Micro Base Stations in Load Constrained Cellular Mobile Radio Networks

Transcription

1 Micro Base Stations in Load Constrained Cellular Mobile Radio Networks Fred Richter,GerhardFettweis, Markus Gruber, and Oliver Blume Vodafone Stiftungslehrstuhl, Technische Universität Dresden, Germany {fred.richter, Alcatel-Lucent Bell Labs, Germany {markus.gruber, Abstract Future cellular mobile radio networks will exhibit a much more dense base station deployment than 2nd or 3rd generation communications systems, particularly with regard to traffic coverage. Hence, a significant increase in power consumption of cellular networks can be expected. In order to counter this trend, energy efficiency of such networks should be increased considerably. Concerning energy efficiency, utilizing micro base stations with their smaller power consumption capabilities appear promising. In this paper we study various homogeneous and heterogeneous deployment strategies incorporating micro base stations with focus on energy efficiency represented by power consumption and throughput. Further, we deal with the impact of different load scenarios on energy efficiency of the various network topologies in more detail. I. INTRODUCTION The global information and communication technology (ICT) industry was responsible for about 2% of the human greenhouse gas emissions world wide already in 27, increasing from 53 Megatons of CO 2 in 22 up to 83 Megatons in 27 [1], [2], which almost equals to those of global aviation [3]. Only recently, the European Union urged the ICT industry to go about energy efficiency of communications networks and of ICT in general [4]. Within the ICT, communications networks contributed only 12% of carbon dioxide in 22, but it is expected to increase by a factor of about 3 by 22. In the mobile radio networks the base stations are the most important contributor to CO 2 emissions with about 8% [5], [6], [7]. In comparison, the energy consumption and, thus, greenhouse gas emission, of both mobile devices and servers is considerably smaller by a factor of about 4 or 5 [8]. Thus, energy efficiency is receiving increasing attention in research on mobile access communications technologies [5], [9]. Also, manufacturer of mobile network equipment have already achieved a recognizable progress in energy efficiency, where most efforts are in more power efficient transceivers in the base stations [1] and alternative cooling techniques for replacing conventional air conditioning of base station housings. Up to now there are more than 4 billion mobile phone subscribers in total, being responsible for the increased energy consumption of the networks [11]. Where mobile devices are designed for energy efficiency due to stringent constraints 1 This work was supported in part by European Community s Seventh Framework Programme (FP7/27-213) under grant agreement n regarding power supply, the base stations as major contributor to greenhouse gas emission in the network are ignored for the most part until recently. This yields potential energy saving capabilities for today s mobile networks. Obviously, optimizing base stations on component level, e.g., utilizing more efficient power amplifiers and load adaptive hard- and software modules, would result in a most significant share of reduced energy consumption. A further source of energy wastage is due to the layout of today s cellular mobile radio networks. It is clear that networks optimized for coverage and capacity need not be energy efficient. Beyond the state-of-theart macro cellular deployment paradigm, there is a large field of most promising deployment strategies such as repeaters and relaying solutions as well as heterogeneous deployments, where the latter includes a sophisticated mix of different cell sizes. For planning and optimizing cellular networks the classical objective is the system s spectral efficiency, implying the maximization of data rates for given transmit power budgets. So far, conventional energy efficiency metrics capture only a small fraction of the whole power budget of the network. In [11] there is a summary of characteristics which should be taken into account in any energy efficiency metric. Amongst others, the energy spent for signal processing, static consumers (e.g., cooling), and the power conversion should be considered. In [12] a power consumption model for typical macro and micro base stations is developed which considers the two mentioned characteristics. This power model will be applied for evaluating the total power consumption of different cellular networks in this paper. An overview of some promising approaches that can lead to energy efficient wireless access networks can be found in, e.g., [6], [7]. In [13] a framework is derived for evaluating a network s efficiency w.r.t. area power consumption and area spectral efficiency and extended by considering the load. In [14] a network densification utilizing micro base stations is studied, considering total power consumption and mean throughput in the system. This approach is studied in more detail in this paper, focusing on different load conditions. With future networks providing high data rates in mind, the energy efficiency of pure micro base station networks is evaluated and compared with the energy saving capabilities of homogeneous macro and heterogeneous cellular networks.

