Wenfeng Shi*, Deyun Gao, Huachun Zhou, Bohao Feng, Haifeng Li, Guanwen Li, Wei Quan

School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing 100044, China

I. INTRODUCTION

Due to the global area coverage, satellite network plays a significant role in many fields such as navigation, broadcast communication and so on. The satellite network can be used as a supplement to terrestrial network and their combination can significantly improve the communication services [1]. The integration of satellite and terrestrial networks has gained a great interest including the combination of satellite network with software dened naval network [2], the combination of high-throughput satellite network with SDN based terrestrial network [3], the combination of satellite network with 5G networks [4], etc. However,these works mainly focus on the integration or the implementations of the terrestrial networks, the design of satellite network especially the multi-layer satellite-terrestrial networks receives little attention.

Different from the single layer satellite networks, the Multi-Layer Satellite Networks(MLSN) which consists of GEO, MEO and LEO layer nodes can make better use of the advantages and complement the defects of the satellites in each layer. MLSN can achieve various advantages which are made up of providing adequate network capacity, increasing the available transmission [5-7], etc. In [5],authors propose a routing algorithm MLSR for multi-layer satellites combined with GEO layer, MEO layer and LEO layer nodes. In MLSN, the MEO and LEO are divided into groups according to the coverage of higher layer. In [6], authors propose a QoS routing for MLSN and suggest that data which are sent to long distance destinations should be forwarded through MEO layer. In [7], authors suggest that data originated from different LEO groups will be delivered through MEO layer. These works mainly focus on the routing schemes,considering less of the long propagation delay and the high bit error in space networks. However, these features prevent TCP/IP protocols from being adopted in satellites networks.

Delay and Disruption Tolerant Networks(DTN) [8] can provide an alternative solution to deal with the space network features and its performance is proved better than some modified TCP protocols [9]. By inserting a layer called “Bundle layer” between application and transport layer，DTN could handle with such challenging environment using a stay-andforward mechanism. Contact Graph Routing(CGR) [10] is a dynamic routing algorithm developed for space DTN. It uses a series of scheduled time-varying contacts to calculate routes. Different from terrestrial DTNs, we can dynamically schedule the moving trajectories of satellites. Thus, the communication opportunity information such as communication duration and rate can be acquired previously. By conguring the previous obtained information into contact plan, CGR could compute routes for messages. In this paper, we use DTN and CGR to handle the space features and focus on the contact plan design problem of multi-layer satellite network.

In multi-layer satellite networks, satellites in lower layer could be covered by multi high layer satellites. Meanwhile, due to the rapid motion among satellites, the network could suffer frequent disruption. It can cause the increase of potential links. If we congure all the contacts into contact plan, it will lead to a very complex contact plan which can incur a signicant increase in route computing delay.Thus, only designing the contact plan in aexible way can DTN and CGR be used ef-ciently in multi-layer satellite network.

Since routing calculation depends on the contact plan, the contact plan design problem have received much attentions [11-13]. In [11]authors proposed a trafc-aware contact plan(TACP) design procedure based on a Mixed Integer Linear Programming (MILP) formulation to improve the delivery efciency. In [12]authors provided the detailed description of the contact plan design (CPD) problem and compared the performance of existed solutions. In[13], authors considered the time-varying of contact capacity when they design the contact plans. However, most of these works are implemented in a small-scale topology. The contact plan design for large-scale networks like multi-layer satellite network received little attention.

Considering above, in this paper we investigate the suitability of DTN and CGR in largescale multi-layer satellite terrestrial network.We propose a distributed contact plan design scheme to reduce the computing delay by dividing the contacts information of multi-layer satellite network into multi small parts. Meanwhile, we propose a duration based inter-layer contact selection algorithm to handle the disruption problem among satellites.

The main contributions of this paper are summarized as follows:

1) A space time graph is used to model the multi-layer satellite network.

2) A distributed contact plan design scheme is proposed to enable CGR to be used efficiently in multi-layer satellite network. Meanwhile, a duration based inter-layer contact selection algorithm is proposed to handle the disruption problem.

