1. Introduction
1.1. Background
From fifth-generation (5G) communication technology and beyond, non-terrestrial networks (NTNs) have become a crucial component of future network architectures, playing a pivotal role in achieving the vision of three-dimensional (3D) global seamless coverage [
1,
2]. NTNs, which include geostationary-Earth-orbit (GEO) satellites, medium-Earth-orbit (MEO) satellites, low-Earth-orbit (LEO) satellites, and high-altitude platforms (HAPs), are capable of providing network connectivity to any corner of our Earth, as illustrated in
Fig. 1. This capability is especially vital in remote areas where terrestrial network (TN) coverage is insufficient or economically unfeasible, such as mountainous regions or oceans [
3]. Moreover, in scenarios where TN infrastructures are compromised by natural disasters or emergencies, NTNs serve as an essential connectivity alternative [
4]. With the advancement of onboard data-processing capabilities and the development of large constellations, satellite communication is becoming a prominent technology in the evolution of NTNs, attracting significant attention from both industry and academia [
5,
6].
The development of LEO constellations is spearheading the evolution of NTN construction due to their proximity, ranging from 500 to 1200 km above the Earth, which enables low-latency network connectivity and increased radio link budgets compared with higher orbit satellites [
7,
8]. The evolution of LEO communication technologies can be characterized in three distinct phases, each of which also reflects advancements in user equipment (UE) types:
- •
The first phase involved dedicated satellite systems, where UE operated independently with both TNs and NTNs—for example, a satellite and a cellular hybrid dual-mode phone, where the latter was often large in size.
- •
The current second phase involves installing a cellphone tower in space. This allows UE to operate without hardware or firmware modifications, enabling NTNs to provide direct-to-device services and support sub-6 GHz cellular data services.
- •
The anticipated third phase, named 5G NTN enhancement, will involve new designs on both satellite and UE to further boost NTN service capabilities and reliability, aiming for a unified TN–NTN constructions.
Historically, these phases also mirror the evolving relationship between TNs and NTNs: starting from independent development with interworking before 5G, moving to TN innovation with minimal changes to support NTNs in 5G and 5G-advanced, and evolving toward a unified TN–NTN design and optimization in six-generation (6G) and beyond. Regarding NTN development in standardization activities, the Three-Generation Partnership Project (3GPP) work in Release 15 (Rel-15) and Rel-16 established key NTN use cases and system architectures while providing satellite-based NTN channel models and system-level simulation assumptions. In Rel-17, 3GPP identified NTNs as part of the 5G new radio (NR) ecosystem, detailing NTN use cases for sub-6 GHz frequencies, including massive access and narrowband Internet of Things (IoT) applications. Rel-18 explored improvements for 5G NR NTN operations for handheld terminals on 10 GHz and above, focusing on spectrum management and co-existence to facilitate efficient integration with TN [[
9], [
10], [
11], [
12]]. As part of ongoing discussions in Rel-19, NTNs are considered a long-term study area, pointing toward their evolving and crucial role in future telecommunications frameworks. Following the NTN development trajectory, the construction of a reliable space Internet to connect every corner of the world through strategic networking technology will soon be realized.
1.2. Challenges and motivation for NTN networking
The extensive coverage and dynamic features of NTNs offer opportunities for ubiquitous access. However, they also present unique challenges in terms of managing real-time, dynamic, massive connectivities due to the large satellite coverage and coverage overlap in multi-constellation environments. Additionally, when satellites are equipped with 5G base station (BS) payloads, unlike TN BSs, NTN BSs must consider dynamic satellite access and mobility management and interference management strategies because of rapid satellite movement and varying coverage status. In addition, for NTNs, it is necessary to consider long-distance and frequent signaling between satellite and UE for real-time data services. These challenges highlight the necessity of efficient NTN networking strategies, particularly regarding access management, satellite mobility, and onboard traffic scheduling, which are detailed below.
1.2.1. Radio resource management in NTNs
The dense access in NTNs often leads to competition among numerous UEs for limited bandwidth, leading to inevitable signal interference issues [[
13], [
14], [
15], [
16], [
17]]. Effective interference management is vital for maintaining the reliability of NTN accessing [[
18], [
19], [
20]]. Here, the challenge primarily involves accurately estimating channel state information (CSI) and developing efficient beamforming strategies [[
21], [
22], [
23]]. CSI provides crucial details about the current state of communication channels, which are typically assessed based on known signals received at the UE and subsequently relayed to the transmitter for use in future transmissions. Accurate CSI is essential, as it allows for informed resource-allocation decisions and the formulation of strategies to mitigate signal interference. However, accurately acquiring CSI is complex in NTNs because of the rapid movement of LEO satellites [[
24], [
25], [
26]]. Traditional CSI estimation techniques that rely heavily on receiver feedback present challenges in NTNs such as latency, resulting in potentially outdated CSI [
27]. Moreover, this challenge is exacerbated by the delay in processing CSI for beamforming design, which must meet quality of service (QoS) requirements while managing interference. Conventional methods such as successive convex approximation (SCA) [
28,
29] and weighted minimum mean square error (WMMSE) [
30] offer solutions but are computationally demanding and sensitive to initialization choices, leading to further delays. These intertwined challenges underline the complexity of radio resource management in NTN and emphasize the need for more adaptive and efficient CSI estimation and beamforming techniques.
1.2.2. Mobility management in NTNs
Traditional TN handover (HO) solutions are insufficient for mobility management in NTNs, due to the distinct dynamics of satellites [
7]. Unlike ground BSs, satellites move at high speeds, leading to frequent HOs and fluctuating signal strengths, influenced by the elevation angle and additional atmospheric losses. Compared with TNs, NTNs also experience more severe Doppler shifts because of the rapid movement of satellites. Coverage by a single LEO satellite might last for only seconds, necessitating timely HO processing, including signaling and synchronization. Moreover, as TN and NTN integration progresses, mobile UEs such as cars and unmanned aerial vehicles (UAVs) may travel from a TN-dense area to a sparse area where both TN and NTN cell coverage exist, as shown in
Fig. 1. Under unified spectrum management, UEs are allowed to HO from a TN to an NTN for better data reception, making it necessary to address both intra-NTN mobility and mobility management between TNs and NTNs to ensure service continuity. Therefore, the ongoing evolution of NTN mobility management poses a key challenge.
