CSI-based Indoor Localization - Kaishun Wu Member, IEEE Jiang Xiao, Student Member, IEEE, Youwen Yi Student Member, IEEE, Dihu Chen , Xiaonan Luo ...
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. XX, NO. X, XXXX 2012 1
CSI-based Indoor Localization
Kaishun Wu Member, IEEE Jiang Xiao, Student Member, IEEE, Youwen Yi Student Member, IEEE,
Dihu Chen , Xiaonan Luo, Lionel M. Ni Fellow, IEEE
Abstract—Indoor positioning systems have received increasing attention for supporting location-based services in indoor environ-
ments. WiFi-based indoor localization has been attractive due to its open access and low cost properties. However, the distance
estimation based on received signal strength indicator (RSSI) is easily affected by the temporal and spatial variance due to the
multipath effect, which contributes to most of the estimation errors in current systems. In this work, we analyze this effect across the
physical layer and account for the undesirable RSSI readings being reported. We explore the frequency diversity of the subcarriers in
OFDM systems and propose a novel approach called FILA, which leverages the channel state information (CSI) to build a propagation
model and a fingerprinting system at the receiver. We implement the FILA system on commercial 802.11 NICs, and then evaluate its
performance in different typical indoor scenarios. The experimental results show that the accuracy and latency of distance calculation
can be significantly enhanced by using CSI. Moreover, FILA can significantly improve the localization accuracy compared with the
corresponding RSSI approach.
Index Terms—Indoor localization, Channel State Information, RSSI, Physical Layer.
✦
1 I NTRODUCTION variance of RSSIs collected from an immobile receiver in
one minute is up to 5dB. Second, RSSI is easily varied by
L O calization is one of the essential modules of many
mobile wireless applications. Although Global Posi-
tioning System (GPS) works extremely well for an open-
the multipath effect. In theory, it is possible to establish
a model to estimate the separation distance using the
received power. In reality, however, the propagation of
air localization, it does not perform effectively in indoor
a RF wave is attenuated by reflection when it hits the
environments due to the disability of GPS signals to
surface of an obstacle. In addition to the line-of-sight
penetrate in-building materials. Therefore, precise in-
(LOS) signal, there are possibly multiple signals arriving
door localization is still a critical missing component
at the receiver through different paths. This multipath
and has been gaining growing interest from a wide
effect is even more severe in indoor environments where
range of applications, e.g., location detection of assets
a ceiling, floor and walls are present. As a result, it is
in a warehouse, patient tracking inside the building of
possible for a closer receiver to have a lower RSSI than
the hospital, and emergency personnel positioning in a
a more distant one. Consequently, a simple relationship
disaster area.
between received power and separating distance cannot
A great number of researches have been done to
be established. Therefore, this time-varying and vulner-
address the indoor localization problem. Many range-
able RSSI value creates undesirable localization errors.
based localization protocols compute positions based on
We argue that a reliable metric provided by commer-
received signal strength indicator (RSSI), which repre-
cial NICs to improve the accuracy of indoor localiza-
sents the received power level at the receiver. According
tion is in need. Such metric should be more temporal
to propagation loss model [1], received signal power
stable and provide the capability to benefit from the
monotonically decreases with increasing distance from
multipath effect. In current widely used Orthogonal Fre-
the source, which is the foundation of the model-based
quency Division Multiplexing (OFDM) systems, where
localization. Most of the existing radio frequency (RF)-
data are modulated on multiple subcarriers in different
based indoor localization are based on the RSSI val-
frequencies and transmitted simultaneously, we have a
ues [1]–[5]. More related work is in the supplemental
value that estimates the channel in each subcarrier called
file. However, we claim that the fundamental reasons
Channel State Information (CSI). Different from RSSI,
why RSSI is not suitable for indoor localization are
CSI is a fine-grained value from the PHY layer which
from two aspects: First, RSSI is measured from the RF
describes the amplitude and phase on each subcarrier in
signal at a per packet level, which is difficult to obtain
the frequency domain. In contrast to having only one
an accurate value. According to our measurement in
RSSI per packet, we can obtain multiple CSIs at one
a typical indoor environment as shown in Fig. 1, the
time. More importantly, the CSIs over multi-subcarriers
will travel along different fading or scattering paths
• The authors are with Department of Computer Science and Engineering,
Hong Kong University of Science and Technology, Hong Kong. Some
on account of the multipath effects. It then naturally
preliminary results of this work were presented in the Proceedings of IEEE brings in the frequency diversity attribute of CSI, which
INFOCOM 2012. each subcarrier has different amplitudes and phases. By
exploiting the frequency diversity, we can construct a
Digital Object Indentifier 10.1109/TPDS.2012.214 1045-9219/12/$31.00 © 2012 IEEE
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. XX, NO. X, XXXX 2012 2
0.4 2 P RELIMINARIES
0.35 In this section, we introduce the CSI value which is
0.3
the foundation of FILA design. And the background
information of the OFDM system can be found in sup-
Probability
0.25
plemental file.
0.2
0.15 2.1 Channel State Information
0.1 Based on OFDM, channel measurement at the subcarrier
0.05 level becomes available. Nowadays, adaptive transmis-
0
sion systems in wireless communication always improve
−34 −32 −30 −28 −26 −24 the throughput by utilizing some knowledge of the chan-
RSS(dBm)
nel state to adapt or allocate transmitter resources [6].
