Open Access
Add to BibSonomy
Abstract
The low-orbit satellite communication can provide users with low-delay and ultra-wideband communication services worldwide. By the wide available bandwidth and immunity to interference, free-space optical (FSO) feeder links are presented as an ideal alternative to radio frequency (RF) feeder links for satellite-to-ground backhaul, and the space-ground integrated optical network (SGION) is gradually formed by using FSO feeder links to integrate the low-orbit satellites and the terrestrial optical network (TON) for backhaul services. The propagation channel characteristics of FSO feeder links vary significantly during low-orbit satellite passes, and atmospheric turbulence causes serious scintillation in FSO feeder links at low elevations, increasing link budget and restricting link capacity. Limiting the observation range of optical ground stations (OGSs) to the high elevation area can establish high-capacity feeder links to improve SGION’s throughput, but inevitably increase the network dynamics and reduce the satellite visibility. This paper trades off SGION’s throughput and dynamics by planning FSO feeder link handover and capacity adjustment. Two baseline schemes and the feeder link handover and capacity adjustment scheme based on non-dominated sorting genetic algorithm (NSGA-FLHCA) are proposed. By finding the Pareto edge of the multi-objective optimization problem, NSGA-FLPCA is more effective in improving network throughput and reducing network dynamics compared with two baselines.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
With the development of optical fiber communication technology, terrestrial optical networks (TONs) can provide reliable broadband services to most people [1]. However, due to the high construction cost and difficulties of network facilities, TONs cannot provide communication services in some remote areas [2]. The development of key technologies, such as aerospace crafts, space communication, and on-board chip, enables the satellite communication to provide a large-capacity and high-flexibility service for ground users [3]. Low-orbit satellites from 500 to 2,000 km above the surface of the Earth have minor transmission delays and broad global coverage [4], ideal for comprehensive covering and low latency communication. At present, some countries and companies have constructed a series of satellite constellations with specific abilities. Leading companies such as Airbus and SpaceX have achieved the wide-area communication coverage through their high-density satellite constellation [5–7]. The development of beamforming and anti-interference technologies will further improve the low-orbit satellite’s communication capacity [8,9].
Limited by geographical and political factors, ground stations are generally deployed on some specific places. The global-distributed traffic converging to the ground stations may result in severe congestion in feeder links [10]. The increasing communication capacity of the low-orbit satellite puts the high-capacity requirement to feeder links. The conventional radio frequency (RF) solutions for feeder links operating at the Ku-band (12-18 GHz), the Ka-band (27-40 GHz), and the Q/V band (40-50 GHz) have the bandwidth limitation and the high risk of interference, requiring to deploy numerous ground stations for satellite-to-ground delivery [11]. Free-space optical (FSO) technologies are being substantially considered as an attractive alternative to the existing RF feeder links for satellite communication, holding the advantages of wide available bandwidth without restriction, low power consumption, and immunity to interference [12]. The space-ground integrated optical network (SGION) is gradually formed by using the FSO feeder links to integrate the TON and low-orbit satellites for backhaul services. Moreover, the FSO inter-satellite links enable the indirect satellite-to-ground delivery for satellites without establishing feeder links, further improving the SGION communication coverage.
The space-ground links providing backhaul services include the user links between satellites and remote terminals, and the feeder links between satellites and ground stations. Considering the random distribution of remote terminals, the RF communication is an ideal choice for user links, because it can provide wide communication coverage for randomly distributed terminals. Unlike the remote terminals, ground stations have the fixed geographical locations, which enables to use line-of-sight FSO links to provide large-capacity feeder communication. Establishing FSO feeder links for low-orbit satellites in the SGION requires to consider the mutual movement between satellites and optical ground stations (OGSs). Moreover, the propagation channel characteristics of FSO feeder links vary significantly during satellite passes, and the caused link impairments strongly depend on the elevations of FSO feeder links [13], increasing link budget and restricting link capacity. Besides free space loss, atmospheric turbulence seriously affects the FSO channel qualities, which is also closely related to the elevations of FSO feeder links. Limiting OGSs’ observation range within the high elevation ranges to establish high-capacity FSO feeder links can improve SGION’s throughput, but this may result in frequent feeder link handover, aggravating SGION’s dynamics. For instance, Acquisition, Pointing, and Tracking (APT) must be required for each feeder link handover. Additionally, due to the frequent variations in connectivity relationships of feeder links, numerous routing tables are required for rerouting, resulting in the significant increase of storage load [14]. Currently, some researches focus on using variable data rate (VDR) architectures to optimize link capacities based on FSO channel characteristics [15–19]. Adaptive-rate FSO terminals have been proposed and developed in the NASA Laser Communications Relay Demonstration (LCRD) [16–19]. However, the time-varying feeder link capacity may lead to traffic lose and invalid routing, and being aware of changes in feeder link capacities in advance for rerouting is also necessary for the satellite-to-ground delivery in the SGION using VDR FSO feeder links.
How to improve the throughput while reducing network dynamics is considered as a type multi-objective optimization problem. This paper proposes two baseline schemes for planning FSO feeder links, which focus on improving feeder link capacity and reducing network dynamics, respectively. Further, we develop the feeder link handover and capacity adjustment scheme based on non-dominated sorting genetic algorithm (NSGA-FLHCA) to trade off SGION’s throughput and dynamics. By finding the Pareto edge of the multi-objective optimization problem, NSGA-FLHCA can provide the solution for planning VDR FSO feeder links to balance the SGION’s throughput and dynamics.
The rest of this paper is organized as follows. Section 2 is the problem description for planning VDR FSO feeder links. Section 3 develops two baselines and NSGA-FLHCA. The performance evaluation and numerical analysis are presented in Section 4. Finally, we discuss and conclude this paper with future research in Section 5.
