S¼ ð5Þ
4.1.2 Estimation phase n−1
During the online phase, we randomly select some un-
known locations to collect CSI data. Similar to the train- where Xm denotes the one of all the measurement, and
ing phase, the collected CSI data is processed using the X is the median. We declare Xm as an outlier if it meets
same method to generate a position fingerprint. A fin- the condition according to (4). The CSIs of four subcar-
gerprint matching algorithm is employed to compare it riers before and after performing outlier removal are
with the location fingerprint database to estimate the shown in Fig. 3. It is observed that the Pauta Criterion
object location. works well to remove the significant outliers.

Fig. 2 System architecture

Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 5 of 14

Fig. 3 Outlier removal. Each red point in this figure denotes the CSI amplitude of the subcarrier of each packet, and the blue circle is the outlier
found by applying Pauta criterion

30 N
4.3 Phase sanitization 1 X d k¼ 1
X
b¼ ∠CSI ∠CSI k þβþZ ð9Þ
The hardware imperfection makes it impossible to gain 30 i¼1 30 k¼1
the genuine CSI phase. Practically, there are two major
causes of the measurement errors. One is the carrier By subtracting the linear term ank + b from the raw
frequency offset (CFO) as the center frequency is not phase, we obtain the calibrated values as follows:
synchronized between the transmitter and receiver. The
sampling frequency offset (SFO) generated by N
non-synchronized clocks of ADC makes the measure- g k ¼∠CSI k − ∠CSI N −∠CSI 1 1 X
∠CSI nk ‐ ∠CSI k ð10Þ
ment errors different for different subcarriers. The mea- nN ‐n1 30 k¼1
sured phase of subcarrier k is written as follows:
The CSI phase after sanitation is shown in Fig. 4a.
d k ¼∠CSI k þ2π nk Δt k þβþZ
∠CSI ð6Þ Calibrated phases are less fluctuant and more concen-
N trated, in comparison with the raw phases which scatter
where the ∠CSId k and the ∠CSIk denote the measured and randomly over many different angles. The calibrated
the genuine values of the subcarrier k, respectively. nk is phases for about 80 packets at 3 different locations are
the index of sub-carriers, and Δtk indicates the time offset shown in Fig. 4b. As can be seen, the calibrated phases
due to SFO. N is the FFT size from the IEEE 802.11n spe- are stable at one given location and differing in different
cification. β is the unknown phase offset for CFO, and Z is locations, suggesting that it can be very suitable for in-
the measurement noise. It is unlikely to obtain the true door fingerprint positioning.
phase information as the unknown Δtk and Δtk. To elim-
inate phase offset Δtk and Δtk, the linear transformation 4.4 Phase decomposition
algorithm [41] is used to achieve this goal. Compared with RSSI, CSI is a fine-grained information,
The offset b and the slope a are defined as follows: yet the CSI value of each subcarrier is also a superpos-
d N ‐∠CSI
∠CSI d 1 ∠CSI N ‐∠CSI 1 2π ition of some paths due to the bandwidth of WiFi net-
a¼ ¼ ‐ Δt ð7Þ work. In this part, we strive to acquire the phase
nN ‐n1 nN ‐n1 N
30 N N
1 X d 1 X 2πΔt X information of as many paths as possible by designing a
b¼ ∠CSI k ¼ ∠CSI k þ nk þβþZ ð8Þ
30 i¼1 30 k¼1 30N nk ¼1 phase decomposition method.
Set CSI0 as the subcarrier of index 0. According to 2),
In the OFDM system, the symmetric frequency of the CSIk is a linear combination of the phase of each path
P
subcarriers leads Nnk ¼1 nk ¼ 0. Then, the b is written as and the phase of subcarrier 0. To simplify the notations,
follows: CSIk is represented as follows:
Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 6 of 14

(a) (b)

Fig. 4 a Raw phase vs. calibrated phase. The orange line in is the raw phase value of each packet and the blue line is presented as calibrated
phase after linear transformation algorithm. b Calibrated phase at three different locations. Red, blue, and yellow line illustrate the calibrated
phase of subcarriers at different locations, respectively

