A Parameterized Base Station Power Model
aa r X i v : . [ c s . I T ] N ov A Parameterized Base Station Power Model
Hauke Holtkamp, Gunther AuerDOCOMO Euro-LabsD-80687 Munich, GermanyEmail: { holtkamp, auer } @docomolab-euro.comVito GianniniIMECLeuven, [email protected] HaasInstitute for Digital CommunicationsJoint Research Institute for Signal and Image ProcessingThe University of Edinburgh, EH9 3JL, Edinburgh, UKE-mail: [email protected] 7, 2014 Abstract
Power models are needed to assess the power consumption of cel-lular base stations (BSs) on an abstract level. Currently availablemodels are either too simplified to cover necessary aspects or overlycomplex. We provide a parameterized linear power model which coversthe individual aspects of a BS which are relevant for a power consump-tion analysis, especially the transmission bandwidth and the numberof radio chains. Details reflecting the underlying architecture are ab-stracted in favor of simplicity and applicability. We identify currentpower-saving techniques of cellular networks for which this model canbe used. Furthermore, the parameter set of typical commercial BSs isprovided and compared to the underlying complex model. The com-plex model is well approximated while only using a fraction of the inputparameters. Introduction
Recently, the power consumption of cellular networks has become a pointof interest in research and even been taken into consideration for the stan-dardization of future cellular networks like Long Term Evolution (LTE)-Advanced [1]. It was found in [2] that in cellular networks the element whichcauses the largest share of overall consumption is the base station (BS).Numerous techniques have consequently been proposed by which the powerconsumption of BSs can be reduced [3–8]. Some of these techniques onlyconsider transmission power while others take into consideration that thegeneration of the radio signal also consumes power in circuitry by employ-ing power models. Such power models describe abstractly how much powera transmitter consumes and how this consumption depends on operatingparameters. In the past, the modelling of BS power consumption often hadto be based on intuition until the first power models were published [9–13].Simple models like [9, 10] allow computing the power consumption of a BSfor specific configurations. In contrast, the non-linear complex model de-scribed by Desset et al. [13] is derived from the combination of each ofa BS’ subcomponents. This allows inspecting the power consumption tosuch detailed level as the effect of giga operations per second or transistorgate lengths, but is unwieldy to apply. In this paper, we extend the workin [11,12] by maintaining simplicity while integrating two relevant operatingvariables into the model, namely the power amplifier (PA)’s output rangeand the transmission bandwidth. The proposed model allows assessing thepower consumption of all techniques that are currently employed to reducethe power consumption of BS while conserving simplicity.The scope of the model is described in Section 2. The model is subse-quently presented in Section 3. In Section 4, it is discussed and comparedto the complex model. The paper is concluded in Section 5.
Power saving techniques in literature can be generally divided into designchanges and operating approaches. Design changes affect the layout of thenetwork or the hardware architecture. For example, in [14] it is proposedthat the use of heterogeneous networks will positively affect the networkpower consumption under certain conditions. Cui et al. [4] show how ad-justing the number of antennas affects power consumption. In contrast todesign changes, operating approaches manipulate the functionality of a BS2uring operation. Here, proposed techniques are the reduction of transmis-sion power [5], the deactivation of unneeded antennas [6], the adaptation ofthe transmission bandwidth [7] and the use of low power consumption sleepmodes of varied durations [8].The model presented in this paper encompasses all of these approachesto allow for a direct comparison while abstracting parameters which caneither be assumed to be constant or have been shown to have little effect inthe studied scenarios, such as modulation and coding settings, equipmentmanufacturing details and leakage powers [13]. To this extent, the followingis covered in the proposed model: • The different BS types of a heterogeneous network are modelled byapplying different parameter sets to the same model equations. • The number of transmission antennas and radio chains affects con-sumption during design and operation. • The same holds true for transmission power, which affects the designindirectly by choice of a suitable power amplifier as well as the opera-tion directly. • Also, transmission bandwidth and sleep modes are modelled in theireffect on BS power consumption.
