LoRa Performance Analysis for LEO Satellite IoT

Networks
1st Wendong Zhou 2nd Tao Hong
2021 13th International Conference on Wireless Communications and Signal Processing (WCSP) | 978-1-6654-0785-4/21/$31.00 ©2021 IEEE | DOI: 10.1109/WCSP52459.2021.9613680

College of Telecommunications and Information Engineering College of Telecommunications and Information Engineering
Nanjing University of Posts and Telecommunications Nanjing University of Posts and Telecommunications
Nanjing, Jiangsu, China Nanjing, Jiangsu, China
[email protected] [email protected]

3rd Xiaojin Ding 4th Gengxin Zhang

School of Internet of Things College of Telecommunications and Information Engineering
Nanjing University of Posts and Telecommunications Nanjing University of Posts and Telecommunications
Nanjing, Jiangsu, China Nanjing, Jiangsu, China
[email protected] [email protected]

Abstract—Due to the wide coverage of satellite communication between symbols and proposed two algorithms to implement
systems, the Satellite Internet of Things (S-IoT) has become a this technology. Laport-Fauret et al. In [7] and Xia et al. In [8]
new hot spot in the field of Internet of Things. Long Range proposed two methods to decode superimposed LoRa signals
(LoRa) signal is one of the main directions of terrestrial Internet
of Things because of the low power wide area network (LPWAN) using the preamble and chirp characteristics of the LoRa
characteristic. In this paper, we present a low earth orbit (LEO) signalrespectively. However, these aforementioned works were
IoT model, based on the capture effect and successive interference performed under the scenario of terrestrial networks.
cancellation (SIC) demodulation technology, to analyze the uplink IoT services based on terrestrial base stations are usually
performance of the LoRa. Furthermore, we derive a connec- affected by geographic environment and natural disasters (such
tion probability equation for LoRa signal from the viewpoint
of received signal strength indicator (RSSI) in Rice channel. as oceans, desert, and earthquake disasters), because terrestrial
Simulation results can be used to guide the Initial transmission base stations in these remote areas are difficult to build and
power setting for satisfying the average access success probability maintain. To solve this problem, scholars began to study
(ASP) requirement of LEO satellite IoT networks. the application potential of Long Range Wide Area Network
Index Terms—Internet of Things, satellite IoT, LoRa, Uplink (LoRaWAN) in low earth orbit (LEO) satellites to supplement
Performance, successive interference cancellation.
terrestrial IoT [9]. In [10], Qian presented a symmetric chirp
I. I NTRODUCTION signal with better cross-correlation performance to overcome
high dynamic characteristic in LEO satellite-based IoT. In
In recent years, the application of the Internet of Things
[11], Yang proposed a folded chirp frequency shift keying
(IoT) has developed rapidly in different fields, such as smart
(FCrSK) modulation with strong immunity to doppler frequen-
buildings, personal medical care, environmental monitoring,
cy shift (DFS).
and logistics transportation [1]. To achieve large scale connec-
Motivated by the LoRa signal performance in the terrestrial
tivity and low energy consumption in IoT networks, according
IoT networks [5], we focus on the access performance from
to the capture effect of Long Range (LoRa) signal [2], the
the viewpoint of received signal strength indicator (RSSI) in
receiver can demodulate the collision packets if one of the
LEO satellite based IoT networks compared with the viewpoint
signal power is suffificiently higher than the other signal. In
of doppler frequency shift in satellite IoT networks [12].
this way, the reliability of the IoT networks based on LoRa
The main difference points between the LEO satellite based
singal [3] is improved by successive interference cancellation
and the terrestrial IoT networks for RSSI: (1) large scale
(SIC) technique. On this basis, Noreen et al. In [4] simulated
fading character: the distance between LEO satellite an the
the performance of LoRa signal combining with SIC technique
ground terminal is in the range of 500-1200km [13], while
in the terrestrial IoT networks. Jean Michel et al. In [5] derived
the terrestrial LoRa network base station covers from the range
a closed-form performance equation and used SIC technique
of several kilometers to about ten kilometers [14]; (2) small
to decode the superimposed signal in the LoRa networks.
scale fading character: compared with rich-scattering channel
Rachkidy et al. In [6] studied the interference cancellation
environment in terrestrial network, light-of-sight (LOS) chan-
supported by the National Science Foundation of China (No. 62171234), nel plays an important role with characteristic of non-rich-
the Jiangsu Province Basic Research Project (No. BK20192002), the Chi- scattering channel in satellite communication channel, which
na Postdoctoral Science Foundation (No. 2018M632347), and the Natural
Science Research of Higher Education Institutions of Jiangsu Province (No. is a typical Rice channel [15]; (3) the number of access
18KJB510030). terminal: compared with the coverage area by the terrestrial

