1
Toward the Internet of Space Things: Performance Analysis of LEO Satellite Relay Networks using mmWave and sub-THz links
arXiv:2605.02061v1 [cs.NI] 3 May 2026
Sergi Aliaga* , Student Member, IEEE, Ahmad Masihi* , Student Member, IEEE, Vitaly Petrov, Member, IEEE, Marc Sanchez Net, Member, IEEE, and Josep M. Jornet, Fellow, IEEE Abstract—As the commercial space economy expands, existing ground-based infrastructure faces severe bottlenecks in supporting the data-intensive continuous connectivity needs of nextgeneration “space users,” including CubeSats, space data centers, and more. Even when utilizing existing Ku-band ground relay networks, the contact time with a CubeSat at low-Earth orbit (LEO) is often still limited to minutes per day only. This paper analyzes an alternative system design that leverages emerging high-rate millimeter-wave (mmWave) and sub-terahertz (subTHz) inter-satellite links to build a high-throughput and highavailability satellite-based relay backbone for space vehicles. To evaluate this concept, we develop a comprehensive mathematical framework that jointly incorporates complex time-variant orbital dynamics and mmWave/sub-THz link characteristics. We then derive the key performance indicators, including contact probability, channel capacity, and energy efficiency. The numerical results, cross-verified by computer simulations, demonstrate that such systems can provide improvements of up to several orders of magnitude compared to existing networks of ground stations. Notably, we identify a fundamental bound on download capacity and show that continuous 24/7 connectivity becomes achievable with only ten LEO relay satellites. These findings establish mmWave and sub-THz satellite relay networks as a promising, scalable, and energy-efficient solution, thus unlocking improved connectivity with various space vehicles of tomorrow.
I. I NTRODUCTION The satellite communications landscape has evolved from a few Geostationary (GEO) satellites providing basic relay services into an ecosystem of massive low-Earth orbit (LEO) constellations comprising thousands of interconnected nodes. While legacy GEO systems suffered from high latency due to their 36,000 km altitude [2], today’s LEO mega-constellations operate between 400–2,000 km, slashing latency to under 10 ms and delivering fiber-like broadband performance [3]. Meanwhile, another ongoing trend in the industry to address continuously growing performance demands is a relatively slow but persistent adoption of new wider frequency bands for the satellite-to-ground and inter-satellite links. While selected millimeter-wave bands (mmWave, such as Ka-band and Ku-band [3]) are already widely used in stateof-the-art systems [4], other candidate bands are also receiving growing attention from academia and industry. These include but are not limited to sub-terahertz (sub-THz, 100 GHz– 300 GHz) [5], terahertz (THz, 300 GHz–3 THz) [6], [7], and free-space optics (FSO, 190 THz–230 THz) [8]. These emerging LEO constellations utilizing wideband wireless links S. Aliaga, A. Masihi, and J. M. Jornet are with Northeastern University, Boston, MA, USA. Email: {aliaga.s, masihi.a, j.jornet}@northeastern.edu V. Petrov is with KTH Royal Institute of Technology, Stockholm, Sweden. Email: [email protected]. M. S. Net is with NASA Jet Propulsion Laboratory at California Institute of Technology, Pasadena, CA, USA. Email: [email protected]. * S. Aliaga and A. Masihi are co-first authors. The work by V. Petrov has been supported by Vinnova 6GSTAR project. The work by M. S. Net was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration (80NM0018D0004). A shorter version of this work has been presented at Asilomar 2024 [1].
are expected to become an essential component of forthcoming non-terrestrial networks (NTNs) – integrated systems of satellites, airborne, and terrestrial nodes – which serve as a cornerstone for 6G and beyond by promising ubiquitous connectivity across land, sea, and air. A. A vision of better connectivity for space users Today, the overwhelming majority of research efforts on NTNs is focused on providing better services for terrestrial nodes (e.g., smartphones and cars), marine traffic (various types of vessels), and airborne users (e.g., planes and UAVs) [9]. However, there exists another group of NTN beneficiaries – currently small in volume but extensively growing – emerging “space users”, which are the main focus of this work. These users primarily represent various LEO satellites owned by companies, research centers, and government bodies, but can ultimately also include orbital data centers, lunar landers, and space telescopes, among others [10]. Such a population of space users is expanding rapidly; as of 2026, there are approximately 15,000 active payloads in orbit, a nearly fourfold increase from just five years ago [11]. Further, modern space vehicles can carry advanced imaging payloads generating terabytes of data per mission [12], yet current Ground Data Systems (GDSs) cannot accommodate this exponential demand. Even NASA’s Deep Space Network (DSN) faces projected capacity shortfalls as client requirements escalate [13]. This bottleneck stems from both spectrum exhaustion in the Ku-band and ground station scheduling conflicts, with access wait times reaching hours or days. We aim to address this limitation in the present study by exploring the following question: Why not leverage existing and future high-rate LEO satellite communication networks to also serve space-based users? First, such advanced networks, illustrated in Fig. 1, can relay data to the backbone with minimal disturbance, ensuring timely delivery to mission centers and scientific teams. Aligning with 3GPP standards for NTNground integration [14], adopting standardized SIM-equivalent radio interfaces would simplify space user hardware, thereby reducing costs and weight and potentially improving scalability, capacity, and energy efficiency. Beyond technical gains, this NTN-centric approach leverages existing commercial infrastructure investment, allowing constellations deployed for terrestrial markets to serve space users without dedicated relay satellites. The resulting space-to-NTN traffic imposes minimal incremental load compared to terrestrial demand, facilitating efficient spectrum sharing—especially over oceans or unpopulated regions where ground user density is low. Finally, properly configured constellations provide continuous coverage, eliminating the scheduling conflicts and communication gaps inherent in sparse ground station networks.