3) The distributed contact plan design scheme and the duration based inter-layer contact selection algorithm were evaluated on a testbed and experiments proved that our methods can reduce data delivery delay.

The remainder of the paper is organized as follows: In part II, we model the satellite network using space time graph. In part III,we introduce the details of the contact design scheme and the duration based inter-layer contact selection algorithm. In part IV, we present the experiments on our satellite-terrestrial network testbed. In part V, we summarize the works in this paper.

II. SYSTEM MODEL

In satellite networks, the topology can be changed all the time. However, as soon as the system period is divided into partial time slots during which the links among nodes aren’t disrupted, the satellite topology could be seen as unchangeable in each time slot. The whole system could be modeled as several static graphs with time sequence using space time graph [14].

Supposing that the whole topology includegGEO satellites,mMEO satellites andlLEO satellites. Letvgi,vmj,vlkdenote theith,jth andkth satellite in the GEO, MEO and LEO layer respectively. Then all the nodes in multi-layer satellite network can be denoted asSupposing that the system period can be divided intoNtime slots according to the topology change andtidenotes theith time slot. We assume thatdenotes the graph of the satellite topology in time slottiand linkdenotes that a pair of nodes such as a GEO satellitegiand a MEO satellitemkcan communicate with each other in the time slotti. Then, the multi-layer satellite network can be modeled as space time graphG=(V,E) withNstatic graph

III. DISTRIBUTED CONTACT PLAN DESIGN FOR MULTI-LAYER SATELLITE NETWORK

For nodes in different layers, we introduce different contact plan design schemes. Meanwhile, we propose a duration based inter-layer contact selection algorithm to keep the integration of the network. By adopting the distributed contact plan design scheme, the complexity of contact plan can be reduced significantly which in turn can reduce the route computing delay.

3.1 Distributed contact plan design for multi-layer satellite network

There exists two kinds of contacts in the satellite network including intra-layer contacts connecting satellites within the same layer and inter-layer contacts connecting satellites within different layers. The intra-layer contacts are stable while the inter-layer contacts can suffer frequent disruption due to fast relative motion among satellites which belong to different layers.

The main principle of the distributed contact plan design scheme is to divide the complete contact information according to which layers the satellite belongs to. By cooperating with each other, satellites in the network can keep the route reachable using their partial contact information.

3.1.1 Contact plan design for LEO layer nodes

Compared to GEO, LEO and MEO layer have shorter round trip delays. Nodes in these two layers are used to take charge of delivering data for ground stations. For inter-layer contacts, the LEO node maintains one contact between it and MEO layer node as well as one contact between it and the GEO layer node in current time slot, ignoring the contacts which are between other LEO nodes and high layer nodes. When the LEO comes out from the communication range of the accessed high layer node, a new time slot will be divided and the GEO will select a new inter-layer contact for this LEO. The selection algorithm adopted by GEO nodes will be discussed in section B.

For intra-layer contacts, a LEO node only maintains the contact information among nodes whose distance is no more than n hop from it in LEO layer. It means that a LEO node can only communicate with LEO nodes withinnhops through LEO layer. Messages transmitted to LEO nodes with a hop count which exceedsnwill be delivered to MEO layer. The suitable value of n will be given according to experiments.

In satellite networks, the total delay is composed of propagation delay, routing calculation delay and transmission delay. Messages transferred through LEO layer can reduce propagation delay in every hop, but it could add transmission hops for long distance bundles which will increase the calculation and transmission delay. Although delivering a message through MEO layer could increase the propagation delay, this can reduce the transmission hop.With the reduction of hop count, the calculation delay and transmission delay can also be reduced.

Figure 1 is a sample example to explain the contact plan design scheme for LEO nodes. Assuming L4 is a LEO layer node and“n hops” is set to 2. The yellow links is the inter-layer contacts selected for current time slot. Then the contact plan of L4 is set to contain all the intra-layer contacts among blue nodes, the yellow inter-layer contacts between L4 and M2 as well as between L4 and G1.If L4 wants to communicate with L2 whose distance is no more than 2 hop from L4, it can compute a reachable route using its contact plan. If L4 wants to communicate with L10 whose distance is more than 2 hops from L4,the message will be sent to MEO node M2 and be delivered through MEO layer.