Fig. 1: Temporal variance of RSSI. Channel state information or channel status informa-
tion (CSI) is information that estimates the channel by
representing the channel properties of a communica-
unique “fingerprinting” indicating each location on the tion link. More specifically, CSI describes how a signal
radio map. According to these advantages, it is favorable propagates from the transmitter(s) to the receiver(s) and
to leverage the CSI to improve the performance of indoor reveals the combined effect of, for instance, scattering,
location fingerprinting. And thus, designing a precise fading, and power decay with distance. In summary,
tracking/localization system becomes possible. the accuracy of CSI greatly influences the overall OFDM
Based on CSI, in this paper, we present the design and system performance. It is worth pointing out that accord-
implementation of FILA, a novel cross-layer approach ing to the definition of CSI, only OFDM-based WLAN
based on OFDM for indoor localization using WLANs. systems can demonstrate the frequency diversity in CSI
since they use multiple subcarriers for data transmission.
In summary, the main contributions of this paper are In another word, other modulation schemes like DSSS
as follows. cannot provide this value.
1) We design FILA, a cross layer approach that en- In a narrowband flat-fading channel, the OFDM sys-
ables fine-grained indoor localization in WLANs. tem in the frequency domain is modeled as
FILA includes two parts, the first one is CSI-based y = Hx + n, (1)
propagation model and the second one is CSI-
based fingerprinting. To the best of our knowledge, where y and x are the received and transmitted vectors,
FILA is the first to use fine-gained PHY layer respectively, and H and n are the channel matrix and
information (CSI) in OFDM to build a propaga- the additive white Gaussian noise (AWGN) vector, re-
tion model so as to improve indoor localization spectively.
performance. And it is also the first time to take ad- Thus, CSI of all subcarriers can be estimated according
vance of the combination of the fine-grained PHY to (1) as
y
layer information CSI with frequency diversity and Ĥ = , (2)
x
multiple antennas with spatial diversity for indoor
location fingerprinting. which is a fine-grained value from the PHY layer that
2) We implement FILA in commercial 802.11 NICs describes the channel gain from TX baseband to RX
and conduct extensive experiments in several typ- baseband.
ical indoor environments to show the feasibility of
our design. 3 S YSTEM D ESIGN
3) Experimental results demonstrate that FILA sig- In this section, we first give an overview of the system
nificantly improves the localization accuracy as architecture. Challenges in the system design are pre-
compared to the corresponding traditional RSSI- sented in the supplemental file.
based approach.
The rest of this paper is organized as follows. In 3.1 System Architecture
Section. 2, we introduce some preliminaries. This is FILA system is built based on the current communi-
followed by the system architecture design in Section 3. cation system and thus compatible to the under layer
In Section 4, we demonstrate the CSI-based propagation design. More precisely, no modification is required at
model. In Section 5, we illustrate the methodology of the transmitter end (TX–the AP), while only one new
CSI-based fingerprinting. The implementation of FILA component for CSI processing is introduced for localiza-
and experimental evaluations are presented in Section 6. tion purposes at the receiver end (RX–the target mobile
Finally, conclusions are presented and suggestions are device). Fig. 2 demonstrates the detailed design of the
made for future research in Section 7. system architecture. For traditional packet transmission
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. XX, NO. X, XXXX 2012 3
TX Distance RX
4 CSI- BASED PROPAGATION MODEL
Stationary RF Signal Propagation Mobile In this section, we leverage the fine grained CSI value in-
stead of RSSI to build a propagation model and address
AP Location
TX
X
Information the indoor localization issue. The CSI-based propagation
model can be built based on three following steps.
CSIeff (2)’ Distance 1) CSI Processing: First, we need to mitigate estima-
(2) Process CSI
Calculator + (3) Locate RX
tion error by effectively processing the CSI value
CSI denoted as CSIef f . This is known as the prerequi-
(1) Collect CSI site of the ongoing two steps.
Channel 2) Calibration: Afterwards, we develop a refined in-
Estimation
door propagation model and a fast training algo-
rithm to derive the relationship between CSIef f
OFDM OFDM Normal
X
RX Demodulator Decoder Data
and distance.
3) Location Determination: By receiving the APs coor-
Localization Block dinates in network layer and CSIef f values from
physical layer, we apply the revised propagation
Fig. 2: System Architecture.
model and trilateration method to accomplish the
localization. This part is in the supplemental file.
in wireless communication, only the demodulated signal 4.1 CSI Processing
is exported to the decoder for message content retrieval. For wireless communication, attenuation of signal
However, a prerequisite in FILA localization system is strength through a mobile radio channel is caused by
that it should be able to export the CSI value after the three nearly independent factors: path loss, multipath
normal demodulation process. Such that we devise a fading, and shadowing. The path loss characterizes the
localization block to exploit the CSI information. property that the signal strength decays as the distance
In our designated localization block, the CSI collected between the transmitter and receiver increases, which is
from 30 groups different subcarriers will firstly be pro- the foundation of our CSI-based localization. Multipath
cessed. After running the proposed algorithm, we can fading is a rapid fluctuation of the complex envelope of
obtain the effective CSI in an efficient time constraint. received signal caused by reception of multiple copies
Then the effective CSI will be used to estimate the of a transmitted signal through multipath propagation.
location of the target object. As mentioned in the pre- Shadowing represents a slow variation in a received
vious section, CSI value is the channel matrix from RX signal strength due to the obstacles in propagation path.
baseband to TX baseband which is needed for channel Therefore, before establishing the relationship between
equalization. Therefore, there is no extra processing over- CSI and distance, we need to mitigate the estimation
head when obtaining the CSI information. Nevertheless, error introduced by multipath fading and shadowing.