X
L Where
CSI k ¼ S l0 Δlk ð11Þ
l¼1 2 3
1 1 1 1 1
For the path l: 6 Δ14 Δ24 Δ34 Δ44 Δ54 7
6 1 7
V¼6
6 Δ18 Δ28 Δ38 Δ48 Δ58 77 ð17Þ
S l0 ¼ αl e− j2π f 0 τl 4Δ Δ212 Δ312 Δ412 Δ512 5
ð12Þ 12
Δlk ¼ e−jkΔ f τl Δ116 Δ216 Δ316 Δ416 Δ516
In the 802.11n standard, when BW = 40 MHz and Ng
= 4 (grouping), the index of the sampled subcarriers And
ranges from − 58 to 58, in which the interval of sub-
 
carrier index is 4. For the convenience of presentation, S0 ¼ diag S 1−29 ; S 2−29 ; S 3−29 ; S 4−29 ; S 5−29 ð18Þ
assuming that there are five paths here, and a Hankel
matrix can be built:
2 3 Thus, X = (VS′)S(VS′)T since S′ = (S′)T X is presented
CSI −58 CSI −54 CSI −50 CSI −46 CSI −42 as follows:
6 CSI −54 CSI −50 CSI −46 CSI −42 CSI −38 7
6 7
X¼6 7
6 CSI −50 CSI −46 CSI −42 CSI −38 CSI −34 7 X¼VΣVT ð19Þ
4 CSI −46 CSI −42 CSI −38 CSI −34 CSI −30 5
CSI −42 CSI −38 CSI −34 CSI −30 CSI −26
It is easy to prove that
ð13Þ
2 3
Actually, X is written as follows: 1 1 1 1 1
6 Δ14 Δ24 Δ34 Δ44 Δ54 7
X ¼ ΔSΔT ð14Þ 6 1 7
V¼66 Δ18 Δ28 Δ38
2 3
Δ48 Δ58 77
4Δ Δ12 Δ12 Δ412 Δ512 5
Where 12
2 3 Δ116 Δ216 Δ316 Δ416 Δ16
5
2 3
Δ1−29 Δ2−29 Δ3−29 Δ4−29 Δ5−29  1 1  12 1  13 1  14 1  15 1
6 Δ1 Δ2−25 Δ3−25 Δ4−25 5 7
Δ−25 7 6 Δ14 Δ Δ Δ Δ 7
6 −25 6  2  4 2  43 2  44 2  45 2 7
Δ¼6
6 Δ−21
1
Δ2−21 Δ3−21 Δ4−21 Δ5−21 7
7 ð15Þ 6 Δ1
¼ 6  4  Δ4 
2
Δ Δ Δ 7 ð20Þ
4 Δ1 Δ2−17 Δ3−17 Δ4−17 Δ5−17 5 6 1 3 3  43 3  44 3  45 3 7
7
−17 4 Δ4 Δ2 Δ Δ Δ 5
Δ1−13 Δ2−13 Δ3−13 Δ4−13 Δ5−13  1 4  42 4  43 4  44 4  45 4
Δ4 Δ4 Δ4 Δ4 Δ4
And
  V is a Vandermonde matrix [42]. Therefore, the prob-
S ¼ diag S 10 ; S 20 ; S 30 ; S 40 ; S 50 ð16Þ
lem arises to find a Vandermonde decomposition of X.
We define Δ = VS′. The decomposition algorithm [43] is shown as follows:
Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 7 of 14

the dimension and yield the robust feature, principal com-

ponent analysis (PCA) is conducted for the location matrix.
For the high dimension data, PCA selects the greatest vari-
ance dimension data as the principal components.
For the location matrix V, we first calculate the median
vector of the matrix and subtract it from each row to
realize the mean-centered value. Set M and σ as the
mean-centered matrix and the convenience matrix of M,
respectively. Assuming that λi denotes the eigenvalue of
σ, and xi is eigenvector corresponding to eigenvalues.
Thus, there is a vector xi which satisfies σxi = λixi. Let T
eigenvectors and eigenvalue form matrix Φ and Λ:
Φ ¼ ½x 1 x2 ⋯ x T 
ð21Þ
Λ ¼ diag ð λ1 λ2 ⋯ λT Þ
By normalizing every eigenvector to unit magnitude
and orthogonal with each other, it is yielded that