2
Envisioned New Use Cases
Scientific telescopes Lunar landers
Orbiting stations
Research & academic small satellites
Space tourism
Ubiquitous coverage
In-flight connectivity Remote users
Existing Use Cases
Direct-todevice
Military operations
Satellite IoT
Internet backhaul
Rescue operations
Existing:
Ground & airborne links
Inter-satellite links
New:
Space user links
Fig. 1: Envisioned scope of 6G+ NTN connectivity to support space users (scientific telescopes and missions, space tourism, lunar landers, and orbital stations, etc.), alongside traditional terrestrial and airborne users. B. Related work In principle, the idea of using satellites for space data relaying is not entirely new and has been explored before through several successful infrastructures. For instance, NASA’s Tracking and Data Relay Satellite System (TDRSS) has utilized GEO satellites for over 30 years to provide Ku-band coverage, with ongoing plans to integrate optical communications [15]. Similarly, the China National Space Administration (CNSA) employs lunar relay satellites in the S- and Ku-bands [16], while the Mars CubeSat One (MarCO) mission recently demonstrated the first use of CubeSats to relay scientific data from Mars during critical mission phases [17]. These early successes have motivated numerous studies on relay architectures for near-Earth and deep-space missions. Wan et al. [18] modeled small solar system relay constellations, while Modenini et al. [19] proposed relays at Lagrange points and Cheung et al. [20] studied heliocentric orbits to mitigate deep-space ground infrastructure and optical conjunction challenges. Elewaily et al. [21] surveyed Delay Tolerant Network (DTN) frameworks as a solution for link disruptions, though these are typically limited to < 100 nodes and have not been evaluated for massive next-generation NTNs. Most of these works focus on dedicated relay constellations operating below 100 GHz and orbiting specific planetary bodies. Palermo et al. [22] considered an Earth-orbiting relay system for space users, but it relied on dedicated architectures operating in the Ku-band. Notably, numerous works explore beyond-100 GHz technology in next-generation NTNs [23], but they predominantly focus on ground or airborne users. Al-qaraghuli et al. [24] proposed a dual Sub-THz/Ka-band system to provide highrate Internet to ground users via massive LEO constellations, while Nie et al. [25] characterized THz inter-satellite links for ultra-high throughput between network nodes. For airborne coverage, literature includes detailed THz channel models [26], stochastic geometry frameworks for satellite-toairplane downlinks [27], and broad reviews of THz aerospace
opportunities [28]. However, these studies ignore the specific peculiarities (e.g., mobility and orbital dynamics) of the growing space users segment. C. Claims and contributions Despite the intuitive advantages of serving space users through (sub-)THz-enabled NTNs ( [29], among others), a comprehensive methodology to quantify the performance boundaries for space users served by mmWave/sub-THz-capable NTNs remains missing. Hence, by extending our preliminary work in [1], we aim to address this gap. The main contributions of this study are thus summarized as follows: • Flexible Mathematical Framework: An elaborate framework is developed to quantify NTN-based space user capacity. The developed framework takes into account all the major static and time-variant orbital parameters, the satellites’ mutual orientation and mutual mobility, as well as the essential radio link parameters. • In-depth Numerical Analysis: A thorough investigation is performed using the developed methodology of the benefits of utilizing existing NTN architectures to serve space users with state-of-the-art mmWave and sub-THz technology. The study highlights time-variant, statistical, and time-averaged Key Performance Indicators (KPIs), including the probability of contact, total download capacity, and energy efficiency. The study particularly emphasizes the deployment configurations and orbital parameters that maximize the outlined KPIs. • Comprehensive Simulation Study: The results delivered with the mathematical framework are further crossverified by system-level computer simulation. We utilize current Ku-band ground data systems as a baseline to reveal substantial improvements in the outlined KPIs with the considered concept. The remainder of this paper details the system model and KPIs in II, derives the mathematical framework in Sec. III, and presents the numerical results and conclusions in Sec. IV and V, respectively
3
II. S YSTEM M ODEL In this section we present the foundational assumptions underlying our model, encompassing the considered system architecture, the adopted propagation and network models, and the key performance metrics of interest. All variables and parameters introduced throughout this work are summarized in Table I for reference. A. Deployment scenarios The considered deployment scenarios, depicted in Fig. 2, involve a LEO CubeSat at an altitude hCS ∈ [400, 2000] km above Earth (RE ). The CubeSat requires a reliable high-rate downlink for scientific data, including measurements, imagery, and telemetry. We evaluate four scenarios to characterize downlink capacity, temporal dynamics, and total data volume, comparing traditional GS-based baselines with emerging NTN mega-constellations. These architectures include: 1) Single Ground Station: The CubeSat communicates directly with a single terrestrial Ground Station (GS) (NG = 1). This represents the baseline architecture, TABLE I: Summary of the notation used. Variable ς k R E , ME G µ PTx , PRx , PN GCS , GRx fc W d(t) L, Lspr , Labs Q, p T0 , Tatm κ r Ta , Tsys FN ρ Tc Tday , T Tres NS , NG NS∗ ω R, h dmin , dmax dB V f, F i Ω M x, y, z φ, Λ β θ, θth τ, ξ u(t) Q C Γ η
Description Physical constants Speed of light Boltzmann’s constant Earth radius and mass Gravitational constant Earth’s standard gravitational parameter Radio and propagation Transmitted, received, and noise power CubeSat and Receiver antenna gain Carrier frequency Bandwidth Distance between CubeSat and receiver Path loss, spreading loss and absorption loss Atmospheric composition and pressure Reference and atmospheric temperature Molecular absorption coefficient propagation path range Antenna and total system noise temperature Receiver noise figure Signal-to-noise ratio System model Contact time between CubeSat and receiver Earth day and scenario repetition period Temporal resolution used for the calculation of T Number of NTN relays and GSs Min. number of NTN relays for no outage Angular velocity Orbital radius and altitude (R = Re + h) Minimum and maximum possible link distances Link blockage distance (blockage by the Earth) Visibility indicator function Probability and cumulative density functions Orbital inclination Right Ascension of the Ascending Node (RAAN) Conversion matrix between 2-D and 3-D systems Orbital coordinates in the 3-D coordinate system Ground station latitude and longitude Rotation of the Earth w.r.t. 3D coordinate system GS elevation angle and Min. GS elevation angle Simple roots of g(t)=dGS (t)−dB GS and dGS (t)−d Unit-step function Performance metrics Contact probability Channel capacity Download capacity over 24h Energy efficiency
where connectivity is limited to the visibility windows of one fixed GS. 2) Multiple Ground Stations: The CubeSat gets served by a ground relay network of NG > 1 GSs. This scenario represents another typical approach, where a mission relies on a network of geographically distributed GSs (typically, third-party) to improve coverage and latency at the cost of increased operational expenses. 3) Single NTN Relay: The CubeSat sends its data through a single NTN satellite relay (NS = 1), which then communicates with the Satellite Service Provider (SSP) ground infrastructure. This configuration facilitates direct comparison of one GS (Scenario 1) vs. one NTN relay. 4) Multiple NTN Relays: This architecture generalizes the previous single-relay scenario by employing NS LEO relays in the same orbit. Similar to multi-GS Scenario 2, this setup is expected to provide increased spatial and temporal diversity, more frequent access opportunities, and enhanced reliability. For clarity and tractability of our first-order analysis, we assume that the NTN relays’ orbit is co-planar to the CubeSat orbit. The rationale here is that typical satellite constellations following a Walker-Delta configuration are composed of multiple orbital planes to provide uniform global coverage [30]. Due to the high density of these planes, a CubeSat’s orbit remains approximately co-planar with that of the nearest relay satellites most of the time, even with orbital precession. This justifies the selected co-planar deployment as a tractable yet representative approximation of the real system dynamics. As illustrated in Fig. 2, we adopt an Earth-centered, righthanded Cartesian coordinate system where the z-axis aligns with Earth’s rotational axis and the x-axis points toward the vernal equinox. The CubeSat and NTN relays occupy circular LEO orbits in the same plane at altitudes hCS and hNTN , with Earth-center distances RE +hCS and RE +hNTN , respectively. Since feeder links are assumed to be seamlessly integrated with terrestrial infrastructure, their detailed modeling is omitted. While Fig. 2 depicts the CubeSat at a lower altitude for clarity, the model applies to any orbital configuration. Relative positions are defined by orbital inclination i and Right Ascension of the Ascending Node (RAAN) Ω. Time-varying communication distances dGS and dNTN drive the scenario dynamics as satellites move along their orbits. B. Propagation and routing assumptions We model the wireless signal propagation using the Friis transmission equation, which describes the received power at a GS or NTN relay as: PRx =
PTx GCS GRx , L(fc , d(t))
(1)
where PTx is the CubeSat’s transmit power, GCS is its antenna gain, and GRx denotes the gain of the receiving node (either a GS or NTN relay). Mutual antenna alignment between the transmitter and receiver is assumed. The channel path loss L(fc , d(t)) depends on the carrier frequency fc and the time-varying distance d(t) between the CubeSat and the receiver.
4
)
b) Instantaneous channel capacity, C(t) This time-varying metric is derived using Shannon’s capacity formula: PRx (t) , (6) C(t) = W log2 (1 + ρ(t)) = W log2 1 + PN
2+/-%
CubeSat
&"#""
Ground Station
&"#"!