RSSI is obtained at the receiver antenna in the 2.4 GHz
radio frequency before down convert to the IF and
baseband. Therefore, the free space model that built 4.1.1 Time-domain Multipath Mitigation
for RSSI-based localization approaches can’t be directly The first concentration of our design is that the sys-
applied to process the CSI value. We need to refine such tem must be capable of dealing with the challenge of
radio propagation model according to the CSI informa- operating over a multipath propagation channel. Since
tion and compute the distance based on the proposed multipath effect will introduce Inter-Symbol-Interference
one. Finally, as the AP location information is obtained (ISI), cyclic prefix (CP) is added to each symbol to
from the network layer while CSI is collected from the combat the time delay in OFDM systems. However,
physical layer, we then use the simplest trilateration the CP technique is helpless for the multiple reflections
method to obtain the location. For the fingerprinting, we within a symbol time.
leverage the CSI values including different amplitudes For narrow-band systems, these reflections will not be
and phases at multiple propagation paths, known as the resolvable by the receiver when the bandwidth is less
frequency diversity, to uniquely manifest a location. We than the coherence bandwidth of the channel. Fortu-
then present a probability algorithm with a correlation nately, the bandwidth of 802.11n waveforms is 20MHz
filter to map object to the fingerprints. (with channel bonding, the bandwidth could be 40MHz),
In our FILA system architecture design, CSI is only which provides the capability of the receiver to resolve
processed by a newly designed localization block if the different reflections in the channel. We propose a
needed. Owing to the fact, FILA can be applied con- multipath mitigation mechanism that can distinguish
currently with the original packet transmission. In other the LOS signal or the most closed NLOS from other
words, it will not introduce additional overhead during reflections in the expectation of reducing the distance
the data transmission. estimation error.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. XX, NO. X, XXXX 2012 4
The commonly used profile of multipath channel in compensation of the fading of received signals in fre-
the time domain is described as follow, quency domain to enhance location accuracy.
Lp −1
In general, when the space between two subcarriers
is larger than the coherence bandwidth, they are fading
h(τ ) = αk δ(τ − τk ), (3)
independently. Since the channel bandwidth of 802.11n
k=0
system is larger than the coherence bandwidth in typical
where Lp is the number of multipath channel compo- indoor environment, the fading across all subcarriers
nent. αk and τk are the amplitude and propagation delay are frequency-selective. To combat such frequency se-
of the k-th path. In practice, OFDM technologies are effi- lectively fading of wireless signals, multiple uncorre-
ciently implemented using a combination of fast Fourier lated fading subchannels (multiple frequency subcarri-
Transform (FFT) and inverse fast Fourier Transform ers), that is 30 groups of CSI values are combined at
(IFFT) blocks. The 30 groups of CSI represent the channel the receiver. Motivation for leveraging the frequency
response in frequency domain, which is about one group diversity stems from the fact that the probability of
per two subcarriers. With IFFT processing of the CSI, we simultaneous deep fading occurring on multiple uncor-
can obtain the channel response in the time domain, i.e., related fading envelopes (in our case, resulting from
h(t). From Fig. 3 we can observe that the LOS signal and frequency diversity) is much lower than the probability
multipath reflections come with different time delay, and of a deep fade occurring on a single frequency system.
generally the LOS signal has higher channel gain, so we Thus, exploiting the wide bandwidth of WLAN that
can use a trunk window with the first largest channel assures sufficiently uncorrelated subcarriers, will reduce
gain in the center to filter out those reflections. If LOS the variance in CSIs owing to small scale factors, which
doesn’t exist, we can identify the shortest path NLOS re- appears to be one of the major sources of location
flection. According to Nyquist sampling theorem, wider determination error. In our FILA system, we weighted
spectrum leads to higher resolution in the time domain. average the 30 groups CSIs in frequency domain so as
Due to the bandwidth limitation of WLAN, we can’t to obtain the effective CSI, which exploits the frequency
distinguish all the reflections but we can use this method diversity to compensate the small-scale fading effect.
to reduce the variance induced by multipath effects. The Given a packet with 30 groups of subcarriers, the
time duration of the first cluster is determined by setting effective CSI of this packet is calculated as
the truncation threshold as 50% of the first peak value.
K
In doing so, we expect to mitigate the estimation error 1 fk
CSIef f = × |Hk |, k ∈ (−15, 15), (4)
introduced by multipath reflection. K f0
k=1
After the time domain signal processing, we reobtain
the CSI using FFT. Fig. 3 shows the CSI results after time where f0 is the central frequency, fk is the frequency of
domain filtering. Note that, commercial NICs embeds the k-th subcarrier, and |Hk | is the amplitude of the k-th
hardware circuits for the FFT and IFFT processing, our subcarrier CSI.
algorithm introduces ignorable latency to the whole Note that selection of weighting factors are based
localization procedure. on the fact that the radio propagation is frequency-
related. According to the free space model, the received
signal strength is related to the frequency the signal is
60
Original value transmitted. So by this weighting method, we transfer
After filtering the channel gain from multiple subcarriers to a single
50
subcarrier, i.e., the central one. Next, we will establish
the relationship between the CSIef f and distance.
