FIFS: Fine-grained Indoor Fingerprinting System

Jiang Xiao∗ , Kaishun Wu∗† , Youwen Yi∗ and Lionel M. Ni∗

∗ Department
of Computer Science and Engineering
Hong Kong University of Science and Technology
† National Engineering Research Center of Digital Life,
State-Province Joint Laboratory of Digital Home Interactive Applications,
School of Physics and Engineering, Sun Yat-sen University

Abstract—WLAN-based indoor location fingerprinting has
been attractive owing to the advantages of open access and high
accuracy. Most fingerprinting-based systems so far rely on the
received signal strength (RSS), which can be easily measured
at the receiver with commercial WLAN equipment. However,
RSS is a coarse value which simply measures the received
power for a whole channel. Thus, it fluctuates over time in
typical indoor environments with rich multipath effects and
not unique for a specific location. In this paper, we present
the design, implementation, and evaluation of a Fine-grained
Indoor Fingerprinting System (FIFS). FIFS explores a PHY-
layer Channel State Information (CSI) that specifies the channel
status over all the subcarriers for location fingerprinting in
WLAN. The system leverages the CSI values including different
amplitudes and phases at multiple propagation paths, known
as the frequency diversity, to uniquely manifest a location.
Moreover, the multiple antennas provides the spatial diversity
that can be further augmented in fingerprinting. We also present
a coherence bandwidth-enhanced probability algorithm with a
correlation filter to map object to the fingerprints. We conducted
experiments in two typical indoor scenarios with commercial
IEEE 802.11 NICs. The experimental results demonstrate that
the overall positioning accuracy can be improved compared with
the RSS-based Horus system.
I. I NTRODUCTION
The advance of wireless technology has fostered the flourish
of indoor location-aware applications, such as indoor naviga-
tion, warehouse management and health care, etc. With the ad-
vent of wireless communications, wireless local area networks
(WLANs) that increasingly being deployed in offices and
homes recently become a means of wireless indoor localization
technique. Due to the open access and low cost properties, it
opens an opportunity for leveraging the existing WLAN IEEE
802.11 [1] infrastructure to provide precise location estimation
in indoor environment. Many WLAN-based indoor positioning
systems [2], [3], [16] adopting ”fingerprinting” technique have
gain popularity due to higher accuracy. The fingerprinting-
based approaches typically determine the location based on
two phases: first, associating location-dependent characteris-
tics to certain locations for constructing a radio map (offline
training phase); then, mapping the characteristic of object to
the radio map to infer the location (online positioning phase).
Radio signal strength (RSS) is widely used in fingerprinting
positioning systems to signify the location-dependent char-
acteristic, such as RADAR [4] and Horus [5]. The striking
point of RSS-based fingerprinting lies in the simplicity of
deployment with no specialized hardware required at the
mobile station except the wireless network interface card
(NIC). However, we claim that the weakness of RSS-based
fingerprint stems from two aspects: First, RSS varies with
time at a fixed position [5] due to the multipath effects,
which including the reflection, diffraction and diffusion in
indoor environments. Second, RSS is a coarse measurement
of the received power at the radio frequency band. For several
locations, the RSS values may be reproducible because it
lacks of the frequency information to capture the multipath
property. Therefore, this time-varying and duplicated RSS
value describes signal characteristics inaccurately and creates
undesirable localization errors.
We argue that a reliable metric provided by commercial
NICs to improve the accuracy of indoor localization is in
need. Such metric should be more temporal stable and pro-
vide the capability to benefit from the multipath effect. In
current widely used Orthogonal Frequency Division Mul-
tiplexing (OFDM) systems, where data are modulated on
multiple subcarriers in different frequencies and transmitted
simultaneously, we have a notation from PHY layer that
represents the channel properties over all the subcarriers called
Channel State Information (CSI). The primary advantage of
CSI over RSSI is that this fine-grained information estimates
the channel on each subcarrier in the frequency domain. In
contrast to only one RSSI per packet, we can obtain multiple
CSI values at one time and the values stay fairly stable over
time [8], [18]. In our pervious work [18], we have built a
propagation model based on CSI. And then we use it for
precise indoor localization by eliminating the multipath effect.
However, we observe that CSIs over multi-subcarrier will have
unique signatures in different locations. These unique features
come from the frequency diversity of CSI, which has different
amplitudes and phases of each subcarrier. By exploiting the
frequency diversity, we can construct a unique “fingerprinting”
indicating each location on the radio map. Motivated by this,
it is favorable to leverage the CSI for location fingerprinting
and thus improve the localization accuracy.
Moreover, MIMO technique that exploits the space dimen-
sion to improve capacity, range and reliability of wireless
systems is widely applied nowadays (e.g., 802.11n, WiMax,
3GPP LTE, etc.). The authors in [6] investigated the impact
of applying multiple antennas on indoor location systems. By
employing multiple antennas, the signal strength variability
can be reduced due to small scale fading compensation. Like-
wise, the stability of a localization system can be enhanced.

