Energy Harvesting Space-Air-Sea Integrated Networks for MEC-Enabled Maritime Internet of Things

Zhijian Lin,Xiaopei Chen,Pingping Chen,*

1 School of Advanced Manufacturing,Fuzhou University,Fujian 362251,China

2 College of Physics and Information Engineering,Fuzhou University,Fujian 350108,China

Abstract: In recent years, various maritime applications such as unmanned surface vehicles, marine environment monitoring,target tracking,and emergency response have developed rapidly in maritime communication networks(MCNs),and these applications are often accompanied by complex computation tasks and low latency requirements.However, due to the limited resources of the vessels,it is critical to design an efficient mobile edge computing (MEC) enabled network for maritime computation.Inspired by this motivation,energy harvesting space-air-sea integrated networks (EH-SASINs) for maritime computation tasks offloading are proposed in this paper.We first make the optimal deployment of tethered aerostats (TAs)with the K-means method.In addition, we study the issue of computation task offloading for vessels, focusing on minimizing the process delay of computation task based on the proposed architecture.Finally,because of the NP-hard properties of the optimization problem,we solve it in two stages and propose an improved water-filling algorithm based on queuing theory.Simulation results show that the proposed EHSASINs and algorithms outperform the existing scenarios and can reduce about 50%of the latency compared with local computation.

Keywords: MCNs; computation offloading; MEC;EH-SASINs

I.INTRODUCTION

Recently, with the rapid development of maritime activities, there has been a growing demand for computation-intensive applications which have various energy consumption, transmission bandwidth,full-coverage and computing latency requirements in marine communication networks (MCNs) [1].Consequently, the MCNs gradually play an increasingly important role in the aspects of data collection and transmission,information storage,and real-time environmental monitoring in a dynamic marine environment.However,the resources of mobile terminals are limited, especially the resources of Internet-of-Thing(IoT)devices on ships, which greatly limit the development of MCNs.At present, the services in MCNs are mainly provided by satellite networks due to the benefit of full coverage, but suffer from many limitations, such as unpredictable latency, user density,complex communication environment and different requirements for maritime services[2].

In the vision of the fifth-generation (5G) and beyond networks, the space-air-sea integrated networks(SASINs)as an integration of satellite systems,aerial networks, and maritime communications, have been becoming an emerging architecture and attracted intensive research interest during the past years [3].It has the characteristics of high throughput and ubiquitous flexible connection and can comprehensively cover space, air, and sea [4].To the best of our knowledge, satellite networks have the disadvantages of long propagation time and cannot meet the rapid growth of maritime communications and computation.Thus, the mobile edge computing (MEC) is taken as a supplement of the SASINs, which is able to meet the requirements of high bandwidth and low delay for the high-speed development of MCNs.Currently,the unmanned aerial vehicle (UAV) is a general solution for MEC deployment in complicated ground environments[5],but this solution will generate high deployment and maintenance costs in the maritime environment.Compared with the terrestrial mobile communication systems,the development of maritime communication systems is facing more challenges due to the harsh environmental conditions, which is not suitable for UAVs.

To the best of our knowledge, a number of related architectures have been studied for MCNs both in academia and industry.In [6], Zenget al.proposed the mobile edge communication,computing and caching technologies to facilitate the MCNs, which have four types of servers, including servers in fixed BSs, in UAVs, in vessels and in cloud.In [7], the marine computing model based on space-air-groundedge integrated network was proposed, in which the UAVs, a communication and computing server at the edge of the core network, and some maritime communication satellites in space are considered as the components of the proposed architecture.In [8], the authors discussed the challenges faced by MCNs and applied three technologies including MEC,space-airground-sea integrated network and blockchain in the proposed scheme.As far as we know, it is obvious that several issues among the existing architectures can be observed.Firstly, placing UAVs as the aerial networks will generate high deployment and maintenance costs in the maritime environment.Secondly,the power supply for BSs and UAVs is difficult and will make a large amount of costs.Thirdly,it is not a cost-effective and energy-efficient way to deploy unmanned surface vehicles(USVs)as the edge computing node.Fortunately, there are many small islands,reefs and drilling platforms in the off-shore area which can be utilized for building the base stations (BSs)and the deployment of tethered aerostats(TAs)[9].In addition,energy harvesting technology has developed rapidly in recent years[10].To tackle the above challenges, in this paper, we propose the deployment of energy-harvesting BS (EH-BS) to facilitate the maritime communication,as well as the TA to serve as the aerial networks.The TAs are tied under the sea and each equips with light energy-harvesting devices.

