Education

A utility minimization approach for energy-aware cooperative content distribution with fairness constraints

Description
A utility minimization approach for energy-aware cooperative content distribution with fairness constraints
Categories
Published
of 17
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  F    o  r    P   e  e  r    R   e  v   i    e  w      ! #$%&%$' (%)%*%+,$%-) !../-,01 2-/ 3)4/5'6!7,/4 8--.4/,$%94 8-)$4)$ :%;$/%<=$%-) 7%$1 >,%/)4;; 8-);$/,%)$;     "#$%&'() !"#$%&'( *#'(+',-.$(+ $( *&/&,$00"(.,'-.$(+  *'&$+,%-./ 01) 23345546557895 :-(;< 4 *'&$+,%-./ /<.;) 9;+;'%,= >%/-,(; ?;<@#%A+) 2&;%B< C-&-C-D'/-#&E C#F-(;4/#4C#F-(; ,##.;%'/-#&E +=#%/ %'&B; ,#CC$&-,'/-#&+E .%#.#%/-#&'( G'-%&;++E C-&4C'H G'-%&;++ http://mc.manuscriptcentral.com/ettEuropean Transactions on Telecommunications  F    o  r    P   e  e  r    R   e  v   i    e  w    EUROPEAN TRANSACTIONS ON TELECOMMUNICATIONS  Euro. Trans. Telecomms.  00 : 1–16 (2007) Published online in Wiley InterScience(www.interscience.wiley.com) DOI: 10.1002/ett.0000 Wireless Systems A Utility Minimization Approach for Energy-Aware Cooperative ContentDistribution with Fairness Constraints Elias Yaacoub 1 ∗ , Lina Al-Kanj 2 , Zaher Dawy 2 , Sanaa Sharafeddine 3 , Fethi Filali 1 , and Adnan Abu-Dayya 1 1 QU Wireless Innovations Center (QUWIC), Qatar Science and Technology Park, Doha, Qatar. Email:  { eliasy, filali, adnan } @quwic.com 2  Department of Electrical and Computer Engineering, American University of Beirut, Beirut, Lebanon. Email:  { lka06, zd03 } @aub.edu.lb 3  Department of Computer Science and Mathematics, Lebanese American University, Beirut, Lebanon. Email: sanaa.sharafeddine@lau.edu.lb SUMMARYCooperation between mobile terminals (MTs) with the objective of energy minimization is studied. Thepurpose is to distribute a content of common interest to collaborating MTs while ensuring at the same timea reduced energy consumption. To reach this goal, the content is sent to selected MTs on a long range link.Then, it is forwarded to other MTs on short range mobile-to-mobile links. The problem is formulated as anoptimization problem and the optimal solution is shown to consist of sending the content to a single MTon the long range link and of having that MT distribute it on the short range links. This leads to an unfairenergy consumption for the selected MT. Thus, to ensure fairness in energy consumption, a low complexityutility minimization algorithm is proposed. Using the appropriate utilities, the algorithm can be used toimplement the optimal greedy energy minimization solution or to ensure different degrees of fairness inenergy consumption. Practical constraints concerning the centralized and distributed implementations of the proposed algorithm are also discussed. Simulation results show that significant energy savings canbe achieved with the proposed approach compared to the non-collaborative case. In addition, a tradeoff between fairness and energy savings is achieved depending on the utility selected. Copyright c ￿ 2007 AEIT 1. Introduction The increase in power demand of future mobile terminals(MTs) due to the high throughput and low latencyrequirements of emerging multimedia services is one of the major challenges towards the development of nextgeneration 4G wireless networks. In fact, studies showthat the high energy consumption of battery-operatedMTs will be one of the main limiting factors for futurewireless communication systems. Emerging multimediaapplications that require the MTs’ wireless interfacesto be active for long periods while downloading largedata sizes require batteries with longer lifetime than ∗ Correspondence to: QU Wireless Innovations Center (QUWIC),Qatar Science and Technology Park, P.O.Box 5825 Doha, Qatar. E-mail:eliasy@quwic.com. what existing battery technologies can provide. In orderto tackle this limitation, mechanisms to reduce energyconsumption appear extensively in the literature, e.g.see [1, 2, 3, 4]. These mechanisms mainly rely on the fact that MTs with multiple wireless interfaces arebecoming common in next generation wireless networks.This results in a heterogeneous network architecture. Aninteresting scenario that is attracting a lot of researchinterest is the case where MTs support multiple radioaccess technologies (RAT) and the best RAT to servean MT is selected according to certain criteria, e.g. asin [5]. Another scenario consists of having a heterogeneousnetwork architecture with MTs that actively use twowireless interfaces: one to communicate with the basestation (BS) or access point (AP) over a long-range (LR)wireless technology (e.g., UMTS/HSPA, WiMAX, orCopyright c ￿ 2007 AEIT  Received 2 February 2011Prepared using  