3.1.2 Contact plan design for MEO layer nodes

Fig. 1. Example of contact plan design for LEO.

For inter-layer contacts between MEO and GEO layer, a MEO node needs to maintain the inter-layer contact between it and a GEO layer node in current time slot. This contact is used as a management contact for contact updating.For inter-layer contacts between MEO and LEO layer, the MEO should maintain all the inter-layer contacts between LEO layer nodes and MEO layer nodes in current time slot.Note that only the inter-layer contacts in one time slot are congured into the contact plan,ignoring inter-layer links in future time slots.When any of the inter-layer contact disrupted,the GEO nodes will select a new contact for the disrupted nodes.

For intra-layer contacts, since the number of nodes is small in MEO layer and the topology is stable compared to inter-layer topology,a MEO node should maintain all the contacts among MEO layer nodes in the system period.

3.1.3 Contact plan design for GEO layer nodes

The GEO layer nodes are used to select inter-layer contacts for low layer satellites due to their ability of providing extensive coverage to low layer satellites. For GEO nodes, they should maintain two kinds of contact plans named Data Contact Plan (DCP) and Handover Contact Plan (HCP). The data contact plan includes all the selected inter-layer contacts of the multi-layer satellite network in the current time slot as well as all the inter-layer contacts in GEO layer. The data contact plan is used to calculate routing and deliver contact updating message to LEO and MEO nodes.

The HCP contains all the potential contacts in the future and it is used for inter-layer contacts selection. By traversing DCP, GEO can find the inter-layer contact which is going to be disrupted and then it could select a new inter-layer contact for the disrupted lower layer node from the potential inter-layer contacts which are recorded in HCP. As soon as the candidate inter-layer contact is selected, the GEO will add the new contact into DCP and send contact updating messages to the associated satellite nodes. The DCP and HCP are formulated by space time graphG=(V,E)discussed in part II.

In this way, nodes in multi-layer satellite network can maintain a partial contact plan of the complete topology and update the contact dynamically to handle the disruption problem.The selection algorithm is described in sections B.

The “distributed” scheme in this paper mainly means that each node maintains a partial part of the whole contact information of the network. We design the contact plan for each node individually. Each node computes paths depending on the local contacts information congured in its own contact plan.

3.2 Duration based inter-layer contact selection algorithm

Since the rapid relative motion among satellites, the inter-layer contacts can suffer frequent disruption. When a low layer node moves out from the communication range of a high layer node, a new inter-layer contact between the disrupted low layer node and the high layer nodes should be selected to keep the integrity of the multi-layer satellite network.Meanwhile, a lower layer node can be covered by multi high layer nodes at the same time. If an unsuitable inter-layer contact such as a contact which is gonging to disrupted is selected,it may incur unnecessary disruptions. Thus, in order to keep the integrity of the network and reduce the disruption times, we design a duration based selection algorithm to handle the disruption problem.

We use the residual communication time to identify the contact that will be disrupted.When a contact between the low layer and the high layer is detected as to be disrupted, the GEO nodes will go through the HCP to find candidate high layer nodes for the disrupted low layer node to connect with.

The meaning of variables in the algorithm is shown in table1. The selection processes are shown in algorithm 1 which can be described from step 1 to step 7:

Step 1: GEO node will open and go through DCP every 5 seconds to calculate the residual time of the inter-layer contact and check whether the start node of the contact is connected with it. If the start node is connected with the GEO, this means that the GEO node can take charge of inter-layer contact selection for this node. When the residual time of an inter-layer contact is less than 5 seconds, the contact will be recognized as to be disrupted.The GEO node will go to step 2 and select a new inter-layer contact for this lower layer node. Or else, GEO node will check next contact in contact plan. When all the contacts have been detected, GEO will sleep 5 seconds for next detection round.

Table I. The meaning of variables in duration based inter-layer contact selection algorithm.