978-1-6654-0785-4/21/$31.00 ©2021 IEEE

Authorized licensed use limited to: Takeshi Ikenaga. Downloaded on October 23,2023 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
base station, the LEO satellite beam covers area with a radius
of hundreds of kilometerswhich results in huge amounts of
terminal in coverage area of LEO satellite. In this access
scenario, the multi-collision packets status may happen to
increase the algorithm complexity or algorithm failure of SIC
compared with only considering the two collision packets
status in [5]. According to these difference points, we presents
a LEO satellite based IoT networks using LoRa signal to
analyze the uplink access performance from the viewpoint
of RSSI. Furthermore, we derive a connection probability
equation for LoRa signal from the viewpoint of RSSI in Rice
channel. Simulation results can be used to guide the Initial
transmission power setting for satisfying the average access
success probability (ASP) requirement of LEO satellite IoT
networks.
Fig. 1. LEO satellite based IoT networks scenario diagram.
II. S YSTEM M ODEL
LoRa signal based on a sub-GHz chirp spread spectrum
(CSS) modulation technology is optimized for long-distance satellite to illustrate the parameter SF setting. The coverage
and low-power transmission application [4]. This technology area is divided into six rings, and each ring area correspond
uses a quasi-orthogonal spreading factor (SF) to increase the to a SF value from 7 to 12. In [16], a SF fair distribution
link budget and the number of virtual channels at the cost 12
ratio was derived as pSFi = 2SF i
P
SFi / 2i , where pSFi is the
i
of the time on air (ToA) of the data packet. LoRa signal has
i=7
three important parameters: sub-channel bandwidth (B), sub- proportion of terminals using ith SF value with the constraint
P12
channel bandwidth of uplink is set to 125kHz or 250kHz, and of i=7 pSFi = 1, to balance the conflict probability of each
sub-channel bandwidth of the downlink is set to 500kHz; SF: SF. The ith ring area is bounded by inner radius li−1 and outer
SF can be set from 7 to 12 corresponding different signal- radius li . For the sake of SIC analysis in the following section,
to-noise ratio (SNR) threshold of LoRa signal; forward error we randomly select a reference terminal with the distance d1
correction (FEC) rate: FEC value varies from 4/8 to 4/5. from the LEO satellite. The traffic of activity terminals in each
Table I shows the relation between SF, Time on Air (ToA), ring is modeled by the homogeneous Poisson point process
bit rate, and link budget. (PPP) Φi with intensity βi = 2pi ρVi , βi > 0, where pi is the
duty-cycle of terminals in ith ring area, ρ = N̄ V is the space
TABLE I density, Vi = π(li2 − li−1
2
) denotes the area of ith ring area.
L O R A U PLINK C HARACTERISTIC :PAYLOAD OF 9 BYTES , B = 125 K H Z ,
CRC AND H EADER M ODE ENABLED
Therefore, the average number of terminals in ith ring area
can be denoted by N̄i = ρVi . Without loss of generality, we
SF Time-on-Air Bite rate Receiver Sensitivity SNR threshold consider that the transmit power of terminals is denoted by Pt
7 41.22ms 5.47kbps -123dBm -6dB
8 72.19ms 3.12kbps -126dBm -9dB
and the bandwidth of the LoRa signal is B in the uplink in
9 144.38ms 1.76kbps -129dBm -12dB the following sections.
10 247.81ms 0.98kbps -132dBm -15dB
11 495.62ms 0.54kbps -134.5dBm -17.5dB
12 991.23ms 0.29kbps -137dBM -20dB

Figure 1 shows the LEO satellite based IoT networkss with

terminals, LEO satellite constellation and gateway station.
Firstly, the IoT terminals transmit data packets to the LEO
satellite via the uplink. Then, IoT terminal data packets are
demodulated by the satellite receiver. Finally, the information
is transmitted to the gateway station by the inter-satellite and
the feeder link which are assumed to be error free. In this
paper, we focus on the access performance of the uplink
between IoT terminals and LEO satellite receiver. We consider
that a beam coverage area V ⊆ R2 , the distance between
the LEO satellite and the sub-satellite point is denoted by H,
the coverage area with a radius of R meters, and the area is
V =πR2 . There are N̄ terminals uniformly deployed in the
area. Figure 2 shows the coverage area top view of LEO Fig. 2. Top view of LEO satellite to illustrate the parameter SF setting.