Authorized licensed use limited to: University of Science & Technology of China. Downloaded on October 03,2023 at 03:39:53 UTC from IEEE Xplore. Restrictions apply.
978-1-4673-1544-9/12/$31.00 ©2012 IEEE
 OFDM TX
Data in
Since the commercial 802.11n NICs in the current market are

Inverse FFT
Coding & s(t)
mostly equipped with multiple antennas, the instinctive spatial

ĂĂ

ĂĂ
S/P

P/S
Mod CP D/A
Interleave
diversity of MIMO lays solid foundations of further enhancing
the localization accuracy. Subcarriers

Channel
fading x
Based on CSI, in this paper, we present an indoor location
fingerprinting scheme. To achieve high accuracy and low noise +
complexity indoor localization, our approach composes of two
parts: first, we process the raw CSI values from measurement

Equalization
Channel
r(t)

ĂĂ
by integrating frequency diversity and spatial diversity, and Decoding & Remove

ĂĂ

FFT
P/S

then build up a radio map; second, we determine the position
of object by correlation calculation augmented with a proba-
bility algorithm.
In summary, the main contributions are as follows.
1) To the best of our knowledge, it is the first time to
take advance of the combination of the fine-grained
PHY layer information CSI with frequency diversity
and multiple antennas with spatial diversity for indoor
location fingerprinting.
2) We carefully design the architecture of FIFS which
consists of two key components: fingerprints generation
and position estimation. First, we process the raw CSI
values by leveraging the diversity in time and spatial
dimensions and store them in the fingerprints database
(radio map). Second, to map the target object to the
radio map, a coherence bandwidth-enhanced probability
algorithm with a correlation filter is proposed.
3) We implement FIFS in commercial IEEE 802.11 NICs.
Experimental results demonstrate that the CSI-based
fingerprinting provided by FIFS can improve the lo-
calization accuracy, and outperform the corresponding
traditional RSSI-based approach.
The rest of this paper is organized as follows. In Section II,
we introduce some preliminaries. Section III presents the
architecture of the FIFS system. This is followed by the
methodology of the CSI-based fingerprinting in Section IV.
The implementation of FIFS and experimental evaluations are
presented in Section V. We summarized the existing work on
fingerprinting in Section VI. Finally, conclusions are presented
and suggestions are made for future research in Section VII.
II. P RELIMINARIES
In this section, we start with an overview of the widespread
OFDM technique in WLAN. Then, we introduce the CSI value
as the fundamental component of our work.
A. Orthogonal Frequency Division Multiplexing
Recently, a worldwide convergence has occurred for the use
of Orthogonal Frequency Division Multiplexing (OFDM) as
a bandwidth-efficient technology for high data rates wireless
communications. It has been endorsed in leading standards
such as IEEE802.11a/g/n, WiMAX, LTE. OFDM is a broad-
band multicarrier modulation scheme combined with multi-
plexing. At the OFDM transeiver, the incoming data stream
is split onto multiple narrow and orthogonally overlapped
subcarriers as depicted in Fig. 1. The data on each subcarrier
is then modulated and converted back to the time domain by
S/P
Demod A/D
Deinterleave CP
Data out
OFDM RX
Fig. 1: OFDM Framework.
an inverse Fast Fourier Transform (IFFT). After Parallel to
Serial (P/S) and Digital to Analog Conversion (DAC) process,
the signals are sent through a frequency-selective channel.
Upon receiving the signals, the receivers sample them and
pass them on to a demodulation block as well as digitize
them using analogue-to-digital converter (ADC). Afterward
FFT procedure processes the data sample blocks to convert
back into the frequency domain.
B. Channel State Information
Channel measurement at the subcarrier level becomes avail-
able based on OFDM in wireless communication. Nowadays,
the measured channel state are widely utilized for adapting or
allocating the transmitter resources [7], [8].
Channel State Information (CSI) also knwon as Channel
Status Information is information that estimates the channel
by representing the channel properties of a communication
link. To be more specifically, CSI describes how a signal
propagates from the transmitter(s) to the receiver(s) and reveals
the combined effect of, for instance, scattering, fading, and
power decay with distance. In summary, the accuracy of CSI
greatly influences the overall OFDM system performance.
In a narrowband flat-fading channel, the OFDM system in
the frequency domain is modeled as
Y = HX + N, (1)
where Y and X are the received and transmitted vectors,
respectively, and H and N are the channel matrix and the
additive white Gaussian noise (AWGN) vector, respectively.
To successfully decode the message X from received signal
Y distorted by fading and noise, we should estimate the chan-
nel distortion first with some symbols known as preambles
or pilots. Thus, CSI of all subcarriers H can be estimated
according to (1) as
Y
Ĥ = . (2)
X
Generally, there are many sophisticated algorithms like
maximum-likelihood (ML), minimum mean square error
(MMSE) to estimate the CSI precisely. Therefore, comparing
with RSSI, CSI is a fine-grained value from the PHY layer that
describes the channel gain from TX baseband to RX baseband.

