northumbria research linknrl.northumbria.ac.uk/34585/1/reward-aided sensing.pdf · care, virtual...

Northumbria Research Link

Citation: Hu, Jiejun, Yang, Kun, Hu, Liang and Wang, Kezhi (2018) Reward-Aided Sensing Task Execution in Mobile Crowdsensing Enabled by Energy Harvesting. IEEE Access, 6. pp. 37604-37614. ISSN 2169-3536

Published by: IEEE

URL: http://dx.doi.org/10.1109/ACCESS.2018.2839582 <http://dx.doi.org/10.1109/ACCESS.2018.2839582>

This version was downloaded from Northumbria Research Link: http://nrl.northumbria.ac.uk/34585/

Northumbria University has developed Northumbria Research Link (NRL) to enable users to access the University’s research output. Copyright © and moral rights for items on NRL are retained by the individual author(s) and/or other copyright owners. Single copies of full items can be reproduced, displayed or performed, and given to third parties in any format or medium for personal research or study, educational, or not-for-profit purposes without prior permission or charge, provided the authors, title and full bibliographic details are given, as well as a hyperlink and/or URL to the original metadata page. The content must not be changed in any way. Full items must not be sold commercially in any format or medium without formal permission of the copyright holder. The full policy is available online: http://nrl.northumbria.ac.uk/pol i cies.html

This document may differ from the final, published version of the research and has been made available online in accordance with publisher policies. To read and/or cite from the published version of the research, please visit the publisher’s website (a subscription may be required.)

http://nrl.northumbria.ac.uk/policies.html

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. Seehttp://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2839582, IEEE Access

1

Reward-Aided Sensing Task Execution in MobileCrowdsensing Enabled by Energy Harvesting

Jiejun Hu, Kun Yang*, Senior Member Liang Hu and Kezhi Wang, Member IEEE

Abstract—Mobile crowdsensing (MCS) is a new sensing frame-work that empowers normal mobile devices to participate insensing tasks. The key challenge that degrades the performanceof MCS is selfish mobile users who conserve the resources (e.g.,CPU, battery and bandwidth) of their devices. Thus, we introduceenergy harvesting (EH) as rewards into MCS and thus providemore possibilities to improve the quality of service (QoS) of thesystem. In this paper, we propose a game theoretic approach forachieving sustainable and higher-quality sensing task executionin MCS. The proposed solution is implemented as a two-stagegame. The first stage of the game is the system reward game,in which the system is the leader, who allocates the task andreward, and the mobile devices are the followers who executethe tasks. The second stage of the game is called the participantdecision-making game, in which we consider both the networkchannel condition and participant’s abilities. We analyse thefeatures of the second stage of the game and show that thegame admits a Nash Equilibrium (NE). Based on the NE of thesecond stage of the game, the system can admit a StackelbergEquilibrium, at which the utility is maximized. Simulation resultsdemonstrate that the proposed mechanism can achieve betterQoS and prolong the system lifetime while also providing aproper incentive mechanism for MCS.

Index Terms—Task execution, energy harvest, game theory,mobile crowdsensing.

I. INTRODUCTION

NOWADAYS smartphones have changed people’s dailylives. The functions of smartphones become increasingly

powerful every day, and they have rich sensory capabilities.Smartphones not only allow us to communicate with eachother but also offer the possibilities of sensing the environmentand collecting, processing and sharing information. Thesetechnologies empower the development of mobile crowdsens-ing (MCS). MCS is widely applied in our daily lives andin areas such as environment monitoring, personal health-care, virtual reality entertainment, transportation monitoringand smart city applications [1]–[4]. It has the advantages ofmobility, scalability and cost effectiveness, comparing to thetraditional Internet of things. Moreover, the integration ofhuman intelligence into the mobile sensing and computingprocess is also a special aspect of MCS. This means thatpeople can decide how to and when to be a member of MCS.

Jiejun Hu and Liang Hu are with the Department of Computer Scienceand Technology, Jilin University, Changchun, Jilin, 130012 China e-mail:[email protected], [email protected]

Kun Yang School of Computer Sciences and Electrical Engineering, Univer-sity of Essex, CO4 3SQ, Colchester, U.K. and School of Communication andInformation Engineering, University of Electronic Science and Technology ofChina.

Kezhi Wang is with Department of Computer and Information Sciences,Northumbria University, Newcastle upon Tyne, NE2, 1XE, U.K.

For a typical MCS procedure, first, the MCS server willpublish the sensing task to a special area. Then, the participantrecruiting procedure based on performance, reputation etc.,will be triggered. All participants will begin to execute thesensing task: sensing, executing sensing data and returning theresults to the server. When the sensing task is relatively easy,sensing and executing will not cause the heavy consumptionof energy or other resources by the mobile devices. However,when a tough sensing task is allocated, such as multimediadata sensing and mining, participants know they have to spendmore resources to accomplish the task, so some of them maychoose to leave without a proper task execution plan andincentive mechanism.

Incentive mechanism is the most important issue whichneeds to be considered in MCS. A proper incentive mechanismcan make sure the participants in MCS to donate their resourceto achieve a common interest. To simulate mobile user tobecome a participant and remain the number of the partici-pants, researchers have designed extensive method to providethe incentive mechanism. They adopt money, reputation andcredit as reward to participants to guarantee the sensing quality[5], [6].

In our work, energy harvesting (EH) is envisioned as apromising way to address the challenge of incentive mech-anism. EH can capture recyclable external energy, includingsolar, indoor lightening, vibrational, chemical and humanmotion energies [7]. The utilization of EH devices in MCSwill provide new features, such as self-sustainability, coverageof sensing area and the quality of sensing result. In this paper,we adopt EH devices in MCS. The system will use the energythat can be harvested by the participants as a reward.

In the traditional MCS sensing task execution, participantsonly can choose to be a member and donate their own resourceor choose not to be involved at all. However the traditionalmethod maybe reduce the possibility of a potential participantby only two choices: to do or not to do. In our work, we in-troduce computation offloading [8], [9] into MCS sensing taskexecution. Thus, the participants can make different choices byanalyse their resource. In addition of EH, participants’ batterycan be maintained and the sensing life of the system can beprolonged.

