Content uploaded by Sheikh Salman Hassan
Author content
All content in this area was uploaded by Sheikh Salman Hassan on Aug 12, 2024
Content may be subject to copyright.
Semantic Enabled 6G LEO Satellite Communication
for Earth Observation: A Resource-Constrained
Network Optimization
†∗ Sheikh Salman Hassan, †∗ Loc X. Nguyen, ‡Yan Kyaw Tun, ††Zhu Han, and †Choong Seon Hong
†Department of Computer Science and Engineering, Kyung Hee University, Yongin, 17104, Republic of Korea
‡Department of Electronic Systems, Aalborg University, A . C. Meyers Vænge 15, 2450 København
††Department of Electrical and Computer Engineering, University of Houston, Houston, TX 77004-4005, USA
Email: {salman0335, xuanloc088, cshong}@khu.ac.kr, ykt@es.aau.dk, zhan2@uh.edu.
Abstract—Earth observation satellites generate large amounts
of real-time data for monitoring and managing time-critical events
such as disaster relief missions. This presents a major challenge
for satellite-to-ground communications operating under limited
bandwidth capacities. This paper explores semantic communica-
tion (SC) as a potential alternative to traditional communication
methods. The rationality for adopting SC is its inherent ability
to reduce communication costs and make spectrum efficient for
6G non-terrestrial networks (6G-NTNs). We focus on the critical
satellite imagery downlink communications latency optimization
for Earth observation through SC techniques. We formulate
the latency minimization problem with SC quality-of-service
(SC-QoS) constraints and address this problem with a meta-
heuristic discrete whale optimization algorithm (DWOA) and a
one-to-one matching game. The proposed approach for captured
image processing and transmission includes the integration of
joint semantic and channel encoding to ensure downlink sum-
rate optimization and latency minimization. Empirical results
from experiments demonstrate the efficiency of the proposed
framework for latency optimization while preserving high-quality
data transmission when compared to baselines.
Index Terms—6G, satellite and semantic communication, Earth
monitoring, disaster relief, scarce network resources optimization.
I. Introduction
Satellite-based Earth monitoring is critical for applications
such as disaster management, particularly when terrestrial
networks fail, and therefore it offers rapid and comprehensive
coverage of disaster zones. High-resolution sensors enable
land surveys and hazard management, generating vast imagery
data. However, transmitting this data directly for analysis
overwhelms current capabilities due to the sheer volume. For
example, the Landsat system gathers extensive data requiring
transmission, highlighting the need for innovative processing
∗Contribute Equally.
This work has been accepted by the 2024 IEEE Global Communications
Conference (GLOBECOM 2024), ©2024 IEEE. Copyright may be transferred
without notice, after which this version may no longer be accessible. ©2024
IEEE. Personal use of this material is permitted. Permission from IEEE must
be obtained for all other uses, in any current or future media, including
reprinting/republishing this material for advertising or promotional purposes,
creating new collective works, for resale or redistribution to servers or lists,
or reusing any copyrighted component of this work in other works. *Dr. CS
Hong is the corresponding author.
techniques [1]. While complete data transmission offers ad-
vantages in change detection, it’s inefficient for storage and
transmission resources. To address this, upcoming research
explores onboard processing and shifting data processing from
ground stations to satellites [2]. This new workflow promises to
improve downlink efficiency and reduce transmission resource
requirements significantly.