Φ0 Φ ¼ ΦΦ0 ¼ I
ð22Þ
σ ¼ ΦΛΦ0
With r largest eigenvalues of σ, T-r projection matrix
is yielded:
 
According to phase decomposition, the phase can be Φ ¼ xp1 xp2 ⋯ xpr ð23Þ
extracted from each path. On the other hand, it cannot
For each location, we project the location matrix into
be used for the calculation of TOF as it contains some
a corresponding matrix as a fingerprint matrix:
unknown phase offset caused by ADC.
F ¼ MΦp ð24Þ

4.5 PCA-based feature extraction

Assuming that our system has N transmitting antennas 5 Experiment implementation and performance
and R receiving antennas. In line with the phase decom- metrics
position method, the dimension of a vector is N*R*L = p 5.1 Experimental equipment and scenarios
where L denotes the assumed number of paths. As each lo- Our system is evaluated in the laboratory, meeting room,
cation consists of T packets, we have T p-dimensional vec- and corridor as shown in Fig. 5. The D-Link dir-859
tor, and the size of a location matrix V is T*P which may be wireless APs and the SONY laptops mounted WiFi
a very high dimension for real time localization. To reduce Wireless Link 5300 Cards serve as transmitters and

Fig. 5 Experimental scenarios

Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 8 of 14

receivers, respectively. The APs operation in IEEE the people at every desk location with the frequent mo-
802.11n mode has three antennas. In some places close bility such as moving in and out of the place, therefore it
to AP, the transmitting antenna and receiving antenna can be a typical dynamic environment. Similar with the
are very close, the signals from different antennas at- laboratory, corridor also can be seen as a dynamic envir-
tenuating very similarly, causing the AP to actively onment since people frequently walk by.
merge similar channels. In this case, we are unable to In training phase, the areas fall into many cells with
obtain the CSI of all three antennas and the channel the size of about 0.6 m × 0.6 m and label each cell
merging process is controlled by internal program of an with a number. At each cell, we collect 1000 packets
AP that could not be intervened. Thereby, we use two of from each AP. During the estimation phase, some
the three antennas as transmitting antennas. cells are randomly selected, and 200 packets are
We select the three experimental scenarios due to dif- gained at each cell. The layout and major parameters
ferent floor areas and mobility. The scenarios can be cat- of three scenarios are shown in Fig. 6 and Table 1,
egorized into static and dynamic based on the mobility, respectively.
which depends on the number of people present and the
frequency of their mobility during experimentation. In
the meeting room, it is almost static with one people sit- 5.2 Performance metrics
ting at their place most of the time during experimenta- Two metrics are used to evaluate the performance of
tion. In contrast, the laboratory is fully occupied with our indoor localization system.

Fig. 6 Floor-plan of the three scenarios. The red points denote the sampling location. At each location, we collect 1000 packets from each AP
Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 9 of 14

Table 1 Main parameters of the scenarios median distance error (MDE) is utilized to represent the
Parameters Laboratory Meeting room Corridor overall performance of positioning system.
Size 23.6 m × 5.8 m 7.2 m × 5.8 m 31.2 m × 1.5 m
Access points 2 2 2
5.3 Overall performance
Cells 81 42 88
The most machine learning algorithms can be catego-
rized into deterministic and probabilistic algorithms, so
5.2.1 Cell estimation accuracy we selected KNN, SVM, and Bayes as matching algo-
The cell estimation accuracy (CEA) suggests the ratio of rithms. KNN and SVM are deterministic algorithms
the number of correctly estimated testing cells to all cell which use the distance between a test fingerprint and
locations during the estimation phase, which is calcu- each training fingerprint from the fingerprint database
lated as follows: to estimate the test location. Bayes algorithm was a
probabilistic algorithm that estimates the test location
1X N
by maximizing the posterior probability.
CEA ¼ I ðyi ¼ ^yi Þ ð25Þ
N i¼1 Three fingerprint matching algorithms are used to
examine the overall performance of our system in the
where N denotes the number of total testing cells. three scenarios with only one AP. The cell estimation
accuracy and median distance error in three scenarios
5.2.2 Median distance error are shown in Fig. 7. It is observed from Fig. 7 that the
The estimation phase may suggest that some testing cells CEA in laboratory is higher than that in the meeting
are being misclassified. Given such circumstances, the aver- room, while the MDE is lower than that in the meeting
age distance between the center points of the estimated cells room and among all scenarios, the corridor also reaches
and true cells are calculated, which is termed as median dis- the maximum MDE and the minimum CEA. This is pri-
tance error. Formally, the median distance error is given by marily because the furniture, the number of people, and
the frequency of their mobility in the meeting room are
1X N
less than those in the laboratory. Thus, the multipath ef-
MDE ¼ ky −^y k ð26Þ
N i¼1 i i 2 fect is small during our experiment. In comparison with
the other matching algorithms, SVM reaches the best
where ‖·‖2 denotes the L2 norm. cell estimation accuracy and the lowest median distance
Combining with the size of cell size, the CEA could be error in both scenarios. The overall performance in the
used to express the precision of positioning and the two different environments is summarized in Table 2.