2 3 Micro sites per macro cell 6 Micro sites per macro cell 9 Micro sites per macro cell Fig. 1: Location of micro sites in a macro cell. The remainder of the paper is organized as follows. In Section II we introduce the system model and define relevant performance measures. In Section III we study the performance of different deployment strategies. Section IV concludes the paper. In the following we use the notations P and E to denote the probability and expectation operator, respectively. II. SYSTEM MODEL AND PERFORMANCE METRICS In this paper the homogeneous macro system is modeled as a cloverleaf layout network due to a three-fold sectorization. The base station density is characterized by the inter site distance D R ++. In homogeneous micro networks, where the micro base stations are equipped with a single omnidirectional antenna, are also located on a hexagonal grid, whereas no sectorization is assumed and comparably smaller inter site distances can be expected. Heterogeneous topologies are based on a homogeneous macro network with a certain number of micro base stations placed uniformly on a circle of sufficiently large radius around each macro base station as depicted in Fig.1. Here, we refer to the notion cell as the area served by one site, i.e., the conjunction of all its sectors, for homogeneous networks. A site is understood as the geographic location of a base station s radio equipment and its antennas. In heterogeneous networks a cell is the area served by both the macro base station and all its micro base stations. A. Propagation Model and Coverage The signal deterioration from transmitter to receiver due to path loss is typically modeled by the relation P rx = Kd λ ΨP tx = P pl Ψ (1) with P pl := Kd λ P tx where P tx, P rx, d, and λ denote transmit and receive power, propagation distance, and path loss exponent, respectively. The lognormal distributed random variable Ψ models the large-scale fading or shadowing. In factor K the following figures are incorporated: antenna heights, carrier frequency, outdoor-to-indoor penetration loss, propagation conditions, and antenna pattern. Note that λ, K, and Ψ depend on line of sight (LoS) conditions as well. The degree of coverage C is defined as the fraction of the reference cell area which enjoys a receive power larger than a given threshold P min from at least one base station. Approximating the macro base stations antenna pattern as circular with full antenna gain in each direction, we can calculate the degree of coverage for homogeneous macro and micro networks by C = 2 R ( ) 1 log1 P min 1 log R 2 rq 1 P pl (r) d r (2) σ ΨdB where R and σ ΨdB denote the radius of the circular assumed cell area and the deviation of the large-scale fading, respectively [15]. The Q function is defined by Q(x) =1 Φ(x) where Φ is the cumulative distribution function of the normal Gaussian distribution. The right side of formula (2) is simply the accumulated probability that the receive power is larger than a given minimal power in the cell, normalized to the cell area. For a given network density and degree of coverage formula (2) provides the required transmit powers of the base stations. In heterogeneous networks the additional micro sites are considered as overlapping in the sense that they do not contribute to the coverage of the underlying macro system, e.g., no transmit power reduction of macro base stations is pursued. B. Power Consumption Model We describe the relation between transmit power and input power of macro and micro base stations using the power consumption model developed in [12]. In general, the main difference between both base station types is the design size where the micro base stations can be considered much more compact, resulting in limited capabilities in radiate power and coverage area and thus considerably smaller power consumption. Another difference is based on the quality and number of hardware components, e.g., macro base stations are equipped with more efficient PAs. For both macro and micro base stations the power consumption depends linearly on the transmit power, i.e., P = ap tx + b. (3) The parameter a accounts for PA efficiency and feeder loss as well as for the overhead due to site cooling, power supply, and battery backup. The power offset b takes into account the signal processing in particular and that part of the overhead which is independent of the transmit power. The power consumption of macro base stations is modeled as independent from the load, i.e., the number of resources allocated for transmission. On the contrary, we assume that micro base stations shall be able to adapt their power consumption to traffic load conditions, e.g., by adapting the PA. The reason is that for a given user density there is a high probability no user is camping in a small cell, so that power adaptivity is the key for efficient small cell deployments. Due to [12] the power consumption of a micro base station with regard to the load L [, 1] then can be written as P = ( ) a dyn L + a stat Ptx + ( ) b dyn L + b stat (4) with certain parameters describing the dynamic (load dependent) and static (load independent) contributions of a and b

