System-Level Metrics for Non-Terrestrial Networks Under

Stochastic Geometry Framework

Journal: IEEE Communications Magazine

Manuscript ID COMMAG-23-00561

Manuscript Type: Series Mobile Communications and Networks

Date Submitted by the

02-Oct-2023
Author:

Complete List of Authors: Huang, Qi; Purdue University

Belmekki, Baha Eddine Youcef; King Abdullah University of Science and
Technology Computer Electrical and Mathematical Science and
Engineering Division
Eltawil, Ahmed M.; King Abdullah University of Science and Technology
Computer Electrical and Mathematical Science and Engineering Division
Alouini, Mohamed-Slim; King Abdullah University of Science and
Technology Computer Electrical and Mathematical Science and
Engineering Division

non-terrestrial network, Stochastic Geometry, system-level metrics, 6G

Keywords:
wireless network

Note: The following files were submitted by the author for peer review, but cannot be converted to PDF.
You must view these files (e.g. movies) online.

System-Level Metrics_NTN_Qi Huang_Baha.zip

ReX_Author_Contributions.xml
Page 1 of 7

1
2
3 System-Level Metrics for Non-Terrestrial Networks
4
5
6
Under Stochastic Geometry Framework
7 Qi Huang, Student Member, IEEE, Baha Eddine Youcef Belmekki, Ahmed M. Eltawil, Senior Member, IEEE, and
8 Mohamed-Slim Alouini, Fellow, IEEE
9
10
11
12 Abstract—Non-terrestrial networks (NTNs) are one of the key The network topology, defined by the number, height, and
enablers in sixth-generation (6G) wireless networks; and with distribution of platforms, is one of the most important factors
13
their rapid growth, system-level metrics analysis adds crucial influencing system-level metrics. Stochastic geometry (SG) is
14 understanding into NTN system design. Applying stochastic
15 geometry (SG) as a system-level analysis tool in the context a powerful mathematical tool for assessing system-level met-
16 of NTN offers novel insights into the network tradeoffs. In rics in large-scale network topology. Moreover, SG is also one
17 this paper, we study and highlight NTN common system-level of a few tools that can provide analytical results for system-
metrics from three perspectives: NTN platform types, typical level metrics while incorporating co-channel interference into
18 communication issues, and application scenarios. In addition to
19 the analysis. Although the SG-based analysis methods have
summarizing existing research, we study the best-suited SG mod-
20 els for different platforms and system-level metrics which have powerful analytical tractability, in certain instances, it ignores
21 not been well studied in the literature. In addition, we showcase the correlation among the platforms. For instance, the binomial
22 NTN-dominated prospective application scenarios. Finally, we point process (BPP), one of the most commonly-used SG
carry out a performance analysis of system-level metrics for these models for satellites, assumed that satellites are distributed
23 applications based on SG models.
24 on the spherical surface independently and uniformly [2],
25 in contrast to the actual distribution where satellites are
26 I. I NTRODUCTION distributed on fixed orbits. Fortunately, the SG-based system-
27 level metrics analysis (such as coverage probability) has been
In recent years, the world has witnessed exponential growth proven to approximate the lower bound of the deterministic