In this paper, we adopt a game theoretic solution to addressthe reward-aided sensing task execution in MCS, in additionto the QoS and system lifetime of MCS. Game theory focuseson the features of participant competition and interactions[7], [10], [11]. In an MCS sensing environment, any actiontaken by a participant affects the decisions of others in themobile network. Thus, game theory is a natural mathematical



2

Fig. 1. Mobile Crowdsensing with Energy Harvesting Scenario

approach for studying this interaction. In this paper, theStackelberg leadership model is a promising way to provide anincentive mechanism and solve the reward allocation problem,which enables the system and the participants to maximizetheir utilities at the same time [12], [13].

In this paper, we study the issue of reward-aided sensingtask execution, in addition to sustainable system lifetime inMCS, and model it as a two-stage Stackelberg game. EH asa reward will also act as the incentive mechanism in MCS.This approach will reduce the energy consumption of mobileusers. By giving different participants different rewards, thisapproach helps to control the sensing quality of the data. Insummary, the contributions of this paper are as follows:• Reward-aided sensing task execution game formula-

tion: The proposed Stackelberg game consists two parts.The first stage of the game is a system reward game(SRG). In the SRG, the system that allocates the tasks andreward is the leader, and the mobile devices that executethe tasks are the followers. The second stage of the gameis called a participant decision-making game(PDG); boththe system and mobile devices are players in the secondstage. The strategy of the system is the optimal reward,and the optimal strategy of the participants is how toexecute the task.

• Properties of the game: The PDG in the frameworkensures that the SRG has an optimal result. Because thesub-game has an NE, all the mobile devices can achievea mutual satisfaction. As a result of the sub-game, theparticipants in MCS will be separated into two groups:those who offload the task and those who execute the tasklocally. Based on their different decisions in sensing taskexecution, the system will allocate the reward to differentparticipants.

• Reward-aided sensing task execution mechanism andmain algorithms: The proposed Stackelberg game isachieved by two main algorithms, which operate intandem: In PDG, algorithm1 takes communication inter-ference, computation time and energy into account andachieves the Nash equilibrium of the decision-makinggame. In SRG, based on the result of algorithm1, al-gorithm 2 can achieve the optimal reward amount andallocate the reward.

The remainder of this paper is organized as follows. Insection II, we discuss the related works. The system model is

introduced in section III. The Stackelberg game and reward-aided sensing task execution mechanism are presented in sec-tions IV and V, respectively. The performance and simulationof the mechanism are analysed in section VI. In section VII,we conclude the paper.

II. RELATED WORK

The incentive mechanism and sensing quality play crucial rolesin MCS. Related works have showed different incentive mech-anisms: they adopt money, reputation and credit as rewards forparticipants [5], [6].

In [14], two types of incentive mechanisms are intro-duced: a user-centric model based on the Stackelberg gameand a system-centric model based on auction. In [15], theauthors proposed a new crowdsensing framework, namely,social network assisted trustworthiness assurance (SONATA),which aims to maximize the crowdsensing platform utilityand minimize the manipulation probability through vote-basedtrustworthiness analysis in a dynamic social network architec-ture. [16] designed an inventive mechanism for discrete crowd-sensing in which each user has a uniform sensing subtasklength. The objection of this work is to maximize the platformutility and achieve perfect Bayesian equilibrium. [17] studiedthe incentive mechanisms for a novel Mobile CrowdsensingScheduling problem, which achieved desirable truthfulness,individual rationality and computational efficiency. Howeverthere is no work which takes energy as a reward in MCS.In our work, energy plays a significant role in MCS, becauseenergy as a reward can guarantee the quantity of the partici-pants, and then sustain the coverage of sensing area, all theseimprove the QoS in MCS.

For the sensing quality of MCS, Jin et al. [18] proposeda reverse auction approach for the incentive mechanism ofMCS based on quality of information (QoI). In [19], theauthors presented a framework for green mobile crowdsensingthat utilized a quality-driven sensor management functionto continuously select the k-best sensors for a predefinedsensing task. In [20], a novel approach was proposed thatuses the techniques of evolutionary algorithms to determinethe optimal trade-off between data quality and cost. In ourwork, we adopt computation offloading as an additional optionfor the participants. Computation offloading make it possibleto execute heavy sensing task in the cloud and return the resultdirectly back to the server. Budgets in MCS is also an essentialparameter when MSC platform maximizes the sensing qual-ity. [21] proposed a novel task allocation framework calledCrowdTasker for MCS. CrowdTasker aims to maximize thecoverage quality of the sensing task under a budget constraintby greedy algorithm. [22] and [23] proposed a framework toachieve the optimal coverage quality under budget constraint.

For EH, this approach is widely applied in many scenariosin wireless sensor networks [24]–[28]. Wang et al. [29] studiedjoint channels and power allocation to improve the energy effi-ciency of user equipment by analysing the batteries of the userequipments. In [30], Mao et al. investigated a green mobile-edge computing system with energy harvesting devices and de-veloped an effective computation offloading strategy. In [31], a



3

tractable model was developed for analysing the performanceof downlink heterogeneous cellular networks with both power-grid-connected base stations and energy harvesting small cellaccess points. In [32], energy harvesting has also been takeninto consideration in cognitive radio sensor networks (CRSNs).It addressed a network utility maximization problem which isgreatly impacted by sampling rate control and channel accessschedule, under the harvested energy, channel capacity andinterference constraints.

Our work is inspired by the works on the interference amongthe users in mobile networks and a game theoretic approach formaximizing the system utility [14], [33]. However, our workdiffers the above mentioned related works in the followingways: (a) we consider interference among participants inMCS to propose an efficient sensing task execution approach.(b) we adopt EH as a reward in the incentive mechanism,which prolongs the system lifetime of MCS. (c) our workfaces a multi-parameter environment where the participantsinformation is multi-dimensional.

