earth_wp4_ead4.3

EARTH PROJECT

EARTH_WP4_D4.3 1 / 121

INFSO-ICT-247733 EARTH

Deliverable D4.3

Final Report on Green Radio Technologies

Abstract:

This deliverable is the final report on energy efficiency improvements enabled by base station radio technologies. Innovations on hardware components lead to a reduction of power consumed by base stations. This benefit is increased by combining them with new interface techniques for maximizing the energy efficiency. Radio transmission techniques are analysed, studying their impact on power consumption and finding their most energy efficient way of use. The concepts and solutions are analysed in terms of their energy efficiency potential and their adaptability to system dynamics for macro- and small-cells. Moreover, each solution is analyzed for specific deployment scenarios, operating conditions, optimization time scales, and traffic loads. The results in this document represent key enablers for energy saving in radio networks and are the basis for the evaluations of the “Integrated Solutions” in the project.

Keyword list: Mobile communications, Green Radios, energy efficiency, radio node, base station, MIMO, macro-cell, small-cell, beamforming, radio interface technologies, transceiver components, power saving, adaptability.

Contractual Date of Delivery: June 30, 2012

Actual Date of Delivery: July 16, 2012

Editor(s): Main Editor: Dieter Ferling (ALUD); Chapter Editors: Dieter Ferling (ALUD), Serge Bories (CEA), Custódio Peixeiro (IST), Mona Hashemi (EAB), Mauro Boldi (TI), Rodolfo Torrea (IMEC)

Author(s): Dieter Ferling, Anton Ambrosy (ALUD), Sven Petersson, Mona Hashemi, Per Burström (EAB), Aykut Erdem, Philippe Maugars, Jean-Marie Retrouvey (NXPFR) , Shinji Mizuta, Guido Dietl (DOCOMO), Mauro Boldi (TI), Alexandre Giry, Serge Bories (CEA), Björn Debaillie, Rodolfo Torrea (IMEC), Tiago Gonçalves, Filipe Cardoso, Luis Correia, Custódio Peixeiro (IST), Jouko Leinonen (UOULU), Yolanda Fernandez (TTI)

Participant(s): ALUD, EAB, NXPFR, DOCOMO, TI, CEA, IMEC, IST, UOULU, TTI

Work package: WP4

Estimated person months: 40 PM

Security: Public

Version: 1.0

Estimated person months: # PM

Security: Confidential

Version: x.x.x

EARTH PROJECT

EARTH_WP4_D4.3 2 / 121

Disclaimer: This document reflects the contribution of the participants of the research project EARTH. The European Union and its agencies are not liable or otherwise responsible for the contents of this document; its content reflects the view of its authors only. This document is provided without any warranty and does not constitute any commitment by any participant as to its content, and specifically excludes any warranty of correctness or fitness for a particular purpose. The user will use this document at the user's sole risk.

EARTH PROJECT

EARTH_WP4_D4.3 3 / 121

Authors

Partner Name Email and Phone

Alcatel-Lucent Deutschland AG (ALUD) Dieter FERLING

Anton AMBROSY

[email protected] +49 711 82142992 [email protected] +49 711 82145631

Ericsson AB (EAB) Sven PETERSSON

Per BURSTRÖM

Mona HASHEMI

[email protected] +46 10 71 23583 [email protected] +46 10 71 44556

[email protected] +46 10 71 39107

NXP Semiconductors France (NXPFR) Aykut ERDEM

Philippe MAUGARS

Jean-Marie RETROUVEY

[email protected] +33 2 31 45 63 96 [email protected] +33 2 31 45 64 67 [email protected] +33 2 31 45 64 13

DOCOMO Communications Laboratories Europe GmbH (DOCOMO)

Shinji MIZUTA

Guido DIETL



Telecom Italia S.p.A. (TI) Mauro BOLDI [email protected] +390112287771

Commissariat à l'Energie Atomique (CEA) Alexandre GIRY

Serge BORIES

[email protected] +33438783965 [email protected] +33 4387851863

Interuniversitair Micro-Electronica Centrum vzw (IMEC)

Björn DEBAILLIE

Rodolfo TORREA

[email protected] +3216281851 [email protected] +32 16288220

Instituto Superior Tecnico (IST) Tiago GONCALVES

Filipe CARDOSO

Luis CORREIA

Custódio PEIXEIRO




[email protected]

mailto:[email protected]



















EARTH PROJECT

EARTH_WP4_D4.3 4 / 121

Partner Name Email and Phone

+351 218 418 478

University of OULU (UOULU) Jouko LEINONEN [email protected] +358 8 553 2968

TTI Norte, SL (TTI) Yolanda FERNANDEZ [email protected] +34 942 29 12 12


EARTH PROJECT

EARTH_WP4_D4.3 5 / 121

Executive Summary

This document describes the final results on energy efficiency improvements enabled by radio base stations technologies. The proposed solutions are focused on base station applications for 4G broadband cellular networks (LTE). They deal with several domains like hardware components and equipment, which provide increased energy efficiency or power management features; with radio interface techniques providing special scheduling techniques or air interface features to maximize the benefit of the power saving features of hardware components; and with radio transmission techniques providing solutions to use multiple antenna techniques in an energy efficient way.

Some solutions to reduce the power consumption show their benefit during high traffic load situation, supporting the requirement for energy efficiency in situations of high throughput. Power amplifiers with increased peak efficiency and antenna interfaces with reduced insertion loss provide 30% of efficiency improvement of these components in small-cell base stations, while a technical solution to increase the radiation efficiency of printed antennas by 15% is applicable for every base station power class. The energy efficient use of multi antenna systems allows for 8-9% of energy savings during high traffic load, when applying beam forming, taking care for typical movement of the traffic in a cell. In MIMO systems, the adaptation of modulation and coding schemes combined with discontinues transmission techniques and component sleep modes allow for energy efficient operation at high traffic load with e.g. 37% of daily energy savings in a dense urban scenario.

Above all, the highest potential to reduce the power consumption is in low and medium traffic load situations. Several EARTH hardware innovations enable sleep modes for components during time periods without signal transmission or energy aware adaptation of component operation to minimize the power consumption at arbitrary load level. Key components of these innovations have been designed and implemented in prototypes. These components for macro- or small-cell base stations have been evaluated providing power characteristics for the base station power models.

Scheduling solutions operating in time and frequency domain follow the principles to increase time slots without signal transmission and apply sleep modes in components; to limit the number of resource elements and adapt the component operation; and to apply energy opportunistic scheduling to benefit from good channel conditions. An optimized combination of these approaches lead to 50-60% reduction of daily energy consumption. These combinations are possible within current LTE standards. Considering the possibility of future modifications in the standard, which allow the application of sleep modes to an increasing degree, additional energy savings are expected.

Current deployments apply multi-antenna techniques for MIMO to optimise spectral efficiency. This comes with an energy efficiency penalty in low load situations. The EARTH concepts for fast deactivation of some antennas in situations of low and medium traffic load, combined with sleep modes for components, provides up to 24% of daily energy savings in medium traffic load scenarios. This approach allows operating base stations designed for highest throughput in an energy efficient way during non-busy hours.

Beyond the direct technical advance in the area of radio base stations, the provided collection of technical solutions and the models for base station power behaviour have been the basis for work in other work packages. The solutions described in the document are key parts of the integrated solutions. Their power models enabled to evaluate the expected overall energy efficiency improvement of radio network solutions and of the integrated solutions.

EARTH PROJECT

EARTH_WP4_D4.3 6 / 121

Table of Contents

1. INTRODUCTION ................................................................................................................... 15

2. TRANSCEIVERS FOR MACRO-CELL BASE STATIONS ............................................................... 17

2.1. DSPC - SOLUTIONS FOR SIGNAL LOAD ADAPTIVE TRANSCEIVERS ........................................................... 19 2.2. DUAL POWER MODE SMALL SIGNAL RF TRANSCEIVER ........................................................................ 20 2.3. ADAPTIVE ENERGY EFFICIENT POWER AMPLIFIER ............................................................................... 23 2.4. ADAPTIVE POWER SUPPLY UNIT .................................................................................................... 25 2.5. BENEFITS ON ENERGY EFFICIENCY IN MACRO-CELL BASE STATIONS ......................................................... 28

3. TRANSCEIVERS FOR SMALL-CELL BASE STATIONS ................................................................. 31

3.1. ADAPTIVE SMALL-CELL BASEBAND PROCESSOR ................................................................................. 32 3.2. ADAPTIVE SMALL-CELL RF TRANSCEIVER .......................................................................................... 34 3.3. ADAPTIVE ENERGY EFFICIENT POWER AMPLIFIER ............................................................................... 36 3.4. TUNEABLE MATCHING NETWORK FOR LOAD ADAPTIVE POWER AMPLIFIER ............................................ 39 3.5. LOW-LOSS ANTENNA INTERFACE .................................................................................................... 41 3.6. BENEFITS ON ENERGY EFFICIENCY IN SMALL-CELL BASE STATIONS .......................................................... 43

4. LOW LOSS ANTENNAS ......................................................................................................... 45

4.1. PRINTED ARRAYS ON FOAM MATERIAL ............................................................................................ 45 4.2. EXPERIMENTAL RESULTS .............................................................................................................. 46 4.3. BENEFITS ON ENERGY EFFICIENCY DUE TO LOW LOSS ANTENNAS ............................................................ 48

5. RADIO INTERFACE TECHNIQUES EXPLOITING HARDWARE FEATURES .................................... 49

5.1. CELL DTX ................................................................................................................................. 50

5.1.1. Long DTX Network Implementation .......................................................................................... 50 5.1.2. Micro, MBSFN-based and Short DTX Evaluation Results Aggregated to Global Scale Using the

E3F ............................................................................................................................................ 52

5.2. SCHEDULING STRATEGIES OPERATING IN FREQUENCY AND TIME DOMAIN .................................................. 53 5.3. ADAPTIVE ENERGY EFFICIENT SCHEDULING ALGORITHM FOR LTE PICO BASE STATIONS ................................. 56 5.4. BENEFIT ON ENERGY EFFICIENCY DUE TO RADIO INTERFACE TECHNIQUES ................................................. 59