28 in network services, driven by emerging use cases of sixth-
29 constellation, such as the Walker constellation, in the case of
generation (6G) wireless networks, such as virtual reality a large-scale satellite constellation [3]. The contributions of
30 and holographic communications. Non-terrestrial networks
31 the paper are as follows:
(NTNs) have shown significant potential as an enabling tech-
32 nology for new wireless communication generations. Com- • While prior works focus on introducing and developing
33 pared to densely deployed terrestrial networks, NTNs are still performance metrics of NTNs, this work evaluates how
34 under development and face various challenges, including in- these metrics influence the entire NTN system perfor-
35 terconnectivity between different airborne platforms, terrestrial mance, rather than an individual link.
36 networks, and among themselves [1]. • To the best of the authors’ knowledge, this is the first
37 There are mainly three types of NTN platforms: Low paper to specifically elucidate the system-level metrics in
38 altitude platform (LAP), high altitude platform (HAP), and the NTNs, and analyze these metrics based on special
39 satellites. These platforms provide coverage for areas that specific application scenarios, including remote and rural
40 the terrestrial network cannot reach. They can also enhance coverage, post-disaster reconstructed networks, and mili-
41 global connectivity with low latency and long-distance reach. tary operations. Finally, the system-level metrics of these
42 Moreover, NTNs benefit from a suitable communication envi- scenarios are simulated according to SG models.
43 ronment where the signal is propagated mostly through free • This paper points out new research directions on the
44 space since it does not experience severe shadowed fading system-level metrics of NTN platforms based on SG mod-
45 or multi-path fading. NTN platforms will be part of the 6G els, which will have a great improvement on the overall
46 architecture networks, therefore, system-level analysis, such performance evaluation of the NTN network combining
47 as large-scale connectivity, becomes more significant than the LAPs, HAPs, and satellites.
48 performance analysis of a single link connectivity. Unlike
49 metrics that highlight the flying attitude and deployment of II. NTN P LATFORMS
50 the platform, system-level metrics focus on the performance
51 NTN platforms have different characteristics and features,
of the entire NTN system. The system-level metrics presented
52 hence, different SG models are used for each type of NTN.
in this paper include coverage probability, availability, channel
53 The radar diagram in the left bottom part of Fig. 1 provides
capacity, propagation, latency, and energy efficiency.
54 the quantitative results of metrics comparison.
55 Qi Huang carried out this work when he was at KAUST, and is
currently with Purdue University (e-mail: huan2065@purdue.edu. B. E.
56 Y. Belmekki, A. M. Eltawil, and M.-S. Alouini are with King Ab- A. Low-Altitude Platform
57 dullah University of Science and Technology (KAUST), CEMSE divi-
58 sion, Thuwal 23955-6900, Saudi Arabia (e-mail: {bahaeddine.belmekki, LAP refers to platforms that typically fly at 50 m – 4 km,
59 ahmed.eltawil,slim.alouini}@kaust.edu.sa. and applies to platforms flying under 10 km. The altitudes
60
 Page 2 of 7