CSI Amplitude
40
4.2 Calibration
30
Since CSI value is obtained from the baseband on the
receiver side, the radio propagation model [7] for RSSI
20
is no longer suitable for our design. So we develop
a refined indoor propagation model to represent the
10
−30 −20 −10 0 10 20 30 relationship between the CSIef f and distance by revising
Subcarrier Index the free space path loss propagation model, given by:
2 n1
Fig. 3: Time Domain Channel Response. 1 c
d= ×σ , (5)
4π f0 × |CSIef f |
4.1.2 Frequency-domain Fading Compensation where c is the wave velocity, σ is the environment
Moreover, since CSI represents the channel responses factor and n is the path loss fading exponent. Both of
of multiple subcarriers, a combination scheme is also the two parameters are dependent on distinctive indoor
introduced to process the CSI value in our system for environments. The environment factor σ represents the
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. XX, NO. X, XXXX 2012 5
gain of the baseband to the RF band at the transmitter where |Hi | is the amplitude response and ∠H is the
side, inversely, the gain of RF band to baseband at the phrase response of the ith subcarrier.
receiver side, and the antenna gains as well. Moreover, Then, it comes to CSI processing which is the prerequi-
for NLOS AP, the σ also includes the power loss due to site of calculating position likelihood in the positioning
wall penetration or the shadowing. The path loss fading phase. We quantify the power of a package, denoted as
exponent n is varying depending on the environment. summational CSI, by adding up the power with respect
For instance, when the RF signal is propagating along a to 30 groups of subcarriers. Specifically,
free space like corridor, the path loss fading exponent
I
n will be around 2. In other cases, such as an office
He = |Hi |2 , i ∈ [1, 30], (8)
that represents a complex indoor scenario, the exponent
i=1
could be larger than 4. In an indoor radio channel with
clutter in medium, where often the LOS path is aug-
mented with the multipath NLOS at the receiver, signal 5.1 CSI-based Fingerprinting Generation
power decreases with a pass loss fading exponent higher As the foundation of fingerprinting approach, the mea-
than 2 and is typically in the order of 2 to 4 [8]. Hence, it sured CSI values are processed to construct a radio
is not trivial to determine the received signal power and map. Since most of the RF-based fingerprint methods
we need to refine the free space propagation model that consider two spatial dimensions for localization [9], we
obeys the analytical and empirical methods. A widely also follow the principle. Therefore, the two-dimension
used simplification is to assume that all the path loss physical space coordinate of a sample position lj is
exponents that model propagations between the specific lj = (lj,x , lj,y ). To generate a radio map, we first extract
receiver and all the APs are equal. This simplification the statistic determine the number of detectable APs
in a typical indoor environment is an oversimplification, for a sample position. At each reference point, we will
since the channel propagation is usually very different estimate the signal strength distribution for each access
depending on the relative position of the mobile client point at each location. In our location system, the signal
with regard to each AP. Therefore, we calibrate both strength over all the subcarriers is represented by He .
environment factor σ and the path loss fading exponent Moreover, another component of the radio map will
n in a per-AP manner. be normalized CSI values at each subcarrier for each
We propose a simple fast training algorithm based AP. The motivation of leveraging the frequency diversity
on supervised learning to retrieve the parameters with stems from the fact that CSI benefits from the mul-
three anchors in offline phase. In the first step, CSIs of tipath effects, because the received signal at different
multiple packets are collected at two of the anchors to positions will be the combination of different reflections.
train the environment factor σ and the path loss fading Therefore, these normalized CSI values will reflect the
exponent n for the refined indoor propagation model. In combination result ignoring the large scale fading.
the second step, CSIs collected at the third anchor are
used to test the efficiency of the parameter estimation.
The two steps run iteratively until convergence. The 5.2 Positioning Phase
experimental results in the next section show that this For object location estimation, the target is required to
simple algorithm can achieve satisfactory accuracy, more be accurately mapped to the radio map.
sophisticated training method will be able to obtain Previous works show that the probabilistic approaches
better performance. such as maximum likelihood provide more accurate
results than deterministic ones do in indoor environ-
ments [9]. Therefore, we adapt the probability model
5 CSI- BASED FINGERPRINTING in [10] except that we use He instead of RSS value.
In this section, we introduce the methodology of CSI- Similarly, we treat He observed from the AP to the
based location fingerprinting approach. We start by us- receiver at a fixed location as a Gaussian variable. In the
ing a mobile device equipped with 802.11 NICs to re- proposed system, we will select K best APs to calculate
ceive the beacon message from nearby APs at each sam- the probability of the MS at each reference point. The
ple position. The message contains CSI that represents criteria for the best AP selection is that those APs with
the channel response of multiple subcarriers. We modify highest He values, because they are more reliable. In our
the driver and divide the CSIs into 30 groups. Hence, experiment, we fix the K to be 3.
N = 30 groups CSI values are collected simultaneously The selected K He values obtained by the terminal to
at the receiver that represented as be located form a vector He = [He,1 , · · · , He,K ]. Then,
the position estimation problem is equivalent to finding
H = [H1 , H2 , · · · , Hi , · · · , HN ]T , i ∈ [1, 30], (6) the l that maximizes the posteriori probability P (lj |He ).