In addition,this paper studies the issue of computation tasks offloading for vessels,focusing on minimizing the total execution delay of computation tasks.The main contributions of this paper can be summarized as follows:

· We propose an architecture of EH-SASINs to satisfy the edge computing in MCNs.

· We formulate an optimization of computation task offloading as an integer linear programming(ILP)problem which aims to minimize the total execution delay of computation tasks.

· We develop an improved water-filling algorithm based on queuing theory to solve the optimization problem,and demonstrate the effectiveness of the proposed algorithm.

The remainder of this paper is organized as follows.The related work in MCNs is reviewed in Section II.Section III describes the proposed energy-harvesting SASINs(EH-SASINs)architecture.In Section IV,the system models for computation offloading are introduced, followed by the problem formulation in Section V.Section VI analyzes the problem and proposes the algorithm.The performance evaluation results are discussed in Section VII.Finally,we conclude this paper and present the future works in Section VIII.

II.RELATED WORKS

In recent years,several technologies and related solutions in MCNs have been proposed.Cuiet al.[11]proposed a marine fog-cloud computing network architecture which is suitable for USV clusters to reduce processing delay and communication load.Liuet al.[12] proposed the space-air-ground-aqua integrated network(SAGAIN)architecture and introduced the composition of SAGAIN.Literature [13] utilized edge computing to enhance the efficiency in the calculation of data contours.In[14],a computation offloading scheme based on heuristic algorithm was proposed to minimize the energy consumption and delay in maritime MEC network.Satellite-based edge computing is another research direction recently.In[15-17],authors discussed the possibility of satellite-based edge computing to promote the service quality of users.Liet al.presented a software-defined MCNs framework in[18].However,the application scenarios of network architecture proposed by the above references are relatively limited, and the network components are not considered comprehensively.Therefore,from the perspective of solving the challenges of MCNs,we comprehensively consider the deployable components of EH-BS and energy-harvesting TA (EH-TA), and propose an architecture of MEC-enabled EH-SASINs for MCNs.

III.THE ARCHITECTURE OF EH-SASINS

As Figure 1 shows,we propose the architecture of EHSASINs for maritime edge computing, which comprises three network layers including space, air, and sea.These layers are able to work independently or collaboratively to provide a MEC service for vessels.

1)Sea layer: The sea layer is composed of EH-BSs,MEC servers and satellite ground stations(SGSs).The EH-BSs are set up on suitable locations such as islands, reefs and drilling platforms.The SGSs and MEC servers are deployed with EH-BSs.It is difficult to implement power supply to sea through onshore grids.So EH-BSs equipped with energy-harvesting devices can harvest energy from wind, tidal and sunshine.The SGSs are dedicated to communicate with satellites.The sea network is able to provide high data rate and high-speed computation for vessels, but its coverage is limited.As Figure 1 shows, vessel 1 is within the coverage of BS 1.Consequently, the computation tasks can be delivered directly to the MEC server,and then BS 1 returns the results to vessel 1.

2) Aerial layer: The aerial layer mainly consists of TAs such as airships or hot air balloons.Because of the strong wind at sea, the UAVs will lead to a lot of energy consumption to keep stability.To tackle the issue,the TAs can be fixed to the seabed by a wire rope in coastal waters.Small wind turbines can be installed on the TAs to generate energy.Considering the generated energy on TAs is limited,the aerial layer is almost used for relay services.Compared with BSs in terrestrial networks,aerial layer has the features of low cost,easy deployment,and large coverage to offer wireless access services.As Figure 1 shows, the computation tasks of vessel 2 are relayed by TA 1 to an idle and suitable MEC server of BS 2 within its coverage.After that, the results are returned by TA 1.The aerial layer is used as a relay and select the task computing nodes on the sea layer.

