sensors

Article
Acoustic Indoor Localization System Integrating
TDMA+FDMA Transmission Scheme and
Positioning Correction Technique
Xiang Chen *, Yuheng Chen, Shuai Cao, Lei Zhang, Xu Zhang and Xun Chen
Department of Electronic Science and Technology, University of Science and Technology of China,
Hefei 230027, China; cyh1102@mail.ustc.edu.cn (Y.C.); caoshuai@ustc.edu.cn (S.C.);
zl2017@mail.ustc.edu.cn (L.Z.); xuzhang90@ustc.edu.cn (X.Z.); xunchen@ustc.edu.cn (X.C.)
* Correspondence: xch@ustc.edu.cn; Tel.: +86-139-6668-9471




Received: 2 April 2019; Accepted: 17 May 2019; Published: 22 May 2019


Abstract: This paper presents a novel audio indoor localization system. In the proposed system,
four speakers placed at known positions transmit chirp signals according to the time-division
multiple access (TDMA) plus frequency-division multiple access (FDMA) transmission scheme.
A smartphone receives the signal via a built-in microphone and calculates the time differences of
arrival (TDOAs). Using TDOA measurements, the position is estimated by the shrinking-circle
method. In particular, to reduce the positioning error in moving conditions, a TDOA correction
method based on Doppler shifts is proposed. The performance of the proposed system was evaluated
in real-world experiments using a 10.971 m × 5.684 m positioning area. The results of the static-target
positioning experiment showed that the TDMA+FDMA transmission scheme has more advantages
in improving the update rate of the positioning system than the TDMA-only transmission scheme.
The results of the moving-target positioning experiment under three different speeds demonstrated
that the positioning errors were reduced by about 10 cm when the Doppler-shift-based TDOA
correction method was adopted. This research provides a possible framework for the realization
of a TDOA-chirp-based acoustic indoor positioning system with high positioning accuracy and
update rate.

Keywords: Indoor localization; Chirp signal; TDOA; TDMA+FDMA; Doppler shifts

1. Introduction
With the development of intelligent information processing technology and smartphones,
location-based services (LBS) have attracted increased attention. The key for LBS is seamless indoor
and outdoor navigation for users. Although a global positioning system (GPS) or Beidou can work
well for outdoor positioning, they are not effective solutions for indoor positioning due to the blockage
by buildings. Bluetooth, Ultra Wideband (UWB), ultrasound, WiFi, etc., have been used successfully to
realize indoor positioning in many systems. The cost of UWB-based systems is relatively high, and
smartphone sensors often cannot meet the bandwidth required by the system [1,2]. Ultrasound-based
systems also cannot be implemented on smartphones because the signal frequency band often is outside
the acceptance range of the mobile phone microphone [3–7]. Although WiFi and Bluetooth-based
systems are easy to set up due to the wide application of WiFi and Bluetooth, the positioning accuracy of
both these systems can only reach the meter level [2,8], which cannot satisfy the needs of high-precision
indoor positioning.
Recently, audio-based indoor positioning technology that operates by emitting and receiving
acoustic signals between nodes and the positioning target has attracted increased attention due to

Sensors 2019, 19, 2353; doi:10.3390/s19102353 www.mdpi.com/journal/sensors

Sensors 2019, 19, 2353 2 of 18

its unique advantages. Related research shows that audio-based indoor positioning systems can
achieve decimeter-level accuracy [9–11]. In addition, audio-based technology is a low-cost indoor
positioning solution because every commercial off-the-shelf (COTS) smartphone has a built-in speaker
and microphone, and many buildings or public places—such as hospitals, shopping malls, airports,
etc.—have their own speaker and microphone systems.
In audio-based indoor positioning systems, code-division multiple access (CDMA) [12–14] and
linear chirp are the two common audio coding methods [9–11,15,16]. Fernando J. Álvarez et al. realized
an audio-based indoor positioning system that utilized CDMA signals [12]. They evaluated the system
performance under different signal-to-noise ratios (SNRs) by setting two criteria, system availability
(SA, defined as the percentage of measurements whose error is below the outlier’s threshold) and mean
positioning error (MPE). At the center of the positioning area, the SA was 90%, and the MPE was 8 cm
when the SNR was 12 dB. Fabian Hoflinger et al. realized an audio-based indoor positioning system
that utilized chirp signals, and the MPEs were observed to be less than 30 cm for static points and 34 cm
for smartphone tracking [9]. Although the positioning for static targets can be highly accurate most of
the time, many problems appear when the positioning target is moving. Sérgio I.Lopes analyzed the
ambiguity functions of CDMA signal and chirp signal with the same frequency band (18–22 kHz) and
time duration and found that the CDMA signal was easily influenced by Doppler shifts when the target
was moving at a high speed [10,17]. Compared to CDMA, chirp signal pulses have a Doppler tolerance
of up to ±bandwidth/10. That is to say, chirp-based systems may be more accurate for moving target
positioning than CDMA-based systems.
Regarding the aspect of acoustic signal measurement, measurements including the angle of
arrival (AOA), the time difference of arrival (TDOA), the time of arrival (TOA), and the frequency
difference of arrival (FDOA) have been applied to realize audio-based indoor positioning in related
research [18–22]. Among these methods, the measurement of TDOA has the advantage of not requiring
clock synchronization between the target and the nodes. The FDOA method, which measures the
frequency difference, can only be applied when there is relative motion between the target and the
nodes. The joint usage of TDOA and FDOA can estimate the target position and the velocity accurately
when the target is moving [23], and has attracted a lot of research interest in the fields of surveillance,
navigation, wireless communications, and sensor networks [24–27].
Recently, to explore decimeter-level accuracy indoor positioning technology, our research group
tried to develop an audio-based system by adopting TDOA measurement and chirp signals [28]. In the
process of system implementation, some main factors were found to affect the system performance.
As suggested in references, the detection performance of chirp signals was influenced by the
time-bandwidth product (TBP), which is defined as the product of the duration of the signal and
the bandwidth. Generally, the larger the TBP is, the higher the time resolution and the accuracy
of chirp signal detection are. In related research [29], 18–22 kHz has been proven to be a suitable
frequency band for indoor positioning, because it can make full use of the audio-signal-receiving ability
of smartphones and can satisfy the insensitivity of human ears. Under the time-division multiple
access (TDMA) transmission scheme, most chirp-based systems use the whole available bandwidth
to generate chirp signals to increase the TBP [10]. However, using TDMA alone can result in some
other problems. For instance, TDOA values calculated from at least three nodes are usually needed for
location algorithms. At each time point, only a signal from one node can be received when the TDMA
transmission scheme is adopted. Consequently, the TDOA values used for positioning are collected at
different positions when the target is moving, which will inevitably bring about a positioning error.
The other problem of TDMA is the low update rate, because the system starts positioning calculation
only after receiving signals from all the nodes.
To improve the positioning update rate and reduce the positioning error caused by target motion,
a novel indoor positioning system based on chirp audio signals and TDOA was implemented in
this study. The main features of this system can be summarized as follows: (1) a TDMA+FDMA
(frequency-division multiple access) signal transmission scheme was adopted to improve the positioning
 Sensors 2019, 19, 2353 3 of 18

update rate; (2) a target-moving-speed-based TDOA correction method was used to improve the
positioning accuracy. The rest of this paper is organized as follows: Section 2 presents the overview
and the detailed description of each part of the system; Section 3 presents location results and analysis
for static targets and moving targets; and Section 4 presents a discussion and conclusions.

2. Methods

2.1. Overview of the TDOA-Chirp-Based System

Sensors 2019, 19, x FOR PEER REVIEW 3 of 18
Figure 1 shows the diagram of the proposed TDOA-chirp-based indoor positioning system.
results and analysis
The system mainly for static of
consisted targets and moving
four modules, namely,targets; and Section
the audio 4 presents amodule,
signal transmission discussion and
the TDOA
conclusions.
estimation module, the localization module, and the positioning correction module.
The audio signal transmission module consisted of a synchronous node, which was used to
2. generate
Methodstrigger signals and four speaker nodes (Node1–4). The hardware structure of the synchronous
node is shown in Figure 2a. The trigger signals were generated by a microcontroller unit (MCU) and
2.1. Overviewby
converted of the
the TDOA-Chirp-Based
transceiver (MAX3362) Systemto a differential signal that controlled the transmission order
of Node1–4. The hardware structure
Figure 1 shows the diagram of the proposed of the speaker node is shown in
TDOA-chirp-based Figure
indoor 2b. Each node
positioning consisted
system. The
mainly
system of four
mainly parts: an
consisted of MCU, an audionamely,
four modules, codec, thean audio
audio power amplifier, and
signal transmission an omnidirectional
module, the TDOA
speaker. module,
estimation After receiving the synchronization
the localization module, and the trigger signal, the
positioning MCU controlled
correction module. the audio codec
(TLV320AIC23B) to convert the digital audio signal into an analog
The audio signal transmission module consisted of a synchronous node, which signal. Then, the analog
was signal
used towas
amplified
generate by an signals
trigger audio power amplifier
and four and drove
speaker nodesthe speaker. The
(Node1–4). Theamplification factor of the
hardware structure of audio
the
synchronous node is shown in Figure 2a. The trigger signals were generated by a microcontrollerby
power amplifier could be adjusted within a certain power range (0–50 W). The system was powered
a power
unit (MCU)overandEthernet
converted (POE) module.
by the The POE
transceiver was used
(MAX3362) to atodifferential
convert 220V alternating
signal current the
that controlled (AC)
to 24V direct
transmission current
order (DC). Each
of Node1–4. Thenode had a buck
hardware module
structure of theto speaker
power the nodeMCU, the audio
is shown codec,
in Figure 2b.etc.
Each node consisted mainly of four parts: an MCU, an audio codec, an audio power amplifier, andthe
Additionally, the synchronous node had two network ports to control the transmission order of
anspeaker node via aspeaker.
omnidirectional standard network
After cable,the
receiving as synchronization
shown in Figure trigger2c. signal, the MCU controlled
The TDOA estimation module was used to detect the
the audio codec (TLV320AIC23B) to convert the digital audio signal into an analog arrival times of thesignal.
chirp signals
Then, the and
calculate the TDOA. The localization module utilized the TDOA to give
analog signal was amplified by an audio power amplifier and drove the speaker. The amplification an initial estimated position.
The positioning
factor of the audiocorrection modulecould
power amplifier corrected the initial
be adjusted withinestimated
a certainposition to get(0–50
power range moreW). accurate
The
positioning results. The detailed description of each module is given
system was powered by a power over Ethernet (POE) module. The POE was used to convert 220V below. A positioning program
consistingcurrent
alternating of a TDOA(AC)estimation
to 24V direct module,
currenta (DC).
localization
Each nodemodule,
had aand
bucka positioning correction
module to power module
the MCU,
thewas written
audio in C++
codec, etc. and was packaged
Additionally, the into the App written
synchronous node had in java.
twoThe whole ports
network App was installedthe
to control in a
smartphone (Huawei P10 Plus, Shanghai, China). The audio signal
transmission order of the speaker node via a standard network cable, as shown in Figure 2c.sampling frequency was 44.1 kHz.