Authorized licensed use limited to: University of Science & Technology of China. Downloaded on October 03,2023 at 03:39:53 UTC from IEEE Xplore. Restrictions apply.

RF Signal
TX RX
Fingerprints Generation Position Estimation
No
Process CSI Calibration?
Fingerprints
Yes
Collect CSI Mapping
Algorithm
Channel
Estimation Radio Map
(Fingerprints
Database)
OFDM
Demodulator Positioning
Fig. 2: FIFS Architecture.
III. A RCHITECTURE
In this section, we describe the architecture of FIFS as
shown in Fig. 2. FIFS has the following two components.
A. Fingerprints Generation Block
During the fingerprints database construction, there exists
two important modules including CSI collection module and
CSI processing module on the mobile device side. Since FIFS
is built based on the current 802.11n communication system,
the transmitter end (TX–the AP) induces no modification.
Once the receiver end (RX–the target mobile device) received
a packet, it will first export the raw CSI value after the normal
demodulation process. In the designated processing module,
the CSI collected from 30 groups different subcarriers will
then be processed. As mentioned in the previous section, CSI
value is the channel matrix from RX baseband to TX baseband
which is needed for channel equalization. Therefore, there is
no extra processing overhead when obtaining the CSI infor-
mation. Afterwards, we introduce a calibration condition to
determine the outgoing of store the CSI value after processing.
The calibrated CSI will be stored in the fingerprints database.
Otherwise, it will be accessed as the input of the mapping
algorithm. This fingerprints generation block serves as the
prerequisite of the positioning block.
B. Positioning Estimation Block
In FIFS, the positioning server will response the mobile
device with the estimated position when it sends out a location
query message. In order to obtain the location information,
the positioning block must be capable of 1)calculating the
similarity between the CSI value measured at the RX and the
fingerprints database, and 2)determining the location of the
RX indicated by the corresponding fingerprint, which has the
highest possibility of the measured CSI. In this manner, the
location of the target device can be estimated by mapping the
CSI value and database.
Authorized licensed use limited to: University of Science & Technology of China. Downloaded on October 03,2023 at 03:39:53 UTC from IEEE Xplore. Restrictions apply.
IV. M ETHODOLOGY
In this section, we describe the design terminology of FIFS.
The methodology of this CSI-based location fingerprinting
approach can be broken down into two following phases.
1) Calibration Phase: First, we need to effectively process
the raw CSI value to generate fingerprints, and then
record the fingerprints corresponding to certain sample
locations for building up a radio map. Note that this is
known as the prerequisite of the ongoing online phase.
2) Positioning Phase: Second, on the basis of this effective
CSI value, in FIFS, we compare it with the CSI fields
of the entries stored in the radio map, and the position
of the object will be extracted from the radio map with
the closest match afterwards.
A. Calibration Phase
In the calibration phase, we denote each sample position
using CSI and construct a radio map in a two-step process.
1) CSI Collection: We start by using a mobile device
equipped with 802.11 NICs to receive the beacon message
from nearby APs at each sample position. The message
contains CSI that represents the channel response of multiple
subcarriers. We modify the chipset firmware and divide the
CSIs into 30 groups. Hence, N = 30 groups CSI values are
collected simultaneously at the receiver that represented as
H = [H1 , H2 , · · · , Hi , · · · , HN ]T , i ∈ [1, 30], (3)
where each subcarrier Hi is defined as
Hi = |Hi |ej sin{∠Hi } , (4)
where |Hi | is the amplitude response and ∠H is the phase
response of the ith subcarrier.
Since multiple antennas can introduce spatial diversity into
communication system, most recent standards like 802.11n
and LTE employ multiple-input-multiple-output (MIMO) tech-
nology to boost the throughput. According to information
theory, the capacity of a MIMO channel is min{M, N } times
of a corresponding channel with single antenna, where M and
N are the number of antennas at receiver and transmitter,
respectively. Note that the channel response of a specific
subcarrier of a MIMO system can be represented by a M × N
matrix, we can expect better location accuracy with additional
CSI information provided by multiple antennas, .
2) CSI-based Fingerprinting Generation: As the founda-
tion of fingerprinting approach, the measured CSI values are
processed to construct a radio map. Since most of the RF-
based fingerprint methods consider two spatial dimensions for
localization [5], we also follow the principle. Therefore, the
two-dimension physical space coordinate of a sample position
lj is lj = (lj,x , lj,y ).
To generate a radio map, we first extract the statistic
determine the number of detectable APs for a sample position.
At each reference point, we will generate a unique fingerprint
from the CSI of all APs and antennas.