Emerging as a key technology for ubiquitous 6G connec-
tivity, non-terrestrial networks (NTNs) mainly composed of
Low-Earth Orbit (LEO) satellite constellations hold promise in
bridging the digital divide for remote and underserved regions
[3]. However, the economic and efficiency-driven surge in
LEO constellations necessitates further research to address the
inherent limitations of these compact satellites. Specifically,
their constrained onboard memory, communication bandwidth,
and associated data transmission costs pose a significant chal-
lenge despite their economic advantages [4]. To address this,
semantic communication (SC) offers a promising solution by
focusing on transmitting only critical information [5]. By
extracting the essence of the data and eliminating redundancies,
SC significantly reduces the transmitted volume. Furthermore,
SC employs a joint source-channel coding design, making it
inherently more robust against environmental noise compared
to traditional communication methods [6]. Integrating SC into
satellite communication has the potential to alleviate bandwidth
limitations and optimize data transmission. Therefore, this
research proposes a solution to optimize data transmission and
minimize latency, which is a critical parameter in time-sensitive
communication for observational LEO satellites where SC goes
beyond simple compression, focusing on understanding the
data’s meaning and transmitting only crucial elements. This
significantly reduces transmission volume, frees up valuable
resources, and ensures SC quality-of-service (SC-QoS). Ad-
ditionally, SC’s unique joint source-channel coding design
enhances reliability by jointly optimizing the data and the trans-
mission channel, leading to robust communication even in chal-
lenging environments. This work presents a novel approach for
LEO satellites that minimizes semantic communication latency
while guaranteeing SC-QoS requirements. To our knowledge,
arXiv:2408.03959v1 [cs.NI] 31 Jul 2024
this is the first to jointly address these aspects in LEO satellite
communications with the following key contributions:
•We propose a novel architecture for a 6G communication
network utilizing SC-enabled LEO satellites, which opti-
mize resource utilization within constrained environments.
•Building upon this architecture, we formulate a critical
latency minimization problem that ensures both SC and
network QoS.
•To address this mixed-integer non-linear programming
(MINLP) problem, we employ a combined approach.
The meta-heuristic discrete whale optimization algorithm
(DWOA) tackles integer variables, while a one-to-one
matching game efficiently handles binary variables.
•The efficacy of our SC-enabled LEO satellite communica-
tion framework is validated through extensive simulations,
demonstrating superior performance compared to the base-
lines across diverse network configurations.
II. System Model
A. Network Model
As illustrated in Fig. 1, we assume a set Kof KLEO
satellites and a set Iof Iimages captured by each satellite,
all of which are positioned within the LEO constellation.
These LEO satellites can establish a communication link with a
ground terminal (GT), (i.e., gateway) during a specific temporal
interval known as the access window, characterized by the
shared visibility of both the satellite and the gateway, which
is defined as Tk. Furthermore, each LEO satellite is equipped
with a sensing camera designed to capture images of the
Earth’s surface and subsequently extract semantic features from
these images through the utilization of an onboard artificial
intelligence (AI) processor. Each satellite will transmit its
computed semantic features during the access window above
via an established communication link.
B. Semantic Communication Framework
Given the image I∈RH×W×Cwith indexing itaken from
the satellite camera, the semantic encoder will extract the
main features nof the image while eliminating the redundancy
information. We denote the semantic encoder as Sα(·)with the
learning parameter α. With the defined notation, the extracted
features are represented as:
fi=Sα(I).(1)
While the semantic encoder’s role is feature selection, the chan-
nel encoder’s responsibilities are compressing the extracted
feature into a lower dimension and protecting the signal from
the physical noise from the environment. The output of the
channel encoder is denoted as follows:
Xi=Cβ(fi),(2)
where Cβ(·)denotes the channel encoder with learnable param-
eter β. To eliminate or at least minimize the physical noise
under various conditions, the proposed channel encoder uses
the signal-to-noise ratio (SNR) feedback information from the
receiver [6].