Fig. 7 Cell estimation accuracy and median distance error in three scenarios. In the top figure blue, orange and yellow bar represent the cell
estimation accuracy for using KNN, Bayes, and SVM as the match algorithm in the laboratory and meeting room, respectively. In the bottom
figure, blue, orange, and yellow bar represent the median distance error for using KNN, Bayes, and SVM as the match algorithm in the laboratory,
in the meeting room and in the corridor, respectively
Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 10 of 14

Table 2 Overall performance prior probabilities, and if there is not enough data for
Scenario Performance metric KNN Bayes SVM calculating the prior probabilities, the location accuracy
Laboratory CEA 0.933 0.918 0.942 would decrease Moreover, in Bayesian algorithm, it is as-
MDE (m) 0.64 1.20 0.60
sumed that the features are independently and identi-
cally distributed; however, this assumption is often not
Meeting room CEA 0.956 0.93 0.97
valid in practice. Therefore, it can be seen from the
MDE (m) 0.58 0.83 0.45 Fig. 9, on the condition of small size of training and esti-
mation samples Bayes achieves the highest MDE com-
5.4 Impact of the number of decomposition path pared to KNN and SVM.
As the number of specific paths in the environment can- KNN calculates the Euclidean distance between the
not be accurately obtained, we can only intuitively as- features of test location and the features in the finger-
sume the number of paths in the environment. We print database to estimate location. The same as Bayes
investigate the impact of the number of decomposition algorithm, it also needs a large amount data to improve
paths, as shown in Fig. 8. It is observed that more de- the location accuracy. SVM is different from KNN, as it
composition paths do not result in higher positioning can build the model for classification only using a few
accuracy. In contrast, due to the noise in the indoor en- samples. Additionally, SVM is good at dealing with
vironment, considerable decomposition path may make high-dimensional and non-linear data.
it difficult for differentiation of each fingerprint, as the The SVM approach performs the best in all of combi-
fingerprint dimension increasing more noise is intro- nations as shown in Table 3. The best accuracy is 0.6 m
duced to the fingerprint. We believe that 2–3 decompos- in the laboratory and 0.45 m in the meeting room. They
ition paths are appropriate in a scenario where the are both reached under the combination of 1 K/200 with
multipath effect is not obvious, and 4–5 decomposition SVM algorithm. In comparison with the other two algo-
paths are appropriate in a multipath effect environment. rithms, Bayes algorithm needs more data to achieve high
accuracy as it calculates the prior and posterior probabil-
5.5 Impact of size of training and estimation samples ities. In brief, a larger amount of data improves the esti-
We have evaluated four combinations: 1 k/200, 500/100, mation of the data characteristics and is conducive to
200/50, and 50/50. Figure 9 presents the MDE of four building a more accurate model to improve the position-
combinations. Bayes algorithm needs to calculate the ing accuracy.