Since the small-scale fading effect occurs at the level of sev-
eral wavelengths (about 12cm at 2.4GHz, and 6cm at 5GHz),
 1
AP1
AP2
0.8
Correlation
0.6
0.4
0.2
0
0 5 10 15 20 25 30
Subcarrier Spacing
Fig. 3: Channel Correlation.
and the distance between multiple antennas is typically much
larger, using multiple antennas presents the opportunity to
smooth out the effect, while maintaining the same calibration
workload required by the localization system. Consequently,
an important open question is if we should aggregate the
CSI measurements obtained from different antennas, or the
localization algorithm should use the CSI from each antenna
independently. In this paper, a simple aggregation scheme is
examined that we perform an averaging over all the antennas
at each sample position, more sophisticated investigation and
schemes are left to our future work.
Moreover, due to the multipath effect of indoor environ-
ment, the wideband channel of 802.11n can provide abundant
diversity in the frequency domain. The metric to evaluate
the frequency diversity is coherence bandwidth [10]. The
coherence of two arbitrary different subcarriers with distance
Δf is defined as
E{H(f )H(f + Δf )∗ }
ρΔf = .
E{|H(f )|2 }
where H(f ) and H(f + Δf ) are the channel responses of the
two subcarriers. As shown in Fig. 3, the channel correlation
decreases as the subcarrier spacing increases. The X% coher-
ence bandwidth is the value of Δf such that ρΔf = X%.
Typically, X is commonly set to be 50 or 90.
Suppose that the X% coherence bandwidth is Bc,X MHz,
if the distance of the i-th subcarrier and the j-th subcarrier
is no less than Bc,X in the frequency domain, these two
subcarriers can be viewed as fading independently. As a result,
the whole channel can be divided to several independently-
fading subbands. To exploit the frequency domain correlation
and diversity, in our experiment, we divided the whole 20MHz
channel into 4 subchannels each with channel bandwidth
5MHz for two reasons. First, according to Fig. 3, when the
subcarrier spacing is larger than 5MHz, the channel response
correlation is low; Second, since the adjacent channels in
802.11n is non-orthogonal, the non-overlap bandwidth is about
5MHz. As a result, the interference [20] situation on each
subband will be different. Then, the channel responses of
subcarriers within the same subchannel is averaged to reduce
the redundance.
Authorized licensed use limited to: University of Science & Technology of China. Downloaded on October 03,2023 at 03:39:53 UTC from IEEE Xplore. Restrictions apply.
Therefore, after averaging over multiple antennas and within
each subchannel, the CSI at each sample position to a specific
AP is represented by {H̃1 , · · · , H̃4 }. Then, we quantify the
power of a package, denoted as effective CSI, by adding up
the power with respect to all subchannels. Specifically, we
have
I
He = Hi |2 , i ∈ [1, 30], (6)
i=1
B. Positioning Phase
For object location estimation, the target is required to be
accurately mapped to the radio map.
Previous works show that the probabilistic approaches such
as maximum likelihood provide more accurate results than
deterministic ones do in indoor environments [5]. Therefore,
we adapt the probability model in [9] except that we use He
instead of RSS value. Similarly, we treat He observed from the
AP to the receiver at a fixed location as a Gaussian variable.
In the proposed system, we will select K best APs to calculate
the probability of the MS at each reference point. The criteria
for the best AP selection is that those APs with highest He
values, because they are more reliable. In our experiment, we
fix the K to be 3.
The selected K He values obtained by the terminal to
be located form a vector He = [He,1 , · · · , He,K ]. Then, the
position estimation problem is equivalent to finding the l that
maximize the posteriori probability P (lj |He ). According to
Bayes’ law,
P (lj )P (He |lj ) P (lj )P (He |lj )
P (lj |He ) =  = (7)
j P (l j )P (H |l
e j ) P (He )
Note that P (lj ) is the prior probability that the terminal lo-
cated at the reference point li . In [5], [9], uniform distribution
(5)
is assumed. In contrast, we will leverage the spatial correlation
of the CSI to determine the P (lj ).
Recall that CSI is a fine-grained information, after process-
ing, we can observe channel response over multiple subbands
represented by Hk = [H̃1 , H̃2 , · · · , H̃4 ]T for the kth AP. We
denote the observed CSI with normalization for each AP as
H(O) ∈ C4×K , and the CSI recorded in the radio map for
the same set of APs at position lj as H(lj ). To quantify the
similarity of the observed CSI and the stored “fingerprints”
for all the APs, we use the Pearson correlation between them
which is defined as
K
 cov(Hk (O), Hk (lj ))
ρH(O),H(lj ) = , (8)
σHk (O) σHk (lj )
k=1
where each AP is considered to be independent. Accord-
ing to the measurement, the spatial channel correlation will
decrease as the distance between the two receiver increases.
Therefore, with higher ρ, the position of the terminal will
be closer to the reference point. Then, the probability of the
terminal on each candidate point is defined as
ρH(O),H(lj )
P (lj ) = J (9)
i=1 ρH(O),H(li )
 3502
3525
3524