It was found in [11] that the supply power consumption of a BS can be ap-proximated as an affine function of transmission power. In other words, theconsumption can be represented by a static (load-independent) share, P ,with an added load-dependent share that increases linearly by a power gra-dient, ∆ p . The maximum supply power consumption, P , is reached whentransmitting at maximum total transmission power, P max . Furthermore, aBS may enter a sleep mode with lowered consumption, P sleep , when it isnot transmitting. Fig. 1 shows an illustration. Total power consumptionconsidering the number of sectors, M sec , is then formulated as P supply ( χ ) = ( M sec ( P + ∆ p P max ( χ − < χ ≤ M sec P sleep if χ = 0 , (1)where P = P + ∆ p P max . The scaling parameter χ is the load share, where χ = 1 indicates a fully loaded system, e.g. transmitting at full power and full3igure 1: Load-dependent power model for an LTE BS.bandwidth, and χ = 0 indicates an idle system. To further understand thecontribution of different parameters on this basic model, we parameterizethe maximum supply power consumption, P . We first establish how powerconsumption scales with the transmission bandwidth in Hz, W , the numberof BS radio chains/antennas, D , and the maximum transmission power in W, P max . This requires to consider the main units of a BS: PA, radio frequency(RF) small-signal transceiver, baseband (BB) engine, direct-current (DC)-DC converter, active cooling and mains supply (MS). The dependence ofthe BS units on W , D and P max can be approximated as follows [13]: • Both the power consumptions of BB and RF, P BB and P RF , respec-tively, scale linearly with bandwidth, W , in and the number of BSantennas D . For some basic consumptions, P ′ BB and P ′ RF , we thusdefine P BB = D W
10 MHz P ′ BB (2)and P RF = D W
10 MHz P ′ RF . (3) • The PA power consumption P PA depends on the maximum transmis-sion power per antenna P max /D and the PA efficiency η PA . Also,possible feeder cable losses, σ feed , have to be accounted for: P PA = P max Dη PA (1 − σ feed ) . (4) • Losses incurred by DC-DC conversion, MS and active cooling scalelinearly with the power consumption of other components and maybe approximated by the loss factors σ DC , σ MS , and σ cool , respectively.These losses are included in the model as losses of a total accordingto [11]. Active cooling is typically only applied in Macro type BSs.4hese assumptions are combined to calculate the maximum power con-sumption of a BS sector, P = P BB + P RF + P PA (1 − σ DC )(1 − σ MS )(1 − σ cool ) (5a)= D W
10 MHz ( P ′ BB + P ′ RF ) + P max Dη PA (1 − σ feed ) (1 − σ DC )(1 − σ MS )(1 − σ cool ) . (5b)An important characteristic of a PA is that operation at lower transmitpowers reduces the efficiency of the PA and that, consequently, power con-sumption is not a linear function of the PA output power. This is resolvedby taking into account the ratio of maximum transmission power of the PAfrom the data sheet, P PA , limit to the maximum transmission power of the PAduring operation P max D . The current transmission power can be adjusted byadapting the DC supply voltage, which impacts the offset power of the PA.The efficiency is assumed to decrease by a factor of γ for each halving of thetransmission power. The efficiency is thus maximal when P max = P PA , limit insingle antenna transmission and was heuristically found to be well-describedby η PA = η PA , max (cid:20) − γ log (cid:18) P PA , limit P max /D (cid:19)(cid:21) , (6)where η PA , max is the maximum PA efficiency.The reduction of power consumption during sleep modes is achieved bypowering off PAs and reduced computations necessary in the BB engine.For simplicity, we only model the dependence on D as each PA is poweredoff. Thus, P sleep , is approximated as P sleep = DP sleep , , (7)where P sleep , is a reference value for the single antenna BS chosen such that P sleep matches the complex model value for two antennas. The parameterized power model is applied to approximate the consumptionof the Macro, Pico and Femto BSs which are described in the complexmodel. Parameters are chosen where possible according to [11], such aslosses, efficiencies and power limits. The remaining parameters are adaptedsuch that a closer match to the complex model could be achieved. Theresulting parameter breakdown is provided in Table 1. The proposed and the5S type P PA , limit η PA , max γ P ′ BB P ′ RF σ feed σ DC σ cool σ MS M Sec P max P /W /W /W /W /WMacro 80.00 0.36 0.15 29.4 12.9 0.5 0.075 0.1 0.09 3 40.00 460.4Pico 0.25 0.08 0.20 4.0 1.2 0.0 0.09 0.0 0.11 1 0.25 17.4Femto 0.10 0.05 0.10 2.5 0.6 0.0 0.09 0.0 0.11 1 0.10 12.0Table 1: Parameter breakdown.complex power models are compared for a bandwidth sweep with a varyingnumber of transmit antennas in Fig. 2, Fig. 3, and Fig. 4, for a Macro, a Picoand a Femto station, respectively. Although two parameters, the bandwidthand the number of BS antennas, are varied, the parameterized model canbe seen to closely approximate the complex model for all BS types. Thelargest deviation of the parameterized model from the complex model occurswhen modeling four transmit antennas. This is caused by the fact that theparameterized model considers a constant slope, ∆ p , which is independentof D . In contrast, the PA efficiency in the complex model decreases withrising D , leading to an increasing slope which can not be matched by aconstant slope. This deviation is a trade-off between simplicity and modelaccuracy.In addition to providing a solid reference, the model and the parameterscan provide a basis for exploration. Individual parameters can be changedto observe the resulting variation in power consumption. With regard to thenumber of antennas, the parameterized model can only be verified up to fourantennas, which is the extent of the complex model. Extending the systembandwidth, for example to 20 MHz, is expected to increase the BB and RFpower consumption. The other parameters such as the transmission powerand losses are expected to remain unaffected by different system bandwidths.Adapting the design maximum transmission power, P max , affects the PAefficiencies, which decrease with P max . In this paper, we have provided a parameterized power model which allowscalculating the power consumption of a modern BS based on importantdesign and operation parameters. The model is much simpler and moreapplicable than the model it was derived from. A comparison of the param-eterized model with the source model is provided.6
Used bandwidth [MHz] S upp l y po w e r c on s u m p t i on [ W ] Figure 2: Comparison of the parameterized with the complex model [13]power models for the Macro BS type with 40 W transmission power.