&&%"
&&%! .'(
%
!! + ℎ"#" (
NTN Relay !! + ℎ$%
'
$
Equatorial Plane
Fig. 2: Modeling the coverage of a CubeSat (space user) using NTNs, compared to GS-based service. For the CubeSat-to-NTN relay link, the propagation occurs entirely in space. As a result, no atmospheric absorption is encountered, and the only relevant attenuation mechanism is free-space spreading loss, modeled 2 as: 4πd(t)fc , (2) Lspr (fc , d(t)) = ς where ς is the speed of light. In contrast, the CubeSat-to-GS link traverses the atmosphere, where molecular absorption may occur. We model the corresponding molecular "Z absorption loss as [31]: # d(t)
Labs (fc , d(t)) = exp
κ(fc , Q(r), p(r), Tatm (r)) dr , (3) 0
where κ is the frequency-dependent molecular absorption coefficient, influenced by the atmospheric composition Q(r), pressure p(r), and temperature T (r) along the path r. These atmospheric quantities are obtained from ITU-R Recommendation P.835 [32]. Therefore, the total path loss for the CubeSat-to-GS link is modeled as: L(fc , d(t)) = Lspr (fc , d(t))Labs (fc , d(t)). (4) To analyze the capacity of relaying data from the CubeSat to a multi-relay NTN system, we assume that the CubeSat always transmits to the nearest visible relay whenever possible. A similar assumption is made for CubeSat-to-GS links, where the CubeSat always communicates with the nearest GS in line of sight. C. Metrics of interest In this study, we evaluate upper bounds for system-level performance, accounting for both wireless technology and architectural alternatives. We primarily focus on the following KPIs: a) Contact probability, Q Defined as the probability that the CubeSat is in line of sight (LoS) with either a GS or an NTN relay at a given time. It is calculated as the ratio between the total contact duration between CubeSat and the GS or NTN relay, Tc , and the scenario repetition period, T : Tc Q= . (5) T A further detailed discussion and formulation for Tc and T is provided in Sec. III.
where W is the system bandwidth, ρ(t) is the instantaneous signal-to-noise ratio (SNR), PRx (t) is the received power at time t, and PN is the system noise power. The received power varies primarily due to changes in the distance d(t) caused by the CubeSat’s orbital motion. The receiver noise power is calculated as PN = kTsys W , where k is the Boltzmann constant, and Tsys is the system noise temperature. Accordingly, the total system noise temperature Tsys is expressed as: Tsys = 10FN /10 − 1 T0 + Ta , (7) where T0 is the reference temperature (typically 290 K), FN is the receiver noise figure in dB, and Ta is the antenna noise temperature, which depends on the receiver’s field of view and pointing direction. c) Download capacity, Γ This metric represents the total amount of data that can be downloaded over a 24-hour period. It is computed by averaging the channel capacity over the scenario repetition period and scaling accordingly: Z Tday T Γ= C(t) dt. (8) T 0 d) Energy efficiency, η Defined as the ratio of the total useful data transmitted to the energy expended. Assuming the CubeSat transmits with a constant power PTx whenever a link is available, the energy efficiency is: Z T 1 η= C(t) dt. (9) PTx Tc 0 This metric reflects the system’s effectiveness in delivering data relative to its energy consumption, which is particularly critical for power- and battery-constrained space users. III. M ATHEMATICAL F RAMEWORK In this section, we present the mathematical analysis of the performance metrics for the system model considered. A. Distance to CubeSat We first analyze the time-varying geometric distance between the CubeSat and the two receiver types—GSs and NTN relays. This analysis is fundamental as it directly influences all performance metrics defined in Sec. II 1) CubeSat-to-Ground Station Distance The distance dGS (t) varies due to the CubeSat’s orbital motion and the Earth’s rotation. To compute dGS (t), we transform the CubeSat’s position from its 2-D orbital plane to the 3-D Earth-centered coordinate system. This involves rotating by the RAAN Ω around the z-axis and the inclination i around the x-axis. The resulting transformation matrix is: cos Ω sin Ω 0 M = − cos i sin Ω cos i cos Ω sin i (10) sin i sin Ω − sin i cos Ω cos i
5
The CubeSat’s position in the 3-D coordinate system is then computed using its position in the 2-D coordinate system as: xCS RCS cos(ωCS t) yCS = M −1 RCS sin(ωCS t) . (11) zCS 0 On the other hand, the position of the GS is defined in the 3-D coordinate system by: xGS = RE cos φ cos(Λ + ωGS t − β) (12) yGS = RE cos φ sin(Λ + ωGS t − β) , zGS = RE sin φ where φ and Λ represent the latitude and longitude of the GS, respectively, and the parameter β represents the angular rotation of the Earth with respect to the 3-D coordinate system at the adopted time origin. The resulting Euclidean distance is then calculated as: p dGS (t) = (xGS−xCS )2 + (yGS−yCS )2 + (zGS−zCS )2 , (13) where the dependency of the position components with time has been committed for clarity. Based on this, the minimum and maximum possible distances are defined by: (p 2 − 2R R RE2 + RCS E CS cos(|φ| − i), |φ| ≥ i min (14) dGS = h , |φ| ≤ i (pCS 2 + 2R R RE2 + RCS E CS cos(|φ| − i), |φ| ≥ i dmax . (15) GS = 2RE + hCS , |φ| ≤ i Notably, we analyze dGS (t) regardless of the Earth’s blockage, which we then take into account as a mask to the possible values of dGS (t). In this regard, we define the corresponding maximum blockage distance between the CubeSat and the GS: B 2 dGS = RE2 + RCS − 1/2 (16) RE −1 2RCS RE sin θth + sin cos θth , RCS where θth indicates the minimum elevation angle of a GS given the surrounding obstacles. Because we are restricting the present analysis to just circular LEO orbits, it is always min B the case that dBGS < dmax GS . Using the notions of dGS , and dGS , we derive the following visibility function: ( B 1, dmin GS < dGS < dGS VGS (dGS ) = (17) 0, Otherwise.
VGS (t) =
Since no closed-form solution exists for the visibility inequality when incorporating (13) into (17), visibility periods must be computed numerically. Let τ1 , τ2 , . . . , τL be the simple roots of g(t) = dGS (t) − dBGS within [0, TGS ], such that g(τl ) = 0 and g ′ (τl ) ̸= 0. The resulting visibility function VGS (t) is defined in (23), where u(t) is the unit step function and g ′ (t) is the derivative of g(t). Using dmax GS , we define the Cumulative Density Function (CDF) of the distance from the CubeSat to the GS, assuming that t ∼ U [0, TGS ]: d ≤ dmin 0, GS FdGS (d) = Fd∗GS (d), dmin (18) < d < dmax GS GS 1, d ≥ dmax , GS where Fd∗GS is defined as follows: Fd∗GS (d) = P (dGS (t) ≤ d) = P (dGS (t) − d ≤ 0) .