2. Problem description
This section first presents the network compositions in the SGION using VDR FSO feeder links, and then introduces the elevation-dependent losses in the FSO feeder link. Finally, we partition the OGSs’ observation range for planning VDR FSO feeder links. Some acronyms are listed in Table 1 in order of appearance in this paper.
Table 1. Acronym List
2.1 Compositions of SGIONs
As illustrated in Fig. 1, the SGION utilizes OGSs and low-orbit satellites to realize the backhaul services from remote terminals to the TON. The OGS refers to an optical gateway with the capability of satellite-to-ground delivery using FSO feeder links. In TONs, some optical nodes configured with FSO terminals are designated as OGSs. Fiber links connecting optical nodes in TONs ensure that multiple OGSs can be used for the satellite-to-ground delivery simultaneously [20]. Considering that FSO links have the immunity to interference, we assume that the number of FSO feeder links one OGS can establish only relates to the satellite visibility and the number of FSO terminals. OGSs continuously establish and remove feeder links with visible satellites to maintain the satellite-to-ground delivery. The process of feeder link update is defined as feeder link handover. Due to the limited number and the uneven distribution of OGSs, not all low-orbit satellites are able to establish feeder links with OGSs. And the inter-satellite links enable low-orbit satellites without feeder links can indirectly deliver services to the ground by inter-satellite communication. This paper uses the grid-mesh pattern to describe the connectivity relationships of inter-satellite links for the low-orbit satellite, where each low-orbit satellite has at most four inter-satellite links (two in-orbit links and two inter-orbit links).
Due to the mutual movement between low-orbit satellites and OGSs, the propagation channel characteristics of FSO feeder links vary significantly during satellite passes. The VDR architecture implemented by changing chip rate [16,21] or resource data rate [15,22] can optimize the link capacity to establish the VDR FSO link. The resource data rate adaptation keeping the fixed chip rate can achieve seamless capacity change through digital baseband operations, while the variable chip rate requires hardware modifications and resynchronization at FSO transceivers. Some works discussing possible implementations of the VDR approach claim that, doubling the symbol period (or spreading/repeating each symbol) can have a significant improvement of up to 3 dB in average received optical power [15–18]. This paper considers implementing the capacity adjustment of FSO feeder links by spreading symbols at different elevations, assuming that appropriate channel interleaving and forward-error-correction (FEC) coding are applied in FSO feeder links, and each satellite can perform the relay function of decoding and forwarding. The criterion for establishing the FSO feeder link successfully is the RX optical power exceeding the Adding-White-Gaussian-Noise (AWGN) threshold under corresponding spreading factor. Considering the time-varying connectivity and capacity of VDR FSO feeder links, being aware of feeder link changes in advance for rerouting is necessary for the satellite-to-ground delivery. Currently, the snapshot policy is commonly to characterize the dynamic network within a short time. A snapshot is defined as the topology of the SON in a period of time, during which all the links are maintaining a fixed state. The snapshot division is defined as the process of determining the start and end time of each snapshot based on link changes, which facilitates network management for routing adjustment. In this paper, only the feeder link change is considered in the snapshot division. The snapshot division that only considers the link connectivity change is not suitable for the SGION using VDR feeder links, as variations in the link capacity may lead to traffic loss and invalid routing. This paper considers the connectivity and capacity changes in VDR FSO feeder links to generate the SGION snapshots, and the number of generated snapshots is used to quantitatively measure the SGION dynamics.
2.2 Elevation-dependent losses
The FSO channel characteristics vary significantly during satellite passes. Elevation-dependent loss LE includes free space loss LFS, atmospheric attenuation LA, and scintillation loss LS. LFS given by Eq. (1) refers to the power loss on the light propagation in an ideal transmission medium, which is determined by propagation distance D and wavelength λ. The distance of the FSO feeder link at elevation angle θ is $D = \sqrt {R_{ER}^2{{\sin }^2}\theta + {H^2} + 2H{R_{ER}}} - {R_{ER}}\sin \theta$, where RER is the earth’s radius and H is the satellite altitude.
Three main factors causing atmospheric attenuation LA are gas molecule absorption, aerosol scattering, and meteorological conditions (such as rain and snow) [23]. LA is calculated using Eq. (2), where $\phi$ represents the zenith angle and $\gamma (\lambda ,h)$ represents the attenuation index. Zenith angle $\phi$ and elevation angle $\theta$ can be converted by Eq. (3).
Scintillation loss LS is caused by atmospheric index-of-refraction turbulence, and its strength depends on height profile, receiver aperture size, wavelength, and link elevation. Assuming Kolmogorov turbulence spectrum for the optical turbulence [24], the intensity scintillation index can be calculated using Eq. (4), where $C_n^2$ profile is described with the Hufnagel-Valley model, ${k_{wave}}$ is wavelength number, and h0 is OGS’s altitude.
For FSO feeder links, ground receiver’s received power scintillation index $\delta _P^2$ is used to estimate scintillation loss LS. The ground receiver’s aperture averaging effect relates $\delta _I^2$ (intensity scintillation index) to $\delta _P^2$ (received power scintillation index) with aperture diameter d by ${A_f}(d) = \delta _P^2(d)/\delta _I^2$. The derivation for LS with a lognormal distribution can be approximated by Eq. (5), where ρthr is the loss fraction and is generally set as 10−6 (as can be recovered by state-of-the-art FEC coding with interleaving) [25].
In addition to elevation-dependent loss, there is also constant loss LC in FSO feeder links, which is not affected by elevation changes, mainly including transmitter-receiver loss LTR at the feeder links’ transmitter and receiver, coupling loss LCO at receiver, and modem implementation loss LMI at receiver. These are elevation-independent losses in FSO feeder links, and this paper sets the fixed values for them in the link budget of FSO feeder links.