III. SYSTEM MODEL

We introduce the system model of MCS in this section. Asshown in Fig. 1, we consider in a framework that consists of aset of participants P = {1, 2, ..., p},where participant i will beassigned a sensing task Ti = {Bi, Di} to be executed, which ispublished by the system. Here, Bi denotes the data size of thecomputation input data (which including the sensing data andthe execution code), and Di denotes the required CPU cyclesof participant i. There is a wireless base station that allowsparticipants to offload computations to the cloud. Participantsin MCS can decide whether to execute locally or to offloadthe task to the cloud via the wireless network. We denoteby di as the sensing task execution decision of participant i.Specifically, we have di = 0 if the participant will execute thesensing task locally, and di = 1 if the participant chooses tooffload the task. The decision profile is d = {d1, d2, ...dn}. Inthis scenario, EH is a special feature of the MCS participants.Hence, the sensing task execution model, EH model andnetwork conditions in MCS will be discussed. We will focuson the efficient sensing tasks execution procedure and theenergy-aided incentive mechanism in MCS.

A. Task Execution model

In MCS, there are two potential ways to execute the sensingtasks. Participants can choose either of them to maximize theirutility. In this section, we introduce the sensing task executionmodel in MCS.

1) Local Execution Model: We consider each i ∈ P has asensing task Ti, which is published by the system. We willdiscuss execution delay and energy consumption of the localexecution model.

For local execution, a participant will execute the task withthe local resources of the mobile device and generate thesensing result ri with data size Bri , Bri can be larger or smallerthan the sensing task size Bi based on the sensing task. Thecomputational capacity of the participant i is denoted as Fi.

The local execution delay T l1i and energy consumption El1ican be expressed as follows:

T l1i =Di

Fi(1)

El1i = ϕDi (2)

where ϕ = KF 2i denotes the energy consumption per CPU

cycle, K is energy coefficient based on the structure ofthe chips [34]. In sensing result transmission process, thetransmission delay T l2i and energy cost El2i can be expressedas follows:

T l2i =BriRi

(3)

where Ri is the data transmission rate of participant i. Basedon (3), we can have the energy consumption of the resulttransmission stage.

El2i = PiBriRi

(4)

where Pi is the transmission power of participant i. Accordingto (1), (2), (3) and (4) we can obtain the local sensing taskexecution cost Cli as:

Cli = w1(Tl1i + T l2i ) + w2(E

l1i + El2i )

= w1(Di

Fi+BriRi

) + w2(ϕDi + PiBriRi

) (5)

where w1, w2 ∈ (0, 1) are the coefficients of the execution de-lay and energy consumption. Ri is the data rate for participanti that we simply assume is the same to every local executionparticipant due to the interference can be discarded. A partic-ipant will compute the sensing cost before deciding whetherto execute the sensing task locally or in the cloud. Note thatexecution delay and energy consumption are parameters withdifferent scales, so we will use a normalization method toconvert the parameters into a common scales.

2) Cloud Execution Model: In the cloud execution model,the sensing task will be executed in the cloud. For a partici-pant, the cost of cloud execution includes two parts: the delaycontributed by transmission in the network and execution inthe cloud and the energy consumption of offloading the sensingtask. The execution delay of the cloud execution model isdenoted as T oi , which is given by

T oi =BiRi(d)

+Di

F ci(6)

where Ri(d) is different from Ri in (4), because interferencewill happen when participants decide to offload the sensingtask. And F ci is the computational ability of the cloud whichwe assume the system will offer every participant the same



4

computation ability. We denoted the energy consumption inthe cloud execution model as Eoi :

Eoi = Pi ·BiRi(d)

(7)

According to (5) and (6), we can compute the cost of the cloudexecution, which is denoted as Coi :

Coi = w1(BiRi(d)

+Di

F ci) + w2Pi

BiRi(d)

(8)

B. Network Model

Due to the scenario of MCS, when a big amount of mobiledevices begin to offload the sensing task to the cloud, thecommunication interference among the participants need tobe considered about. In the mobile network, the base stationcan manage the communications of all mobile users, includinguplink and downlink communications. The data transmissionrate of participant i is a logarithmic function of SINR. SINRis denoted as γi(d) [33], [35]

Ri(d) = f(γi(d))

= f(PiHi,b

σ2 +∑m∈P,m6=i,dm=1 PmHm,b

)

=W log2(1 +PiHi,b

σ2 +∑m∈P,m6=i,dm=1 PmHm,b

)

(9)

where Pi and Hi,b represent the transmission power of par-ticipant i and channel gain between the participant i andbase station, respectively; σ2 denotes the noise power level,including noise power and interference power. Accordingto the equations above, if too many participants decide tooffloading the sensing task, SINR will decrease, interferencewill incur and lead to low data rates, which negatively affectMCS. Thus in the offloading phrase, the interference needs tobe taken into consideration.

IV. REWARD-AIDED SENSING TASK EXECUTION GAME

In MCS, the system is interested in maximizing its utilitywhile publishing the tasks and rewards for the participants.At the same time, the participants who own the mobiledevices are both selfish and rational; hence, they also want tomaximize their own utility. The participant must compute thecost of execution based on communication interference, energyconsumption and battery level. If the system will give the hima reward that is not less than the cost, then he will participatein MCS. Firstly in our work, we assume that the system wouldlike to have more participants to offload the sensing task to thecloud, due to the capacity of the cloud and the transmissiondelay of local execution. Thus there will be more reward thatwill be allocated to the offloading participants.

The proposed reward-aided sensing task execution mecha-nism is achieved by Stackelberg game, where the system isthe leader who moves first in the game and the participantsare the followers. In Stackelberg game, there are two stages:

Stage 1: SRG

Strategy: rewardObjective: system cost minimization

Stage 2: PDGStrategy: how to participate

Objective: profit maximization

s.t. the interference of the channel

Stackelberg Game

Rewardd=f(reward)

Fig. 2. Stackelberg Game

an SRG, in which the system provides reward for users whoparticipate in computation and a PDG, in which participantscan decide by themselves whether to offload the sensing taskor execute it locally according to the reward from the system.In order the solve the whole Stacketlberg game, we need tosolve the second stage of the game (PDG) first based on thereward from the system [36].