3 Tab. 1: Power consumption model parameters per macro and micro base station antenna derived from [12] Macro Micro a ma 3.77 a mi,stat 4.44 a mi,dyn 1.11 b ma W b mi,stat W b mi,dyn W to the power consumption. Note that the power consumption models (3) and (4) can be applied per base station, per sector, or per antenna, respectively. In Tab.1 the parameter values for an individual antenna of a macro and micro base station are summarized. Multiplying these parameters with the number of antennas yields the total power consumption of a base station. C. Load Dependent Throughput Let I and I A denote the index set of all sectors in the network and in the reference cell A, respectively. Each sector A i with i Iis described by its associate area where the receive power from the base station serving this sector is largest compared to any other base station, i.e., { [ A i = x E Ψ Prx,i (x) ] [ E Ψ Prx,j (x) ] } j i [ with E Ψ Prx,i (x) ] = K(x)d(x) λ(x) E [ Ψ ] P tx using (1). Since we are interested in the SINR over a large time period and disregard any variation on short time scales, we define the so called long term average SINR in location x A i as [ E Ψ Prx,i (x) ] γ i (x) := j I\{i} E [ Ψ Prx,j (x) ] + σ, (5) 2 where σ 2 denotes the noise power. The definition in (5) can be seen as worst case since interference from all base stations, transmitting with full transmit power, is considered. The average throughput per subcarrier of bandwidth B sc in x is now calculated as T i (x) :=B sc min [ log 2 ( 1+γi (x) ),S max ]. (6) The parameter S max models the usage of finite modulation schemes in practice. In order to derive a load dependent throughput figure, consider a user process which generates random coordinates of mobile terminals in the reference cell. This user process induces random variables N and N i for i I A denoting the number of users in the reference cell and in sector i, respectively. By this definition it holds N = i I A N i.let further denote X i the random coordinates of a single user in sector i I A according to the user process. Under a full buffer assumption, i.e., all resources are allocated for transmission if there is at least one user requesting data, the throughput in the reference cell can be computed by T := i I A T i (X i )P [ N i > ]. (7) As figure of merit we choose the mean of the throughput T := E [ T ]. In this paper we assume a uniform user distribution in the reference cell which corresponds to a Poisson point process of users with a certain intensity μ which can be characterized, [ ] e.g., by the user density in the area, i.e., μ = E N A with A denoting the size of the reference cell. This leads to an expression for sector i being nonempty by P [ N i > ] =1 P [ N i = ] =1 e μ Ai. (8) D. Load Dependent Power Consumption The total power consumption equals the sum of power consumption figures of the individual base stations in the reference cell, i.e., P = i I A P i. The individual power consumption values P i are based on the power models (3) and (4) for macro and micro base stations, respectively, as well as the transmit powers calculated for the corresponding network density, coverage constraint, and load. Due to consistency reasons we employ the power model per antenna (refer to Tab.1), yielding for the individual power consumptions { N ma,ant,sec P ma if A i is served by a macro BS, P i = N mi,ant,sec P mi (L i ) if A i is served by a micro BS, where N ma,ant,sec and N mi,ant,sec denote the number of antennas per sector of macro and micro base stations, respectively. With regard to the spectral efficiency, it is convenient to identify the load in the corresponding sector as L i = P [ N i > ]. E. Energy Efficiency The efficiency of a system is inherently the ratio of a desired output to a given input. As commonly used we consider energy efficiency as the ratio of achieved average throughput to power consumption in the network, i.e., Q := T P, (9) measured in bit per second per watt. Although this metric is simple and intuitive, it does not capture network specific aspects such as coverage and user fairness. Also higher layer aspects, e.g., quality of service, are also neglected. Hence, the metric (9) should be complemented by other metrics. The question arises whether the metric Q can be chosen as an objective for optimizing a network s density, i.e., solving for an optimal inter site distance. Consider first the extreme case of an infinitely dense network, i.e., decreasing cell sizes. Since the throughput T i (x) in any location x in a sector of the reference cell is bounded by and B sc S max, the same is true for the mean throughput E [ ] T i in this sector. The weights P [ N i > ] in the summation of the individual throughput figures tend to zero for decreasing reference cell size. Hence the throughput T goes to zero. On the other hand, the power consumption of each base station tends to some finite positive

