IEEE SENSORS JOURNAL, VOL. 19, NO.

13, JULY 1, 2019 5141

3D-WiFi: 3D Localization With Commodity WiFi

Lingyan Zhang and Hongyu Wang , Member, IEEE

Abstract— Recently, 3D localization has attracted intense inter- accuracy [5]–[7]. Due to the requirement of extremely high
ests with its enabled applications, such as smart healthcare, sampling rate and strict time-synchronization between UWB
smart home, and smart robots. Conventional 3D localization transceivers, the sophisticated location system is designed and
technologies require to utilize the special-dedicated infrastruc-
tures with a large-scale deployment, which results in high implemented by using the dedicated infrastructures. More-
cost. However, in order to accomplish an appealing tradeoff over, current location systems also adopt the mechanism of
between location accuracy and infrastructure cost, designing and TOA measurements to locate the target in 3D space using
implementing a 3D localization system are still a challenging frequency modulated continuous wave (FMCW) technology
task to solve. To address this issue, we develop a novel 3D [8], [9], which can achieve decimeter-level accuracy. Since
localization system by using commercial off-the-shelf (COTS)
WiFi infrastructures at a low cost. First, we design a cost- FMCW has to take a narrow-band signal to sweep the
effective hardware scheme to derive the orientation information carrier frequency through a wide bandwidth, FMCW signal
for 3D localization. Then, we adopt a fusion method of multiple generator is designed and developed by the special-purpose
channel state information measurements to improve multipath hardware platform. In addition, radio frequency identification
resolution and identify the target direct path by the azimuth devices (RFIDs) can be attached to the target object for
and elevation angle-of-arrival (AOA) and equivalent time-of-
arrival estimations, which can be formulated as subspace-based 3D localization with a larger-scale deployment [10]–[12].
sparse reconstruction with high accuracy. Finally, according to In indoor scenarios, RFID signal is harshly impaired by com-
both azimuth and elevation AOA estimations of the target direct plicated multipath propagation, so that the location systems
path, the target can be localized in 3D space at decimeter-level. provide coarse-grained 3D location information. In order to
We implement the proposed 3D localization system using COTS improve the location accuracy, the emerging location scheme
WiFi infrastructures and validate its performance with extensive
evaluations in two indoor scenarios. exploits a hybrid technique with both RFID-based subsys-
tem and computer-vision-based subsystem [13], which adopts
Index Terms— 3D, wireless localization, 2D-AOA, sparse recon- fingerprinting approach to localize the stationary object at
struction, COTS.
decimeter level with not only a cumbersome training proce-
dure, but also an additional auxiliary device. Taken together,
I. I NTRODUCTION existing 3D localization systems require the special-dedicated
infrastructures, which results in high cost to make them

W IRELESS localization technology has become a key

enabler for location-based services in a wide range
of applications. With multi-dimensional location information
unattractive in commercial applications. Therefore, designing
and implementing 3D localization system in terms of a tradeoff
between infrastructure cost and location accuracy are still a
or location-aware context, 3D localization has an augmented challenging task to solve.
ability to its enabled applications, such as service-oriented Fortunately, ubiquitous WiFi is the most promising
localization and navigation, unmanned aerial vehicle localiza- technology to realize GPS-like localization using the off-
tion, smart home, smart healthcare and smart robots [1]–[4]. the-shelf infrastructures in indoor scenarios [14], [15], which
Recently, 3D localization has been considered as an impor- makes it feasible to implement 3D localization at low cost.
tant aspect on wireless indoor localization by standardization With the proliferation of WiFi on the multi-antenna system,
committees, e.g., Federal Communications Commission (FCC) it can provide a new mechanism to accurately localize
[1], [5], and thus, it has aroused considerable interests. the target by deriving angle-of-arrival (AOA) information
3D localization can be defined as 3D locations determi- [16]–[23]. The accuracy of AOA measurement grows with the
nation relative to nearby reference nodes of known loca- number of antennas. Due to the limited number of antennas
tion in an interested space. Conventionally, 3D localization on the commodity WiFi card, current solutions [16], [20], [21]
based on ultra-wide bandwidth (UWB) technology utilizes fully utilize diverse phase information of WiFi channel state
time of arrival (TOA) or time difference of arrival (TDOA) information (CSI) measurements across both multiple antennas
measurements with centimeter-level or even millimeter-level and multiple subcarriers to emulate and enlarge the number of
Manuscript received December 17, 2018; revised January 26, 2019; accepted antennas, which can offer joint AOA and TOA estimation with
February 1, 2019. Date of publication February 21, 2019; date of current two-dimensional MUSIC algorithm. While classical MUSIC
version June 4, 2019. This work was supported by the National Natural algorithm is insensitive to the coherence of multipath signal in
Science Foundation of China under Grant 61671103. The associate editor
coordinating the review of this paper and approving it for publication was indoor scenarios, it results in AOA accuracy degradation, even
Dr. Wan-Young Chung. (Corresponding author: Hongyu Wang.) failures with insufficient antennas. To effectively enhance AOA
The authors are with the Faculty of Electronic Information and Electrical accuracy, the state-of-the-art location system [19] leverages the
Engineering, Dalian University of Technology, Dalian 116023, China (e-mail:
[email protected]; [email protected]). spatial sparsity of multipath signals with sparse reconstruction,
Digital Object Identifier 10.1109/JSEN.2019.2900511 which enables the sparsity in spatial dimension by partitioning
1558-1748 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app
 5142 IEEE SENSORS JOURNAL, VOL. 19, NO. 13, JULY 1, 2019