Calculating Fd∗GS (d) requires identifying the simple roots ξ1 , ξ2 , . . . , ξK of the inequality in (19). As shown in (24), the expression is derived by integrating over all time intervals within TGS where dGS (t) ≤ d. If multiple GSs are considered, the CubeSat connects to the closest visible station. Thus, dGS (t) is defined as the minimum distance between the CubeSat and the NG GSs: dGS (t) = min{dGS1 (t), dGS2 (t), ..., dGSNG (t)},
n=1 (L−1)/2 P
TNTN =
n=1
2π . NS |ωCS − ωNTN |
(21)
The angular velocities of the CubeSat and the NTN relays are derived from Kepler’s third law, given by: r µ ωCS/NTN = , (22) 3 RCS/NTN
(L+1)/2 P n=1 u(t − τ2n−1 ) − n=1 u(t − τ2n ) − u(t − TGS ), L/2 L/2 P P u(t) + u(t − τ2n ) − u(t − τ2n−1 ) − u(t − TGS ), n=1
(20)
and the rest of the analysis is equally derived. 2) CubeSat-to-NTN Relays Distance dNTN (t) is a periodic function of the relative motion between co-planar orbits. Assuming alignment at t = 0 and NS evenly spaced relays, the angular distance between consecutive nodes is 2π/NS . As the CubeSat overtakes (or is overtaken by) the relays due to altitude-dependent angular velocities, the link distance repeats with period TNTN . This period, representing the time to sweep a relative angular displacement of 2π/NS , depends on NS and the relative angular velocity ωCS − ωNTN :
L/2 L/2 P P u(t − τ2n−1 ) − u(t − τ2n ), n=1 n=1 (L−1)/2 (L+1)/2 P P u(t − τ2n ) − u(t − τ2n−1 ), u(t) + n=1
(19)
if g ′ (τ1 ) < 0 and g ′ (τL ) > 0 if g ′ (τ1 ) > 0 and g ′ (τL ) > 0 (23) ′
′
if g (τ1 ) < 0 and g (τL ) < 0 if g ′ (τ1 ) > 0 and g ′ (τL ) < 0
6
where µ = GME is the Earth’s standard gravitational parameter, G is the gravitational constant and ME the mass of Earth (µ ≈ 3.986 × 1014 m3 /s2 ). The distances between the CubeSat and two consecutive relays are calculated via the law of cosines in (25). The CubeSat connects to relay n during [0, TNTN /2] and switches to relay n + 1 during (TNTN /2, TNTN ] as it becomes the closer node. Due to the circular symmetry of the constellation, the distance function in the second interval is a mirrored and shifted version of the first, a pattern that repeats periodically. Equations (26) and (27) define the link distance bounds for (25). The minimum, dmin NTN , occurs when the CubeSat and relay are radially aligned, while the maximum, dmax NTN , occurs when the CubeSat is equidistant between two relays. These limits are essential for subsequent visibility and capacity evaluations. dmin NTN = |hNTN − hCS | s dmax NTN =
2 + R2 RCS NTN − 2RCS RNTN cos
π In (30), if NS ≥ ⌈ α ⌉, the first condition is impossible, rendering the visibility function equal to 1 for all t. Ensuring permanent CubeSat coverage requires dmax NTN < B dNTN , meaning the maximum distance to any relay must remain below the Earth’s blockage threshold. This yields the minimum number of NTN relays required for continuous coverage:
π NS
p hCS hNTN (2RE + hCS )(2RE + hNTN )
NTN
(27)
1/2 .
This distance represents the segment between the CubeSat and an NTN relay when tangent to the Earth’s surface. As shown in (28), the blockage distance depends primarily on the orbital altitudes hCS and hNTN . Similar to the GS case, we max B define a visibility function using dmin NTN , dNTN , and dNTN : ( B max 1, dmin NTN < dNTN < min{dNTN , dNTN } VNTN (dNTN ) = (29) 0, otherwise.
' π r (32) R2 R2 R2 1 − R2E cos−1 RNTNERCS − 1 − R2E
& NS∗ =
(26)
These limits analyze dNTN (t) regardless of Earth’s blockage, though LoS is not always guaranteed. For example, if relay density is low, certain orbital positions may lack coverage. To account for this, we define the maximum distance for unobstructed LoS as: B dNTN = h2CS + h2NTN + 2RE (hCS + hNTN )+ (28) +2
The parameter α captures the dependency on the number of relays and the altitudes of both the CubeSat and NTN relays: 2 ! 2 2 RCS + RNTN − min{dBNTN , dmax NTN } −1 . (31) α = cos 2RCS RNTN
CS
We study the statistical properties of dNTN using (25) and assuming again a uniform distribution of time, i.e. t ∼ U [0, TNTN ]. The resulting CDF is reported in (33), with the corresponding derivations included in Appendix A. The statistical models for dGS and dNTN provide a basis for evaluating distance-dependent performance metrics and their distributions. These models enable the analysis and comparison of different architectures, including ground-based and NTN-assisted scenarios. The derived distributions are used in subsequent sections to compute key performance metrics. B. Contact Probability As introduced in Section II-C, the contact probability Q represents the ratio of time the CubeSat maintains LoS with a receiving node and the scenario repetition period T . For the CubeSat-to-GS links, the scenario repetition period TGS corresponds to the least common multiple (LCM) of the Earth’s rotation period, Tday , and the CubeSat orbital period, TCS = 1/ωCS , as follows: Tday TCS TGS = Tres LCM , , (34) Tres Tres
Incorporating the time-dependence of dNTN (t) into (29) yields a visibility function that depends directly on time: ( Sα Sα 0, N2π TNTN < t < (1 − N2π )TNTN VNTN (t) = (30) for which we use a temporal resolution Tres set to 1 min. 1, Otherwise, ! K K 2 2 P P 1 ξ2n − ξ2n−1 , if g ′ (ξ1 ) < 0 and g ′ (ξK ) > 0 TGS n=1 n=1 ! K+1 K−1 2 2 P P 1 ξ2n−1 − ξ2n , if g ′ (ξ1 ) > 0 and g ′ (ξK ) > 0 TGS n=1 n=1 ! Fd∗GS (d) = (24) K+1 K−1 2 2 P P 1 ′ ′ ξ2n − ξ2n−1 , if g (ξ1 ) < 0 and g (ξK ) < 0 TGS + TGS n=1 n=1 ! K K 2 2 P P 1 ξ2n−1 − ξ2n , if g ′ (ξ1 ) > 0 and g ′ (ξK ) < 0 TGS TGS + n=1
dNTN (t) =
n=1
p 2 + R2 RCS NTN − 2RCS RNTN cos(ωCS − ωNTN )t,
0 < t < TNTN 2
p
TNTN 2 < t < TNTN
2 + R2 RCS NTN − 2RCS RNTN cos(ωCS − ωNTN )(TNTN − t),
(25)
7
By integrating the visibility function in (17) over a full cycle and dividing by TGS , we obtain the contact probability expressed in (35). R TGS VGS (t) dt QGS = 0 = Fd∗GS (dBGS ) (35) TGS Similarly, the contact probability QNTN is the ratio of contact time to the NTN scenario periodicity, TNTN , derived in (21). By integrating the visibility function in (30) over a full cycle, we determine the contact time as follows: R TNTN VNTN (t) dt QNTN = 0 TNTN R N2πS α TNTN R TNTN 1 dt + (1− 1 dt NS α 0 NS α 2π )TNTN = = . (36) TNTN π As expected,, the contact probability QNTN depends on the number of relays NS and orbital altitudes through α (31). If NS < NS∗ , LoS availability is intermittent due to geometric constraints, resulting in QNTN < 1. Conversely, if NS ≥ NS∗ , the CubeSat maintains continuous LoS, and QNTN = 1. C. Channel capacity The channel capacity C(t) is inherently time-dependent due to the motion-driven variation of d(t). While PN remains constant—determined by bandwidth and receiver noise temperature—PRx (t) varies with the orbital geometry, making C(t) highly sensitive to the satellites’ positions. By substituting (1) into (6), the channel capacity for both GS and NTN cases is derived as: PTx GCS GRx V (t) C(t) = W log2 1 + PN L(fc , d(t)) (37) b = W log2 1 + V (t), Labs (fc , d(t))d2 (t) where b encapsulates the effects of transmit power, noise power, carrier frequency, and transmitter and receiver antenna gain as follows: 2 PTx ς b= GCS GRx . (38) PN 4πfc Unlike the GS case, molecular absorption is negligible outside the atmosphere (Labs (fc , d) ≈ 1). Consequently, for the NTN case, (37) simplifies to: b VNTN (t). CNTN (t) = W log2 1 + 2 (39) dNTN (t) Using the statistical characterization of d(t), we derive the univariate distribution of C(t). Specifically, the CDF of channel capacity is determined using the visibility functions from Sec. III-A as follows: FC (c) = P (C ≤ c) = P (C ≤ c | V (d) = 0)P (V (d) = 0)
(40)
+ P (C ≤ c | V (d) = 1)P (V (d) = 1),
FdNTN (d) =
0,
NS
π 1,
where c ≥ 0. Given that C = 0 during non line of sight (NLoS) events, P (C ≤ c | V (d) = 0) = 1. The probabilities of NLoS and LoS are 1 − Q and Q, respectively, where Q is the contact probability. Applying these to (37), we obtain: FC (c) = 1 − QP d ≤
b Labs (fc , d)(2c/W − 1)
0.5
! V (d) = 1 .