4 Tab. 2: Simulation assumptions Tab. 3: Base station and mobile terminal relevant parameters LTE specific parameters Carrier frequency 2. GHz Bandwidth 5 MHz FFT size 512 # Subcarriers occupied 3 Subcarrier spacing B sc 15 khz Transmission Downlink, OFDMA Deployment and model specific parameters D ma (hom. macro) m D mi (hom. micro) 5-35 m D het (heterogeneous) m # Micros per macro (heterogeneous) 3, 6, 9 (refer to Fig.1) User density per km 2 (uniform) 11, 5, 9, 138 Coverage C 95% Max. spectral efficiency S max 6 bit/s/hz Traffic model Full buffer Propagation specific parameters Macro propagation model Urban macro (UMa) [16] Micro propagation model Urban micro (UMi, hex.) [16] Outdoor-to-indoor penetration loss 2 db Macro antenna height 25 m Micro antenna height 1 m Macro antenna pattern typ. horiz. 3-sector [16] Statistics Outdoor users only Fading margins Fast fading margin Slow fading margin Inter-cell interference margin Mobile terminal sensitivity Thermal noise SNR required Noise per subcarrier Receiver sensitivity per subcarrier 2 db Considered in path loss model 3 db -174 dbm/hz db -132 dbm -12 dbm value due to the offset powers b and the lower boundedness of the transmit powers because of coverage constraints. This leads to Q going to zero. For the other extreme case, i.e., increasing cell sizes, we observe that the throughput T is bounded as well with P [ N i > ] 1, but the power consumption tends to infinity due to increasing transmit powers. We conclude that Q D and Q D, (1) which provides the metric with an optimum for finite network density for each homogeneous and heterogeneous deployment strategy. III. NUMERICAL RESULTS Our simulation assumptions equal the ones in [14] which are summarized in Tab.2. The lower and higher user densities of 11 and 138 users per square kilometer are based on typical simulation assumptions of 1 active users per macro sector, where the inter site distances are 5 m (3GPP Case 1) and 173 m (3GPP Case 3), respectively [16]. Relating the 1 users to the corresponding area of the sector yields the given user densities. The characteristics of the considered base stations and mobile terminals are provided in Tab.3 and the parameters of the power consumption model are given in Tab.1. Parameter Macro BS Micro BS MS # Antennas (per sector) # Sectors 3 1 Antenna gain (main lobe) 15 dbi 2 dbi -1 dbi Noise figure 4 db 4 db 7 db Tab. 4: Transmit power figures in homogeneous networks D 3 m 5 m 1 m 15 m 173 m P tx,ma 4 mw 31 mw 4.7 W 23.9 W 4. W D 5 m 1 m 2 m 3 m 35 m P tx,mi 5 mw 21 mw 4.1 W 19.1 W 34.5 W A. Throughput against Network Density In Tab.4 the transmit powers, calculated for the homogeneous macro and micro networks with different inter site distances by means of (2), are listed. In heterogeneous networks the transmit powers of the macro base stations correspond to the ones in homogeneous macro deployments. For micro sites in a heterogeneous deployment, the transmit power is calculated by requiring a cell radius of about 1 m and a coverage of 95%, yielding 2.3 W. In Fig.2 the mean throughput per subcarrier for the homogeneous micro network is plotted. It can be seen that in the considered range the throughput increases almost linearly for very low load. For moderate and high load situations a maximum can be observed. This becomes clear by taking a closer look at (7), where the throughput for small micro cells is mainly affected by the probability of the micro cell being loaded at all. Hence, the load dependent maximal throughput can be observed for fully loaded micro cells, where decreasing the micro cell size would lead to a significant decrease in the load. The decrease for higher inter site distance is due to the capacity limitation of the cells. In that regime a higher user density does not lead to a higher served throughput and thus the curves converge to a value independent of the user density. T (kbit/s/subcarrier) users per km users per km Fig. 2: Mean throughput per subcarrier of homogeneous micro networks as function of inter site distance for different load conditions.