the sampling spatial grids. The prominent location systems implementation of the proposed 3D-WiFi localization system.
achieve decimeter-level accuracy, preferably using the Section IV validates the performance of the proposed 3D-WiFi
commercial off-the-shelf (COTS) WiFi infrastructures without localization system with experimental evaluations. Finally,
any other additional auxiliary. However, the emerging location the conclusion is drawn in section V.
scheme only focus on the plane-domain (2D) localization.
Therefore, the favorable performance of WiFi CSI-based
II. R ELATED W ORK
localization motives us to reach and realize 3D localization
with AOA measurements using COTS WiFi infrastructures at In this section, we firstly review 3D localization technolo-
low cost. gies with enabled applications, and then introduce WiFi-based
In this paper, we develop a novel 3D localization system, indoor localization with AOA measurements.
named 3D-WiFi that localizes the target in 3D space by Recent 3D localization systems are designed and imple-
using COTS WiFi infrastructures. We firstly design a cost- mented with diverse wireless technologies. Conventional 3D
effective hardware scheme to derive the orientation informa- location systems adopt TOA or TDOA measurements to
tion, i.e., the azimuth and elevation AOA estimations, for achieve an excellent accuracy using UWB technology [5]–[7]
3D localization. Then, we adopt a fusion method of multiple or FMCW technology [8], [9]. High accurate location sys-
CSI measurements to improve multipath resolution with the tems can be applicable for human body localization such
azimuth and elevation AOA and equivalent TOA measure- as elder or patient monitoring, motion tracking, and even
ments, which can be formulated as a problem of subspace- smart surgical tools [6], [8], [17], [18]. In these 3D local-
based sparse reconstruction. After that, the target direct path ization systems, RF signal acquisition requires to design and
can be identified by the minimal TOA measurement. Finally, develop the special-purposed infrastructures components. In
according to the azimuth and elevation AOA estimations of addition, RFIDs can be attached to the target object for 3D
the target direct path from two or more receivers, the target localization with a larger-scale deployment in the interested
can be localized in 3D space at a decimeter level by using environment [10]–[12]. In order to improve the location accu-
COTS WiFi infrastructures. racy, RFID-based scheme adopts fingerprinting approach to
The main contributions of the paper can be summarized as localize the object in 3D space with an additional vision-
follows. assisted technique [13]. Therefore, existing 3D localization
1) We develop a novel 3D localization system, named systems are designed and implemented by using the dedicated
3D-WiFi that first realizes to localize the target in 3D infrastructures, which results in high cost for potential and
space by using COTS WiFi infrastructures at low cost. practical location-based applications.
The proposed 3D-WiFi localization system can achieve Recently, ubiquitous WiFi infrastructures with multiple
decimeter-level accuracy and further accomplish the antennas bring about an emerging solution on wireless local-
tradeoff between location accuracy and infrastructure ization by deriving AOA information. In complicated mutil-
cost. path indoor environments, since the AOA resolution relies on
2) We design a cost-effective hardware scheme to derive the number of antennas, the location systems [17], [22], [23]
the azimuth and elevation AOA information for 3D have achieved high accuracy with a lot of antennas using
localization. The developed L-shaped array consists of the special-purpose hardware platform. However, advanced
two commodity WiFi cards with external uniform linear location schemes can achieve decimeter-level accuracy by
subarrays in the orthogonal directions by stitching one using the COTS WiFi infrastructures, even without any other
shared antenna. The advantage of the L-shaped array lies auxiliary [16], [20], [21]. With the limited number of antennas
in that the azimuth and elevation AOA estimations can on the COTS WiFi devices, these systems effectively leverage
be separately obtained using each uniform linear sub- rich phase information of CSI readings to emulate and enlarge
array with higher accuracy than other structured arrays, the number of antennas, which leads to joint AOA and
effectively reducing the computational complexity for TOA measurement with two-dimensional MUSIC algorithm.
practical 3D localization. To enhance AOA accuracy, the art [19] explores and exploits
3) In order to improve multipath resolution, we adopt the spatial sparsity of multipath signal to realize joint AOA and
a fusion method of multiple CSI measurements with TOA estimations with two-dimensional sparse reconstruction.
singular value decomposition (SVD). With the robust- Using the COTS WiFi infrastructures, the favorable location
ness to noises, the signal subspace is utilized to systems can achieve an attractive tradeoff between location
realize joint AOA and equivalent TOA estimations accuracy and infrastructure cost. In this regard, it motives us to
using each uniform linear subarray, which can be evolve WiFi CSI-based location systems into 3D localization
formulated as a problem of subspace-based sparse with AOA measurements by using COTS WiFi infrastructures
reconstruction. at low cost.
4) We implement the proposed 3D-WiFi localization sys- Therefore, we design and implement a novel 3D localization
tem on commercial WiFi devices, carry out experiments system that realizes the azimuth and elevation AOA measure-
in two indoor scenarios, and validate its performance ments by using COTS WiFi infrastructures. The developed
with extensive evaluations. L-shaped array can be utilized to independently obtain the
The rest of this paper is structured as follows. Section II azimuth and elevation AOA estimations using each uniform
reviews the related work. Section III describes the system linear subarray with higher accuracy than other structured

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app
 ZHANG AND WANG: 3D-WiFi: 3D LOCALIZATION WITH COMMODITY WiFi 5143

arrays. In order to improve multipath resolution for 3D local-

ization, we adopt a fusion method of multiple CSI measure-
ments that the signal subspace is utilized to realize the accurate
azimuth and elevation AOA estimations with subspace-based
sparse reconstruction. The proposed 3D localization system
can achieve decimeter-level accuracy with the tradeoff between
location accuracy and infrastructure cost.

III. S YSTEM I MPLEMENTATION

In this section, we introduce the detailed implementation
of 3D-WiFi system that can localize the target in 3D space
by using the COTS WiFi infrastructures. Firstly, we design
a hardware scheme to derive the azimuth and elevation AOA
measurements for 3D localization. Two commodity WiFi cards Fig. 1. L-shaped array with two commodity WiFi NICs by one shared
with external uniform linear subarrays in the orthogonal direc- antenna.
tions are consisted of the developed L-shaped array, which can
separate the azimuth and elevation AOA estimations using the same incident signal arrived at the developed L-shaped
each uniform linear subarray with high accuracy. Then, we array. Thus, it is necessary to sanitize phase noises and then
adopt a fusion method of multiple CSI measurements to align CSI measurements between two commodity WiFi NICs,
improve multipath resolution, which formulates the azimuth before performing the azimuth and elevation AOA estimations
and elevation AOA estimations as a subspace-based sparse for 3D localization.
reconstruction problem. With each uniform linear subcarray on 1) Phase Sanitization: Using the developed L-shaped array
the commodity WiFi device, the signal subspace is utilized to with commodity WiFi NICs, raw phase measurements include
realize joint AOA and equivalent TOA estimation of each path random phase offsets incurred by the sampling time off-
based on 1 -methodology SVD. After that, the target direct sets (STOs), which result from an extra delay except for the
path can be identified by minimal TOA measurement. Finally, signal propagating time in the interested environment. The
according to both azimuth and elevation AOA estimations of STOs are different from two commodity WiFi NICs, but the
the target direct path from two or more receivers, the target STO is identical to the particular subcarriers on the individual
can be localized in 3D space at decimeter-level by using COTS WiFi NIC. Therefore, the random phase offsets by the STO at
WiFi infrastructures. multiple subcarriers are linear in frequency. We can adopt a
linear fitting of raw unwrapped phase measurement to sanitize
A. Hardware Design such random phase offsets with each commodity WiFi NIC,
respectively [16].
In order to realize 3D localization system based on AOA
2) CSI Alignment: After sanitizing random phase offsets
measurements, it requires the orientation information, as both
with each commodity WiFi NIC, we require to align CSI
azimuth and elevation AOA estimations. Researches [24]–[26]
readings between two commodity WiFi NICs for the azimuth
point out that the L-shaped geometry array has the obvious
and elevation AOA estimations of the same incident signal.
advantage in terms of both azimuth and elevation AOA estima-
The sanitized phase measurements on the shared antenna
tions with higher accuracy than other structured arrays. More
should be reconciled using the developed L-shaped array. Due
importantly, the azimuth and elevation AOAs can be sepa-
to unsynchronized clocks between two commercial WiFi NICs,
rately estimated using each uniform linear subarray, effectively
there is a time shifting τ0 in the time-domain sampling, when
reducing the computational complexity. To this end, we design
the same incident signal arrives at the developed L-shaped
a cost-effective hardware scheme to form the L-shaped array
array. We can leverage the time-shift property of Discrete
using two commercial WiFi network interface cards (NICs).
Fourier Transform (DFT) to investigate the effect of τ0 in the
As shown in Fig. 1, two commercial WiFi NICs with external
frequency domain. The CSI reading H (k) at kth subcarrier on
uniform linear subarrays in the orthogonal directions construct
the shared antenna from one WiFi NIC can be expressed by
the L-shaped array by stitching one shared antenna using the
DFT as
power splitter.
As we all know, AOA measurements can be obtained by 
N−1
the phase differences across the antennas on an array. Due to H (k) = h(n)e− j 2πkn/N , (1)
hardware imperfection on commodity WiFi NICs, there are n=0
two critical challenges to realize both azimuth and elevation
where h(n) is the discrete sampling of channel impulse
AOA estimations using the developed L-shaped array. Firstly,
response and N is the DFT length. In this way, DFT by a
raw phase measurements on the commercial WiFi NICs are
time shifting τ0 is represented as
extremely noisy, which seriously deteriorate an accuracy of
AOA estimation. Secondly, CSI readings from two indepen- 
N−1
− j 2πkτ0 /N
dent WiFi NICs have unsynchronized clocks that impose H (k)e = h[(n − τ0 ) N ]e− j 2πkn/N , (2)
restrictions on the azimuth and elevation AOA estimations of n=0

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app
 5144 IEEE SENSORS JOURNAL, VOL. 19, NO. 13, JULY 1, 2019

Fig. 2. CSI phase measurements of the shared antenna on the developed L-shaped array. (a) Raw phase measurements on the share antenna. (b) Sanitized
phase measurements on the shared antenna. (c) Aligned phase measurements on the shared antenna between two commodity WiFi NICs.

where 2πkτ0 /N is the phase shift caused by the time shifting

τ0 at the kth subcarrier. It intuitively indicates that the time
shifting τ0 in the time domain causes a linear phase shift in
the frequency domain, which can be examined by the phase
difference of the sanitized phase measurements between two
WiFi NICs on the shared antenna [27]. Let (k) denotes
the sanitized phase difference at kth subcarrier on the shared
antenna between two WiFi NICs. The phase shifts caused by
the time shifting τ0 can be calculated as
  
K
â, b̂ = arg min ((k) + 2π(k − 1) f δ · a + b)2 , (3)
a,b
k=1

where f δ is the frequency interval between the adjacent

subcarriers. With the linear fitting parameters â and b̂, we can
turn away the phase shifts caused by the time shifting τ0 to
align the sanitized phase measurements on the shared antenna
from two commodity WiFi NICs. Fig. 3. Framework of the L-shaped array for both azimuth and elevation
AOA estimations.
As shown in Fig. 1, the developed L-shaped array is
composed of two commodity WiFi NICs that one can be calibrated [28], [29] as
set as the primary, whereas the other one is considered as  
− j 2π(k−1) f δ â+b̂
the secondary. To validate the effectiveness of CSI align- H2 (m, k) = H2 (m, k)e . (4)
ment between two commodity WiFi NICs, we illustrate CSI
Thus, the alignment of CSI measurements can come true with
phase measurements of the shared antenna on the developed
the developed L-shaped array for both azimuth and elevation
L-shaped array in Fig. 2. Firstly, raw phase measurements
AOA measurements of the same incident signals, i.e., 2D-AOA
on the shared antenna are noisy with random phase offsets
estimation.
in Fig. 2(a). Then, we remove random phase offsets on the
shared antenna with each commodity WiFi NIC. The results
B. Signal Model for 2D-AOA Estimation
of the sanitized phase measurements on the shared antenna are
shown in Fig. 2(b). The phase difference between two WiFi As shown in Fig. 3, the L-shaped array consists of two
NICs on the particular subcarrier is the phase shift caused uniform linear subarrays with an equal spacing d on both x and
by the time shifting τ0 . Finally, we eliminate the phase shift z axes. Each uniform linear subarray has M omnidirectional
caused by τ0 on the shared antenna, and thus, the sanitized antennas, e.g., typically M = 3 for the commodity WiFi
phase measurements from two commodity WiFi NICs can be NIC. Suppose that there are L incident signals arrived at the
reconciled with each other in Fig. 2(c). azimuth AOA θl and the elevation AOA φl , l = 1, 2, . . . , L,
Furthermore, CSI readings from two commodity WiFi on the L-shaped array. The steering vectors containing AOA
NICs can be aligned for the azimuth and elevation AOA information with each uniform linear subarray can be defined,
estimations of the same incident signal using the developed respectively as
L-shaped array. Let H1(m, k) at the kth subcarrier on the  T
a1 (θl , φl ) = 1, 1 , . . . , 1M−1 , (5)
mth antenna be CSI readings from the primary WiFi NIC.
 T
In the meanwhile, the other CSI readings H2 (m, k) at the kth
a2 (φl ) = 1, 2 , . . . , 2M−1 , (6)
subcarrier on the mth antenna from the secondary one can be

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app
 ZHANG AND WANG: 3D-WiFi: 3D LOCALIZATION WITH COMMODITY WiFi 5145