audio
signal
audio signal
transmission module

Node 1 Node 2

TDOA
estimation

synchronous node localization

positioning
Node 3
Node 4 correction

Figure 1. Diagram
Figure of the
1. Diagram time
of the difference
time of of
difference arrival (TDOA)-chirp-based
arrival system.
(TDOA)-chirp-based system.
Sensors 2019,19,
Sensors2019, 19,2353
x FOR PEER REVIEW 44 of 18
18

(a) (b)

(c)
Figure 2.
Figure 2. Diagram
Diagram of
of the
the system
system hardware.
hardware. (a)
(a) Synchronous
Synchronous node;
node; (b)
(b) speaker
speaker node;
node; (c)
(c) diagram
diagram of
of
node connection.
node connection.

2.2. FDMA+TDMA
The TDOA estimationAudio Signal Transmission
module was used Scheme
to detect the arrival times of the chirp signals and
calculate the TDOA. The localization module
The frequency range of the audio chirp signal wasutilized the TDOA to givekHz
set to 15–22 an initial
in theestimated
proposedposition.
system.
The positioning correction module corrected the initial estimated position to
In FDMA, the 15–22 kHz frequency band was divided into two sub-bands to generate chirp signals. get more accurate
positioning
The frequency results.
of theThe detailed
chirp description
signal shown of each
in Figure module is
3a dropped given
from below.
18 to A positioning
15 kHz, program
and that of the chirp
consisting of a TDOA estimation module, a localization module, and a positioning
signal shown in Figure 3b rose from 19 to 22 kHz. In order to avoid the spectrum leakage caused correction module
was
by written
a sharp in C++ in
increase and was packaged
amplitude of the into
chirpthe Appand
pulse, written in java.
to keep The whole
the pulse App wasainstalled
non-invasive, combined in
a smartphone (Huawei P10 Plus, Shanghai, China). The audio signal sampling
window function (hamming window plus rectangular window) was used to modulate the amplitude frequency was 44.1
kHz.
of the chirp pulse. The 18-19 kHz protection bandwidth was used to avoid interference between the
two chirp signals. In TDMA, the four speaker nodes were divided into two groups to transmit the
2.2. FDMA+TDMA Audio Signal Transmission Scheme
two kinds of chirp signals. As shown in Figure 4, which gives an example of the TDMA+FDMA
The frequency
transmission scheme, range
Node of1the audio18–15
emitted chirp kHz
signal wassignal,
chirp set to 15–22 kHz 2inemitted
and Node the proposed system.
19–22 kHz chirpIn
FDMA,synchronously;
signal the 15–22 kHz frequency bandNode
after 200 ms, was divided
3 emittedinto two sub-bands
18-15 kHz chirp to generate
signal, andchirp
Nodesignals. The
4 emitted
frequency
19–22 of thesignal
kHz chirp chirp synchronously.
signal shown inFor Figure 3a dropped
a given from
positioning 18 tothe
period, 15 kHz,
signaland that ofthe
duration, theguard
chirp
signaland
time, shown in Figure 3b
the remaining rose
time fromto
needed 19be
todesigned.
22 kHz. InAsorder
shownto avoid the 4,
in Figure spectrum leakageperiod
the positioning caused by
was
a sharp
set to 1000increase
ms. Both in amplitude of the had
the chirp signals chirpa pulse, and to
50 ms time keep theApulse
duration. 150 msnon-invasive, a combined
guard time was used to
window function
completely separate (hamming window
the two chirp
Sensors 2019, 19, x FOR PEER REVIEW
plus rectangular
signals, and a 600 ms window)
remainingwastime
usedwas
to modulate the amplitude
used to distinguish the
5 of 18
of the chirpperiod.
positioning pulse. The 18-19 kHz protection bandwidth was used to avoid interference between the
two chirp signals. In TDMA, the four speaker nodes were divided into two groups to transmit the
two kinds of chirp signals. As shown in Figure 4, which gives an example of the TDMA+FDMA
transmission scheme, Node 1 emitted 18-15 kHz chirp signal, and Node 2 emitted 19-22 kHz chirp
signal synchronously; after 200 ms, Node 3 emitted 18-15 kHz chirp signal, and Node 4 emitted 19-
22 kHz chirp signal synchronously. For a given positioning period, the signal duration, the guard
time, and the remaining time needed to be designed. As shown in Figure 4, the positioning period
was set to 1000 ms. Both the chirp signals had a 50 ms time duration. A 150 ms guard time was used
to completely separate the two chirp signals, and a 600 ms remaining time was used to distinguish
the positioning period. (a) (b)
Figure
Figure 3. Waveformillustrations
3. Waveform illustrationsofofchirp
chirp signals.
signals. (a)(a) 18–15
18–15 kHzkHz chirp
chirp signal;
signal; (b) 19–22
(b) 19–22 kHz signal.
kHz chirp chirp
signal.
 (a) (b)
Figure
Sensors 3. 2353
2019, 19, Waveform illustrations of chirp signals. (a) 18–15 kHz chirp signal; (b) 19–22 kHz chirp5 of 18
signal.

Figure
Figure 4. An
An example
example of
of time-division
time-division multiple
multiple access
access plus
plus frequency-division
frequency-division multiple
multiple access
access
(TDMA+FDMA)
(TDMA+FDMA)transmission
transmissionscheme.
scheme.

2.3. TDOA
2.3. TDOAEstimation
Estimation
When the
When the microphone
microphone receivedreceived audio
audio signals,
signals, thethe TDOA
TDOA estimation
estimation module
module could
could detect
detect the
the
arrival times of the chirp signals from different nodes. Generally, the arrival times
arrival times of the chirp signals from different nodes. Generally, the arrival times of chirp signals of chirp signals
were obtained
were obtained as as the
the lag time 𝜏τthat
lag time thatmaximized
maximizedthe thecross
crosscorrelation
correlation(CC)
(CC)between
betweenthethereceived
receivedsignal
signal
and the
and the emitted
emittedsignal,
signal,asas shown
shown in Equation
in Equation Rpq was
(1). (1). 𝑅 thewasCC between
the the received
CC between signal, denoted
the received signal,
as p [ n ] , and the emitted signal, denoted as q [ n ] . E{·} stood for the mathematical
denoted as 𝑝[𝑛] , and the emitted signal, denoted as 𝑞[𝑛] . 𝐸{∙} stood for the mathematical expectation.
expectation.  
τ = argmax Rpq(m)
𝜏 = 𝑎𝑟𝑔𝑚𝑎𝑥
m (𝑅 ( ))

Rpq [m] = E p[n]q (1)
𝑅 [𝑚] =m
[n + 𝐸{𝑝[𝑛]𝑞
]
(1)

[𝑛 + 𝑚]}
However, in reality, the τ computed using Equation (1) was generally not the true time delay,
sinceHowever, in reality,
there existed the 𝜏 computed
a multipath effect. Large using
errorsEquation
may have (1) was generally
occurred whennotthethe true timeCC
maximum delay,
was
used to determine the τ. Considering that the time of the direct path must have been earlier than τ,
since there existed a multipath effect. Large errors may have occurred when the maximum CC was
used to determine
an alternative the 𝜏.
method Considering
was that the
to use the time of atime of the direct
corresponding pathwhose
point must have
valuebeen earlier than
was greater than or𝜏,
equal to the product of a weight coefficient (0 < α < 1) and the maximum CC as the arrival time. α was
an alternative method was to use the time of a corresponding point whose value was greater than or
equal to the product
determined according ofto
a the
weight (0 < 𝛼 < 1) and the maximum CC as the arrival time. α
coefficientenvironment.
experimental
was determined according  to the experimental environment.
 𝑇𝐷𝑂𝐴 ij==𝑡 t−
TDOA i −𝑡 t j 𝑖 i≠,𝑗;j; |𝑖i −
− 𝑗|
j ≤ 1; and
≤ 1; 𝑎𝑛𝑑i,𝑖,j 𝑗,≠2,2,3
3



𝑇𝐷𝑂𝐴
TDOAij==𝑡 t− 200 𝑖 i≠,𝑗;j; |𝑖i − (2)(2)
i −𝑡 t +
j +200 − 𝑗|
j ≥≥ 1;
1; and
𝑎𝑛𝑑i,𝑖,j 𝑗==2,2,3
3


When and t𝑡j were

When 𝑡ti and were used
used to
to represent
represent the
the arrival
arrival times
times of
of the
the chirp
chirp signals
signals the
the smartphone
smartphone
received
received from the ith node and the jth node, respectively, the 𝑇𝐷𝑂𝐴 between the two nodes
node and the jth node, respectively, the TDOAij between the two nodes could could
be
be calculated
calculated according
according to Equation
to Equation (2).(2).