Considering that the inter-layer contacts can suffer frequent disruption, the contact plan should be updated dynamically. The propagation delay among GEO and LEO nodes are about 120ms, the delivery delay of the contact updating message should be considered.When the updating messages are influenced by the bit error, the retransmission can cost more time. Meanwhile, the computation of CGR depends on the communication starting time and ending time which requires the time synchronization of nodes. However, in the actual deployment of satellites, the time synchronization may be inuenced. Since the new contact selected for next time slot has already been used at the selecting time, the inuence of time synchronization mainly occurs at the ending time of contact. Thus, the detecting duration is set to “5 seconds” to consider the propagation delay and avoid the synchronization problem.

Step 2: GEO node will open and go through HCP tond the candidate inter-layer contacts for the disrupted low layer node. Firstly the contact must start with the disrupted low lay-er node and end with a node which belongs to the high layer that the disrupted low layer node connected. Then the GEO node will check whether the contact is subject to the time constraints: the start time of the candidate inter-layer contact is before the disrupted contact’s end time, the end time of the candidate inter-layer contact is after the end time of the disrupted contact. This could promise that the contact can be used in next time slot. If the contact is subject to the time constraints, then GEO node will go to step 3, or else it will check next contact in HCP.

Step 3: GEO node will calculate the contact duration as the end time of the contact subtract current time and add the contact as well as its duration into Alternative Contact Set (ACS).Then GEO nodes will return to step 2 to check next potential contact in HCP. When all the contacts in HCP have been checked, it will go to step 4.

Step 4: The GEO node should check which layer the disrupted contact belongs to. If the start nodethis means that the contact is an inter-layer contact between LEO and MEO layer and then GEO node will go to step 5, Or else go to step 6.

Step 5: GEO node will select the contact with the longest duration in ACS as the inter-layer contact in next time slot. They will add the new inter-layer contact into the data contact plan. Meanwhile, GEO node will create the contact updating message and send them to the disrupted LEO node, the other GEO nodes as well as all MEO layer nodes. In this way, the integrity of the multi-layer satellite network can be guaranteed. Then GEO node will clean ACS and return to step1 to check next inter-layer contact record in DCP.

Step 7: GEO node will select the inter-layer contact with the longest duration in ACS.Once the contact is selected, the GEO node will create a contact updating message and send it to the disrupted low layer node and all the GEO nodes. Then the ACS will be cleaned and return to step1 to check next inter-layer contact record in DCP. The time complexity of the scheme is O(n2).

Algorithm 1:Duration based inter-layer contact selection

Input: data contact plan; handover contact planOutput: contact update message

IV. EXPERIMENTS AND PERFORMANCE

The performance of the distributed contact plan design scheme and the duration based inter-layer contact selection algorithm were evaluated on our testbed which is implemented using OpenStack [15]. The constellation of multi-layer satellite network adopts the design proposed in [16]. The LEO layer uses an Iridium constellation with 66 nodes distributed in six orbits. The MEO layer uses odyssey system with 10 nodes distributed in two or-bits. The GEO layer uses a constellation with 3 nodes distributed in one orbit. We use STK[17] to achieve the contact information among satellites in 12 hours. The orbit parameters and STK model are presented in table 2 andg 2.

We use 79 virtual nodes to simulate the satellites and 16 nodes to simulate the terrestrial network which adopts Identier/Locator split scheme. We use ION3.3.1 [18] to realize the proposed distributed contact plan design and the duration based inter-layer contact selection scheme. For satellite nodes, we use DTN (BP/LTP) protocol to handle the long propagation delay and high bit error. For terrestrial nodes,we adopt the TCP/IP protocols. In this paper,we focus on the contact plan design problem of multi-layer satellite network. We assume axed satellite terminal which works as a gateway. The user communicate with the server in data center through the gateway and the satellite network. The communication details are described in our previous work in [19].The experiment topology is shown ingure 3.In this paper, we mainly focus on the works of network layer. The inuences of physical layer characteristics such as the frequency, Doppler shifts are represented by the network layer features including the transfer rate, bit error and the propagation delay. The satellite parameter is referenced to an experimental project developed by NASA [20]. The satellite adopts the symmetric channel whose frequency is Ku-band. The data rate is set to 250kB/s. The bit error is set to 10-6which is a typical value in satellite networks [21]. The delays among nodes are acquired from the STK simulation and shown in table 3. The rate, propagation delay and bit error are simulated using the Linux tool “TC”.