where the phase difference between the adjacent antennas with

the corresponding subarray is expressed as
 
2πd cos θl cos φl
1 = exp − j , l = 1, 2, . . . , L, (7)
λ
 
2πd sin φl
2 = exp − j , l = 1, 2, . . . , L, (8)
λ
Fig. 4. Illustration of AOA estimation. (a) MUSIC. (b) Sparse reconstruction.
that λ is the wavelength of transmitted WiFi signal. The Its performance of AOA estimation is better than MUSIC algorithm with sharp
peak.
elevation angle φl is stand-alone in the steering vector 2 ,
which can be directly estimated using the uniform linear spatial grids, it can obtain the sharp peeks of AOA spectrum
subarray on the z axis. While it is hard for the other subarray and overcome rough AOA estimation with a modest set of
on the x axis to estimate the azimuth angle θl , there is the antennas. By comparing with MUSIC, higher AOA accuracy
mixture of the azimuth AOA θl and the elevation AOA φl in the with sharper peak can be achieved using sparse reconstruction
steering vector 1 . We can denote the associated AOA ϕl as with one uniform linear subarray on commodity WiFi NIC,
sin ϕl = cos θl cos φl , l = 1, 2, . . . , L. (9) as illustrated in Fig. 4. The algorithm of sparse reconstruction
for AOA estimation is elaborated in the following Section.
Therefore, the steering vector for the uniform linear subarray
on the x axis can be rephrased as C. Direct Path Identification
 T In complicated indoor environments, we require to identify
a1 (ϕl ) = 1, 1 , . . . , 1M−1 , (10) the direct path to localize the target, which is a crux of 3D
where localization. In fact, the direct path is much more stable than
  other reflected paths with the shortest path. Even when the
2πd sin ϕl
1 = exp − j , l = 1, 2, . . . , L. (11) direct path is in the none-line-of-sight condition, the shortest
λ
path can be most neighboring to the target direct path [35].
With the developed L-shaped array, the received sig- Based on this criteria, the direct path or the path closest to the
nal at the uniform linear subarray on the x axis is direct path should be refined from all paths by the minimal
H1 = [H1(1), H1(2), . . . , H1(M)]T . Meanwhile, the received TOA measurement.
signal at the other subarray on the z axis is H2 = Fortunately, fine-grained CSI readings also provide the
T
H2 (1), H2 (2), . . . , H2 (M) . The received signals can be the measurable phase information on multiple subcarriers, which
linear combination of the L incident signals, respectively as can be employed for TOA measurements. Due to hardware
imperfection on the commodity WiFi NIC, the TOA estimation
H1 = A(ϕ)s1 + n1 , (12)
cannot capture the true time taken by the signal propagating
H2 = A(φ)s2 + n2 , (13) in the environment, but it can introduce the equivalent TOA
where s1 and s2 are the L × 1 signal vectors for the incident measurement to differentiate multipath in indoor scenarios.
signals, and the array manifold matrixes for each uniform Therefore, we can realize and utilize the equivalent TOA esti-
linear subarray are represented as mation to identify the direct path, which should demonstrate
the smallest TOA measurement than other paths in indoor
A(ϕ) = [a(ϕ1 ), a(ϕ2 ), . . . , a(ϕ L )], (14) scenarios.
A(φ) = [a(φ1 ), a(φ2 ), . . . , a(φ L )]. (15) For the equivalent TOA estimation, the phase difference
between the adjacent subcarriers is defined as
In this framework, the associated AOA ϕl and the elevation
AOA φl can be independently estimated using each uniform τ = exp (− j 2π f δ τ ), (16)
linear subarray with an ordinary estimation algorithm, and
where τ is denoted as the equivalent TOA of the path in the
further we can obtain the azimuth AOA estimation θl by (9).
environment. We fully utilize the abundant phase information
Therefore, we can realize both azimuth and elevation AOA
of CSI measurements across not only multiple antennas but
estimations using the develop L-shaped array, simultaneously.
also multiple subcarriers to realize joint AOA and equivalent
Classical MUSIC algorithm [30] is commonly adopted to
TOA estimation (JATE) with each uniform linear subarray,
realize AOA estimation by eigenvalue decomposition, which
respectively. The joint steering vectors for JATE are composed
relies on the orthogonality of the signal subspace and the
of the phase differences across both K subcarriers and M
noise subspace. In effect, this orthogonality is insensitive to
antennas on each uniform linear subarray, respectively as
the coherence of multipath signals in an indoor scenario,
a(ϕ, τ ) = [1, . . . , τK −1 , . . . , 1M−1 , . . . , 1M−1 τK −1 ] ,
T
which results in AOA accuracy degradation, even a failure
of AOA estimation without sufficient antennas, e.g. typical  
antenna 1 antenna M
three antennas on the commodity WiFi card. In an alternative (17)
perspective, since multipath signals are highly sparse in the
a(φ, τ ) = [1, . . . , τK −1 , . . . , 2M−1 , . . . , 2M−1 τK −1 ] .
T
spatial dimension, AOA accuracy can be effectively enhanced  
with sparse reconstruction in a compressive sensing frame- antenna 1 antenna M
work [32]–[34]. By enabling the sparsity on the sampling (18)

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app
 5146 IEEE SENSORS JOURNAL, VOL. 19, NO. 13, JULY 1, 2019

Owing to the accessibility to the commodity WiFi chipsets, with second-order cone (SOC) programming [37], [38]. Once
each CSI measurement usually includes 30 subcarriers on the spatial sparse spectrums could be work out, the sharp peaks
3 antennas using the standard Intel 5300 WiFi NIC, so that of spatial sparse spectrums can be translated to JATE of the
the CSI vectors H 1 and H 2 from the developed L-shaped array each path with each uniform linear subarray. Therefore, JATE
can be expressed as (19) and (20), as shown at the bottom of with two commodity WiFi NICs can be obtained and then the
this page, respectively. direct path can be identified by the minimal TOA measurement
According to the spatial sparsity of multipath signals, for 3D localization.
we adopt the sparse reconstruction algorithm to realize JATE
of each path with high accuracy. To cope with the prob- D. Subspace Sparse Reconstruction Based on
lem of sparse reconstruction, the CSI measurements should 1 -Methodology SVD
be sparsely represented. We require to partition the two- In order to realize the target localization, the adequate pairs
dimensional grids of all possible angles and time delays of JATE from dozens of CSI measurements are clustered to
with an equal spacing, i.e., a sampling spatial grid for AOA discriminate multipath and identify the direct path [16], [20],
estimation as { ϕ1 ,  ϕ Na } and {
ϕ2 , . . . ,  2 , . . . , 
φ1 , φ φ Na } with [21]. However, when the target is slowly moving or static
each uniform linear subarray respectively, and a sampling in a short time, we prefer to combine such independent
time grid of all possible delays for equivalent TOA estimation estimations from multiple CSI measurements and refine one
as {τ1 ,
τ2 , . . . ,
τ Nτ }1 . The bases of sparse representation are representative estimation by coherent combined processing
comprised of the joint steering vectors of all possible angles in a fusion method, effectively reducing computational com-
and time delays as (21) and (22), as shown at the bottom plexity. Firstly, we combine the coherence of multiple CSI
of this page, respectively, where Aϕτ and Aφτ are the size measurements using SVD. Then, the signal subspace is uti-
of 90×Nτ Na . With the over-completed bases, the aligned lized to improve multipath resolution with the azimuth AOA,
CSI measurements from each uniform linear subarray can be elevation AOA, and equivalent TOA estimations, which can
sparsely represented as be formulated as subspace-based sparse reconstruction based
1 = Aϕτ sϕτ + n1 , on 1 -methodology optimization.
H (23)
The CSI matrixes with multiple snapshots are collected with
2 = Aφτ sφτ + n2 , each uniform liner subarray, as H1 = H 11 , H
21 , . . . , H
1Nt
H (24)  
and H2 = H 12 , H
22 , . . . , H
2Nt . At the same time, S1 =
where the sparse spectrums sϕτ and sφτ are the Nτ Na ×1 vec-    
Nt Nt
tors that should own non-zero elements mapping the JATE of s1ϕτ , s2ϕτ , . . . , sϕτ and S2 = s1φτ , s2φτ , . . . , sφτ are defined,
each path in the indoor scenario, and zero otherwise. In com- respectively. With the over-complete bases of sparse repre-
pressive sensing theory, the sparse spectrums sϕτ and sφτ can sentation, the CSI matrixes can be expressed with the noise,
be reconstructed by regularizing it based on 1 -methodology respectively as
[33], [34], [36] that is formulated by the appropriated objective
H1 = Aϕτ S1 + N1 , (27)
functions respectively as
 2  
min  H1 − Aϕτ sϕτ 2 + κ sϕτ 1 , (25) H2 = Aφτ S2 + N2 . (28)
 2   We perform the SVD of H1 and H2 that are decomposed
min H2 − Aφτ sφτ 2 + κ sφτ 1 , (26) into the signal subspace and the noise subspace. With robust-
ness to the noise, the signal subspace can be used to estimate
where κ is a regularization factor that controls a tradeoff
JATE with each uniform linear subarray, which models the
between the sparsity of spatial spectrum and the residual norm.
problem as subspace-based sparse reconstruction. We hold on
Finding the numerous solutions of the sparse spectrums is a
the over-complete bases for sparse representation of the signal
convex optimization problem that can be effectively tackled
subspace, and thus, the signal subspace HSV1 and H2 can be
SV