Authorized licensed use limited to: Takeshi Ikenaga. Downloaded on October 23,2023 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
 The large scale path loss can be written as: where z is the collision ratio, and sign() is an indicator
λ η function. It is notes that we consider that the LEO satellite
g(dk ) = ( ) (1) only can demodulate two packets collision status because of
4πdk
the LEO satellite resource-constrained. In this way, the access
where dk is the distance between kth and the LEO satellite, success probability C1SIC can be expressed as:
λ = fcc is the wavelength, c is the speed of light, fc is
the carrier frequency, and η is the path loss coefficient. The C1 SIC ≈ H1 P2 Q1 + H1 P2 Q2 (8)
Small scale fading hk follows Rice fading with probability where Q2 is the connection probability of SIC algorithm for
distribution function (PDF) as: two packets collision. The equation of H1 , Q1 and Q2 will be
x x 2 + A2 x·A derived in the following subsection III(A), III(B) and III(C),
fhk (x) = 2
exp(− ) · I0 ( 2 ), x ≥ 0 (2) respectively.
σ 2σ 2 σ
where A is the amplitude of the line of sight (LOS) A. Connection Probability without Packet Collision
signal, σ 2 is the power of scattering signals, and I0 ()
If the receive signal power Pr at the LEO satellite receiver
is the modified 0-order first-kind Bessel function. Accord-
is larger than the value of receiver sensitivity and no packets
ing to the empirical formula in [17], the Rice factor
collision happens, we consider that the terminal successfully
K(α) = 2.731 − 0.1074α + 0.002774α2 , where α is the el-
access the LEO satellite receiver. The connection probability
evation angle between terminals and LEO satellite. Therefore,
is denoted by H1 = P [Pr ≥ Si |d1 , i] ,where Si is the value
the PDF of Rice channel power gain |hk |2 can be expressed
of receiver sensitivity for different SF value as shown in Table
as:
√ I. Therefore, H1 can be written as:
1 x + A2 x·A  Si −GR

f|hk |2 (x) = 2
exp(− 2
) · I 0 ( ), x ≥ 0 (3) 2 10( 10 −3)
2σ 2σ σ2 H1 = P |h1 | ≥ Pt g(d1 )
We consider the transmit signal by the reference terminal S −GR
( i −3)
(9)
10
is s1 . Thus, the receive signal r1 at the LEO satellite receiver R 10 Pt g(d1 )
= 1− 0 f|h1 |2 (x)dx
can be written as: 2
p X p where GR = 10log10 Gr , and Pr = 10log10 ( Pt |h110
| g(d1 )Gr
).
r1 = Pt g(d1 )Gr h1 s1 + Pt g(dk )Gr hk sk + ω
−3