Fig. 8 MDE for different number of decomposition path. In the top figure, blue, orange, and yellow bar represent the median distance error of
different decomposition path by using KNN, Bayes, and SVM as the match algorithm in the laboratory, respectively. In the middle figure, blue,
orange, and yellow bar represent the median distance error of different decomposition path by using KNN, Bayes, and SVM as the match
algorithm in the meeting room, respectively. In the bottom figure, blue, orange, and yellow bar represent the median distance error of different
decomposition path by using KNN, Bayes, and SVM as the match algorithm in the corridor, respectively
Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 11 of 14

Fig. 9 MDE for different size of training and estimation samples. In the top figure, blue, orange, and yellow bar represent the median distance
error of different size of training and estimation samples by using KNN, Bayes, and SVM as the match algorithm in the laboratory, respectively. In
the middle figure, blue, orange, and yellow bar represent the median distance error of different size of training and estimation samples by using
KNN, Bayes, and SVM as the match algorithm in the meeting room, respectively. In the bottom figure, blue, orange, and yellow bar represent the
median distance error of different size of training and estimation samples by using KNN, Bayes, and SVM as the match algorithm in the
corridor, respectively

5.6 Impact of the number of APs KNN and SVM approaches are used for localization.
Although our system only uses one AP to reach ac- This level of reduction is inconsequential in some ap-
curate localization, the number of APs is also a cru- plication scenarios. However, it would be very useful
cial factor for localization. In all scenarios, two APs for Bayes to increase the accuracy whose MDE drop
are used to investigate the impact for positional ac- from 1.21 to 1.08 m in the laboratory and from 0.83
curacy. The MDE for different numbers of APs is to 0.63 m in the meeting room, respectively.
demonstrated in Fig. 10. It is observed that adding an
AP will only reduce the MDE by about 0.05 m when 5.7 Impact of the size of cell
The size of cell is also an important factor that affects
the positon accuracy. We conduct experiments in la-
Table 3 MDE for different sizes of training and estimation boratory, meeting room, and corridor respectively to
samples
analyze the impact of the sizes of cell. We select three
Scenario Sample size KNN (m) Bayes (m) SVM (m)
size of cell which are 0.3 m × 0.3 m, 0.6 m × 0.6 m, and
Laboratory 1K/200 0.64 1.21 0.6 1.2 m × 1.2 m. The results are shown in Fig. 11. As can
500/100 0.79 1.53 0.7 be seen from the figure, the size of 0.6 m × 0.6 m reaches
200/50 1.02 2.31 0.76 the minimum mean distance error. Since the variance of
50/50 1.53 6.54 1.45 the feature in a fingerprint is not small enough com-
Meeting room 1K/200 0.58 0.83 0.45
pared to the distance between two adjacent cells, if the
size is too small, the features of the fingerprint would be
500/100 0.63 1.83 0.54
very similar to that of the fingerprint of the adjacent cell
200/50 1.22 2.95 0.75 in feature space, which would make it difficult to distin-
50/50 1.38 7.36 1.31 guish. Hence, too small size would not help to improve
Corridor 1K/200 1.26 1.58 1.08 positioning accuracy, but would only increase the labor
500/100 1.31 2.03 1.11 and time during building the fingerprint database. On
200/50 1.42 5.34 1.36
the other hand, if the size is too big, the measurement
scale would be more extensive which would not be
50/50 1.49 8.91 2.06
beneficial to improve the position accuracy, too.
Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 12 of 14

Fig. 10 MDE for different number of APs. In the top figure, blue and orange represent the median distance error of different match algorithm
under the one AP and two APs conditions in the laboratory, respectively. In the middle figure, blue and orange represent the median distance
error of different match algorithm under the one AP and two APs conditions in the meeting room, respectively. In the bottom figure, blue and
orange represent the median distance error of different match algorithm under the one AP and two APs conditions in the corridor, respectively

5.8 Comparison with other localization system CSI-based system, we use the same CSI data as that in
Our localization system is compared with two our system. KNN approach is introduced to all the three
fingerprint-based systems, i.e., a RSSI-based system systems for matching fingerprints. The performance of
(RADAR) [33] and a CSI-based system (CSI-MIMO) the three systems under the conditions of one AP and
[20] in the three environments. In the experiments, the two APs are shown in Fig. 11.
RSSI used in the RADAR system is the mean signal Since an AP only provides one RSSI, the RADAR sys-
strength of the three receiving antennas and in the tem lacks features for localization whose MDE is

Fig. 11 Different size of cell. In the figure, blue, orange, and yellow line represents the experiment scenarios. Blue represents the experiment
done in the laboratory, orange line represents the experiment done in the meeting room and yellow line represents the experiment done in the
corridor. Star represents the size of cell is 0.3 m × 0.3 m, circular represents the size of cell is 0.6 m × 0.6 m, and triangle represents the size of cell
is 1.2 m × 1.2 m
Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 13 of 14