2.3 Observation range partitioning
The minimum elevation angle (MEA) determines OGSs’ observation range. To plan FSO feeder link handover and capacity adjustment, OGSs’ observation range can be partitioned into multiple parts based on different criteria, such as equal time intervals for satellite passes, equal loss intervals, or equal elevation intervals. L0 (dB) denotes the elevation-dependent loss at MEA θ0. We partition OGSs’ observation range using a 3 dB elevation-dependent loss interval and adjust the spreading factor configuration in feeder links to adapt to the FSO channel changes. As shown in Fig. 2, after partitioning OGSs’ observation range based on the elevation-dependent loss, we can determine the elevation range of each partitioned part. The configured spreading factor in the FSO feeder link must ensure that the RX optical power exceeds the corresponding AWGN threshold. For the VDR FSO feeder link, the spreading factor can be adjusted when the access satellite is in different partitioned parts to change the link capacity and adapt to the FSO channel.
3. Baselines and NSGA-FLHCA
This section first presents two optimization objectives for planning VDR FSO feeder links in the SGION, and then introduces two baselines and NSGA-FLHCA.
3.1 Optimization objectives
In this paper, the total capacities of FSO feeder links and the number of generated snapshots in the given time are used to quantitatively measure the network throughput and dynamics in the SGION. The related notations and variables are shown in Table 2.
Table 2. Notations and Variables for Planning VDR FSO Feeder Links
In Table 2, $e_{k,i}^{s,g}$ is a binary constant indicating the partitioned part of OGS’s observation range where the low-orbit satellite locates, obtained by querying ephemeris within a given time T. This paper focuses on the VDR FSO feeder link implemented of spreading symbols, and $h_{k,j}^{s,g}$ is a binary constant that indicates whether the jth spreading factor is available for establishing FSO feeder link, as calculated by Eq. (6).
Improving the capacity sum of feeder links and reducing generated snapshots within the given time are two objectives for planning VDR FSO feeder links in the SGION (the changes in connectivity relationship or link capacity of any FSO feeder link will generate a new snapshot). Considering that the APT process of FSO feeder link handover may cause the transient link unavailability, we calculate the average capacity sum of feeder links within the given time as the one objective using Eq. (7), where Δt denotes the time of APT process, $({t_k} - y_k^{g,n} \times \Delta t)$ is the link availability time, and $B/sf_k^{g,n}$ is the current link capacity when the spreading factor of $sf_k^{g,n}$ configured in the FSO feeder link. Due to that the PAT delay of the feeder link handover at low elevations is greater than that at high elevations, this paper sets an approximation to distinguish the PAT time of access satellites at different elevations as Δt = t0-ηθ, where t0 is the time to adjust the optical terminal to the horizontal direction (maximum PAT time), η is a constant weight in seconds per degree, and θ is the next access satellite’s elevation in degree. In this paper, we set t0 = 10 Sec and η=0.1 Sec/Deg. The number of snapshots generated within the given time is calculated by Eq. (8) as another objective.
The capacity sum of feeder links is not exactly equivalent to SGION’s throughput, but it can be used as an important criterion to trade off throughput and dynamics in the SGION using VDR FSO feeder links. In addition to feeder link capacities, uplink capacities and inter-satellite link capacities also determine the SGION’s throughput. S’ denotes the set of access satellites of OGSs for establishing feeder links, and Es is the set of inter-satellite links of access satellite s$\in$S’. The theoretical maximum of SGION’s throughput Otheory can be expressed by Eq. (9), where Cfeeder, Cinter, and Cuplink denote the capacities of the FSO feeder link, the FSO inter-satellite link, and the RF uplink, respectively. To ensure that the capacity sum of FSO feeder links is a reasonable measure of SGION’s throughput, Eq. (10) limits the maximum capacity of each FSO feeder link based on the capacities of RF uplinks and FSO inter-satellite links (in the grid-mesh pattern, each access satellite at most establishes four inter-satellite links with adjacent satellites).
The OGS can only establish FSO links with its available satellites. $S_{g,n}^k \subseteq S$ denotes the available satellites for the nth FSO terminal on OGS g at the kth moment, and each satellite $s \in S_{g,n}^k$ must be observed by g and hasn’t established the FSO feeder link with other OGSs’ FSO terminals at the kth moment. Assuming there is at most one FSO feeder link on each satellite, the corresponding constraints for establishing FSO feeder links are presented in Eqs. (11-13). The criterion for establishing the FSO feeder link successfully is the RX optical power exceeding the AWGN threshold under its used spreading factor. $J_{g,s}^k = \{ j|h_{k,j}^{s,g} > 0,\forall j \in J\}$ is the index set of available spreading factors for the FSO feeder link from access satellite s to OGS g at the kth moment.
Binary variables $y_k^{g,n}$ and $v_k^{g,n}$ are used to express the changes in the connectivity relationships and link capacities in VDR FSO feeder links. When the feeder link on OGS g changes its connectivity or capacity at the kth moment, $y_k^{g,n}$ or $v_k^{g,n}$ is set to 1 correspondingly. Binary variable ${w_k}$ is used to judge whether a new snapshot needs to be generated for the SGION at the kth moment, as detailed by Eq. (14).
3.2 Two baselines
Two baseline schemes are proposed in this paper for planning FSO feeder links, namely the minimum-snapshot scheme (MSS) and the maximum-capacity scheme (MCS). Two baselines focus on different objectives for planning FSO feeder links. To minimize the number of generated snapshots, MSS first plans the feeder link handover with the least changes of access satellites for OGSs, and then reduces the capacity changes in VDR FSO feeder links. MSS gets the satellite change sequence of feeder link handover using Algorithm 1.