1 The range of equivalent TOA is normally up to the maximum of sampling expressed, respectively as
delay as [0, τmax ], where τmax = 1/ f δ . As IEEE 802.11a/g/n standards, f δ HSV
1 = Aϕτ S1 + N1 ,
SV SV
(29)
depends on the CSI settings. In the proposed 3D-WiFi system, Intel 5300 WiFi
NIC is set up in 5GHz WiFi with 40MHz bandwidth, f δ = 1.25MHz, thus
τmax = 800ns. HSV
2 = Aφτ S2 + N2 .
SV SV
(30)

1 = [H1 (1, 1), H1(1, 2), . . . , H1 (1, 30), . . . , H1(3, 1), H1(3, 2), . . . , H1(3, 30)]T ,
H (19)
2 = H2 (1, 1), H2 (1, 2), . . . , H2 (1, 30), . . . , H2 (3, 1), H2 (3, 2), . . . , H2 (3, 30) T .
H (20)
   
Aϕτ = a (
ϕ1 ,τ1 ), . . . , a ϕ1 ,
τ Nτ , a (
ϕ2 ,
τ1 ), . . . , a ϕ2 ,
τ Nτ , . . . , a(
ϕ Na , τ1 ), . . . , a(
ϕ Na ,
τ Nτ ) , (21)
  
size of 90×N size of 90×N size of 90×N
  τ    τ    τ 
Aφτ = a  τ1 , . . . , a 
φ1 , φ1 ,
τ Nτ , a φ τ1 , . . . , a 
2 , φ2 ,
τ Nτ , . . . , a φ τ1 , . . . , a 
Na , φ Na ,
τ Nτ . (22)
  
size of 90×Nτ size of 90×Nτ size of 90×Nτ

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app
 ZHANG AND WANG: 3D-WiFi: 3D LOCALIZATION WITH COMMODITY WiFi 5147

Fig. 5. Joint AOA and equivalent TOA estimations with one uniform linear
subarray. (a) Sparse reconstruction using one CSI measurement. (b) Subspace-
based sparse reconstruction using 50 CSI measurements with finer accuracy
in a fusion method.

where HSV1 and H2 are the sizes of 90×L that are effectively
SV

reduced from Nt to L. Differently from a single CSI measure-

Fig. 6. 3D localization algorithm.
ment by (23) and (24), SSV 1 and S2 have to be enforced in
SV

the spatial dimension by computing the 2 -norm of all sparse

time-sample vectors, i.e., the singular vectors in SSV identifying the target direct path, we obtain the azimuth and
1 and S2 ,
SV
(2 ) (2 ) elevation AOA estimations of the target direct path from
respectively. The sparsity of resultant vectors sϕτ and sφτ
two or more receivers to localize the target in 3D space. Let
by Nτ Na ×1 can reveal the sparsity of spatial spectrums. By
( ) ( ) us first localize the target in 2D plane defined by the x and
penalizing the 1 -norm of sϕτ2 and sφτ2 , we can reconstruct
y axes. By deploying all receivers in the same horizontal
the sparse spectrum as
 2 plane, the target location is estimated by the intersection of
 SV  (2 ) at least two azimuth AOA beams or by the centroid of more
min HSV1 − Aϕτ S1  + κsϕτ 1 , (31)
f than two intersections. We calculate the distance between each
 2 receiver and the 2D location estimation of the target. And then,
 SV  (2 )
min HSV
2 − A φτ S2  + κsφτ 1 . (32) with both such distances and the corresponding elevation AOA
f
estimation from each receivers, the height location of the target
where · f is defined as the Frobenius norm2 and the sparse by the z axis can be estimated by the intersection of each two
vectors can be calculated, respectively as elevation AOA beams in the 3D space as shown in Fig. 6.

   L 
τ Na 
N 2
Therefore, 3D locations of the target can be determined with
 (2 )   the azimuth and elevation AOA measurements of the target
sϕτ  = SV (l) , ∀i,
sφτ i (33)
1 direct path from two or more receivers.
i=1 l=1
3D-WiFi localization system can straightforwardly augment

   L
τ Na 
N
an ability of WiFi-based location in 3D space with AOA mea-
 (2 )   (s SV (l))2 , ∀i. surements. In practice, the developed L-shaped array should
sφτ  = ϕτ i (34)
1 be placed at relative low height location to contribute the
i=1 l=1
elevation AOA estimation, while all receivers are on the same
The objective functions can be formulated as a convex horizontal plane. 3D-WiFi localization system requires the
optimization problem, which can be solved by an interior point minimum number of two receivers to resolve 3D location of
implementation in SOC constraints [33]. In this framework, the target, adding more receivers with more accurate location
the spatial sparse spectrums can be reconstructed with sharp estimation in 3D space.
peaks, which corresponds JATE of the each path sing each
uniform linear subarray. Therefore, the pairs of JATE from IV. E XPERIMENTAL E VALUATION
two commodity WiFi NICs can be incorporated to obtain the
azimuth and elevation AOA estimation with high accuracy for To validate the effectiveness of the propose 3D-WiFi local-
3D localization. As illustrated in Fig. 5, we demonstrate the ization system, we conduct the extensive experiments in two
improvement of JATE with subspace-based sparse reconstruc- indoor scenarios: the lobby and the corridor. The size of the
tion. The signal subspace without the noise is utilized to realize lobby is deployed by 6m × 7.8m. The other scenario is the
JATE estimation, which can improve multipath resolution with corridor as the size of 3.5m × 4.2m, which is a rich multipath
finer accuracy as shown in Fig. 5(b). scenario. Each indoor scenario is placed by one transmitter
and up to three receivers. For 3D localization, we deploy
E. 3D Localization test locations in each indoor scenario with both different
height locations and different plane locations. 3D localization
The goal of this section is to realize 3D localization with examples are demonstrated in two indoor scenarios as shown
the informative azimuth and elevation AOA estimations. After in Fig. 7.
2 The Frobenius norm is the matrix norm, defined as  A
 f = We develop a prototype 3D-WiFi system using commercial