Before starting, there are serval important questions regard-ing the game, the first of which is how to choose a reasonableUtility Function (UF). After a UF has been selected, we mustdetermine what kind of strategy a participant can choose tomaximize or minimize the UF; in another words, what is thebest response of a participant? Moreover, will a stable state(NE) exist for all the participants? If the system can achieveNE, will it be unique? In this section, we will address severalimportant questions while formulating the two-stage game.

A. Utility Functions

1) Stage One of Stackelberg Game : In an SRG, the leaderis the system, who makes the first move in the game, andthe followers are the mobile users. The UF of the systemis the cost of the system after allocating the reward to theparticipants, which can be formulated as

US =

p∑i=1

EiI(di=1) +

p∑i=1

E′iI(di=0) (10)

where I is an indicator function, if di = 1 is true, then theindicator function is true and vice versa. In equation (10), Eidenotes as the reward for offloading participants and E′i as thereward to local participants. E′i = εEi for local participants



5

TABLE INOTATION AND DESCRIPTIONS

Notation Description

P = {1, 2, ...n} set of participators

Fi Participator i’s computation ability

Di i’s Required CPU circles of the sensing task

Bi computation input data size

Bri Sensing result of participant i

Ri Uplink data rate for participator i

F c Computation ability of the cloud

W Channel bandwidth

Pi Transmission power of participant i

Hi,b Channel gain between participant i and base station

a1, a2, b1, b2, w1, w2 Weights

d = d1, d2, ..., dn Decision profile of participators

eci i’s energy consumption

ehi i’s harvested energy

eni i’s new energy level after energy harvesting

eii i’s initial energy level

eb i’s battery ability

where ε is constant. In the first stage of Stackelberg game, thegoal is to minimize the cost of the system, which implies

mind,Ei

US

s.t. Ei ≥ Eoi ∀i ∈ PE′i ≥ El1i + El2i ∀i ∈ P

(11)

2) Stage two of Stackelberg Game : The participant i’sutility function is the profit he can make by obtaining rewardfrom the system minus the cost of sensing task executions

UPi =

{Ei − Coi if di = 1

E′i − Cli if di = 0(12)

where Ei and E′i are different rewards he can obtain fromdifferent sensing task execution plan. In the second stage ofStackelberg game, the utility function of participant should bemaximized for each participants.

maxd

UPi (13)

B. Game formulation and Property

Definition 1 (Stackelberg Equilibrium). (d∗, E∗i ) is aStackelberg Equilibrium for the proposed game if it satisfiedthe following conditions for any value of (d, Ei)

US(E∗i , d∗) ≥ US(Ei, d∗)

UPi (d∗i , d−i) ≥ UPi (d′i, d−i)(14)

According to definition 1, the Stackelberg Equilibrium canbe obtained as follow: in PDG, equilibrium depends on thefollowers’ optimal response of sensing task execution planwhere they will obtain the optimal strategy profile d∗. In SRG,system uses the optimal strategy profile of PDG to obtain the

optimal reward E∗i for the participants. Therefore, we need toanalysis the PDG first, to solve the SRG.

First, we define the Best Response (BR) in the PDG. TheBR is a central concept in game theory. It is a strategy thatproduces the maximum profit for a player in the game, giventhe other players’ strategies.

Definition 2 (Best Response in a Decision-making Game).Participant i’s strategy d∗i is the Best Response to strategiesd−i of other participants, if

UPi (d∗i , d−i) > UPi (d′i, d−i) ∀di ∈ d, d∗i 6= d′i (15)

In a PDG, all the participants will act according to their BRwhen playing the game. Hence, the PDG will eventually reacha stable point, which we call it NE.

C. Solution of Stackelberg Game

1) Stage2 – PDG: In order to solve the Stackelberg game,we apply backward induction. First we need to solve thesecond stage of the game. We take the reward Ei and E′ias given in stage 2.

Theorem 1 (Best Response strategy) According to Defi-nition 2 and the action profile of participant i is di = {0, 1},where 0 means local execution and 1 means cloud execution.Here we will discuss which action of the participant is the bestresponse towards to other participants’ actions. We assume thebest response of participant i is d∗i , according to the fixedstrategy profile d−i of other participants. Thus we have

UPi (d∗i , d−i) =

{Ei − Coi + UPi (d−i) if d∗i = 1

E′i − Cli + UPi (d−i) if d∗i = 0(16)

in BR d∗i 6= d′i, thus we assume when d∗i = 1, we obtain



6

UPi (1, d−i) > UPi (0, d−i)

where we have

Ei − Coi + UPi (d−i) ≥ E′i − Cli + UPi (d−i)

Coi − Cli ≤ Ei − E′i

on the other hand, when d∗i = 0, we have

Coi − Cli ≥ Ei − E′i (17)

As it shows in (17), we need to discuss the relation betweenCli and Coi . According to (5) and (8), we have the left side of(17)

w1(BiRi(d)

+Di

F c) + w2Pi

BiRi(d)

− Cli

where we can find out the data rate when participants offloadthe sensing task to the cloud will strongly effect the cost ofsensing task execution. Based on (9) when all the participantswant to offloading the task, they will suffer a high interferencein the channel. This will cause the increasing of the cost ofoffloading. Thus in PNG, the best response is when participanttry to choose the cloud execution, which implies

d∗i =

{1 if Coi − Cli ≤ Ei − E′i0 otherwise

(18)

To simplify the problem, we assume all the participants arehomogenous, which means they are under the same chan-nel condition and same transmission power, thus P1H1,s =P2H2,s = ... = PpHp,s = k. Based on d∗i = 1 is the thebest response of participants i. According to (9) and (18), weobtain

0 <

p∑i=1

kI(di=1) ≤k

2

w1Bi+w2BiPi

(Cli+Ei−E′i−w1

DiFci

)W − 1

− σ2 (19)