Substituting (18) and (33) into (41) yields the capacity CDFs for GS and NTN in (46) and (47). For brevity, the frequency and distance dependence of absorption loss is omitted in (46). These formulations enable quantifying the impact of relay density and altitude on throughput, as explored in Sec. IV. D. Total download capacity Average system performance is evaluated through the total download capacity Γ, introduced in (8). This metric is obtained by integrating C(t) over the scenario’s repetition period and scaling to a 24-hour duration. Substituting (37) into (8), Γ is calculated as: Z Tday T b V (t)dt (42) W log2 1 + Γ= T 0 Labs (fc , d(t))d2 (t) For the NTN case, (42) can be further developed by neglecting absorption loss and leveraging the symmetry of dNTN (t) around TNTN /2. By substituting (21), (25), and (30) and applying a change of variables, we obtain: Tday W ΓNTN = π ! Z NS α b dϕ. log2 1 + 2 2 RCS + RNTN − 2RCS RNTN cos( NϕS ) 0 (43) Since the integrals for the NTN and GS cases lack analytical solutions, they must be computed numerically. While (43) shows that total download capacity increases monotonically with the number of relays, we must determine if this growth is bounded. To evaluate whether capacity increases indefinitely with NS , we examine the limit: W b lim ΓNTN = log2 1 + min 2 . (44) N →∞ Tday (dNTN ) This result reveals a fundamental bound: the total download capacity cannot be arbitrarily increased by adding more satellites. Instead, the download capacity is bounded by the channel capacity at the minimum distance between the CubeSat and the NTN relays, dmin NTN . This is an important theoretical insight from our work that is further elaborated on in Sec. IV. E. Energy efficiency Using the general definition in (9), we compute the energy efficiency for both cases by substituting the contact time Tc = d ≤ dmin NTN
cos−1
2 2 RCS +RNTN −d2 2RCS RNTN
(41)
,
B max dmin NTN < d < max{dNTN , dNTN }
d ≥ max{dBNTN , dmax NTN }.
(33)
8
QT and the total download capacity over one period. Thus, (9) simplifies to: Γ η= . (45) PTx QTday
efficiency will soon be available through advanced precision manufacturing. TABLE II: Key radio technology parameters
Accordingly, the energy efficiency expressions for the GS and NTN cases are obtained from (45) by using the respective values of Γ, T , and Q previously derived. IV. N UMERICAL R ESULTS
Parameter fc W PTx GTx GRx FN
Ku Band 18 GHz [35] 400 MHz [37] 40 dBm [37] 23 dBi 39 dBi 3 dB [37]
sub-THz sub-THz NG 220 GHz [36] 5 GHz [38] 20 dBm [36] 27 dBm [39] 45 dBi 47 dBi 61 dBi 63 dBi 7 dB [36]
In this section, the mathematical results from Section III are numerically elaborated. We particularly compare the system A. Simulation setup performance dynamics for four modeled CubeSat connectivity Our mathematical results are cross-verified using the MATscenarios (see Section II), contrasting the existing GS-centric LAB Satellite Communications Toolbox. We primarily ensure and alternative NTN-assisted configurations. that the analytically derived distances between all communi1) Ground-Station Configuration cating nodes match simulation results, specifically addressing The single-GS scenario uses a station in Svalbard, whose the complex scenario geometry and non-linear orbital dynamnear-polar latitude provides excellent Sun-synchronous orbit ics described in Section III. Following position and distance (SSO) coverage. For the multi-GS configuration, we select 17 verification, propagation and link-budget computations are locations from the Leaf Space ”Leaf Line” network—a global performed according to the methodology in Sec. II. Ground-Segment-as-a-Service (GSaaS) provider widely used in small-satellite missions [33]. This architecture provides op- B. Link distance analysis erationally validated worldwide coverage aligned with current This subsection evaluates the temporal evolution and statistical distribution of link distances for both NTN-based and mission practices. GS-based scenarios. We compare results from the analytical 2) CubeSat and NTN Satellites Orbital Configuration The CubeSat follows a SSO, typical for missions requiring framework against the simulation procedure to validate the consistent illumination and predictable revisits. The co-planar link distance modeling for both scenarios. Fig. 3 illustrates the link distance over time for CubeSat NTN relays are positioned in circular LEO orbits at hNTN = 550 km. This co-planar assumption is physically sound for altitudes of 400 km and 2000 km. GS contacts are brief, SSO orbits, reflecting the high inclination of polar orbits in typically between 11 and 22 minutes for a single station. While increasing the number of GSs adds more contact modern constellations [3]. We evaluate three wireless connectivity options: Ku-band, events, continuous connectivity remains largely unavailable. Sub-THz, and Sub-THz Next-Generation (NG). The corre- In contrast, a single NTN relay provides significantly longer sponding link-budget parameters for these three radio setups, contact durations (approx. 2.25 to 5.7 hours), albeit often based on state-of-the-art mmWave and sub-THz hardware, at greater distances than GS links. Continuous coverage is are summarized in Table II. The difference between the achieved with NS = 10 relays, and further increasing relay two sub-THz options is that the former assumes state-of-the- density reduces the average link distance, which converges art sub-THz hardware, while the latter assumes prospective toward the altitude difference (|hCS − hNTN |) at NS = 50. Fig. 4 validates the statistical link distance analysis by better and higher-output components to become available soon [34]. We assume antenna diameters of 10 cm (CubeSat) comparing analytical CDFs from (18) and (33) with simuand 60 cm (NTN relay or GS), with an antenna temperature lation results. Shaded areas denote NLoS regions, where the Ta = 300 K [35]. The aperture efficiency is 60% for Ku and intersection of each curve with the boundary represents the standard Sub-THz, and 95% for Sub-THz NG, assuming such contact probability Q. For GS links (Fig. 4a), Q remains below 0, c<0 b 1 − QGS , 0 ≤ c ≤ W log2 1 + Labs (d B )2 q GS (46) FCGS (c) = b b b 1 − QGS Fd∗GS , W log 1 + < c < W log 1 + min )2 B )2 c/W −1) 2 2 L (2 L (d L (d abs abs GS abs GS 1, c ≥ W log2 1 + Labs (db min )2 GS
0, 1 − QNTN , FCNTN (c)= 1 − QNTN NπS cos−1 1,
c<0 0 ≤ c ≤ W log2 1 + (min{dB b ,dmax })2 NTN NTN ! 2 2 RCS +RNTN − cb b b 2 W −1 , W log 1+ < c < W log 1+ 2 2 2 2 2RCS RNTN (min{dBNTN ,dmax (dmin NTN }) NTN ) b c ≥ W log2 1 + (dmin 2 NTN ) (47)
9
12000
≈ 22 min
≈ 5 h 38 min
2000 km
8000
0.8
1500 1000
4000
≈ 11 min
500 6
0.4
6.5
7
0.2
2000
0
0
3
6
9
12
15
18
21
24
NLoS
1440 km
0.6 400 km
6000
0
NLoS
1
4437 km
10000
0
2000
4000
6000
8000
10000
(a) CubeSat-to-GS
(a) hCS = 400 km
12000 10000
7
7.5
4000
0.4
2000
0.2
0
0
8138 km
0.6
2790 km
≈ 22 min
1450 km
0.8
5000 km
≈ 2 h 15 min
≈ 37 min
2149 km
6000
1
150 km
5000 4000 3000 2000 6.5
8000
NLoS
0
3
6
9
12
15
18
21
24
0
2000
4000
6000
NLoS
8000
10000
(b) hCS = 2000 km
(b) CubeSat-to-NTN
Fig. 3: Link distance for different CubeSat altitudes hCS .