1.2.3. Traffic scheduling in NTNs
Current NTN systems with multiple constellations often have significant limitations in flexibility and adaptability due to the constellations being custom-built for specific tasks, such as Earth observation or broadcasting. This leads to underutilization of onboard resources, as these satellites cannot easily be reconnected for other tasks during idle periods [[
31], [
32], [
33], [
34], [
35], [
36], [
37], [
38], [
39]]. To overcome these limitations, virtualization technologies such as network function virtualization (NFV) and software-defined networking (SDN) have been introduced, creating software-defined NTNs (SD-NTNs) for centralized inter-satellite networking and efficient resource scheduling [[
40], [
41], [
42], [
43]]. An important feature enabled by SD-NTNs is network slicing, which allows for the creation of multiple isolated virtual networks, known as network slices, over a shared physical infrastructure [[
44], [
45], [
46]]. Each network slice is tailored to meet the specific requirements of different services or applications, such as enhanced mobile broadband (eMBB), ultra-reliable low-latency communication (URLLC), or massive machine-type communication (mMTC) [[
45], [
46], [
47], [
48]], significantly increasing NTN operational efficiency, responsiveness, and resource utilization. However, network slicing in SD-NTNs presents challenges due to satellite fast motion, resource constraints, and diverse service demands [[
49], [
50], [
51], [
52]]. Efficient resource allocation and the resolution of service conflicts are essential for optimizing NTN networking, ensuring resource efficiency, and supporting diverse applications.
In the next section, we first summarize recent NTN advances with a focus on radio resource management, satellite mobility, and onboard traffic scheduling. Our aim is to provide a comprehensive overview of the evolution of NTN technology in satellite access, HO mechanisms, and onboard transmissions, while highlighting innovative perspectives on these elements. Finally, we identify open problems and future research directions that could foster significant breakthroughs in NTN networking.
2. Related advancements
2.1. Radio resource management
2.1.1. CSI estimation
Real-time and accurate CSI estimation is becoming one of the biggest challenges for radio resource management in highly dynamic mobile systems such as NTNs. As a means of coping with outdated CSI, many researchers have focused on the long-term estimation or prediction of CSI. CSI prediction can be categorized into two classes: stochastic modeling and artificial intelligence (AI)-based methods. Conventional stochastic methods rely on the statistical modeling of wireless channels to predict CSI. The most widely used method is the auto-regressive (AR) model, which uses a weighted linear combination of historical CSI data to forecast future CSI and approximates the channel as an AR process [
53,
54]. However, the AR model can only enable one-step prediction and incurs huge computational complexity. To tackle this problem, neural networks—with their inherent ability to capture temporal dependencies—are emerging as a powerful new paradigm for accurate and adaptive channel prediction. Based on multilayer perceptron (MLP), Turan and Coleri [
55] proposed a novel framework for channel modeling in vehicle-to-everything networks. In Ref. [
56], a convolutional neural network (CNN)-based predictor is proposed to improve the prediction quality for high mobility scenarios. Ding and Hirose [
57] were the first to consider a recurrent neural network (RNN) for channel prediction in a fading environment; because of its memory mechanism, long short-term memory (LSTM) achieves better performance in predicting sequential data. In Ref. [
58], Peng et al. proposed an LSTM-based framework for predicting the channel in massive multiple-input multiple-output (MIMO) communications under imperfect CSI. Recently, an attention-based deep learning (DL) approach known as the transformer was designed for processing sequential data. It has been demonstrated that a transformer-based structure can significantly increase the accuracy of channel prediction [
59].
2.1.2. Beamforming design
Precoding plays a pivotal role in enhancing the quality of transmitted signals by mitigating interference and improving the spectrum efficiency. One of the conventional precoding techniques is zero-forcing (ZF) precoding, which aims to cancel out interference at the receivers [
60]. However, the performance of ZF precoding is known to be sensitive to the quality of CSI. To address this limitation, iterative algorithms such as SCA [
28,
29] and WMMSE [
30] have been proposed for optimizing precoders under imperfect CSI conditions. Nevertheless, due to the substantial number of users and antennas in NTNs, the computational complexity of these methods can be prohibitively high. Thus, incentive approaches have been proposed as a means of efficiently solving the precoding design and interference management [
13]. Inspired by the superiority of DL for solving non-convex problems, researchers have introduced DL-based approaches for efficient precoder design. For example, Liu et al. [
23] introduced a DL-based approach for precoder design in vehicle-to-infrastructure (V2I) networks, which was later extended to address UAV communications [
61]. Additionally, an LSTM-based precoding framework has been proposed [
62], with subsequent applications in vehicular networks [
63]. These DL-based approaches offer promising solutions for optimizing precoder design in highly dynamic mobile systems, indicating the potential of using AI-based methods for NTN radio resource management to reduce complexity while improving performance.
2.2. NTN mobility management
2.2.1. From TNs to NTNs
Regarding NTN mobility management, LEO satellite constellations create overlapping coverage areas, presenting multiple HO candidates for each UE, which calls for more efficient HO strategies [
64]. Traditional TN HO strategies based on radio quality have been adapted for NTNs [
65,
66], resulting in high HO rates due to minimal signal variation. Studies have investigated how NTN capacity and throughput change with satellite–UE distance, proposing threshold triggers to increase HO effectiveness [
67,
68]. In addition, NTN-specific HO criteria have been developed based on satellite and UE positions, utilizing predictable satellite orbital movement to reduce HO overhead [
69,
70].