When∑pi=1 kI(di=1) is out of the range of (19), the utility

function will decrease, so the participants will choose toexecute locally. We can obtain the strategy profile d∗ of theparticipants

p∑i=1

I(di=1) =1

2

w1Bi+w2BiPi


DiFci

)W − 1

− σ2

k(20)

2) Stage1 – SRG: From Stage2, we obtain the optimalstrategy profile d∗. Now we will solve Stage1. According to(10) and (19), in addition with E′i = εEi, we obtain

mind∗,Ei

n∑i=1

EiI(di=1) +n∑i=1

εEiI(di=0)

= (1

2Ai

Bi+CiEi − 1− σ2

k)Ei

+ (n− (1

2Ai


k))εEi

= (1− ε)( 1

2Ai


k)Ei

+ εnEi

(21)

whereAi = w1Bi + w2BiPi

Bi = (Cli − w1Di

F ci)W

Ci = (1− ε)Wand n is the total number of the participants in MCS. In Stage1,we want to find out the smallest value of function (24).Thus,the derivative of (24) is

d(US)

d(Ei)=

(1− ε)(2Ai

Bi+CiEi − 1)

(2Ai

Bi+CiEi − 1)2

+AiCiln2(1− ε)Ei2

AiBi+CiEi

1(Bi+CiEi)2

(2Ai

Bi+CiEi − 1)2

− [(1− ε)σ2

k− εn]

(22)

where ε ∈ (0, 1),Ai, Bi, Ci > 0. And then we need to discussabout the second derivative of (23), namely

d2(US)

d(Ei)2=

d

d(Ei)[(1− ε)(2

AiBi+CiEi − 1)+

(2Ai

Bi+CiEi − 1)2

+AiCiln2(1− ε)Ei2

AiBi+CiEi

1(Bi+CiEi)2

(2Ai

Bi+CiEi − 1)2]

(23)

According to (23), we can easily find out d2(US)d(Ei)2

is positive,thus (22) is concave. Setting the first derivative of US to 0,we obtain the estimated Ei

Emini =Ai −Bi

Ci(σ2

k −εn1−ε )

− BiCi

(24)

According to the range of Ei, namely

Ei =

{max[Eoi , E

l1i + El2i ] if Emini < Eoi , (E

l1i + El2i )

Emini else(25)

According to (25), we can obtain the optimal reward fromsystem to every participants individually based on how theparticipants execute the sensing task.



7

D. Energy Harvest Model

Now we introduce the battery level update of each participantafter reward allocated. The mobile devices in the MCS areequipped with EH components.

We use ECi , Ei, ENi , EIi and EBi to denote the energyconsumption, the energy that can be harvested as reward, thenew energy level, the initial energy level and battery capabilityof the participants, respectively. According to (2), (4), (7), wehave

ECi =

{EoiEl1i + El2i

(26)

According to energy aided reward in MCS, we have

ENi =

{EBi if EIi − ECi + EHi ≥ EBiEIi − ECi + EHi otherwise

(27)

Equation (27) shows the new battery level of participants afterenergy harvesting.

V. REWARD-AIDED SENSING TASK EXECUTIONMECHANISM

In this section, we introduce the reward-aided sensing taskexecution mechanism and the main algorithms. We supposethat the system publishes a sensing task that requires themobile devices in the specific area to record a video for datamining.

A. Mechanism Design

• The system will announce the task to all the participantsin the specific area, including all the parameters of thesensing task (the size of the task, required CPU, etc.) andthe reward to the participants based on different sensingtask execution plan;

• Based on the parameters of the sensing task, the param-eters of the participants (battery level, bandwidth, dataplan, and CPU), the channel in the mobile network andthe reward, the participants will decide whether to executethe sensing task locally or offload it to the server;

• The system will allocate the rewards to different par-ticipants based on the sensing task execution plan. Theparticipants will get the energy and renew their batterylevel;

• The cost of system is optimized and the lifetime of thesystem will be prolonged.

B. Main Algorithms

Based on Theorem 1, Algorithm 1 is designed to achieveoptimal strategy profile in a decision-making game. In otherwords, after the algorithm 1 has been carried out, all the par-ticipants will be separated into two groups. In this algorithm,first the cost of execution will be calculated and the differencewill be ranked in ascending order; then, based on the bestresponse, the participants will make different decisions. Here

Algorithm 1 Participators Decision-making Game1: Input: metrics of channel, task, participators’ ability and

reward2: Output: d∗

3: Let Coi [co1 . . . c

oi ], C

li [c

l1 . . . c

li] and d[d1 . . . di] be new

arrays4: Let i = 15: for i = 1 to n do6: Ci = Coi − Cli7: end for8: Sort Ci in ascending order9: Let j = 1

10: for j = 1 to n do11: if

∑ni=1 kI(di=1) ≤ k

2

w1Bi+w2BiPi


DiFci

)W

−1

− σ2 then

12: dj = 113: d∗ = dj

⋃d

14: end if15: end for

in Algorithm1, we obtain the optimal decision profile of thesecond stage of Stackelberg game. Algorithm 2 determines thereward that will be allocated to the participants. We adopt theresult of the solution of stage1 to find the optimal reward forthe system to allocate to the participants. It is based on theutility functions of the SRG.

Algorithm 2 System Reward Allocation1: Input: d2: Output: reward Ei and the new battery level ENi3: Let reward = 14: for i = 1 to n do5: if di 6= 0 then6: i++7: end if8: end for9: Ei = max( Ai

σ2

k −εn1−ε− Bi

Ci, Pi

BiRi(d)

, 1ε (ϕDi + PiBriRi

)

10: for j = 1 to n do11: if di = 1 then12: EIi − ECi + Ei13: else14: EIi − ECi + εEi15: end if16: end for

Here, we analyse the battery levels of all the participantsto determine whether the proposed mechanism can achievea longer system lifetime. Although system lifetime is animportant parameter in MCS, there has been little work in thisarea. Therefore, a deeper collaborative approach for energyharvesting in MCS will be developed in the future work.