According to Bayes’ law,
where each subcarrier Hi is defined as
P (lj )P (He |lj ) P (lj )P (He |lj )
P (lj |He ) = = (9)
Hi = |Hi |ej sin{∠Hi } , (7) i P (l i )P (H |l
e i ) P (He )
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. XX, NO. X, XXXX 2012 6
Note that P (lj ) is the prior probability that the termi- 6 E XPERIMENTAL R ESULTS
nal located at the reference point li . In [9], [10], uniform
distribution is assumed. In contrast, we will leverage the In this section, we present the implementation and ex-
spatial correlation of the CSI to determine the P (lj ). perimental evaluation of FILA. First, we describe the
experimental setup which can be found in the supple-
Recall that CSI is a fine-grained information, we can
mental file. Then we illustrate the validation results for
observe channel response over multiple subcarriers rep-
our refined propagation model. Finally, we evaluate the
resented by Hk = [H1 , H2 , · · · , Hi , · · · , HN ]T for the kth
performance of CSI-based propagation model and its fin-
AP. We denote the observed CSI with normalization for
gerprinting. In our evaluation, we use the performance
each AP as H(O) ∈ CN ×K , and the CSI recorded in the
of corresponding RSSI-based approach based on radio
radio map for the same set of APs at position lj as H(lj ).
propagation model and trilateration as baseline .
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 II. Experimental Scenarios
K We conduct experiments to show the performance and
cov(Hk (O), Hk (lj ))
ρH(O),H(lj ) = , (10) robustness of our FILA system in four different scenarios
σHk (O) σHk (lj ) in the campus of Hong Kong University of Science and
k=1
Technology as follows:
where each AP is considered to be independent. Ac- 1) Chamber First, we set up a testbed in a 3m × 4m
cording to the measurement, the spatial channel cor- Chamber to collect the RSSI and CSI as shown in
relation will decrease as the distance between the two Fig. 4. In general, Chamber is an enclosure that
receiver increases. Therefore, with higher ρ, the position used as environmental conditions for conducting
of the terminal will be closer to the reference point. Then, testing of specimen. In our experiment, chamber
the probability of the terminal on each candidate point represents the ideal free space indoor environment
is defined as which means only LOS signal exists without other
ρH(O),H(lj ) multipath reflection or external interference.
P (lj ) = J (11) 2) Research Laboratory Then, we deployed FILA in
i=1 ρH(O),H(li )
an identical indoor scenario – a 5m × 8m research
where J is the size of the candidate reference points set. laboratory as shown in Fig. 5. In the laboratory
Considering uncorrelated property between each AP, region, we place three APs on the top of three shel-
the likelihood P (He |lj ) can be calculated as, ters in three dimensions. The experiment was con-
ducted on a weekday afternoon when there were
K
a couple of students sitting or walking around,
P (He |lj ) = P (He,k |lj ), (12) which will show the robustness of our system to
k=1 temporal dynamics of the environment. The laptop
was placed at a fixed position at the beginning of
Since the signal strength at each reference point is the experiment and then moved to the next point
modeled as a Gaussian variable which requires less sam- along a pre-measured path.
plings than the histogram approach [11]. At the offline 3) Lecture Theatre In addition, we chose a larger
phase, we can obtain the expectation H̄e,k and variance lecture hall to conduct the localization experiments,
σe,k corresponding to the He,k , and the P (He,k |lj ) is which is a 20m × 20m lecture theatre. Since the
obtained as space is relatively large, the influence of the room
size can be explored.
1 −(He,k − H̄e,k )2
P (He,k |lj ) = exp . (13) 4) Corridor Finally, we performed experiments in a
2πσe,k 2σe,k corridor environment with multiple offices aside
in our academic building, which is 32.5m × 10m
The location estimation of the terminal is the weighted covering corridors, rooms and cubicles. In this
average over the whole candidate set, scenario, we expect to illustrate the impact of the
J
absence of LOS APs on the location accuracy.
l̂ = P̄ (lj |He )lj (14)
j
6.1 Validate the Refined Model
For the fingerprinting method, the terminal can pro- As the target for precise indoor localization, two most
cess CSI by itself and then check the globe or local important metrics are used to testify FILA: the accuracy
database for localization. It doesn’t need to rely on one and the temporal stability of location estimation. After-
pubic server. wards, we compared the performance of our CSI-based
The performance of the proposed fingerprinting localization system with the corresponding RSSI-based
methodology is evaluated in the following section. approach.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. XX, NO. X, XXXX 2012 7
50
CSIeff amplitude
45 Exponential Fitting
40
CSIeff amplitude
35
30
25
20
15
10
2.5 3 3.5 4 4.5 5 5.5 6
Distance (meters)
Fig. 4: Chamber Fig. 5: Research Laboratory Fig. 6: Relation between CSIef f and
Distance.
20 −25 3
19 Lecture Theatre −26 Lecture Theatre RSSI CSI
Chamber Chamber
18 −27 2.5 RSSI CSI
Lab Lab
CSIeff amplitude
17 −28 2
RSSI(dBm)
Error (m)
16 −29
15 −30 1.5
14 −31
13 −32 1
12 −33 0.5
11 −34
10 −35 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 2 4 6 8 10 12 14 16
Time Duration(s) Time Duration(s) Time Duration(min)
Fig. 7: CSIef f on Temporal Variance Fig. 8: RSSI on Temporal Variance Fig. 9: CSI vs. RSSI on distance esti-
mation in different environments.