ettauth.cls   [Version: 2007/01/05 v1.00] Revised  Accepted  Page 1 of 16http://mc.manuscriptcentral.com/ettEuropean Transactions on Telecommunications 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960  F    o  r    P   e  e  r    R   e  v   i    e  w    2 YAACOUB ET AL. LTE) and one to communicate with other MTs over ashort-range (SR) wireless technology (e.g., Bluetooth orWLAN). Hence, the throughput and power limitations of a given wireless technology can be overcome by allowingcooperation among MTs over other wireless interfaces[6, 7]. It is this latter scenario that is investigated in this paper.Cooperative wireless networks proved to have a lot of advantages in terms of increasing the network throughput[8, 9, 10, 11, 12, 13], extending the network coverage [9, 14], decreasing the end user communication cost [15, 16], decreasing the file download time [10, 11, 12], and decreasingenergyconsumptionatMTs[17,18].TheiCAR architecture integrates an ad hoc component into a cellularsystem by placing stationary special-purpose relay nodesin order to help improve network throughput [19, 20]. The ICAM architecture presents an integrated cellularand ad hoc multicast, to increase the cellular multicastthroughput through the use of ad hoc relays which areMTs themselves [8]. In the UCAN architecture [9], the MTs use their WLAN interface to increase the coverageof a wireless wide area network (WWAN) and to enhancethe network throughput. The authors in [10] present acooperative mobile-to-mobile (M2M) file disseminationarchitecture over a UMTS wireless interface to increasethe network throughput and decrease the file downloadtime. In [15], an MT is assumed to be connected toseveral wireless networks with different characteristicsin terms of bandwidth, packet loss probability, andtransmission cost. Due to the complexity of theoptimization framework, a near-optimal solution shows areduction in end user cost while meeting the distortion anddelay constraints. In [13], an optimization framework forcooperative relay node selection in heterogeneous wirelesscommunication networks is presented and a suboptimalcooperative relay node selection algorithm is proposed.The considered heterogeneous network scenario assumesthat BSs communicate with MTs over a UMTS wirelessinterface whereas MTs communicate with each other overa WiMAX wireless interface. In [21], Data sub-streamdistribution and energy consumption are studied in anoptimized way using integer linear programming (ILP).The energy minimization problem is solved using CPLEXfor both unicasting and multicasting on the SR, assumingthe same energy consumption on the LR and SR links. Thebenefits of collaboration for content distribution have beeneven investigated in wired networks, e.g. in [22], wherebroadband access sharing is studied.The advantages of cooperative wireless networksare particularly important for cooperative M2M videostreaming. For example, the CHUM [18, 23, 24] and COSMOS [16] architectures assume that all users areinterested in the same video that is divided intomultiple descriptions. In the CHUM architecture, each MTrandomly selects and pulls a video description through anLR cellular link and multicasts it to all members in itscooperation group which is formed in an ad-hoc manner.In [1], a cooperative network architecture composed of anLR link technology and an SR link technology is presentedin order to reduce energy consumption among MTsduring real-time video streaming. Results show promisingopportunities to decrease the total energy consumed byincreasing the number of collaborative MTs. In [25],preliminary experimental analysis for a collaborative videostreaming architecture using test bed implementation ispresented. A group of MTs interested in the same videoare connected to a WLAN AP through which they pull oneof the video descriptions that they share with other MTsusing their Bluetooth interface. This collaborative schemeproved to be more energy efficient than pulling all thevideo over the WLAN interface. A more comprehensivestudy is conducted in COMBINE [26] where experimentalresults are presented for a test bed composed of a GPRSLR interface and a WLAN SR interface.The previous references do not present the energyminimization solution in closed form, neither do theyinclude fairness consideration as part of the problemformulation or solution approach. In this paper, we presentthe optimal solution for energy minimization in contentdistribution with M2M collaboration in a single cluster of cooperating MTs. The problem is formulated in a generalsetup with different possible wireless technologies on theLR and SR. Scenarios with multicasting and unicastingon the SR links are studied. The optimal solution