Fig. 4: Link distance CDF at distinct CubeSat altitudes hCS .
20% for NG = 1, and while a multi-GS network improves Q, continuous coverage is not reached. In contrast, Fig. 4b shows that higher CubeSat altitudes increase Q at the expense of longer link distances. Notably, multiple relays enable continuous connectivity (Q = 1), as the CDFs saturate before the NLoS threshold. For NS = 10, the maximum distance remains below 2149 km and 2790 km for 400 km and 2000 km altitudes, respectively—well within the corresponding NLoS boundaries of 5000 km and 8138 km. As shown in Figs. 3 and 4, the analytical and simulated results match closely across all altitudes and NTN configurations. This high correlation validates the geometric foundations of our mathematical model. Consequently, subsequent evaluations for channel capacity, total download capacity, and energy efficiency rely exclusively on the analytical framework.
Expanding the GS network to 17 stations improves Q by 13%–55%, yet performance remains inferior to even a small relay constellation. Notably, NS = 3 relays outperform the 17-node GS architecture, while NS = 10 achieves 100% contact probability across all altitudes. This demonstrates that continuous coverage is feasible with a modest NTN deployment, a result practically unattainable with groundbased infrastructures. Fig. 5b illustrates the minimum number of relays NS∗ required for continuous connectivity, as derived in (32). The requirement for NS∗ decreases as either the NTN altitude (hNTN ) or CubeSat altitude (hCS ) increases. For hNTN = 550 km, the relay requirement drops from five to three as the CubeSat altitude rises from 400 km to 2000 km. Notably, even at the very low LEO altitudes (160 km), continuous connectivity is achievable with fewer than ten relays.
C. Probability of contact To better understand the dynamics of the probability of contact beyond Fig. 4, we plot this performance metric as a function of CubeSat orbital altitude in Fig. 5a. As shown, the probability of contact Q increases monotonically with CubeSat altitude, as higher orbits extend contact duration. While this increase is mild for single-node configurations, the single-relay NTN consistently provides a ≈ 20% higher Q than a single GS.
D. Channel capacity This section evaluates the joint impact of orbital dynamics and wireless technologies on the available CubeSat channel capacity. Fig. 6 evaluates the impact of orbital dynamics and wireless technology on channel capacity, marking the boundaries for power-limited (PL, C/B < 1 bps/Hz) and bandwidth-limited (BL, C/B > 2 bps/Hz) regimes. While Ku-
10
100 29 %
80 60
54 %
40 20
55 % ≈ 20 %
13 %
0 400
600
800
1000 1200 1400 1600 1800 2000
(a) Probability of contact.
8 7 6 5 4 3 2
200
400
600
800 1000 1200 1400 1600 1800 2000
(b) Minimum number of relays for 24/7 contact.
Fig. 5: Probability of contact and minimum number of relays for 24/7 contact. band performance is stable due to negligible absorption, SubTHz capacity varies significantly, improving drastically when transitioning from absorption-prone, low-elevation GS links to absorption-free NTN scenarios. Current Sub-THz hardware (100 mW) matches or exceeds Ku-band capacity in NTN configurations for distances below 3200 km and achieves an eightfold gain at 50 km.Sub-THz NG provides the most significant gains, maintaining a capacity approximately one order of magnitude higher than Ku-band across all distances. Both
Sub-THz NG and Ku-band exhibit similar spectral efficiency transitions—entering the PL region above ≈ 2400–2900 km and the BL region below ≈ 1400–1700 km. This indicates that the Sub-THz capacity advantage is driven primarily by its significantly larger available bandwidth rather than superior spectral efficiency. To capture the temporal dynamics of each scenario, Fig. 7 illustrates the instantaneous channel capacity for hCS of 400 km and 2000 km. Higher altitudes provide longer contacts and more stable average capacities, though lower altitudes yield higher absolute peaks due to reduced path loss. Spatial diversity is critical; at t = 4 h 46 min (hCS = 400 km), NS = 10 relays maintain Ku capacity near 3.4 Gbps, whereas a single node drops to 0.23 Gbps. Sub-THz NG in GS scenarios generally outperforms Ku-band but suffers from severe absorption at low elevation angles, causing rapid degradation away from the zenith. In contrast, Sub-THz NTN architectures decouple high-frequency performance from atmospheric impairments, providing consistent gains—up to 7-fold for SubTHz and 12-fold for Sub-THz NG in peak capacity. These results demonstrate that NTN configurations are essential for achieving the reliable, multi-gigabit throughput required for data-intensive space applications. Fig. 8 provides a statistical characterization of channel capacity for NS = 10 and NG = 17 configurations. For GS architectures, the offset FC (0) = 1 − Q represents outage periods where no contact exists. The results confirm that lower CubeSat altitudes (400 km) yield higher peak capacities and greater variability than higher orbits, as lower altitude CDF curves exhibit a broader capacity range and saturate at higher values compared to those for 2000 km. Notably, all GS-based curves saturate well below NTN counterparts. Sub-THz NTNs emerge as the only viable solution for highdata-rate demands, with P (C ≥ 10 Gbps) reaching 17% for standard Sub-THz and 64% for Sub-THz NG. Conversely, while Ku-band GS systems remain suitable for applications below 1 Gbps, they fail to achieve the 10 Gbps threshold with meaningful probability. E. Total download capacity
×8
102
2400 km
101
Sub-THz BL
≤ 2790 km: NTN Dist. Range
×10
Sub-THz PL
100
Ku BL 2900 km
1400 km
10-1
1700 km
Ku PL
1000 2000 3000 4000 5000 6000 7000 8000
Fig. 6: Chanel capacity offered by the different wireless technologies considered.
To move beyond time-varying dynamics and evaluate the net throughput of the system, we analyze the 24-hour total download capacity Γ. As shown in Fig. 9, baseline singleGS performance ranges from 140–750 GB/day for Ku and Sub-THz, reaching ≈ 2 TB/day for Sub-THz NG. MultiGS configurations increase Γ by a factor of 4–4.5 across all technologies due to increased contact availability. Notably, a single-GS Sub-THz NG setup or a single-node Ku NTN achieves performance comparable to a 17-node Ku GS network. This highlights the core potential of the proposed framework: both the transition to NTN-based architectures and the adoption of Sub-THz technology yield substantial gains over state-of-the-art single-GS infrastructures. The most significant performance increase occurs when combining NTN-based architectures with Sub-THz radios, achieving more than an order of magnitude improvement in Γ over Ku-band across all altitudes. Current Sub-THz technology yields gains of ×1.4 to ×4, while Sub-THz NG
11
≈40 Gbps
102
≈ 23 Gbps ≈6 Gbps
≈3.4 Gbps
≈0.76 Gbps
101
×12
×7 ×10
10
×1.25
0
×15
≈0.6 Gbps
≈0.23 Gbps
-1
10
4 h 46 min
0
1
2
3
4
5
6
≈4 Gbps
≈0.4 Gbps
7
8
7
8
(a) hCS = 400 km
102
≈9.6 Gbps ≈0.9 Gbps
10
≈1.6 Gbps
1
×10
10-1
×10
×1.8
100
0
1
2
3
4
5
6
(b) hCS = 2000 km
Fig. 7: Channel capacity over time for different CubeSat orbital altitudes hCS .