m n H Intel 5300 WiFi NICs in several mini-desktops as shown
j =1 |ai, j | = tr( A A), where A is the conjugate transpose
2 H
i=1
matrix of A. in Fig. 8(a). Linux CSI Tool [39] is utilized to obtain

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app
 5148 IEEE SENSORS JOURNAL, VOL. 19, NO. 13, JULY 1, 2019

Fig. 7. 3D localization examples in two indoor scenarios. (a) Lobby, z = 0.6m. (b) Lobby, z = 1.2m. (c) Corridor, z = 0.6m. (d) Corridor, z = 1.2m.

PHY layer CSI measurements. Because of firmware limi-

tation in 2.4GHz [16], [29], we measure CSI readings on
the Channel 132 in 5.66GHz WiFi with 40MHz bandwidth.
The WiFi transmitter transmits 2500 packets per second
using packet injection mode, and all the receivers mon-
itor the traffic and record CSI measurements. As shown
in Fig. 8(b), the developed L-shaped array consists of two
commodity WiFi NICs with external uniform linear sub-
array at the spacing of half-wavelength (2.6cm). One the
shared antenna stitches two commodity WiFi NICs by the
power splitter. The central server incorporates CSI measure-
ments from all receivers to realize 3D location estimation
Fig. 8. Experiment settings. (a) Mini-desktop with Intel 5300 NIC.
accordingly. The localization algorithm of 3D-WiFi system is (b) Developed L-shaped array with one shared antenna stitching two com-
implemented using MATLAB in current prototype. We use a modity WiFi NICs by the power splitter.
MATLAB-based CVX model to tackle with subspace-based
sparse reconstruction for the azimuth and elevation AOA MUSIC algorithm on each uniform linear subarray. Firstly,
estimation. JATE-MUSIC uses the developed L-shaped array to receive
CSI readings and align CSI measurements with our proposed
A. Localization Performance scheme in Section III-A. Then, JATE-MUSIC enlarges the
To validate the localization performance of the pro- CSI values in a certain order by smoothing processing for
posed 3D-WiFi system, we compare it with JATE-MUSIC JATE with each uniform linear subarray and obtains both
that obtains JATE of each path with two-dimensional azimuth and elevation AOA estimations on the developed

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app
 ZHANG AND WANG: 3D-WiFi: 3D LOCALIZATION WITH COMMODITY WiFi 5149

Fig. 10. CDFs of 2D-AOA estimation errors in the lobby scenario.

Fig. 9. CDFs of 3D localization errors. (a) Lobby. (b) Corridor. (a) Azimuth AOA estimation. (b) Elevation AOA estimation.

L-shaped array. By clustering the sufficient azimuth and ele- Furthermore, the proposed 3D-WiFi localization system
vation AOA estimations from multiple CSI measurements, uses the off-the-shelf infrastructures without other auxil-
multipath signals in the indoor scenarios can be discriminated iary devices at low cost and can achieve a satisfied accu-
and the direct path can be identified for 3D localization. In racy at decimeter-level. Therefore, the favorable performance
the experimental evaluations, we evaluate the effectiveness of of 3D-WiFi localization system accomplishes an attractive
the proposed 3D-WiFi system and validate the localization tradeoff between location accuracy and infrastructure cost.
performance in different indoor scenarios.
Fig. 9 illustrates the cumulative distribution func-
B. Direct Path Identification Accuracy
tions (CDFs) of 3D localization errors in two indoor sce-
narios. From the results, we can discover that our proposed The azimuth and elevation AOA estimations are a key factor
3D-WiFi outperforms JATE-MUSIC in all scenarios. The of 3D localization with AOA measurements. We evaluate the
median localization errors of 3D-WiFi system are 0.78m and accuracy of AOA estimation by comparing the difference
1.06m in the lobby and the corridor, respectively. Compared between the ground truth AOA of direct path and the identified
with JATE-MUSIC, the median localization errors of the AOA of direct path with both 3D-WiFi and JATE-MUSIC.
proposed 3D-WiFi system reduce about 18% and 32% in the Fig. 10 illustrates the CDFs of 2D-AOA estimation errors
lobby and the corridor, respectively. These results of 3D-WiFi in the lobby scenario. With the proposed 3D-WiFi system,
localization accuracy demonstrate the effectiveness of the pro- the median errors are 8.9◦ for the azimuth AOA and 6.7◦
pose 3D-WiFi localization system with AOA measurements on for the elevation AOA in the lobby, respectively. Compared
not only the designed hardware scheme, but also the azimuth with JATE-MUSIC, the proposed 3D-WiFi system reduces the
and elevation AOA estimation with subspace-based sparse median errors as 17% for the azimuth AOA estimation and
reconstruction. Therefore, the proposed 3D-WiFi localization 23% for the elevation AOA estimation, respectively. In the
system can achieve good localization performance in indoor corridor, Fig. 11 illustrates the CDFs of 2D-AOA estimation
scenarios. errors. The proposed 3D-WiFi system realizes an improvement

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app
 5150 IEEE SENSORS JOURNAL, VOL. 19, NO. 13, JULY 1, 2019

Fig. 11. CDFs of 2D-AOA estimation errors in the corridor scenario. Fig. 12. Impact of the deployment density. (a) Lobby. (b) Corridor.
(a) Azimuth AOA estimation. (b) Elevation AOA estimation.