3) Space layer: The space layer uses satellites and their corresponding terrestrial infrastructures as carriers for information acquisition,transmission,and processing,which can provide extensive and flexible network connectivity to serve remote vessels.Satellites are in different orbits and with various characteristics.In terms of the altitude,satellites can be classified into three types: geostationary earth orbit(GEO),medium earth orbit(MEO)and low earth orbit(LEO)satellites.Compared with GEO/MEO satellites, LEO satellites have further quantity and lower latency.Note that the speed of LEO satellites is very high, therefore, LEO satellites equipped with MEC servers are only used for collecting and computing tasks.For a single satellite,the coverage of MEO/GEO is more larger than LEO.Besides, the random access memory (RAM), central processing units(CPUs)and graphic processing units(GPUs) on MEO/GEO are vulnerable to solar flares and cosmic radiation[19].Thus,the MEO/GEO satellites only act as relay nodes to return results to vessels.As Figure 1 shows,vessel 3 is outside the coverage of both the BSs and TAs,the computation service is provided by the space layer.There are two cases of the computation service for vessel 3:

Case 1.The task is computed on LEO satellite 1’s MEC server.LEO satellite 1 sends results to MEO satellite 2,and then MEO satellite 2 returns the results to vessel 3.

Case 2.LEO satellite 1 offloads the task to the terrestrial MEC server of BS 3.BS 3 sends the results to GEO satellite 3, and GEO satellite 3 returns the results to vessel 3.

IV.SYSTEM MODEL

4.1 Network Model

The network model of this paper is illustrated in Figure 1.We consider the EH-SASINs which consist of a sea layer,an aerial layer and a space layer.

Figure 1.The energy harvesting space-air-sea integrated networks.

In the space layer, there are LEO, MEO, and GEO satellites.The LEO satellites are equipped with MEC servers whose computing speed arevLEO.Due to the large number and high moving speed of satellites,we assume that all the vessels are within the coverage of satellites.The aerial layer consists ofKTAs with the same altitudehTA.We assume that the coverage radius of TA is denoted byrT A.The location of TAs is denoted by(xT A[k],yT A[k]).The sea layer consists ofNBSs with the same heighthBS.We assume that the coverage radius of a BS is denoted byrBS.The location of BSs is(xBS[n],yBS[n]).So far as we know, the further offshore,the fewer locations are suitable for BSs deployment.Therefore, we suppose thatxBS[n] follows a half-normal distribution andyBS[n] follows random distribution.Each BS is equipped with one MEC server whose computing speed isvBS.Due to the limitation of energy and computing resources, one MEC server only can process Ω tasks simultaneously.

We consider that all theMvessels are equipped with the same single antenna of heighthξand each generates one task at once.Letξm={sizem,cyclem,Lmξ }indicates the taskm,sizemdenotes the size of task measured in bits,cyclemdenotes the number of CPU cycles required for computing one bit of the task andLmξ= (xξ[m],yξ[m]) following random distribution is the location of taskm.In addition,the size of taskmis related to its computation.Hence,we havesizem=g·cyclem.

4.2 Communication Model

In our scenario, we consider three channel cases in the computation offloading process, namely vessel to satellite,satellite to BS and sea channel.Note that different spectrum bands are occupied among them,thus lead to no interference to each other.