Table II. The parameters of the experimental constellation.

Table III. Delays of the links in multi-layer satellite network.

Fig. 2. Multi-layer satellite network.

Fig. 3. Experiment topology.

The experiments in this paper include four parts. Firstly, we evaluate the influence of different values of hop count n and determine the suitable value according to the minimum data delivery delay. Secondly, we compare the distributed contact plan design scheme with the method that congure all the contact information into contact plan. Thirdly, we compare the disruption times of inter-layer contacts when GEO nodes adopt duration based selection algorithm and that of a random selection scheme. Fourthly, we analyze the throughput of the satellite-terrestrial network.

4.1 Influence of different value of n hop

In distributed contact plan design scheme,LEO nodes only maintain the contact information among LEO nodes whose distance is withinnhops from the local node. Data whose destination nodes exceednhop from the route computing node in LEO layer will be sent through MEO layer. If the hop limitationnis set to an improper value, the delivery delay may increase. Thus, we should select the most suitable value ofnto minimize the delivery delay.

In DTN, application layer data are divided into bundles as the transport units. In this part, we use one LEO node to send bundles to all the other 65 LEO satellites and acquire the average delivery delay of these bundles.The LEO node sends one bundle to one other LEO every second and the sending duration is 3600s. The valuenis set from 1 to 8. We use three bundle sizes including 100k, 150k and 200k to evaluate the inuence of the value ofnhop on bundle delivery delay. The integrity of the network can be guaranteed by the duration based inter-layer contact selection algorithm.

Fig. 4. Delivery delay with different value of n.

Figure 4 is the experimental results of delivery delay with different value of hop count limitation n. We can see that the delay is decreased with the increase of the value ofnrstly. When the value exceeds 4, the delay is increased with the increase ofn. It is because the LEO layer has lower propagation delay while the MEO layer exists longer propagation delay. If the number ofnis set to a small value, many bundles are sent to MEO layer and this can increase the total propagation delay. With the increase of the value ofn, the number of bundles delivered through MEO layer will decrease. Thus, the propagation delay will decrease. When the value ofnis set to 4, the average delay is the smallest among the three bundle sizes. Although MEO layer has longer propagation delay, data delivered through MEO layer can reduce the hop counts.This could decrease the total computing delay and transmission delay for long distance data.When the value ofnexceeds 4 hops, with the increase ofn, bundles delivered through LEO layer will increase. The computing delay,transmission delay and total propagation delay can also increase caused by the increase of delivery hops. Experimental results showed that the hop limitation value ofnwhich is set to 4 is most suitable. We use the value 4 as the hop count limitation of our distributed contact plan design scheme.

4.2 The delivery delay of different contact plan design scheme

In figure 5, we compare the delivery delay of distributed contact plan design scheme(DCPS) with the contact plan design scheme that congures all the LEO layer contacts into contact plan (LEO-CP). We use one LEO node to send 100 bundles to the other LEO nodes whose distance is from 1 hop to 8 hops away from the sending node. In each destination node, we compute the average delivery delay of the 100 bundles. The bundle size is set to 100k. We can see that when the distance is less than 4 hops, the difference of average delay is rather small when we use the two contact plan design scheme. When the distance is more than 4 hops, with the increase of hop counts, the delivery delay of LEO-CP can increase constantly. Oppositely, the delay is stable relatively when we use distributed contact plan design scheme. It is because when we congure all the LEO contacts information into contact plan, all data will be delivered through LEO layer. When the destination is far away from the source node, data will be delivered through a long path. With the increase of hop count, the total transmission delay can increase. As we use distributed contact plan design scheme, when hop exceeds 4, data will be sent through MEO layer and the hop count can be decreased. The total delay will decrease as well. From figure 6 we can conclude that the distributed contact plan design scheme can reduce the delivery delay of bundles which are sent to remote destinations.