1
2 GEO: 35786 km LEO: 160-2000 km
3
Satellite Coverage
4
form
m 5
Relay
6 HAP 17-22 km
7
8
9 LAP
<10 km Routing Source
10

SI
NR
11


th
re
12

sh
Energy

ol
efficiency 4

d
LEO satellite
13 3
Min latency Destination 160–2,000 km
GEO satellite
35,786 km
14GEO satellite 2
1 Coverage routing
15 35,786 km Availability
0
range

16
17
Max energy routing Gateway
18
19 Satellite
HAP
20 Capacity Latency
LAP

21 Higher score means better performance or larger latency

22
23 Fig. 1: System framework diagram.
24
25
26 of most flying platforms, such as unmanned aerial vehi- The performances of HAPs are ranked between LAPs and
27 cles (UAVs), networked tethered flying platforms (NTFPs), satellites regarding coverage area, propagation latency, channel
28 aerostats, and balloons, vary from tens of meters to hundreds capacity, and energy efficiency due to their altitudes which
29 of meters. LAPs are widely used to form a temporary network are lower than satellites and higher than LAPs. HAPs fly
30 to enhance coverage owing to their flexible deployment. They at high altitudes so they do not undergo frequency shadow
31 are generally deployed in small regions at low altitudes, fading caused by objects such as buildings and trees. Thus, the
32 forming a rapidly changing topology due to their frequent links between HAPs and service targets have a higher line-of-
33 movement, making the density of LAPs more important sight (LoS) probability than LAPs. Consequently, HAPs have
34 compared with their numbers. Therefore, a two-dimensional a stable signal-to-noise ratio (SNR) and negligible fluctuations
35 Poisson point process (PPP) is suitable for modeling a LAP of system-level metric values. They also benefit from higher
36 network [4]. In typical NTN application scenarios, including availability (both transmitter and receiver are above the hori-
37 remote areas, post-disaster areas, and military communications zon) due to their greater LoS probability if deployed strategi-
38 (See Section IV), LAPs are often carried on vehicles (such cally. Higher altitude leads to better availability. Furthermore,
39 as trucks) and need frequent charging, making it difficult to HAPs can be powered by solar energy, hence, they can stay
40 provide stable coverage and availability for LAPs. Moreover, aloft for several years in the stratosphere.
41 the coverage area of a single LAP is significantly smaller than
42 that of a HAP or a satellite due to its low altitude. In addition,
43 a small channel capacity suffices to support serving targets C. Low Earth Orbit Satellite
44 within LAPs coverage. In comparison, LAPs possess high- Low Earth orbit (LEO) satellites orbit around the Earth
45 energy efficiency and low propagation latency due to their with an altitude between 160–2000 km above the Earth’s
46 lower altitude. surface. Examples of LEO satellite networks include Starlink
47 and OneWeb. LEO satellites are not stationary relative to the
48 Earth while geosynchronous equatorial orbit (GEO) satellites
B. High-Altitude Platform are geosynchronous at 35786 km. Since most GEO satellite
49
50 HAPs are flying platforms between 17 and 22 km in the constellations are only composed of a few satellites, SG
51 stratosphere. It has not been implemented until recent years models are not accurate and suitable for GEO satellites. For
52 with technological progress due to the harsh environmental small-scale LEO satellite constellations composed of dozens
53 conditions such as low temperature, high wind speed, and of satellites (e.g., Iridium), the orbit geometry model is the
54 long exposure to radiation. HAPs are distributed on a thin closest one to approximate the actual distribution of satellites
55 hollow spherical shell with 5 km thickness above the Earth. [5]. Unlike BPP and PPP, the orbit geometry model is a
56 Therefore, HAPs are modeled approximately by BPP over a semi-stochastic model, that is, the positions of orbits are
57 sphere. BPP is more accurate than PPP for closed surfaces deterministic, while the distribution of satellites in orbit is
58 such as spheres [3]. We note that BPP can also be used to stochastic. As a stochastic model, details about orbits are
59 model LAPs distributed in a finite area. ignored in BPP. However, spherical BPP, as is described in
60
Page 3 of 7

