Opportunities for Energy Savings in Pico/Femto-cell Base-Stations

Transcription

1 Future Network & MobileSummit 211 Conference Proceedings Paul Cunningham and Miriam Cunningham (Eds) IIMC International Information Management Corporation, 211 ISBN: Opportunities for Energy Savings in Pico/Femto-cell Base-Stations Björn DEBAILLIE 1, Alexandre GIRY 2, Manuel J. GONZALEZ 3, Laurent DUSSOPT 2, Min LI 1, Dieter FERLING 4, Vito GIANNINI 1 1 imec, Kapeldreef 75, 31 Leuven, BELGIUM. bjorn.debaillie@imec.be 2 CEA-LETI, rue des Martyrs 17, 3854 Grenoble 9, FRANCE. 3 TTI, calle Albert Einstein 14, 3911 Santander, SPAIN. 4 Alcatel-Lucent Bell Lab, Lorenzstr.1,7435 Stuttgart,GERMANY. 1. Introduction Abstract: To support the growth and the dynamism in today s wireless communication, networks are evolving towards smaller cells and base-stations closer to the mobile users. This evolution opens the possibility to deal with one of the main problems in mobile networks: the energy consumption of the base-stations. In this paper, different opportunities are proposed to save energy in small-cell base-stations. These solutions, targeting energy adaptation, focus on the most power greedy components. The energy savings are shown both for the individual components as for a typical pico-cell base-station operating on LTE signals and considering a daily data traffic profile. This offers an average energy saving of 3%, but the flexible nature of the proposed solutions can be further exploited in heterogeneous networks. Keywords: energy efficiency, pico-cell base station, femto-cell base station, transceiver, baseband, power amplifier, EARTH. The wireless communication scene is both diverse and dynamic: the users show large variations over time [1], space [2] and frequency. Serving them with large homogeneous cells and static approaches is therefore inefficient [3]. Evolving towards a network of smaller cells and base-stations closer to the user facilitates to reduce the radiation power. Moreover, these smaller base-stations can be implemented as flexible and scalable solutions that adapt their operation to the dynamic wireless scene with high energy efficiency [4]. This efficiency is crucial because the continuous growth of the wireless communication scene leads to an intolerable ecological footprint and electricity costs. Based on this concern, the Energy Aware Radio and network technologies (EARTH) [5] project sets the ambitious objective to reduce the energy consumption of the 4 th generation mobile wireless communication network by 5%. This paper contributes to this objective by saving energy in the small-cell base-stations. First, the pico/femto base-station architecture is described and the components are identified that dominate the power consumption. Then, flexible analaog and digital solutions are presented that allow dynamic and energy-efficient smallcell operation. Finally, the energy savings when applying all these solutions in a pico-cell base-station are evaluated considering realistic LTE [6] traffic load and a daily cellular data traffic profile [7]. 2. Energy distribution The power consumption of a base-station (BS) is mainly dominated by the radio equipment. Such radio equipment is especially critical as it provides a physical interface between the Copyright 211 The authors Page 1 of 8