6. BEAMFORMING IN RECONFIGURABLE AND ADAPTIVE ANTENNA SYSTEMS .......................... 61

6.1. RECONFIGURABLE ANTENNA SYSTEMS ............................................................................................. 61

6.1.1. Offered traffic ........................................................................................................................... 63 6.1.2. Number of hotspots and hotspot probability ............................................................................ 64

6.2. ADAPTIVE BEAMFORMING ON ACTIVE ANTENNAS .............................................................................. 65

6.2.1. Active Antennas ........................................................................................................................ 66 6.2.2. Adaptive Beamforming ............................................................................................................. 67

6.3. BENEFIT ON ENERGY EFFICIENCY DUE TO BEAMFORMING AND ACTIVE ANTENNAS ...................................... 69

7. SMART MIMO OPERATION .................................................................................................. 70

7.1. ENERGY EFFICIENT RESOURCE ALLOCATION USING MULTIPLE TRANSMIT ANTENNAS .................................. 70

7.1.1. MIMO Power-Performance Optimization in Picocells ............................................................... 70 7.1.2. MIMO and Resource Allocation Optimization in Macrocells ..................................................... 72

7.2. ANTENNA MUTING IN LTE............................................................................................................ 75

7.2.1. Link Level Evaluation Results .................................................................................................... 77

EARTH PROJECT

EARTH_WP4_D4.3 7 / 121

7.2.2. System Level Evaluation Results ............................................................................................... 79

7.3. ZERO-FORCING LIKE MULTIUSER MIMO PRECODING SUPPORTING THE CONSTANT MODULUS PROPERTY ....... 80 7.4. BENEFIT ON ENERGY EFFICIENCY DUE TO SMART MIMO USAGE ............................................................ 83

8. CONCLUSIONS ..................................................................................................................... 85

9. REFERENCES ........................................................................................................................ 88

10. APPENDIX............................................................................................................................ 90

10.1. SIGNAL LOAD ADAPTIVE TRANSCEIVER ........................................................................................ 90

10.1.1. Impact of gain variation during component reconfiguration ..................................................... 90 10.1.2. Circuitries for the small signal RF transceiver ............................................................................ 91

10.1.2.1. Receiver ................................................................................................................................... 91 10.1.2.2. Dual Mode Transmitter ............................................................................................................ 94

10.1.3. AEEPA module .......................................................................................................................... 96 10.1.4. Power characteristic of the APSU .............................................................................................. 97

10.2. LOW-LOSS ANTENNA INTERFACE ............................................................................................... 98 10.3. LOW LOSS FOAM PRINTED ANTENNAS ...................................................................................... 100

10.3.1. Foam Printed Antennas .......................................................................................................... 100 10.3.2. Foam Characterization ............................................................................................................ 100 10.3.3. Software Validation ................................................................................................................ 101 10.3.4. Antenna Application Examples ............................................................................................... 101

10.3.4.1. Dual-Band Element ................................................................................................................ 102 10.3.4.2. Dual-Band Array ..................................................................................................................... 103 10.3.4.3. Final Remarks ......................................................................................................................... 104

10.4. ACTIVE ANTENNAS ............................................................................................................... 105 10.5. BEAMFORMING METHODS ..................................................................................................... 107

10.5.1. Reconfigurable “per cell” ........................................................................................................ 107

10.5.1.1. Introduction. .......................................................................................................................... 107 10.5.1.2. Simulator details .................................................................................................................... 107

10.5.2. Single DoA, Eigen BF ............................................................................................................... 109

10.5.2.1. DoA based beamforming ....................................................................................................... 109 10.5.2.2. Eigen Beamforming algorithm ............................................................................................... 110

10.5.3. MMSE ..................................................................................................................................... 112

10.5.3.1. Introduction ........................................................................................................................... 112 10.5.3.2. Model and simulator implementation ................................................................................... 112 10.5.3.3. Simulation scenarios and results ........................................................................................... 113 10.5.3.4. Conclusions ............................................................................................................................ 117

10.6. CELL DTX .......................................................................................................................... 119

10.6.1. Micro DTX ............................................................................................................................... 119 10.6.2. MBSFN-based DTX .................................................................................................................. 120 10.6.3. Short DTX ................................................................................................................................ 121

EARTH PROJECT

EARTH_WP4_D4.3 8 / 121

List of Figures

FIGURE 1. 24 HOURS TRAFFIC LOAD PROFILE. .................................................................................................................................. 15 FIGURE 2. RELATIVE RF OUTPUT POWER LEVEL ACCORDING TO THE LTE FRAME STRUCTURE. .................................................................... 17 FIGURE 3. BLOCK DIAGRAM OF A TRANSCEIVER SYSTEM FOR MACRO-CELL BASE STATIONS. ....................................................................... 18 FIGURE 4. ARCHITECTURE OF A SCC BLOCK OF THE DIGITAL TRANSMIT BLOCK FOR SLA-TRX SOLUTIONS. ................................................... 19 FIGURE 5. ACTIVATION AND DEACTIVATION TIME OF THE MODULATOR BLOCK VIA ON/OFF CONTROL PIN. ................................................. 21 FIGURE 6. ACTIVATION AND DEACTIVATION TIME OF THE TX VGA BLOCK VIA ON/OFF CONTROL PIN. ....................................................... 22 FIGURE 7. ACTIVATION AND DEACTIVATION TIME OF THE FEEDBACK MIXER BLOCK VIA ON/OFF CONTROL PIN. ............................................ 22 FIGURE 8. BLOCK DIAGRAM FOR ADAPTIVE ENERGY EFFICIENT PA FOR MACRO BS. ................................................................................ 23 FIGURE 9. DEACTIVATION (A) AND ACTIVATION (B) BEHAVIOUR OF THE AEEPA. .................................................................................... 24 FIGURE 10. POWER CONSUMPTION OF AN AEEPA WITH 20 W MAXIMUM AVERAGE OUTPUT POWER. ...................................................... 24 FIGURE 11. POWER EFFICIENCY OF AN AEEPA WITH 20 W MAXIMUM AVERAGE OUTPUT POWER. ........................................................... 25 FIGURE 12. POWER SUPPLY SYSTEM OF THE AEEPA FOR MACRO BASE STATION COMPOSED OF 2 MAIN BLOCKS: THE ISOLATED STAGE AND THE

APSU. ....................................................................................................................................................................... 26 FIGURE 13. STAGES 2 AND 3 ARCHITECTURE. .................................................................................................................................. 27 FIGURE 14. EFFICIENCY RESULTS OF STAGES 2 AND 3. ....................................................................................................................... 27 FIGURE 15. APSU POWER DISSIPATION FOR STAGES 2 AND 3. ........................................................................................................... 28 FIGURE 16. POWER CONSUMPTION OF SLA-TRX SYSTEM WITH 20 W MAXIMUM AVERAGE OUTPUT POWER. ............................................. 29 FIGURE 17. SIMPLIFIED BLOCK DIAGRAM OF A SMALL-CELL BASE STATION (A) AND THE BS POWER CONSUMPTION BREAKDOWN FOR DIFFERENT

CELL-SIZES (B). ............................................................................................................................................................. 31 FIGURE 18. POWER CONSUMPTION OF A 2X2 MIMO PICO-CELL BASEBAND PROCESSOR OVER THE SIGNAL LOAD. ........................................ 34 FIGURE 19. SMALL-CELL RF TRANSCEIVER BLOCK DIAGRAM BASED ON A DIRECT-CONVERSION ARCHITECTURE. THE ENERGY EFFICIENCY

OPTIMIZATION TIMESCALE OF THE ENERGY FLEXIBLE SUB-COMPONENTS ARE INDICATED ON AN INDICATIVE TIMESCALE. ..................... 35 FIGURE 20. SMALL-CELL RF TRANSCEIVER PICTOGRAM. .................................................................................................................... 35 FIGURE 21. POWER CONSUMPTION OF A 2X2 MIMO PICO-CELL BASE-STATION RF TRANSCEIVER OVER THE TRAFFIC LOAD. ........................... 36 FIGURE 22. BLOCK DIAGRAM FOR ADAPTIVE ENERGY EFFICIENT PA FOR SMALL-CELL BS. ........................................................................ 37 FIGURE 23. PHOTOGRAPH OF ADAPTIVE ENERGY EFFICIENT PA FOR SMALL-CELL BS. ............................................................................. 38 FIGURE 24. POWER CONSUMPTION OF 0.25W ADAPTIVE ENERGY EFFICIENT PA FOR SMALL-CELL BS. ...................................................... 38 FIGURE 25. PAE OF 0.25W ADAPTIVE ENERGY EFFICIENT PA FOR SMALL-CELL BS................................................................................ 39 FIGURE 26. LOAD ADAPTIVE PA BLOCK DIAGRAM. ........................................................................................................................... 40 FIGURE 27. PHOTOGRAPH OF THE LOAD ADAPTIVE PA: BOARD (A) AND CHIP-ON-BOARD ASSEMBLY (B). .................................................. 40 FIGURE 28. LOAD ADAPTIVE PA: (A) POWER CONSUMPTION (PDC) AND (B) PAE VS OUTPUT POWER (POUT) FOR FIXED MATCHING (BLUE

CURVE) AND OPTIMIZED MATCHING (RED CURVE) ................................................................................................................ 41 FIGURE 29. COMPARISON BETWEEN STANDARD (A) AND PROPOSED (B) RF FRONT-END ARCHITECTURES WITH DIFFERENT PLACES OF THE

DUPLEXING FUNCTION. .................................................................................................................................................. 42 FIGURE 30. PROTOTYPE OF A LOW-LOSS ANTENNA INTERFACE FOR SMALL-CELL BS. ............................................................................... 43 FIGURE 31. MULTILAYER ANTENNA PROFILE. .................................................................................................................................. 45 FIGURE 32. THE 4 METAL LAYERS OF THE ARRAY CONFIGURATION. ...................................................................................................... 46 FIGURE 33. ARRAY PROTOTYPE BEFORE ASSEMBLING (A) AND AFTER ASSEMBLING (B). ............................................................................ 46 FIGURE 34. EXPERIMENTAL INPUT REFLECTION COEFFICIENT OF THE ANTENNA ARRAY. ............................................................................ 47 FIGURE 35. EXPERIMENTAL RADIATION PATTERN OF THE ANTENNA ARRAY (1.95GHZ). .......................................................................... 47 FIGURE 36. EXPERIMENTAL EFFICIENCY OF THE ANTENNA ARRAY. ........................................................................................................ 47 FIGURE 37. ILLUSTRATION OF LONG DTX. ...................................................................................................................................... 50 FIGURE 38. TIMED UE CELL SEARCH RELYING ON A LOW ACCURACY UE TIME. ACTIVE PERIODS ARE SHIFTED DIFFERENTLY FOR EACH FREQUENCY.