1
2 subsection II-B, is well-suited to approximate massive dense B. Relay-Based: Dual-Hop Communication
3 constellations. To account for the circular shape and multiple
Relay communication can be viewed as the combination
4 altitudes of satellite orbits, the Cox point process model was
of an uplink and a downlink coverage, or a basic unit of
5 presented, where orbits are created based on PPP on a cuboid
routing. Compared with coverage and routing, availability and
6 and satellites are distributed conditionally on these orbits [6].
channel capacity are more suitable for relay-based analysis.
7 For well-designed constellations, the unavailability of satel-
The results of these two system-level metrics obtained in relay-
8 lites on the horizon hardly occurs owing to the non-temporary
based analysis provide better insight than a coverage-based one
9 nature of satellite networks. The coverage area and capacity
and can be easily extended to results in multi-hop routing.
10 of a single LEO or GEO satellite are much larger than those
We start with HAP availability between two ground users.
11 of HAPs [7]. A single LEO satellite system provides seamless
Due to blockage caused by the Earth, platforms below the
12 coverage for the whole Earth, while global coverage can be
horizon are out of the ground user LoS range and cannot
13 achieved with only three GEO satellites. The channel capacity
provide a stable service. Therefore, the HAP has to be at the
14 of LEO satellite systems suffices to support communications in
overlapping region composed of two spherical caps above the
15 remote areas. The tens to hundreds of milliseconds propagation
horizons. The area of this overlapping region is significantly
16 latency of LEO satellites can support most real-time commu-
smaller than a single-user LoS spherical cap when considering
17 nication scenarios. Satellites, including LEO and GEO, have
single-hop communications. Therefore, it is more relevant to
18 a low energy efficiency due to the satellite-ground distance.
discuss HAP availability in relay-based analysis. When HAPs
19 Even with a large transmission power, the received SNR
are modeled as BPP, SG analysis can provide concise ana-
20 may still not exceed the decoding threshold. Due to severe
lytical results for HAP availability using the null probability.
21 attenuation in the ground clutter layer, satellite communication
The SG analysis can also provide the probability of HAPs
22 often requires HAPs or ground gateways for relaying.
availability in the overlapping region.
23 In terms of satellite system availability, we focus on LEO
24 III. T YPICAL C OMMUNICATION A NALYSES
satellites that are not geosynchronous. LEO satellite availabil-
25 Depending on the number of hops, there are three commu- ity is called pass duration. Pass duration is the continuous
26 nication analyses: coverage, relay, and routing. The coverage time interval when the LEO satellite is serving a given user.
27 analysis using coverage probability as a metric has been the Denoting the arc drawn by the satellite during this interval
28 most studied system-level metric in SG literature. However, as the session arc, the duration can be given by a fraction of
29 the system-level metrics in relay communications (namely the orbital period, which is equal to the ratio of the session
30 availability and channel capacity) and routing (namely latency arc length to the orbit length [10]. The SG-based framework
31 and energy efficiency) still need further investigations. provides tools to derive analytical expressions of the session
32 arc length and pass duration. Finally, similar to the concept of
33 LEO satellite pass duration, LAP availability is defined as the
A. Coverage-Based Analysis: Single-Hop Communication
34 ratio of serving time to the whole period, since LAPs need to
35 Coverage probability is a key metric in coverage-based anal-
ysis, which is the probability of received signal-to-interference be recharged frequently. The whole period is the summation
36 of serving time, back-haul duration, and charging duration [9].
37 plus noise ratio (SINR) being larger than a predefined thresh-
old. Factors that affect coverage probability include [7]: 1) the Channel capacity is defined as the ergodic capacity given
38 by the Shannon-Hartley theorem over a fading channel [3].
39 interference power; 2) the distance to the serving platform;
3) and small-scale fading. An important feature of SG-based Generally, a large packet is decomposed into many small
40 packets. The channel capacity in a multi-hop link is determined
41 analysis is that the aforementioned three factors are assumed
to be random variables subject to certain distributions. by the hop with the lowest capacity. In addition, since the
42 channel capacity in uplink and downlink might be different, it
43 Interference power is the sum of the received powers of all
the transmitting nodes in the network excluding the serving is more relevant to investigate channel capacity in relay-based
44 analysis. Under limited transmitting power, the SG-based
45 platform power. The authors in [3] show that, by taking
the Laplace transform of the interference power, the specific method maximizes channel capacity by selecting platforms at
46 optimal locations. To make the analysis of channel capacity
47 distribution of the interference will not affect the analytical
expression of the coverage probability. Second, in [2], the tractable, the SG-based method often selects a suboptimal
48 relay strategy [11]. Such suboptimal relay strategy selects
49 distance to the serving platform in the SG model (contact
distance distribution), is the distribution of the distance be- the platform with the maximum receiver capacity as the
50 relay among a subset of platforms that has a communication
51 tween the user and the serving platform with the strongest
average received power. Finally, small-scale fading in the SG- capacity exceeding a given threshold.
52
53 based channel model also contributes to the randomness of
54 received power. The air-to-ground link passing through the
C. Routing-Based Analysis: Multi-Hop Communication
55 ground clutter layer induces a severe multi-path effect, leading
56 to small-scale fading. Shadowed-Ricean fading is the most Path planning and relay selections are the main focus
57 accurate small-scale fading model satellite to ground user link when conducting a routing-based analysis [12]. Unlike relay
58 or HAP to ground [8] while that from LAP to ground is best selection in channel capacity analysis, path planning and relay
59 captured by Nakagami-m fading [9]. selection in routing is based on all links of the entire network.
60
 Page 4 of 7