VI. PERFORMANCE EVALUATION

We evaluate the performance of the reward-aided sensing taskexecution mechanism and its algorithms by performing sim-ulations. The simulation results are obtained using MATLABon a computer with Intel i5 at 1.3 GHz.



8

A. Simulation Setup

We simulate an MCS environment that consists of 20 partici-pants with a sensing task with a data size of 420KB and 1000Megacycles of required CPU. The computational capability ofeach participant is randomly selected from 1.0, 0.8 and 0.5GHz. The bandwidth of the channel is set as 5.5 MHz, thetransmission power is 100 mWatts, and the background noiseis -100 dBm. The capability of the cloud is 100 GHz.

To represent the sensing quality, we proposed the sensingquality metric to measure the QoS of the system [37], whichis denoted by ψ

ψ =

∑pi=1 σ1F

ci I(di=1) +

∑pi=1 σ2FiI(di=0)

Expected(28)

where Expected is the expected quality level of the MCSsystem; here, we set Expected = 100. The QoS indexindicates how MCS executes the sensing task; since the cloudhas more computation power for better executing the sensingtask, we set 1 > σ1 > σ2 > 0.

B. Performance Evaluation of the Rewarded-aided SensingTask Execution Mechanism

The following simulation results provide an insight intothe performance of the reward-aided sensing task executionmechanism. The metrics that interest us include the utilitiesof the system and participants, system lifetime and sensingquality, along with the effects of the numbers of participants,sensing task size and reward. The performance of the proposedmechanism is also compared with that of traditional MCSsensing task execution strategy, in which all tasks are sensedand executed locally. In addition to the local strategy, we alsoadopt a non-cooperative game as a new comparison [38]. In anon-cooperative game, based on equation (12) the participant’sutility function, the participants will compete according to afixed reward to maximize the utility by offloading sensingtask.Thus, we set a non-cooperative algorithm and local sens-ing task execution as the benchmarks.

In Fig. 4, we show the utilities of system and participantsin the scenario in which the MCS system consists of 20participants and a set reward, along with a task size range from0 KB to 5000 KB. According to the figure, when the task sizeis small, the values of the utilities are stable because most ofthe participants will execute a small task locally. As the tasksize increases, the participants’ utility decreases, while the sys-tem’s utility increases. Note that the system’s and participants’utilities intersect when the task size is approximately 1250 KB,which means that in this scenario, the task size is optimal forthe MCS system. Compared to the proposed mechanism, theutility of the traditional sensing task execution method is muchsmaller due to the low computational capability.

System lifetime is a crucial metric in MCS and can indicatehow healthy an MCS is. In this paper, we adopt the reward-aided mechanism to prolong the system lifetime of MCS.We define that the system lifetime in terms of the lowestbattery level of the participants. Fig. 5, Fig. 6 and Fig. 7 showthe impacts of the sensing task size, reward and number of

0 1000 2000 3000 4000 5000

Task Size

0

1

2

3

4

5

6

7

8

Utilit

y

×104 Utility and tasksize

System UtilityParticipants UtilitySystem Utility(local)Participants Utility(local)

Fig. 3. System/ Participants Utility and Task Size

0 5 10 15 20 25 30 35 40 45

Participants

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

Syste

m u

tilit

y

×104 System utility and participants

Energy aided reward gameLocal execution

Fig. 4. System Utility and Participants

400 600 800 1000 1200Task size

0

10

20

30

40

50

60

Syste

m lifetim

e

System lifetime and task size

with EHwithout EHlocal with EHlocal without EH

Fig. 5. System Lifetime and Task Size

participants, respectively, on system lifetime. Fig. 5 shows theresult of both mechanism, with and without energy harvesting.The proposed mechanism achieves a slower decrease and abetter lifetime than the other methods. Fig. 6 shows that withthe same reward, the proposed mechanism achieves a longerlifetime than the traditional one. Moreover, it shows that whenthe number of participants increases, the proposed mechanismbecomes less competitive since the number of participantsis not optimal for this special scenario. Fig. 7 shows howthe number of participants affects the system lifetime. As thenumber of participant increases, the system life time stabilizes.

Fig. 8 shows the relation between reward and sensing qual-ity. According to the figure, the sensing quality increases whenthe reward increases because more participants will offloadthe sensing task for execution, which achieves better quality.



9

0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000

Reward

0

5

10

15

20

25

30

Syste

m lifetim

e

System lifetime and reward

Energy aided reward game(20 participants)Local with EH(20 participants)Energy aided reward game(40 participants)Local with EH(40 participants)

Fig. 6. System Lifetime and Reward

0 10 20 30 40

Participants

0

10

20

30

40

50

60

70

80

90

100

Syste

m life

tim

e

System lifetime and participants

Energy aided reward gamelocal executionnon-cooperative game

Fig. 7. System Lifetime and Participants

0 2000 4000 6000 8000 10000 12000

Reward

0

5

10

15

20

25

Ave

rag

e s

en

sin

g q

ua

lity

Average sensing quality and reward

Energy aided reward game(tasksize 1000)local execution(tasksize 1000)Energy aided reward game(tasksize 2000)Non-cooperative game

Fig. 8. Sensing Quality and Reward

However, after a certain point, as the reward increases, thesensing quality does not change. This is due to the interferenceof the channel, as no more participants can offload the sensingtask.

VII. CONCLUSION

In this paper, a reward-aided sensing task execution mecha-nism for MCS has been proposed. The proposed mechanismaims to improve the sensing quality and prolong the systemlifetime by assigning energy as a reward to the participants.The proposed mechanism adopts game theory to solve thismulti-objective optimization problem. The mechanism pre-sented in this paper considers both the network channel condi-tions and participants’ abilities and adopts energy as the rewardfor the participants. Extensive simulations have demonstrated

the advantages of the proposed mechanism, which yields abetter QoS and a longer system lifetime for MCS.