2 cloud of mobile users and the network; it must guarantee a continuous flow of information while providing sufficient quality-of-service. To reduce the power consumption of the radio equipment efficiently, it is essential to quantify the power consumption over the different radio equipment components and to focus on the main consumers. Figure 1 (left) shows a simplified block diagram of a pico/femto base-station transceiver supporting multiple transceivers, antennas and antenna interfaces (L). Each transceiver comprises a Power Amplifier (PA), a Radio Frequency (RF) transceiver, a baseband (BB) interface and a DC-DC power regulator. Both the RF and BB components offer reception and transmission capabilities for Up-Link (UL) and Down-Link (DL) operation respectively. Figure 1 shows on the right the energy distribution over the different basestation components for different cell sizes [7]. For macro-cell BSs, the power consumption is mainly concentrated in the PA. Scaling down the cell size reduces the relative weight of the power consumption from the PA towards the BB and RF components. For pico and femto-cell base-stations, the power consumption of the RF component is more than 12% and the BB and PA consume about 3% each. Therefore, for pico/femto-cell BSs, it is opportune to investigate energy enhancement techniques for these three components. Note that the power consumption of the antennas and their interface are minor consumers, but as they co-determine the energy efficiency of the PA, they are also considered as potential energy savers. PA Main Supply DC DC RF BB Cooling 1% 8% Pico/Femto-cell Base station BB-DL BB-UL Main supply DC-DC RF-DL RF-UL PA LLL Power consumption breakdown 8% 6% 4% 2% 9% 7% 29% 33% 5% 7% 12% 5% 16% 7% 7% 9% 64% 47% 39% 13% 6% 1% 36% 32% % Macro Micro Pico Femto Figure 1: Simplified block diagram of a pico/femto-cell base-station (left) and the BS power consumption breakdown for different cell-sizes (right). 3. Energy adaptation opportunities Normally, base-stations are designed for maximal traffic load and maximal performance. Unfortunately, the daily data traffic profile for cellular systems in a dense urban environment (Figure 2) shows high variations. The highest utilisation ratio is achieved only between 18. and 24.. The remaining time (18 hours), the base-station is under-utilized which opens the opportunity for different techniques to enhance the BS energy efficiency while maintaining the required traffic load. Sleep modes: During low traffic periods ( ) no or very few users are served such that parts of the BS components can be switched off. A low energy sensing system will monitor the network activity and wake-up the BS if required, while taking into account transition times for activation and deactivation. Run time savings: During reduced traffic periods, varying capacity should be provided to the area and/or timeframe. This capacity is scaled by tuning the transceiver RF power or Signal to Noise and Distortion (SiNAD), activation/deactivation the amount of antennas in Copyright 211 The authors Page 2 of 8

3 a MIMO system, by scaling the bandwidth and by frequency and time-duty-cycling. According the data frame structure, some specific BS components can also be discontinuously activated (DTX) at symbol rate speed. Heterogeneous deployment: At the network level, the BS energy efficiency scalability can be further exploited. With a proper heterogeneous network, it could for example be opportune during low traffic to deactivate several pico-cell BSs and to transfer their traffic to a nearby macro-cell BS. 1% Normailized Traffic Load [%] 8% 6% 4% 2% % Time [hours] Figure 2: Daily (24h) data traffic profile for cellular systems in a dense urban environment [7]. 4. Energy adaptation solutions In this section, attractive and realistic solutions are presented to adapt and enhance the energy efficiency of the main power consumers in the pico/femto BSs: BB, RF and PA Digital baseband engine Due to the shrinking cell size and the rapidly growing signal processing complexity, the energy consumption of digital baseband implementation is becoming more and more dominant. Hence, optimizing the energy efficiency of digital baseband is crucial. The key technique is to enable energy scalability for both the dynamic wireless communication environment and the dynamic user requirement, such as the load of the system. In general, baseband processing can be split into many components. Among other timedomain processing for filtering and up/down-sampling, frequency-domain processing for modulation/demodulation or equalization, channel coding/decoding, pre-distortion, platform control and backbone network serial link. These components have the potential for Energy Adaptation (EA) according to the signal load (defined as the output power related to the maximum specified transmission power) by adaptation of the bandwidth, modulation, coding rate, number of antennas, duty-cycling in time or frequency etc. Figure 3 shows on the left the power consumption of a 2x2 MIMO pico-cell BS BB engine over the signal load. The UL power is almost double the DL given the more complex signal processing needed at receiver side (MIMO signal detection, channel decoding...). The potential for EA is mainly concentrated in the UL case because most of its signal processing is proportional to the data throughput. In the DL case, BB processing is relatively independent of the throughput (time-domain processing of the signal, platform control overhead...). The overall power reduction ranges up to 2% at signal loads <2%. At higher signal loads, the power reduction decreases. Copyright 211 The authors Page 3 of 8