4.2.1 Vessel to Satellite

According to[19],the Ku-band is utilized in the transmission link of vessel-to-satellite,since it has no interference to the terrestrial wireless communication systems(e.g.,4G,5G and Wi-Fi).However,the channel condition of Ku-band is easily affected by the rain attenuation [20].Therefore, the data rate of vessel -tosatellite link denoted byRvsis given as follow:

where Λ refers to the rain attenuation ratio,Bvsdenotes the channel bandwidth of the vessel-to-satellite link,Pvis the transmission power of vessels,σ2indicates the power of noise,hvsdenotes the channel fading coefficient between the vessel and the LEO satellite.Note that the estimation of rain attenuation ratio has been studied in[21],which is beyond the scope of this paper.For simplicity,we set Λ to be a fixed value in this paper.

4.2.2 Satellite to BS

In this case,LEO satellites offload tasks or results directly to the BSs or through the relay of MEO/GEO satellites.The satellite-to-BS link occupies Ka-band[22], which can also be used in MEO/GEO satellite systems to provide high data rate.The channel of Kaband is also easily affected by the rain attenuation.The data rate of the satellite-to-BS link denoted byRsBis given by:

where Λ refers to the rain attenuation ratio,BsBdenotes the channel bandwidth of the satellite-to-BS link,Psis the transmission power of vessels,σ2indicates the power of noise,hsBdenotes the channel fading coefficient between the LEO satellite and the BS.

4.2.3 Sea Communication

The sea layer and aerial layer offloading are mainly through the sea communication.Unlike land,there are fewer obstacles on the sea surface,but there are a large amount of scattered waves[23].The path loss can be calculated by the following two-ray reflection model[24]:

whereλis the wavelength,ddenotes the distance between the transmitter and receiver,htdenotes the height of the transmitting antenna andhrdenotes the height of the receiving antenna.

LetPtbe the transmitted power, then the received powerPris given by:

We consider that the sea communication are carried out in the mode of orthogonal frequency division multiplexing (OFDM) with low-density parity-check(LDPC)codes[25,26],thus all the transmission links have no interference to each other[27].LetBscdenote the channel bandwidth of the link andσ2be the power of noise.The data rate of the sea link denoted byRscis given by:

4.3 Offloading Model

In the coverage of the sea and aerial layer, we defineY={0,1}to indicate the choice,Ym= 0 denotes that taskmis directly offloaded to the BSs,Ym= 1 denotes that taskmis offloaded through the relay of TA.If the vessel is within the coverage of the BS,the BS provides services directly,otherwise,the TA act as relay services.χ ∈{1,2,...K}represents the relay selection of tasks.

If the vessel is outside of the coverage of sea layer and aerial layer,the tasks are only able to be offloaded by LEO satellites.In the space, the computing speed of MEC server at LEO satellites is lower than the one at BSs(vLEO ＜vBS)[19].LEO satellites can choose to operate tasks at their own or offload them to the BSs.We denoteS={0,1}as the offloading strategy of the tasks.Sm= 0 represents that taskmis computed at a LEO satellite,Sm= 1 represents that the LEO satellite offload taskmto a BS.

Based on the above analysis, in a time slott, we define aM × Nmatrix Φ(t) as the task offloading matrix in sea layer.It is given as follows:

V.PROBLEM FORMULATION

In this section,we formulate the optimization problem for the EH-SASINs.Due to the limitation of vessel resources,the computing tasks can not be finished locally within the tolerance of time.Consequently, the tasks need to be offloaded to MEC servers.Therefore,in the EH-SASINs, we consider to optimize the total delay of computation tasks.The delay of taskmcan be expressed as follows:

wheredenotes the delivery delay of taskm,is the computation delay andpresents the delay of the results return to the vesselm.Generally, the size of results is small, we can neglect the.Thus, the equation (7) can be simplified as:

The optimization problem for the total delay of computation tasks can be given as follows:

VI.PROBLEM ANALYSIS AND SOLUTION DESIGN

In the EH-SASINs, the BS and LEO satellites equipped with MEC servers can provide computing services.The total execution delay of computation tasks depends on the offloading strategy.Therefore,the key to the problem is to determine the offloading strategy matrix.We can observe that the optimization problem is NP-hard.To tackle this issue, we divide this problem into two offloading stages: (1) The first one is the tasks from the vessels in coastal waters(e.g.near the coastline, covered by BSs or TAs).We denoteas the total delay of these tasks; (2) The second one is the tasks from the vessels in ocean(e.g.far from the coastline,only covered by satellites).We denoteas the total delay of these tasks.