isshown to be unfair, since it consists of sending all thedata to a single MT on the LR. In order to add fairnessto the energy minimization problem, we present a lowcomplexity algorithm that performs utility minimization.With an appropriate choice of the utility, the algorithm canachieve the unfair greedy energy minimization solution.However, the main purpose of the algorithm is to beused in conjunction with utilities leading to fairness inenergy consumption. The algorithm can be used withutilities ensuring min-max fairness or corresponding toa game theoretical formulation where MTs are assumedto play a bargaining game to reach the Nash bargainingsolution (NBS).This paper is organized as follows. The system model ispresented in Section 2. The network energy minimizationformulation and optimal solution are presented in Copyright c ￿ 2007 AEIT Prepared using  ettauth.cls   Euro. Trans. Telecomms.  00 : 1–16 (2007) DOI: 10.1002/ett Page 2 of 16http://mc.manuscriptcentral.com/ettEuropean Transactions on Telecommunications 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960  F    o  r    P   e  e  r    R   e  v   i    e  w    ENER. AWARE COOP. CONTENT DIST. 3 Figure 1. General system model. Section 3. In Section 4, utility minimization is formulated and a low complexity utility minimization algorithm ispresented. Different utilities that can be used with theproposed algorithm are discussed in Section 5. Severalsimulation scenarios are studied and analyzed in Section 6.Practical considerations concerning MT grouping intocooperating clusters in addition to feedback overhead arediscussed in Section 7. Future extensions of this work arepresented in Section 8. Finally, conclusions are drawn inSection 9. 2. System Model The system model adopted in this work is depictedin Figure 1. The design consists of a number  K   of cooperating MTs in the range of a BS. The BS is connectedvia wired LAN to the server that holds the content.Terminals can communicate with each other over SR links.Unicasting is considered on the LR. This allows the BSto transmit at the rate supported by each MT insteadof forcing all MTs to receive at the rate of the MThaving the worst LR channel gain, as is the case in LRmulticasting. On the SR, both unicasting and multicastingare investigated in this paper.In a traditional setup, the server either separately streamsthe complete content to each requesting MT or multicaststhe content once to all requesting MTs. In both cases, thecommunication interface of each MT remains active for thewhole reception duration, which depends on the length of the content and the transmission rate. This results in highenergy consumption due to the required processing duringdata reception.In this work, we assume the establishment of an M2Mnetwork between the MTs over SR wireless links that aremore energy-efficient than the LR wireless link. In thisscheme, the content is divided into  N   parts. If   K   MTsare requesting the content, then, each will be receivinga subset of the  N   data parts from the server. Over theSR wireless links, each MT receives the remaining datasubsets from the other cooperating MTs in the M2Mnetwork. Being exchanged over an energy-efficient SRwireless technology, the SR exchanged subsets requirelower reception power at the communication interface of the MTs. However, an additional overhead in this case isthat each MT needs to spend additional energy to transmitits received data subset to the other cooperating MTs.In this work, we consider a single cluster of connectedMTs where each MT can communicate with every otherMT in the cluster. Furthermore, a low mobility scenariois adopted. Thus, it can be assumed that the channelconditions on the MT-MT links and BS-MT links remainapproximately constant during the distribution of a singlefile. Furthermore, we assume that all collaborating MTsexchanging a certain file remain available during theexchange, i.e., an MT does not leave the cluster while itis in the midst of a collaborative content exchange process.  2.1. Channel Model  The channels on the LR and SR links are assumed to beorthogonal and are modeled by pathloss, shadowing andfading. Thus, the received power  P  r  can be linked to thetransmitted power  P  t  by a pathloss model as in [27]: P  r P  t (dB) = 10log 10 κ − 10 ν   · log 10 d  ￿ 󿿿􏿿 󟿿  distance based pathloss +  h dB  + f  dB ( a )  ￿     󿿿􏿿    󟿿  random variables (1)where  κ  is a unitless constant which depends onthe antenna characteristics and the average channelattenuation,  ν   is the path loss exponent,  d  is the distancewhere the received power is calculated,  h  is a Gaussianrandom variable representing shadowing or slow fadinghaving a zero mean and a variance σ 2 h dB , and f   is a randomvariable representing Rayleigh fading with a Rayleighparameter  a .  