4 Power Consumption [W] UL UL EA Potential for Energy Savings 1% 2% 15% Power reduction 2% 4% 6% 8% 1% DL DL EA Power Consumption [W] % UL+DL UL+DL EA 39% 32% Potential for Energy Savings Power reduction % 2% 4% 6% 8% 1% Signal Load [%] Signal Load [%] Figure 3: Power consumption of a 2x2 MIMO pico-cell base-station BB engine (left) and RF transceiver (right) over the traffic load Analog RF transceiver The introduction of energy adaptation in the base-station RF transceiver is inspired by recent work in the area of handheld communication devices, where an analog software defined radio (SDR) is implemented in low-cost and power-efficient CMOS technology [8]. This SDR builds on a simple zero-if architecture which is highly reconfigurable over all its analog building blocks. This flexibility permits to configure the radio over diverse standards and operation conditions. The hardware provides flexibility over the operation frequency and offers to control the filtering and amplification in the cascaded transmitter and receiver stages. This control is particularly interesting in the context of energy optimisation, as both the bandwidth and the gain/sinad performance are key components for energy adaptation. To efficiently exploit the transceiver flexibility for run-time energy optimization according to the traffic load, an off-line pruning is required to select all relevant configurations. In traditional base-stations, the RF transceiver targets the best SiNAD performance independent of the signal load. From an energy consumption perspective, it is more advantageous to scale the transceiver to provide a just good enough SiNAD performance. This is illustrated in Figure 3 (right), which shows the measured power consumption on [8] for a 2x2 MIMO pico-cell base-station RF transceiver over the signal load. This figure depicts the power consumption for the UL receiver, the DL transmitter and the complete transceiver (UL+DL). The dotted lines correspond to the traditional approach whereas the solid lines consider energy adaptation of scaling the SiNAD performance. Energy adaptation is mainly beneficial at lower traffic load, where the transceiver power consumption is reduced beyond 3% (signal load <5%) and up to 55% (signal load <5%) Power amplifier In pico and femto base stations, the power amplifier does not represent the main consumer block. However, because of the spectral mask restrictions and the lack of digital predistortion techniques, the operating peak-to-average-power-ratio (PAPR) must be increased causing that the energy efficiency of the PA goes down significantly. The concept Adaptive Energy Efficient Power Amplifier is proposed to lead the power consumption reduction when the RF output power is lower than the maximum. The main features are: Copyright 211 The authors Page 4 of 8

5 Operating Point Adjustment PA operating point is optimised according to required RF output power level (signal load). At the same time, spectral mask and PAPR specification are fulfilled. Deactivation of PA Stages Enabling a fast switching on/off in the RF power transistor, the consumption is reduced to the minimum when no RF output power is required. The implementation of this concept is made in cooperation with digital baseband and power supply unit blocks. A line-up power amplifier with at least two stages is required. The first stage has a fixed bias point; however, the second stage operating point is optimized through variable bias circuitry, which is controlled remotely by digital base band. Based on a 24dBm maximum rms RF output power (AB-class, PAPR=12dB) [9], the benefits of Adaptive Energy Efficient (EE) PA (Figure 4) has been evaluated in five different operating points (OP1 to OP5) and in switch off condition. Given that the operating point for maximum signal load (OP1) represents the performance without adaptation, the comparison between OP1 and the others provides the enhancement of the proposed concept. The power consumption when PA is deactivated (% signal load), is drastically reduced around 8%. The improvements when PA is adjusted in function of the signal load, also provides high reductions, mainly, for medium and low signal loads. For instance, below 2% signal load, in OP5, the expected reduction is in the range of 55%. In OP4, for signal load range 2-4%, the benefit is around 37%, while for OP3 (22%) and OP2 (11%), the improvement is still quite significant. 3. Adaptive EE PA Performance 2.5 Power Consumption, W OP1 OP3 OP5 OP2 OP4 SWITCH OFF Signal Load, % Matching network Figure 4: Power consumption performance of Adaptive EE PA. Adaptive power amplifiers scale their RF power according to the signal load, but classically, their load impedance is fixed. From an energy efficiency point of view, the load impedance should however scale inversely proportional to the RF power. Recent technological evolution facilitates high quality variable passive components to design low loss Tunable Matching Networks (TMN) that allow dynamic load modulation. Figure 5 shows the PA efficiency curves with a classical fixed matching network (blue curve) and a tunable matching network (green curve). The green curve is obtained from selecting the optimal load impedance and input drive levels over the RF power. At 12dB back-off, which corresponds with a realistic PAPR for e.g. LTE signals in pico/femto BSs, the energy efficiency increases from 17% up to 32%, which corresponds to a 9% improvement. This efficiency improvement however does not include the TMN loss and its tuning range limitations. Figure 6 shows the estimated power consumption of a 24dBm PA for different signal loads, considering.5db and 1dB loss in the fixed matching and tunable matching Copyright 211 The authors Page 5 of 8