In future work, we plan to make possible deeper collabora-tion between energy harvesting and MCS. We will investigatethe sensing data and energy transmission in MCS to achievea better QoS of MCS.

ACKNOWLEDGMENT

This work is funded by China Scholarship Council, NationalKey R&D Plan of China under Grant No. 2017YFA0604500,National Sci-Tech Support Plan of China under Grant No.2014BAH02F00, and by National Natural Science Foundationof China under Grant No. 61701190, and by Youth ScienceFoundation of Jilin Province of China under Grant No.20160520011JH, and by Youth Sci-Tech Innovation Leaderand Team Project of Jilin Province of China under GrantNo. 20170519017JH, and by Key Technology InnovationCooperation Project of Government and University for thewhole Industry Demonstration under Grant No. SXGJSF2017-4. and China Postdoctoral Science Foundation under GrantNo. 2018M631873 .

REFERENCES

[1] Bin Guo, Zhu Wang, Zhiwen Yu, Yu Wang, Neil Y Yen, Runhe Huang,and Xingshe Zhou. Mobile crowd sensing and computing: The reviewof an emerging human-powered sensing paradigm. ACM ComputingSurveys (CSUR), 48(1):7, 2015.

[2] Dong Zhao, Huadong Ma, and Shaojie Tang. Coupon: Cooperativelybuilding sensing maps in mobile opportunistic networks. In MobileAd-Hoc and Sensor Systems (MASS), 2013 IEEE 10th InternationalConference on, pages 295–303. IEEE, 2013.

[3] Yoshito Tobe, Itaru Usami, Yusuke Kobana, Junji Takahashi, GuillaumeLopez, and Niwat Thepvilojanapong. vcity map: Crowdsensing towardsvisible cities. In SENSORS, 2014 IEEE, pages 17–20. IEEE, 2014.

[4] Shu-Ting Wang, Ching-Ling Fan, Yi-Ying Huang, and Cheng-Hsin Hsu.Toward optimal crowdsensing video quality for wearable cameras insmart cities. In Computer Communications Workshops (INFOCOMWKSHPS), 2015 IEEE Conference on, pages 624–629. IEEE, 2015.

[5] Luis G Jaimes, Idalides J Vergara-Laurens, and Andrew Raij. A surveyof incentive techniques for mobile crowd sensing. IEEE Internet ofThings Journal, 2(5):370–380, 2015.

[6] Cem U Saraydar, Narayan B Mandayam, and David J Goodman.Efficient power control via pricing in wireless data networks. IEEEtransactions on Communications, 50(2):291–303, 2002.

[7] Farhad Meshkati, H Vincent Poor, Stuart C Schwartz, and Radu V Balan.Energy-efficient resource allocation in wireless networks with quality-of-service constraints. IEEE Transactions on Communications, 57(11),2009.

[8] Yating Wang, Ray Chen, and Ding-Chau Wang. A survey of mobilecloud computing applications: perspectives and challenges. WirelessPersonal Communications, 80(4):1607–1623, 2015.

[9] Karthik Kumar, Jibang Liu, Yung-Hsiang Lu, and Bharat Bhargava. Asurvey of computation offloading for mobile systems. Mobile Networksand Applications, 18(1):129–140, 2013.

[10] Andrea Galeotti, Sanjeev Goyal, Matthew O Jackson, Fernando Vega-Redondo, and Leeat Yariv. Network games. The review of economicstudies, 77(1):218–244, 2010.

[11] Hai-Yan Shi, Wan-Liang Wang, Ngai-Ming Kwok, and Sheng-YongChen. Game theory for wireless sensor networks: a survey. Sensors,12(7):9055–9097, 2012.

[12] Heinrich Von Stackelberg. Market structure and equilibrium. SpringerScience & Business Media, 2010.

[13] John Nash. Non-cooperative games. Annals of mathematics, pages 286–295, 1951.

[14] Dejun Yang, Guoliang Xue, Xi Fang, and Jian Tang. Crowdsourcingto smartphones: Incentive mechanism design for mobile phone sensing.In Proceedings of the 18th annual international conference on Mobilecomputing and networking, pages 173–184. ACM, 2012.



10

[15] Burak Kantarci, Kevin G. Carr, and Connor D. Pearsall. Sonata:SocialNetwork Assisted Trustworthiness Assurance in Smart City Crowdsens-ing. International Journal of Distributed Systems and Technologies,7(1):59–78, 2016.

[16] Shiyu Ji and Tingting Chen. Incentive mechanisms for discretizedmobile crowdsensings. IEEE Transactions on Wireless Communications,15(1):146–161, 2016.

[17] Kai Han, Chi Zhang, Jun Luo, Menglan Hu, and Bharadwaj Veeravalli.Truthful scheduling mechanisms for powering mobile crowdsensing.IEEE Transactions on Computers, 65(1):294–307, 2016.

[18] Haiming Jin, Lu Su, Danyang Chen, Klara Nahrstedt, and JinhuiXu. Quality of information aware incentive mechanisms for mobilecrowd sensing systems. In Proceedings of the 16th ACM InternationalSymposium on Mobile Ad Hoc Networking and Computing, pages 167–176. ACM, 2015.

[19] Martina Marjanovic, Lea Skorin-Kapov, Kresimir Pripuzic, AleksandarAntonic, and Ivana Podnar Zarko. Energy-aware and quality-drivensensor management for green mobile crowd sensing. Journal of networkand computer applications, 59:95–108, 2016.

[20] Hong Nhat Pham, Back Sun Sim, and Hee Yong Youn. A novelapproach for selecting the participants to collect data in participatorysensing. In Applications and the Internet (SAINT), 2011 IEEE/IPSJ11th International Symposium on, pages 50–55. IEEE, 2011.

[21] Haoyi Xiong, Daqing Zhang, Guanling Chen, and L Wang. Crowdtasker:Maximizing coverage quality in piggyback crowdsensing under budgetconstraint. In IEEE International Conference on Pervasive Computingand Communications, pages 55–62, 2015.