2.4. Positioning
2.4. PositioningAlgorithm
Algorithm
The localization
The localization module
modulewaswasused to get
used the target
to get position
the target through
position the TDOA.
through When the
the TDOA. TDOA
When theij
𝑇𝐷𝑂𝐴
was obtained, the distance
was obtained, the difference between the
distance difference ith andthe
between jth ith
nodes
andcould be calculated
jth nodes could beaccording to
calculated
Equation (3).
according to Equation (3). 
→ → → →
 dij = c × TDOAij = kq x − x ik − k x − x jk



 (3)
T
c = 331.45 × 1 + 273.15



→ → →
where x was the position to locate, x i and x j stood for the positions of the ith and the jth nodes,
respectively, T was the temperature, and c represented the velocity of voice. The essence of localization
using TDOA was an optimization problem, and many methods have been proposed to solve this
problem, including the maximum-likelihood (ML) method [30], the non-linear least-squares (NLS)
approach, the weighted least-squares (WLS) approach, the two-step weighted least-squares (2WLS)
approach, the constrained weighted least-squares (CWLS) approach, the separated CWLS (SCWLS)
approach [31–34], etc.
Sensors 2019, 19, 2353 6 of 18

In our system, the shrinking-circle (SC) method was used to realize target positioning due to its
low complexity and high robustness [28]. As shown in Figure 5, Oi (xi , yi ) denoted the coordinate of
the ith node, T (x, y) denoted the target’s coordinate, ri represented the distance between the target and
the ith node, the target T was on the circumference of a circle with Qi as the center and ri as the radius.
Equation (4) depicts the relationship between these parameters. Taking the 1st node as the reference,
the radius of circle Oi could be determined, as shown in Equation (5). As the di1 could be computed
from the TDOA values, the radius r1 was the only variable required to solve T (x, y). Consequently, the
basic idea of the SC method is to find the perfect radius ri for which all the circles intersect at the same
point. More details about the SC strategies can be found in our previous work [28].

(x − xi )2 + ( y − yi )2 = r2i (4)
Sensors 2019, 19, x FOR PEER REVIEW 7 of 18

ri = r1 + di1 (5)

O2

r2
Target
O1 r1 T(x,y)

r3

O3

Figure 5. Basic
Figure5. Basic idea
idea of
of the
the shrinking-circle
shrinking-circle (SC)
(SC) method.
method.

2.5. Positioning Correction Considering Moving Speed of the Target

As mentioned above, the localization module utilized the TDOA to give an initial estimated
position. However, the Doppler shift caused by target motion could influence TDOA estimation,
resulting in a positioning error. Therefore, we proposed a positioning correction module to modify
the initial estimated position and get more accurate positioning results. The details of the positioning
correction are shown in Figure 6.
In the positioning correction, it was indispensable to estimate the Doppler shift. The Doppler
shift ∆ f caused by a moving target could be depicted using Equation (6) [35].

f
∆ f = of×positioning
Figure 6. Diagram v × cosθ correction. (6)
c
whereTo analyze
f was the effect
the carrier of the Doppler
frequency, c was theshift, we conducted
velocity of voice, v simulations under
was the speed and θ stood
the following
of target, two
situations:
for the angle(1)between
the emitted signal wasof18–15
the directions targetkHz chirp
motion signal,
and audiowhere
signalone received signal
propagation. was 18–15
Supposing that
◦
= 18chirp
fkHz v = 1 m/s
kHz, signal with
and50θms= time delay,
0 , 52.9 and the shift
Hz Doppler othercould
received signal was 18–15 kHz chirp signal
be calculated.
with 52.9 Hz Doppler shift and 50 ms time delay; (2) the emitted signal was 19–22 kHz chirp signal,
where one received signal was 19–22 kHz chirp signal with 50 ms time delay, and the other received
signal was 19–22 kHz chirp signal with 52.9 Hz Doppler shift and 50 ms time delay. With weight
coefficient α = 1, the CC between the emitted and the received signals and the estimated arrival times
were computed. As shown in Figure 7, the arrival time for the 18–15 kHz chirp signal with Doppler
shift was about 50.8 ms, whereas the arrival time for the 19–22 kHz chirp signal with Doppler shift
was about 49 ms. The simulation results show that the 52.9 Hz Doppler shift could cause about 1 ms
error in arrival time estimation. Consequently, when the two kinds of chirp signals with Doppler
shift were the received signals corresponding to two nodes, there existed about 2 ms TDOA
estimation error, which could result in about 68 cm distance difference.
 O3

Sensors 2019, 19, 2353 7 of 18

Figure 5. Basic idea of the shrinking-circle (SC) method.

Figure 6. Diagram of positioning correction.

To
To analyze
analyze the the effect
effect of
of the
the Doppler
Doppler shift,shift, we
we conducted
conducted simulations
simulations under the following two
situations:
situations: (1)(1) the
theemitted
emittedsignal
signalwas was18–15
18–15 kHz
kHz chirp signal,
chirp where
signal, oneone
where received signal
received was was
signal 18–1518–15
kHz
chirp signalsignal
kHz chirp with with
50 ms50time delay,delay,
ms time and the andother received
the other signalsignal
received was 18–15 kHz chirp
was 18–15 signalsignal
kHz chirp with
52.9
withHz 52.9Doppler
Hz Dopplershift and
shift50and
ms50timemsdelay; (2) the(2)
time delay; emitted signal signal
the emitted was 19–22
was kHz
19–22chirp
kHzsignal, where
chirp signal,
one
wherereceived signal was
one received 19–22
signal waskHz 19–22chirp
kHzsignal
chirp with
signal50with
ms time
50 msdelay,
time and theand
delay, other
thereceived signal
other received
was
signal19–22
waskHz19–22chirp
kHzsignal
chirpwith 52.9with
signal Hz Doppler shift and shift
52.9 Hz Doppler 50 msand
time50delay. With
ms time weight
delay. coefficient
With weight
α = 1, the αCC
coefficient = 1,
between
the CC the emitted
between the and the and
emitted received signals signals
the received and theandestimated arrivalarrival
the estimated times times
were
computed.
were computed. As shown in Figure
As shown 7, the7,arrival
in Figure time for
the arrival time the
for18–15 kHz chirp
the 18–15 signalsignal
kHz chirp with Doppler
with Dopplershift
was about 50.8 ms, whereas the arrival time for the 19–22 kHz chirp signal
shift was about 50.8 ms, whereas the arrival time for the 19–22 kHz chirp signal with Doppler shiftwith Doppler shift was
about
was about49 ms.49 The
ms. simulation results
The simulation showshow
results that the
that52.9
theHz52.9Doppler shift could
Hz Doppler causecause
shift could aboutabout
1 ms error
1 ms
in arrival
error time estimation.
in arrival Consequently,
time estimation. when the
Consequently, two kinds
when the twoof chirp
kindssignals
of chirpwith Doppler
signals withshift were
Doppler
the
shiftreceived
Sensors were signals
2019, 19,the
corresponding
received
x FOR
to two nodes,tothere
signals corresponding
PEER REVIEW twoexisted
nodes,about
there2existed
ms TDOA estimation
about 2 ms TDOAerror,
8 of 19
which coulderror,
estimation result in about
which could68 result
cm distance
in about difference.
68 cm distance difference.

(a) (b)
Figure 7. Simulation
Simulation results
results of
of influences
influences of Doppler shift on arrival time detection. (a)
(a) 18–15
18–15 kHz
chirp signal; (b) 19–22 kHz chirp signal.

In this study, an effective method that considered the moving speed speed of the target was used to
correct the positioning error caused by the Doppler shift. The The basic
basic idea
idea was
was to convert
convert the Doppler
shifts in the frequency domain into time differences to update the TDOA
in the frequency domain into time differences to update the TDOA values. values. To simplify the process,
To simplify the
the linearthe
process, chirp signal
linear with
chirp Doppler
signal shift wasshift
with Doppler still was
supposed to be a linear
still supposed to bechirp signal.
a linear chirpThe Doppler
signal. The
shift ∆ f was
Doppler ∆𝑓 was into
shiftconverted the time
converted into the time∆tdifference
difference ∆𝑡 using
using Equation (7), where F was
Equation thewhere
(7), frequency
F was range
the
of the chirprange
frequency signals,
of the Tchirpsignals,
andchirp stood for the𝑇duration
and stoodof the chirp
for the signal. Since
duration of thethe Doppler
chirp signal.shift
Sincecould
the
Doppler shift could cause a delay in arrival of the 18-15 kHz chirp signal and advance the arrival of
the 19–22 kHz chirp signal, the 𝑇𝐷𝑂𝐴 between the ith and the jth nodes was updated according to
Equation (8).
∆𝑓
∆𝑡 = ×𝑇 (7)
𝐹
Sensors 2019, 19, 2353 8 of 18

cause a delay in arrival of the 18–15 kHz chirp signal and advance the arrival of the 19–22 kHz chirp
signal, the TDOAij between the ith and the jth nodes was updated according to Equation (8).