6 networks respectively. The main energy savings are observed at high signal load: 4% power reduction is achieved at full signal load and 3% power reduction at half signal load. 3, with Fixed Matching Network (IL=.5dB) 2,8 with Tunable Matching Network (IL=1.dB) Efficiency Improvement +9% Power Consumption [W] 2,6 2,4 2,2 2, 1,8 1,6 1,4 3% Power Reduction 4% 12dB back-off 1,2 1, 1% 2% 3% 4% 5% 6% 7% 8% 9% 1% Signal load [%] Figure 5: PA efficiency curve (excluding matching loss) with fixed matching (blue) and tunable matching (green); black dots represent efficiency data for different load impedances and input power levels. Figure 6: Power consumption for different signal loads Tx/Rx antenna port Standard RF front-end architectures for FDD communications are based on a 5 reference impedance and make use of a duplexer to connect the Tx and Rx chains to the antenna(s) (Figure 7). While this approach allows the separate optimization of each block of the system, significant efficiency losses are caused in the matching network and duplexer by the requirements on impedance matching and isolation. Typical state-of-the-art duplexers, based on SAW or BAW technology, exhibit isolation levels between Tx and Rx ports in excess of 5 db. Insertion losses depend on the Tx and Rx frequency-bands separation and are typically in the range db. IL 1-2dB IL 1.5-3dB IL 1dB IL 1dB Loss <.5dB PA LNA TMN Tx Rx Isolation > 5dB Loss.5-1dB Figure 7: Typical RF front-end architecture of a pico/femto base station for FDD. PA LNA TMN IL 1dB Tx Isolation > 3dB Figure 8: Low loss RF front-end architecture with separate Tx/Rx antenna ports and relaxed filter specifications. Since the duplexer is a critical piece of the front-end, one may question the interest of using a single antenna. Separate Tx/Rx antennas can provide some level of isolation that may reduce the requirements of the Tx and Rx filters, and in turn result in lower insertion losses for these filters. The antennas can be implemented as identical antenna elements separated in space or can be the combination of different antenna elements into the same volume in order to minimize the size. Custom antenna design allows the development of dual-access (Tx/Rx) antennas (Figure 8) with a significant isolation between ports and low impedance levels (<2 ohms). With such antenna, the PA matching network and Rx filter isolation specifications can be significantly relaxed with a net benefit on insertion losses and a global efficiency of the Tx front-end: the total insertion loss is reduced from 3.5dB down to 2.5dB and the efficiency improvement ranges between 25% and 1%. Rx Copyright 211 The authors Page 6 of 8