Fig. 5. Delivery delay of different distance.

Fig. 6. Delivery delay of different distance.

Fig. 7. Disruption times of different selection scheme.

4.3 Disruption times of inter-layer contact

In this part, we compare the disruption times of inter-layer contact among LEO nodes and MEO nodes when GEO adopts duration based inter-layer contact selection algorithm and a random inter-layer contact selection scheme.The random selection scheme is designed to select an inter-layer contact which is therst contact that acquired by GEO nodes. The experimental duration is set from 3 hours to 10 hours. Figure 7 shows the total disruption times between the 66 LEO nodes and the 10 MEO nodes in the experimental duration.Since every inter-layer contact selected by duration based selection algorithm has the longest communication duration, the total inter-layer contact disruption time could be decreased. From theg we can see that when GEO nodes adopt duration based inter-layer contact selection algorithm, the disruption times could be reduced by half than that of the random selection scheme.

4.4 The throughput of the satelliteterrestrial network

In this part, we evaluate the throughput of the satellite-terrestrial network. We use GW1 to simulate the sender in terrestrial network and GW2 to simulate the receiver in terrestrial network. We use iperf at GW1 to send data to GW2. All the data will be sent through the satellite network.

In figure 8 we compare the throughput of the duration based inter-layer contact selection algorithm and the random inter-layer contact selection scheme. Since most of the inter-layer contacts will experience disruption after 4000s when the ION started, the sending time is set to that time. The duration of the experiment is set to 1200s and bit error is set to 10-6. We compare the two selecting schemes at the rate of 50 kbyte/s. From figure 8 we can see that the throughputs of the two selection schemes are similar. This is because that the duration based inter-layer selection scheme can mainly reduce the updating messages compared with the random selection scheme. While the updating message has less impact on the throughput. Therefore, the differences of the throughput between the two selection schemes are little.

Figure 9 shows the throughput of the duration based inter-layer contact selection scheme when faced with different bit error. The bit error is set to 10-6and 10-5respectively. When the bit error is 10-5, the data can suffer more retransmission and the throughput is more dispersive than that of 10-6. Thanks to the ability of dealing with bit error, DTN and LTP are enable to transfer the data integrally.

Fig. 8. Throughput of different inter-layer contact selection algorithms.

Fig. 9. Throughput of different bit errors.

V. CONCLUSION

In this paper, we have studied the contact plan design problem for multi-layer satellite network. We propose a distributed contact plan design scheme to divide the complete contact information into several partial parts to reduce its complexity. Meanwhile, a duration based inter-layer selection algorithm is proposed to deal with the disruption of inter-layer contacts.We evaluate the performance of the distributed contact plan design in a satellite-terrestrial network testbed. Extensive experiments demonstrate that the distributed contact plan design can signicantly reduce the data delivery delay.

ACKNOWLEDGEMENT

This paper is supported by National High Technology of China (“863 program”) under Grant No. 2015AA015702, NSAF under Grant No. U1530118, NSFC under Grant No.61602030 and National Basic Research Program of China (“973 program”) under Grant No. 2013CB329101.

[1] R. Ferrs, H. Koumaras, O. Sallent, G. Agapiou, T.Rasheed, M.-A. Kourtis, C. Boustie, P. Glard, and T. Ahmed, “Sdn/nfv-enabled satellite communications networks: Opportunities, scenarios and challenges,”Physical Communication, vol. 18,Part 2, 2016, pp. 95 – 112.

[2] Nazari S, Du P, Gerla M, et al. “Software Dened naval network for satellite communications(SDN-SAT),”Proc. Military Communications Conference, MILCOM 2016, 2016, pp. 360-366.

[3] Mongelli M, De Cola T, Cello M, et al. “Feeder-link outage prediction algorithms for sdnbased high-throughput satellite systems,”Proc.Communications (ICC), 2016 IEEE International Conference on communications, 2016, pp. 1-6.