∆f
∆t = × Tchirp (7)
F
 →



 ti = ti − ∆t i = 1, 3
→


= ti + ∆t i = 2, 4 (8)



 ti

 → →
TDOAij_new = t i − t j



Based on the above introduction, to achieve the correction of the TDOA value, calculation of the
Doppler shift was very critical. According to Equation (6), the Doppler shift was determined by v and
θ. Assuming the current position and the previous position were denoted as P̂ and Pprev , respectively,
the moving speed v of the target could be computed using Equation (9), where t0 was the time interval
between P̂ and Pprev .

 →
 P̂−Pprev
 v = t0

(9)
 v = k→


 vk
→
θ could be calculated using two vectors, as shown in Equation (11). The vector v is shown in
→
Equation (9) and the vector d propagation is shown in Equation (10), where Pi was the position of the ith
speaker node. According to Equations (7)–(11), Doppler disturbances were compensated by using the
location and the speed of the mobile target estimated from the pre-positioning result before correction.
However, large positioning errors caused by reverberation, outliers in the propagation times and
occlusions, etc., may have caused large speed estimation errors. Therefore, a corresponding strategy
should have been formulated to abandon the speed estimation results of obvious anomalies. Taking
the indoor navigation of human walking as an example, in normal circumstances, the walking speed of
a person is under 1.5 m/s. Therefore, 2 m/s could be set as a speed threshold. When the speed estimated
by Equation (9) was above 2 m/s, it was considered to be inappropriate for Doppler shift calculation.

→ P̂ − Pi
d propagation = (10)
kP̂ − Pi k

→→
 
 v · d propagation 
θ = acos

→
 (11)
→ 
k v k ∗ k d propagation k


2.6. Parameters for System Performance Evaluation

The localization error (LE) defined in Equation (12) and the cumulative distribution function
(CDF) of the LE erre used to evaluate the system positioning performance in static positioning
experiments. In Equation (12), x denoted the true position, and x̂ was the estimate of the true
position. In moving-target positioning experiments, the positioning error of each positioning, denoted
as the moving-target localization error (MTLE), was defined as the shortest distance between the
estimated positions and the track. The mean squared error (MSE) was used to evaluate the positioning
performance in moving-target positioning experiments, as shown in Equation (13), where N was the
number of estimated positions and xi was the estimated position.

LE = kx̂ − xk (12)

N
1 X
MSE = MTLE(xi ) (13)
N
i=1
 𝐿𝐸 = ‖𝑥 − 𝑥‖ (12)

1
𝑀𝑆𝐸 = 𝑀𝑇𝐿𝐸(𝑥 ) (13)
𝑁
Sensors 2019, 19, 2353 9 of 18

3. Results
3. Results
The TDOA-chirp-based system was placed in a hall, as shown in Figure 8. The four speakers
The TDOA-chirp-based system was placed in a hall, as shown in Figure 8. The four speakers
(Node 1 to Node 4) were placed at (0, 0), (0, 5.684), (10.971, 5.684), and (10.971, 0), respectively, to
(Node 1 to Node 4) were placed at (0, 0), (0, 5.684), (10.971, 5.684), and (10.971, 0), respectively, to form
form a positioning area with a size of 10.971 m × 5.684 m. The synchronous node was set to control
a positioning area with a size of 10.971 m × 5.684 m. The synchronous node was set to control the
the speakers to emit chirp signals according to the TDMA+FDMA or the TDMA transmission
speakers to emit chirp signals according to the TDMA+FDMA or the TDMA transmission schemes.
schemes. Considering the effect of temperature on the speed of sound, the positioning experiment
Considering the effect of temperature on the speed of sound, the positioning experiment was generally
was generally carried out at the same temperature, around 15 C. In addition, there was no other
carried out at the same temperature, around 15 ◦ C. In addition, there was no other equipment that
equipment that may have caused temperature inconsistencies in the experimental area. The
may have caused temperature inconsistencies in the experimental area. The smartphone collected
smartphone collected audio signals through the built-in microphone. At the same time of collecting
audio signals through the built-in microphone. At the same time of collecting the chirp signals,
the chirp signals, a lot of noise—including talking sounds, footsteps, and ringtones of other
a lot of noise—including talking sounds, footsteps, and ringtones of other smartphones—was also
smartphones—was also being collected. Since the noises were usually low-frequency noises, the
being collected. Since the noises were usually low-frequency noises, the received audio signal was
received audio signal was preprocessed first by a finite impulse response (FIR) band-pass filter (200-
preprocessed first by a finite impulse response (FIR) band-pass filter (200-order, 15–22 kHz) to reduce
order, 15–22 kHz) to reduce the influence of the noise. The SNR before filtering was 15 dB–20 dB,
the influence of the noise. The SNR before filtering was 15 dB–20 dB, while it was 45 dB–50 dB after
while it was 45 dB–50 dB after filtering. Considering the experimental environment, the value of α
filtering. Considering the experimental environment, the value of α in TDOA estimation was set as 0.2.
in TDOA estimation was set as 0.2.

Sensors 2019, 19, x FOR PEER REVIEW 10 of 18

Figure 8.
Figure 8. Photo
Photo of
of experimental
experimental setup.
setup.
3.1. Signal Preprocessing and TDOA Estimation
3.1. Signal Preprocessing and TDOA Estimation
An example of the audio signal collected by the smartphone during one positioning period is
shown Anin
example
Figure of
9. the
Afteraudio signalthe
filtering, collected
arrivalby the of
time smartphone
the signal during one positioning
transmitted periodnode
by each speaker is shown
was
in Figure 9. After filtering, the arrival time of the signal transmitted by each
estimated, as shown in Figure 10, in which the red point corresponded to the maximum of CC and speaker node was estimated,
as
theshown
greeninpoint
Figure
was10,the
in which
first pointthe red pointthan
greater corresponded
0.2 times theto the maximumThe
maximum. of CC and
time the green point
corresponding to
was the first point greater than 0.2 times the maximum. The time corresponding
the green point, which was denoted as 𝑡 , 𝑡 , 𝑡 , 𝑡 , respectively (𝑡 = 0.2527s, 𝑡 = 0.2638s, to the green point,
𝑡 =
which was
0.4842s, 𝑡 denoted as t1 ,was
= 0.4789s), t2 , t3 ,considered
t4 , respectively
as the( t1 = 0.2527
signal s, t2 =time.
arrival 0.2638 s, t3 = 0.4842
According s, t4 = 0.4789
to Equation s),
(2), the
was
TDOA considered
of anyas twothe signal
speaker arrival time. could
nodes According to Equation ((2),
be calculated the TDOA
𝑇𝐷𝑂𝐴 of any two
= −0.0111𝑠 speaker
, 𝑇𝐷𝑂𝐴 =
nodes could be calculated
−0.0315𝑠,𝑇𝐷𝑂𝐴 = −0.0262𝑠). (TDOA 12 = −0.0111 s, TDOA 13 = −0.0315 s, TDOA 14 = −0.0262 s).

Figure 9.
Figure 9. An
An example
example of
of the
the received
received audio
audio signal
signal during
during one
one positioning
positioning period.
period.
Sensors 2019, 19, 2353
Figure 9. An example of the received audio signal during one positioning period. 10 of 18

t1 t2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time(s) Time(s)

(a) (b)

t3 t4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time(s) Time(s)

(c) (d)
Figure 10.
Figure 10. Estimation
Estimation of
of arrival
arrival times
times for
for different nodes. (a)
different nodes. (a) Cross
Cross correlation (CC) of
correlation (CC) of Node
Node 1;
1; (b)
(b) CC
CC
of Node
of Node 2;
2; (c)
(c) CC
CC of
of Node
Node 3;
3; (d)
(d) CC
CC of
of Node
Node 4.4.

3.2. Static-Target Positioning

Positioning Experiments
Experiments under
under Different
Different Update
Update Rates
Rates
To demonstrate
demonstrate thethe positioning
positioning characteristics
characteristics ofof the
the proposed
proposed FDMA+TDMA
FDMA+TDMA scheme under
high update rate, static-target positioning experiments with three different positioning periods were
conducted for both TDMA+FDMA and TDMA-only schemes.
As shown
showninin Figure 4, the
Figure 4, specific transmission
the specific schemescheme
transmission of TDMA+FDMA
of TDMA+FDMAwas related to parameters
was related to
such as the period,
parameters such asthe signal
the duration,
period, the guard
the signal time, the
duration, andguard
the remaining
time, and time.
the As mentioned
remaining above,
time. As
the accuracy of TDOA estimation was related to the chirp signal TBP. In the TDMA+FDMA
mentioned above, the accuracy of TDOA estimation was related to the chirp signal TBP. In the transmission
scheme,
TDMA+FDMA the chirp signals had
transmission a narrow
scheme, the frequency
chirp signalsband.
had Hence,
a narrowin frequency
the process of shortening
band. Hence, in the
positioning period, thethe
process of shortening signal duration
positioning and the
period, theguard
signaltime should
duration andhave
theremained
guard time unchanged for
should have
effective TDOA estimation and signal separation. In other words, the positioning period could
be shortened only by shortening the remaining time. The specific parameter values of the three
TDMA+FDMA transmission schemes are described in Table 1, where Scheme 1, Scheme 2, and
Scheme 3 represent 1000 ms, 600 ms, and 450 ms positioning periods, respectively.