1 0.8
17 %
64 %
0.6 0.4 0.2 0 0.1
1
10
100
Fig. 8: CDF of the channel capacity for the different architectures and technologies considered. technology extends this to ×10–×11. These results underscore that the maximum potential of NTN architectures is realized by leveraging Sub-THz bands. Notably, the curves in Fig. 9a exhibit local maxima due to the trade-off between contact probability and link distance. While higher CubeSat altitudes increase the probability of contact (Fig. 5a), they simultaneously increase link distances (Fig. 4). The interplay of these opposing effects yields a peak in Γ, which for NTN architectures occurs at 550 km—the altitude of the relay nodes. As shown in Fig. 9b, increasing the number of NTN relays enhances Γ, yet (44) and Fig. 9b confirm this performance is bounded. These upper bounds are significantly higher for
Sub-THz than for Ku-band. At 2000 km, the maximum daily download capacities are 10, 17, and 103 TB for Ku, SubTHz, and Sub-THz NG, respectively. Reducing the altitude to 400 km increases these bounds by ≈ ×4 for Ku and Sub-THz NG, while standard Sub-THz experiences a ×14.5 gain as the link transitions from a PL to a BL regime. The rate at which Γ approaches its bound varies by configuration; round markers in Fig. 9b indicate the 75% capacity threshold. Higher CubeSat altitudes reach the threshold with fewer NTN relays. Comparing technologies, Ku-band and Sub-THz NG saturate earlier than current Sub-THz hardware, which remains power-limited—underscoring the need for higher-power SubTHz sources. Notably, the threshold of diminishing returns often occurs below NS = 20, where the marginal capacity gains from additional relays may not justify the increased cost. F. Energy efficiency To wrap up the results, we analyze the energy efficiency η across technologies and architectures in Fig. 10. For GS architectures, η decreases monotonically with altitude because the gains in total download capacity Γ fail to offset the increased link distance and contact time trade-offs described in (45). In contrast, NTN architectures achieve peak energy efficiency at approximately 550 km, directly coinciding with the altitude of the relay nodes. Ku-band energy efficiency remains low, between 0.036 and 0.16 GbpJ, while Sub-THz technologies achieve orders-of-magnitude higher efficiency. This advantage is driven by vast available bandwidths, en-
12
4
10
Fig. 9—provided by next-generation hardware. 103
V. C ONCLUSIONS
102
× 11
×4
× 10
× 1.4
101
×4
100 ×4
10-1 400
600
800
× 4.5
1000 1200 1400 1600 1800 2000
(a) Across different architectures and technologies
103
10
𝑁𝑆 =39
≈ 433 TB ≈ 246 TB 2 ≈ 103 TB
𝑁𝑆 =62 × 4.2 × 14.5
𝑁𝑆 =11 𝑁𝑆 =36 𝑁𝑆 =14
101 𝑁𝑆 =10
10
≈ 17 TB ≈ 10 TB
0
100
101
≈ 37 TB
× 3.7
𝛤𝑁𝑇𝑁 = 0.75 × lim 𝛤𝑁𝑇𝑁 𝑁𝑆 𝑁𝑠→∞
102
103
(b) As a function of the number of NTN satellites N
Fig. 9: Total download capacity Γ over one day of coverage. abling superior total download capacity despite using transmit powers 100 and 20 times lower than Ku-band for current and next-generation cases, respectively. Notably, increasing SubTHz transmit power slightly reduces energy efficiency but allows the link to escape the power-limited regime. This tradeoff is justified by the substantial throughput gains—as seen in
103 102
≈ 0.16 GbpJ ≈ 0.036 GbpJ
101 100 10-1 10-2 400
600
800
As the commercial space economy grows, 6G and beyond NTNs offer a vital paradigm shift to overcome the bandwidth constraints and GS congestion facing emerging “space users”. In this study, we evaluated this new concept via a comprehensive mathematical framework that incorporates satellite orbital dynamics, mutual mobility, and radio link characteristics. We then quantified the key performance indicators for such a system, including the contact probability, capacity, daily download volume, and energy efficiency. To ensure accuracy, our analytical results were also cross-verified via computer simulations. We finally compared the performance boundaries for the considered NTN-relay-assisted deployment scenarios with state-of-the-art Ku-band GDSs (a representative baseline [4]). Our key findings from this study are: 1) NTN-based relays provide substantial improvements in service availability compared to ground relays. Particularly, the probability of contact in Fig. 5 for as few as 3 NTN relays is already much higher than the one with as many as 17 ground relays. 2) Continuous 24/7 coverage is achievable with as few as 8– 10 NTN satellites per orbit – in contrast to only up to 40% contact probability with roughly twice as many GSs for CubeSats at altitudes under 1000 km (hence, over 60% of time, the GS-connected CubeSat is unavailable). 3) The utilization of NTN-based relays also leads to up to a tenfold increase in the downlink capacity and the daily download volume (as illustrated in Fig. 7–Fig. 9). 4) Assuming continuous beam alignment, the use of sub-THz radio links, especially, the forthcoming sub-THz NG-class equipment may lead to up to 40x increase in the total download volume of data per day compared to state-ofthe-art Ku radio in similar conditions1 . 5) While a higher number of NTN relays naturally increases capacity, the system reaches diminishing extra benefits at approximately 20 nodes per orbit, bounding the benefit of further densification (as best illustrated in Fig. 9b). 6) Last but not least, the use of NTN-based space relays also notably improves the energy efficiency of the data exchange with the CubeSat (as summarized in Fig. 10). The shift toward mmWave- and sub-THz-enabled NTNs marks a transformative milestone for space exploration and commercialization. By dismantling the ground-based download bottleneck, the concept evaluated in this article provides a scalable, multi-gigabit backbone for next-generation data-intensive missions. As 6G matures, high-frequency relay constellations will serve as the foundational infrastructure for a connected space economy. Ultimately, combining the benefits of mmWave and sub-THz systems with the ubiquitous coverage of LEO satellite communication networks transforms emerging space users (CubeSats, space telescopes, orbital stations, etc.) from only sporadically-connected individual
1000 1200 1400 1600 1800 2000 1 Imperfect beam alignment is expected to lead to performance gains of sub-THz over Ku becoming lower than 40x but still staying substantial.
Fig. 10: Energy efficiency of the CubeSat links.