Used bandwidth [MHz] S upp l y po w e r c on s u m p t i on [ W ] Figure 3: Comparison of the parameterized with the complex model [13]power models for the Pico BS type with 0.25 W transmission power.
Used bandwidth [MHz] S upp l y po w e r c on s u m p t i on [ W ] Figure 4: Comparison of the parameterized with the complex model [13]power models for the Femto BS type with 0.1 W transmission power.7 eferences [1] 3GPP TR 36.927, “Evolved Universal Terrestrial Radio Access (E-UTRA); Potential solutions for energy saving for E-UTRAN,” Sep.2012.[2] A. Fehske, J. Malmodin, G. Bicz´ok, and G. Fettweis, “The Global Car-bon Footprint of Mobile Communications - The Ecological and Eco-nomic Perspective,”
IEEE Communications Magazine , 2010.[3] H. Claussen, L. Ho, and F. Pivit, “Effects of Joint Macrocell and Res-idential Picocell Deployment on the Network Energy Efficiency,” in
PIMRC 2008 , 2008, pp. 1–6.[4] S. Cui, A. J. Goldsmith, and A. Bahai, “Energy-Efficiency of MIMOand Cooperative MIMO Techniques in Sensor Networks,”
IEEE Journalon Selected Areas in Communications , 2004.[5] H. Holtkamp, G. Auer, and H. Haas, “On Minimizing Base StationPower Consumption,” in
Proceedings of the IEEE VTC 2011-Fall , 2011.[6] Z. Xu, C. Yang, G. Y. Li, S. Zhang, Y. Chen, and S. Xu, “Energy-Efficient MIMO-OFDMA Systems based on Switching Off RF Chains,”in
Proceedings of the VTC Fall 2011 , 2011.[7] A. Ambrosy, M. Wilhelm, O. Blume, and W. Wajda, “Dynamic Band-width Management for Energy Savings in Wireless Base Stations,” in
Proceedings of Globecom 2012 , 2012.[8] P. Frenger, P. Moberg, J. Malmodin, Y. Jading, and I. G´odor, “Reduc-ing Energy Consumption in LTE with Cell DTX,” in
Proceedings of theIEEE VTC 2011-Spring , 2011.[9] M. Deruyck, W. Joseph, and L. Martens, “Power Consumption Modelfor Macrocell and Microcell Base Stations,”
Transactions on EmergingTelecommunications Technologies , pp. n/a–n/a, 2012.[10] O. Arnold, F. Richter, G. Fettweis, and O. Blume, “Power ConsumptionModeling of Different Base Station Types in Heterogeneous CellularNetworks,” in
Future Network and Mobile Summit , 2010.[11] Gunther Auer et al. , “How Much Energy is Needed to Run a WirelessNetwork?”
IEEE Wireless Communications , vol. 18, no. 5, pp. 40 –49,Oct 2011. 812] G. Auer, V. Giannini, I. G´odor, P. Skillermark, M. Olsson, M. Imran,D. Sabella, M. J. Gonzalez, and C. Desset, “Cellular Energy EfficiencyEvaluation Framework,” in
Proceedings of the VTC 2011-Spring , 2011.[13] Claude Desset et al. , “Flexible Power Modeling of LTE Base Stations,”in
IEEE WCNC 2012 , 2012.[14] A. Fehske, F. Richter, and G. Fettweis, “Energy efficiency improve-ments through micro sites in cellular mobile radio networks,” in