2.2. Data Rates Given for each MT: the transmit power  P  t  the sender istransmitting with, the pathloss, shadowing, and fading onthe channel, and the thermal noise power  σ 2 , the receivedsignal-to-noise ratio (SNR)  γ   can be calculated following γ   =  P  r σ 2 . Given the target bit error rate  P  e  and the SNR, Copyright c ￿ 2007 AEIT Prepared using  ettauth.cls   Euro. Trans. Telecomms.  00 : 1–16 (2007) DOI: 10.1002/ett Page 3 of 16http://mc.manuscriptcentral.com/ettEuropean Transactions on Telecommunications 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960  F    o  r    P   e  e  r    R   e  v   i    e  w    4 YAACOUB ET AL. the bit rates on the LR and SR links can be calculatedaccording to the following: R  =  B  · log 2 (1 + βγ  )  (2)In (2),  B  is the passband bandwidth of the channel, and  β  is called the SNR gap. It indicates the difference betweenthe SNR needed to achieve a certain data transmission ratefor a practical M-QAM system and the theoretical Shannonlimit [27, 28]. It is given by:  β   =  − 1 . 5ln(5 P  e ) .  2.3. Parameters and Variables The parameters that affect the energy consumption in thechosen system model are the following: •  K  : the number of requesting MTs. •  S  T  : the size of the content to be sent in onetransmission interval. This depends on the contentserver transmission rate. •  N  : the number of parts the content is divided into.Thus, the size of one part is  S  T  /N  . •  R L ,k : transmission rate on the LR links from the BSto MT  k . •  R S ,kj : transmission rate on the SR links from MT  k to MT  j . •  P  L , Rx : power consumed by the MT during receptionon the LR link. •  P  S , Rx : power consumed by the MT during receptionon the SR links. •  P  S , Tx ,kj : power consumed by MT  k  whiletransmitting to MT  j  on the SR links. •  Decision variables: the decision variables are  x k ,with  x k  an integer variable that determines thenumber of parts received by MT  k  over the LR link.It should be noted that  P  L , Rx ,  P  S , Rx , and  P  S , Tx ,kj correspond to the power consumed by the MT, i.e., energydrained per second from its battery, during reception andtransmission, respectively. They are not to be confusedwith  P  r  and  P  t , the respective receive and transmit powersover the air, measured at the antenna. It should be notedthat  P  S , Tx ,kj  can be expressed as: P  S , Tx ,kj  =  P  S , Tx , 0  + P  t,kj  (3)where  P  S , Tx , 0  corresponds to the power consumed by thecircuitry of the MTs during transmission on the SR links,and  P  t,kj  corresponds to the power transmitted over the airon the SR links from MT  k  to MT  j .Power consumption of the BS is not considered since theinterest in this paper is in the battery life of the MTs. Thiscan be justified by the fact that most BSs rely on power linecables and not on batteries and thus do not haveas stringentpower limitations as the MTs. 3. Optimal Energy Minimization In this section, the energy minimization problem isformulated and the optimal solution is presented.Unicasting and multicasting are considered on the SR, bothwith rate adaptive and power adaptive transmissions.  3.1. Energy Minimization with Unicasting Considering  K   requesting MTs interested in downloadinga content from a server on the internet in a cooperativemanner, and assuming that the content is divided into  N  parts of equal size and importance, the time  t k  required tosend  x k  parts over a link with rate  R k  is: t k  =  x k  · S  T  N   · R k (4)Multiplying the power drained from the MT battery bythe the time needed, then the expression of the energyconsumed can be obtained. Consequently, the energyconsumed by MT  k  is: E  k  =  x k  · S  T  N   · R L ,k P  L , Rx  +  x k  · S  T  N  K   j =1 ,j ￿ = k P  S , Tx ,kj R S ,kj +  S  T  N  P  S , Rx K   j =1 ,j ￿ = k x j R S ,jk (5)In (5), the first term corresponds to the energy consumedby MT  k  for receiving  x k  parts over the LR, the secondterm corresponds to sending the data received by MT  k on the LR to the other MTs on the SR, and the lastterm corresponds to the energy consumed by MT  k  whilereceiving the data parts from the other MTs on the SR, i.e.,receiving  x j  data parts from each MT  j  on the SR.The total energy consumed by the requesting MTs is: E  coop  = K   k =1 E  k  (6) Copyright c ￿ 2007 AEIT Prepared using  ettauth.cls   Euro. Trans. Telecomms.  00 : 1–16 (2007) DOI: 10.1002/ett Page 4 of 16http://mc.manuscriptcentral.com/ettEuropean Transactions on Telecommunications 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Search
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks