Power Modeling of Base Stations

Transcription

1 Power Modeling of Base Stations Björn Debaillie, Claude Desset Imec, Belgium 5GrEEn Summerschool, August 2014, Stockholm, Sweden

2 imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 2

3 Massive amount of communications devices Massive growth in traffic volume (both in data & signaling) Challenging requirements (latency, coverage, throughput availability,...) This evolution should be affordable and sustainable imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 3

4 Power consumption of mobile communications Energy Use Total Energy = 4 TWh/yr 0.1W per user for 5 billion Subscriptions Total Energy = 75 TWh/yr 1.7kW per each of the 5 million Base Stations Total Energy = <1 TWh/yr 1kW per each of the 17,000 Controllers Total Energy = 14 TWh/yr 10kW per each other elements Users Base Station Network Control Core & Servers With 80%, the base stations are by far the main consumers Based on: ETSI RRS05_024, NSN version 2011 imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 4

5 Base station functional components Radio Heads at least one per sector Can be: integrated with BBU (classic BS) independent (RRH) integrated with antenna (AAA) Can be multi-technology (e.g. LTE + GSM) # antennas depend on: number of sectors MIMO, beamforming supported antennas Analog connection Feeder loss ifo cable length antenna antenna antenna Radio Heads Base Band Unit Power Amplifier Power Amplifier Power Amplifier Cooling Analog TRx Analog TRx Analog TRx Power Supply Unit From/to backhaul and neighbor cells Digital Signal Digital Signal Digital processing Signal processing processing Digital control I/Q samples A/D conversion A/D conversion A/D conversion Digital connection No loss Internal/external (CPRI) imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 5

6 Base station power breakdown a quantified example Indicative values based on EARTH power model 2012 (macro-cell baseline scenario full load) antenna antenna antenna 3x20W EIRP 50% feeder loss Radio Heads Base Band Unit Power Amplifier Power Amplifier Power Amplifier Cooling 85W 20% efficiency 205W 205W Power Supply Unit 115W Analog TRx Analog TRx 25W Analog TRx 25W Digital Signal Digital Signal Digital processing Signal processing processing 90W A/D conversion A/D conversion A/D conversion 3x230W in RRH Digital control ~300W in cabinet 70% to RRH 980W Only 6% of the power is transmitted into the air imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 6

7 Power efficiency evolution of base stations imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 7

8 The EARTH project EARTH project (energy aware radio and network technologies): 50% reduction of the energy consumption in LTE based access networks effective and collaborative energy saving mechanisms in the wireless networks, their components, and its radio interfaces, while maintaining the users perceived quality of service and system capacity January 2010 June 2012 Deployment Network Management Components Zzz DC supply DC supply small cells off Small cell RF in PA Small Cells with Overlay Macro Cell Dynamic operation; Sleep modes, Bandwidth Adaptation, Power Amplifier & Transceiver, Load-adaptive Hardware imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 8

9 EARTH power model A matlab tool that provides realistic power consumption values of the base station and its components over different scenarios: Focuses on opportunities in LTE networks Covers different base station types (macro, micro, pico, femto-cells) Considering BAU (Earth OFF) and novel hardware (EARTH ON) Provides network mgnt and deployment layers realistic hardware values Enables analysis of the power consumption at component level The main scaling parameter is the traffic load imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 9

10 Load adaptive base stations Maximal load Data Traffic Day 1 Network capacity Day 2 Day 3 Daily data traffic profile for cellular systems in a dense urban environment Power Consumption Minimal load Sleep mode Data Load Data traffic varies during the day Wireless access networks are dimensioned for estimated peak demand Power amplifier efficiency decreases at low load Signaling traffic should be preserved imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 10

11 Load adaptive base stations a quantified example (continued) Pmax Total BS power consumption (EARTH 2012 macro-cell baseline system, BAU) ~60% Pmax Sleep mode ~60% of Pmax does not scale with the data traffic DSP, cooling, and power supply is poorly dependent on the traffic Signaling is continuously emitted (10% of P RF ) 0% Data load [%] 100% ETSI load definitions - Low load =10% P RF (no data; only signaling) - Medium load = 30% P RF (data + signaling) - Busy hour = 50% P RF (data + signaling) - Average = 6/24 low + 10/24 medium + 8/24 busy = 31.7% P RF imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 11

12 Power modeling can be easy Simply measure the base station power consumption 3 point measurements: mute, no load, full load Linear interpolation But... Load is the only scaling parameter; covers only limited scenarios Are power values representative for other base stations (types/size)? Power breakdown over the different hardware components? Impact of technology evolution? Etc.... has limited usage capabilities Power Consumption Traffic Load imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 12

13 Power breakdown depends on base station type PA Main Supply DC-DC RF BB Cooling 100% 8% Power consumption breakdown 80% 60% 40% 20% 9% 29% 7% 5% 7% 12% 5% 7% 64% 47% 33% 39% 16% 13% 7% 6% 9% 10% 36% 32% 0% Macro Micro Pico Femto Basic power model is insufficient with future base stations imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 13

14 EARTH power model First extensive base station power model Addressing flexible load-adaptive adaptations Multiple BS types, scenarios, parameters... Includes power optimization strategies (e.g. depending on traffic load) Embeds duty-cycle scenarios over the traffic load Contains hidden parameters and assumptions Macro-Cell Baseline System (EARTH OFF) CO BS Power Consumption [W] PS DC BB RF PA Relative RF Output Power [%] imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 14

15 Beyond the EARTH power model Shortcomings when going beyond EARTH No consistent definition of the model parameters User interface sometimes mixes scenarios and design Limited support for new systems and technologies No (de)activation information Need to go beyond EARTH: from 2x to 1000x EE improvements Much broader range of scenarios needed 2020 extrapolation needed Basic power model structure can be reused Split into main components, reference power and scaling rules Power optimization enhancements imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 15

16 How to improve the BS power and scalability Total BS power consumption (power scalable architectures) Scalability Performance scaling (MCS, SiNAD) Scalable MIMO Efficiency Smaller cells BS architecture Beamforming Massive MIMO Deactivation Multiple levels + fast reconfiguration (mirco-sleep) Components and subcomponents Process technology (static HW improvements) 0% Data load [%] 100% Indicative list of techniques only imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 16

17 The GreenTouch project Deliver architectures, specifications and roadmap of demonstrated key technologies to increase the network EE by a factor 1000 from 2010 to 2015 Bell Labs initiated Global Research Consortium representing industry, government and academic organizations New innovation Model for sustainability May Focus on energy efficiency, sustainability and growth Holistic and ambitious: Goal of 1000x 60 member organizations with 350+ leading scientist IMEC s power modeling received 9-months funding from GreenTouch in 2013 imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 17

18 Why IMEC? Imec is a research institute in nano-electronics and -technology, delivering industry-relevant technology solutions for ICT, healthcare and energy. We are not a network vendor or operator We have no specific activities on base station design or access networks In-house expertise in future processing technologies Green radio program focusing on radio solutions for handsets High and practical expertise in energy efficient radio system design 5+ years ahead of the component market Gained substantial knowledge and credibility over different projects High interaction and openness with industrial partners imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 18

19 GreenTouch power model outline Key model capabilities and features Hierarchical model architecture and parameters Base station and network architectures Technology evolution Base station (de)activation and sleep levels User interface and practical examples imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 19

20 GreenTouch power model Matlab tool which quantifies the power consumption and (de)activation delays of base stations and its (sub-)components Flexible model - Multiple base-station architectures and components - Embeds hardware energy-scaling (traffic load, MIMO, deployment...) - Hardware technology evolution (towards 2020 and beyond) Include transition effects - Active, idle, sleep... incl. reactivation delay Clear and user friendly interface - Separating the user/scenario and hardware parameters - Enables co- and re-simulation imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 20

21 GreenTouch power model outline Key model capabilities and features Hierarchical model architecture and parameters Base station and network architectures Technology evolution Base station (de)activation and sleep levels User interface and practical examples imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 21

22 Power model architecture Layer 1 Layer 2 Layer 3 Layer 4 Base station definition System scalability Translation layer Physical scalability Base station hardware (static due to installation) Model users Dynamic system configuration (dynamic to network variability) Translation from user-defined parameters to physical components Physical parameters to (sub)components configuration Goal: user friendly interface and convenient configuration & usage Hierarchical architecture with layers corresponding to abstraction levels Specific layers for normal users and model designers Each layer comes with specific parameters Layer 5 Power model core Power consumption tables and algorithms Model designers imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 22

23 Layer 1 Base station definition Base station hardware (static due to installation) Base station definition specifies the installed hardware Static base station characteristics Maximal capabilities (antenna chains, nominal output power...) Parameter Values Default Base station type large, small, data (BCG2), signal (BCG2), LSAS large Year of deployment Number of sectors any integer >= 1 Number of antennas (per sector) Maximum output power 1-8 for all types except LSAS (any integer >= 1) any (dbm, limited to the implemented PA model) 3 for large or signal, 1 for small, data or LSAS 4 for large, 2 for small and data, 1 for signal, 200 for LSAS Maximum bandwidth 1.4, 5, 10, 20 [MHz] 10 for all except 1.4 for signal Feeder loss >= 0 (db) 3 db for large, 0 db for other types TDD/FDD operation TDD, FDD FDD imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 23

24 Layer 2 System scalability Dynamic system configuration (dynamic to network variability) Specifies the current hardware configuration Covers link or network variability Covers high level hardware flexibility Parameter Values Default Notes System load fractional between 0 and 1 100% load (value 1) total system load Data load fractional between 0 and 1 100% load (value 1) Traffic throughput value in Mbps / Data-related signaling between 0 (full signaling) and 1 (full data) 0 data + data-related signaling load, after removing fixed signaling will determine load based on average MCS Fixed signaling overhead between 0 and system_load 0.1 Idle time profile any value > e-6 will be rounded to integer number of OFDM symbols (default = 1) Idle reactivation constraint any value between 0 and idle_time = idle_time Reduced bandwidth between 0 and 1 1 RF power control power modification [db] 0 negative value for reducing power MIMO configuration integer in [1 ; antennas] = antennas not used for LSAS MCS 2-6 for modulation, 0 and 1 for coding_rate, integer between 1 and 15 for CQI 6 for modulation, 1/2 for coding rate specified as either modulation and coding rate or CQI (not both) Spatial multiplexing imec integer 2014 in Confidential [1 ; antennas] Personal use only Power Modeling 30 of Base Stations 5GrEEn Summerschool LSAS only Aug page 24

25 Layer 2 System scalability Dynamic system configuration (dynamic to network variability) Advanced data/signal load definition Multiple parameters to accommodate various systems/users 100% Total resources Remaining resources 100% 80% 20% 0% Data + data-related signalling Fixed signalling System load Fixed signaling Data load 75% 0% 100% fractional signalling 10% 0% imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 25

26 Layer 3 Translation layer Translation from user-defined parameters to physical components Layer 3 defines the energy-scaling strategy when translating the user-defined parameters to physical components - Deactivation strategies Sleeping strategy (duty-cycling) or continuous operation Optimizing sleeping strategy based on scenario and components - Automatic transmit power control strategy Output power scaling (Y/N) at reduced load, bandwidth, #antennas - Power-performance trade-offs Scalable components (switching modes at, e.g., reduced bandwidth) Down-scaling signal accuracy in order to save power - Less PA linearity, fewer digital quantization bits, reduced dynamic range or EVM of analog components... - Special role for LSAS imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 26

27 Layer 4 & 5 Layer 1 Layer 2 Layer 3 Layer 4 Base station definition System scalability Translation layer Physical scalability Base station hardware (static due to installation) Model users Dynamic system configuration (dynamic to network variability) Translation from user-defined parameters to physical components Physical parameters to (sub)components configuration For model designers! Model of each (sub-)component Power consumption in reference scenario Scaling rules w.r.t. scenario parameters (layer 2) Extrapolation to different hardware designs (layer 1) Levels of deactivation and delay Local trade-offs (PA linearity, quantization, analog accuracy...) Layer 5 Power model core Power consumption tables and algorithms Model designers imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 27

28 Layer 4 & 5 Source of power consumption values and scaling factors Imec expertise on design of scalable radios in advanced technology Industrial partner interaction Literature (publications, conferences, data sheets,...) Extrapolation, estimated guesses,... EARTH power model Benchmark with other power models imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 28

29 GreenTouch power model outline Key model capabilities and features Hierarchical model architecture and parameters Base station and network architectures Technology evolution Base station (de)activation and sleep levels User interface and practical examples imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 29

30 Architectures and components of future base stations Large-cell base station BCG2-signalling Small-cell base station LSAS BCG2-data Large-cell base station Discrete components, heterodyne architecture, feedback path for calibration, cooling elements, multiple supply structures,... Small-cell base station Integrated analog/digital circuitry, direct conversion, light calibration,... BCG2 and LSAS architectures are more unique BCG2 for signaling and data: large and small-cell with specific features LSAS a bit of both, with specific features (no PA, relaxed dyn. range resol.) imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 30

31 GreenTouch power model outline Key model capabilities and features Hierarchical model architecture and parameters Base station and network architectures Technology evolution Base station (de)activation and sleep levels User interface and practical examples imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 31

32 Technology evolution towards 2020 Process and design technology evolution Process technology (silicon) Technology scaling (Moore s law) per scaling step: - 50% increases the Gops/W - 20% reduction in analog circuit Leakage problem! power consumption - Passive power vs active power - In todays baseband processors, 30% power loss due to leakage Emerging technologies required to sustain scaling - Material, processing, interconnection,... - e.g. FinFET, 3D stacking,... imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 32

33 Technology evolution towards 2020 Process and design technology evolution Design technology (architecture) Main trend in large-cell base station - Main focus on performance, less on efficiency and size - PA linearization: Doherty structure, Digital Predistortion Main trend in small-cell base station - Power efficiency (battery lifetime) : switching and digital PA s - Reconfigurable radios (multi-mode): simple circuits with massive control - Design flexibility and accuracy: digital transceivers Distribute and weight the process and design technology over the base station (sub-)components imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 33

34 GreenTouch power model outline Key model capabilities and features Hierarchical model architecture and parameters Base station and network architectures Technology evolution Base station (de)activation and sleep levels User interface and practical examples imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 34

35 Base station (de)activation and sleep levels EARTH indicated that base station deactivation is most promising for EE enhancement during low traffic load, but could not be fully quantified because of limited capabilities of the power model GreenTouch targets more advanced network architectures and mgmt Power model should embed base station deactivation information! Implemented at (sub-)component level of high accuracy Note: (sub-)component deactivated = UL/DL radio not operational imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 35

36 Component (de)activation time and power Quantified at (sub-)component level deactivation and reactivation delay on/off or multiple sleep levels power consumption over sleep levels and transitions imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 36

37 Base station (de)activation and sleep levels imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 37

38 GreenTouch power model outline Key model capabilities and features Hierarchical model architecture and parameters Base station and network architectures Technology evolution Base station (de)activation and sleep levels User interface and practical examples imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 38

39 GT_Power script (layer 1) Characterization of installed base station Base station type Antennas Bandwidth Sectors Output power Advantages of using script Configuring a scenario and saving the script for future use/reference Scenarios very different from default without long command-line Empty value [] possible to use default or dependent settings imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 39

40 GT_Power script (layer 2) Configuration of the base station in operation Load (data + signaling) Fractional use of: Antennas Bandwidth Output power Dynamism (idle time) Spectral efficiency (MCS) LSAS frame structure (if applicable) imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 40

41 GT_Power script (layer 3) Specific power-saving and architecture-related options Output power adaptation With load, bandwdith, antennas Duty-cycling and sleeping Specific architecture parameters PA control optimization RRH selection (reduced cooling) Backhauling model Derivation of asymmetric MIMO modes imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 41

42 Function call illustrating model outputs [Power in W, 3 sectors] Function call without parameters: default configuration (2020, 4x4, 10 MHz...) Details (for 1 sector) Per component For [downlink, uplink] Throughput [Mbps] is indicative only, based on input parameters Power model does not consider link budget, coverage, error rate... Coupling needed between power model and system/network simulator imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 42

43 Load reduction impact 50% load (continuously) 50% load (duty-cycled) Some more reduction from duty-cycling Default sleep time = 1 OFDM symbol, 71 µs Limits the power savings imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 43

44 Deeper sleep reduces power More savings 1 frame sleep time = 10 ms System enters deeper sleep modes imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 44

45 References for further readings Power model related projects This power model Other power models This power model references Claude Desset, Björn Debaillie, Filip Louagie, Towards a Flexible and Future-Proof Power Model for Cellular Base Stations, Tyrrhenian International Workshop on Digital Communications (TIWDC), Sept Claude Desset, Björn Debaillie, Vito Giannini, Albrecht Fehske, Gunther Auer, Hauke Holtkamp, Wieslawa Wajda, Dario Sabella, Fred Richter, Manuel J. Gonzalez, Henrik Klessig, Istvan Godor, Magnus Olsson, Muhammad Ali Imran, Anton Ambrosy, and Oliver Blume., "Flexible power modeling of LTE base stations, in WCNC, Paris, France, April Gunther Auer, Vito Giannini, Istvan Godor, Per Skillermark, Magnus Olsson, Muhammad Ali Imran, Dario Sabella, Manuel J. Gonzalez, Claude Desset, Oliver Blume, and Albrecht Fehske, How much energy is needed to run a wireless network?, IEEE Wireless Communications Magazine, special issue on Technologies for Green Radio Communication Networks, vol. 18, no. 4, Oct Dietrich Zeller, Magnus Olsson, Oliver Blume, Albrecht Fehske, Dieter Ferling, William Tomaselli, István Gódor, "Sustainable wireless broadband access to the future Internet - The EARTH project, chapter in book "The future Internet - Future Internet Assembly 2013: Validated Results and New Horizons," by Alex Galis and Anastasius Gavras, pp imec 2014 Confidential Personal use only Power Modeling of Base Stations 5GrEEn Summerschool Aug page 45