1
TABLE I: System framework diagram
2
3 Platform Altitude SG Model Advantages Challenges Fading model
4 UAV 150 m–200 m
• Limited coverage
Balloon 150 m–700 m • PPP • High energy efficiency Shadowed-Racian
5 LAP*
Blimps 100 m–5 km • BPP for a finite area • Low propagation latency
area, availability, and
fading
6 Helikites 100 m-1.5 km
channel capacity
7 Aerostat • More LoS probability • High deployment
Shadowed-Racian
8 HAP Airship
17 km–22 km
• BPP than LAP; expense and difficulty;
fading
9 Balloon • Longer lifetime; Stable • Low mobility
system metrics values
10 LEO 160 km–2000 km • BPP • Large coverage area • Long propagation latency; Nakagami-m
11 Satellite
GEO • Semi-stochastic model and availability • Low energy efficiency; fading
35786 km
12 • Cox point process • Severe attenuation
13 * All LAP systems can be tethered to the ground [1].
14
15
When selecting a relay platform, the best relay locations for system in these areas. Covering a large remote area with a
16
the next few hops have to be considered in advance. limited number of devices at a manageable cost is challeng-
17
Because each hop has an equal effect on the total latency, ing. Hence, studying the coverage probability under different
18
propagation latency is a typical system-level metric considered platform distributions is necessary to design a robust NTN
19
in routing. Propagation latency is defined as the time spent system. Furthermore, access equity in remote areas ensures
20
by the signal to propagate through the transmission medium. fairness among all the users in the network. Access equity
21
Propagation latency in densely distributed LAP networks has can be measured by availability, i.e., ensuring that at least
22
been comprehensively studied using SG framework [13]. As one platform is available to access the network via an NTN
23
for the propagation latency as a system-level metric in HAP platform.
24
and satellite networks, SG is the only mathematical tool pro-
25
viding analytical results for spherical routing. Through contact
26 B. Post-disaster Reconstructed Networks
distance distribution, the authors in [12] provide accurate lower
27 When a disaster destroys terrestrial communication infras-
and upper bounds for propagation latency in multi-hop routing.
28 tructures, NTNs play a pivotal role in providing coverage
Energy efficiency is the average number of bits that can
29 for the disaster area. Considering that, satellite networks can
be transmitted per unit of energy consumed, which can be
30 achieve seamless and ubiquitous global coverage; the post-
measured as the ratio of channel capacity to total transmis-
31 disaster network will utilize satellites as the primary NTN
sion power. Generally, there are two main strategies of path
32 platform for communication. LAPs and HAPs can form a
and relay selection: minimum propagation latency routing
33 temporary network in a short period of time to increase the
and maximum-efficiency routing. The minimum propagation
34 capacity of the post-disaster network.
latency routing uses the least number of hops. Smaller number
35 When the original communication networks are destroyed
of hops leads to lower latency, but with the same transmission
36 by a disaster, the affected areas will face a high demand
distance, it has a lower energy efficiency. In comparison, larger
37 for communication services in a short period. Therefore, the
number of hops leads to larger energy efficiency since the
38 channel capacity of NTN is a key system-level metric to assess
transmission of each bit requires less energy, but the latency
39 the quality of communications in post-disaster scenarios. Fur-
is larger when the transmission distance is the same.
40 thermore, satellites rely on solar power for recharging, which
41 limits the amount of energy they can store. On the other
42 IV. A PPLICATION S CENARIOS hand, HAPs and LAPs require frequent recharging, which
43 In this section, we first describe three relevant application is challenging in post-disaster areas. Therefore, improving
44 scenarios of NTNs under the SG framework. Then, we select energy efficiency is critical in post-disaster scenarios.
45 the most relevant system-level metrics for each application ac-
46 cording to its specific requirement. Finally, we put forward the
C. Military Operations
47 prospect of future research directions under the SG framework.
48 During military operations, communication and surveillance
49 are of the utmost importance and play a decisive role in
A. Remote and Rural Coverage warfare. Communication consists of the transmission of com-
50
51 The cost of laying terrestrial infrastructures, such as optical mands, orders, and instructions to their soldiers. It also consists
52 fibers, is high. Remote areas, such as mountain areas, have of the transmission of information gathered through surveil-
53 sparse populations and usually low income compared to urban lance. Surveillance, on the other hand, consists of persistent
54 populations. Consequently, these areas are lacking economic and prolonged monitoring of a target or a region. Commu-
55 incentives for mobile network operators (MNOs) to deploy nication through terrestrial networks especially in an enemy-
56 terrestrial infrastructure. occupied area is difficult and dangerous during military ac-
57 The most crucial communication aspects in remote areas tions. This is because the network can be easily compromised,
58 are coverage and access equity. Therefore, investigating these hijacked, or eavesdropped on. In addition, surveillance requires
59 two aspects is of utmost importance when designing an NTN visual observation and can be conducted using soldiers on
60
Page 5 of 7