Table 1. Specific parameter values of three TDMA+FDMA transmission schemes.

Period (ms) Signal Duration (ms) Guard Time (ms) Remaining Time (ms)
Scheme 1 1000 50 150 600
Scheme 2 600 50 150 200
Scheme 3 450 50 150 50

In the TDMA-only transmission scheme, the frequency band of the chirp signal was about twice
that in the TDMA+FDMA transmission scheme. Hence, the positioning period could be shortened by
reducing the signal duration, the guard time, and the remaining time. Taking the 1000 ms positioning
period as an example, the TDMA-only transmission scheme is shown in Figure 11. Within the 1000 ms
positioning period, the speakers Node 1, Node 2, Node 3, and Node 4 emitted 50 ms 15–22 kHz chirp
signals in turn at 200 ms intervals. The specific parameter values of the three TDMA-only transmission
schemes are described in Table 2, where Scheme 1, Scheme 2, and Scheme 3 represent 1000 ms, 600 ms,
and 450 ms positioning periods, respectively.
reducing the signal duration, the guard time, and the remaining time. Taking the 1000 ms positioning
remained
period as anunchanged
example, for effective TDOA
the TDMA-only estimation
transmission and signal
scheme separation.
is shown in FigureIn11.other words,
Within the
the 1000
positioning period could be shortened only by shortening the remaining time. The specific
ms positioning period, the speakers Node 1, Node 2, Node 3, and Node 4 emitted 50 ms 15–22 kHz parameter
valuessignals
chirp of the three TDMA+FDMA
in turn transmission
at 200 ms intervals. schemesparameter
The specific are described in Table
values of the1,three
where Scheme 1,
TDMA-only
Scheme 2,
transmission
Sensors
and Scheme
2019, 19, 2353
3 represent 1000 ms, 600 ms, and 450 ms positioning periods, respectively.
schemes are described in Table 2, where Scheme 1, Scheme 2, and Scheme 3 represent 11 of 18
1000 In
ms,the
600 TDMA-only
ms, and 450transmission scheme,
ms positioning the frequency
periods, band of the chirp signal was about twice
respectively.
that in the TDMA+FDMA transmission scheme. Hence, the positioning period could be shortened by
reducing the signal duration, the guard time, and the remaining time. Taking the 1000 ms positioning
period as an example, the TDMA-only transmission scheme is shown in Figure 11. Within the 1000
ms positioning period, the speakers Node 1, Node 2, Node 3, and Node 4 emitted 50 ms 15–22 kHz
chirp signals in turn at 200 ms intervals. The specific parameter values of the three TDMA-only
transmission schemes are described in Table 2, where Scheme 1, Scheme 2, and Scheme 3 represent
1000 ms, 600 ms, and 450 ms positioning periods, respectively.

Figure 11. TDMA-only

Figure 11. TDMA-only transmission
transmission scheme.
scheme.

Table 2. Specific
In the positioning area, 40parameter values
test points of threeinTDMA-only
(marked transmission
red in Figure schemes.
12a) were selected. During the
experiment, the smartphone was placed on the test points to finish the positioning. Approximately
Period (ms) Signal Duration (ms) Guard Time (ms) Remaining Time (ms)
90 positioning results were collected at each point. Figure 12b gives the CDF of the LE under TDMA-
Scheme 1 1000 50 150 200
only transmission schemes, and Figure 12c shows the CDF of the LE under TDMA+FDMA
Scheme 2 600 20 100 120
transmission
Scheme 3schemes.450It is worth noting that
20 the x-axis units are70different in Figure 12b,c.
90
Figure 11. TDMA-only transmission scheme.

In the
In the positioning
positioning area,
area, 40
40 test
test points
points (marked
(marked in in red
red in
in Figure
Figure 12a)
12a) were
were selected.
selected. During
During the
the
experiment, the
experiment, the smartphone
smartphonewas wasplaced
placedon onthe
thetest
testpoints
pointstoto
finish thethe
finish positioning. Approximately
positioning. Approximately90
positioning results were collected at each point. Figure 12b gives the CDF of the LE under
90 positioning results were collected at each point. Figure 12b gives the CDF of the LE under TDMA- TDMA-only
transmission
only schemes,
transmission and Figure
schemes, and 12c shows
Figure 12ctheshows
CDF of the LE under
theCDF of theTDMA+FDMA transmission
LE under TDMA+FDMA
schemes. It isschemes.
transmission worth noting that the
It is worth x-axisthat
noting units
theare different
x-axis unitsin Figure
are 12b,c.
different in Figure 12b,c.

(a) (b) (c)

Figure 12. Results of static-target positioning experiment. (a) Test points; (b) TDMA-only scheme; (c)
TDMA+FDMA scheme.

Table 1. Specific parameter values of three TDMA+FDMA transmission schemes.

Period (ms) Signal Duration (ms) Guard Time (ms) Remaining Time (ms)
Scheme 1 1000
(a) 50 (b) 150 (c) 600
Scheme 2 600 50 150 200
Figure
Figure3 12.
12. Results
Results of static-target positioning experiment. (a)(a)
Test points; (b) TDMA-only scheme; (c)
Scheme 450of static-target positioning
50 experiment. Test
150 points; (b) TDMA-only50scheme;
TDMA+FDMA
(c) scheme.
TDMA+FDMA scheme.
Table 2. Specific parameter values of three TDMA-only transmission schemes
As shown Table 1. Specific
in Figure 12b, parameter values performance
the positioning of three TDMA+FDMA transmission
decreased schemes.
upon shortening the positioning
period in the TDMA-only
Period (ms) scheme. In the
Signal 1000 ms(ms)
Duration positioning
Guardperiod,
Time96%(ms)of the LE was below
Remaining Time20(ms)
cm,
Period (ms) Signal Duration (ms) Guard Time (ms) Remaining Time (ms)
and 99% was
Scheme 1 below1000 50 cm. In the 600 ms50positioning period, the 150positioning accuracy200 was slightly
Scheme 1 1000 50 150 600
reduced;
Scheme 93%2 of the600 LE was below 20 cm,20and 99% was below 50 100cm. When the positioning
120 period
Scheme 2 600 50 150 200
was shortened to 450 ms, the positioning performance of the system was poor; only about 30% of the
Scheme 3 450 50 150 50
LE was less than 20 cm, and only about 33% was less than 50 cm. However, as shown in Figure 12c,
in the TDMA+FDMA transmission scheme, the positioning performance was almost the same under
Table 2. Specific parameter values of three TDMA-only transmission schemes
the three different positioning periods; about 90% of the LE was less than 20 cm, and about 99% was
less than 50 cm.Period
Hence,(ms) Signal Duration
the TDMA+FDMA (ms) Guard
transmission Time
scheme (ms)
was moreRemaining
powerful inTime (ms)
realizing
Scheme 1 1000 50
centimeter-level accuracy positioning under a high update rate. 150 200
Scheme 2 600 20 100 120
3.3. Moving-Target Positioning Experiments
To demonstrate the feasibility and the validity of the proposed positioning correction scheme,
moving-target positioning experiments were conducted using the TDMA+FDMA transmission scheme
with a 1000 ms positioning period. Three tracks were planned in the positioning area. A user with
less than 50 cm. Hence, the TDMA+FDMA transmission scheme was more powerful in realizing
centimeter-level accuracy positioning under a high update rate.

3.3. Moving-Target Positioning Experiments

SensorsTo demonstrate
2019, 19, 2353 the feasibility and the validity of the proposed positioning correction scheme,
12 of 18
moving-target positioning experiments were conducted using the TDMA+FDMA transmission
scheme with a 1000 ms positioning period. Three tracks were planned in the positioning area. A user
the smartphone, as shown in Figure 13, moved on each track according to the direction specified
with the smartphone, as shown in Figure 13, moved on each track according to the direction specified
by the black arrow at three different speeds, namely, slow-speed, medium-speed, and high-speed.
by the black arrow at three different speeds, namely, slow-speed, medium-speed, and high-speed.
The approximate ranges were 0.4–0.5 m/s for slow-speed, 0.6–0.7 m/s for medium-speed, and 0.8–0.9 m/s
The approximate ranges were 0.4–0.5 m/s for slow-speed, 0.6–0.7 m/s for medium-speed, and 0.8–0.9
for high-speed. The user walked five laps on each track at each speed.
m/s for high-speed. The user walked five laps on each track at each speed.

Figure13.
Figure 13. Diagram
Diagram of
of smartphone’s
smartphone’sposition
positionduring
duringexperiment.
experiment.