UE CELL SEARCH WINDOWS FOR 2,10 ,, fff AND 3f ARE SHOWN IN RED, YELLOW, GREEN, AND BLUE. .................................... 52

FIGURE 39. GLOBAL ENERGY CONSUMPTION PER AREA-TIME UNIT AND PER BIT IN HIGH TRAFFIC DENSITY NETWORK, USING

DTXMicroDTXShort PP __ . ..................................................................................................................................... 53

FIGURE 40. PA CHARACTERISTICS OVER RF OUTPUT, DEPENDING ON OPERATION POINT SETTINGS (OPI WITH I=1,…,8). .............................. 54 FIGURE 41. RESOURCE UTILISATION PER SYMBOL AT 40% USER LOAD: (A) BW ADAPTATION, (B) MICRO DTX. ............................................ 54 FIGURE 42. DAILY POWER CONSUMPTION OF A CELL FOR SOTA AND IMPROVEMENTS BY THE PA ONLY BASED ON DIFFERENT SCHEDULING

STRATEGIES IN DENSE URBAN SCENARIO (A) AND RURAL SCENARIO (B). .................................................................................... 55 FIGURE 43. DAILY ENERGY SAVINGS FOR DIFFERENT SCHEDULING STRATEGIES BY PA ONLY (A), BY PA AND ALL OTHER COMPONENTS (B). ......... 55

EARTH PROJECT

EARTH_WP4_D4.3 9 / 121

FIGURE 44. ENERGY EFFICIENT SCHEDULING ALGORITHM: SEND PACKETS WHEN HIGH MCS CAN BE USED AND SLEEP THE REST OF THE TIME. ..... 56 FIGURE 45. ENERGY PER INFORMATION BIT FOR ALL THE CQI VALUES. ................................................................................................. 57 FIGURE 46. CQI REPORTING FOR A RESOURCE BLOCK FOR THREE UES. ................................................................................................. 58 FIGURE 47. INSTANTANEOUS AND AVERAGE ENERGY CONSUMPTION PER INFORMATION BIT FOR THE PROPOSED APPROACH WITH DIFFERENT CQI

THRESHOLDS COMPARED TO THE NO-THRESHOLD AND THE LOW-POWER APPROACHES. .............................................................. 58 FIGURE 48. RECONFIGURABLE AND ADAPTIVE BEAMFORMING. .......................................................................................................... 61 FIGURE 49. EXAMPLE OF SYSTEM LAYOUT WITH RECONFIGURABLE ANTENNAS (A) AND CONVENTIONAL ANTENNAS (B). ................................. 62 FIGURE 50. POWER CONSUMPTION FOR RAS AND REF SYSTEMS AS A FUNCTION OF OFFERED TRAFFIC. ...................................................... 63 FIGURE 51. RELATIVE POWER CONSUMPTION FOR RAS VERSUS REF SYSTEMS AS A FUNCTION OF OFFERED TRAFFIC. ..................................... 63 FIGURE 52. RELATIVE POWER CONSUMPTION FOR RAS VERSUS REF SYSTEMS OVER 24H FOR OFFERED TRAFFIC OF 92 MBPS/KM

2 (UPPER) AND

194 MBPS/KM2 (LOWER). ............................................................................................................................................. 64

FIGURE 53. POWER CONSUMPTION PER CELL FOR RAS AND REF SYSTEMS AS A FUNCTION OF NUMBER OF HOT SPOTS. .................................. 65 FIGURE 54. POWER CONSUMPTION PER CELL FOR RAS AND REF SYSTEMS AS A FUNCTION OF HOTSPOT PROBABILITY. ................................... 65 FIGURE 55. ENERGY CONSUMPTION INDICATOR OF AAS WITH REFERENCE TO THE BASELINE. ................................................................... 66 FIGURE 56. DAILY ECI OF AAS WITH REFERENCE TO THE BASELINE. ..................................................................................................... 67 FIGURE 57. ENERGY CONSUMPTION INDICATOR OF DOA BF WITH REFERENCE TO THE BASELINE. .............................................................. 67 FIGURE 58. DAILY ECI OF DOA BF WITH REFERENCE TO THE BASELINE. ............................................................................................... 68 FIGURE 59. ENERGY CONSUMPTION INDICATOR OF DOA BF AND AAS WITH REFERENCE TO THE BASELINE. ................................................ 68 FIGURE 60. DAILY ECI OF DOA BF AND AAS WITH REFERENCE TO THE BASELINE. .................................................................................. 69 FIGURE 61. POWER CONSUMPTION AS A FUNCTION OF SNR BY USING THE MOST ENERGY-EFFICIENT LINK ADAPTATION................................. 72 FIGURE 62. BS’S MEAN POWER LEVEL VERSUS SYSTEM TRAFFIC LOAD WITH CAP. .................................................................................. 74 FIGURE 63. DAILY BS’S MEAN POWER CONSUMPTION WITH PEAK DATA RATE 155 MBPS/KM

2. ............................................................... 75

FIGURE 64. ADAPTIVE MULTI-ANTENNA TRANSMISSION. ................................................................................................................... 76 FIGURE 65. A) LOGICAL AND PHYSICAL ANTENNA MUTING, B) LOGICAL ANTENNA PORT MERGING AND PHYSICAL ANTENNA PORT MUTING. ........ 76 FIGURE 66. PDCCH BLER VERSUS SINR, 4 TX AND 2 RX ANTENNA CONFIGURATION. ........................................................................... 77 FIGURE 67. PBCH BLER VERSUS SINR, 2 TX AND 2 RX ANTENNA CONFIGURATION............................................................................... 78 FIGURE 68. PDCCH BLER VERSUS SINR, 2 TX AND 2 RX ANTENNA CONFIGURATION. ........................................................................... 78 FIGURE 69. ENERGY CONSUMPTION PER AREA UNIT. ........................................................................................................................ 79 FIGURE 70. RESOURCE UTILIZATION AS FUNCTION OF SYSTEM THROUGHPUT. ........................................................................................ 79 FIGURE 71. POWER CONSUMPTION PER AREA UNIT AND POWER CONSUMPTION PER BIT. ........................................................................ 80 FIGURE 72. NORMALIZED TRANSMIT POWERS FOR THE CVQ ZF AND THE PROPOSED METHOD IN THE UNCORRELATED CHANNEL CASE. ............ 81 FIGURE 73. TOTAL PA DC POWER CONSUMPTION VERSUS PAPR BASED ON THE EARTH PA MODEL. ....................................................... 82 FIGURE 74. PERFORMANCE COMPARISON BETWEEN THE CONVENTIONAL AND THE PROPOSED METHOD. .................................................... 82 FIGURE 75. IMPACT ON LTE SIGNAL SPECTRUM (40 MS; 30% LOAD) AFTER PASSING THE AEEPA. ........................................................... 90 FIGURE 76. BLOCK DIAGRAM OF THE HETERODYNE RECEIVER FOR THE SLA-TRX IN MACRO BS. ................................................................ 91 FIGURE 77. PROTOTYPE OF THE DUAL CHAIN RECEIVER AS A PART OF THE SLA-TRX FOR MACRO BS. ......................................................... 92 FIGURE 78. PERFORMANCE AND DC POWER CONSUMPTION PARTITIONING OF THE RECEIVER. .................................................................. 93 FIGURE 79. BLOCK DIAGRAM OF THE SMALL SIGNAL SLA_TX FOR MACRO BS. ...................................................................................... 94 FIGURE 80. PROTOTYPE OF THE SMALL SIGNAL SLA_TX FOR MACRO BS. ............................................................................................. 95 FIGURE 81. PHOTOGRAPH OF THE ADAPTIVE ENERGY EFFICIENT PA PROTOTYPE FOR MACRO BS. ............................................................. 96 FIGURE 82. APSU POWER CHARACTERISTIC. .................................................................................................................................. 97 FIGURE 83. RETURN LOSS AND TX/RX ISOLATION FOR THE FULL BAND1 (A) AND DEDICATED UP/DL ANTENNAS (B). ................................... 98 FIGURE 84. ANTENNA STRUCTURE: LAYERS DESCRIPTION WITH MATERIAL (A), 3D VIEW OF THE SIMULATED STRUCTURE (B). .......................... 98 FIGURE 85. TWO DUPLEXING ANTENNAS STRUCTURES WITH SPATIAL DIVERSITY (MIMO CONFIGURATION). ................................................ 99 FIGURE 86. EXPERIMENTAL RELATIVE DIELECTRIC CONSTANT (ΕR) OF STYROFOAM™. ............................................................................ 100 FIGURE 87. EXPERIMENTAL LOSS TANGENT (TAN Δ) OF STYROFOAM™. .............................................................................................. 101 FIGURE 88. ANTENNA PROTOTYPES USED TO VALIDATE SOFTWARE TOOLS. ......................................................................................... 101 FIGURE 89. SINGLE ELEMENT PROTOTYPE BEFORE ASSEMBLING (A) AND AFTER ASSEMBLING (B). ............................................................ 102 FIGURE 90. EXPERIMENTAL INPUT REFLECTION COEFFICIENT OF THE SINGLE ELEMENT ANTENNA. ............................................................. 103 FIGURE 91. MEASUREMENT SETUPS OF S11, RADIATION PATTERN AND EFFICIENCY. .............................................................................. 103 FIGURE 92. EXPERIMENTAL FAR FIELD RADIATION PATTERN OF THE ANTENNA ARRAY (1.95 GHZ). .......................................................... 104 FIGURE 93. EXPERIMENTAL FAR FIELD RADIATION PATTERN OF THE ANTENNA ARRAY (2.14 GHZ). .......................................................... 104 FIGURE 94. SCHEMATIC CONCEPT OF ACTIVE ANTENNA SYSTEM. ...................................................................................................... 105 FIGURE 97. LINEARIZED EARTH POWER MODEL, YEAR 2012. THE CIRCLES AT RESOURCE UTILIZATION = 0 INDICATE POWER CONSUMPTION WHEN