$S_{g,n}^k$ can be obtained by traversing whether $x_{k,s}^{g^{\prime},n}$ and $e_{k,i}^{s,g}$ of each satellite satisfy Eqs. (11-13). qg,n = [$s_{g,n}^1$, $s_{g,n}^2$, …, $s_{g,n}^{|K|- 1}$, $s_{g,n}^{|K|}$] denotes the satellite change sequence for feeder link handover, where $s_k^{g,n}$ is the access satellite at the kth moment. In Fig. 3, the k0th moment serves as the initial moment, and R = [s1, s2, s3, s4] and DS = [k0] are initialized at the k0th moment, which are changed at different moments to record the satellite visibility changes and the corresponding moments. Before R being empty, one satellite in it is chosen to access g for feeder link handover, and the access time can be obtained in RS. R is [s4] at the (k0 + 2)th moment and empty at the (k0 + 3)th moment, so we choose s4 to access g at the k0th moment. The next moment to change the access satellite is the (k0 + 3)th moment, and R and DS are re-initialized as [s5,s6,s7] and [(k0 + 3)] at this moment. After getting the satellite change sequence of feeder link handover, MSS configures the fixed spreading factor in feeder links to minimize the link capacity changes.
OGSs may observe the available satellites in partitioned parts with different elevation-dependent losses. For MCS, the available satellite in the partitioned part with the minimum elevation-dependent loss is chosen as the access satellite for the OGS to establish the feeder link, and the minimum spreading factor meeting the AWGN requirement in this partitioned part is configured in the FSO feeder link. If multiple available satellites are in the partitioned part with the least elevation-dependent loss at the same time, MCS chooses the one with the longest visible time to access the OGS. MCS gets the satellite change sequence of feeder link handover using Algorithm 2, which is presented as follows.
After getting the satellite change sequence of feeder link handover, MCS configures the minimum spreading factor meeting the AWGN requirement in the FSO feeder link, which can maximize the link capacity.
3.3 NSGA-FLHCA
For the optimization problem with multiple objectives, it is a common method to find the Pareto edge in the optimization problem’s feasible solution space and choose the solution in the Pareto set according to different criteria for objectives. Based on two baselines MSS and MCS, we design NSGA-FLHCA to find the Pareto sets and the Pareto edges for planning VDR FSO feeder links. A solution in NSGA-FLHCA is represented by vector X = [S, SF], where S is the vector consisting of OGSs’ access satellite sequences, and SF is the vector consisting of OGSs’ spreading factor configuration sequences. The length of X is |X|=2*|G|*N*|K|, where |G| is the number of OGSs, N is the number of FSO terminals on each OGS, and |K| is the number of partitioned moments within the given time T. The evaluation of two objective functions f1 and f2 produces the corresponding value f(X)= (f1(X), f2(X)), i.e., f1() and f2() represent the two components of f(). Let Xa and Xb denote two different link planning solutions, Specifically, f(Xa) = (f1(Xa), f2(Xa)) and f(Xb) = (f1(Xb), f2(Xb)). The values of f(Xa) and f(Xb) are used to evaluate two solutions Xa and Xb. The concept of dominating is defined as: (i) Xa dominates Xb if no component of f(Xa) is better than the corresponding component of f(Xb), and at least one component is absolutely better; (ii) All non-dominated solutions are the optimal solutions of the problem, the set of these solutions is the Pareto set, and the corresponding objective values are the Pareto edges. The flowchart of NSGA-FLHGA is shown in Fig. 4. First, NSGA-FLHGA processes individual coding for FSO feeder link planning. Then, two baselines are used to generate multiple individuals for population initialization. Non-dominated sorting is to evaluate individuals according to two objective functions f1 and f2. In selection, some individuals are chosen as parents of the next generation of individuals, where gene mutation and crossover are key operations to generate new individuals. After non-dominated sorting and fitness assessment, the new individuals merged with the individuals in the previous generation, which is generation merge (excellent individuals have the greater probability to be remained to the next generation). By repeating generation iterations, the fitness of remaining individuals continues to increase.
Individual coding: We take each solution as one individual, and the individual coding under N = 2 and |G|=4 is shown in Fig. 5. The feeder link configuration is coded as the gene, which refers to the access satellite and the spreading factor used to establish the FSO feeder link at a single moment. S[g,n,k] and SF[g,n,k] denote the access satellite and the spreading factor used to establish the feeder link of the nth FSO terminal on OGS g at the kth moment. Chromosome can be considered as one gene combination which consists of the genes at different moments, and the moment index is set as the gene index at each chromosome. An individual is a chromosome combination that contains the feeder link configurations of all OGSs in the SGION at different moments.
Population initialization: For initializing population, we use MSS and MCS to generate different individuals. Population initialization is detailed as Algorithm 3, where ψ denotes the number of individuals in each population.
In Step 6, ρ is a scheme probability to choose different baselines to initialize individuals.
Non-dominated sorting: We set different non-dominated ranks to sort each individual (i.e., rank 1 is the best). Algorithms about non-dominated sorting are referred to [26].
Crowding degree: To ensure individual diversity in population, the crowding degree is used to evaluate individuals under the same rank. The calculation of one individual’s crowding degree is to sum the normalized difference between two adjacent individuals under different objectives. Under the jth objective (j$\in$ [1,2]), individuals are sorted according their value of objective function fj(). For individual ${\textrm{X}^i}$, let $\textrm{X}_j^{i - 1}$ and $\textrm{X}_j^{i + 1}$ denote two adjacent individuals in the sorting under the jth objective, i.e., ${f_j}(\textrm{X}_j^{i - 1}) \le {f_j}({\textrm{X}^i}) \le {f_j}(\textrm{X}_j^{i + 1})$. The crowding degree of ${\textrm{X}^i}$ is calculated as $CD({\textrm{X}^i}) = \sum\limits_{j \in [1,2]} {norm({f_j}(\textrm{X}_j^{i - 1}) - {f_j}(\textrm{X}_j^{i + 1}))}$, where norm() is the normalization function. Under the same rank, the individual having a higher crowding degree is preferred.