3519
3518

3514
3513

3510
3509

3504

3503
Servers Servers

5 10 15 20 25 30 35 40 AP6 AP1

4 9 14 19 24 29 34 39 25
AP2
10m

3523

3520
3517

3512

3508

3505
7m AP5
1 3 8 6 7 13 18 12 13 23 28 18 19 33 38 24 AP4 AP3

3511
3515
AP2
2 11 14 17 20 23

3522

3521
2 5 8

3516
7 12 17 22 27 32 37

3507
3506

3501
AP3 AP1

32.5m
3 1 6 4 9 11 16 10 15 21 26 16 21 31 36 22

11m Fig. 5: The layout of the corridor

Fig. 4: The layout of the laboratory

1 1
trace 1 trace 1
0.9
trace 2
0.9
trace 2
A. Experimental Scenarios
0.8 0.8
trace 3 trace 3
0.7 trace 4 0.7 trace 4
In our experiment, the proposed methodology was imple-
Probability

Probability

0.6 trace 5 0.6 trace 5

0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4
Normalized changes of CSI
0.5
(a) CSI
Fig. 6: Temporal Stability
where J is the size of the candidate reference points set.
Considering uncorrelated property between each AP, the
likelihood P (He |lj ) can be calculated as,
P (He |lj ) =
Since the signal strength at each reference point is modeled
as a Gaussian variable which requires less samplings than the
histogram approach [11]. At the offline phase, we can obtain
the expectation H̄e,k and variance σe,k corresponding to the
He,k , and the P (He,k |lj ) is obtained as
1 −(He,k − H̄e,k )2
P (He,k |lj ) =  exp
2πσe,k
Formally, the location estimation of the terminal is the
weighted average over the whole candidate set,
J 60 samples for both the CSI values and RSS values.

ˆl =
j B. Performance Evaluation
The performance of the proposed fingerprinting methodol-
ogy is evaluated in the following section.
V. P ERFORMANCE E VALUATION
In this section, we first introduce our experimental scenarios
and the data collection procedure of FIFS. Afterwards, we will
show the performance of our proposed CSI-based fingerprint-
ing approach by comparing against best known RSS-based
Horus system [5]. For fair comparison, we consider the con-
tinuous space estimator in Horus system without consideration
of the temporal correlation.
0.5
0.4
mented on the the TL-WR941ND router manufactured by
0.3 TP-LINK technologies CO., Ltd. as the transmitted APs. A
0.2
0.1 HP laptop served as the receiver object, which equipped
0
0 0.1 0.2 0.3 0.4
Normalized changes of RSSI
0.5 with an Intel WiFi Link 5300 (iwl5300) 802.11n NICs. We
modified the driver as in [7] to collect the raw CSI values. The
(b) RSSI
experiments were conducted in two different scenarios in the
campus of Hong Kong University of Science and Technology
as follows:
1) Research Laboratory First, we set up a testbed in a
7m × 11m research laboratory covering by three APs
as shown in Fig. 4. Three APs were fixed on the top
K of the shelters. In the offline phase, the CSI values