SLEEP MODE IS ON. .................................................................................................................................................... 108

EARTH PROJECT

EARTH_WP4_D4.3 10 / 121

FIGURE 98. RADIATION PATTERN COMPARISON BETWEEN SINGLE DOA AND EIGEN BF ALGORITHMS. ...................................................... 111 FIGURE 99. SIMULATION MODEL. ............................................................................................................................................... 112 FIGURE 100. RADIATION PATTERNS, WITH 8 ANTENNA ELEMENTS..................................................................................................... 113 FIGURE 101. VARIATION OF POWER IMPROVEMENT AS A FUNCTION OF CELLS’ RADIUS. ......................................................................... 115 FIGURE 102. VARIATION OF POWER IMPROVEMENT AS A FUNCTION OF NUMBER OF USERS FOR LTE. ...................................................... 116 FIGURE 103. LTE DOWNLINK RADIO FRAME WITH 10 SUB-FRAMES SHOWING CELL SPECIFIC REFERENCE SYMBOLS (CSRS) FOR ONE ANTENNA

PORT, DOWNLINK CONTROL REGION (PDCCH) WITH A SIZE OF ONE OFDM SYMBOL, PRIMARY AND SECONDARY SYNCHRONIZATION

SIGNALS (PSS AND SSS), AND BROADCAST CHANNEL (BCH). ONLY MICRO DTX IS POSSIBLE. .................................................... 119 FIGURE 104. EXAMPLE OF A DOWNLINK RADIO FRAME IN LTE WITH 6 MBSFN SUB-FRAMES SHOWING DOWNLINK CONTROL REGION (PDCCH)

WITH A SIZE OF ONE OFDM SYMBOL, PRIMARY AND SECONDARY SYNCHRONIZATION SIGNALS (PSS AND SSS), AND BROADCAST

CHANNEL (BCH). ....................................................................................................................................................... 120 FIGURE 105. PROPOSED LTE DOWNLINK RADIO FRAME WITHOUT CSRS. SHORT DTX CAN BE APPLIED. ................................................... 121

EARTH PROJECT

EARTH_WP4_D4.3 11 / 121

List of Tables

TABLE 1. POWER CONSUMPTION BREAKDOWN FOR DUAL MODE SMALL SIGNAL TRANSMITTER IN MACRO-CELL BS. ....................................... 21 TABLE 2. SCALING EXPONENTS FOR THE BASEBAND SUB-COMPONENTS IN DOWNLINK .............................................................................. 33 TABLE 3. LOSS FOR THE DIFFERENT COMPONENTS AND ENERGY EFFICIENCY OF THE PROPOSED CONCEPT ..................................................... 42 TABLE 4. MAIN EXPERIMENTAL CHARACTERISTICS OF THE ANTENNA ARRAY ........................................................................................... 48 TABLE 5. RADIATION EFFICIENCY OF ARRAYS OF PATCHES WITH DIFFERENT NUMBER OF ELEMENTS (N) ....................................................... 48 TABLE 6. ENERGY PER BIT REDUCTION AND AVERAGE PACKET DELAY FOR DIFFERENT CQI THRESHOLDS. ....................................................... 59 TABLE 7. RESULTS OF SISO AND MIMO 2X2 AND4X4 MODES FOR MINIMAL POWER CONSUMPTION AS FUNCTION OF SNR AND DUTY-CYCLING

FOR A GIVEN THROUGHPUT. ............................................................................................................................................ 71 TABLE 8. DAILY ENERGY SAVINGS OF ADAPTIVE MIMO SCHEMES WHEN PEAK DATA RATE IS 155 MBPS/KM

2 .............................................. 75

TABLE 9. SIMULATION PARAMETERS .............................................................................................................................................. 77 TABLE 10. COMPARISON OF METHODS IN TERMS OF TOTAL PA DC POWER CONSUMPTION. .................................................................... 83 TABLE 11. CHEAP SUBSTRATE CHARACTERISTICS AND ANTENNA EFFICIENCY (AT 2 GHZ). ........................................................................ 100 TABLE 12. ANTENNA PARAMETER RANGES. ................................................................................................................................... 108 TABLE 13. MODELING OF, CLUSTERED, OFFERED TRAFFIC. ................................................................................................................ 109 TABLE 14. SIMULATION SET FOR GENERAL SCENARIOS. .................................................................................................................... 114

EARTH PROJECT

EARTH_WP4_D4.3 12 / 121

Acronyms and Abbreviations

3G AA AAS ACLR AEEPA AMC

Third Generation Active Antenna Active Antenna System Adjacent Channel Leakage Ratio Adaptive Energy Efficient PA Adaptive Modulation and Coding

ANT ASIP

Antenna Application Specific Instruction Processors

BAW Bulk Acoustic Wave filter BCH Broadcast Channel BB BF BS CAP CD CDF

Baseband Beamforming Base Station Capacity adaptation Component Deactivation Cumulative Distribution Function

CMOS CoMP CQI CSI

Complementary Metal Oxide Semiconductor Coordinated Multi-Point Channel Quality Indicator Channel State Information

CSRS Cell Specific Reference Signal CW DC DL DoA DPD

Continuous Wave Direct Current Down-link Direction of Arrival Digital Pre Distortion

DTX Discontinuous Transmission E3F EARTH

Energy Efficiency Evaluation Framework Energy Aware Radio and neTwork tecHnologies

ECI Energy Consumption Indicator EDF Earliest Deadline First EA EE

Energy Adaptive Energy Efficiency

EIRP Equivalent Isotropic Radiated Power EPA Extended Pedestrian A ETU Extended Typical Urban EVA Extended Vehicular A EVM FAMUP

Error Vector Magnitude Frequency Adaptive Multi-User Priority

FBAR FD FDD

Film Bulk Acoustic Resonator Frequency Domain Frequency Division Duplexing

FDPS Frequency Dependent Packet Scheduling FEC Forward Error Correction FPGA Field Programmable Gate Array GPIO General Purpose Input/Output GSM Global System for Mobile communications

EARTH PROJECT

EARTH_WP4_D4.3 13 / 121

HARQ HPA

Hybrid Automatic Repeat Request High Power Amplifier

HSPA High Speed Packet Access IQ Q LNA LO LTE MCS MCI

Real and Imaginary Quality factor Low Noise Amplifier Local Oscillator Long Term Evolution Modulation and Coding Scheme Maximum Channel to Interference ratio

MIMO Multiple-Input-Multiple-Output MMSE Minimmum Mean-Square Error MN MU NACK NHS OFDMA OPA PA

Matching Network Multi User Negative ACKnowledge Number of Hotspots Orthogonal Frequency Division Multiple Access Operating Point Adjustment Power Amplifier

PAPR Peak-to-Average Power Ratio PBCH Physical Broadcast CHannel PBO PDCCH

Power Back-Off Physical Downlink Control Channel

PER Packet Error Rate PMI Precoding Matrix Indicator PF Proportional Fair PHS PHY

Probability of Hotspots PHYsical

PLL PRB

Phase Lock Loop Physical Resource Block

PSS Primary Synchronization Signals RAN Radio Access Network RAS RF RI RoF Rx QoS

Reconfigurable Antenna System Radio Frequency Rank Indicator Radio over Fiber Receive Quality-of-Service

SAW Surface Acoustic Wave filter SDM Spatial Division Multiplexing SiNAD Signal-to-Noise And Distortion ratio SIMO SINR

Single-Input-Multiple-Output Signal-to-Interference-plus-Noise Ratio

SISO SNR SOTA

Single-Input-Single-Output Signal-to-Noise-Ratio State-Of-The-Art

SOI SPI SRS

Silicon-on-Insulator Serial Peripheral Interface Bus Sounding Reference Signal

SS Small Signal

EARTH PROJECT

EARTH_WP4_D4.3 14 / 121

SSS SU TDD

Secondary Synchronization Signals Single User Time Division Duplexing

TMN TRX Tx UE UL UMa UMi

Tunable Matching Network Transceiver Transmit User Equipment Up-Link Urban Macro Urban Micro

VCO VM WCDMA

Voltage Control Oscillator Vector Modulator Wideband Code Division Multiple Access

EARTH PROJECT

EARTH_WP4_D4.3 15 / 121

1. INTRODUCTION

The EARTH project aims to address the global environmental challenge by investigating and proposing effective mechanisms to drastically reduce energy consumption of mobile broadband communication systems by a factor of at least 50% compared with the current ones, without compromising users’ perceived quality of service and system capacity [EARTH-Leaf].

Most of the equipment of the current systems have static configuration that can support the maximum traffic load. The load variation of the daily traffic is very significant as shown in FIGURE 1. This characteristic is based on 15 minutes average and is part of the reference scenario defined for performance evaluation in the project [EARTH-D2.3]. The maximum traffic load values in the evening hours differ due to the deployment achieving for example 60% in dense urban or only 30% in suburban scenarios. This shows that in average the equipment is operating significantly below its maximum capacity with significantly reduced energy efficiency, defined as metric in [EARTH-D2.4].

FIGURE 1. 24 hours traffic load profile.