Gene mutation: Gene mutation is performed by adjusting the access satellites of the terminal on OGSs, as shown in Fig. 6. The original individual will produce a new individual after gene mutation. Gene mutation is detailed as Algorithm 4.
In Algorithm 4, δthr is the probability threshold of gene mutation, which is set as 0.1. To accelerate the convergence of NSGA-FLHCA, gene mutation is to change the access satellite at several consecutive moments rather than a single moment, as detailed in Steps 6-7. To replan feeder link handover using two baselines, the newly chosen access satellites must ensure that they are visible for OGSs.
Crossover: Crossover is the exchange of two chromosomes with the same index in different individuals, as shown in Fig. 7. Crossover is detailed as Algorithm 5.
In Algorithm 5, φthr is the probability threshold of crossover, which is set as 0.2. For crossover, two new individuals must satisfy Eq. (11) to ensure that each satellite establishes at most one FSO feeder link to OGSs.
Selection: The next population is formed by excellent individuals under the front ranks, individuals under the same rank are selected by their crowding degree.
Fitness assessment: Each individual is assessed according to the rank and crowding degree.
Generation merge: In generation merge, the number of individuals in each generation is fixed. The fitness is used in roulette to select the next generation of individuals from the newly generated and parent individuals, where roulette can reduce the probability of the population iteration falling into the local optimization.
The time complexities of Algorithm 1 and Algorithm 2 of MSS and MCS are both O(|K|*|S|), where |K| is the number of time indexes, and |S| is the number of satellites. Based on the time complexities of Algorithm 1 and Algorithm 2, the time complexities of population initialization, gene mutation, and crossover in NSGA-FLHCA are all O(|K|*|S|*|J|), and the time complexity of NSGA-FLHCA is O(Λ*(ψ*|G|*N*|K|*|S|*|J|+ψ2)), where Λ is the number of population iterations, ψ is the number of individuals in each population, |G| is the number of OGSs, N is the number of terminals in each OGS, |J| is the number of spreading factors, and O(ψ2) is the time complexity in sorting. NSGA-FLHCA can achieve high-performance solutions by emulating population evolution. However, the population iterations and individual operations make a high time complexity in NSGA-FLHCA.
4. Performance evaluation
In this section, numerical results and analysis are presented to demonstrate the performance of different schemes for planning VDR FSO feeder links.
4.1 Simulation setup
To simulate atmospheric attenuation, we set the atmospheric visibility height as 30 km and only consider the atmospheric attenuation caused by aerosol scattering. To simulate atmospheric scintillation, the $C_n^2$ profile of the refractive index structure parameter $C_n^2(h)$ described as HV5/7 model [27] is used in this paper. The intensity structure size parameter in the FSO receiver is derived in a rough approximation through the Fresnel size given by the distance from a dominant turbulent layer, and the height of the dominant turbulent layer is assumed as 12 km referring to [28]. The height difference between OGSs is ignored in this paper. For the low-orbit satellite distribution, two constellations Starlink and Iridium are considered in this paper. Some configuration parameters used in the simulation are shown in Table 3. Considering that the FSO feeder link may be affected by the cloud shading, the cloud margin Mcloud of 3 dB is required in the feeder link budget, and we classify this margin into the constant loss of feeder links for simplicity. This paper assumes that there are some OGSs deployed in the gateway nodes in the TON. Figure 8 shows the topology of the TON national science foundation network (NSFNET), and the nodes 0-13 are the gateway nodes of NSFNET. Considering the limitation of the number of satellite-ground communication terminals and on-board energy in low-orbit satellites, we set that each satellite can only establish a feeder link. For OGSs with close geographic locations, there may be some overlap between their observation areas. OGSs with overlapping observation areas may compete for the same satellite, and this condition easily happens in the SGION with low satellite density. To compare the performance of two baselines and NSGA-FLHCA when there are different degrees of satellite competition between OGSs, we simulate the feeder link planning under different OGS deployments, and two deployments of four OGSs in NSFNET are shown in Fig. 8. The difference between two deployments is that the OGS distribution of deployment-2 is more dispersed than that of deployment-1, and the OGSs in deployment-1 are more likely to compete the same satellites for establishing feeder links.
Table 3. Configuration Parameters
4.2 Link budget of FSO feeder links
Figure 9 shows the total losses in FSO feeder links versus elevation in different constellations. As mentioned in Section 2.2, the elevation-dependent losses include the free space loss, the atmospheric attenuation, and the scintillation loss. The elevation-independent losses are considered constant in this paper, including transmitter-receiver loss LTR, coupling loss LCO, and modem implementation loss LMI. The elevation-independent losses are set as the fixed values in simulation. Obviously, the high constellation aggravates the elevation-dependent losses of FSO feeder links, which are more severe at small elevations. This paper only considers planning VDR FSO feeder links under ideal clear weather conditions, when free space loss accounts for a major portion of the link budget. Under the strong turbulence condition, atmospheric attenuation and scintillation loss rise sharply and even cause the link unavailability. The smart ground diversity can alleviate this problem, which is not covered in this paper.
