蜂窝小区频率有效性的基站三模式休眠机制设计

蜂窝小区频率有效性的基站三模式休眠机制设计 //.paper.edu - 1 - 中国科技 论文 政研论文下载论文大学下载论文大学下载关于长拳的论文浙大论文封面下载 在线 Triple-mode Sleeping Mechanism Design for Energy Efficiency in Cellular Network# Xu Feng 1 , Kai Niu 1 , Ping Gong 1 , Baoyu Tian 1 , Shaohui Sun 2** 5 (1. Information Theory and Technology Center, BUPT; 2. State Key Laboratory of Wireless Mobile Communications, CATT) Foundations: This work was supported by National Science and Technology Major Project of China (2012ZX03004005-002,2012ZX03003003-005); National Natural Science Foundation of China (No. 61171099); State Key Laboratory of Wireless Mobile Communications, CATT. Abstract: The cellular network energy needs urgently to be saved because the carbon emissions continue to increase. Therefore, many kinds of energy saving methods in bases are widely researched and designed. In this study, we propose a different kind of method to realize the energy saving in base 10 stations. Three base station modes are defined: Normal Work Mode, Light Sleep Mode and Deep Sleep Mode. By conditionally switching among these modes according to the traffic load, base stations will fall asleep when it’s necessary to save energy. State set is brought in to analyze the process of state transition. And we propose a novel way to evaluate the traffic load of the network. Moreover, a general method of optimizing all bases’ modes is put forward and the energy-efficiency is considered in the 15 optimizing process. The proposed sleep and wake-up mode have been validated by some simulations and experimental results under the LTE scenarios. Key words: ktraffic load; state set ;sleep mode; energy efficiency 20 0 Introduction Beyond all question, our human has reached at the epoch of carefully calculating and strictly budgeting on account of constrained resource. Traditional energy is gradually exhausting while new energy source isn't dependable enough. At the meantime, the energy consumption on telecommunication has an upward trend. More and more networks and base stations are built to 25 meet higher demand on QoS of wireless communication, intelligent mobiles popularization, and the revenues of telecom service providers. Statistics indicate that cellular network infrastructures around the world consume a total energy of 60 TWh per year in recent years. The radio access networks consumed 80% of the total energy[1]. And the number of this is steadily on the increase. A typical universal mobile 30 telecommunication system cell site, which consists of power amplifier, air conditioner, digital signal processor and feeder from radio network controller to e-NodeB, consumes an average of 6kW power. Moreover, the e-NodeB needs power which is between 500W and 3KW, while transmitting power of carrier of BTSs varies from 1 to 60W[2] [3]. It can be seen that the base stations consume the largest portion of energy in the telecommunication progress. So it gains a 35 significant attention on how to increase the energy saving efficiency of the base stations. For energy-efficient design, the most straight and efficient method is decreasing the hardware expense. Currently, energy reduction strategies are defined in wireless communication standards and supported by user terminals. For example, the LTE system adopts discontinuous transmission and reception techniques[4]. With regard to mobile terminals, IEEE 802.16e proposed several kinds 40 of sleep mode mechanisms[5]. Many kinds of approaches, which are about sleep mode design in radio base station, have contributed towards energy saving. In [6], switching off individual base station at traffic load is discussed. Some cells are turned off to accomplish the energy consumption, while radio coverage and users services are taken care of by the neighbor active cells. More exquisitely, it's allowed to switch off a number of frequency subcarriers rather than a whole base 45 //.paper.edu - 2 - 中国科技论文在线 station during off-peak hours[7]. This makes base stations more flexible and bring better QoS. With MIMO systems widely deployed, spatial-domain sleep modes have also been investigated. As extra energy for transceiver circuits and control signaling overhead are required in MIMO systems, switching off