Two basic strategies, “duty-cycling in time” and “in frequency”, allow to exploit the power saving potential provided by low and medium traffic load. With the first one, time slots of no data transmission are maximized enabling for efficient application of “sleep modes” in base station components, while with the second one, the number of resource elements is limited, which limits the RF transmission power and allows thereby for optimizing the component operation for increased energy efficiency.

‘Adaptability’ is the key concept met in most of the solutions for Green Radios presented in this document. This means that hardware or software solutions can decide and adjust their configuration following the traffic load variation in order to improve their energy efficiency. While radio interface techniques are mainly based on the average traffic load variation and act on several time scales, the component solutions consider the short term variation of transmitted data and take benefit from the specifics of LTE signals, which are considered as reference in the project.

All presented green radio technologies focus on base stations for 4G broadband cellular networks (LTE). They deal with several domains like hardware components and equipment, radio transmission or adequate interface technologies and multiple antennas techniques. But all of them are quantitatively analysed as regards their energy efficiency potentials and/or their adaptability to dynamic systems for macro- and small-cell base stations. Consequently for each solution, the areas of application and application conditions are depicted according to deployment scenarios, optimization time scale and traffic load.

For mobile base stations, new power saving potentials have been identified to be exploited in addition to the continuously decrease of power consumption due to the technological evolution observed over the years and

EARTH PROJECT

EARTH_WP4_D4.3 16 / 121

due to the aim of minimizing the power consumption for maximum signal load (instantaneous level of the transmitted signal related to the maximum RF output power defined for transmission). New power-saving features take benefit by the non-uniform load distribution over the day and the short-term signal characteristics. Investigating the BS radio equipment aspects, a distinction is done between components for macro-cell and small-cell BS due to the different origins and their relative weight of power consumption. For macro-cell BS, the application of a signal load adaptive transceiver exploits the power saving potential in medium and low loads according to the optimization of operating points and the deactivation of certain components (sleep modes). For small-cell BS, all the components from baseband engine to antenna interface have been designed and optimized to be energy efficient and/or energy flexible. Considering the printed antenna array BS equipment, the use of a low loss foam substrate is studied for improving energy efficiency.

A further group of solutions deals with radio interface technologies that exploit the interaction between higher protocol layers and the hardware to improve the energy efficiency. Scheduling solutions operating in time and frequency domain and energy opportunistic scheduling for exploiting good channel conditions, enables to maximize the power saving due to adaptive hardware components.

Multiple antenna solutions have been analysed for improving the link budget in a cell. Beamforming considering cell specific or user specific approaches allow for power saving especially when using active antenna systems. MIMO solutions, that use the channel properties to support high throughput, have been analysed for assuring their energy efficient operation considering the selection of the number of active transmit antennas and resource allocation strategies or a downscaling to SISO/SIMO when possible.

Each presented radio technology is structured to express the motivation, the solution concept, the expected benefits and its relation to other EARTH solutions. They serve for the evaluation of “Integrated Solutions” done on project level and presented in [EARTH-D6.4].

EARTH PROJECT

EARTH_WP4_D4.3 17 / 121

2. TRANSCEIVERS FOR MACRO-CELL BASE STATIONS

Solutions for increasing the energy efficiency of transceivers exploit the potential to reduce the power consumption in low and medium load situations. Considering in EARTH LTE as the reference system [EARTH-D2.3] for defining and evaluating the proposed concepts and solutions to increase the energy efficiency in wireless mobile communication systems, the characteristics of LTE signals have been analysed for translating the averaged traffic load shown in FIGURE 1, into short term signal characteristics.

A principal diagram with short-term signal characteristics is presented in FIGURE 2 showing the signal load for different transmission states. In time slots of data transmission the maximum (100%) signal load is achieved but even lower signal levels are possible. During data transmission, user information and physical signals and channels (CSRS, PSS, SSS and BCH) are scheduled. In time slots of no data, only signalling (physical signals and channels) is transmitted consisting in symbols of different levels lower than 100% (i.e. 27% for CSRS-BCH, 17% for CSRS only, and 12% for PSS|SSS signals or BCH channel, in 10 MHz bandwidth configurations) and even empty symbols (0% signal load). Their level depends on the bandwidth configuration.

FIGURE 2. Relative RF output power level according to the LTE frame structure.

The short-term characteristics of LTE signals enable reduction in DC power consumption according to the presented frame structure. The empty symbols open the opportunity to apply sleep modes on dedicated components in the equipment during these short time slots providing a certain reduction of the power consumed during operation. The different signal levels below 100% gives the possibility of applying a further solution to reduce the power consumption by adapting the operating points of some components to optimize the energy efficiency.

For exploiting the identified potential for energy efficiency improvement, new techniques for reducing the energy consumption have been defined and evaluated for different transceiver components. They show key features of power efficient transceiver systems [FeBBWP10], and can be used for different wireless standards by applying them in specific ways.

The power reduction techniques proposed for macro base station transceivers are based on the deactivation of components and the adjustment of their operating points in medium and low load situations. These

No data

CSRS + BCHDATA MODE

CSRS EMPTY SYMBOL

80%

100%

Signal Load

27%17%12%

Time(OFDMA symbol resolution)

DataData

Data

BCH

PSS | SSS

EARTH PROJECT

EARTH_WP4_D4.3 18 / 121

techniques are applied to different transceiver components which define the Signal Load Adaptive Transceiver System (SLA-TRX) and integrate the following transceiver blocks (see FIGURE 3):

The Digital Signal Processing and Control (DSPC) unit shows the digital interface to the baseband (BB) for receiving and transmitting BB-signals and a system control interface (SC) to layer 1 and 2 of the base station. The DSPC is controlling the adaptation of analogue transceiver components for minimizing their power consumption and provides the required control signals (CS).

The Small Signal RF-Transceiver (SS RF-TRX) does the signal conversions between the digital and analogue domains and the frequency conversions between BB and RF frequencies. It allows the deactivation of some components, controlled by the DSPC for minimizing the power consumption.

The Adaptive Energy Efficient Power Amplifier (AEEPA) amplifies the signal to be transmitted and provides the output power level requested for the sector to be operated. It allows the adaptation or deactivation of power amplifier stages in correlation with the signal level.

The Adaptive Power Supply Unit (APSU) delivers the supply voltages and currents required by the TRX system. It supports the reconfiguration of voltages supplied to the AEEPA.

FIGURE 3. Block diagram of a transceiver system for macro-cell base stations.

The main idea of the SLA-TRX is to adapt the operation of the analogue hardware components to the signal load for minimizing the consumed power. A concerted definition of the functionality required in the different transceiver blocks enables the realisation of operating point adjustment (OPA) and component deactivation (CD) in strong correlation to the variation of the signal level. OPA and CD show the key features for an energy efficient transceiver operation in macro BSs.

For operating these features, the DSPC block is receiving control signal information from layers 1 and 2 of the BS and is analysing the characteristics of the BB signals in order to adapt the signal processing and to control the adaptation process of the analogue hardware components mentioned above. The component adaptation is to be done synchronized with the LTE signal pattern.

The CD shall be applied in time slots of empty symbols which show a minimum duration of two symbols (ca. 140 µs). For exploiting the potential on energy saving during such short time slots, a transition time lower than 17 µs is envisaged, which is the upper limit of the transmitter transient period defined in [LTE-SPEC].

For OPA two application strategies are possible. One considers a fast adjustment of the operating points (fast OPA) of PA stages with transition times shorter than 30 µs. It would allow the application for every arbitrary variation of signal level but it is challenging for implementation, as a sophisticated biasing circuitry is required. The second strategy, a slow OPA, allows for a much longer transition time. It is easier to implement and less susceptible regarding distortions during the transition as the operating point modifications happen slower, allowing for a more accurate operation of the DSPC algorithms. This strategy requires a more careful

DSPC SS RF-TRX

TX-DigitalModule

RX-DigitalModule

TX-ConversionModule

RX-ConversionModule

AEEPA

LNA

AntennaRX-BB

TX-BB

APSU

CSSC

EARTH PROJECT

EARTH_WP4_D4.3 19 / 121

coordination of the scheduling algorithms and the component control as the information required for component adaptation must be available a longer time in advance as the transition time is longer.

The concepts proposed for DSPC, SS RF-TRX, AEEPA and APSU are defined and presented in the following sections of this document. The power performance of some transceiver blocks and the complete SLA-TRX system is defined by power characteristics showing the consumed power related to the signal load. The expected benefit is determined by comparing these characteristics with a state-of-the-art characteristic, which do not consider the proposed concepts.

2.1. DSPC - SOLUTIONS FOR SIGNAL LOAD ADAPTIVE TRANSCEIVERS

The basic role of the DSPC (digital transceiver part) in the transmit part is to adapt the signals for transmission to the properties of SS RF-TRX and power amplifier (PA) assuring the quality of signals to meet the standard specifications. Such features are typically signal synthesis for multicarrier applications, digital predistortion (DPD) and crest factor reduction for enabling a higher maximum power efficiency of the PA, and IQ imbalance compensation to support direct frequency conversion architectures. A typical DSPC block is presented in [EARTH-D4.1].

Supporting the SLA-TRX concept presented above, the DSPC provides additional functionality by controlling the component reconfiguration of the other transceiver blocks, by adapting the signal conditioning algorithms to the reconfiguration procedures and by assuring the quality of transmitted signals which can be impacted by imperfections of the reconfigured analogue components. Basically three functional blocks are included in the signal conditioning and control (SCC) block (see FIGURE 4) to enable the required features:

The Sample Based Signal Scan (SBS) executes the required analysis of the received BB signals. The achieved information is used in conjunction with layers´ 1 or 2 information for defining the appropriate action to adapt the transceiver operation to the instantaneous signal load for minimized power consumption. It assures a component reconfiguration synchronized to the signal stream.

The Adaptive Clipping (ACLP) adjusts the clipping threshold to the signal load. This leads to a reduction of peak values in correlation to the reduced average signal level allowing operating point adaptation of PAs for minimized power consumption.