2.2.2. Mobility management with network optimization
Optimizing target satellite selections to boost service performance is crucial for efficient NTN mobility management. Researchers have incorporated satellite HO strategies into multi-metric network optimizations to refine radio resource allocation [[
71], [
72], [
73]] or facilitate traffic offloading [
74,
75]. For example, Ji et al. [
8] developed a dynamic NTN management architecture for global information acquisition; it supports satellite HO schemes aimed at minimizing signaling overhead and transmission delays. In addition, NTN HO decisions have been treated as multi-objective optimization problems to reduce HO failure rates [
76]. Network flow models have also been utilized to optimize satellite HOs, ensuring efficient resource use and reducing conflicting selections [
74,
77]. Furthermore, tailored HO strategies have been developed to meet diverse UE requirements, employing approaches such as game theory to derive HO decisions that maximize UE satisfaction [
78].
2.2.3. NTN service continuity advancements
Researchers have also focused on extending the time of stay of satellites through optimized NTN HOs for a stable data service. Al-Hourani [
79] developed an analytical model to estimate session durations between consecutive HOs, further influencing the strategic deployment of LEO constellations [
80]. Predictive HO decisions have been introduced to extend service durations and minimize disruptions [[
81], [
82], [
83]]. Machine learning has proven instrumental in heterogeneous networking for optimizing HO decisions within NTNs, enhancing system responsiveness and reliability [[
84], [
85], [
86]]. Moreover, researchers are exploring TN–NTN HO strategies to support seamless mobility for moving UEs such as cars, high-speed trains, and UAVs, thereby ensuring sustainable and high-quality data reception in integrated systems [
87,
88].
Table 1 [[
65], [
66], [
67], [
68], [
69], [
70], [
71], [
72], [
73], [
74], [
75], [
76], [
77], [
78], [
79], [
80], [
81], [
82], [
83], [
84], [
85], [
86], [
87], [
88]] summarizes advancements related to NTN mobility management. While existing works focus on specific network performance optimizations within NTN mobility management, they often divide the whole service duration into discrete snapshots to decide each HO target sequentially. This approach generally overlooks the interdependencies between consecutive HOs and potentially better future selections. That is, each HO decision is interconnected, potentially leading to various satellite HO sequences and influencing the overall system service continuity. Moreover, effective TN–NTN HO strategies must consider the distinct coverage and signal variations between TNs and NTNs.
2.3. SD-NTNs and traffic scheduling
Recently, the SD-NTN architecture has been intensively investigated [
34,
37,
43,
89,
90]. In Ref. [
34], a cross-domain SDN architecture for a multi-layered space–terrestrial integrated network is proposed to increase flexibility and efficiency in managing diverse network domains, with improvements in configuration updates and decision-making processes. In Ref. [
7], a software-defined broadband satellite network architecture is investigated to enable flexible and scalable resource management, and an optimal allocation strategy is proposed to improve resource sharing and cooperation among various network resources. In Ref. [
43], a software-defined space–terrestrial integrated network architecture is proposed to improve network management, flexibility, and service quality. In Ref. [
89], a software-defined space–air–ground integrated vehicular network architecture is proposed to support diverse vehicular services, utilizing network slicing and hierarchical controllers to manage a dynamic resource pool while ensuring service isolation across different network segments. In Ref. [
90], a software-defined satellite–terrestrial integrated network architecture is proposed, integrating information-centric networking and SDN technologies to provide flexible management and efficient content retrieval, along with cooperative caching schemes to reduce traffic load.
Network slicing in SD-NTNs has been intensively investigated [
35,
38,
39,
42,[
91], [
92], [
93], [
94]]. In Ref. [
35], network slicing in software-defined space–air–ground integrated networks is investigated to achieve a trade-off between communication and computing resource consumption. In Ref. [
94], network slicing in software-defined satellite–terrestrial integrated networks is examined as a means of minimizing link resource utilization and the number of servers used. However, the strategies for network slicing in Refs. [
35,
94] do not account for the time-varying nature of SD-NTNs. In contrast, several works have considered the dynamic characteristics of SD-NTNs [
38,
39,
42,[
91], [
92], [
93]]. Specifically, Yang et al. [
38] investigated network slicing in software-defined space information networks, jointly utilizing communication, storage, and computing resources to maximize the number of completed missions while ensuring end-to-end latency requirements. Jia et al. [
39] reported on network slicing in software-defined LEO satellite networks to minimize satellite-to-satellite resource consumption while efficiently serving terrestrial tasks. In Ref. [
42], network slicing in a software-defined satellite–terrestrial integrated network is investigated to optimize virtual network function (VNF) deployment and routing, with the goal of minimizing service latency. Yang et al. [
91] studied network slicing in software-defined space information networks to optimize routing with service function chain constraints, aiming to maximize network flow. In Ref. [
92], group sparse network slicing in software-defined space information networks is researched with the aim of achieving a trade-off between network maximum flow and network coordination overhead. In Ref. [
93], potential game-based network slicing in software-defined satellite edge computing is proposed to minimize deployment costs and maximize network payoff.
Current works largely overlook the efficient management of NTN connectivity in large-scale, highly dynamic environments. A promising direction is to exploit satellite orbital motion to design a predictable or preconfigured satellite connectivity strategy that trades off these dynamics to some extent. Radio resource management, satellite mobility, and onboard traffic scheduling ultimately impact NTN networking, spanning from UE access to inter-satellite connectivity. Building on this foundation and considering the latest trends in NTN development, we provide innovative perspectives on these elements in the following sections to help advance the evolution of space Internet management.
3. GenAI-empowered satellite radio resource management
3.1. GenAI for NTNs
Recently, generative artificial intelligence (GenAI) has gained significant attention within the field of wireless communication, opening up new frontiers for intelligent automation and network optimization [
95]. Originally popular in applications such as natural language processing, image generation, and creative design, GenAI techniques are now starting to have a strong impact on wireless communication systems by enabling more efficient and adaptive resource management, interference mitigation, and network planning [
21]. These advancements come from the generative models’ ability to learn complex patterns, generate new data samples, and predict network behaviors, which are key in addressing the growing complexity of dynamic communication systems. The rapid evolution of wireless communication technologies—including the shift toward 6G and NTNs—has further fueled the integration of GenAI into the industry, allowing for more agile, data-driven decisions for radio resource allocation.