1
2 the ground or a UAV. NTNs are a preferable alternative for Considering that availability is a key system-level metric
3 terrestrial networks for this type of operation. NTNs have a in both relay communications and remote area networks, we
4 larger coverage than terrestrial networks, perform surveillance constructed a hybrid HAP and LEO satellite network to carry
5 with a large visual range, are easy and fast to deploy during out an availability analysis. Ground users are modeled as a
6 a military operation, and have a higher LoS probability. homogeneous PPP on a spherical cap with density 1/km2 .
7 When monitoring fast-moving military equipment, a low The area of the spherical cap is 1.55 × 105 km2 , its dome
1
8 propagation latency is necessary so that the target will not angle is 45 π, and its radius is 6371 km, which is equal to the
9 leave the detection range, making it critical to form a high- Earth radius.
10 quality network. In addition, high bit rate images and videos Assuming that the nearest terrestrial base station from a
11 help military commanders complete military deployments, core network is 500 km away from the center of the spherical
12 take strategic decisions, and locate enemy forces. These high- cap so that there is no terrestrial base station serving this
13 resolution images or video streams rely heavily on high area. Therefore, users in a remote area cannot establish direct
14 channel capacity. When the military has to transmit mes- communication links with the core network and must be
15 sages, orders, or reports (both in the field and at sea) or is relayed by HAPs or LEO satellites. The satellites form a
16 faced with a demanding surveillance mission, the k-coverage homogeneous spherical BPP in the whole sphere with a radius
17 probability is a more pertinent system-level metric than the of 6921 km, which is the same radius as the recently deployed
18 coverage probability for these types of operations. The k- Starlink constellation. Unlike satellites, HAPs can be deployed
19 coverage is achieved when an object or region is covered by stationary relative to the ground, thus we deploy HAPs directly
20 at least k platforms at the same time [14]. NTN platforms above the user. HAPs are modeled by a homogeneous BPP on
21 have advantageous anti-reconnaissance ability and mobility a spherical cap with a radius of 6391 km (the HAPs altitude
22 compared with the terrestrial network platforms to form the is 20 km above Earth’s mean sea level) and the same dome
23 high-quality network and satisfy the k-coverage probability to angle as users.
24 reliably support military operations. A randomly selected user in a remote area initiates a
25 communication request. The relay communication link is
26 V. S IMULATION R ESULTS available when there is at least one NTN platform in the
27 common LoS region between the user and the ground base
In this section, we provide comprehensive simulation results station. Fig. 2 shows the ratio of the number of available
28 that include several system-level metrics, NTN platform types,
29 communication links to the number of communication requests
communication analysis, and application scenarios. The mod- under the different numbers of HAPs and satellites. Even
30 els used to characterize NTN platforms and users are based
31 though LEO satellites are distributed on the whole sphere,
on the existing SG models. To enhance the richness of the they still contribute more to availability than HAP. Satellites
32 simulation results, we choose communication analysis with
33 have always been the first choice to promote equal access
system-level metrics that have not been well-investigated in to networks. A small-scale LEO satellite constellation of 40
34 the literature. In the simulations, we consider decoding and
35 satellites assisted by 50 HAPs is sufficient to provide 95%
forward relaying scheme, and the power consumption at each coverage over a region with 1.55 × 105 km2 area.
36 node is not considered.
37
B. Capacity and Energy Efficiency of Post-disaster Networks
38
39 A. Relay Communications Availability in Remote Areas We consider an aftermath scenario in which a ground
40 network in a 2-dimensional plane with a radius of 30 km was
41 ４０ destroyed by a disaster. The locations of the ground residents
０．９６
42 in this disaster area follow a homogeneous PPP with a density
３６
43 ０．９４ of 100/km2 . LAPs form a homogeneous BPP at an altitude of

ＮｕｍｂｅｒｏｆＳａｔｌｉｓ２１
44 ３２ 80 m in the disaster area. HAPs and satellites follow the same
０．９２
45 distribution as subsection V-A. Therefore, the deployment area
46 ０．９ of HAP will be larger than the disaster area.
47 In this NTN system composed of LAPs, HAPs, and satel-
０．８８ lites, all the platforms transmit in an RF frequency of 2 GHz.
48
49 ０．８６
LAPs adopt the same channel model as in [9], and the channel
50 model adopted by HAPs and satellites is given in [2]. The
51 ０．８４ transmission powers of LAPs, HAPs, and satellites are 1 dBW,
52 8 dBW, and 15 dBW, respectively, and their antenna gains
８ ０．８２
53 are 10 dB, 20 dB, and 70 dB, respectively. The small-scale
54 ４ ０．８ fading parameters of the LAP channel are the same as [9],
１０ ２０ ３０ ４０ ５０ ６０ ７０ ８０ ９０ １００ and the small-scale fading parameters of the HAP and satellite
55 Ｎｕｍｂｅｒ ｏｆ ＨＡＰｓ
56 channels are given in [2]. The environmental noise is set to
Fig. 2: Availability analysis of relay communications in -174 dBm/Hz.
57
remote areas. As shown in the lines with circle and star marks in Fig. 3(a),
58
59 when the number of LAPs increases, the channel capacity
60
 Page 6 of 7