6.1.1 Robustness of the Refined Model the variance of CSIef f is within 1 dBm. Therefore,
One essential aspect that needs to be determined before the proposed new matric is much more stable in time
the localization experiments is whether CSI value can domain compared with RSSI.
build a relationship with distance. In general, indoor
RF signal strength is a non-monotonic function with
distance due to multipath and shadowing effects. Fig. 6 In Fig. 9, we further investigate the CSIef f value
illustrates the CSI value approximated by a power func- and RSSI value in the chamber and research laboratory
tion of distance according to our refined propagation so as to discover the effect of any temporal instability
model. In diverse scenarios with corresponding environ- on distance estimation. Chamber provides a free space-
ment factor σ, the path loss fading exponent n varies in like environment as it uses specific material that can
a range of [2, 4]. It is shown that our refined model prop- absorb the non-LOS signals. Thus, multipath effect can
erly fits the relationship between CSIef f and distance. be eliminated in chamber environment. However, the
result from our experiment shows that even in chamber
the RSSI is also varied significantly from time to time
6.1.2 Temporal Stability of CSI
due to the inaccurate measurement. In contrast, research
Temporal stability is a fundamental criteria in validating lab is a typical multipath environment. Both the static
the robustness of the localization systems. We thus set obstacles and dynamic walking around individuals exert
out to examine the stability of the proposed new metric the influence on multipath and bring in more intense
CSIef f and RSSI value in time series. It is well-known path loss. In this way, the variance of RSSI becomes even
that RSSI is a fickle measurement of the channel gain larger and the performance of distance estimation is even
because of its coarse packet-level estimation and easily worse.
varied by multipath effect. As CSI is fine-grained PHY
layer information that provides detailed channel state
information in subcarrier level, it is of great importance Fig. 7, Fig. 8 and Fig. 9 lead to an essential conclusion:
to figure out whether it will remain in a stable manner in comparison with RSSI, CSI is more temporally stable
in practical environment. in different environments and helps maintain the per-
Fig. 7 and Fig. 8 illustrate both the interactions of formance over time. Therefore, FILA can achieve accu-
CSIef f and RSSI values on temporal variance, respec- rate location more quickly than the RSSI-based scheme,
tively. It is shown that the received signal power cal- which is very crucial for some location-based application
culated by RSSI has a variance up to 5 dBm, while like search and rescue.This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. XX, NO. X, XXXX 2012 8
4 falls within the range of 1 meter, and the 50% accuracy
Chamber is less than 0.5 m. In such a dynamic environment with
3.5
Mean distance error (m)
Laboratory
3 Lecture Theatre lots of factors interfering the propagation of signals,
FILA exhibits a preferable property indicating that the
2.5
fine-grained nature of CSI is beneficial to improve the
2
accuracy of corresponding RF-based approach.
1.5 Fig. 11(b) depicts the cumulative distribution function
1 errors per location detected at the university lecture
0.5 theatre. Even for such a much larger space, FILA can
0 locate objects in the range within 1.8 meters of their
RSSI CSI
RSSI vs CSI actual position with 90 percent probability, which is
acceptable for most location-based applications.
Fig. 10: Mean distance error. Across the above two typical single room indoor sce-
narios, FILA achieves median accuracy of 0.45 m and 1.2
6.2 Performance Evaluation of CSI-based Propaga- m, respectively. It is therefore safe to conclude that the
tion Model proposed CSI-based scheme performs much better than
As a target for precise and fast indoor localization, two RSSI-based one when locating objects in a typical indoor
most important metrics are used to testify this model: the building with multipath effect.
accuracy and the latency of localization, afterwards, we
Localization Accuracy in Multiple Rooms
compared the performance of our CSI-based localization
systemwith the corresponding RSSI-based approach by In our previous experiments, the APs and client are
using the propagation model. placed in the same room. We also examined the corridor
scenario where several APs are deployed in the multiple
6.2.1 Accuracy rooms. Specially, we take into consideration both the
complicated multipath effect and the shadowing fading
Accuracy over a Single Link
brought by wall shield. We first fix the position of the
As the premise of indoor localization, we first inves- object at some reference nodes, and collects the AP
tigated the distance determination accuracy of FILA coordinate and CSI value for offline training. Then, we
compared with the corresponding RSSI-based approach. move the object to arbitrary positions for online tracking.
The primary source of error in indoor localization is The moving speed is around 1m/s and we collect 20 CSIs
multipath propagation caused by multiple reflections and RSSIs at each position.
that overlap with the direct LOS subcarrier at the re- In Fig. 11(c), we plot the cumulative distribution of
ceiver. FILA takes advantage of the fine-grained trait location errors across 10 positions. It is shown that
to mitigate such multipath effect, and exploits the fre- multipath propagation does degrade the accuracy of
quency diversity to compensate the frequency-selective object localization as well as the shadowing in the mul-
shading. We repeated the distance measurement experi- tiple rooms scenarios. However, FILA is robust enough
ments across 10 different locations in chamber, research to maintain the degradation. More importantly, FILA
laboratory and lecture theatre, respectively. For some can achieve median accuracy of 1.2 m in this corridor
positions with serious multipath effect, FILA achieves up environment. This result indicates that FILA is able to
to 10 times accuracy gain over the corresponding RSSI- effectively estimate and compensate for gain differences
based scheme. Fig. 10 illustrates the mean distance errors across multiple rooms.
in three different environments. Our evaluation shows
that FILA can outperform the corresponding RSSI-based 6.2.2 Latency of CSI-based propagation model localiza-
scheme by around 3 times for the distance determination tion
of a single link. Two main phases contribute to the latency of FILA,
To assess the effectiveness of the CSI-based localiza- the calibration phase and location determination phase.