Fig. 12 Comparison with other localization system. In the top figure, blue, red, and orange bar represent the median distance error of different
localization system with one AP in the laboratory, meeting room, and the corridor, respectively. In the top figure, blue, red, and orange bar
represent the median distance error of different localization system with two AP in the laboratory, meeting room and the corridor, respectively

15.3 m in laboratory, 12.5 m in the meeting room, and Our future work will focus on more complex scenarios
16.5 m in the corridor, respectively. Two APs contrib- where multiple sensing elements are used to obtain
ute to reducing the MDE for RADAR, the MDE de- more exact information for localization. Research is on-
creases significantly in all three scenarios. Similar to going to verify whether the current system can also be
RADAR, more APs are also conducive to improving the extended to a multi-level floor.
localization accuracy for our system. With two APs, the
MDE of our system is 0.61 m in the laboratory, which Abbreviations
CSI: Channel state information; LBS: Location-based service; MIMO: Multiple
is better than 0.82 m reached by CSI-MIMO. In terms input multiple output; OFDM: Orthogonal frequency division multiplexing;
of one AP, the advantage of our system is apparent, RSSI: Received signal strength indication
reaching 0.64 m in the laboratory, 0.58 m in the meet-
ing room, and 1.12 m in the corridor. Moreover, this is Acknowledgements
The authors would like to thank the reviewers for their helpful advice.
superior over 1.53 m, 1.78 m, and 3.15 m of
CSI-MIMO system, respectively (Fig. 12). Funding
This work is supported by the National key research and development plan
under Grant (No: 2017YFC0804401).
6 Conclusion
Availability of data and materials
In this study, a fingerprint-based localization system is The data used to support the findings of this study can be obtained by
presented that only using a single AP. We use the phase emailing the corresponding author.
information for fingerprinting localization and a linear
transformation algorithm to eliminate the noise in the Authors’ contributions
All authors read and approved the final manuscript.
phase. A new phase decomposition method is proposed to
acquire the phase information of multipath, and a Competing interests
PCA-based algorithm is used to extract features for the The authors declare that they have no conflict of interests.
generation of a fingerprint. The performance in the la-
boratory, the meeting room, and the corridor is validated. Publisher’s Note
The results reveal a minimum mean distance error of Springer Nature remains neutral with regard to jurisdictional claims in
0.6 m in the laboratory, 0.45 m in the meeting room, and published maps and institutional affiliations.
1.08 m in the corridor. We also analyze the impact factors Author details
on the localization accuracy of our system and compare 1
School of Information and Control Engineering, China University of Mining
our system with a RSSI-based and a CSI-based fingerprint and Technology, Xuzhou 221000, China. 2IoT Perception Mine Research
Center, China University of Mining and Technology, Xuzhou 221000, China.
localization system. The results demonstrate that our sys- 3
National and Local Joint Engineering Laboratory of Internet Application
tem outperforms the other two systems. Technology on Mine, Xuzhou 221008, Jiangsu, China.
Zhang et al. EURASIP Journal on Wireless Communications and Networking (2019) 2019:51 Page 14 of 14