In particular, conventional DL methods have played a vital role in solving resource-allocation problems by learning from historical data and predicting optimal resource-usage patterns. However, they often have limitations when dealing with highly complex and changing network conditions, especially those found in NTNs, where the rapid movement of satellites, fluctuating channel states, and varying traffic demands require real-time adaptability [
16]. While DL excels in classification and prediction tasks, it struggles with generating novel solutions for unseen scenarios, which are common in NTNs. GenAI, on the other hand, represents an evolution in AI’s role in radio resource allocation by offering more powerful capabilities in generating new, optimized resource-allocation strategies on the fly. Unlike DL, which typically relies on predefined datasets and supervised learning approaches, GenAI can autonomously create new data points, simulate network behaviors, and predict optimal resource allocations even in the absence of complete or up-to-date information. This generative capacity is essential in NTNs, where conditions change rapidly. By leveraging GenAI, NTNs can benefit from more proactive and flexible resource management, reducing latency and improving network efficiency in real time.
For instance, in the context of NTN CSI estimation and beamforming design, GenAI is leveraged to predict and generate accurate real-time CSI in dynamic and unpredictable NTN environments. Traditional methods often struggle with outdated CSI because of the high mobility of satellites, but GenAI models can dynamically generate predictive CSI estimates, ensuring that resource-allocation decisions are based on up-to-date information. This capability has also been applied to beamforming design, where GenAI autonomously generates optimized beamforming vectors that not only meet QoS requirements but also mitigate interference across multiple NTN layers. The following subsections dig deeper into these use cases, highlighting the crucial role GenAI plays in advancing NTN operations.
3.2. Channel estimation
Imperfect CSI at the transmitter (CSIT) originates from a satellite’s mobility and report delay due to the long propagation distance. To deal with this problem, a satellite could predict the long-term CSIT to avoid using outdated CSI. In particular, future CSIT can be predicted according to historical CSI samples. Given the volume of non-linear CSI data, applying a GenAI model is considered to be one of the most promising approaches for satellite CSI prediction.
Fig. 2 illustrates a comprehensive framework for GenAI-enabled CSI estimation in NTNs. This framework leverages the data-collection phase before the training process, which typically involves gathering trajectory and CSI data from previous communication scenarios. These collected data points serve as the foundation for pretraining. GenAI autonomously generates new data points, simulating a wide range of conditions to construct an enriched training dataset for various scenarios encountered in NTN communication systems. This ability to generate synthetic yet representative data allows GenAI to address the lack of labeled datasets, particularly in highly dynamic NTNs. Once the dataset is constructed, the GenAI model undergoes a fine-tuning process. Fine-tuning helps the model adapt its parameters to the specific NTN communication scenario, further improving CSI estimation accuracy.
A commonly used GenAI architecture, such as transformer or diffusion models, typically consists of three fundamental components: the input layer, encoder layer, and decoder layer. These components work together to facilitate the generation and prediction processes. The input layer receives the raw input data, such as past trajectory information, historical CSI, or even random noise, depending on the specific GenAI model in use. Then, the encoder processes the input data to produce a rich and compact representation of the underlying features. In transformers, the encoder employs mechanisms such as self-attention to capture complex relationships within the input sequence. This allows the model to understand long-range dependencies and contextual information, which is crucial for accurate satellite CSI prediction. In the context of diffusion models, the encoder progressively refines noisy inputs, learning to denoise and reconstruct meaningful representations at each stage. Subsequently, the decoder layer is responsible for transforming the encoded data back into meaningful output—in this case, satellite CSI data. For transformers, this involves generating sequences of data based on the encoded information and the model’s internal learned representations [
17]. In diffusion models, the decoder gradually refines the denoised representation, ultimately converging on the final predicted CSI data. Finally, the GenAI model undergoes an evaluation phase to assess its accuracy and generalization capabilities across different NTN scenarios. To minimize the signaling and computing overhead while ensuring real-time CSI estimation, GenAI modules are deployed on each satellite, while the model pretraining and fine-tuning process is deployed on regional ground controllers, which manage and collect historical data from all regional NTN nodes. The model update process involves uploading the latest model to one satellite per orbit and then distributing it to all nodes, as satellites in the same orbit have stable connections and share coverage status and traffic demand in sequence.
By automating data generation and improving the prediction process through advanced GenAI architectures, this framework demonstrates how GenAI can revolutionize satellite CSI estimation, offering immediacy and adaptability in complex and dynamic NTN environments.
3.3. Predictive beamforming
Nevertheless, effectively processing the predictive CSI results remains a formidable challenge, introducing additional processing delays that consequently compromise the NTN radio performance. To address this problem, GenAI-based predictive beamforming has been introduced as a promising solution. Predictive beamforming offers a comprehensive approach in which a GenAI model is leveraged to directly generate beamforming parameters from historical CSIT, eliminating the need for separate CSI prediction and precoder design stages. In this context, GenAI emerges as a powerful tool for tackling the complexities of this approach, as its prowess in time-series prediction and capturing non-linear relationships proves advantageous [
95], reducing signal processing overhead with higher beamforming accuracy [
63].
Fig. 3 presents a comprehensive overview of proposed predictive satellite beamforming protocol, specified for a frequency division duplex (FDD) system in which the channel conditions—including user deployment and channel coefficients—remain constant within each communication round. For the conventional FDD protocol, each communication round
t is structured into three phases: the data phase
At, the control phase
Bt, and the processing phase
Ct. In particular, in the last communication round (
t – 1), the receiver estimates the CSI from the received pilot signal and reports the estimated CSI to the transmitter during the control phase
Bt–1. Upon receiving the reported CSI from the last round, the BS predicts the CSIT in the next time slot via the CSI prediction method and generates the beamforming design accordingly. This involves additional write/read operations between different modules, leading to additional processing delay, which is more severe in dynamic NTN environments.