Withthe
With themeasured
measuredTDOA TDOAdata, data,twotwokindskindsof ofpositioning
positioningcorrection
correctionexperiments
experiments werewerecarried
carried
out. In the first experiment, the Doppler shifts were calculated at different fixed
out. In the first experiment, the Doppler shifts were calculated at different fixed speeds but not at speeds but not atthe
the
estimated speed.
estimated speed. In In other
other words,
words, the
the vv in in Equation
Equation (6) (6) was
was not
not calculated
calculated according
according to to Equation
Equation (9)(9)
but was
but was set
set to
tofixed
fixedvalues
valuesin inthe
therange
rangeofof0–1.20-1.2 m/sm/satatintervals
intervalsofof0.1
0.1m/s.
m/s. As
As the
the target
target moving
moving
directionswere
directions weredetermined
determinedin in advance
advance for each track, the θ𝜃 atateach
for each eachpositioning
positioningpoint
pointwaswascalculated
calculated
accordingto
according toEquation
Equation(11)(11)directly.
directly.Figure
Figure14 14gives
givesthe therelationship
relationshipbetween
betweenthethepositioning
positioningMSE MSEand
and
v value for the three tracks and the three different speeds. When v = 0, the MSE was the average error
v value for the three tracks and the three different speeds. When v = 0, the MSE was the average error
withoutpositioning
without positioningcorrection.
correction.ItItwas
wasfound
foundthat thatthethegreater
greaterthe
thetarget
targetmoving
movingspeed
speedwas,
was,thethe greater
greater
theMSE
the MSEbefore
beforepositioning
positioning correction
correction was.
was. In In all
all cases,
cases, the
the MSE
MSE decreased
decreased first
firstand
andthen
thenincreased
increased
asvvincreased,
as increased,and andthe
thevvvalue
valuecorresponding
correspondingto tothetheminimum
minimumMSE MSEapproximated
approximatedthe theactual
actualtarget
target
movingspeed.
moving speed. Generally,
Generally, thethe optimal
optimal value
value of of vv for
for slow-speed
slow-speed was was near 0.5 m/s, except for Track Track 1,
1,
andthe
and theoptimal
optimalvalues
values of
of vv for medium-speed
medium-speed and high-speed were near 0.6 and 0.8 m/s, m/s, respectively.
respectively.
Thisexperimental
This experimentalresultresultverifies
verifiesthat
thatititisispossible
possibleto toobtain
obtaingood
goodpositioning
positioningcorrection
correctionperformance
performance
whenthe
when thevvisisestimated
estimatedfromfromthe theactual
actualspeedspeedofofthethemoving
movingtarget.
target.
In the second positioning experiment, the Doppler shifts were calculated according to Equations
(6)–(11) where 2 m/s was set as the speed threshold. When the speed estimated by Equation (9) was
above 2 m/s, no TDOA update and positioning correction were made. Figure 15 demonstrates the
positioning results for the three tracks under the three different speeds. The red points are the positions
before correction and the green points represent positions after corrections. Besides, the blue lines
are real tracks. For the sake of clear display, the results with positioning error above 50 cm were not
presented in Figure 15. From Figure 15, it can be found that positioning results without correction were
always on one side of the real track due to the influence of the Doppler shift for all the three tracks.
After correction, the positioning results were closer to the real track and distributed on both sides of
the real track. As a supplement, Table 3 gives the percentage of the abnormal positioning with errors
greater than 50 cm. As can be seen from Table 3, TDOA correction based on Doppler shift could greatly
reduce the proportion of results with positioning errors greater than 50 cm. Table 4 summarizes the
MSE for all cases. The MSE values before correction are given on the left side, and those after correction
are on the right side in parentheses. The data in Table 4 show that, after correction, the positioning
errors were reduced by about 10 cm.
Sensors
Sensors 2019,
2019, 19,
19, x2353
FOR PEER REVIEW 1313of
of 18
18

0.35 0.45 0.4

0.3 0.35
MSE(m)

0.4

MSE(m)
0.3

MSE(m)
0.25 0.35
0.25
0.2 0.3 0.2
0.15 0.25 0.15
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
v(m/s) v(m/s) v(m/s)

(a) (b) (c)

0.26 0.28 0.3

0.24 0.26

MSE(m)
0.25
MSE(m)
0.24
MSE(m)

0.22
0.22
0.2 0.2
0.2
0.18 0.18 0.15
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
v(m/s) v(m/s) v(m/s)

(d) (e) (f)

0.24 0.3 0.35

0.22 0.3
0.25
MSE(m)

MSE(m)

0.2 0.25

MSE(m)
0.2
0.18 0.2
0.16 0.15 0.15
0.14 0.1 0.1
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
v(m/s) v(m/s) v(m/ s)

(g) (h) (i)

Figure
Figure 14.
14.Relationship
Relationshipbetween
betweenpositioning
positioningmean squared
mean error
squared (MSE)
error and and
(MSE) v values. (a) Track
v values. (a) 1 slow-1
Track
speed; (b) Track 1 medium-speed; (c) Track 1 high-speed; (d) Track 2 slow-speed; (e) Track 2
slow-speed; (b) Track 1 medium-speed; (c) Track 1 high-speed; (d) Track 2 slow-speed; (e) Track 2medium-
speed; (f) Track 2(f)high-speed;
medium-speed; (g) Track(g)
Track 2 high-speed; 3 slow-speed; (h) Track(h)
Track 3 slow-speed; 3 medium-speed; (i) Track(i)
Track 3 medium-speed; 3 high-
Track
speed.
3 high-speed.

The percentage
Table 3. positioning
In the second (%) of the abnormal
experiment, positioning
the Doppler with calculated
shifts were errors greater than 50 cm.
according to Equations
(6)–(11) where 2 m/s was set as the speed threshold.
Track 1 (before/after)
When the speed
Track 2 (before/after)
estimated by Equation
Track 3 (before/after)
(9) was
above 2 m/s, no TDOA update and positioning correction were made. Figure 15 demonstrates the
Slow-speed 15.20/7.35 7.17/5.02 3.77/3.14
positioning results for the three tracks under the three different speeds. The red points are the
Medium-speed 29.11/14.56 9.44/6.67 7.14/0
positions High-speed
before correction and 25.87/4.90
the green points represent positions after corrections.
10.14/4.05 17.17/0 Besides, the
blue lines are real tracks. For the sake of clear display, the results with positioning error above 50 cm
were not presented in Table
Figure 15. From
4. MSE Figure
(cm) of 15, it can
moving-target be foundexperiments.
positioning that positioning results without
correction were always on one side of the real track due to the influence of the Doppler shift for all
Track 1the
the three tracks. After correction, (before/after) Track 2were
positioning results (before/after)
closer to theTrack 3 (before/after)
real track and distributed
on both sides of the real track.32.72/20.02
Slow-speed As a supplement, Table 3 gives the percentage
23.99/18.80 of the abnormal
23.21/15.02
Medium-speed
positioning 40.43/27.03
with errors greater than 26.62/18.49
50 cm. As can be seen from Table 3, TDOA 26.02/12.03
correction based on
High-speed 38.76/18.21 29.52/16.57
Doppler shift could greatly reduce the proportion of results with positioning errors 30.37/13.85
greater than 50
cm. Table 4 summarizes the MSE for all cases. The MSE values before correction are given on the left
side, and those after correction are on the right side in parentheses. The data in Table 4 show that,
after correction, the positioning errors were reduced by about 10 cm.

Table 3. The percentage (%) of the abnormal positioning with errors greater than 50 cm.

Track 1 Track 2 Track 3

(before/after) (before/after) (before/after)
Slow-speed 15.20/7.35 7.17/5.02 3.77/3.14
Medium-speed 29.11/14.56 9.44/6.67 7.14/0
High-speed 25.87/4.90 10.14/4.05 17.17/0
 Track 1 Track 2 Track 3
(before/after) (before/after) (before/after)
Slow-speed 32.72/20.02 23.99/18.80 23.21/15.02
Medium-speed
Sensors 2019, 19, 2353 40.43/27.03 26.62/18.49 26.02/12.03 14 of 18
High-speed 38.76/18.21 29.52/16.57 30.37/13.85

6 6 6
5 5 5
4 4 4

y(m)
y(m)

3 3

y(m)
3
2 2 2
1 1 1
0 0 0
-1 -1 -1
0 1 2 3 4 5 6 7 8 9 10 11 0 1 2 3 4 5 6 7 8 9 10 11 0 1 2 3 4 5 6 7 8 9 10 11
x(m) x(m) x(m)

(a) (b) (c)

6 6 6
5 5 5
4 4 4
y(m)

3 3
y(m)

y(m)
3
2 2 2
1 1 1
0 0 0
-1 -1 -1
0 1 2 3 4 5 6 7 8 9 10 11 0 1 2 3 4 5 6 7 8 9 10 11 0 1 2 3 4 5 6 7 8 9 10 11
x(m) x(m) x(m)

(d) (e) (f)

6 6 6
5 5 5
4 4 4
3
y(m)

y(m)

3 3
y(m)
2 2 2
1 1 1
0 0 0
-1 -1 -1
0 1 2 3 4 5 6 7 8 9 10 11 0 1 2 3 4 5 6 7 8 9 10 11 0 1 2 3 4 5 6 7 8 9 10 11
x(m) x(m) x(m)

(g) (h) (i)

Figure
Figure 15.
15. Results
Results ofof moving-target
moving-target positioning
positioning experiments.
experiments. (a)(a)Track
Track11slow-speed;
slow-speed;(b)
(b)Track
Track 11
medium-speed; (c) Track 1 high-speed; (d) Track 2 slow-speed; (e) Track 2 medium-speed;
medium-speed; (c) Track 1 high-speed; (d) Track 2 slow-speed; (e) Track 2 medium-speed; (f) Track(f) Track2
2high-speed;
high-speed;(g)
(g)Track
Track3 3slow-speed;
slow-speed;(h)
(h)Track
Track 3 medium-speed;
3 medium-speed; (i)(i) Track
Track 3 high-speed.
3 high-speed.