The Signal Quality Check (SQC) is analysing the quality of the signal TXS prepared for transmission, assuring that the transmitted signals meet the standard specifications.

FIGURE 4. Architecture of a SCC block of the digital transmit block for SLA-TRX solutions.

ACLP—Adaptive clippingBBS—Baseband signalSS—Signal synthesis

SBS—Sample based signal scanSC—System control

SCC—Signal conditioning and controlSQC—Signal quality checkSVC—Supply voltage control

SVS—Supply voltage switchTX—Transmitter

TXS—Transmission signal

SS ACLP

SCC

SC

SBS SQC

SVC

TX-BBS

TX-BBS

. .

SVS

TXS

EARTH PROJECT

EARTH_WP4_D4.3 20 / 121

The DSPC supports the two features defined for energy efficiency optimization in the SLA-TRX: the OPA of analogue components at medium and low signal loads and the CD in time slots without signal transmission.

For enabling the operating point adjustment, the following DSPC operations are defined:

The optimum clipping threshold is determined and applied for minimizing the peak values of the signal. This allows the optimal adjustment of the PA operating point for maximum power efficiency.

Control information is provided to the components to be adjusted, which defines a set of positions of new operating parameters available for following reconfigurations.

Reconfiguration signals are transmitted which initiate the reconfiguration, selects the position of the new operating parameters and synchronizes the adaptation procedure with the transmitted signal.

The deactivation of components is managed by the DSPC as following:

The time periods without signal transmission are determined based on SC information and SBS signal analysis.

Control information is provided to components designated for deactivation and reactivation.

Starting signals are transmitted to initiate the deactivation or reactivation procedures synchronized with the transmitted signal.

For enabling the above operations, the DSPC provides control signals to the components designated for adaptation. Starting signals for initiating the reconfiguration are correlated with the signal variation for avoiding an inacceptable impact on the signal quality. The determination of the starting time considers parameters defined by the hardware implementation like time delays and transition slopes.

By applying the features proposed for power reduction, some analogue components show variable behaviour during adaptation procedures by changing their parameters for different operating points. As this impacts the algorithms running in the DSPC block for mitigation of analogue imperfections, e.g. digital pre-distortion, these algorithms must be adapted for assuring their stable operation.

A significant impact on the signal quality can be caused by a gain variation of the power amplifier when the operating point is changed. During the reconfiguration process the variation of the gain within single LTE symbols leads to increased noise outside the signal band with the risk of violating the spectrum emission mask. When modifying the PA operating point individually for successive LTE symbols, the gain variation in the transceiver implies a path loss variation for successive symbols in the user equipment leading to deviations in the channel estimation results and to increased Error Vector Magnitude (EVM) values for the signal received by the mobile equipment which can exceed about 10%. To minimize this impact, a compensation of the gain variation is required for the DSPC algorithms by allowing a residual gain variation in the range of 10%. Then the standard requirement would be met regarding the spectrum emission mask and the EVM. See appendix section 10.1.1 for more details.

2.2. DUAL POWER MODE SMALL SIGNAL RF TRANSCEIVER

To reduce the power consumption in the SS RF-TRX during low and medium load situations, the deactivation of such components in the transmit path has been proposed, which allow for fast reactivation without long settling times or a need of synchronisation. In addition, they have to show a dedicated on/off control input. Such feature has not been proposed for the receive path, as in the considered scenario of very short sleeping periods, it is not possible to get knowledge about time periods when now signal is to be received.

A small signal transmitter has been designed, which can be configured by the DSPC according to the component control algorithms which enable the deactivation of transmitter blocks when there is no symbol to be transmitted. The IQ-modulator, the variable gain amplifier (VGA) and the mixer in the feedback path can be

EARTH PROJECT

EARTH_WP4_D4.3 21 / 121

deactivated. The power consumption of these components during operation (ON) as well as during the deactivation state (OFF) is shown in TABLE 1, where the complete power consumption breakdown of the dual mode SS transmitter is given.

TABLE 1. Power consumption breakdown for dual mode small signal transmitter in macro-cell BS.

In order to support the CD feature, the small signal blocks considered for deactivation have to show a transition time shorter than the 17 µs limit defined in [LTE-SPEC] for the transmitter transient period. The evaluation done on a prototype show short enough transition times as presented in FIGURE 5, FIGURE 6, and FIGURE 7. “Power_on” marks the transition during the activation procedure while the transitions for the deactivation procedure are marked with “Power_off”.

FIGURE 5. Activation and deactivation time of the modulator block via ON/OFF control pin.

Block Name: Vcc (V): Icc_ON (mA): Pdc (W)_ON Icc_OFF (mA):

IQ Modulator 5 185 0.925 7

VGA (single block) 5 115 0.575 15

VCO&Synth_frwd 3.3 90 0.297 not applicable

VCO&Synth_fb 3.3 90 0.297 not applicable

Feedback Mixer 5 200 1 40

Clock Clock Distributor 3.3 250 0.825 not applicable

1.8 640 1.152 not applicable

3.3 43 0.1419 not applicable

ADC 1.8 280 0.504 not applicable

Icc_total_5V (mA): 500 Pdc_5V (W): 2.50 Pdc_total (W):

Icc_total_3V3(mA): 473 Pdc_3V3 (W): 1.56

Icc_total_1V8 (mA): 920 Pdc_1V8 (W): 1.66

Icc_total_5V (mA): 62 Pdc_5V (W): 0.31 Pdc_total (W):

Icc_total_3V3(mA): 473 Pdc_3V3 (W): 1.56

Icc_total_1V8 (mA): 920 Pdc_1V8 (W): 1.66

RF

DA

C

and

AD

C Dual DAC

Idc and Pdc in

ACTIVE mode5.72

Idc and Pdc in

DEACTIVATED

mode 3.53

Power_on:<1µsec Power_off:<500nsec

1.1µs/1.1µs/

EARTH PROJECT

EARTH_WP4_D4.3 22 / 121

FIGURE 6. Activation and deactivation time of the TX VGA block via ON/OFF control pin.

FIGURE 7. Activation and deactivation time of the feedback mixer block via ON/OFF control pin.

As presented in TABLE 1, the CD feature applied for the dual power mode small signal transceiver provides around 2.2 W of DC power saving, which corresponds to a 38% reduction of the consumed power compared to the SOTA solution, while the reconfiguration can be achieved during time duration shorter than 3.5 µs for activation and shorter than 0.5 µs for deactivation. These performance figures meet the requirements for applying the CD feature and are considered for the performance evaluation of the whole SLA-TRX.

Although the component deactivation concept cannot be applied to the receiver chain, the power consumption of a respective circuit is to be considered for the energy efficiency performance evaluation of the whole SLA-TRX with a value of 7.3 W.

A dual power mode SS RF-Transceiver has been designed and realized for performance evaluation. The circuitries for transmitter and receiver are presented in the appendix (10.1.2).

Power_on: <3.5µsecPower_off: <70nsec

Power_on:<2µsec

Power_off:<100nsec

EARTH PROJECT

EARTH_WP4_D4.3 23 / 121

2.3. ADAPTIVE ENERGY EFFICIENT POWER AMPLIFIER

Supporting the SLA-TRX concept, an Adaptive Energy Efficient PA has been defined for minimizing the power consumption in a macro BS at arbitrary load levels. This PA solution supports the key features of the SLA-TRX concept: Operating Point Adjustment and Component Deactivation.

In order to evaluate these concepts, an AEEPA has been designed for LTE in the frequency band from 2.11 to 2.17 GHz with maximum average output power of 20 W. The maximum peak output power is 120 W for supporting up to 8 dB PAPR required for LTE signals.

The AEEPA shows different stages:

The Pre-DRIVER is supplied with fixed bias, supporting the CD feature only.

The DRIVER enables the OPA and CD features. This stage can be adapted to different maximum RF output power levels, optimizing its power efficiency for arbitrary signal load and can be switched off during empty LTE symbols.

The HPA (high power amplifier) is a Doherty amplifier of high power efficiency and enables the OPA and CD features as the previous stage.

The block diagram of the AEEPA for macro BS is presented below in FIGURE 8.

FIGURE 8. Block diagram for Adaptive Energy Efficient PA for macro BS.

An AEEPA prototype has been realized for proving the concept and for evaluating the performance. This hardware development is presented in the appendix in section 10.1.3. It shows RF and bias connections with different modules in the SLA-TRX: DSPC, SS RF-TRX and APSU according to the block diagram shown in FIGURE 3.

Applying the CD feature to the AEEPA, the PA stages are deactivated based on control signals received from the DSPC. For operating the OPA feature, the PA stages 2 and 3 use the control signals received from the DSPC and the drain voltages adjusted by the APSU as presented below in the section 2.4.

The deactivation and activation speed of the AEEPA has been evaluated with the hardware prototype showing the transient behaviour presented in FIGURE 9. It shows the PA output signal of a continuous RF signal at 2.14 GHz during the deactivation respectively the activation procedure of the AEEPA. The deactivation occurs

Pre-DRIVER DRIVER HPA

Variable BIAS

Variable BIAS Fixed BIAS

POWER AMPLIFIER

Supply Voltage V3

Supply Voltage V1

Supply Voltage V2

BIAS Control

BIAS Control

EARTH PROJECT

EARTH_WP4_D4.3 24 / 121

within 3 µs while the activation takes about 10 µs. This allows the component deactivation during empty LTE symbols and fulfils the 17 µs of maximum transient period requested by the LTE-TDD specification [LTE-SPEC].

a) b)

FIGURE 9. Deactivation (a) and activation (b) behaviour of the AEEPA.

For OPA, a slow transition has been considered for the prototype realization providing a suitable performance regarding oscillations and noise at the drain voltage defined at the interface between the AEEPA and the APSU. Fast OPA would require a sophisticated development of this interface and was not in the focus of this activity.