At the GT, the receiver will be equipped with two modules:
channel decoder and semantic decoder. These modules will try
to reverse the encoding process on the satellite. The follow-
ing equation can denote the received signal over a wireless
communication link:
ˆ
Yi,k=Xi,kHk+Nk,(3)
where HI,kis the channel between satellites kand the GT, Nkis
the noise from physical environment. In this paper, we assume
that the channel state information (CSI) is known at the GT
(i.e., the receiver), and therefore, the signal can be transformed
as follows: [7]:
ˆ
Xi,k=(HHH)−1HHˆ
Yi,k=Xi,k+ˆ
Nk,(4)
where ˆ
Xi,kdenotes the estimation of the encoded symbols of
image iat the LEO satellite (i.e., transmitter) k. The encoded
symbols will be the input of the channel decoder to eliminate
the noise and also decompress them to recover the extracted
features of the original image:
ˆ
fi,k=C−1
θ(ˆ
Xi,k),(5)
where the C−1
θdenotes the channel decoder with learning
parameter θand ˆ
fi,kis the estimated semantic features of the
image I. Later, the image will be reconstructed by the semantic
decoder using the above semantic features, i.e.,
ˆ
ik=S−1
γ(ˆ
fi,k),(6)
where S−1(·)is the semantic decoder at the GT with the leaning
parameter γ. To train the whole network in an end-to-end
manner, the most common loss is the mean squared error
(MSE) loss between the original and the reconstructed one:
d(i,ˆ
i)=MSE(i,ˆ
i).(7)
To be more specific, this loss will calculate the value difference
between each pixel in both images and force the network to
reproduce the images that are identical to the original image.
The quality of the reconstructed image is evaluated by the peak
SNR (PSNR) metric as follows:
PSNR =10 log10
MAX2
MSE ,(8)
where MAX denotes the highest value of the image pixel per
channel, which is determined 2n−1; for example, each color
channel is presented by eight bits, and the value of nis 8.
C. Semantic Feature Transmission Model
We consider an orthogonal frequency-division multiple ac-
cess (OFDMA) due to its flexibility in allocating limited net-
work resources. This allows satellites to be assigned subcarriers
based on their specific data requirements, efficiently handling
uneven traffic patterns. Additionally, OFDMA inherently avoids
inter-user interference within the spectrum. A set Uof U
subcarriers is considered where a subset of these subcarriers
will be assigned to each LEO satellite k.
Semantic
Feature
Encoder
Channel
Encoder
Earth’s Captured
Image 1
In,1
Xn,1
On-board
Semantic Process
Earth Observation
Satellite
Semantic
Feature
Encoder
Channel
Encoder
Earth’s Captured
Image 2
In,2
Xn,2
Physical Wireless
Channel
Satellite Orbit
Semantic Feature
Decoder
Channel Decoder
Reconstructed Earth’s
Captured Images
𝑰n,K
Xn,K
Ground Gateway -
Semantic Destination
Feature
Vector
Transmission
Satellite-Gateway Access Window - Association Time
Gateway
Processing for
Image
Reconstruction
t1t2
t
Fig. 1: The architectural framework for facilitating multi-satellite semantic communication in Earth observation for disaster relief systems.
The downlink rate from the LEO satellite kto the GT within
subcarrier uto transmit semantic feature information ˆ
fi,kas:
Ru
k(γk)=
U
X
u=1
γk,uBu
klog21+ Γk,(9)
where Bkis the subcarrier bandwidth, γk=[γk,1,· · · , γk,U]
denotes the subcarrier assignment tuple for each LEO satellite
k, which can be defined as:
γk,1={0,1},(10)
where γk,u=1denotes that subcarrier uis utilized by LEO
satellite k, and γk,u=0represents otherwise. We also provide
the condition that only one subcarrier uis assigned to each
LEO satellite at each time slot to avoid network interference:
K
X
k=1
γk,u≤1.(11)
Similarly, each LEO satellite kcan utilize at most a single
subcarrier, which is given as follows:
U
X
u=1
γk,u≤1.(12)
Moreover, the SNR Γbetween LEO satellite kand the GT can
be defined as follows:
Γk=GkGGTPkc
4πdkfcN02,(13)
where Guand GGT denote the wireless power antenna gains for
the LEO satellite and GT respectively. Pkis the LEO satellite
transmit power, cindicates the speed of light, fcdenotes the
carrier frequency, dkdenotes the distance between the LEO
satellite kand the GT, and N0denotes the spectral density of
the noise power to capture the interference generated by the
LEO satellites in other constellations which utilize the same
bandwidth spectrum and subcarrier u.