7 5. System evaluation This paragraph provides a system evaluation based on [7] for a typical 1MHz bandwidth 2x2 MIMO pico-cell with 24dBm maximum rms output power. The main objective is to map a certain percentage of traffic load expressed in terms of relative rms RF output power to the actual power consumption of the small-cell BS. FIGURE 9 shows a comparison between a conservative state-of-the-art implementation (Earth OFF) with the power scalable implementations we propose in this paper (Earth ON). Savings up to 5% can be achieved in low load conditions. Furthermore, advanced sleep-modes can be easily implemented that make the small BS consumes a stand-by power that is a small fraction of the maximum. This is critical when small-cells are deployed in heterogeneous networks; at low capacity, selected small-cell BS will be put to sleep to save energy. ] [W n p tio m u n s o C w er o B SP ] [W n p tio m u n s o C w er o B SP Earth OFF RF Output Power [%] Earth ON RF Output Power [%] PS CO DC BB ] RF [W PA n p tio m u n s o C PS w er COo B SP DC BB RF PA PicoCell - 24dBm Max Earth OFF Earth OFF+Sleep Mode Earth ON Earth ON+Sleep Mode RF Output Power [%] FIGURE 9 Pico-cell BS power consumption versus the signal load 1% 12 9% 8% 1 7% 6% 5% 4% 3% no output PSS, SSS, BCH CSRS CSRS and BCH ] [W n o p ti m u n s o 8 6 2% 1% % P(user data) w erc o B SP 4 2 Load [%] FIGURE 1 Example of FTP resource utilization model Day [h] FIGURE 11 Possible savings over typical daily traffic-load In order to quantify the average savings over the dense-urban scenario described in section 3, an FTP resource utilization model is used [7], shown in FIGURE 1, which distributes Copyright 211 The authors Page 7 of 8

8 probabilities of user data versus LTE signalling over the relative traffic-load that a pico-cell can support. FIGURE 11 shows how both power model and resource utilization result in an average power consumption distribution over the day that allows in relative terms savings at the base-station up to 3%. From a system network perspective, this energy adaptation is exploited by building heterogeneous networks with a flexible deployment and scheduling depending on the user profiles, the traffic load and the environmental conditions (channel, interference, etc.). 6. Conclusion This paper presents new concepts to save energy in small-cell wireless communication base-stations. The power consumption of these base-stations is dominated by three components: the digital baseband engine (~3%), the analog RF transceiver (~12%) and the power amplifier (~3%). For these components, energy adaptation solutions are identified and quantified in function of the signal load. These solutions offer a high potential in energy savings beyond the technological improvements while enhancing the flexibility to adapt to the dynamic wireless communication scene. A system evaluation has been performed on a pico-cell base-station (2x2 MIMO, 1MHz bandwidth) operating on LTE signals and considering a daily data traffic profile in a dense-urban environment. This evaluation indicates an average power consumption reduction of 3% when exploiting the energy adaptation solutions described in this paper. Acknowledgements The research leading to these results has received funding from the European Community's Seventh Framework Programme FP7/ under grant agreement n project EARTH. References [1] D. Willkomm, S. Machiraju, J. Bolot, A. Wolisz, Primary Users in Cellular Networks: A Large-Scale Measurement Study, New Frontiers in Dynamic Spectrum Access Networks, 28. DySPAN 28. 3rd IEEE Symposium on, Oct. 28. [2] D. Tang, M. Baker, Analysis of a Local-Area Wireless Network, International Conference on Mobile Computing and Networking, 2. [3] S. Boiardi, A.Capone, B. Sanso, Radio Planning of Energy-Aware Cellular Networks, IEEE Journal on Selected Areas in Communications, 21. [4] L. Van der Perre, J. Craninckx, A. Dejonghe, Green Software Defined Radios, Springer 29. [5] EARTH project under European Community's Seventh Framework Programme FP7/27-213, grant agreement n [6] Ericsson White Paper: Long Term Evolution (LTE): an introduction, October 27. [7] INFSO-ICT EARTH-Report D2.3, "Energy efficiency analysis of the reference systems, areas of improvements and target breakdown," Dec. 21. [8] M. Ingels, V. Giannini, J. Borremans, G. Mandal, B. Debaillie, P. Van Wesemael, T. Sano, T. Yamamoto, D. Hauspie, J. Van Driessche, J. Craninckx, A 5mm 2 4nm LP CMOS.1-to-3GHz multi-standard transceiver, IEEE International Solid-State Circuits Conference, pp , Feb. 21. [9] NPDTB4 Datasheet. Gallium Nitride 28V, 5W RF Power Transistor. Nitronex Corporation. Copyright 211 The authors Page 8 of 8