4. Discussion
4. Discussion

4.1. Feasibility
4.1. Feasibilityand
andSuperiority
Superiorityof
of TDMA+FDMA
TDMA+FDMATransmission
TransmissionScheme
Schemein
inImproving
Improving System
System Update
Update Rate
Rate
The TDOA-chirp-based
The TDOA-chirp-based indoor indoor positioning
positioning systems
systems used used inin related
related studies
studies generally
generally adopted
adopted
the TDMA-only
the TDMA-only transmission
transmission scheme scheme to to control
control chirp
chirp signals.
signals. For Forinstance,
instance,Sérgio
SérgioI.I.Lopes
Lopesetetal.al.
designed a system that used four speakers to transmit chirp signals of
designed a system that used four speakers to transmit chirp signals of 18–22 kHz frequency under 18–22 kHz frequency under
theTDMA-only
the TDMA-onlyscheme scheme[5]. [5].InIn their
their research,
research, thethe positioning
positioning period,
period, defined
defined as the
as the timetime between
between the
the signal transmission and the position estimation, had a mean value
signal transmission and the position estimation, had a mean value of 350 ms, and 96% of the LE was of 350 ms, and 96% of the
LE was below 20 cm in a positioning area of 9 m × 8 m. In this study, the
below 20 cm in a positioning area of 9 m × 8 m. In this study, the positioning performance decreased positioning performance
decreased
upon upon shortening
shortening the positioningthe positioning
period in the period in the TDMA-only
TDMA-only scheme. When scheme. When the period
the positioning positioning
was
period was 1000 ms, 96% of the LE was below 20 cm, and 99% was
1000 ms, 96% of the LE was below 20 cm, and 99% was below 50 cm. Considering the positioning below 50 cm. Considering the
accuracy and the update rate, it seems that our positioning accuracy is inferior to that in Sérgio in
positioning accuracy and the update rate, it seems that our positioning accuracy is inferior to that I.
Sérgioet
Lopes I. al.’s
Lopes et al.’s research.
research. However,However, the different
the different experimental
experimental environmentsenvironments
make themake theofresults
results these
of these
two twoincomparable.
studies studies incomparable. The experiments
The experiments in Lopes inet Lopes et al.’s research
al.’s research wereout
were carried carried
in a out in a
regular
regular room with concrete walls and ceiling, windows, linoleum floor, and
room with concrete walls and ceiling, windows, linoleum floor, and different types of office furniture different types of office
furniture providing relatively less reflection. However, the experiment environment of our system was
in the lobby of an office building with a great deal of noise and significant reflection. As mentioned
in the introduction, the detection performance of chirp signals is influenced by the TBP, which is
defined as the product of the duration of the signal and the bandwidth. Generally, the larger the TBP
is, the higher the time resolution and the accuracy of chirp signal detection will be. In the research of
Sensors 2019, 19, 2353 15 of 18

Lopes et al., a 10 ms signal duration, 4 kHz bandwidth, and 40 kHz·ms TBP was used for localization.
In our experiments, the minimum positioning period was 450 ms, in which the signal duration time,
the bandwidth, and the TBP were 20 ms, 7 kHz, and 140 kHz·ms, respectively, and about 30% of
the LE was less than 20 cm, and 33% was less than 50 cm. The results show that the experimental
environment in our study puts forward higher requirements for TBP. When the positioning period is
less than 450 ms, the positioning accuracy cannot be guaranteed.
Although the combination of TDMA and FDMA has been widely used in related fields, such as
wireless networks [36], microcellular systems [37], underwater acoustic networks [38], and mobile
radio systems [39], it is the first attempt to realize the scheme in indoor localization systems. In FDMA,
the 15–22 kHz frequency band was divided into two sub-bands: 18–15 kHz and 19–22 kHz. In TDMA,
four nodes were divided into two groups to transmit the two sub-band chirp signals. The results of the
static-target positioning experiments showed that the TDMA+FDMA transmission scheme has more
advantages in improving the update rate than the TDMA-only scheme. The positioning performances
were almost the same under three different positioning periods, and about 90% of the LE was less than
20 cm, and 99% was less than 50 cm, even for the 450 ms positioning period. For the 450 ms positioning
period, the signal duration time, the guard time, the remaining time, and the TBP were 50 ms, 150 ms,
50 ms, and 150 kHz·ms, respectively. Thus, the TDMA+FDMA scheme had a higher TBP value than
the TDMA-only scheme, which may have been one of the reasons why the former exhibited higher
positioning accuracy.
However, although the TDMA+FDMA scheme was verified to be an effective solution to improve
the system update rate, as a preliminary work, this study failed to show the extent to which this scheme
can shorten the positioning period. Since the frequency band of chirp signals in the TDMA+FDMA
scheme was less than half of that in the TDMA-only scheme, the signal duration time was set to
50 ms to ensure the TBP value, and a 150 ms guard time was adopted. In future work, the optimal
combination of the signal duration time, the protection time, and the remaining time will be explored
to improve the system update rate further.

4.2. Feasibility and Superiority of Target-Moving-Speed-Based TDOA Correction Method in Improving

Positioning Accuracy
It is generally known that, for a moving target, the signal received exhibits Doppler shifts, which
can influence the performance of TDOA estimates [10,40]. The research on Doppler shift has mainly
focused on outdoor positioning systems. For instance, Slamet Widodo et al. designed a GPS-based
positioning system that used Doppler shift compensation for moving target positioning [41]. In their
research, the Doppler shift was estimated by detecting the maximum value of the power spectrum and
was then used to re-generate a new referred signal to estimate the arrival time. The results showed
that, when the target moved in a circular path with an angular velocity of 0–1.3 rad/s, the positioning
error before compensation was about 50 mm at 0.2 rad/s and 120 mm at 0.4 rad/s. When Doppler shift
compensation was used, the target could be localized to within 25 mm of the actual position. However,
there are very few studies on reducing the effect of the Doppler shift on the positioning accuracy of
indoor positioning systems.
To the best of our knowledge, this study is the first to deal with Doppler frequency shift in acoustic
indoor positioning systems. In particular, a target-moving-speed-based TDOA correction method was
used to improve the positioning accuracy of a TDOA-chirp-based indoor positioning system. The basic
idea of this method is to convert the Doppler shifts in the frequency domain into time differences
to update the TDOA values. The results of moving-target positioning experiments verified that it is
possible to obtain correction performance when the v is estimated from the actual speed of the moving
target, and the positioning error was reduced by about 10 cm when the proposed method was adopted.
In other words, the proposed target-moving-speed-based TDOA correction method can be considered
as an effective method to reduce the positioning error caused by the Doppler shift.
Sensors 2019, 19, 2353 16 of 18

However, although its effectiveness was verified, the proposed correction method is still very
preliminary, and there are at least two problems that should be paid attention to and solved to improve
its performance in the future.
Firstly, when the Doppler shift was calculated using Equation (6), the speed v of the moving
target and the angle θ between the directions of target motion and audio signal propagation were
estimated based on the previous position and the current position. Taking the 1000 ms positioning
period and the speed of 0.6–0.7 m/s as an example, the position difference between the two positionings
was about 60–70 cm. As shown in Table 4, the positioning errors ranged from 30–40 cm. These large
positioning errors may have caused relatively large speed and direction estimation errors, thereby
affecting the accuracy of estimation of the Doppler frequency shift. Therefore, more effective methods
for the accurate estimation of Doppler frequency shifts caused by target motion should be explored.
Accelerometers and gyroscopes can be considered for accurate detection of the target moving speed
and direction.
Secondly, the Doppler shift ∆f was converted into the time difference ∆t using Equation (7) based
on the assumption that the linear chirp signal with Doppler shift was still a linear chirp signal, and
the TDOA correction based on Doppler shifts was carried out according to Equation (8). However,
the assumption that the received signal was still a linear chirp signal was made for the convenience
of calculation. In fact, it is not very rigorous. Usage of the Doppler frequency shift to achieve
more accurate TDOA correction is of great significance to improving the positioning accuracy for
moving targets.

5. Conclusions
Taking an indoor positioning system based on audio chirp signals and TDOA as the research
target, we proposed a TDMA+FDMA signal transmission scheme to improve the positioning update
rate and a target-moving-speed-based TDOA correction method to reduce the positioning error caused
by Doppler frequency shift. The feasibility and the superiority of the proposed methods were verified
in static-target positioning and moving-target positioning experiments. This research provides a
possible framework for the realization of a TDOA-chirp-based audio indoor positioning system with a
high positioning accuracy and update rate.

Author Contributions: X.C. and Y.C. designed the correction algorithms, conducted the experiment, collected and
analyzed the data, and wrote the draft of the manuscript. S.C. built the equipment of the indoor localization system.
L.Z. offered help to complete experiment. X.Z. and X.C. gave guidance on the presentation of the localization
results. The submitted manuscript was approved by all authors.
Funding: This study was supported by the National Key Research and Development Program of China
(No. 2016YFB0502202).
Acknowledgments: We would like to thank Songyu Cong, Chi Wu for their help during the experiment. We also
appreciate the support of the other fellows in our lab.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Liu, K.; Liu, X.; Xie, L.; Li, X. Towards accurate acoustic localization on a smartphone. In Proceedings of the
2013 Proceedings IEEE INFOCOM, Turin, Italy, 14–19 April 2013; pp. 495–499.
2. Alarifi, A.; Al-Salman, A.; Alsaleh, M.; Alnafessah, A.; Al-Hadhrami, S.; Al-Ammar, M.; Al-Khalifa, H. Ultra
wideband indoor positioning technologies: Analysis and recent advances. Sensors 2016, 16, 707. [CrossRef]
3. Kim, H.-S.; Choi, J.-S. Advanced indoor localization using ultrasonic sensor and digital compass.
In Proceedings of the 2008 International Conference on Control, Automation and Systems, Seoul, Korea,
14–17 October 2008; pp. 223–226.
4. Kim, S.J.; Kim, B.K. Dynamic ultrasonic hybrid localization system for indoor mobile robots. IEEE Trans.
Ind. Electron. 2013, 60, 4562–4573. [CrossRef]
Sensors 2019, 19, 2353 17 of 18