Based on the semantic features obtained in (1) and the
achievable data rate calculated in (9), we can calculate the
latency of transmitting the features from LEO satellite kin
subcarrier uto the GT is:
t(lX,γk)=lX
Ru
k
,(14)
where lXdenotes the length of the transmitted data, and this
length can be reduced or increased by adjusting the compres-
sion ratio. Extended signals provide receivers with additional
information about channel noise, enhancing their ability to
reconstruct the original signal accurately. Therefore, there will
be a trade-offbetween the transmission latency and the quality
of the reconstructed image.
III. Latency Minimization Problem Formulation
Leveraging the aforementioned system model, we aim to
minimize the average latency for transmitting semantic features
from LEO satellites. However, this optimization must adhere
to the network model’s practical constraints. Specifically, suf-
ficient semantic information must be transmitted within the
access window to allow the GT to reconstruct the original
image. Additionally, reliability and SC-QoS must be ensured
by guaranteeing a minimum PSNR threshold that meets the
desired image recovery quality. Under these assumptions, the
average latency can be given as:
O(lX,γk)=1
KX
k∈K
t(lX,γk).(15)
The latency minimization problem can be formulated as:
P1: min
lXI,γk
O(lXI,γk)(16a)
s.t. γk,1={0,1},∀k∈ K,∀u∈ U,(16b)
U
X
u=1
γk,u≤1,∀k∈ K,(16c)
K
X
k=1
γk,u≤1,∀u∈ U,(16d)
E(PSNRk(X)) ≥Ψk,∀k∈ K,(16e)
0≤γktk≤Tk,∀k∈ K,(16f)
where constraints (16b), (16c), and (16d) guarantee that the
ground station will assign only a single subcarrier to each LEO
satellite, and every subcarrier could only be utilized by a single
LEO satellite at each time slot for semantic features trans-
mission. Constraint (16e) ensures that the semantic features
should be reliable and meet the SC-QoS threshold from each
LEO satellite k. Constraint (16f) ensures the communication
efficiency of LEO satellites within the access window, i.e.,
the coverage period of the GT. It can be observed that the
formulated problem is MINLP which is difficult to solve due
to NP-hard. To address this problem, we provide a solution
approach in the next section.
IV. Proposed Algorithm
To address the formulated MINLP, we utilize the block
coordinated descent (BCD) approach, where we decomposed
the main problem in (16) into two subproblems according to
the nature of decision variables and then solved iteratively.
A. Image Reliable Retrieval (SC-QoS) Problem
In this subproblem, we minimize the total transmission
latency while ensuring the quality of the received images
by determining the transmitted length of the data. Given the
feasible value γkof subcarrier assignment, the problem can be
decomposed as follows:
P1.1: min
lX
O(lX,γk)(17a)
s.t.(16e)and (16 f).(17b)
The compression ratio affects not only the transmission latency
but also the quality of the reconstructed images. As shown
in (16e), the GS demands different service quality from each
satellite, which results in different transmitted lengths for each
satellite. To tackle the challenge of integer variables in problem
P1.1, we leverage the meta-heuristic discrete whale optimiza-
tion algorithm (DWOA) [8]. DWOA adapts the operators of
the standard WOA to work effectively with integer decision
variables. This process involves initializing a population of
Algorithm 1 DWOA for SC-QoS Problem
1: Input: N: number of searching agents, CRS: compression
ratio set, K,Ψkand the access window Tk,γk, the maximum
number of iterations MaxIT.
2: Output: The optimal transmitted length l∗
Xand the com-
munication latency.