Received: 27 September 2018 Accepted: 14 February 2019 24. J. Xiong, K. Jamieson, ArrayTrack: A fine-grained indoor location system [C]//
Usenix conference on networked systems design and implementation.
USENIX Association, 71–84 (2013)
25. J. Xiao, K. Wu, Y. Yi, et al., Pilot: Passive device-free indoor localization using
References channel state information [C]//distributed computing systems (ICDCS), 2013
1. G.A. Amagsila, M.E. Cabuhat, J.E. Tigbayan, et al., A framework for mobile IEEE 33rd international conference on. IEEE, 236–245 (2013)
application of flood alert monitoring system for vehicle users using Arduino 26. M. Seifeldin, A. Saeed, A.E. Kosba, et al., Nuzzer: a large-scale device-free
device [C]//humanoid, nanotechnology, information technology, passive localization system for wireless environments [J]. IEEE Trans. Mob.
communication and control, environment and management (HNICEM), Comput. 12(7), 1321–1334 (2013)
2017 IEEE 9th international conference on. IEEE, 1–6 (2017) 27. J. Wang, L. Zhang, X. Wang, et al., A novel CSI pre-processing scheme for
2. F.A. Milaat, H. Liu, Decentralized detection of GPS spoofing in vehicular ad device-free localization indoors [C]//proceedings of the eighth wireless of the
hoc networks [J]. IEEE Commun. Lett. 99, 1–1 (2018) students, by the students, and for the students workshop. ACM, 6–8 (2016)
3. S. He, S.H.G. Chan, Wi-fi fingerprint-based indoor positioning: recent 28. G. Yang, WiLocus: CSI based human tracking system in indoor environment
advances and comparisons [J]. IEEE Communications Surveys & Tutorials [C]//measuring technology and mechatronics automation (ICMTMA), 2016
18(1), 466–490 (2017) eighth international conference on. IEEE, 915–918 (2016)
4. K. QIAN et al., Enabling phased array signal processing for mobile WiFi 29. J.W. Yang, G. Cho, Utilizing CSI to improve distance estimation precision in
devices. IEEE Trans. Mob. Comput. 17(8), 1820–1833 (2018) the indoor environment [J]. International Journal of Software Engineering
5. M.O.G.H.T.A.D.A.I.E.E. Vahideh, G. DEMPSTER Andrew, Determining the best and Its Applications 9(3), 49–56 (2015)
vector distance measure for use in location fingerprinting. Pervasive and 30. S. Shi, S. Sigg, Y. Ji, Probabilistic fingerprinting based passive device-free
Mobile Computing 23, 59–79 (2015) localization from channel state information [C]//vehicular technology
conference (VTC spring), 2016 IEEE 83rd. IEEE, 1–5 (2016)
6. S. He, S.H.G. Chan, L. Yu, et al., SLAC: calibration-free pedometer-fingerprint
31. R. Zhou, J. Chen, X. Lu, et al., CSI fingerprinting with SVM regression to
fusion for indoor localization [J]. IEEE Trans. Mob. Comput. 17(5), 1176–1189
achieve device-free passive localization [C]//a world of wireless, mobile and
(2018)
multimedia networks (WoWMoM), 2017 IEEE 18th international symposium
7. S. He, W. Lin, S.H.G. Chan, Indoor localization and automatic fingerprint
on. IEEE, 1–9 (2017)
update with altered AP signals [J]. IEEE Trans. Mob. Comput. 16(7), 1897–
32. X. Li, S. Li, D. Zhang, et al., Dynamic-music: accurate device-free indoor
1910 (2017)
localization [C]//proceedings of the 2016 ACM international joint
8. X. Tian, M. Wang, W. Li, et al., Improve accuracy of fingerprinting localization
conference on pervasive and ubiquitous computing. ACM, 196–207 (2016)
with temporal correlation of the RSS [J]. IEEE Trans. Mob. Comput. 17(1),
33. P. Bahl, V.N. Padmanabhan, RADAR: an in-building RF-based user location
113–126 (2018)
and tracking system [C]//INFOCOM 2000. Nineteenth annual joint
9. W. Su, E. Liu, A.M. Calveras Augé, et al., Design and realization of precise
conference of the IEEE computer and communications societies.
indoor localization mechanism for Wi-Fi devices [C]//KSII transactions on
Proceedings. IEEE. 2, 775–784 (2000)
internet and. Inf. Syst. 10(12), 5422–5441 (2016)
34. X. Wang, L. Gao, S. Mao, CSI phase fingerprinting for indoor localization with
10. M. Youssef, A. Agrawala, The Horus location determination system [J]. Wirel.
a deep learning approach [J]. IEEE Internet Things J. 3(6), 1113–1123 (2016)