tion approach, in the following we evaluate the accuracy Since the environment factor and the fading exponent
of FILA in different typical indoor environments. vary in different environments, we need to conduct
calibration to train these two parameters for the refined
Localization Accuracy in Single Room propagation model. We should collect CSIs at some pre-
In the experiments conducted in the research laboratory, known positions to calculate these two parameters using
we fix three APs on the top of the shelters. The mobile our fast training algorithm. Actually, this process can
laptop with iwl5300 NICs is first fixed at one location be finished before localization as an offline task since
and then moved to another. We repeated this process and the APs can use each other’s information for the cali-
placed the device at 10 different positions respectively. bration. In our FILA system, the AP takes about 0.8ms
Fig. 11(a) illustrates the cumulative distribution (CDF) of to transmit a packet with 100 bytes beacon message in
localization errors across the 10 positions. In our experi- IEEE 802.11n. Each time we collect 20 CSIs and the time
ments, for over 90% of data points, the localization error will be 0.8 × 20 = 16ms. The calibration process can beThis article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. XX, NO. X, XXXX 2012 9
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7
Probability
Probability
Probability
0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 FILA 0.2 0.2
FILA FILA
0.1 RSSI−based 0.1 RSSI−based 0.1 RSSI−based
0 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 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4
Distance error (m) Distance error (m) Distance error (m)
(a) Research Laboratory (b) Lecture Theatre (c) Corridor
Fig. 11: CDF of localization error in different indoor environments.
done within 2ms according to our measurement on a is shown in Fig. 12. Our evaluation shows that CSI-
HP laptop with 2.4GHz dual-core CPU. In the location based method can achieve the median accuracy of 0.65m,
determination phase, the IFFT and FFT process can lever- which outperforms Horus by about 0.2m, and the gain
age the according hardware blocks in the wireless NICs is about 24%. Moreover, in the corridor scenario, where
whose running time is ignorable. While we conduct covered by 6 APs and 3 APs were taken into computa-
these signal processing on laptop consuming around tion, the mean accuracy of our approach is 1.07m which
2ms, including the time needed for the trilateration is 0.35m lower than Horus system, about 25% gain over
location calculation. Therefore, the time consumption Horus.
for both training and location determination is within In addition, we compare the two approaches con-
several ms. In our experiment, the walking speed of user cerning different numbers of APs in corridor. Fig. 14
is around 1m/s. The average tracking error is around depicts the average accuracy according to the amount
1.2m as shown in Figure 11(c). For multiple rooms envi- of APs varying from 1 to 6. Since richer information to
ronment, the simple alpha-tracking algorithm [12] can be estimate the location can be obtained from the more APs,
applied for triggering the training. We only need to train both lines demonstrate the accuracy improvement. In
these parameters once unless the environment changes particular, our approach reduced the mean distance error
greatly. In summary, our system can reach the average by 29% on the average. Obviously, these results show the
time tracking latency to as fast as about 0.01s, which effectiveness of the proposed CSI-based location system
significantly outperforms previous RSSI-based tracking and indicate the benefits from indoor environment with
systems [4] (usually 2 − 3s). dense-deployed APs. When the AP is sparse, our scheme
performs much better than the RSS-based one.
1.5
Horus 6.3.2 Precision
Mean distance error (m)
1.25 CSI−based
Fig. 15 illustrates the cumulative distribution (CDF) of
1 localization errors in the laboratory. The data were col-
0.75 lected across the 28 positions in the laboratory. From
Fig. 15, we can observe that the confidence interval with
0.5 confidence level 90% for the error is (-1.3m, +1.3m),
0.25 which means the CSI-based localization error falls within
the range of 1.3 meters, and the 50% accuracy is less than
0 0.6 m. However, the Horus can locate objects in the range
Laboratory Corridor
CSI−based vs Horus within 1.6 meters of their actual position with 90 percent
probability, and the median accuracy is 0.8.
Fig. 12: Mean distance error.
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
6.3 Performance Evaluation of CSI-based Finger-
rooms. Fig. 13 depicts the cumulative distribution of
printing
positioning errors across 20 positions. We can easily ob-
6.3.1 Accuracy serve that both our approach and Horus can achieve the
First, we evaluate the accuracy of the proposed CSI- median accuracy less than 1.25m. However, the accuracy
based probability algorithm and compared with Horus, improvement of our approach over Horus for 90% of
the widely used RSSI-based fingerprinting system. The data points is 0.55m. We can conclude that our approach
mean distance error in the two different environments exhibits a preferable property since the fine-grained andThis article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. XX, NO. X, XXXX 2012 10
1 4.5 1
CSI−based CSI−based CSI−based
0.9 4 0.9
Horus Horus Horus
Mean distance error (m)
0.8 3.5 0.8
0.7 3 0.7
Probability
Probability
0.6 0.6
2.5
0.5 0.5
2
0.4 0.4
0.3 1.5 0.3
0.2 1 0.2
0.1 0.5 0.1
0 0 0
0 0.5 1 1.5 2 2.5 3 1 2 3 4 5 6 0 0.25 0.5 0.75 1 1.25 1.5 1.75 2
Distance error (m) Number of APs Distance error (m)
Fig. 13: CDF of localization error in Fig. 14: Mean distance error with dif- Fig. 15: CDF of localization error in
Corridor. ferent numbers of APs. Laboratory.
frequency diversity nature of CSI is beneficial to improve [3] D. Moore, J. Leonard, D. Rus, and S. Teller, “Robust distributed
the precision of location fingerprinting. network localization with noisy range measurements,” in Proc. of
ACM Sensys, 2004.