5 Comparing heterogeneous deployments with the homogeneous macro and micro scenario, the introduction of additional micro base stations increases system throughput even for low traffic scenarios as shown in Fig.3 which is a consequence of the increased spatial reuse. Note that in order to facilitate reading, the low traffic figures for the heterogeneous networks with 6 and 9 micro base stations per macro cell are omitted. The huge gap between low and high traffic can be attributed to the probability that each sector in the reference cell contains at least one user. Hence, the gain in throughput due to additional micro sites in an existing macro deployment is quite high already for low load conditions. B. Power Consumption against Network Density Since the power consumption of macro base stations is modeled as constant regardless of the activity level, we can omit a detailed analysis for pure macro deployments with regard to different load scenarios. On the contrary, we have micro base stations whose consumed power consists of a static and a dynamic part. Therefore, we expect different system power consumption values for different traffic demands. In homogeneous micro networks we observe that for decreasing cell size the probability of the cell being empty becomes 1 very fast, providing almost identical power consumption values for different user densities. On the other hand, the power consumption figures for low and high load scenarios converge for sufficiently large cell sizes due to full load. In Fig.4 the power consumption in the reference cell for each deployment and load scenario is depicted. Note that the reference cell size varies for different inter site distances. The ability of micro base stations to adapt the power consumption to different load conditions does not lead to significant savings as can be seen for homogeneous micro networks and heterogeneous deployments as well. We come to the conclusion that although micro base stations have load dependent power consumption the total power consumption is not markedly affected. This means that only the throughput really depends on the load condition. T (kbit/s/subcarrier) users per km 2 9 users per km 2 Heterogeneous, 3 micros Homogeneous micro Heterogeneous, 6 micros Homogeneous macro Heterogeneous, 9 micros Fig. 3: Mean throughput per subcarrier as function of inter site distance for different deployments and load conditions. C. Energy Efficiency against Network Density In Fig.5 the Q metric is depicted for the various deployment strategies. As theoretically shown, the metric has an optimum for each topology (refer to (1)). This can be observed at least for the homogeneous networks. The impact, i.e., the location and the height of the maximum, is mostly affected by the nonempty cell probability (8) in homogeneous micro systems due to comparably small cell sizes. In the considered range of inter site distances, the sensitivity of Q on the load is relatively small in heterogeneous systems compared to homogeneous micro systems, at least for higher loads. The significant difference there between low and high load is due to the huge throughput variation. In general, the sensitivity of Q on the load is due to the high sensitivity of the throughput T since the power consumption is almost independent of the load. Note that the simulation is assuming a full buffer model. In reality the offered traffic of the mobile users may be limited. In this case, the energy efficiency gain of a deployment with micro cells is only beneficial as long as there is demand for the increased capacity. Otherwise the throughput will stay behind the capacity of the system while the power consumption increases by the added micro cells without gain in user experience. We conclude that the placement of micro base stations is very sensitive to traffic patterns regardless whether an existing macro network is to be supported or a new network of micro base stations is to be planned. Regarding the ratio Q, we further conclude that for high load scenarios homogeneous micro networks with optimal inter site distance are superior to homogeneous macro deployments. Also, macro networks gain in efficiency when additional micro sites are brought in the network, i.e., the gain in throughput justifies the additional power consumed by the micro base stations. This gain is increasing with the number of micro sites. We here discussed the energy efficiency. For very dense de- P (W) users per km 2 9 users per km 2 Homogeneous macro Heterogeneous, 3 micros Homogeneous micro Heterogeneous, 9 micros Heterogeneous, 6 micros Fig. 4: Power consumption per reference area as function of inter site distance for different deployments and load conditions (note that this area increases with the inter site distance).