individual antenna ports can also reduce considerable energy consumption[8]. And in [9], a mechanism is put forward to make mobile terminals switch between MIMO and SIMO to 50 conserve energy. We design a more smooth base station sleep mechanism on basis of these studies. So bases will sleep in different levels according to the traffic load change. We divide the base stations status into three modes--Normal Work Mode, Light Sleep Mode, and Deep Sleep Mode. In normal traffic load, the base stations are all in Normal Work Mode, then 55 we switch off the control channel of some base stations. We call this state which base stations turn into Light Sleep Mode. The RF coverage of the base stations whose control channels have been turned off is taken care of by their neighbor base stations. When the load varies from normal to low, these base stations which in their Light Sleep Mode are sequentially switched to Deep Sleep Mode. In the period of switching from Light Sleep Mode to Deep Sleep Mode, the base stations 60 gradually deactivate their communication systems, air conditions and other power consumption services. While a base station in Deep Sleep Mode, all the services of it are delivered to the neighbor base stations. So it needs quite a little energy just for its restart. When the load varies from low to high, the base stations wake up. In this way, we could decrease the delay and suddenly severe power consumption. 65 The rest of paper is organized as follows: Section 1 provides a system mode in which a base station could accomplish its sleep process. Section 2 provides the consideration of energy efficiency, followed by the simulation results to prove the feasibility of our study in Section 3. Section 4 summarizes this paper and draw our conclusions. 1 System Description 70 1.1 Base stations's switch among triple modes The traffic load always floats in a cellular network. In some time like 8:00 a.m. in the morning or 12:30 p.m. at noon, the whole network is always in heavy load. While in the midnight, mobile phones are almost turned off. The load is not only changing over the time but also over the location. Workplaces need better telecommunication QoS in the daytime while residential suburbs 75 need that in the night. Thus it can be seen that turning off the base stations or their control channels in certain time and place will save quite a lot of energy. In a novel way, we divide the base stations status into three modes--Normal Work Mode (NWM), Light Sleep Mode (LSM), and Deep Sleep Mode (DSM). Fig.1 is the state transition diagram of base stations' triple modes. It not only contains these 80 three states of a base, but also includes three self-sustaining processes and six mutual transformation processes. These three states will keep in the self-sustaining processes when the traffic rarely vary. This conforms to the Markov property. When the traffic load appears obvious change, these three states will transform into each other. We assume that in the case of all transformation satisfying their corresponding threshold, the six mutual transformation processes 85 have different priorities: the transformation from DSM to NWM and the one from LSM to NWM have the highest priorities, the one from DSM to LSM has the second highest priority, and the rest ones from NWM to LSM, from NWM to DSM and from LSM to DSM have the lowest priorities. Suppose that the state of a certain base in time t is tS . When the traffic load near it varies in //.paper.edu - 3 - 中国科技论文在线 time t+1 , the state will transform in time t+1. The t+1S is determined only by two factors: the 90 state set(which we will explain in next subsection) of this base in time t and the cost function(which we will explain in next section) of all bases in time t . So we can see that the states of bases in time t+1 are just concerned with the states of bases and traffic load change in time t . This conforms to the Markov property, too. The states will stay steady in time t when they are determined. So, the process shown in Fig.1 is a Markov process. 