1 0.9 1
2 1
(Sat, HAP, LAP)
( 80, 40 , 4000)
0.9
3
0.8
0.9 ( 20, 10 , 4000)
( 80, 40 , 1000)

4
0.8 ( 20, 40 , 1000)
0.7 0.8 ( 80, 10 , 1000)
( 20, 10 , 1000)

5
0.7
CDF of Channel Capacity

0.7

CDF of Energy Efficiency

0.6

4-Coverage Probabality
6
0.6
0.6
0.5

7
0.5
0.5
0.4

8 0.3
0.4
0.4

9 (Sat, HAP, LAP)

0.3
(Sat, HAP, LAP)
0.3

10
0.2 ( 80, 40 , 4000) ( 80, 40 , 4000)
0.2 0.2
( 20, 10 , 4000) ( 20, 10 , 4000)
( 80, 40 , 1000) ( 80, 40 , 1000)

11 0.1 ( 20, 40 , 1000)

( 80, 10 , 1000)
0.1 ( 20, 40 , 1000)
( 80, 10 , 1000)
0.1

12
( 20, 10 , 1000) ( 20, 10 , 1000)
0 0 0
150 200 250 300 350 0 100 200 300 400 500 0 5 10 15 20

13
Channel Capacity [Mbps] Energy Efficiency [Mb/J] SNR Threshold [dB]

(a) Capacity (b) Energy Efficiency

14
15 Fig. 3: Capacity and energy efficiency analysis in post-disaster
Fig. 4: k-coverage probability
16 reconstructed network.
under different thresholds
17 (k = 4).
18
19
20 increases significantly. The cumulative distribution function coverage, relay, and routing, were introduced and the system-
21 (CDF) curves corresponding to the red and green lines con- level metrics of each issue were elaborated. The requirements
22 verge, indicating that almost only LAPs can achieve a channel for system-level metrics were further discussed for the three
23 capacity greater than 300 Mbps. With the same number of representative application scenarios of NTNs. Finally, we
24 LAPs, increasing the number of HAPs brings more benefits to conducted some new combinations of system-level metrics
25 channel capacity than increasing the number of satellites. The with communication issues. The detailed SG-based analysis
26 curves in Fig. 3(b) show the same convergence characteristics framework and numerical results of the metrics are presented.
27 as in Fig. 3(a), which means that only LAPs can achieve
28 an energy efficiency of more than 250 Mb/J. The slope of
the curves in the interval of 80 − 200 Mb/J is steep, further R EFERENCES
29
30 indicating that the energy efficiency that HAP and satellites [1] B. E. Y. Belmekki et al., “Unleashing the potential of networked tethered
31 can achieve is lower than 80 Mb/J. When the number of UAVs flying platforms: Prospects, challenges, and applications,” IEEE Open
is 1000, HAPs are more likely to be associated with satellites. Journal of Vehicular Technology, vol. 3, pp. 278–320, 2022.
32 [2] A. Talgat et al., “Stochastic geometry-based analysis of LEO satellite
33 communication systems,” IEEE Communications Letters, vol. 25, no. 8,
pp. 2458–2462, 2021.
34 C. k-Coverage Probability in Military Operations
[3] N. Okati et al., “Downlink coverage and rate analysis of low Earth orbit
35 This subsection investigates the k-coverage probability satellite constellations using stochastic geometry,” IEEE Transactions on
36 when k = 4 during a military operation, that is, the probability Communications, vol. 68, no. 8, pp. 5120–5134, 2020.
[4] Z. Lou et al., “Green tethered UAVs for EMF-aware cellular networks,”
37 that a military target in a given ground region is covered by IEEE Transactions on Green Communications and Networking, vol. 5,
38 at least four NTN platforms simultaneously. The distribution no. 4, pp. 1697–1711, 2021.
39 model and channel model for the three types of platforms in [5] J. Lee et al., “Coverage analysis of LEO satellite downlink
networks: Orbit geometry dependent approach,” available online:
40 NTN are similar to those in the previous two subsections. It https://arxiv.org/abs/2206.09382, 2022.
41 is assumed that the target is distributed in the same spherical [6] C.-S. Choi et al., “Cox point processes for multi-altitude leo satellite
42 cap as the users in subsection V-A. networks,” arXiv preprint arXiv:2301.02469, 2023.
43 Fig. 4 shows the simulation results of 4-coverage probability
[7] R. Wang et al., “Ultra-dense LEO satellite-based communication sys-
tems: A novel modeling technique,” IEEE Communications Magazine,
44 as a function of the SNR threshold. NTN platforms that vol. 60, no. 4, pp. 25–31, 2022.
45 are closer to the target can achieve better communication [8] D.-H. Jung et al., “Performance analysis of satellite communication sys-
46 performance. The average SNR that LAP can provide is 10–
tem under the shadowed-rician fading: A stochastic geometry approach,”
IEEE Transactions on Communications, vol. 70, no. 4, pp. 2707–2721,
47 20 dB, while the SNR of the HAP signal is greater than 5dB. 2022.
48 When the military targets can demodulate signals with SNR [9] Y. Qin et al., “Performance evaluation of UAV-enabled cellular networks
49 over 5 dB, deploying more than 40 HAPs and 1000 LAPs over
with battery-limited drones,” IEEE Communications Letters, vol. 24,
no. 12, pp. 2664–2668, 2020.
50 an area of 1.5−2×105 km2 can achieve 4-coverage for almost [10] A. Al-Hourani, “Session duration between handovers in dense LEO
51 all military targets. satellite networks,” IEEE Wireless Communications Letters, vol. 10,
52 no. 12, pp. 2810–2814, 2021.
[11] K. Belbase et al., “Coverage analysis of millimeter wave decode-and-
53 VI. C ONCLUSION forward networks with best relay selection,” IEEE Access, vol. 6, pp.
54 22 670–22 683, 2018.
55 In this paper, we started with an introductory discussion [12] R. Wang et al., “Stochastic geometry-based low latency routing in
56 about the concepts of NTNs and system-level metrics, and massive LEO satellite networks,” IEEE Transactions on Aerospace and
the motivation to apply SG tools. Then, we discussed suit- Electronic Systems, pp. 1–14, 2022.
57 [13] M. Haenggi, “On routing in random Rayleigh fading networks,” IEEE
58 able SG models for different platforms and compared their Transactions on Wireless Communications, vol. 4, no. 4, pp. 1553–1562,
59 performance. Three typical communication issues, namely 2005.

60
Page 7 of 7

1
2 [14] H. P. Keeler et al., “SINR-based k-coverage probability in cellular
networks with arbitrary shadowing,” in International Symposium on
3 Information Theory. IEEE, 2013, pp. 1167–1171.
4
5
BIOGRAPHIES
6
7
8
9
10
11
12
13
14
15
16 Qi Huang received his B.Eng degree from SWPU, China in 2017, and is
pursuing his master’s degree in KAUST.
17
18
19
20
21
22
23
24
25
26
27 Baha Eddine Youcef Belmekki received the M.Sc. degree in wireless
communications and networking from the USTHB, Algeria, in 2013, and the
28 Ph.D. degrees in wireless communications from INPT, France, in 2020. He
29 is currently a postdoctoral fellow at KAUST.
30
31
32
33
34
35
36
37
38
39 Ahmed M. Eltawil received his Ph.D. degree from UCLA in 2003. He was
40 with UCI since 2015 as a professor and now is a professor of electrical
engineering at KAUST.
41
42
43
44
45
46
47
48
49
50
51 Mohamed-Slim Alouini received his Ph.D. degree in electrical engineering
from the California Institute of Technology, Pasadena, in 1998. He joined
52 KAUST as a professor of electrical engineering in 2009.
53
54
55
56
57
58
59
60