To solve this problem, we introduce a novel predictive beamforming protocol, as illustrated in
Fig. 3. The primary distinction between the proposed and conventional protocols lies in the processing phase
Ct. In traditional beamforming, the beamformer is predominantly generated based on the CSIT of the previous communication round. In contrast, the predictive beamforming protocol retains the CSIT reported over the last
T0 communication rounds
†1, and the GenAI model utilizes this accumulated historical CSIT to generate the beamformer. This increased context awareness leads to better anticipation of channel variations, thereby providing a more robust solution to dynamic satellite mobility and fluctuating channel conditions in NTNs. Therefore, a GenAI model can be efficiently utilized in NTNs to predict time-series data—that is, the satellite CSI sequence—thus reducing the impact of satellite mobility on real-time radio resource management and improving NTN networking accuracy.
4. Improving mobility management
In TNs, HO mechanisms enable a service BS to maintain communication when a mobile UE moves away from the current cell upon reaching cell overlap, which triggers the HO process, while the BSs remain fixed in place [
65,
67,
96]. Conversely, in NTN systems, multiple satellites from different orbits rapidly pass over the target area, each providing temporary cell coverage [
83,
97]. The serving satellite maintains communication within its coverage window until the UE is under multi-coverage for HO. Thus, traditional TN HO strategies cannot be directly applied to NTNs due to the latter’s dynamic nature, making specific mobility management strategies necessary for NTN.
4.1. Seamless HOs in NTNs
4.1.1. Distinct NTN signal variations
As depicted in
Fig. 4, a UE (in this case, a UAV) moving from the center to the edge of a TN cell will experience significant signal strength reduction, primarily due to increased propagation distances [
3,
10]. However, signal variations in an NTN are relatively minor due to the satellites’ high altitude [
98,
99]. In the temporal dimension, the service time of a LEO satellite may only last from several seconds to a minute. Given the extended HO signaling delays and the limited satellite onboard capability, the time available for executing an HO in an NTN is compressed even further. These factors complicate the determination of the best HO target for UEs in rapidly changing environments, necessitating different link measurement and HO conditions in NTNs.
4.1.2. Preconfigurable HO in NTNs
NTN mobility management cannot solely rely on signal quality metrics such as reference signal received power (RSRP) because of the rapid movement of satellites, which can affect real-time signal measurements. Yet each satellite follows a stable orbital path, allowing for predictable coverage over target areas. This predictability enables the pre-calculation of multi-satellite coverage and the accompanying durations. Additionally, satellites on the same orbit pass over a target area in sequence, exhibiting similar signal quality changes. These factors make it possible for advance HO preparations to be made in an NTN. Hence, 3GPP Rel-16 introduced the conditional handover (CHO) technique, which involves preemptively executing the HO preparation phase and then monitoring the conditions of all candidate BSs [
10]. Once a candidate meets the HO conditions, only a final detach and synchronize step with the source BS is needed [
10]. A CHO is designed to reduce HO failures in dynamic scenarios, making NTN mobility enhancements possible [
96,
100]. It is natural for UEs in NTNs to use satellite motion information (e.g., ephemeris data) as the measurement condition for a location-based CHO strategy. However, relying solely on this condition could cause UEs to trigger HOs to low-quality cells. For fast-moving satellites, the time of stay is also a crucial metric to ensure NTN service stability and minimize the HO number. Similarly, since satellites passing over a target area create a signal quality pattern that rises and then falls, the change in signal quality is an important HO evaluation condition. Hence, the monitoring condition should combine several types of metrics, but all should account for predictable satellite movements.
4.1.3. NTN HO optimization
Orbital satellite movements facilitate not just CHOs but also the prediction of potential future HOs, establishing a preconfigured NTN HO sequence, as studied in Refs. [
87,
101]. For a LEO constellation, the coverage status over a target area within a specific time duration is predictable. Thus, the satellites that the UEs in the area connect to within this period are in a time-ordered sequence. Due to the dynamics, managing a high HO rate is essential in an NTN and often leads to frequent HO signaling storms. A preconfigured HO sequence can reduce signaling interactions between satellites and UEs and among the satellites themselves, increasing the HO success rate. Moreover, a preconfigured NTN HO sequence can ensure better service continuity. Traditional HO strategies sequentially select the best target when under multiple coverages but may overlook the dependence between successive HOs, impacting network service continuity. This issue can be mitigated by configuring the NTN HO sequence, which can be integrated with technologies such as radio resource allocation and network slicing to expand NTN service capabilities.
4.2. Mobility support between TNs and NTNs
4.2.1. Integrated TN–NTN architecture
Since the inception of 5G and beyond, the integration of TNs and NTNs has played a key role in the evolution of mobile systems [
5,
8,
88]. As shown in
Fig. 5, satellites can now act as space stations managed under a unified spectrum with ground BSs, capable of signal interaction and jointly managed backhaul by network operators [
102]. In the fronthaul, UEs perceive coexisting TN and NTN cells and can select either segment for access based on the signal conditions and their requirements, which necessitates unified TN–NTN mobility management [
12,
103].
Consider a TN–NTN system with a shared spectrum. The ground BS and satellite BS share the control unit deployment; that is, a satellite gateway is deployed at each ground-control unit site, facilitating efficient and seamless HOs between the TN and NTN. In the integrated system, the NTN’s role extends beyond merely providing network connectivity in areas beyond the reach of the TN. It is increasingly valued for complementing the network capacity—especially in sparse TN distributions or when UEs are moving to cell edges—and thereby improving the overall service capabilities [
2]. This integration is particularly beneficial for mobile UEs across regions. For example, in the increasingly emphasized low-altitude economy, UAVs traveling from urban to rural areas must maintain a constant connection with a mobile network for the real-time reception of task-specific payloads and control commands. Current TNs may only ensure network coverage in urban areas. An integrated TN–NTN system ensures complete connectivity coverage along the entire UAV corridor, potentially operating in a service model that transitions from a TN to a mix of TN–NTN, and then solely to an NTN. From a coordination perspective, the integrated TN–NTN controller obtains the connectivity density distribution of the TN and NTN to determine where and how to opportunistically utilize the NTN to expand the service coverage of the TN. This integration introduces new challenges in efficient mobility management between TNs and NTNs to ensure the reliability and continuity of data services.
4.2.2. Unified assessment for TN–NTNs
UEs in integrated TN–NTNs experience uninterrupted data service through seamless HOs between TN and NTN segments [
3,
99]. However, as illustrated in
Fig. 4, there are distinct variations in signal strength between TNs and NTNs, both temporally and spatially. Spatially, TN signal variations primarily depend on the propagation distance from the cell center to the cell edge. In contrast, NTN signals are mainly affected by the elevation angle, which affects the line-of-sight (LoS) probability and shadow fading value, with lower angles inhibiting valid satellite communications. Temporally, the rapid movement of satellites may necessitate multiple NTN HOs within the duration of a single TN cell’s service. The preconfigured NTN HO sequence mentioned above is helpful for addressing these difficulties, as it can align with the TN service duration to uniformly assess HO candidates between the TN and NTN. Each UE is expected to receive high service rates, while the HO overhead for mobility support is reduced. Hence, the HO condition should jointly consider the signal-to-inference-plus-noise ratio (SINR), time of stay, and signaling dynamics for both TN and NTN candidates. It should also take the differences in TN and NTN service capability into account, with the NTN not only supplementing but also potentially competing with the TN to ensure sustainable and capable data service upon switching segments. More specifically, when requesting an HO under the coverage of both TN and NTN cells, UE will assess the potential service capability of both the TN BS candidate and the NTN service satellite sequence and then select the optimal segment, ensuring continuous and reliable data service.
4.2.3. HO between TNs and NTNs
When mobile UE is under heterogeneous HO candidate coverage, the TN–NTN HO mechanism activates [
99]. The serving BS sends an HO request to both the TN and NTN candidates. The satellite candidate computes NTN HO sequence for the UE based on predictable satellite movements for a comparison with the TN BS candidate. The UE then assesses the HO benefits of the TN and NTN under unified conditions and selects the most suitable next service segment. The preconfigured HO approach extends the HO preparation signaling to involved satellites that will fly over the UE in a time sequence to design a satellite service chain. The planned service duration is aligned with a TN cell for effective comparison between the TN and NTN. It is important to note that preconfiguring the satellite serving chain involves reserving resources in subsequent satellites, a process that incorporates constellation management techniques. This is crucial when managing large-scale LEO constellations, in order to calculate the best NTN serving chain for UEs with diverse requirements while maintaining efficient network resource allocation. Therefore, mobility management in integrated TN–NTN systems must not only consider the dynamic access between UEs and BSs but also synchronize with resource management across the heterogeneous network to achieve joint TN–NTN optimization. This technique aligns with the 6G vision and future network developments aiming for TN–NTN unification.
5. Network slicing for SD-NTNs
5.1. SD-NTN architecture
The SD-NTN architecture employs virtualization technologies to convert physical network resources—such as communication, storage, and computing—across space, air, and ground layers into virtual resources [
89], as depicted in
Fig. 6. These resources are pooled into a unified resource pool that supports networking across diverse network segments [
37,
89]. This unification facilitates dynamic resource allocation and optimization, tailored to meet the varied demands of different services. Key to this architecture are two pivotal virtualization technologies, SDN and NFV, which facilitate effective resource management and agile network orchestration [
35,
36,
89]:
(1)
Software-defined networking. SDN decouples the control plane from the data plane, enabling centralized and programmable control over the entire network [
41,
104]. In the SD-NTN architecture, SDN provides a centralized view of both physical and virtual resources across space, air, and ground segments, allowing network operators to dynamically manage and optimize resource allocation in real time [
34,
37,
52]. This centralized control is necessary to manage the highly dynamic and distributed nature of NTNs, where satellite constellations, HAPs, UAVs, and terrestrial nodes need to be coordinated seamlessly. Leveraging the extensive coverage of air and space nodes, an SD-NTN enables direct resource condition delivery between nodes and controller under the coverage. For nodes outside of direct reach, resource updates and control messages are relayed in a minimal-hop transmission fashion. Moreover, SDN controllers can monitor network conditions, such as link availability, bandwidth usage, and node health [
34,
89]. By enabling programmable control, SDN increases network flexibility, allowing the network to dynamically adjust flow routing and efficiently adapt to varying traffic demands and frequent topology changes [
34,
43]. SDN also simplifies QoS management, strengthens network security, and supports rapid fault recovery, further increasing overall network scalability and flexibility [
34,
43,
52].
(2)
Network function virtualization. NFV decouples network functions from dedicated physical hardware, allowing them to be flexibly deployed as software instances, referred to as VNFs, on various physical nodes in the NTN [
40,
105]. In the SD-NTN architecture, this decoupling enables the flexible deployment of VNFs on satellites, HAPs, UAVs, or ground stations, depending on the service demands and resource availability [
46]. This flexibility allows network operators to scale and optimize network functions without requiring specialized hardware upgrades, thereby reducing operational costs and simplifying network maintenance [
106,
107]. Furthermore, NFV enables efficient resource utilization by allowing multiple VNFs to share the same physical node, improving network adaptability to the evolving needs of different services [
38]. NFV also increases network agility by facilitating the rapid deployment and migration of VNFs across different nodes in the NTN [
107]. This dynamic allocation and redeployment of VNFs allow SD-NTNs to maintain high levels of service availability and reliability, even under changing network conditions [
107].
By integrating NFV and SDN technologies, SD-NTNs can achieve flexible, scalable, and efficient resource management across the network [
34,
37,
43,
44], ultimately improving the adaptability and scalability of NTNs to handle diverse and complex tasks.
5.2. Network slicing
In SD-NTNs, network slicing plays a pivotal role in enabling flexible, scalable, and efficient service provisioning, allowing the network to meet diverse service requirements over a shared physical infrastructure [[
44], [
45], [
46]]. More specifically, network slicing refers to the creation of multiple isolated virtual networks, or slices, that can operate simultaneously over the same physical infrastructure [[
44], [
45], [
46]], as shown in
Fig. 7. Each network slice is customized to meet the specific requirements of different services. By creating distinct virtual networks, SD-NTNs ensure that each slice has dedicated resources and configurations optimized for specific service requirements.
SD-NTNs allow VNFs to be flexibly deployed on different physical nodes, with each slice meeting the specific service requirements [
35,
38,
46]. Each requested service can be represented by a service function chain (SFC), which consists of a sequence of VNFs in a predefined order [
35,
38,
108,
109]. To complete the requested service, the mission flow must traverse each VNF in the specified sequence defined by the SFC [
35,
38,
108,
109]. Therefore, in SD-NTNs, there are two key technologies essential for network slicing: the first is VNF deployment, which involves strategically placing virtual functions on suitable physical nodes to maximize service performance [
35]; and the second is flow routing, which requires networking schemes in SD-NTNs that meet both the SFC constraints and the multidimensional resource constraints [
35,
38].
5.2.1. Efficient VNF deployment
Different VNF deployment strategies are crucial for optimizing network performance in SD-NTNs [
35,
38,
92], which must align with the specific requirements of each service type, such as low-latency, high-data-throughput, and IoT or distributed services. The following are several VNF deployment cases based on the service characteristics.
(1)
Low-latency services. For network slices with strict low-latency requirements, such as those supporting URLLC services, VNFs should be deployed as close to the end-users as possible. In SD-NTNs, this can be achieved by deploying VNFs on HAPs, UAVs, or ground nodes near the end-users, thus significantly reducing transmission distance and minimizing propagation delays. This VNF deployment strategy ensures that latency-sensitive services, such as autonomous driving and remote medical procedures, maintain the ultra-low latency necessary for real-time data processing and decision-making [
91].
(2)
High data throughput services. For network slices supporting services that require high data throughput, such as eMBB services, VNFs should be deployed on physical nodes with significant processing, storage, and bandwidth capabilities. In SD-NTNs, this can be achieved by deploying VNFs on nodes with higher communication capacity, such as MEO or GEO satellites, where larger amounts of data can be transmitted more efficiently across vast areas [
92]. Additionally, deploying VNFs on central data centers or ground stations with abundant computing resources allows the network to handle data-intensive services, ensuring that large-scale data transmissions can occur without congestion or bottlenecks. This strategy supports high-throughput applications such as video streaming, augmented reality, and cloud-based services, ensuring minimal congestion and maintaining high-performance levels.
(3)
IoT and distributed services. For network slices supporting IoT and distributed services, such as mMTC services, VNFs should be deployed on nodes optimized for scalability and energy efficiency [
38]. In SD-NTNs, this can be achieved by deploying VNFs on LEO satellites, UAVs, or edge nodes that can support a large number of low-data-rate IoT devices. These nodes distribute processing closer to the devices, reducing latency and alleviating congestion in the core networks. By deploying VNFs across multiple nodes, the network can efficiently manage the high-density connectivity required by IoT devices, while optimizing resource allocation and minimizing energy consumption. This strategy supports seamless operation for applications such as smart city infrastructure, industrial automation, and environmental monitoring in vast, decentralized networks.
5.2.2. SD-NTN flow routing
In SD-NTNs, the mission flow must traverse each VNF in the specified sequence defined by the SFC [
35,
38,
108,
109]. However, due to the high-speed movement of satellites, UAVs, and HAPs, the connectivity between nodes in SD-NTNs is intermittent, resulting in time-varying topology [
44,
45,[
49], [
50], [
51]]. Therefore, flow-routing strategies satisfying SFC constraints, designed for static network topologies, are not directly applicable to time-varying SD-NTNs [
38,
42,
91]. In addition, due to the size, weight, and cost limitations of satellites, UAV, and HAPs, the payloads they can carry are strictly limited [
45,
51,
52]. Furthermore, in SD-NTNs, fulfilling complex service requests typically requires the collaboration of multidimensional heterogeneous resources—including communication, storage, and computing—across different nodes. Therefore, in time-varying SD-NTNs, it is essential to design optimized flow-routing strategies that not only meet SFC constraints but also ensure the efficient allocation of limited communication, storage, and computing resources. These strategies must consider the dynamic and intermittent connectivity between nodes in order to maximize resource utilization and improve network performance. One approach is to utilize a multifunctional time-expanded graph (MF-TEG) to model the time-varying SD-NTN topology with communication, storage, and computing resources [
110]. An MF-TEG can be further utilized to design VNF deployment and flow-routing strategies to ensure the efficient utilization of network resources and improve overall network performance.
6. Future research and open problems
6.1. Future Research
6.1.1. Interference management in NTNs
While the aforementioned studies have addressed intra-system interference within NTNs, the broader challenges of managing inter-system interference remain unresolved. In a 6G system, the co-existence of LEO satellites and UAVs in the same frequency bands transforms inter-system interference into a significant challenge. Some early work relied on cognitive radio approaches for inter-system interference migration, typically designating one system as primary and the other as secondary [[
111], [
112], [
113]]. These approaches aimed to maximize the throughput of the primary system while guaranteeing the QoS for the secondary. However, these efforts suffered from inherent limitations. By simply adopting intra-system interference management strategies such as non-orthogonal multiple access (NOMA) and one-layer rate-splitting multiple access (RSMA) for inter-system scenarios, they fail to fully exploit the potential of NTNs. Consequently, they often achieve suboptimal performance, either by prioritizing the primary system while neglecting the secondary’s potential or by constraining the secondary’s capabilities to minimize interference. This binary approach ultimately hinders the true potential of both systems and falls short of addressing the intricacies of inter-system interference in NTNs. To address this issue, an effective and efficient interference management strategy is necessary in future research. For example, in satellite multi-coverage environments, efficient interference management first necessitates capturing different physical features of nodes such as signal strengths and coverage durations; then, a focus on UE demands can help in designing a more refined and accurate interference management strategy.