The AEEPA allows the adjustment of arbitrary operating points in contrast to the APSU which is defined for 8 discrete drain voltage levels. These are considered for defining the 8 operating points for the SLA-TRX as a whole. According to this, the AEEPA has been evaluated for 8 different operating points defined with 1dB steps referred to the maximum level of 20 W average output power. The adjustment is performed by modifying the drain voltage in the driver respectively the gate voltages and the drain voltages in the Doherty amplifier. The drain voltage changes are performed in the APSU, while the gate voltage changes are performed in the AEEPA. The reconfiguration procedure is coordinated by the DSPC, controlling the AEEPA and the APSU in a harmonized way.

The power performance of the AEEPA has been analyzed by considering the features CD and OPA and is presented for 8 operating points in FIGURE 10.

FIGURE 10. Power consumption of an AEEPA with 20 W maximum average output power.

EARTH PROJECT

EARTH_WP4_D4.3 25 / 121

Eight different operating points are presented showing the performance for different levels of signal load below their individual maximum levels. Following the 1 dB steps, the operating points are defined for different signal load ranges: 100-79% for OP1, 79-63% for OP2, 63-50% for OP3, 50-39% for OP4, 39-31% for OP5, 31-25% for OP6, 25-20% for OP7 and 20-0% for OP8.

Without applying OPA, the AEEPA is operating in OP1. Using this technique, the power consumption reductions are around 6% for OP2, 9% for OP3, 13% for OP4, 18% for OP5, 23% for OP6, 25% for OP7 and 35% for OP8. While the power consumption reduction for CD reaches up to 94%.

OPA and CD can be applied individually or in a combined way. They are mainly beneficial at low and medium traffic load and allow for significant energy saving, improving the energy efficiency in macro BS.

Following the power efficiency metric [EARTH-D2.4], the evaluated performance is presented as power efficiency in FIGURE 11 for the eight operating points. The efficiency improvement is around 7% for OP2, 12% for OP3, 16% for OP4, 21% for OP5, 25% for OP6, 28% for OP7 and 34% for OP8.

FIGURE 11. Power efficiency of an AEEPA with 20 W maximum average output power.

The features presented with the AEEPA show promising solution for energy efficient PA operation in macro-cell base stations. The energy efficiency performance is used as input for the energy efficiency performance analysis of the SLA-TRX presented in section 2.5.

2.4. ADAPTIVE POWER SUPPLY UNIT

For the SLA-TRX a specific power supply unit is proposed to adjust the power supply voltage for the AEEPA, supporting the OPA feature.

The adaptive power supply unit (APSU) defined for the SLA-TRX includes the specific power supply dedicated to the AEEPA and an isolated stage to provide a protection to the batteries as presented in FIGURE 12. The APSU block is to be connected to the AEEPA, supporting OPA. The APSU contains 4 DC-DC converters to supply several stages of the PA:

EARTH PROJECT

EARTH_WP4_D4.3 26 / 121

The OPA feature is supported by “Stage 2” and “Stage 3” which supply the driver and the HPA of the AEEPA. The output supply voltages of these DC/DC converters are controlled by the DSPC and can be adjusted for eight operating points (OP1 to OP8).

The stage with 24 V output supplies the fans in the AEEPA.

The block “Stage 1” supplies the pre-driver of the AEEPA with 5 V.

OP1...OP8

48V supply

(6A needed max)

DC/DCLM5088

Fans Supply

Iso Stage DC/

DC

enable

enable

DC/DCLM5088

Stage 3

DC/DCLM5088

Stage 2

42V, 5.5 A

DSPC

(FSL)

APSU

SPI

PA3READY.TEMP3

PA2READY.TEMP2

Logic

Block

Stage 3

Logic

Block

Stage 2

State Select

State SelectA2B2C2

B3C3

A3

s

s

C B A State

0 0 0 OP1

0 0 1 OP2

0 1 0 OP3

0 1 1 OP4

1 0 0 OP5

1 0 1 OP6

1 1 0 OP7

1 1 1 OP8

P2_7

& P2_14

P2_8

& P2_15

P3_4

P3_2

P3_1

P3_5

P2_12

Reset Lines Isos

Reset sub-block

rs

rs

rs

Imax 4A

Imax 1AOP1...OP8

DC/DCLM5088

Stage 1

P2_1

P2_2

P2_4

P2_3 P2_5

P2_6

P2_9

P2_10 P2_11

rs

Average 1A

Max. 1.5 A

5V

24V1A average

1.5A peak

P2_13

24V, 1.9A max

FIGURE 12. Power Supply System of the AEEPA for macro base station composed of 2 main blocks: the isolated stage and the APSU.

The concepts and solutions used for the core of the APSU circuit, the Stages 2 and 3, enable the adaptation of the voltages supplied for the AEEPA. The operating points adjusted by Stage 2 and 3 can be defined individually. The eight operating points supported by the APSU, namely OP1 to OP8, can provide voltage levels between 28 V and 12 V with the maximum current pointed out in FIGURE 12.

As high power is to be supplied (max 260 W) by the DC-DC converters, a Buck converter has been chosen due to its efficiency characteristics. FIGURE 13 shows the solution for an adaptive power supply for stages 2 and 3 supporting OPA operation. With eight couples of resistors and transistors in parallel the output voltages of the stages are changed by acting on a feed-back loop. The reconfiguration speed of the system is enhanced by using only one inductor (L) and one capacitor (C). The main challenge is to define the values for L and C for good efficiency and to define the compensation loop especially for achieving a low ripple on the output voltage and a good stability of the circuit.

EARTH PROJECT

EARTH_WP4_D4.3 27 / 121

Rload

iin

iout

VinVout

+

IC

Control

+

-

+

-

Decoder

3 to 8

Current

Feedback

Voltage Feedback and

compensation loop

Vref

FIGURE 13. Stages 2 and 3 architecture.

First evaluations done on a hardware prototype show a quite short transition time during reconfiguration. For example, only 50 µs are required to change the voltage from 20 V to 28 V with a load of 4.5 A. The power performance evaluated for the APSU is presented in different ways in the following diagrams.

FIGURE 14. Efficiency results of stages 2 and 3.

The efficiency performance presented in FIGURE 14 shows for Stages 2 and 3 a maximum of 92% power efficiency for 28 V regulation. The performance has been evaluated for 8 operating points (OP1 to OP8) defined for 8 different levels of maximum output power. These efficiency curves follow the basic metric used for describing the energy performance of power supplies.

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

0 500 1000 1500 2000 2500

Effi

cie

ncy

Current Load, (mA)

APSU Efficiency

OP1 OP2 OP3 OP4 OP5 OP6 OP7 OP8

0,75

0,8

0,85

0,9

0,95

400 900 1400 1900 2400

Zoom On Efficiency

EARTH PROJECT

EARTH_WP4_D4.3 28 / 121

Following the power performance presentation as used in EARTH power model [EARTH-D2.3], the power characteristics of the APSU has been defined as the power consumed in the APSU related to the power delivered to the AEEPA. Accordingly, a diagram with the DC input power and the supplied output power is presented in the appendix (section 10.1.4) while in FIGURE 15 the power dissipated in the APSU is shown for the eight evaluated operating points and arbitrary signal load as defined for the AEEPA in section 2.3.

FIGURE 15. APSU power dissipation for stages 2 and 3.

The presented concept for an APSU allows the realization of the defined SLA-TRX. The supported 8 operating points show a granularity which is high enough to exploit the energy efficiency potential of OPA. The presented power performance characteristics are used for generating the power characteristics of the whole SLA-TRX.

2.5. BENEFITS ON ENERGY EFFICIENCY IN MACRO-CELL BASE STATIONS

Considering the SLA-TRX system as an integration of the transceiver components the performance of the whole transceiver has been analysed by making use of the individual performance figures presented in the sections above.

The power performance of the SLA-TRX system is defined by power characteristics showing the consumed power related to the signal load. They are generated for the 8 operating points considered for the OPA feature and for the CD feature as presented in FIGURE 16.

The SLA-TRX solution enables significant energy saving for medium and low loads according to the optimization of operating points and the deactivation of certain components. The characteristics are based on a SLA-TRX system with a power amplifier designed for 20 W maximum average output power. OPA provides up to 23% of power reduction while this benefit is minimum when applying OP2 and increase at lower signal load showing the highest values for OP8. The CD feature provides 55% of power savings in time periods of no signal transmission. These characteristics are based on experimental evaluations.

0

1

2

3

4

5

6

0 10 20 30 40 50 60

Po

we

r Lo

st in

AP

SU (

W)

Output Power (W)

APSU Power performance

OP1 OP2 OP3 OP4 OP5 OP6 OP7 OP8 CD

EARTH PROJECT

EARTH_WP4_D4.3 29 / 121

FIGURE 16. Power consumption of SLA-TRX system with 20 W maximum average output power.

The elaborated power characteristics are used in the base station power model [EARTH-D2.3], generated in the project for energy efficiency evaluation on network level. In this way the impact of the SLA-TRX solution will be considered by the evaluation of the integrated solutions which are going to provide the final results of the project. Therefore, the impact of the SLA-TRX concept on network level can be evaluated by comparing the power consumption of an integrated solution when applying the SLA-TRX solution with the consumption when it is not applied.

The OPA feature of the transceiver system can be exploited for energy efficiency improvement at low and medium traffic load by applying scheduling algorithms which reduce the instantaneous transmitted output power allowing for transceiver operation at reduced signal load and with reduced power consumption according to the applied operating point.

The CD feature of the transceiver system takes benefit from the short transition times (3µs for deactivation and 10 µs for activation) allowing the deactivation of the AEEPA stages and some components in the SS RF-TX block during time slots of 2 or 3 LTE symbols. With special scheduling algorithms based on different DTX solutions, even longer time slots without signal transmission can be produced leading to increased benefit in energy efficiency, as shown in section 5.1.

A combined application of OPA and CD can be also reasonable to optimize the power efficiency. The benefit expected when applying the mentioned energy aware scheduling solutions is shown in chapter 5.

CD and OPA are key power saving solutions in macro-cell transceivers allowing for a maximum power reduction in combination with energy aware scheduling solutions like Cell DTX or Bandwidth Adaptation to be applied for different cell-sizes from dense urban to wilderness. But as basic solutions, their application is not restricted to macro-cell base BSs as shown in section 3, dealing with transceivers for small-cell BSs.