13
devices into a continuously-available integral part of 6G+ heterogeneous non-terrestrial networks of tomorrow. A PPENDIX A CDF OF THE DISTANCE TO AN NTN SATELLITE The CDF of the distance between the CubeSat and an NTN satellite is defined as FdNTN (d) = P (dNTN (t) ≤ d). We incorporate (25) in the expression above to obtain: TNTN F dNTN (d) = P (cos(∆ωt)≥σ) P 0 < t < + 2 (48) TNTN P (cos(∆ω(TNTN − t))≥σ) P < t < TNTN , 2 where ∆ω = |ωCS − ωNTN |, and: σ=
2 2 RCS + RNTN − d2 . 2RCS RNTN
Therefore, solving the inner inequalities for t: 1 1 cos−1 (σ) FdNTN (d) = P 0 ≤ t ≤ 2 ∆ω 1 1 −1 cos (σ) ≤ t ≤ TNTN . + P TNTN − 2 ∆ω
(49)
(50)
Here, we incorporate the assumption that t ∼ U [0, TNTN ] to obtain: 2 cos−1 (σ) . (51) FdNTN (d) = TNTN ∆ω Finally, incorporating (21) we obtain the final expression of B max the CDF when d ∈ (dmin NTN , max{dNTN , dNTN }): 2 2 RCS + RNTN − d2 N −1 FdNTN (d) = cos . (52) π 2RCS RNTN R EFERENCES [1] S. Aliaga et al., “Non-Terrestrial Networks for Space Vehicles Beyond 6G: Applications, Architecture, and Capacity Limits,” in 2024 Asilomar Conf on Signals, Systems, and Computers, Oct. 2024, p. 850–857. [2] G. Quaglione, “Evolution of the Intelsat system from Intelsat IV to Intelsat V,” J. of Spacecraft and Rockets, vol. 17, no. 2, pp. 67–74, Mar. 1980. [3] N. Pachler et al., “An Updated Comparison of Four Low Earth Orbit Satellite Constellation Systems to Provide Global Broadband,” in 2021 IEEE ICC Workshops, Jun. 2021, pp. 1–7. [4] O. Kodheli et al., “Satellite Communications in the New Space Era: A Survey and Future Challenges,” IEEE Communications Surveys & Tutorials, vol. 23, no. 1, pp. 70–109, Oct. 2021. [5] R. De Gaudenzi et al., “The Open Challenge of Integrating Satellites into (Beyond-) 5G Cellular Networks,” IEEE Network, vol. 36, no. 2, p. 168–174, Mar. 2022. [6] M. Civas and O. B. Akan, “Terahertz wireless communications in space,” ITU J. on Future and Evolving Technologies, vol. 2, no. 7, pp. 31–38, Oct. 2021. [7] S. Aliaga et al., “Joint Terahertz Communication and Atmospheric Sensing in Low Earth Orbit Satellite Networks: Physical Layer Design,” in Proc. of the IEEE WoWMoM Workshops, Jun. 2022, pp. 457–463. [8] A. U. Chaudhry and H. Yanikomeroglu, “Temporary Laser Inter-Satellite Links in Free-Space Optical Satellite Networks,” IEEE Open Journal of the Communications Society, vol. 3, pp. 1413–1427, Aug. 2022. [9] M. M. Azari et al., “Evolution of non-terrestrial networks from 5g to 6g: A survey,” IEEE Communications Surveys & Tutorials, vol. 24, no. 4, pp. 2633–2672, Fourthquarter 2022. [10] I. F. Akyildiz and A. Kak, “The internet of space things/cubesats,” IEEE Network, vol. 33, no. 5, pp. 212–218, 2019. [11] ESA Space Debris Office, “ESA’s Annual Space Environment Report,” European Space Agency (ESA), Darmstadt, Germany, Report GENDB-LOG-00288-OPS-SD, Mar. 2025, accessed: 2026-04-04. [Online]. Available: https://sdup.esoc.esa.int/discosweb/statistics/
[12] D. Tiede, “Big Earth Data: From Data to Information,” Journal for Geographic Information Science, vol. Volume 6,, pp. 184–193, Jun. 2018. [13] D. S. Abraham et al., “Recommendations emerging from an analysis of NASA’s deep space communications capacity,” Space Operations: Inspiring Humankind’s Future, pp. 475–511, May 2019. [14] 3GPP, Study on New Radio (NR) to support non-terrestrial networks, Oct. 2020, no. TR 38.811. [15] D. J. Israel and H. Shaw, “Next-generation NASA Earth-orbiting relay satellites: Fusing optical and microwave communications,” in 2018 IEEE Aero. Conf. Proc., Mar. 2018, pp. 1–7. [16] L. Zhang, “Development and prospect of Chinese lunar relay communication satellite,” Space: Science & Technology, Apr. 2021. [17] J. Schoolcraft et al., “MarCO: Interplanetary Mission Development on a CubeSat Scale,” in Space Operations: Contributions from the Global Community, C. Cruzen et al., Eds. Cham: Springer, 2017, pp. 221–231. [18] P. Wan and Y. Zhan, “A structured Solar System satellite relay constellation network topology design for Earth-Mars deep space communications,” Int. J. of Satellite Comms. and Networking, vol. 37, no. 3, pp. 292–313, Nov. 2019. [19] D. Modenini et al., “Two-leg deep-space relay architectures: Performance, challenges, and perspectives,” IEEE Trans. on Aerospace and Electronic Systems, vol. 58, no. 5, pp. 3840–3858, May 2022. [20] K.-M. Cheung et al., “Deep space relay architecture for communications and navigation,” in 2023 IEEE Aero. Conf. Proc., Mar. 2023, pp. 1–19. [21] D. I. Elewaily et al., “Delay/disruption-tolerant networking-based the integrated deep-space relay network: State-of-the-art,” Ad Hoc Networks, vol. 152, p. 103307, 2024. [22] G. Palermo et al., “Earth Orbiting Support Systems for commercial low Earth orbit data relay: Assessing architectures through tradespace exploration,” Acta Astronautica, vol. 111, p. 48–60, Jun. 2015. [23] I. Mehdi et al., “THz Technology for Space Communications,” in 2018 Asia-Pacific Microwave Conf. (APMC), Nov. 2018, p. 76–78. [24] A. J. Alqaraghuli et al., “Performance Analysis of a Dual Terahertz/Ka Band Communication System for Satellite Mega-Constellations,” in Proc. of the IEEE WoWMoM Workshops, Jun. 2021, p. 316–322. [25] S. Nie and I. F. Akyildiz, “Channel Modeling and Analysis of InterSmall-Satellite Links in Terahertz Band Space Networks,” IEEE Trans. on Communications, vol. 69, no. 12, p. 8585–8599, Dec. 2021. [26] J. Kokkoniemi et al., “Channel modeling and performance analysis of airplane-satellite terahertz band communications,” IEEE Trans. on Vehicular Technology, vol. 70, no. 3, pp. 2047–2061, Feb. 2021. [27] X. Wang et al., “Coverage and Rate Analysis of LEO Satellite-toAirplane Communication Networks in Terahertz Band,” IEEE Trans. on Wireless Communications, vol. 22, no. 12, p. 9076–9090, Dec. 2023. [28] W. Gao et al., “Terahertz aerospace communications: enabling technologies and future directions,” Science China Information Sciences, vol. 68, no. 12, p. 220302, Nov. 2025. [29] S. U. Hwu et al., “Terahertz (THz) wireless systems for space applications,” in 2013 IEEE Sensors Applications Symposium Proceedings, Feb. 2013, p. 171–175. [30] J. G. Walker, “Satellite Constellations,” Journal of the British Interplanetary Society, vol. 37, no. 12, pp. 559–572, 1984. [31] “Attenuation by atmospheric gases and related effects,” International Telecommunication Union, vol. P.676-12, 2019. [32] “Reference standard atmospheres,” International Telecommunication Union, vol. P.835-6, 2017. [33] Leaf Space, “Leaf Line,” https://satsearch.co/products/leafspace-leafline, Accessed: Feb. 2026. [34] J. M. Jornet et al., “The Evolution of Applications, Hardware Design, and Channel Modeling for Terahertz (THz) Band Communications and Sensing: Ready for 6G?” Proceedings of the IEEE, vol. 113, no. 9, pp. 920–951, Sept. 2025. [35] G. Maral et al., Satellite communications systems: Systems, techniques and technology. Wiley, 2009. [36] J. V. Siles et al., “A New Generation of Room-Temperature FrequencyMultiplied Sources With up to 10× Higher Output Power in the 160GHz–1.6-THz Range,” IEEE Trans. on Terahertz Science and Technology, vol. 8, no. 6, p. 596–604, Oct. 2018. [37] J. Chen et al., “Catalyzing satellite communication: A 20W Ku-Band RF front-end power amplifier design and deployment,” PLoS ONE, vol. 19, no. 4, p. e0300616, Apr. 2024. [38] P. Sen et al., “A versatile experimental testbed for ultrabroadband communication networks above 100 GHz,” Computer Networks, vol. 193, p. 108092, Jul. 2021. [39] K. B. Cooper et al., “A Power-Combined 240 GHz Frequency-Multiplier Source for Cloud Radar Applications,” IEEE Journal of Microwaves, vol. 5, no. 6, p. 1253–1259, Nov. 2025.