P (He,k |lj ). (10) were collected at 40 locations with 1.2m spacing to
k=1 build up the radio map. Specifically, each location has
the raw CSI of 60 packages in a format as described
in Section IV-A1. Simultaneously, the counterpart RSSI
values were recorded for comparison.
2) Corridor Second, we performed experiments in a cor-
ridor environment with multiple offices aside in our
academic building, which is 32.5m × 10m covering
corridors, rooms and cubicles. In this scenario, there are
. (11) 6 APs that can be detected. We collected fingerprints
2σe,k
at 28 different reference positions in this scenario, and
these 28 positions are 2m apart along the corridor as
shown in Fig. 5. At each reference point, we also took
P̄ (lj |He )lj (12)
1) Stability: To ensure the robustness of the location finger-
printing systems, temporal stability is a foremost criteria we
need to validate. We thus set out to investigate the stability
of the proposed new metric CSI and the widely applied RSSI
value in time series. Due to the coarse packet-level estimation
and easily varied by multipath effect, RSSI is well known to be
a fickle measurement of the channel gain. In particular, this
instability of RSSI induces inevitable errors in localization.
Thus, we need to figure out whether the fine-grained PHY
layer information CSI will remain in a stable manner in
practical indoor environment.

Authorized licensed use limited to: University of Science & Technology of China. Downloaded on October 03,2023 at 03:39:53 UTC from IEEE Xplore. Restrictions apply.
 1.5 4.5
FIFS, 1×1 FIFS, 1×1
4

Mean distance error (m)

Mean distance error (m)

1.25 FIFS, 2×2 FIFS, 2×2
3.5
Horus Horus
1 3
2.5
0.75
2
0.5 1.5
1
0.25
0.5
0 0
Laboratory Corridor 1 2 3 4 5 6
FIFS vs Horus Number of APs

Fig. 7: Mean distance error. Fig. 8: Mean distance error with different numbers of APs.

1 1
FIFS, 1×1 FIFS, 1×1
0.9 0.9
FIFS, 2×2 FIFS, 2×2
0.8 0.8
Horus Horus
0.7 0.7
Probability

Probability
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 0 0.5 1 1.5 2 2.5 3
Distance error (m) Distance error (m)

Fig. 9: CDF of localization error in Laboratory. Fig. 10: CDF of localization error in Corridor.

Fig. 6(a) plots the CDF of the amplitude change of CSI
between two successive packets in 5 mobile traces. It is
shown that the amplitude variance of CSI is within 15%.
The temporal variance of RSSI in corresponding traces is
much larger within 30% as presented in Fig. 6(b). Therefore,
the relatively stability for CSI is an essential advantage for
localization while comparing with the traditional RSSI in time
domain.
2) Accuracy: First, we evaluate the mean accuracy of the
proposed CSI-based probability algorithm and compare it with
that of Horus, the widely used RSSI-based fingerprinting
system. Fig. 7 presents the mean distance error obtained by
FIFS for both single antenna (FIFS, 1 × 1) and multiple
antennas (FIFS, 2 × 2) settings in two different environments.
As shown in the figure, FIFS with single antenna can achieve
the median accuracy of 0.65m, which outperforms Horus by
about 0.2m, and the gain is about 24%. And this gain can be
further improved by 8% with multiple antennas. Moreover, in
the corridor scenario, where covered by 6 APs and 3 APs were
taken into computation, the mean accuracy of our approach is
1.07m and 0.96m which are about 0.35m lower than Horus
system (around 25% accuracy gain).
In addition, we compare the two approaches concerning
different numbers of APs in corridor. Fig. 8 depicts the average
accuracy according to the amount of APs varing from 1
to 6. Since richer information to estimate the location can
be obtained from the more APs, both lines demonstrate the
accuracy improvement. In particular, our approach reduced the
mean distance error by 29% (FIFS, 1×1) and 32% (FIFS, 2×2)
on the average. Obviously, these results show the effectiveness
of the proposed CSI-based location system and indicate the
benefits from indoor environment with dense-deployed APs.
When the AP is sparse, our scheme performs much better
than the RSS-based one.
Fig. 9 illustrates the cumulative distribution function(CDF)
of localization errors in the laboratory. The fingerprints were
collected across the 28 positions in the laboratory. In our
experiments, for over 90% of data points, the error of FIFS,
1 × 1 and FIFS, 2 × 2 falls within the range of 1.3 meters
and 1.1 meters, respectively. However, the Horus can locate
objects in the range within 1.6 meters of their actual position
with 90 percent probability.
Unlike the first scenario that 3 APs and client are placed
in the same room, we also examined the corridor testbed
where the 6 APs are deployed in the multiple rooms. Fig. 10
depicts the cumulative distribution of positioning errors across
20 positions. We can easily observe that both our approach
and Horus can achieve the median accuracy less than 1.25m.
However, the accuracy improvement of our approach (for
the FIFS, 2 × 2 case)over Horus for 90% of data points is
up to 0.55m. We can conclude that our approach exhibits
a preferable property since the fine-grained and frequency
diversity nature of CSI is beneficial to improve the precision
of location fingerprinting.