95 Fig. 1 State transition of triple modes Fig.2 shows how its channels migrate in a base station’s triple mode. Base station ?? in Fig.2 is our consideration. When it is in the state of NWM, base station ?? provides services for the UE which is under the pilot coverage. If the traffic load of its area is relatively low, we will 100 turn off the control channel of base station ?? . It comes into its state of LSM. At this time, base station ?? still renders services while its control channel suspends to sleep. Its neighbor base station ?? will zoom out and establish the control links with the previous UE. Then if it is under low traffic load for quite a long time, base station ?? will be turned off totally. It comes into its state of DSM. Being totally turned off doesn’t mean closing the whole base station in traditional 105 sense. We will power down communication equipments, air conditions and conventional power, but hold a certain amount of power for the bases to restart and interactive information of traffic load to be communicated. Because the power of restarting the bases is much lower than the one of those high-power devices, base station ?? could save considerable energy when it is in the state of DSM. At this moment, all the services of base station will be delivered to its neighbor base station 110 ?? . //.paper.edu - 4 - 中国科技论文在线 Fig. 2 Channels migration of triple modes All the bases work together to make sure that they stay in their best states while energy is saved in transforming process. Bases are in different states in a certain period. And the states will 115 change in the next period. The system is always in Markov steady state. The probability of every state is concerned with traffic load of the network. It’s obvious that bases will stay at NWM more probably if the network traffic load is very heavy. 1.2 The change process of state set We use a state set to indicate a base’s current state and possible state change according to the 120 variation of traffic load near iB . This state set consists of three sets as shown in table 1. Tab. 1 STATE SET Current Set Reserve Set Taboo Set The current time t states (N short for NWM, L short for LSM, and D short for DSM) of iB are stored in current set. The possible state of iB in time 1t ?? are stored in reserve set. The states which iB will not belong to in time 1t ?? are stored in taboo set. The reserve set and the 125 taboo set of iB are determined by the traffic load near it in time t while the current set is determined by the state set of time 1t ?? and the optimization of cost function about all the bases in time 1t ?? . Suppose a scene that the current time t state of iB in on mode of LSM and the traffic load nearby becomes heavier, but it’s not so heavy that iB of time 1t ?? must be in NWM. We stored ”L” in current set, ”L” and ”N” in reserve set and ”D” in taboo set. So in time 130 1t ?? , we should guarantee that the current set of Bi in time 1t ?? isn’t the taboo set of iB in time t . Actually, not a single base iB is stored in the state set, but all the bases (assume the number is N, that is i = 1…N ) in topology are stored in the set. 1.3 The estimate on traffic load The traffic load that we consider has a great deal of differences with the definition of 135 traditional way. For example, in [10] and [11], traffic load is produced from fitting bandwidth occupancy rate, signal-to-interference and noise ratio (SINR), and cell capacity. It indicates the congestion of some certain area. Actually, this definition aims at finding the potential performance in an evaluated area. That manifests that we could calculate the number of users which a base station is braced for by assessing the traffic load. 140 //.paper.edu - 5 - 中国科技论文在线 But in the process of base stations sleeping, the traffic load in a certain area indicates how many services are provided by the central base station and how much congestion it brings. That means even if a base station is over its load, we will still turn off the station because of its low throughput and weak user bearing capacity which would not happen if we use the traditional definition of traffic load. In fact, over load base stations will cost much more energy according to 145 the nonlinear characteristics of electrical equipments[12]. It’s appropriate to turn these stations off out of consideration of energy saving. This example also explains the importance of redefining the traffic load from another side. In recognition of this fact, a general formula is given to assess the traffic load as follows: max(0, ) (1 ) i i i ii i total base total L D S H WD H N W ?? ?? ???? ?? ?? ?? ?? ?? ?? ?? 