4.3 Capacity range of FSO links
The FSO inter-satellite link capacity is also an important factor affecting SGION’s throughput. It should be noted that the FSO inter-satellite links in two constellations have different distance ranges. According to the data in Satellite Tool Kit (STK), the distance range of inter-satellite links in Iridium is approximately [1700km, 4400 km], and that in Starlink is [350 km, 610 km]. We assume that FSO inter-satellite links’ losses only include constant losses and free space loss. The range of FSO inter-satellites links’ free space loss in Iridium is approximately [262.756 dB, 271.026 dB], and that in Starlink is [249.036 dB, 253.856 dB]. Assuming that the same FSO terminals are used in inter-satellite and feeder links, the range of FSO inter-satellite links’ RX optical power in Iridium is [-41.726 dB, -33.456 dB], and that in Starlink is [-24.556 dB, -19.736 dB]. This paper sets the values of spreading factors as 16, 8, 4, 2, and 1, and the corresponding AWGN thresholds on RX optical power are -49.2dBm, -46.2dBm, -43.2dBm, -40.2dBm, and -37.2dBm, respectively. To set the constant capacity of the inter-satellite link, the fixed spreading factor is configured in inter-satellite links according to the maximum free space loss. In Iridium, the spreading factor of 4 is configured in inter-satellite links and the corresponding capacity is Cinter = B/4 = 5Gbps. In Starlink, the spreading factor of 1 is configured in inter-satellite links and the corresponding capacity is Cinter = B = 20Gbps. For the SGION following the grid-mesh pattern, the capacity requirement of FSO feeder links doesn’t exceed (4×Cinter + Cuplink). In OGSs’ observation range, multiple adjacent partitioned parts offering the link capacity exceeding (4×Cinter + Cuplink) can be integrated into one large part and configure the same spreading factor to reduce the planning complexity.
4.4 Elevation ranges of partitioned parts
According to RX optical powers of FSO feeder links at different elevations and the AWGN thresholds of different spreading factors, we partition OGSs’ observation range into multiple parts, as shown in Fig. 10. The elevation ranges of partitioned parts in Iridium and Starlink are shown in Table 4. Due to the height difference between two constellations, OGSs can establish FSO feeder links at lower elevations in Starlink compared to Iridium. Figure 11 shows the elevation changes of OGSs’ visible satellites versus time in two constellations, where the simulated OGS is deployed on the optical node Boulder in NSFNET. Due to the higher satellite density, the OGSs in Starlink can observe more low-orbit satellites at high elevations, which allows OGSs to continuously establish high-capacity FSO feeder links with low elevation-dependent losses.
Table 4. Elevation Ranges of Partitioned Parts
4.5 Performance of NSGA-FLHCA
The best edges of population generations directly reflect the performance of NSGA-FLHCA. We first take the OGS deployment-1 in Iridium as the example.
The best edges under different generations and individuals are compared in Figs. 12 (a-d), where the horizontal coordinate represents objective f1, the vertical coordinate represents objective f2, and C0 = B/sfmax is the FSO feeder link capacity configured with the maximum spreading factor sfmax. We can find that two objectives are positively correlated, and this verifies the improvement of SGION’s throughput is often accompanied by intensifying dynamics. With the increase of generations, objective f1 increases to a certain extent, and this represents the expansion of the search domain of NSGA-FLHCA by gene mutation and crossover. The increase in individuals makes the best edge of each generation clearer. When the number of individuals is 100, it can be clearly observed that the best edge gets closer to the bottom right of the solution domain with the generation increasing, and the individuals under the higher generation dominate those under the lower generation. As shown in Fig. 12 (d), the nodes (2.77, 22) and (16.38, 113) are the elevation values of two baselines MSS and MCS, and the best edges of each generation are on the lower right of the straight line passing these two nodes, which means NSGA-FLHCA performs better than two baselines.
Fig. 12. Best edges in the NSGA-FLHCA, (a) 10th\20th\30th\40th\50th generation and 20 individuals, (b) 10th\20th\30th\40th\50th generation and 100 individuals, (c) 30th\60th\90th\120th\150th generation and 20 individuals, (d) 30th\60th\90th\120th\150th generation and 100 individuals.
To further evaluate NSGA-FLHCA’s performance, we take objective f1 as the revenue and objective f2 as the cost, and the solutions in the best set of NSGA-FLHCA are evaluated by f’=f1-α×f2, where α is the cost weight. To set weight α reasonably, we compare the ratio of f1 to f2 under two baselines, as shown in Fig. 13. In Iridium, the revenue-to-cost ratios of two baselines tend to be the same. However, in Starlink, the revenue-to-cost ratios differ greatly, this is because each OGS has sufficient visible satellites in Starlink, and MCS can find the solution with a lower cost while increasing the revenue.
Based on the revenue-to-cost ratios of two baselines, we respectively set different weights to choose the solution with greatest f’ in the best edge. In Table 5, α1, α2, and α3 are three weights in f’, and β1 and β2 are the revenue-to-cost ratios of MSS and MCS.
Table 5. Objective Weight Setting
We select the solution with the greatest f’ in the best edge to compare the performance with two baselines under different conditions. Performance comparison in different constellations is shown in Figs. 14 and 15. Due to that Starlink has higher satellite density than Iridium, there are sufficient visible satellites at the partitioned part with low elevation-dependent losses. Therefore, in Starlink, the performance of MSS and MCS are very close to that of NSGA-FLHCA with weights of α3 and α1. In Iridium, multiple OGSs may compete for the same visible satellite. After using MSS and MCS to plan access satellites for some OGSs, there may be no available satellites leaving for other OGSs. NSGA-FLHCA eases this problem through gene mutation and crossover. In Iridium, the performance of NSGA-FLPCA with weights of α3 and α1 are obviously better than MSS and MCS. Compared with MSS, NSGA-FLPCA with weight α3 improves network throughput by an average of 8.01% and reduces snapshots by an average of 5.3%. Compared with MCS, NSGA-FLPCA with weight α1 improves network throughput by an average of 7.05% and reduces snapshots by an average of 2.85%. Compared with MSS, NSGA-FLPCA with weight α2 improves network throughput by more than 200% with the cost of increasing 114.29% snapshots. Compared with MCS, NSGA-FLPCA with weight α2 reduces snapshots by 57.14% but reduces network throughput by 41.91%. In addition to the performance improvements than baselines, NSGA-FLHCA also provides referable solutions of feeder link handover and capacity adjustment for the SGION requiring different network throughputs.