Netw 14(3), 357–374 (2008)
35. J. Xiong, K. Sundaresan, K. Jamieson, Tonetrack: Leveraging frequency-agile
11. Z. Yang, Z. Zhou, Y. Liu, From RSSI to CSI:Indoor localization via channel radios for time-based indoor wireless localization [C]//proceedings of the
response [J]. ACM Comput. Surv. 46(2), 1–32 (2013) 21st annual international conference on Mobile computing and networking.
12. S. Yiu, M. Dashti, H. Claussen, et al., Wireless RSSI fingerprinting localization ACM, 537–549 (2015)
[J]. Signal Process. 131, 235–244 (2017) 36. D. Vasisht, S. Kumar, D. Katabi, Decimeter-level localization with a single WiFi
13. A.K.M.M. Hossain, Y. Jin, W.S. Soh, et al., SSD: a robust RF location fingerprint access point [C]//NSDI, vol 16 (2016), pp. 165–178
addressing mobile devices' heterogeneity [J]. IEEE Trans. Mob. Comput. 37. S. Cai, W. Liao, C. Luo, et al., CRIL: an efficient online adaptive indoor
12(1), 65–77 (2013) localization system [J]. IEEE Trans. Veh. Technol. 66(5), 4148–4160 (2017)
14. "802.11n-2009 - IEEE Standard for Information technology-- Local and 38. D. Halperin, W. Hu, A. Sheth, et al., Predictable 802.11 packet delivery from
metropolitan area networks-- Specific requirements-- Part 11: Wireless LAN wireless channel measurements [C]//ACM SIGCOMM computer
Medium Access Control (MAC) and Physical Layer (PHY) Specifications communication review. ACM 40(4), 159–170 (2010)
Amendment 5: Enhancements for Higher Throughput," 2009, 1–565 39. D. Tse, P. Viswanath, Fundamentals of wireless communication [M]
15. S. Sen, J. Lee, K.H. Kim, et al., Avoiding multipath to revive inbuilding WiFi (Cambridge university press, 2005)
localization [C]// proceeding of the, international conference on Mobile 40. C. Shen, X. Bao, J. Tan, et al., Two noise-robust axial scanning multi-image
systems, applications, and services. ACM, 249–262 (2013) phase retrieval algorithms based on Pauta criterion and smoothness
16. D. Halperin, W. Hu, A. Sheth, et al., Tool release: gathering 802.11 n traces constraint [J]. Opt. Express 25(14), 16235–16249 (2017)
with channel state information [J]. ACM SIGCOMM Computer 41. S. Sen, B. Radunovic, R.R. Choudhury, et al., You are facing the Mona Lisa:
Communication Review 41(1), 53–53 (2011) spot localization using PHY layer information [C]//proceedings of the 10th
17. R. Zhou, J. Chen, X. Lu, et al., CSI fingerprinting with SVM regression to international conference on Mobile systems, applications, and services.
achieve device-free passive localization [C]// IEEE, international symposium ACM, 183–196 (2012)
on a world of wireless, Mobile and multimedia networks. IEEE, 1–9 (2017) 42. M.K. Ng, R.H. Chan, W.C. Tang, A fast algorithm for deblurring models with
18. Z. Tian, Z. Li, M. Zhou, et al., Sub-meter localization using CSI from Neumann boundary conditions [J]. SIAM J. Sci. Comput. 21(3), 851–866 (1999)
commodity Wi-Fi devices [J]. Sensors 16(10), 1664 (2016) 43. D.L. Boley, F.T. Luk, D. Vandevoorde, A general vandermonde factorization of
19. X. Wang, L. Gao, S. Mao, et al., DeepFi: deep learning for indoor a Hankel matrix [C]//Int’l Lin. Alg. Soc.(ILAS) Symp. on Fast Algorithms for
fingerprinting using channel state information [C]// wireless Control, Signals and Image Processing (1997)
communications and NETWORKING conference. IEEE, 1666–1671 (2015)
20. Y. Chapre, A. Ignjatovic, A. Seneviratne, et al., CSI-MIMO: An efficient Wi-fi
fingerprinting using channel state information with MIMO [J]. Pervasive &
Mobile Computing 23, 89–103 (2015)
21. J. Wang, H. Jiang, J. Xiong, et al., LiFS: low human-effort, device-free
localization with fine-grained subcarrier information [C]// international
conference on Mobile computing and NETWORKING. ACM, 243–256 (2016)
22. K. Wu, J. Xiao, Y. Yi, et al., CSI-based indoor localization [J]. IEEE Transactions
on Parallel & Distributed Systems 24(7), 1300–1309 (2013)
23. J. Xiao, K. Wu, Y. Yi, et al., FIFS: Fine-grained indoor fingerprinting system
[C]// international conference on computer communications and networks.
IEEE, 1–7 (2012)