150 Here, iL represents the traffic load of the thi base station, ?? is the factor of throughput which represents how important the throughput is in measuring the performance of a network, iH is the throughput of the thi base station accumulates for the whole network, totalH is the total throughput of current network, iD is the average delay caused by turning off the thi base station, iS is the traffic services sensitivity of the users under the thi base station, max(0, )x 155 means that the result is the larger one between 0 and x, baseN is the number of base stations which might be sleeping, iW is the bandwidth which the thi base station uses, totalW is the system total bandwidth, ?? is a factor which represents the proportion of every base station in the network, and it can be obtained from the following equation: 1 base total base N i i N W W ?? ?? ?? ?? ?? 160 It is obvious that the range of the traffic load defined here is between 0 and 1. The traffic load we defined here represents how much important a role the thi base plays in the system rather than the degree of congestion near the thi base. So it’s proper to express as a percentage. The load is the sum of two parts. The first part is the throughput provided by the observed base for the whole system. The second part is sum delay of users of observed base if the base turns into sleep. 165 ?? is used to control the proportion of these two parts. With regard to these services of high latency requirement like VOIP, we’ll choose a big ?? ; while with regard to these service of low latency requirement like email service, we’ll choose a small one. 2 Energy Efficiency 2.1 The process of optimizing sleep states 170 As analyzed in previous section, the sleep state changes of every base station are Markov processes. And all the states in a certain time constitute the recursive solution of the optimizing problem. Because the constraint condition to this problem is non-convexity, the optimizing problem is non-convexity, too. The base stations are interacted in a certain time, so we have to consider all the states to accomplish the bases’ adaptive process. 175 For example: all bases states in time t are indicated in Table 2 and the traffic load in entire topology is changing. Tab. 2 ALL STATE SETS IN TIME T //.paper.edu - 6 - 中国科技论文在线 Current Set: B1=N,B2=L,B3=D,...,BN=N Reserve Set: B1=N and L,B2=N and L,...,BN=N Taboo Set: B1=D,B2=D,B3=D and L,...,BN=L and D So we adjust all the states by making decisions as follows: 1) Calculate the coverage and make sure that the coverage doesn’t change when we adjusting 180 the bases’ states; 2) Give priority to adjusting these stations whose current set and taboo set are in conflict. It guarantees that these states of stations which have to vary change first; 3) Change the other bases to save more energy or provide better QoS; 4) Calculate the cost function to achieve the maximum by repeat step2 and step3. 185 After this optimizing process, we may get all new bases states in time 1t ?? as indicated in Table 3. And this is just a sketch. We can see that there is no intersection between the taboo set in time t and the current set in time 1t ?? . Tab. 3 ALL STATE SETS IN TIME T+1 Current Set: B1=L,B2=L,B3=N,...,BN=N Reserve Set: B1=N and L,B2=N and L,...,BN=N Taboo Set: B1=D,B2=D,B3=D and L,...,BN=L and D We can always get good states of base stations to save energy and provide good QoS by 190 repeating the optimizing process. 