k6=1,k∈φi B. Connection Probability with Capture Effect of LoRa Signal

(4)
According to [18], for two packets collision status, the
where sk is the kth terminal transmit singal, ω is the system
signal-to-interference ratio (SIR) can be written as:
noise power, and Gr denotes the receive gain of LEO satellite
antenna, which satisfies as: P |h |2 g(d1 ) |h1 |2 d1 −η
SIR = Pt 1 = (10)
Pt |hk |2 g(dk ) |hk |2 dk −η
 2 P
πD 2 J1 (µ) J3 (µ)
Gr = ξ( ) + 36 3 (5) k6=1,k∈φi k6=1,k∈φi
λ 2µ µ
If the SIR value is larger than the capture effect threshold
where µ = 2.07123 sin(θ)/sin(θ3dB ), θ3dB is the beam gain γ, the probability of demodulation strong signal in presence
relative to the angle corresponding to the 3dB attenuation of of interference is expressed as:
the beam center, J1 and J3 are the 1st and 3rd order Bessel
functions of the first kind, D is the antenna aperture, and ξ is Q1 = P [SIR ≥ γ|d1 ] (11)
the antenna efficiency. where γ is the capture effect threshold. In order to evaluate this
III. U PLINK ACCESS P ERFORMANCE FOR L O R A S IGNAL statuswe consider
p the worst-case for interfering signal k ∗ =
arg max{ Pt g(dk )Gr hk sk }. Thus Q1 can be written as:
The access performance of LoRa signal in LEO satellite
2
h h ii
based IoT can be divided into three categories: (1) connection Q1 = E|h1 |2 P Xk∗ < |h1 | γg(d1 ) |h1 |2 , d1
probability without pactet collision; (2) connection probability R∞ (12)
1)
with capture effect of LoRa signal; (3) connection probability = 0 f|h1 |2 (x)FXk∗ ( xg(dγ )dx
with SIC algorithm. Therefore, the connection probability C1 where Xk∗ = |hk∗ |2 g(dk∗ ), and FXk ∗ is the cumulative
can be written as: distribution function (CDF) of Xk∗ .
C 1 ≈ H1 P 2 Q 1 (6)
C. Connection Probability with SIC Algorithm
where H1 is the connection probability without packet colli-
On the basis of status III(B), if the LEO satellite receiver
sion, Q1 is the connection probability with capture effect, and
can recover the reference signal s1 and perform SIC algorithm
P2 denotes the probabilty of two pactets collision. According
to demodulate the interfering signal sk , the demodulation
to the [16], the PDF of P2 follows as:
probability can be written as:
2
1 X n 2

|hk |2 g(dk )

f (z) = · (−1) ( )(z − n) · sign(z − 2) (7) Q2 = P ≥ γ ||φi = 1|, d1 (13)
2 n=0 n |h1 |2 g(d1 )

Authorized licensed use limited to: Takeshi Ikenaga. Downloaded on October 23,2023 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
According to [5], consideing the LEO satellite based IoT
scenario, Q2 is derived as follows:
  √ η

βi e−βi 2 2

2 2 γ li2 +H 2
Q2 = li 2 −li−1 2 li + H 2 F1 1, η ; η + 1; − d1 η ...
 √ 2 η

γ li−1 +H 2
2
+ H 2 2 F1 1, η2 ; η2 + 1; −


− li−1 d1 η
(14)
where 2 F1 is the Gaussian hypergeometric function [19]. li−1
and li are the boundaries of the SFi ring.

Fig. 3. Q1 and Q2

Figure 3 shows that Q1 and Q2 represent the probabilities

of the disjoint event. When SIR value is larger than γ, the
reference signal can be decoded due to the capture effect.
When the SIR value smaller than −γ, the interference signal
can be decoded due to the SIC.
IV. S IMULATION R ESULTS AND D ISCUSSION

TABLE II
S IMULATION PARAMETERS

Parameters Parameter setting value

H 600km
R 318km-2160km
fc 470MHz
Pt 0.1W-1W Fig. 4. Access success performance of LoRa signal with R = 600km,
B 125kHz Pt = 0.3W. (a)N̄ = 10000, (b)N̄ = 50000.
η 2
D 0.6m-1.6m
ξ 55% P2 Q2 is around 1% and C1SIC is around 23%. Figure 4(b)
γ 1dB
shows that H1 is almost 100% and P2 Q1 is around 4%. If
SIC is considered, P2 Q2 is around 5% and C1SIC is around
In this section, we analyze the access performance of LoRa 9%.
signal in LEO satellite based IoT networks. The access perfor-
mance is performed in the MATLAB simulation platform and
the Monte Carlo method is adopted to verify the theoretical
access performance. The simulation conditions are supposed
in Table II.
Figure 4 shows the access success probability performances
of LoRa signal varying with the distance between terminal
and sub-satellite point when the transmit power Pt euqals to
0.3W, R is 600km, antenna aperture D is 1m, and the number
of terminals N̄ equals to 10000 and 50000, respectively. In
the figure 3(a) and (b), different colored line denotes different
access theoretical performances of success probability for H1 ,
P2 Q1 , P2 Q2 , and C1SIC , respectively, while the circular marks
represents the these probability performances obtained by
Monte Carlo simulation. It is found that theoretical perfor-
mance curves of success probability are well matched with
those circular marks by Monte Carlo simulatuion method. For Fig. 5. Probabilities at the cell border (d1 = 600km), γ = 1dB.
the terminal at d1 = R = 600km, Figure 4(a) shows that H1 is
almost 100% and P2 Q1 is around 22%. If SIC is considered, Figure 5 shows that the access success probability perfor-