3: Initialize the whale population, lX=[ln1,xn2, ..., xnk ]∈ CRS
(n=1,2, ..., N). Calculate the communication latency of
each agent and determine the optimal decision l∗
X,τ←0.
4: while τ≤MaxIT do
5: for n←1to N(each searching agent) do
6: Update vector a, A, C, m and p.
7: if p<0.5then
8: if |A|<1then
9: Update the lXby the first sub-equation of (18).
10: else
11: Update the transmitted length by (19).
12: end if
13: else if p≥0.5then
14: Update the lXby the second sub-equation of (18).
15: end if
16: end for
17: Calculate the transmission latency of each agent and
update l∗
Xif there is a better transmission length.
18: τ←τ+1.
19: end while
20: Output: Optimal transmitted length l∗
X.
candidate integer solutions within the feasible search space.
Subsequently, DWOA iteratively refines these solutions through
exploration and exploitation phases, which are specifically
adapted to handle discrete values. The exploitation can be
expressed by the following equations:
lX(τ+1) =
l∗
X(τ)−
A·
D,p<0.5,
D′·ebm ·cos(2πm)+
l∗
X(τ),p≥0.5,
(18)
where
l∗
X(τ)denotes the best solution at the current iteration τ.
pis a random variable, which is used to decide between two
hunting behaviors of whales. On the other hand, the searching
agents explore other solutions by updating their location toward
a random agent in the population and can be expressed as:
lX(τ+1) =
lXrand (τ)−
A·
D.(19)
Each candidate’s fitness is evaluated based on the objective
function. In addition, to deal with the constraints, we integrate
them into the objective function as a punishment value if vio-
lated. The algorithm then iteratively updates solution positions
and selects the best solution found so far, continuing until a pre-
defined stopping criterion is met. This approach allows DWOA
to search for high-quality solutions to problems with integer
constraints efficiently, which is described in Algorithm 1.
B. Subcarrier Allocation to LEO Satellite
This subproblem optimizes the binary optimization part of
the P1 to address the subcarrier allocation to each LEO satellite
ufrom GT. Similarly, by having the optimal PSNR solution X∗
I
from the subproblem in (17), we can decompose the subcarrier
assignment problem as follows:
P1.2: min
γk
O(l∗
X,γk)(20a)
s.t.(16b),(16c),(16d),and (16 f),(20b)
To solve this binary optimization problem, we utilize a match-
ing game for the subcarrier allocation to the LEO satellite.
According to constraints (16b) and (16c) in problem P1.2, each
subcarrier can be allocated to at most one LEO satellite, and
each LEO satellite can only have one subcarrier allocated to
it. Therefore, we reformulate the subcarrier allocation problem
as a two-sided one-to-one matching game. We first define a
one-to-one matching game for the subcarrier allocation to the
LEO satellite at the GT.
Definition 1. Consider two disjoint sets of players, Kand U.
The two-sided one-to-one matching game ς:K → U for the
subcarrier allocation is defined as:
1) ς(u)⊆ K and |ς(u)|∈{0,1},∀u∈ U;
2) ς(k)⊆ U and |ς(k)|∈{0,1},∀k∈ K ;
3) k=ς(u)⇔u=ς(k),∀u∈ U,∀k∈ K .
Here, |ς(.)|is the cardinality of the matching outcome ς(.),
where the outcome of the matching is the allocation mapping
between a set Kof LEO satellites and Uof subcarriers.
Furthermore, conditions (1) and (2) in the definition guarantee
that one subcarrier can be allocated to at most one satellite
at a time and that one satellite can only have one subcarrier
allocated to it. Finally, condition (3) ensures that if LEO
satellite kis matched with subcarrier uthen subcarrier umust
also be matched with LEO satellite k.
Preference of players: Matching is carried out based on the
preference profiles constructed by the LEO satellites and the
GT over subcarriers to prioritize potential pairings based on
their information. To define the matching game, let us denote
≻kand ≻uthe preference profiles of LEO satellite k∈ K
and subcarrier u∈ U. Furthermore, let ζk(u)and ζu(k)be the
preference functions of LEO satellite k∈ K for subcarrier uand
subcarrier u∈ U for LEO satellite k, respectively. Note that the
preference function of each satellite must satisfy the constraint
(16f), therefore the preference function of LEO satellite kfor
subcarrier uis given by:
ζk(u)=ϱk(u)Ru
k(γk),(21)
where ϱk(u)is an indicator function with the value of 1if
constraint (16f) for LEO satellite kis satisfied and 0otherwise.
Each LEO satellite prefers to be allocated to the subcarrier that
provides the highest communication rate. For instance, u≻ku′
implies that LEO satellite kprefers subcarrier uover subcarrier
u′to be allocated, i.e., ζk(u)> ζk(u′). Similarly, we can define
the preference function of subcarrier ufor LEO satellite kas:
ζu(k)=Ru
k(γk).(22)
Algorithm 2 One-to-One Matching for subcarrier Allocation
1: Initialization: K,U,P,Kun=K,Uk=U,∀k∈ K , a set of
LEO satellites requested to subcarrier u,Ku,req =∅, and a set
of rejected LEO satellites from subcarrier u,Ku,rej,∀u∈ U;
2: LEO satellite kdesigns the preference list ≻kwith (21);
3: while Pk∈K Pu∈U qku ,0do
4: for k=1to |Kun |do
5: Find u=argmax
u∈U
ζk(u).
6: Make a request to the GT by setting qku =1.
7: end for
8: for u=1to Udo
9: Update Ku,req ← {k:qku =1,∀k∈ K} .
10: Construct the preference of GT for its available sub-
carrier according to (22).
11: Find k=argmax
k∈K
ζu(k).
12: Allocate subcarrier uto LEO satellite k.
13: Update Ku,rej ← {Ku,req \k}.
14: Update Uk← {Uk\u}∀k∈ Ku,rej .
15: end for
16: Update Kun ← Kun ∩ {K1,rej ∪ · · · ∪ KU,rej }.
17: end while
18: Output: Optimal subcarrier allocation γ∗.
Algorithm 3 The sequence of overall Proposed Scheme
Step 1: Given the feasible value γkof subcarrier assignment,
we run DWOA to determine the length of transmitted signals.
Step 2: With the temporary transmitted lengths from
DWOA, we optimize the subcarrier for each satellite by the
One-to-One Matching from Algorithm 2.
Step 3: Finally, we re-run the DWOA to determine the
final transmitted lengths for each satellite given the optimal
subcarrier.
This preference reflects that the GT desires to match each
subcarrier uto a LEO satellite kthat achieves the maximum
communication rate on that subcarrier to maximize the total
communication rate which ultimately minimizes the latency,
which is the objective function of P1.
Definition 2. A stable matching ς∗is achieved if there is no
blocking pair (k,u), where a pair (k,u)is a blocking pair when
u<ς(u),k<ς(k), and u≻kς(u)and k≻uς(k).
Since the proposed game is implemented exactly like the
standard deferred acceptance algorithm [9], it guarantees a
stable matching. The output of the algorithm is the optimal
subcarrier assignment γ∗. The pseudocode of the proposed one-
to-one matching game-based subcarrier assignment algorithm
is shown in Algorithm 2. The overall solution approach is
summarized in Algorithm 3.
TABLE I: Simulation Parameters
Notation Definition Value
fcCarrier Frequency [20 −30] GHz
BuBandwidth 500 MHz
GuLEO Satellite Antenna Gain 33.13 dBi
TkAccess Window 60 Seconds
dLEO Satellite Altitude 786 km
N0Noise Power −43 dB
GGT Ground Terminal Antenna Gain 34.2 dBi
puLEO satellite Transmit Power 10 W
Torb Orbital Period 100 min
CRS Compression Ratio Set [ 4
128 ;5
128 ... 11
128 ;12
128 ]
Fig. 2: Communication performance evaluation.