In the first stage, the tasks are directly offloaded to the BSs directly or by the relay TAs.The delayTjcan be expressed as:

wheredenotes the data rate of vesseljto BSn,denotes the data rate of vesseljto relay TA anddenotes the data rate of relay TA to BSn.

However,due to the limitation of the computing resources in BS,equation(10)is only valid for the case ofJ ≤NΩ.IfJ ＞NΩ,all the tasks cannot be allocated simultaneously in one time slot and some tasks have to wait.Thus,we defineTwaitas the waiting time of tasks.The total delay of the tasks can be expressed as:

Then the optimization problem in the first stage is:

In this stage, the delivery delay is related to the date rateof relay link.According to equation(3)(4)(5),we observed that the distance between BS and TA effects.Therefore, in order to obtain the optimal location of TAs, we design Algorithm 1 to make the optimal deployment of TAs.According to the location of BSs,we use the K-means algorithm to cluster the BSs, and find the cluster center as the deployment point of the TAs to maximize the average channel capacity of the relay links.

In stage 2,the tasks outside sea layer and aerial layer are offloaded by LEO satellites.In this model,we can neglect the location of the tasks and consider the distance between tasks and LEO satellites as the altitude of LEO.Due to the fact that LEO satellites have a large number and a high speed, we consider that the LEO satellites can set up the link with the vessels and the BSs fast.Therefore, the delay of taskican be expressed as:

Thus,the optimization problem in the second stage can be given as equation (14).If the LEO satellites offload the tasks to the BSs,there will be more delivery delay compared to the tasks covered by BSs or TAs.Therefore, after the accomplishment of the first stage offloading,the second stage will be continued.Due to the NP-hard property of the problem, we propose an improved water-filling algorithm based on the queuing theory to solve the optimization problem,as shown in Algorithm 2.

Algorithm 1.TA deployment algorithm based on K-means clustering.Require: K (the number of TAs), LBS (the location of BSs),m(the number of iterations)Ensure: LTA (the location of TAs),C (the cluster of BSs),Rkn sc (the data rate of relay links)1: Ck ←∅,k =1,2,...,K 2: Randomly select K LBS as LTA 3: while i ＜m do 4: for k =1,2,...,K do 5: for n=1,2,...,N do 6: dnk =|LnBS -LkTA|7: end for 8: mark LnBS of min(dnk),Ck ←Ck ∪LnBS 9: According to (3)(4)(5),calculate Rkn sc 10: end for 11: for n=1,2,...,K do 12: LnTA = 1 images/BZ_61_1669_1312_1717_1358.png LBS∈Cn LBS 13: end for 14: if LiTA =Li-1TA then 15: break 16: end if 17: i=i+1 18: end while 19: for k =1,2,...,K do 20: Sort BSs in Ck according to Rknsc in descending order 21: end for|Cn|

VII.NUMERICAL RESULTS AND ANALYSIS

In this section,we evaluate the performance of the proposed algorithm.The simulation is completed by Matlab 2019a.We consider a sea region of 100 km×100 km and set the experimental conditions as Table 1.

Table 1.Simulation parameters.

As Figure 2 shows, we study the impact of three methods on TA deployment on the average channel capacity including K-means, random and regional center.We assume the number of TAs is 2.They can be randomly deployed or in the regional center of two equal sub-areas.Compared with two other deployment methods,the K-means clustering algorithm outperforms two others.Consequently,it will increase the transmission rate and reduce delivery delay.