6 Q (kbit/s/subcarrier/w) Homogeneous micro Homogeneous macro Heterogeneous, 3 micros 5 users per km 2 9 users per km 2 Heterogeneous, 6 micros Heterogeneous, 9 micros stations are thus better suited than homogeneous micro cell deployments. Future work should focus on the sensitivity of these results on the power consumption model parameters. For instance, it can be expected that future macro base stations will be able to scale their power consumption with the traffic load. Moreover, the location of micro sites should be extended to the whole macro cell area. Another aspect worth further consideration is heterogeneous networks with reuse larger than 1, e.g., heterogeneous deployments where macro and micro base stations transmit in different frequency bands Fig. 5: Energy efficiency as function of inter site distance for different deployments and load conditions. ployments it is necessary also to regard the power consumption per area [13]. In spite of high efficiency, the high number of cell sites comes along with a huge power consumption per area. In case of limited capacity demand heterogeneous deployments with optimal numbers of additional micro base stations make more sense than homogeneous micro cell deployments. IV. SUMMARY AND CONCLUSIONS In this paper we studied networks consisting of a mix of macro and micro base stations in comparison with homogenous systems. Using the three key performance indicators coverage, throughput, and power consumption, we defined energy efficiency as the ratio between throughput as desired output and power consumption as required input. Moreover, we investigated the impact of the load conditions on the network performance by changing the user density and modeling the micro base stations power consumption as load adaptive. We have theoretically shown that the energy efficiency metric as defined in this paper consists of a maximum for each deployment strategy with varying density, where the optimal network density depends, among others, on the load. Simulation results revealed that the throughput is very sensitive to the load, i.e., there is a significant difference between low and high load scenarios. However, although the micro base stations power consumption scales with the load, the total power consumption varies only very slightly with the load. The load dependency of the energy efficiency metric hence originates from the throughput. We further conclude that a network densification using micro base stations only is beneficial for medium and high load scenarios, at least in the considered range of network densities. For low load, a network densification using a mix of macro and micro cells is almost as efficient as a pure micro system. On the other hand, very dense networks come along with a huge power consumption per area [13], [14]. In case of limited capacity demand, heterogeneous deployments with optimal numbers of additional micro base REFERENCES [1] McKinsey & Company, The impact of ICT on global emissions, on behalf of the Global esustainability Initiative (GeSI), Tech. Rep., November 27. [2] SMART 22: Enabling the low carbon economy in the information age, The Climate Group, June 28, [3] Gartner Study, [4] EU Commissioner calls on ICT industry to reduce its carbon footprint by 2as early as 215, EC press release: MEMO/9/14, 3/3/29. [5] G. P. Fettweis and E. Zimmermann, ICT energy consumption - trends and challenges, in Proceedings of the 11th International Symposium on Wireless Personal Multimedia Communications, Lapland, Finland, September 28. [6] M. Gruber, O. Blume, D. Ferling, D. Zeller, M. Imran, and E. Calvanese Strinati, EARTH - Energy Aware Radio and Network Technologies, in Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), Cannes, France, Sep 28. [7] J. Louhi and H.-O. Scheck, Energy efficiency of cellular networks, in Proc. Int. Symp. Wireless Personal Multimedia Communications (WPMC), Lapland, Finland, Sep 28. [8] B. Gorini, ETSI work programme on energy saving, presented at Telecommunications Energy Conference, 27. [9] Ericsson, Sustainable energy use in mobile communications, August 27, White paper. [1] J. Gao, Analysis of energy consumption for ad hoc wireless sensor networks using a bit-meter-per-joule metric, IPN Progress Report 42-15, Aug 22. [11] G. Auer, I. Gódor, L. Hévizi, M. Imran, P. Fazekas, G. Biczók, and R. Tafazolli, The EARTH Project: Towards Energy Efficient Wireless Networks, in Proc. of 19th Future Network & MobileSummit 21, Florence, Italy, June 21. [12] O. Arnold, F. Richter, G. Fettweis, and O. Blume, Power consumption modeling of different base station types in heterogeneous cellular networks, in Proc. of 19th Future Network & MobileSummit 21, Florence, Italy, June 21. [13] F. Richter, A. J. Fehske, P. Marsch, and G. P. Fettweis, Traffic demand and energy efficiency in heterogeneous cellular mobile radio networks, in Proceedings of the 71th Vehicular Technology Conference (VTC Spring), Taipei, Taiwan, May 21. [14] F. Richter and G. P. Fettweis, Cellular mobile network densification utilizing micro base stations, in Proceedings of IEEE International Conference on Communications (ICC), Cape Town, South Africa, May 21. [15] A. Goldsmith, Wireless Communications. New York, NY, USA: Cambridge University Press, 25. [16] Technical Specification Group Radio Access Network, TR Further Advancements for E-UTRA: Physical Layer Aspects (Release 9), 3rd Generation Partnership Project, Tech. Rep., 29.