5. Saad, M.M.; Bleakley, C.J.; Ballal, T.; Dobson, S. High-accuracy reference-free ultrasonic location estimation.
IEEE Trans. Instrum. Meas. 2012, 61, 1561–1570. [CrossRef]
6. Carotenuto, R.; Merenda, M.; Iero, D.; Della Corte, F.G. Ranging RFID tags with ultrasound. IEEE Sens. J.
2018, 18, 2967–2975. [CrossRef]
7. Díaz, E.; Pérez, M.; Gualda, D.; Villadangos, J.; Ureña, J.; García, J. Ultrasonic indoor positioning for smart
environments: A mobile application. In Proceedings of the 2017 4th Experiment@ International Conference
(exp. at’17), Algarve, Portugal, 6–8 June 2017; pp. 280–285.
8. Kim, Y.; Shin, H.; Chon, Y.; Cha, H. Smartphone-based Wi-Fi tracking system exploiting the RSS peak to
overcome the RSS variance problem. Pervasive Mob. Comput. 2013, 9, 406–420. [CrossRef]
9. Höflinger, F.; Zhang, R.; Hoppe, J.; Bannoura, A.; Reindl, L.M.; Wendeberg, J.; Bührer, M.; Schindelhauer, C.
Acoustic self-calibrating system for indoor smartphone tracking (assist). In Proceedings of the 2012
International Conference on Indoor Positioning and Indoor Navigation (IPIN), Sydney, Australia,
13–15 November 2012; pp. 1–9.
10. Lopes, S.I.; Vieira, J.M.; Reis, J.; Albuquerque, D.; Carvalho, N.B. Accurate smartphone indoor positioning
using a WSN infrastructure and non-invasive audio for TDoA estimation. Pervasive Mob. Comput. 2015,
20, 29–46. [CrossRef]
11. Peng, C.; Shen, G.; Zhang, Y. BeepBeep: A high-accuracy acoustic-based system for ranging and localization
using COTS devices. ACM Trans. Embedded Comput. Syst. (TECS) 2012, 11, 4. [CrossRef]
12. Álvarez, F.J.; Aguilera, T.; López-Valcarce, R. CDMA-based acoustic local positioning system for portable
devices with multipath cancellation. Digit. Signal Process. 2017, 62, 38–51. [CrossRef]
13. Sertatıl, C.; Altınkaya, M.A.; Raoof, K. A novel acoustic indoor localization system employing CDMA.
Digit. Signal Process. 2012, 22, 506–517. [CrossRef]
14. Liu, K.; Liu, X.; Li, X. Guoguo: Enabling fine-grained smartphone localization via acoustic anchors. IEEE Trans.
Mob. Comput. 2016, 15, 1144–1156. [CrossRef]
15. Tan, C.; Zhu, X.; Su, Y.; Wang, Y.; Wu, Z.; Gu, D. A low-cost centimeter-level acoustic localization system
without time synchronization. Measurement 2016, 78, 73–82. [CrossRef]
16. Höflinger, F.; Bordoy, J.; Simon, N.; Wendeberg, J.; Reindl, L.M.; Schindelhauer, C. Indoor-localization system
for smart phones. In Proceedings of the 2015 IEEE International Workshop on Measurements & Networking
(M&N), Coimbra, Portugal, 12–13 October 2015; pp. 1–6.
17. Skolnik, M.I. Radar Handbook; McGraw-Hill Education: New York, NY, USA, 1970.
18. Sathyan, T.; Sinha, A.; Kirubarajan, T. Passive geolocation and tracking of an unknown number of emitters.
IEEE Trans. Aerosp. Electron. Syst. 2006, 42, 740–750. [CrossRef]
19. Ran, Q.; Feng, R.; Yu, N.; Wu, Y. A weighted least squares source localization algorithm using TDOA
measurements in wireless sensor networks. In Proceedings of the 2016 6th International Conference on
Electronics Information and Emergency Communication (ICEIEC), Beijing, China, 17–19 June 2016; pp. 10–13.
20. Qu, X.; Xie, L.; Tan, W. Iterative constrained weighted least squares source localization using TDOA and
FDOA measurements. IEEE Trans. Signal Process. 2017, 65, 3990–4003. [CrossRef]
21. Fowler, M.L.; Hu, X. Signal models for TDOA/FDOA estimation. IEEE Trans. Aerosp. Electron. Syst. 2008,
44, 1543–1550. [CrossRef]
22. Yeredor, A.; Angel, E. Joint TDOA and FDOA estimation: A conditional bound and its use for optimally
weighted localization. IEEE Trans. Signal Process. 2011, 59, 1612–1623. [CrossRef]
23. Ho, K.; Xu, W. An accurate algebraic solution for moving source location using TDOA and FDOA
measurements. IEEE Trans. Signal Process. 2004, 52, 2453–2463. [CrossRef]
24. Zhu, M.; Yao, H.; Wu, X.; Lu, Z.; Zhu, X.; Huang, Q. Gaussian filter for TDOA based sound source localization
in multimedia surveillance. Multimedia Tools Appl. 2018, 77, 3369–3385. [CrossRef]
25. Wu, R.; Zhang, Y.; Huang, Y.; Xiong, J.; Deng, Z. A Novel Long-Time Accumulation Method for
Double-Satellite TDOA/FDOA Interference Localization. Radio Sci. 2018, 53, 129–142. [CrossRef]
26. Bull, J.F.; Ward, M.L. Interference Detection, Characterization and Location in a Wireless Communications or
Broadcast System. Google Patents US9121926B2, 1 September 2015.
27. Zhai, X.; Yang, J.; Cui, L. Wireless Network Localization via Alternating Projections with TDOA and FDOA
Measurements. Ad Hoc Sens. Wirel. Netw. 2017, 38, 1–20.
28. Luo, M.; Chen, X.; Cao, S.; Zhang, X. Two new shrinking-circle methods for source localization based on
TDOA measurements. Sensors 2018, 18, 1274. [CrossRef]
Sensors 2019, 19, 2353 18 of 18

29. Ens, A.; Höflinger, F.; Wendeberg, J.; Hoppe, J.; Zhang, R.; Bannoura, A.; Reindl, L.M.; Schindelhauer, C.
Acoustic self-calibrating system for indoor smart phone tracking. Int. J. Navig. Obs. 2015, 2015, 1–9.
[CrossRef]
30. Chan, Y.-T.; Hang, H.Y.C.; Ching, P.-c. Exact and approximate maximum likelihood localization algorithms.
IEEE Trans. Veh. Technol. 2006, 55, 10–16. [CrossRef]
31. Friedlander, B. A passive localization algorithm and its accuracy analysis. IEEE J. Ocean. Eng. 1987, 12, 234–245.
[CrossRef]
32. Chan, Y.-T.; Ho, K. A simple and efficient estimator for hyperbolic location. IEEE Trans. Signal Process. 1994,
42, 1905–1915. [CrossRef]
33. Cheung, K.W.; So, H.-C.; Ma, W.-K.; Chan, Y.-T. A constrained least squares approach to mobile positioning:
algorithms and optimality. EURASIP J. Adv. Signal Process. 2006, 2006, 020858. [CrossRef]
34. Lin, L.; So, H.-C.; Chan, F.K.; Chan, Y.-T.; Ho, K. A new constrained weighted least squares algorithm for
TDOA-based localization. Signal Process. 2013, 93, 2872–2878. [CrossRef]
35. Khalid, G.A. The Effect of Doppler Phenomenon on the Speed of Blood Flow. Al-Khwarizmi Eng. J. 2012,
8, 26–39.
36. Salajegheh, M.; Soroush, H.; Kalis, A. HYMAC: Hybrid TDMA/FDMA medium access control protocol
for wireless sensor networks. In Proceedings of the 2007 IEEE 18th International Symposium on Personal,
Indoor and Mobile Radio Communications, Athens, Greece, 3–7 September 2007; pp. 1–5.
37. Akaiwa, Y.; Andoh, H. Channel segregation-a self-organized dynamic channel allocation method: Application
to TDMA/FDMA microcellular system. IEEE J. Sel. Areas Commun. 1993, 11, 949–954. [CrossRef]
38. Casari, P.; Marella, S.; Zorzi, M. A comparison of multiple access techniques in clustered underwater acoustic
networks. In Proceedings of the OCEANS 2007-Europe, Aberdeen, Scotland, 18–21 June 2007; pp. 1–6.
39. Arslan, H.; Gupta, S.; Bottomley, G.; Chennakeshu, S. Successive cancellation of adjacent channel signals in
FDMA/TDMA digital mobile radio systems. In Proceedings of the VTC’98. 48th IEEE Vehicular Technology
Conference. Pathway to Global Wireless Revolution (Cat. No. 98CH36151), Ottawa, ON, Canada, 18–21
May 1998; pp. 1720–1724.
40. Miao, F.; Yang, D.; Wang, R.; Wen, J.; Wang, Z.; Lian, X. A moving sound source localization method based
on TDOA. In Proceedings of the INTER-NOISE and NOISE-CON Congress and Conference Proceedings,
Melbourne, Australia, 16–19 November 2014; pp. 4159–4165.
41. Widodo, S.; Shiigi, T.; Hayashi, N.; Kikuchi, H.; Yanagida, K.; Nakatsuchi, Y.; Ogawa, Y.; Kondo, N. Moving
object localization using sound-based positioning system with doppler shift compensation. Robotics 2013,
2, 36–53. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).