Algorithm 2.Improved water-flling algorithm based on queuing.Require: ξ (the set of tasks), LTA (the location of TAs),C (the cluster of BSs)Ensure: Φ 1: According to Lξ,LBS and LTA,obtain Y 2: for m=1,2,...,M do 3: if Ym =0 then 4: for n=1,2,...,N do 5: According to (3)(4)(5),calculate Rmn sc 6: end for 7: mark the BS α of max(Rmnsc ), Φmα = 1,Ωα =Ω-1 8: else if Ym =1 then 9: for k =1,2,...,K do 10: According to (3)(4)(5),calculate Rmk sc 11: end for 12: mark the TA β of max(Rmksc ),χm =β 13: for k =1,2,...,K do 14: ΦmCk(i) =1,ΩCk(i) =ΩCk(i)-1 15: if ΩCk(i) =0 then 16: i=i+1 17: end if 18: end for 19: else 20: Sm =0 21: end if 22: end for 23: ifimages/BZ_62_318_1986_330_2031.pngimages/BZ_62_330_1951_378_1997.pngN n=1 Ωn /=0 then 24: A=images/BZ_62_427_2010_475_2055.pngN n=1 Ωn 25: for m=1,2,...,M do 26: if Sm =0 then 27: Sm =1,A=A-1 28: if A=0 then 29: break 30: end if 31: end if 32: end for 33: end if

As Figure 3 shows, we explore the relationship between the number of tasks and the total latency.We can observe that, with the increased number of tasks,the total delay is also increasing.If the tasks are computed locally, due to the relatively poor computation ability of the vessels, its total delay is much higher(about 50%)than two other strategies.Compared with the traditional water-filling algorithm, the total delay is close to the proposed algorithm when the number of tasks is small.With the growth of the number of tasks, the MEC server reaches its limits and will be overloaded.Therefore, some tasks need to wait.The proposed algorithm can save the waiting time and reduce the total delay of the system.

Figure 2.The average channel capacity of different deployments.

Figure 4 shows how the total delay varies with the size of tasks.In order to better match the actual situation, we model the size of the task as a Poisson distribution.We can observe that as the size of tasks increases, the total delay also increases.Compared with the case of local computation,as the size of tasks increases, the time-saving effect becomes better.Besides, the performance of the proposed algorithm is also better than the traditional water-filling algorithm.In general,applying the proposed network architecture and algorithm for computation offloading can save about 50% (when the size is 25 Mb) of the total execution delay than the method of local computation.

Figure 3.Total delay versus task number.

Figure 4.Total delay versus task size.

Figure 5 shows the changes of total delay under the different capacity of a BS (the number of tasks that the BS can compute simultaneously).We observe that with the increase of capacity of BSs, the total delay can be greatly improved.However, the improvement becomes not obvious if the capacity of BS is larger than 11.The reason is as follows:when the total number of tasks is greater than the total capacity of BSs,all the tasks cannot be allocated simultaneously,which results in a waiting delay for some tasks.In addition,the waiting time of the tasks can be reduced as the capacity of the BS increases.We utilize queuing theory and adopt the principle of first-in and first-out,and prioritize larger tasks.Compared with the traditional waterfilling algorithm,the proposed algorithm can improve the waiting delay and achieves better performance if BS capacity is small.

Figure 5.Total delay under different capacity of a BS.

VIII.CONCLUSION

In this article, a novel architecture is designed to facilitate the maritime activities in MCNs, which includes space layer,aerial layer and sea layer with energy harvesting BSs and TAs.In order to meet the requirements of low latency for the rapid development of MCNs, MEC is taken as a promising paradigm to help the vessels via computation offloading.To investigate the offloading performance in term of latency in the proposed network architecture,we first deploy TAs with the optimal locations via K-means method.Secondly,we formulate the offloading optimization problem as two stage sub-problems and propose an improved water-filling algorithm based on queuing theory.Simulations show that the proposed architecture and algorithms can achieve better results than other existing schemes.Movement of ships and satellites can affect their communication links.Thus in the future work,mobility will be considered in our scenarios.

ACKNOWLEDGEMENT

This work was supported in part by 2020 Science and Technology Innovation Team from Universities of Fujian Province, the NSF of China (Nos.61871132,62171135)and the Project of Science and Technology of Quanzhou City 2021N050.