[4] D. Zhang, J. Ma, Q. B. Chen, and L. M. Ni, “An rf-based system for
7 C ONCLUSIONS AND F UTURE WORK tracking transceiver-free objects,” in Proc. of IEEE PerCom, 2007.
[5] D. Zhang and L. M. Ni, “Dynamic clustering for tracking multiple
Localization is one of the most appealing applications transceiver-free objects,” in Proc. of IEEE PerCom, 2009.
and becomes increasingly common in our daily life. [6] D. Halperin, W. J. Hu, A. Sheth, and D. Wetherall, “Predictable
802.11 packet delivery from wireless channel measurements,” in
RSSI-based schemes have been widely used to provide Proc. of ACM SIGCOMM, 2010.
location-aware services in WLAN. However, in this pa- [7] A. Goldsmith, Wireless Communications and Networks:3G and be-
per, we observe that RSSI is roughly measured and easily yond. Tata McGraw Hill Education Private Limited, 2010.
[8] K. Fazal and S. Kaiser, Multi-carrier and spread spectrum systems :
affected by the multipath effect which is unreliable. We from OFDM and MC-CDMA to LTE and WiMAX. Wiley, 2008.
then use the fine-grained information, that is, Channel [9] M. Youssef and A. Agrawala, “The horus wlan location determi-
State Information (CSI), which explores the frequency nation system,” in Proc. of ACM MobiSys, pp. 205–218, 2005.
[10] S. Fang, T. Lin and K. Lee, “A novel algorithm for multipath
diversity characteristic in OFDM systems to build the fingerprinting in indoor wlan environments,” IEEE Trans. Wireless
indoor localization system FILA. In FILA, we process the Commun., 2008.
CSI of multiple subcarriers in a single packet as effective [11] A. Haeberlen, E. Flannery, A. M. Ladd, A. Rudys, D. S. Wallach,
and L. E. Kavraki, “Practical robust localization over large-scale
CSI value CSIef f , and develop a refined indoor radio 802.11 wireless networks,” in Proc. of ACM MobiCom, pp. 70–84,
propagation model to represent the relationship between 2004.
CSIef f and distance. Based on the CSIef f , we then [12] T. He, S. Krishnamurthy, T. Luo, L. andYan, K. B., L. Gu,
R. Stoleru, G. Zhou, Q. Cao, P. Vicaire, J. A. Stankovic, T. F.
design a new fingerprinting method which leverages Abdelzaher, and J. Hui, “VigilNet: An Integrated Sensor Network
the frequency diversity. To demonstrate the effectiveness System for Energy-Efficient Surveillance.,” in ACM Transactions on
of FILA, we implemented it on the commercial 802.11n Sensor Networks, 2006.
NICs. We then conducted extensive experiments in typ-
ical indoor environments and the experimental results
show that the accuracy and speed of distance calculation
can be significantly enhanced by using CSI. Kaishun Wu is currently a research assistant
In this work, we just use the simplest trilateration professor in Fok Ying Tung Graduate School with
the Hong Kong University of Science and Tech-
method to illustrate the effectiveness of CSI in indoor nology. He received the Ph.D. degree in com-
localization. The future research in the new and largely puter science and engineering from Hong Kong
open areas of wireless technologies can be carried out Uinversity of Science and Technology in 2011.
along the following directions. First, we can leverage the His research interests include wireless commu-
nication, mobile computing, wireless sensor net-
available multiple APs to improve the location accuracy works and data center networks.
in some extent. Second, in this paper we only leverage
the frequency diversity, however, the spatial diversity
can also be exploited in MIMO to enhance the indoor
localization performance. Third, since some of the smart
phones have 802.11n chipset, the next step of our work Jiang Xiao is currently a first year Ph.D student
is to implement FILA in smart phone. in Hong Kong University of Science and Tech-
nology. Her research interests mainly focused
on wireless indoor localization systems, wireless
R EFERENCES sensor networks, and data center networks.
[1] P. Bahl and V. N. Padmanabhan, “Radar: an in-building rf-based
user location and tracking system,” in Proc. of IEEE INFOCOM,
2000.
[2] J. Liu, Y. Zhang, and F. Zhao, “Robust distributed node localiza-
tion with error management,” in Proc. of ACM MobiHoc, 2006.This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. XX, NO. X, XXXX 2012 11
Youwen Yi received his B.Sc (Hons) degree
from Harbin Institute of Technology in 2007, and
receive his M.Sc degree from Huazhong Univer-
sity of Science and Technology in 2009. From
2009 and 2011, he was a research assistant at
Hong Kong University of Science and Technol-
ogy . Currently, he is a senior research engineer
at Huawei Technologies. His research interests
include real-time indoor localization system, and
next generation self-organized network.
Dihu Chen is a Professor in the School
of Physics and Engineering, Head of Micro-
electronic Department, Sun Yat-sen University,
China. His research interests include microelec-
tronics, IC design.
Xiaonan Luo is a Professor in the School of
Information Science and Technology, Director of
the Computer Application Institute, Sun Yat-sen
University, China. His research interests include
mobile computing, computer graphics and CAD.
Lionel M. Ni is Chair Professor in the Depart-
ment of Computer Science and Engineering at
the Hong Kong University of Science and Tech-
nology (HKUST). He also serves as the Special
Assistant to the President of HKUST, Dean of
HKUST Fok Ying Tung Graduate School and
Visiting Chair Professor of Shanghai Key Lab of
Scalable Computing and Systems at Shanghai
Jiao Tong University. A fellow of IEEE, Dr. Ni has
chaired over 30 professional conferences.You can also read