Fig. 14. Simulation results under OGS deployment-1 in Iridium (N = 1), (a) objective f1, (b) objective f2.
Fig. 15. Simulation results under OGS deployment-1 in Starlink (N = 1), (a) objective f1, (b) objective f2.
The performance comparison under different OGS deployments is shown in Figs. 14 and 16. The OGS deployment-2 is more dispersed than deployment-1, which eases the competition for available satellites among different OGSs, improving objective f1. The performance comparison under different numbers of FSO terminals on each OGS is shown in Figs. 16 and 17. Being consistent with the above analysis, multiple OGSs may compete for the same visible satellite in Iridium. This situation may be worsened as more FSO terminals configured on OGSs. As shown in Figs. 16 (a) and 17 (a), increasing FSO terminals does not significantly improve f1, which indicates that the satellite density is also an important factor limiting SGION’s throughput.
Fig. 16. Simulation results under OGS deployment-2 in Iridium (N = 1), (a) objective f1, (b) objective f2.
Fig. 17. Simulation results under the OGS deployment-2 in Iridium (N = 2), (a) objective f1, (b) objective f2.
5. Discussion and conclusion
5.1 Discussion
In the SGION using VDR FSO feeder links, the increase of network dynamics inevitably has an adverse effect on Quality of Services (QoS), that is, services require to be rerouted due to the changing link relationship and link capacity, thus aggravating delay jitter. How to process QoS routing in the SGION using VDR FSO feeder links is the focus of our next research. In addition, the free space loss of inter-satellite links is time-vary due to the mutual movement between adjacent satellites. For simplicity, we only focus on the influence of changing feeder link capacities on the SGION’s throughout, keeping the constant capacity of inter-satellite links by configuring the fixed spreading factor based on the maximum free space loss. How to further adjust inter-satellite link’s capacities according to the changing free space loss to improve the SGION’s throughput is the difficulty of future research.
5.2 Conclusion
This paper trades off the network throughput and dynamics in the SGION using VDR FSO feeder links, where the network throughput has a positive correlation with dynamics. Specifically, increasing the capacity sum of feeder links requires more frequent link handover and spreading factor adjustment. Two baselines and NSGA-FLHCA are proposed in this paper for planning FSO VDR feeder links to balance SGION’s throughput and dynamics. Compared with two baselines, NSGA-FLPCA is more effective in improving network throughput and reducing network dynamics. On the network throughput in SGIONs, NSGA-FLPCA has an average improvement of 8.01% than MSS and an average improvement of 7.05% than MCS. On the dynamics in SGIONs, NSGA-FLPCA has an average reduction of 5.3% than MSS and an average reduction of 2.85% than MCS. In addition, NSGA-FLHCA also provides referable solutions for different requirements of network throughput in SGIONs, which is to reduce the network dynamics while ensuring that the total capacities of FSO feeder links meet the requirement of network throughput.
Funding
National Natural Science Foundation of China (62125103, 62171050); Fund of Key Laboratory of Computer System and Architecture (CARCH201906); State Key Laboratory of Information Photonics and Optical Communications (IPOC2021ZT15); BUPT Excellent Ph.D. Students Foundation (CX2023232).
Acknowledgments
The authors thank the National Natural Science Foundation of China (62171050, 62125103), the Fund of Key Laboratory of Computer System and Architecture (CARCH201906), the Open Fund of State Key Laboratory of IPOC (BUPT) (IPOC2021ZT15), the BUPT Excellent Ph.D. Students Foundation for supporting this work (CX2023232).
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. M. Gerstel, A. Jinno, S. J. B. Lord, and Yoo, “Elastic optical networking: A new dawn for the optical layer?” IEEE Commun. Mag. 50(2), s12–s20 (2012). [CrossRef]
2. K. C. Tsai, L. Fan, L. C. Wang, R. Lent, and Z. Han, “Multi-Commodity Flow Routing for Large-Scale LEO Satellite Networks Using Deep Reinforcement Learning,” in 2022 IEEE Wireless Communications and Networking Conference (WCNC), 626–631 (2022).
3. Q. Huang, M. Lin, W.-P. Zhu, S. Chatzinotas, and M.-S. Alouini, “Performance analysis of integrated satellite-terrestrial multiantenna relay networks with multiuser scheduling,” IEEE Trans. Aerosp. Electron. Syst. 56(4), 2718–2731 (2020). [CrossRef]
4. M. Handley, “Delay is not an option: Low latency routing in space,” in Proceedings of the 17th ACM Workshop on Hot Topics in Networks, 85–91 (2018).
5. N. P. Joseph and J. Bernard, “Distributed internet-optimized services via satellite constellations,” Handbook of Satellite Applications, Springer, 249–269 (2017).
6. J. Radtke, C. Kebschull, and E. Stoll, “Interactions of the space debris environment with mega constellations-using the example of the OneWeb constellation,” Acta Astronaut. 131, 55–68 (2017). [CrossRef]
7. N. Kato, Z. Fadlullah, F. Tang, et al., “Optimizing Space-Air-Ground Integrated Networks by Artificial Intelligence,” IEEE Wirel. Commun. 26(4), 140–147 (2019). [CrossRef]
8. K. Y. Zhong, Y. J. Cheng, H. N. Yang, and B. Zheng, “LEO Satellite Multibeam Coverage Area Division and Beamforming Method,” IEEE Antenn. Wirel. Pr. 20(11), 2115–2119 (2021). [CrossRef]
9. S. Juneja, R. Pratap, and R. Sharma, “Design of a Passive Beamforming Circuit with -37.5 dB Return Loss and -42 dB Isolation Loss for a Multibeam Antenna in 5 G Applications,” in 2022 IEEE International Conference of Electron Devices Society Kolkata Chapter (EDKCON), 553–557 (2022).