2.2 The estimate of QoS It has better flexibility to only turn off the control channel first compared with conventional way of simply directly turning off a whole base station. It can help save some energy in heavy traffic load. Moreover, if the QoS of network becomes very poor abruptly after the several base 195 stations being straightly shut off, it will bring much delay and power overhead. But turning off the control channel will provide a buffer space for further shutting off because it brings lower latency. So this kind of multi-lever sleeping mechanism doesn’t change the network traffic services much, and our energy saving process is quite gentle. In order to estimate QoS, we use the formula below: 200 min max min1 1 ( ) (1 ) (1 ) min( , )M Mi i i i Q t D R R D R ?? ?? ?? ?? ?? ?? ?? ?? ?? ???? ?? Here, ?? is the weight coefficient of delay, M is the number of users in a cell, iD is the delay of thi user’s traffic, maxD is the allowable maximum delay, iR is current traffic transmission rate of thi user, minR is minimum traffic transmission rate. The QoS here considers two aspects, one is the delay and the other is the transmission rate. Through considering 205 delay we could estimate the users’ feeling about mobile service. Different kinds of traffic service have various required delay. We could estimate the performance supplied by the system through monitoring the transmission rate. 2.3 The trade-off between energy saving and QoS We should not sacrifice the QoS of users when saving the energy of many bases. In our 210 system mode, bases save their energy through two kinds of sleeping in different traffic load. That means if the traffic load of the observation area is always heavy, all the bases should be in NWM so as to guarantee the QoS of this area. While if the traffic load is always light, some bases would be in some kind of sleeping to save energy. We will also improve the traffic quality in the service area by reducing the bases’ mutual interference. Our consideration is as follows: 215 //.paper.edu - 7 - 中国科技论文在线 1 2w P(t) + w Q(t)C ?? 1 ( ) (( ( , ) ( , ) ( , )) ( ( , ) ( , ) ( , ))) N D L D N L N N L N D L D i P t P i t P i t P i t P i t P i t P i t?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ???? Here, C is the value of cost function, 1w and 2w are the weighting coefficients. Q is the QoS, P is the power which could be saved, t represents time, N is the number of base stations, ( , )D LP i t?? ?? is the power saved of thi base in time t from its DSM to LSM, 220 ( , )D NP i t?? ?? , ( , )L NP i t?? ?? , ( , )N LP i t?? ?? , ( , )N DP i t?? ?? and ( , )L DP i t?? ?? are also power saved among mutual states transforming. By calculating the cost function, we could both save the energy and take care of the QoS of users. And we could also adjust the 1w and 2w to make the green system more flexible. 3 Simulation Results 225 In this paper, extensive simulations have been carried out for evaluating the proposed mechanism. We adopt the model of heterogeneous network deployments in 3GPP 36.814[13]. The simulation parameters are as shown in Table 4. Tab. 4 SIMULATION PARAMETERS Parameters The width of system Topology Distance-dependent path loss Lognormal shadowing values 10MHz regular hexagon, 19 cells, 6 femto cells per cell L = 127 + 30log10;R for 2GHz,R in km, the number of floors in the pathis assumed to be 0 2 ( ) cor x in dR x e ?? ?? ?? ?? ?? Parameters Shadowing standard deviation Antenna pattern Total BS TX power UE power class values 108><#004699'>dB Omnidirectional 46 dBm 23dBm (200mW) Parameters Shadowing standard deviation Antenna pattern Total BS TX power UE power class values 10<#004699'>dB Omnidirectional 46 dBm 23dBm (200mW) Fig.3 shows the average amount of power change in the different process. It is calculated by 230 the statics of all the bases. We can see that it’s the best to let bases sleep in DSM. And it had better not to transform the bases which are changing frequently. Fig. 3 The process of base sleeping //.paper.edu - 8 - 中国科技论文在线 Fig.4 shows the relationship between energy saving and the system traffic load. And the 235 bases and users we simulate are uniform relatively. We can see that the more the system have traffic surplus, the more energy we could save by switching off base stations. It saves a considerable amount of energy when the system is in low load. Fig. 4 Energy saving varying with different load surplus 240 According to our statics, we can switch 10 base stations into LSM in normal traffic load on our simulation scene. That means we can save 8%-12% energy of bases totally consuming just by turning off base stations’ control channels. When the system is in low traffic load, we can switch 4-5 base stations into DSM. That means we can save 20%-30% energy of bases totally consuming by switching base stations among triple modes. 