Authorized licensed use limited to: University of Science & Technology of China. Downloaded on October 03,2023 at 03:39:53 UTC from IEEE Xplore. Restrictions apply.
 VI. R ELATED W ORK
Many researches have been presented by using WiFi infras-
tructure for indoor localization. Due to the severe multipath ef-
fect, fingerprinting approach based on RSS is widely adopted.
RADAR [4] is the primal location fingerprinting system that
measures RSS at multiple base stations as to offline build a
radio map. It adopts the kNN algorithm to online match the
position of the target to the radio map and achieves meadian
accuracy of 3m. LANDMARC [12] is a RFID localization
system which requires densely deployment. Horus [5] pro-
poses a probabilistic algorithm for fingerprinting to enhance
the accuracy. It applies joint clustering for classifying the
reference positions on radio map covering by a common set
of APs and then selected one cluster. Therefore, it reduces
the computational overhead for searching the radio map and
refines accuracy of position estimation as 2.1m for 90%.
RSS [13], [14], [15] is considered as the simplest and cost
effective RF measurement technique. Our work is different
from the aforementioned work in that we use fine-grained
CSI instead of RSS for fingerprinting, which benefits from
the multipath effect.
Recently, a technique called channel impulse response (CIR)
has been proposed to used for fingerprinting. It denotes
the signal variation between the sender and receiver due
to multipath effects. Location performance can be enhanced
by distinguishing CIR as an identical fingerprint instead of
erroneous RSS. In [16], [17], it leverages vector network
analyzer to obtain the CIR with 200MHz and adopts the
nerual networking training algorithm to improve the accuracy.
Different from the above works required specific hardware,
our work is more practical because it is compatible with
commodity NICs.
Our previous FILA [18] system leverages the frequency
diversity of fine-grained CSI and the refined free space path
loss propagation model to increase the accuracy of indoor
localization. This work is an nature extension in that, beneficial
from the multipath effects, we further investigate the property
of CSI that provides unique amplitude and phrase per location
for applying fingerprinting. PinLoc [19] also exploits the phys-
ical layer channel information to enhance the area granularity
accuracy by only considering the frequency diversity. How-
ever, the spatial diversity, one of the most important features
of current commercial APs, is lack of investigation. Moreover,
in our work, we determine the position of object by correlation
calculation augmented with a probability algorithm, where the
coherence bandwidth is applied to reduce the computational
complexity.
VII. C ONCLUSIONS
A novel RF-based indoor location system FIFS is presented
in this paper, which is based on fingerprinting through CSI
values obtained in real time from the PHY layer. The charac-
teristic of the propagation channel between the received object
and each AP is dynamically estimated from these fine-grained
CSI values. We collected and proceeded the CSI values as
”fingerprints” by leveraging both the frequency diversity and
Authorized licensed use limited to: University of Science & Technology of China. Downloaded on October 03,2023 at 03:39:53 UTC from IEEE Xplore. Restrictions apply.
spatial diversity to uniquely present the reference points and
generate the radio map. Moreover, we adapted a probabilistic