10. W. Liu, Y. Tao, and L. Liu, “Load-Balancing Routing Algorithm Based on Segment Routing for Traffic Return in LEO Satellite Networks,” IEEE Access 7, 112044–112053 (2019). [CrossRef]
11. R. M. Calvo, J. Poliak, J. Surof, et al., “Optical technologies for very high throughput satellite communications,” Proc. SPIE 10910, 31 (2019). [CrossRef]
12. E. Zedini, A. Kammoun, and M. S. Alouini, “Performance of Multibeam Very High Throughput Satellite Systems Based on FSO Feeder Links With HPA Nonlinearity,” IEEE Trans. Wirel. Commun. 19(9), 5908–5923 (2020). [CrossRef]
13. F. Moll, “Experimental analysis of channel coherence time and fading behavior in the LEO-ground link,” in Proc. Int. Conf. Space Opt. Syst.Appl., S4-2.1-7 (2014).
14. Z. Ma, Y. Zhao, W. Wang, X. Xin, and J. Zhang, “Adaptive Snapshot Routing Based on Space Debris Risk Perception in Satellite Optical Networks,” in 2021 International Conference on Optical Network Design and Modeling (ONDM), 1–6 (2021).
15. P. D. Arapoglou, G. Colavolpe, T. Foggi, N. Mazzali, and A. Vannucci, “Variable Data Rate Architectures in Optical LEO Direct-to-Earth Links: Design Aspects and System Analysis,” J. Lightwave Technol. 40(16), 5541–5556 (2022). [CrossRef]
16. N. W. Spellmeyer, C. A. Browne, D. O. Caplan, J. J. Carney, M. L. Chavez, A. S. Fletcher, J. J. Fitzgerald, R. D. Kaminsky, G. Lund, S. A. Hamilton, R. J. Magliocco, O. V. Mikulina, R. J. Murphy, H. G. Rao, M. S. Scheinbart, M. M. Seaver, and J. P. Wang, “A multi-rate DPSK modem for free-space laser communications,” Proc. SPIE 8971, 89710J (2014). [CrossRef]
17. D. O. Caplan, J. J. Carney, J. J. Fitzgerald, I. Gaschits, R. Kaminsky, G. Lund, S. A. Hamilton, R. J. Magliocco, R. J. Murphy, H. G. Rao, N. W. Spellmeyer, and J. P. Wang, “Multi-rate DPSK optical transceivers for free-space applications,” Proc. SPIE 8971, 89710K (2014). [CrossRef]
18. J. P. Wang, C. A. Browne, C. D. Burton, D. O. Caplan, J. J. Carney, M. L. Chavez, J. J. Fitzgerald, I. Gaschits, D. J. Geisler, S. A. Hamilton, S. R. Henion, G. Lund, R. J. Magliocco, O. V. Mikulina, R. J. Murphy, H. G. Rao, M. M. Seaver, and N. W. Spellmeyer, “Performance and qualification of a multi-rate DPSK modem,” Proc. SPIE 8971, 89710Z (2014). [CrossRef]
19. D. M. Hall, R. K. DeLong, D. G. Heflflinger, and P. M. Tanner, “Multi-rate variable duty cycle modem for use in an optical communication system,” U.S. Patent 6 661 975B1, Dec. 2003. [Online]. Available: https://patents.google.com/patent/US6661975B1/en.
20. Y. Ning, Y. Zhao, X. Li, S. Rahman, H. Zhang, and J. Zhang, “Load-Balancing Routing Algorithm Against Inter-Satellite Link Congestion in LEO Satellite Optical Networks,” in Optical Fiber Communications Conference and Exhibition (OFC), W2A.22 (2022).
21. G. Rao, C. Browne, D. Caplan, et al., “Electronics design of a multi-rate DPSK modem for free-space optical communications,” Proc. SPIE 8971, 89710Y (2014). [CrossRef]
22. G. Colavolpe, T. Foggi, and A. Vannucci, “Synchronization for Variable Data Rate LEO Direct-to-Earth Optical Links,” in IEEE International Conference on Communications (ICC), 3064–3069 (2022).
23. H. Yao, Q. Jiang, Q. Hao, Z. Liu, P. Zhang, Y. Chang, G. Zhang, and T. Gemg, “Theoretical Calculations and Simulations for Oblique Atmospheric Transmittance,” in 2022 IEEE 14th International Conference on Advanced Infocomm Technology (ICAIT), 138–141 (2022).
24. D. R. Kolev and M. Toyoshima, “Received-Power Fluctuation Analysis for LEO Satellite-to-Ground Laser Links,” J. Lightwave Technol. 35(1), 103–112 (2017). [CrossRef]
25. Giggenbach and Dirk, “Fading-loss assessment in atmospheric free-space optical communication links with on-off keying,” Opt. Eng. 47(4), 046001 (2008). [CrossRef]
26. N. Srinivas and K. Deb, “Multiobjective Optimization Using Non-Dominated Sorting in Genetic Algorithms,” Evol. Comput. 2(3), 221–248 (1994). [CrossRef]
27. L. C. Andrews and R. L. Phillips, “Laser beam propagation through random media,” SPIE Press (2005).
28. L. C. Andrews, R. L. Phillips, and C. Young Hopen, “Scintillation model for a satellite communication link at large zenith angles,” Opt. Eng. 39(12), 3272–3280 (2000). [CrossRef]