245 Fig.5 shows the pilot frequency CIR of the system when some base stations are in LSM and DSM. It can be seen that the system coverage changes a little while QoS doesn’t almost change in our turning-off process. Fig. 5 System pilot coverage 250 From these above simulation results, we can save energy and guarantee users’ QoS through bases multi-level sleeping. And the mechanism could be used in more complex situation. //.paper.edu - 9 - 中国科技论文在线 4 Conclusion In this paper, we proposed triple modes of base stations-NWM, LSM and DSM, and energy is saved by switching among these modes. The system mode is analyzed as Markov process. We 255 also provide new definition about the traffic load of cellular network and our consideration is given in optimizing process. The simulation shows that our mechanism has significant effect. We will try to bring in the mechanism of Pico sleeping in the next step, and more irregular scene will be simulated on. References 260 [1] M. Hossain, K. Munasinghe, and A. Jamalipour. ;A protocooperationbased sleep-wake architecture for next generation green cellular access networks[j].in Signal Processing and Communication Systems (ICSPCS),2010 4th International Conference on. IEEE, 2010:1-8. [2] J. Louhi.Energy efficiency of modern cellular base stations[A], in Telecommunications Energy Conference, 2007. INTELEC 2007. 29th International. IEEE, 2007:475-476. 265 [3] L. Chiaraviglio, D. Ciullo, M. Meo, et al. Energy-efficient management of umts access networks[J].in Teletraffic Congress, 2009. ITC 21 2009. 21st International. IEEE, 2009:1-8. [4] T. Kolding, J. Wigard, and L. Dalsgaard.Balancing power saving and single user experience with discontinuous reception in lte[Z]. in Wireless Communication Systems. 2008. ISWCS'08. IEEE International Symposium on. IEEE, 2008:713-717. 270 [5] S. Vuyst, K. Turck, D. Fiems, et al.Delay versus energy consumption of the ieee 802.16 e sleep-mode mechanism[Z]. Wireless Communications, IEEE Transactions on, vol. 8, no. 11:5383-5387. [6] M.A. Marsan, L. Chiaraviglio, D. Ciullo, et al.Optimal energy savings in cellular access networks[D]. in Communications Workshops, 2009. ICC Workshops 2009. IEEE International Conference on. IEEE, 2009:1-5. [7] Z. Xu, C. Yang, G.Y. Li, et al.Energy-efficient mimo-ofdma systems based on switching off rf chains[J]. in 275 Vehicular Technology Conference (VTC Fall), 2011 IEEE:pp. 1-5. [8] V. Nguyen and P. Ookpact switched and reconfigurable 4-ports beam antenna array for mimo applications[J]. in Intelligent Radio for Future Personal Terminals (IMWS-IRFPT), 2011 IEEE MTT-S International Microwave Workshop Series on. IEEE, 2011: pp. 1-3. [9] H. Kim, C.B. Chae, G. De Veciana, et al.A cross-layer approach to energy efficiency for adaptive mimo 280 systems exploiting spare capacity[Z].Wireless Communications, IEEE Transactions on, vol. 8, no. 8:4264-4275. [10] L. Zhang, F. Liu, L. Huang, et al.Traffic load balance methods in the lte-advanced system with carrier aggregation[J]. in Communications, Circuits and Systems (ICCCAS), 2010 International Conference on.IEEE, 2010:63-67. [11] V. Capdevielle, A. Feki, and E. Temer.Enhanced resource sharing strategies for lte picocells with 285 heterogeneous traffic loads[J]. in Vehicular Technology Conference (VTC Spring), 2011 IEEE 73rd. IEEE, 2011: 1-5. [12] W. Patterson and M. Kuhn.An electronic system for measuring the electrical characteristics of nonlinear devices[Z]. Review of Scientific Instruments, vol. 40, no. 7, pp. 960-961, 1969. [13] E. Access.3GPP TR 36.814.Further advance-ments for e-utra physical layer aspects[S].Tech. Rep., 2010. 290 蜂窝小区频率有效性的基站三模式休眠机 制设计 冯旭1，牛凯1，龚萍1，田宝玉1，孙绍辉2 295 (1. 北京邮电大学信息理论与技术教研中心; 2. 大唐无线通信重点实验室) 摘要:由于碳排放的持续增加，减少蜂窝小区的能量消耗成为迫在眉睫的一件工作。本篇提 出了一种实现基站节能的新型方法,定义了三种基站的模式:正常工作模式，轻度睡眠模式 和重度睡眠模式。在不同的情形下，根据不同的地区负载情况，通过在三种模式之间互相切300 换以使基站休眠从而达到节省能力的目的。状态表也被引入以 分析 定性数据统计分析pdf销售业绩分析模板建筑结构震害分析销售进度分析表京东商城竞争战略分析 状态转移过程。同时还提 出一种更新颖的方法以评估一个网络的负载情况。不仅如此，我们还提出一种通用方法以优 化所有基站状态，而且在优化过程中奖能量的有效性作为我们衡量的重要因素之一。最后， 在 LTE 场景进行验证，效果良好。 关键词:负载;状态表;休眠模式;能量有效性 305 中图分类号:TN929.53