0.67m using 2, and 3 receivers in the lobby, and 1.12m and

of 4.8◦ for the azimuth AOA and 5.5◦ for the elevation AOA, 0.98m in the corridor, respectively.
respectively.
The performance of 3D-WiFi system gains benefits from
D. Impact of CSI Measurements Fusion
not only the developed L-shaped array but also subspace-
based sparse reconstruction, which can realize 2D-AOA esti- The proposed 3D-WiFi localization system adopts a fusion
mation by leveraging a fusion method of multiple CSI method of multiple CSI measurements with SVD and utilizes
measurements. The experimental results can verify that the the signal subspace to realize the azimuth and elevation
proposed 3D-WiFi system achieves the azimuth and eleva- AOA estimation for 3D localization. Therefore, the number
tion AOA estimations with a reasonable accuracy for 3D of multiple CSI measurements determines 3D localization
localization. performance to some extent. We carry out the experiments
with the various number of CSI measurements in each indoor
scenario. Fig. 13 illustrates the median errors of 3D local-
C. Impact of Deployment Density ization in two indoor scenarios. In particular, the median
The proposed 3D-WiFi system realizes 3D localization with localization errors are 1.62m in the lobby and 2.41m in the
the azimuth and elevation AOA estimations from two or more corridor using one CSI measurement, i.e., without a fusion
WiFi receivers. The localization performance should be method of CSI measurements. In stark contrast, 3D localiza-
impacted by the number of available receivers. With more tion performance should be quite better with a larger number
receivers, the localization algorithm will have more valu- of CSI measurements. It is shown that 50 CSI measurements
able location measurements, which can improve the location are sufficient for the propose 3D-WiFi system to achieve a
accuracy. Thus, the localization performance should be better reasonable localization performance, which can guarantee the
with more receivers. The results of 3D localization errors tradeoff between the location accuracy and infrastructure cost.
in Fig. 12 confirm our analysis in two indoor scenarios. We can Therefore, 3D-WiFi localization system can benefit from a
discover that the median localization errors are 0.86m and fusion method of multiple CSI measurement.

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app
 ZHANG AND WANG: 3D-WiFi: 3D LOCALIZATION WITH COMMODITY WiFi 5151

[3] D. Moore, J. J. Leonard, D. Rus, and S. Teller, “Robust distributed

network localization with noisy range measurements,” in Proc. ACM
Conf. Embedded Net. Sens. Syst. Baltimore, MD, USA, Nov. 2004,
pp. 50–61.
[4] A. J. Ruiz-Ruiz, P. E. Lopez-De-Teruel, and O. Canovas, “A multi-
sensor LBS using SIFT-based 3D models,” in Proc. Int. Conf. Indoor
Positioning Indoor Navigat. (IPIN). Sydney, NSW, Australia, Nov. 2012,
pp. 1–10.
[5] C. Meier, A. Terzis, and S. Lindenmeier, “A robust 3D high preci-
sion radio location system,” in IEEE MTT-S Int. Microw. Symp. Dig.
Honolulu, HI, USA, Jun. 2007, pp. 397–400.
[6] M. R. Mahfouz, C. Zhang, B. C. Merkl, M. J. Kuhn, and
A. E. Fathy, “Investigation of high-accuracy indoor 3D positioning using
UWB technology,” IEEE Trans. Microw. Theory Techn., vol. 56, no. 6,
pp. 1316–1330, Jun. 2008.
[7] C. Zhang, M. Kuhn, B. Merkl, M. Mahfouz, and A. E. Fathy, “Develop-
ment of an UWB indoor 3D positioning radar with millimeter accuracy,”
in IEEE MTT-S Int. Conf. Microw. Symp. Dig. San Francisco, CA, USA,
Jun. 2006, pp. 106–109.
[8] F. Adib, Z. Kabelac, D. Katabi, and R. C. Miller, “3D tracking via body
radio reflections,” in Proc. USENIX Symp. (NSDI), Seattle, WA, USA,
Apr. 2014, pp. 317–329.
[9] A. Stelzer, K. Pourvoyeur, and A. Fischer, “Concept and application of
LPM—A novel 3D local position measurement system,” IEEE Trans.
Microw. Theory Techn., vol. 52, no. 12, pp. 2664–2669, Dec. 2004.
[10] J. Maneesilp, C. Wang, H. Wu, and N. F. Tzeng, “RFID support
for accurate 3D localization,” IEEE Trans. Comput., vol. 62, no. 7,
pp. 1447–1459, Jul. 2013.
[11] C. Wang, H. Wu, and N.-F. Tzeng, “RFID-Based 3-D positioning
schemes,” in Proc. IEEE 26th IEEE Int. Conf. Comput. Commun.
(INFOCOM). Barcelona, Spain, May 2007, pp. 1235–1243.
[12] J. Hightower, R. Want, and G. Borriello, “SpotON: An indoor 3D loca-
tion sensing technology based on RF signal strength,” Univ. Washington,
Seattle, WA, USA, Tech. Rep. UW CSE 2000-02-02, Feb. 2000.
[13] E. L. Berz, D. A. Tesch, and F. P. Hessel, “Machine-learning-based
system for multi-sensor 3D localisation of stationary objects,” IET
Cyber-Phys. Syst., Theory Appl., vol. 3, no. 2, pp. 81–88, Jun. 2018.
[14] Z. Yang, Z. M. Zhou, and Y. H. Liu, “From RSSI to CSI: Indoor
localization via channel response,” ACM Comput. Surv., vol. 46, no. 2,
Nov. 2013, Art. no. 25.
[15] C. Yang and H.-R. Shao, “WiFi-based indoor positioning,” IEEE Com-
mun. Mag., vol. 53, no. 3, pp. 150–157, Mar. 2015.
Fig. 13. Impact of CSI fusion method. (a) Lobby. (b) Corridor. [16] M. Kotaru, K. Joshi, D. Bharadia, and S. Katti, “SpotFi: Decimeter level
localization using WiFi,” in Proc. ACM SIGCOMM Comput. Commun.
V. C ONCLUSION New York, NY, USA, Aug. 2015, pp. 269–282.
[17] J. Xiong and K. Jamieson, “ArrayTrack: A fine-grained indoor location
We have designed and developed a novel 3D localization system,” in Proc. USENIX Symp. (NSDI). Chicago, IL, USA, Apr. 2013,
system, named 3D-WiFi, that realizes the azimuth and ele- pp. 71–84.
[18] S. Sen, J. Lee, K. Kim, and P. Congdon, “Avoiding multipath to revive
vation AOA estimations to localize the target in 3D space by inbuilding WiFi localization,” in Proc. ACM Mobile Syst. Appl. Serv.
using the off-the-shelf WiFi infrastructures. We implement the Taipei, Taiwan, Jun. 2013, pp. 249–262.
proposed 3D-WiFi system and validate its performance in two [19] W. Gong and J. Liu, “Robust indoor wireless localization using sparse
recovery,” in Proc. IEEE 37th Int. Conf. Distrib. Comput. Syst. (ICDCS).
indoor scenarios. Experimental results demonstrate that the Jun. 2017, pp. 847–856.
proposed 3D-WiFi system can achieve the median error of 3D [20] L. Zhang, Q. Gao, X. Ma, J. Wang, T. Yang, and H. Wang, “DeFi:
localization as 0.78m in a typical indoor scenario, which can Robust training-free device-free wireless localization with WiFi,” IEEE
Trans. Veh. Technol., vol. 67, no. 9, pp. 8822–8831, Sep. 2018.
accomplish an attractive tradeoff between location accuracy [21] X. Li, S. Li, D. Zhang, J. Xiong, Y. Wang, and H. Mei, “Dynamic-
and infrastructure cost. MUSIC: Accurate device-free indoor localization,” in Proc. ACM Int.
3D-WiFi system explores on AOA measurements to localize Conf. Pervasive Ubiquitous Comput. Heidelberg, Germany, Sep. 2016,
pp. 196–207.
the target in 3D space by using the COTS WiFi infrastructures, [22] K. Joshi, D. Bharadia, M. Kotaru, and S. Katti, “WiDeo: Fine-grained
which can provide a wide range of location-based applications device-free motion tracing using RF backscatter,” in Proc. USENIX
such as device-free 3D localization, smart health monitoring Symp. (NSDI). Oakland, CA, USA, May 2015, pp. 189–204.
[23] Y. Xie, J. Xiong, M. Li, and K. Jamieson, “xD-track: Leveraging multi-
and motion tracking. We will try to solve these issues in our dimensional information for passive Wi-Fi tracking,” in Proc. ACM Int.
future work. Conf. HotWireless, New York, NY, USA, Oct. 2016, pp. 39–43.
[24] Y. Hua, T. K. Sarkar, and D. D. Weiner, “An L-shaped array for esti-
R EFERENCES mating 2-D directions of wave arrival,” IEEE Trans. Antennas Propag.,
vol. AP-39, no. 2, pp. 143–146, Feb. 1991.
[1] A. Yassin et al., “Recent advances in indoor localization: A survey on [25] N. Tayem and H. M. Kwon, “L-shape 2-dimensional arrival angle
theoretical approaches and applications,” IEEE Commun. Surv. Tuts., estimation with propagator method,” IEEE Trans. Antennas Propag.,
vol. 19, no. 2, pp. 1327–1346, 2nd Quart., 2017. vol. 53, no. 5, pp. 1622–1630, May 2005.
[2] H. Liu, H. Darabi, P. Banerjee, and J. Liu, “Survey of wireless indoor [26] J. Liang and D. Liu, “Joint elevation and azimuth direction finding
positioning techniques and systems,” IEEE Trans. Syst., Man, Cybern. using L-shaped array,” IEEE Trans. Antennas Propag., vol. 58, no. 6,
C, Appl. Rev., vol. 37, no. 6, pp. 1067–1080, Nov. 2007. pp. 2136–2141, Jun. 2010.

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app
 5152 IEEE SENSORS JOURNAL, VOL. 19, NO. 13, JULY 1, 2019

[27] Y. Xie, Z. Li, and M. Li, “Precise power delay profiling with commodity [38] A. Nemirovski and A. Ben Tal, Lectures Modern Convex Optimization:
WiFi,” in Proc. ACM Int. Conf. Mobile Comput. Netw., Paris, France, Analysis, Algorithms, Engineering Applications. Philadelphia, PA, USA:
Sep. 2015, pp. 53–64. SIAM, 2001.
[28] Y. Xie, Y. Zhang, J. C. Liando, and M. Li, “SWAN: Stitched Wi-Fi [39] D. Halperin, W. Hu, A. Sheth, and D. Wetherall, “Tool release: Gath-
ANtennas,” in Proc. ACM Int. Conf. Mobile Comput. Netw. New Delhi, ering 802.11n traces with channel state information,” ACM SIGCOMM
India, Oct. 2018, pp. 51–66. Comput. Commun. Rev., vol. 41, no. 1, p. 53, Jan. 2011.
[29] J. Gjengset, J. Xiong, G. McPhillips, and K. Jamieson, “Phaser: Enabling
phased array signal processing on commodity WiFi access points,”
in Proc. ACM Int. Conf. Mobile Comput. Netw. Honolulu, HI, USA, Lingyan Zhang received the B.S. degree from the
Sep. 2014, pp. 153–164. Nanjing University of Technology, Nanjing, China,
[30] R. Schmidt, “Multiple emitter location and signal parameter estima- in 2007, and the M.S. degree from the Shenzhen
tion,” IEEE Trans. Antennas Propag., vol. AP-34, no. 3, pp. 276–280, Graduate School, Harbin Institute of Technology,
Mar. 1986. Shenzhen, China, in 2010. She is currently pursuing
[31] J. M. Kim, O. K. Lee, and J. C. Ye, “Compressive MUSIC: Revisiting the Ph.D. degree with the Department of Electri-
the link between compressive sensing and array signal processing,” IEEE cal Engineering, Dalian University of Technology,
Trans. Inf. Theory, vol. 58, no. 1, pp. 278–301, Jan. 2012. Dalian, China. Her research interests include wire-
[32] I. Bilik, “Spatial compressive sensing for direction-of-arrival estimation less localization and wireless sensor networks.
of multiple sources using dynamic sensor arrays,” IEEE Trans. Aeros.
Elect. Syst., vol. 47, no. 3, pp. 1754–1769, Jul. 2011.
[33] D. Malioutov, M. Cetin, and A. S. Willsky, “A sparse signal reconstruc-
tion perspective for source localization with sensor arrays,” IEEE Trans.
Signal Process., vol. 53, no. 8, pp. 3010–3022, Aug. 2005. Hongyu Wang (M’03) received the B.S. degree in
[34] E. J. Candes and M. B. Wakin, “An introduction to compressive electronic engineering from the Jilin University of
sampling,” IEEE Signal Process. Mag., vol. 25, no. 2, pp. 21–30, Technology in 1990, the M.S. degree in electronic
Mar. 2008. engineering from the Graduate School of the Chi-
[35] S. Kumar, E. Hamed, D. Katabi, and L. E. Li, “LTE radio analytics made nese Academy of Sciences in 1993, and the Ph.D.
easy and accessible,” in Proc. ACM Int. Conf. SIGCOMM. Chicago, IL, degree in precision instrument and optoelectronics
USA, Aug. 2014, pp. 211–222. engineering from Tianjin University in 1997.
[36] D. L. Donoho and M. Elad, “Optimally sparse representation in general He is currently a Professor with the Dalian Uni-
(nonorthogonal) dictionaries via 1 minimization,” Nat. Acad. Sci., versity of Technology. He serves as the Dean of
vol. 100, no. 5, pp. 2197–2202, Mar. 2003. the School of Information and Communication Engi-
[37] M. Lobo, L. Vandenberghe, S. Boyd, and H. Lebret, “Applications neering. His research interests include algorithmic,
of second-order cone programming,” Linear Algebra Appl., vol. 284, optimization, and performance issues in wireless ad hoc, mesh, and sensor
nos. 1–3, pp. 193–228, 1998. networks.

horized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on September 24,2024 at 09:40:23 UTC from IEEE Xplore. Restrictions app