model to accurately map the observed CSI into the stored
”fingerprints” and use the coherence bandwidth to reduce the
complexity of the algorithm. Experimental results show that a
mean error slightly lower than 1m is obtained in an unmodified
FIFS WLAN network deployment.
VIII. ACKNOWLEDGEMENTS
This research was supported in part by Pearl River
New Star Technology Training Project, Hong Kong RGC
Grant HKUST617710, 617811, S&T Project of Guangdong
Province, China(Grant No. 2011A011302001), NSFC Grant
(No.60933011,61027009), NSFC-Guangdong Joint Fund
(U0935004, U1135003), 973 Program (2011CB302204), the
National Key Technology R&D Program (2011BAH27B01,
2011BHA16B08), and the Major Science and Technology
Projects of Guangdong (2011A080401007).
R EFERENCES
[1] M.B. Kjargaard,“A Taxonomy for Radio Location Fingerprinting,” in
Proc. of LoCA, 2007.
[2] M.B. Kjargaard, G. Teru and C. Linnhoff-Popien, “Zone-based RSS
Reporting for Location Fingerprinting,” in Proc. of IEEE PerCom, 2007.
[3] K. Azadeh, K. N. Plataniotis, A. N. Venetsanopoulos, “Kernel-based
Positioning in Wireless Local Area Networks,” in IEEE Transactions on
Wireless Communnication, 2007.
[4] P. Bahl and V. N. Padmanabhan, “Radar: an in-building RF-based User
Location and Tracking System,” in Proc. of IEEE INFOCOM, 2000.
[5] M. Youssef and A. Agrawala, “The Horus WLAN Location Determina-
tion System,” in Proc. of ACM MobiSys, 2005.
[6] K. Kleisouris, Y. Chen, J. Yang, R.P. Martin, “The Impact of Using
Multiple Antennas on Wireless Localization,” in Proc. of IEEE SECON,
2008.
[7] D. Halperin, W. J. Hu, A. Sheth, and D. Wetherall, “Predictable 802.11
Packet Delivery from Wireless Channel Measurements,” in Proc. of ACM
SIGCOMM, 2010.
[8] A. Bhartia, Y. Chen, S. Rallapalli, and L. Qiu, “Harnessing Frequency
Diversity in Wi-Fi Networks,” in Proc. of ACM MobiCom, 2011.
[9] S. Fang, T. Lin, K. Lee, “A Novel Algorithm for Multipath Fingerprint-
ing in Indoor WLAN Environments,” in IEEE Transactions on Wireless
Communnication, 2008.
[10] T. Rappaport, “Wireless Communications: Principles and Practice (2nd
ed.),” Prentice Hall PTR, J, USA. 2001.
[11] A. Haeberlen, E. Flannery, A.M.Laddand, and et al., “Practical Robust
Localization over Large-scale 802.11 Wireless Networks,” in Proc. of
ACM MobiCom, 2004.
[12] Lionel M. Ni, Y. Liu, Y. Lau, and A. Patil, “Landmarc: Indoor Location
Sensing using Active RFID,” in Proc. of IEEE PerCom, 2003.
[13] Z. Yang and Y. Liu, “Understanding Node Localizability of Wireless
Ad-hoc Networks,” in Proc. of IEEE INFOCOM, 2010.
[14] J. Park, B. Charrow, D. Curtis and et al., “Growing an Organic Indoor
Location System,” in Proc. of ACM MobiSys, 2010.
[15] Y. Cheng, Y. Chawathe, A. LaMarca, and J. Krumm, “Accuracy Chrac-
terization for Metropolitan-scale Wi-Fi Localization,” in Proc. of ACM
MobiSys, 2005.
[16] C. Nerguizian, C. Despins, and S. Affes, “Geolocation in Mines with
an Impulse Response Fingerprinting Technique and Neural Networks,”
in Proc. of IEEE TWC, 2006.
[17] S. Dayekh and et al., “Cooperative Localization in Mines using Finger-
printing and Neural Networks,” in Proc. of IEEE WCNC, 2010.
[18] K. Wu, J. Xiao, Y. Yi, M. Gao and L. Ni, “FILA: Fine-grained Indoor
Localization,” in Proc. of IEEE INFOCOM, 2012.
[19] S. Sen, B. Radunovic, R. Choudhury, and T. Minka, “Precise Indoor
Localization using PHY Layer Information,”in Proc. of ACM HotNets,
2011.
[20] K. Wu, H. Tan, Y. Liu, J. Zhang, Q. Zhang and L. Ni “Side Channel:
Bits over Interference”, in IEEE Transactions on Mobile Computing,
2012.