Authorized licensed use limited to: Takeshi Ikenaga. Downloaded on October 23,2023 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.
mances of LoRa signal varying with the average network usage
βi when the terminal are selected at the edge of the beam R EFERENCES
coverage range (d1 = 600km), the transmit power Pt euqals
[1] T. T. Nguyen, H. H. Nguyen, R. Barton and P. Grossetete, ”Efficient
to 0.3W, antenna aperture D is 1m, the SIR threshold γ is Design of Chirp Spread Spectrum Modulation for Low-Power Wide-
1dB, respectively. In the figure 5, different colored line denotes Area Networks,” in IEEE Internet of Things Journal, vol. 6, no. 6, pp.
different access theoretical performances of success probabil- 9503-9515, Dec. 2019, doi: 10.1109/JIOT.2019.2929496.
[2] W. Ayoub, A. E. Samhat, F. Nouvel, M. Mroue, and J.-C. Prvotet,
ity for H1 , H1 P2 Q1 , and H1 P2 (Q1 + Q2 ), respectively. It ”Internet of Mobile Things: Overview of LoRaWAN, DASH7, and NB-
is found that when the usage rate βi increases gradually, H1 IoT in LPWANs standards and supported mobility,” in IEEE Commun.
remains unchanged, and H1 P2 Q1 and H1 P2 (Q1 + Q2 ) are Surveys Tuts., vol. 21, no. 2, pp. 1561-1581, Secondquarter 2019, doi:
10.1109/COMST.2018.2877382.
in decline. When the usage rate is at 100% (βi = 1), H1 is [3] LoRaWAN 1.0.3 Specification, LoRaWAN Alliance, Fremont, CA,
almost 100%, H1 P2 Q1 is around 22%, and H1 P2 (Q1 + Q2 ) USA, Jul. 2018.
is around 30 %. [4] U. Noreen, L. Clavier and A. Bounceur, ”LoRa-like CSS-based PHY
layer, Capture Effect and Serial Interference Cancellation,” European
Wireless 2018; 24th European Wireless Conference, 2018, pp. 1-6.
[5] J. M. de Souza SantAna, A. Hoeller, R. D. Souza, H. Alves and S.
Montejo-Snchez, ”LoRa Performance Analysis with Superposed Signal
Decoding,” in IEEE Wireless Communications Letters, vol. 9, no. 11,
pp. 1865-1868, Nov. 2020, doi: 10.1109/LWC.2020.3006588.
[6] N. E. Rachkidy, A. Guitton and M. Kaneko, ”Decoding Superposed Lo-
Ra Signals,” 2018 IEEE 43rd Conference on Local Computer Networks
(LCN), 2018, pp. 184-190, doi: 10.1109/LCN.2018.8638253.
[7] B. Laporte-Fauret, M. A. Ben Temim, G. Ferre, D. Dallet, B. Minger
and L. Fuch, ”An Enhanced LoRa-Like Receiver for the Simultaneous
Reception of Two Interfering Signals,” 2019 IEEE 30th Annual Inter-
national Symposium on Personal, Indoor and Mobile Radio Communi-
cations (PIMRC), 2019, pp. 1-6, doi: 10.1109/PIMRC.2019.8904258.
[8] X. Xia, Y. Zheng and T. Gu, ”FTrack: Parallel Decoding for LoRa
Transmissions,” in IEEE/ACM Transactions on Networking, vol. 28, no.
6, pp. 2573-2586, Dec. 2020, doi: 10.1109/TNET.2020.3018020.
[9] Z. Qu, G. Zhang, H. Cao and J. Xie, ”LEO Satellite Constellation for
Internet of Things,” in IEEE Access, vol. 5, pp. 18391-18401, 2017,
doi: 10.1109/ACCESS.2017.2735988.
[10] Y. Qian, L. Ma and X. Liang, ”The Acquisition Method of Symmetry
Chirp Signal Used in LEO Satellite Internet of Things,” in IEEE
Fig. 6. At the edge of the beam coverage range, the relation diagram of Communications Letters, vol. 23, no. 9, pp. 1572-1575, Sept. 2019, doi:
transmit power and access success probability without packet collision for 10.1109/LCOMM.2019.2926262.
different LEO satellite antenna apertures. [11] C. Yang, M. Wang, L. Zheng and G. Zhou, ”Folded Chirp-Rate Shift
Keying Modulation for LEO Satellite IoT,” in IEEE Access, vol. 7, pp.
Figure 6 shows that the access success probability perfor- 99451-99461, 2019, doi: 10.1109/ACCESS.2019.2930327.
[12] Wu T, Qu D, Zhang G. Research on LoRa Adaptability in the LEO
mances of LoRa signal without packet collision varying with Satellites Internet of Things[C]// 2019 15th International Wireless Com-
the transmit power Pt when the terminal are selected at the munications and Mobile Computing Conference (IWCMC). 2019.
edge of the beam coverage range. In the figure 6, different [13] H. Mroue, A. Nasser, B. Parrein, S. Hamrioui, E. Mona-Cruz and G.
Rouyer, ”Analytical and Simulation study for LoRa Modulation,” 2018
colored line denotes different access theoretical performances 25th International Conference on Telecommunications (ICT), 2018, pp.
of success probability for D = 0.6m, D = 0.8m, D = 1m, 655-659, doi: 10.1109/ICT.2018.8464879.
D = 1.2m, D = 1.4m, and D = 1.6m, respectively. It is found [14] Y. Qian, L. Ma and X. Liang, ”Symmetry Chirp Spread Spectrum
Modulation Used in LEO Satellite Internet of Things,” in IEEE Com-
that under different antenna apertures, the lines in the Figure munications Letters, vol. 22, no. 11, pp. 2230-2233, Nov. 2018, doi:
6 show an upward trend. When Pt = 0.3W and D is more 10.1109/LCOMM.2018.2866820.
than 1m, the access success probability is more than 90%. [15] Y. Qian, L. Ma and X. Liang, ”The Acquisition Method of Symmetry
Chirp Signal Used in LEO Satellite Internet of Things,” in IEEE
V. C ONCLUSION Communications Letters, vol. 23, no. 9, pp. 1572-1575, Sept. 2019, doi:
10.1109/LCOMM.2019.2926262.
According to the channel characteristic of satellite commu- [16] B. Reynders, W. Meert and S. Pollin, ”Power and spreading factor
nication system, we analyze the uplink access performance control in low power wide area networks,” 2017 IEEE International
Conference on Communications (ICC), 2017, pp. 1-6, doi: 10.1109/IC-
of LoRa signal in LEO satellite based IoT networks in this C.2017.7996380.
paper. We also derived a connection probability equation for [17] G. E. Corazza and F. Vatalaro, ”A statistical model for land mobile
LoRa signal from the viewpoint of RSSI based on the capture satellite channels and its application to nongeostationary orbit systems,”
in IEEE Transactions on Vehicular Technology, vol. 43, no. 3, pp. 738-
effect and SIC technology in Rice channel. Simulation results 742, Aug. 1994, doi: 10.1109/25.312773.
can be used to guide the Initial transmission power setting for [18] A. Mahmood, E. Sisinni, L. Guntupalli, R. Rondn, S. A. Hassan and
satisfying the average access success probability requirement M. Gidlund, ”Scalability Analysis of a LoRa Network Under Imperfect
Orthogonality,” in IEEE Transactions on Industrial Informatics, vol. 15,
of LEO satellite IoT networks. Further works on Increasing no. 3, pp. 1425-1436, March 2019, doi: 10.1109/TII.2018.2864681.
the access performance need to be discussed such as IoT [19] A. Daalhuis, ”Hypergeometric function,” in NIST Handbook of Mathe-
terminal power control strategy based on machine learning matical Functions. New York, NY, USA: Cambridge Univ. Press, 2010,
ch. 15.
algorithm, adaptive SF selection method and joint optimization
of transmission rate and SF.

Authorized licensed use limited to: Takeshi Ikenaga. Downloaded on October 23,2023 at 10:42:51 UTC from IEEE Xplore. Restrictions apply.