6.1.2. Fine-tuned TN–NTN unification
The unification of TNs and NTNs goes beyond integrated architectures and connectivity management, extending to fine-grained orchestration of applications and services. Due to the ubiquitous coverage of NTNs, UEs typically experience coexisting coverage from both a TN and an NTN. In such scenarios, the NTN must play the role of service availability and reliability reinforcement, which is particularly valuable in applications such as highway vehicular networks. Additionally, as BSs support multicast-broadcast services (MBS), an NTN can enable service scalability by leveraging its natural wide coverage to provide efficient multicasting/broadcasting for data delivery in cooperation with TNs [
114,
115]. Thus, beyond merely supplementing connectivity, NTNs and TNs enhance the overall system performance through strategic service orchestration, leveraging their respective strengths.
6.1.3. Reliability-enhanced network slicing
In SD-NTNs, ensuring network reliability presents significant challenges due to the network’s heterogeneous composition, which includes satellites, UAVs, HAPs, and ground stations. These nodes are prone to disruptions caused by various factors such as hardware failures, link interruptions, environmental conditions, or mobility-induced disconnections. For example, satellites may encounter malfunctions or orbital adjustments, while UAVs and HAPs can experience power loss or weather-related disturbances. Ensuring that network slices remain resilient and can quickly recover from such failures is crucial for maintaining service continuity in SD-NTN. Future research should focus on developing robust, fault-tolerant mechanisms that can detect and respond to failures in real time, ensuring minimal service disruption. These strategies include dynamic reconfiguration of VNFs, allowing services to migrate seamlessly across available nodes, such as satellites, UAVs, and HAPs. In addition, real-time fault-detection strategies are essential for improving the overall reliability of SD-NTNs. These strategies should be used to continuously monitor network conditions and predict potential failures before they occur. By integrating predictive models and machine learning techniques, the network can proactively identify early signs of hardware malfunctions, link instability, or resource exhaustion. This enables preemptive actions, such as reconfiguring network slices or adjusting resource allocation, to prevent failure and enhance the overall reliability of the network.
6.2. Open problems
6.2.1. Integrated sensing and communications in NTNs
The integration of sensing and communication (ISAC) in NTNs is a highly promising advancement that holds the potential to revolutionize both communication and sensing capabilities in spaceborne and airborne platforms. By optimizing the spectrum, power, and hardware for dual functionality, ISAC enables NTNs to support tasks such as environmental monitoring, object detection, and surveillance within a unified framework [
95,
116]. Moreover, ISAC can drive NTN advancements in autonomous navigation, disaster management, and real-time situational awareness, significantly increasing operational efficiency. However, ISAC introduces major challenges into NTNs due to the dynamic propagation environment. Frequent HOs and long distances increase latency and complicate synchronization, while limited power further constrains ISAC operations. Achieving effective ISAC in NTNs will require advanced signal processing and adaptive beamforming techniques to manage these complexities.
6.2.2. TN–NTN competitive integration
The deployment and service of TNs and NTNs differ significantly in spatiotemporal aspects, including distribution patterns, fixed versus dynamic BSs, coverage status, and cell service durations [
8,
85]. When arranging both TNs and NTNs, operators must swiftly design transmission schemes and manage mobility for the heterogeneous system. A key challenge is the strategic assignment of transmission tasks for TNs and NTNs. Theoretically, any task that can be transmitted via a TN could also be offloaded to an NTN, given the global coverage. When TNs and NTNs alternately have transmission advantages, there must be a foresighted strategy for task division and joint transmission. Advanced computational and analytical technologies, such as federated learning and digital twins, may be of assistance here to ensure efficient data delivery.
6.2.3. Security and privacy in NTN slicing
In SD-NTNs, the diverse cyber threats inherent in the distributed architecture make it a significant challenge to increase security and privacy across network slices. These networks are particularly vulnerable to security risks such as eavesdropping, data tampering, and denial-of-service attacks, which are exacerbated by a shared infrastructure that increases the risk of cross-slice attacks and data exposure. Moreover, the time-varying topology of SD-NTNs further complicates real-time security measures. It is necessary to develop secure isolation mechanisms that ensure that data and control flows remain segregated even on shared resources, along with advanced encryption methods, authentication protocols, and secure resource-management strategies for the network dynamics. Furthermore, slice-specific security policies should be dynamically applied, ensuring that security measures adapt to the unique requirements of each slice. For example, slices handling URLLC services may require more stringent security protocols than slices designed for low-latency entertainment services.
7. Conclusions
In this paper, we explored the critical role of NTN deployment in the evolution of 6G mobile systems. Given satellites’ extensive coverage and rapid dynamics, we reviewed the unique challenges these pose to NTN networking, including the development of and issues related to radio access, satellite HO, and constellation slicing mechanisms. Furthermore, we introduced innovative perspectives on managing NTN radio resources, system mobility, and onboard traffic scheduling, aiming to achieve efficient, high-quality, and reliable global NTN networking and data services. Finally, we identified several open problems and highlighted useful techniques and directions for advancing the deployment of space Internet. It is our hope that this study provides valuable insights for the design, operation, and optimization of NTNs.
CRediT authorship contribution statement
Feng Wang: Writing – original draft, Methodology, Investigation, Conceptualization. Shengyu Zhang: Writing – original draft, Methodology. Huiting Yang: Writing – original draft, Methodology. Tony Q.S. Quek: Supervision.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
The author is an Editorial Board Member/Editor-in-Chief/Associate Editor/Guest Editor for this journal and was not involved in the editorial review or the decision to publish this article.
Acknowledgments
This work was supported by the National Research Foundation, Singapore and Infocomm Media Development Authority under its Future Communications Research & Development Programme.
PDF