[22] Maotian Zhang, Panlong Yang, Chang Tian, Shaojie Tang, XiaofengGao, Baowei Wang, and Fu Xiao. Quality-aware sensing coverage inbudget-constrained mobile crowdsensing networks. IEEE Transactionson Vehicular Technology, 65(9):7698–7707, 2016.

[23] Zhenzhe Zheng, Fan Wu, Xiaofeng Gao, Hongzi Zhu, Guihai Chen,and Shaojie Tang. A budget feasible incentive mechanism for weightedcoverage maximization in mobile crowdsensing. IEEE Transactions onMobile Computing, PP(99):1–1, 2017.

[24] Sennur Ulukus, Aylin Yener, Elza Erkip, Osvaldo Simeone, MicheleZorzi, Pulkit Grover, and Kaibin Huang. Energy harvesting wirelesscommunications: A review of recent advances. IEEE Journal on SelectedAreas in Communications, 33(3):360–381, 2015.

[25] Kun Yang, Qin Yu, Supeng Leng, Bo Fan, and Fan Wu. Data and energyintegrated communication networks for wireless big data. IEEE access,4:713–723, 2016.

[26] Kaibin Huang, Marios Kountouris, and Victor OK Li. Renewablepowered cellular networks: Energy field modeling and network cover-age. IEEE Transactions on Wireless Communications, 14(8):4234–4247,2015.

[27] Berk Gurakan, Omur Ozel, Jing Yang, and Sennur Ulukus. Energycooperation in energy harvesting two-way communications. In Commu-nications (ICC), 2013 IEEE International Conference on, pages 3126–3130. IEEE, 2013.

[28] Shuguang Cui, Andrea J Goldsmith, and Ahmad Bahai. Energy-efficiency of mimo and cooperative mimo techniques in sensor networks.IEEE Journal on selected areas in communications, 22(6):1089–1098,2004.

[29] Feiran Wang, Chen Xu, Lingyang Song, and Zhu Han. Energy-efficientresource allocation for device-to-device underlay communication. IEEETransactions on Wireless Communications, 14(4):2082–2092, 2015.

[30] Yuyi Mao, Jun Zhang, and Khaled B Letaief. Dynamic computationoffloading for mobile-edge computing with energy harvesting devices.IEEE Journal on Selected Areas in Communications, 34(12):3590–3605,2016.

[31] Pei-Shan Yu, Jemin Lee, Tony QS Quek, and Y-W Peter Hong. Trafficoffloading in heterogeneous networks with energy harvesting personalcells-network throughput and energy efficiency. IEEE Transactions onWireless Communications, 15(2):1146–1161, 2016.

[32] Ju Ren, Yaoxue Zhang, Ruilong Deng, Ning Zhang, Deyu Zhang, andXuemin Shen. Joint channel access and sampling rate control in energyharvesting cognitive radio sensor networks. IEEE Transactions onEmerging Topics in Computing, PP(99):1–1, 2016.

[33] Xu Chen. Decentralized computation offloading game for mobile cloudcomputing. IEEE Transactions on Parallel and Distributed Systems,26(4):974–983, 2015.

[34] Mingbo Xiao, Ness B Shroff, and Edwin KP Chong. A utility-based power-control scheme in wireless cellular systems. IEEE/ACMTransactions on networking, 11(2):210–221, 2003.

[35] Theodore S Rappaport et al. Wireless communications: principles andpractice, volume 2. prentice hall PTR New Jersey, 1996.

[36] Walid Saad, Zhu Han, Merouane Debbah, Are Hjorungnes, and TamerBasar. Coalitional game theory for communication networks. IEEESignal Processing Magazine, 26(5):77–97, 2009.

[37] Liang Xiao, Tianhua Chen, Caixia Xie, Huaiyu Dai, and Vincent Poor.Mobile crowdsensing games in vehicular networks. IEEE Transactionson Vehicular Technology, 2017.

[38] Hamed Shah-Mansouri, Vincent W. S. Wong, and Jianwei Huang. Anincentive framework for mobile data offloading market under pricecompetition. IEEE Transactions on Mobile Computing, PP(99):1–1,2017.

Jiejun Hu received her B.E. degree in Computer Science from Jilin Universityin 2012. Currently she is a PhD student in Jilin University, China. Her currentresearch areas are computer network, future network and technology, resourceallocation in network, and related applications.

Kun Yang received the PhD degree from the Department of Electronic andElectrical Engineer- ing, University College London (UCL), United Kingdom.He is currently a full professor and the head of Network Convergence Labo-ratory (NCL) in the School of Computer Science and Electronic Engineering,University of Essex, United King- dom. Before joining in the University ofEssex at 2003, he worked at UCL on several European Union ResearchProjects such as FAIN, MAN- TRIP, and CONTEXT. His current majorresearch interests include heterogeneous wireless networks, fixed mobileconver- gence, and cloud computing. He has published more than 150 journaland conference papers in the above areas. He serves on the editorial boardsof both IEEE and non-IEEE journals. He is a senior member of the IEEE anda fellow of the IET. Xiaohui

Liang Hu received the B.S. degree in computer systems from the HarbinInstitute of Technology in 1993, and the PhD degree in computer soft- wareand theory in 1999. Currently, he is the professor and PhD supervisor ofthe College of Computer Science and Technology, Jilin Univer- sity, China.His main research interests include distributed systems, computer networks,communications technology and information security system, etc. As aperson in charge or a principal participant, he has finished more than 20national, provincial, and ministerial level research projects of China.

Kezhi Wang received his B.E. and M.E. degrees in College of Automationfrom Chongqing University, P.R.China, in 2008 and 2011, respectively. Hereceived his Ph.D. degree from the University of Warwick, U.K. in 2015. Hewas a senior research officer in University of Essex, U.K. Currently he is alecturer in Department of Computer and Information Sciences, NorthumbriaUniversity. His research interests include wireless communication, signalprocessing and mobile cloud computing.

northumbria research linknrl.northumbria.ac.uk/34585/1/reward-aided sensing.pdf · care, virtual...

Documents