V. Simulation Results Evaluation
To validate the proposed joint DWOA and matching game
scheme for SC in GS-LEO networks, we conduct simulations
and results analyses. We establish the simulation environment
and introduce benchmark schemes for comparison. Subse-
quently, we evaluate the performance of the proposed scheme.
The main parameters are summarized in Table I. To evaluate
the effectiveness of SC-enabled GS-LEO communication, we
compare our proposed scheme against the following bench-
marks:
•OnlyWhale: In this scheme, the GS solely employs the
DWOA to optimize the transmit length for guaranteeing
the PSNR requirements. While the resource block is
randomly allocated to each satellite.
•MatchingOnly: Here, the GS utilizes only a matching
scheme to allocate each resource block for the satellites,
while a greedy algorithm randomly searches the transmis-
sion length of the data.
•Random: This baseline randomly assigns the resource
blocks to the satellites and the transmitted length is
achieved by the greedy algorithm.
Fig. 2 depicts the communication latency comparison of the
proposed joint DWOA and matching game scheme against
three baseline approaches. The objective is to minimize la-
tency while ensuring both SC and network QoS requirements.
As evident from the figure, the proposed algorithm achieves
superior performance in terms of communication latency for
the required transmission. In simpler terms, the proposed
scheme transmits data with lower latency compared to the
Proposed WhaleOnly MatchingOnly Random
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
Compression Ratio
The Compression Ratio for Tranmsitting Data
Fig. 3: Compression ratio comparison.
baselines while still meeting the necessary SC and network
QoS constraints.
Fig. 3 shows the average compression ratio of the satellites
for all the scenarios. The proposed scheme achieves the lowest
compression ratio, which equivalently indicates the lowest
transmit length while satisfying the requirement of QoS for
each satellite. Compared to the baseline schemes, the pro-
posed scheme exhibits a significant advantage. The WhaleOnly
scheme achieves a relatively large compression ratio perfor-
mance due to insufficient resource allocation. Therefore, it
suffers from higher communication latency as shown in Fig. 2.
The MatchingOnly and Random perform poorly in the problem
of determining the transmitted length, which leads to a high
compression ratio. In conclusion, the proposed scheme offers
a superior trade-offbetween image quality and communication
efficiency.
Fig. 4 presents original and reconstructed images from the
FloodNet dataset [10], which offers high-resolution unmanned
aerial system (UAS) imagery with comprehensive semantic
annotations detailing post-disaster damage. This showcases the
generalizability of our semantic model to satellite datasets,
as evidenced by its effectiveness on UAS imagery. It can be
observed that the semantic communication model has achieved
an outstanding visual result for different compression ratios,
which makes it difficult to spot the difference by the human
eyes. However, with the PSNR metric, there is a significant
difference in quality for those compressions. To be specific,
the PSNR and MS-SSIM values are [27.23, 0.84], [28.42,
0.89], [29.34, 0.91] corresponding to the compression ratios:
4/128, 8/128, 12/128, respectively. These results have effec-
tively demonstrated the effect of transmitting lengths on the
performance of the receiver. The longer the transmitted signals
are transmitted, the better the receiver performs in capturing
the channel noise and constructing more quality images.
Fig. 5 further validates the effectiveness of our proposed
algorithm by assessing its performance in achieving PSNR
requirements for individual satellites under network dynamics.
Unlike some baseline algorithms that exhibit inconsistent per-
formance across satellites, the proposed scheme demonstrates
(a) Original Image (b) Compression Ratio: 4/128 (c) Compression Ratio: 8/128 (d) Compression Ratio: 12/128
Fig. 4: The original image and the reconstruction images with different compression ratios under the SNR =3dB.
12345678910
Satellite Number
0
5
10
15
20
25
30
35
40
PSNR
PSNR Results for Different Algorithms
Requirement
Proposed
WhaleOnly
Matching
Random
Fig. 5: PSNR evaluation.
consistent results. It successfully meets or comes very close
to the required PSNR for most satellites in various network
conditions. This consistency highlights the robustness of our
approach in handling diverse PSNR demands across the net-
work while ensuring quality. Additionally, the proposed scheme
guarantees that the PSNR threshold is never violated.
VI. Conclusion
This work presented a novel approach for Earth observa-
tional data transmission using an SC-enabled LEO satellite
network for GS communication. The proposed framework
leverages multiple LEO satellites with onboard processing
capabilities. These satellites extract critical semantic features
from captured images and transmit them to the GS, enabling
reconstruction of the original image. This strategy achieves
efficient spectrum utilization and significantly reduces net-
work latency. Consequently, it overcomes the limitations of
traditional LEO satellite communication for Earth observation
applications, where high latency is particularly detrimental due
to the large data volumes involved. Furthermore, extensive
simulations demonstrate the superior performance of the pro-
posed algorithm compared to baseline approaches, validating
its effectiveness.
References
[1] “U.S. geological survey: What are the landsat collection 1
level-1 data product file sizes?” https://www.usgs.gov/faqs/
what-are- landsat-collection- 1-level-1- data-product- file-sizes, 2022.
[2] V.-P. Bui, T. Q. Dinh, I. Leyva-Mayorga, S. R. Pandey, E. Lagunas,
and P. Popovski, “On-board change detection for resource-efficient earth
observation with LEO satellites,” in Proc. of the IEEE Global Commu-
nications Conference, Kuala Lumpur, Malaysia, Dec. 2023.
[3] S. S. Hassan, Y. M. Park, Y. K. Tun, W. Saad, Z. Han, and C. S.
Hong, “SpaceRIS: LEO satellite coverage maximization in 6G sub-THz
networks by MAPPO DRL and whale optimization,” IEEE Journal on
Selected Areas in Communications, Feb. 2024, early access.
[4] R. Xie, Q. Tang, Q. Wang, X. Liu, F. R. Yu, and T. Huang, “Satellite-
terrestrial integrated edge computing networks: Architecture, challenges,
and open issues,” IEEE Network, vol. 34, no. 3, pp. 224–231, Mar. 2020.
[5] L. X. Nguyen, Y. L. Tun, Y. K. Tun, M. N. H. Nguyen, C. Zhang,
Z. Han, and C. S. Hong, “Swin transformer-based dynamic semantic
communication for multi-user with different computing capacity,” IEEE
Transactions on Vehicular Technology, Feb. 2024, early access.
[6] D. B. Kurka and D. G¨
und¨
uz, “Deepjscc-f: Deep joint source-channel
coding of images with feedback,” IEEE Journal on Selected Areas in
Information Theory, vol. 1, no. 1, pp. 178–193, Apr. 2020.
[7] H. Xie, Z. Qin, and G. Y. Li, “Task-oriented multi-user semantic com-
munications for vqa,” IEEE Wireless Communications Letters, vol. 11,
no. 3, pp. 553–557, Dec. 2021.
[8] Y. Li, Y. He, X. Liu, X. Guo, and Z. Li, “A novel discrete whale opti-
mization algorithm for solving knapsack problems,” Applied Intelligence,
vol. 50, no. 10, pp. 3350–3366, Jun. 2020.
[9] A. E. Roth, “Deferred acceptance algorithms: History, theory, practice,
and open questions,” International Journal of Game Theory, vol. 36,
no. 3, pp. 537–569, Jan. 2008.
[10] M. Rahnemoonfar, T. Chowdhury, A. Sarkar, D. Varshney, M. Yari, and
R. R. Murphy, “Floodnet: A high resolution aerial imagery dataset for
post flood scene understanding,” IEEE Access, vol. 9, pp. 89644–89 654,
Jun. 2021.