[4] Jia M, Gu X, Guo Q, et al. “Broadband Hybrid Satellite-Terrestrial Communication Systems Based on Cognitive Radio toward 5G,”IEEE Wireless Communications, 23(6), 2016, pp. 96-106.

[5] Akyildiz I F, Ekici E, Bender M D. “MLSR: a novel routing algorithm for multilayered satellite IP networks,”IEEE/ACM Transactions on networking,10(3), 2016, pp. 411-424.

[6] Lee J, Kang S. “Satellite over satellite (SOS) network: A novel architecture for satellite network,”Proc. INFOCOM 2000. Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies, 2000, pp. 315-321.

[7] Zhou Y, Sun F, Zhang B. “A novel QoS routing protocol for LEO and MEO satellite networks.”International Journal of Satellite Communications and Networking, 25(6), 2007, pp. 603-617.

[8] S. Burleigh, A. Hooke, L. Torgerson, K. Fall, V.Cerf, B. Durst, K. Scott, and H. Weiss, “Delay-tolerant networking: an approach to interplanetary internet,”IEEE Communications Magazine, vol.41, no. 6, 2003, pp. 128–136.

[9] C. Caini, H. Cruickshank, S. Farrell, and M.Marchese, “Delay- and disruption-tolerant networking (dtn): An alternative solution for future satellite networking applications,”Proceedings of the IEEE, vol. 99, no. 11, 2011, pp. 1980–1997.

[10] G. Araniti, N. Bezirgiannidis, E. Birrane, I. Bisio,S. Burleigh, C. Caini, M. Feldmann, M. Marchese,J. Segui, and K. Suzuki, “Contact graph routing in dtn space networks: overview, enhancements and perfor- mance,”IEEE Communications Magazine, vol. 53, no. 3, 2015, pp. 38–46.

[11] J. A. Fraire, P. G. Madoery, and J. M. Finochietto,“Trafc-aware contact plan design for disruption-tolerant space sensor networks,”Ad Hoc Networks,vol. 47, 2016, pp. 41 – 52.

[12] J. A. Fraire and J. M. Finochietto, “Design challenges in contact plans for disruption-tolerant satellite networks,”IEEE Communications Magazine, vol. 53, no. 5, 2015, pp. 163–169.

[13] Y. Wang, M. Sheng, J. Li, X. Wang, R. Liu, and D.Zhou, “Dynamic contact plan design in broadband satellite networks with varying contact capacity,”IEEE Communications Letters, vol. 20,no. 12, 2016, pp. 2410–2413.

[14] M. Huang, S. Chen, Y. Zhu, and Y. Wang, “Topology control for time-evolving and predictable delay-tolerant networks,”IEEE Transactions on Computers, vol. 62, no. 11, 2013, pp. 2308–2321.

[15] H. Li, H. Zhou, H. Zhang, B. Feng, and W. Shi,“Emustack: An openstack-based dtn network emulation platform (extended version),”Mobile Information Systems, vol. 2016, 2016.

[16] F. Long, “Satellite Network Robust QoS-aware Routing,” Springer Publishing Company, Incorporated, 2014.

[17] Satellite tool kit (stk). [Online]. Available: http://www.agi.com/products/ stk/,2016

[18] Interplanetary overlay network (ion). [Online].Available: https: //sourceforge.net/projects/iondtn/,2015

[19] B. Feng, H. Zhou, G. Li, H. Li, and S. Yu, “Sat-grd:An id/loc split network architecture interconnecting satellite and ground networks,”Proc.2016 IEEE International Conference on Communications (ICC), 2016, pp. 1–6.

[20] IVANCIC W D, SULLIVAN D V. “Delivery of Unmanned Aerial Vehicle DATA,” Sandusky, Ohio,USA: National Aeronautics and Space Administration, Glenn Research Center, 2011.

[21] Wang R, Burleigh S C, Parikh P, et al. “Licklider transmission protocol (LTP)-based DTN for cislunar communications,”IEEE/ACM Transactions on Networking (TON), 2011, 19(2), pp. 359-368.