The solutions support energy efficiency enablers acting on different time scales as presented in [EARTH-D6.4]. By considering the three time scale categories their impact is as following:

As they represent hardware solutions it is a matter of deployment by considering equipment implementation. The impact is on long time scale of months and years.

EARTH PROJECT

EARTH_WP4_D4.3 30 / 121

Supporting solutions acting on medium time scale like slow BW adaptation, long Cell DTX, slow MIMO muting or macro/small-cell management, the impact is within minutes or hours.

Supporting solutions acting on short time scale like power control, fast bandwidth adaptation, fast MIMO muting, micro Cell DTX and special scheduling, the impact is within milliseconds and seconds.

The benefit of the SLA-TRX solution on the overall energy efficiency improvement has been analyzed in the project with the evaluation of “Integrated Solutions” based on system level simulations and delivering the final project results [EARTH-D6.4].

EARTH PROJECT

EARTH_WP4_D4.3 31 / 121

3. TRANSCEIVERS FOR SMALL-CELL BASE STATIONS

In areas with urban and dense urban population densities, the wireless communication scene is growing rapidly. To support this trend, the traditional approach of adding powerful base-stations might not be the best option because the capacity is mainly consumed in-doors and shows large variation in time, space and frequency. In a densely populated office environment for example, users will move indoor with their communication devices. Depending on their activities, the users might require low or high capacity. To support this dynamism, adding small-cell base-stations closer to the users is more beneficial to increase the link performance and to reduce the radiation power. Although the power consumption of an individual small-cell base-station is relatively low, a heterogeneous network of some large-cell and multiple small-cell base-stations might lead to an excessive ecological footprint and electrical cost [EARTH-D2.1]. From an energy efficiency perspective, it is essential that the small-cell base-stations are not only energy efficient but also energy flexible to adapt to the capacity dynamics.

Different energy efficiency and adaptive solutions have been proposed for Green Radio Technologies as described in [EARTH-D4.1]. The tracks related to hardware components for small-cell base-stations (BS) are grouped in the topic ‘transceiver for small-cell base-stations’. This topic is covered in this chapter by identifying the solutions and representing their energy efficiency potential. The architecture of the small-cell base-station is illustrated in FIGURE 17, 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 supply. Both, the RF and BB components offer reception and transmission capabilities for Up-Link (UL) and Down-Link (DL) operation respectively. FIGURE 17 shows on the right the energy distribution over the different base stations components for different cell sizes [EARTH-D2.3]. For small-cell base-stations such as pico- and femto-cell BSs, the power consumption of the RF component is more than 12% and the BB and PA consume about 30% each. For small-cell base-stations, it is opportune to investigate energy enhancement and adaptation 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 efficiency enablers.

(a) (b)

FIGURE 17. Simplified block diagram of a small-cell base station (a) and the BS power consumption breakdown for different cell-sizes (b).

The ‘transceiver for small-cell base-stations’ topic consists of five different tracks, covering the BB processor, the RF transceiver, two approaches on the PA and an antenna interface solution. All tracks target an innovative

BB-DL

BB-UL

RF-DL

RF-UL

PA

Main supply

DC-DC

Small-cell Base station

LLL

0%

20%

40%

60%

80%

100%

Macro Micro Pico Femto

64%

47%36% 32%

7%

7%

9%10%

5%

5%

7% 6%

7%

12%16%

13%

9% 29% 33%39%

8%

Po

we

r co

nsu

mp

tio

n b

reak

do

wn

PA Main Supply DC-DC RF BB Cooling

EARTH PROJECT

EARTH_WP4_D4.3 32 / 121

architecture that combines high performance with energy efficient operation. All tracks, except the antenna interface, are energy flexible depending on the traffic load. Thus the power performance of these four different transceiver blocks is defined by power characteristics showing the consumed power related to the signal load. The expected benefit is determined by comparing these characteristics with a state-of-the-art characteristic which does not consider the proposed concepts.

In the following subsections, the small-cell transceiver component tracks are discussed and quantified in terms of energy flexibility and energy efficiency.

3.1. ADAPTIVE SMALL-CELL BASEBAND PROCESSOR

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 as indicated in FIGURE 17. Hence, optimizing the energy efficiency of digital baseband processing is crucial. Digital baseband processing is also increasingly diverse. The different digital baseband processing sub-components considered for small-cell applications comprise:

Digital front-end processing and Analogue-Digital Conversion

Up/Down sampling and filtering (Filtering)

FFT and OFDM-specific processing

Frequency Domain (FD) processing (mapping/demapping, MIMO equalization), which scales partly linearly and partly non-linearly with the number of antennas

Forward Error Correction (FEC)

Platform control processing (CPU)

Optimizing the digital baseband processing power consumption can be achieved on technology level and with dynamic scaling.

On technology level, the energy efficiency of baseband processing is improved by using ASIP’s (Application Specific Instruction Processors) rather than FPGA’s. ASIP based baseband platforms are dedicatedly optimized for the workload of signal processing tasks. For instance, consider four categories of ASIPs for energy efficient signal processing:

Digital front end processors and Analogue-Digital Convertors that are optimized for mixed signal filtering, synchronization and data conversion.

Baseband processors (time and frequency domain) that are optimized for various diversified signal processing tasks in the inner modem, such as MIMO-OFDM processing, channel estimation, equalization etc.

Channel error correction processors that are optimized for various FEC en-/de-coding tasks.

Platform control processors such as power regulation and micro-processors.

Due to the high degree of specialization, higher intrinsic energy efficiency is achieved. The energy efficiency improvement when porting a small-cell baseband processor from an FPGA platform towards an ASIP platform exceeds 20%. Additionally, technology scaling enhances the dynamic energy efficiency, expressed in Giga Operation per Watt (Gops/W). Each scaling step (every two years) increases the Gops/W with 20%. This increase of energy efficiency is however tempered by the increase of leakage when migrating to dense technologies. This leakage increases the static power consumption and doubles (or even triples) per scaling step. In 2012 technology, about 30% of the consumed power in the baseband processor is dedicated to leakage. It is expected that technological research will tackle this problem and reduce the impact of leakage. From a design perspective, the stand-by leakage (when part of the processor is in sleep) can be reduced by

EARTH PROJECT

EARTH_WP4_D4.3 33 / 121

power gating: transistor switches are placed in-between the supply/ground connections and the logic to additionally cut-off leakage currents when the circuit is not operating.

Different dynamic scaling techniques are identified to improve the power consumption of the digital baseband processing hardware in function of the traffic load. These dynamic scaling techniques supports different higher-level techniques proposed in the project: duty-cycling in time and in frequency, modulation and coding-rate adaptation, bandwidth adaptation (deactivation based on resource elements over multiple frames) and antenna muting.

The dynamic hardware scaling techniques create instances in time and frequency where parts of the processor operations get no or low priority. Then, the dynamic power consumption is reduced by applying clock gating and voltage- and frequency scaling. These techniques require a specific circuit topology. For clock gating, different clock lines are implemented that can be deactivated to avoid digital logic from switching when they are not addressed. For dynamic voltage scaling, the processor is designed in different regions with dedicated supplies. Depending on the importance or intensity of the operation, the multi-level general supply will distribute the voltages over the different regions – and a similar strategy applies for the frequency scaling. From a signal processing point of view, dynamic scaling is achieved by changing the time- and frequency accuracy by means of the dynamic range and the FFT size respectively.

The diversity and interaction of hardware and software techniques have been implemented in a power model [DeDG12]. This model calculates the digital processing dynamic power consumption for UL and DL, covering frequency-domain processing (up/down-sampling and filtering, MIMO-OFDM processing with an FFT, equalization and mapping/demapping), channel en/decoding (FEC) and platform control. This model also considers the static power consumption due to leakage. The power consumption dynamics are based on scaling exponents between each baseband sub-component and each scaling parameter. TABLE 2 depicts the scaling exponents for DL: it shows whether the amount of computations performed in each sub-component was depending on each of the scaling parameters or not. Filtering and OFDM are put together as time domain processing category, with the same scaling properties. Leakage power is also a category of its own. Clock gating is assumed when duty-cycling components in time domain, but not power gating (leakage remains). In UL, all exponents are the same, except for the non-linear frequency-domain processing (cubic dependency (^3) in uplink instead of the quadratic dependency (^2) in downlink).

TABLE 2. Scaling exponents for the baseband sub-components in downlink

Scaling exponents Time

duty-cycling Freq.

Duty-cycling Mod.&Coding Ant. muting BW adapt.

Filtering and OFDM 1 0 0 1 1

FD (linear) 1 1 0 1 1

FD (non-linear) 1 1 0 2 1

FEC 1 1 1 1 1

CPU 0 0 0 1 0

Leakage 0 0 0 1 1

FIGURE 18 shows the power consumption of a 2x2 MIMO pico-cell baseband processor over the signal load. It is observed that the UL consumes almost twice as much power then the DL because MIMO processing of equalization and detection is more complex and because the FEC decoding is significantly more complex than encoding. The Energy Adaptive (EA) results consider the energy flexible processor solutions discussed in this subsection. 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, baseband processing is relatively independent of the

EARTH PROJECT

EARTH_WP4_D4.3 34 / 121

throughput (time-domain processing of the signal, platform control overhead...). The overall power reduction ranges up to 20% at signal loads <20%. At higher signal loads, the power reduction decreases.

FIGURE 18. Power consumption of a 2x2 MIMO pico-cell baseband processor over the signal load.

It can be concluded that for small-cell baseband processing, part of the energy efficiency is gained from technology: more than 20% from using ASIP’s instead of FPGA’s and 20% in Gops/W from technology scaling. Contrarily, technology scaling also increases the static power consumption due to leakage. Further, energy efficiency is gained both in hardware as in software. In hardware, innovative techniques to slower or shut-down parts of digital switching logic can be implemented. In software, scalable algorithms to ease computations while offering the required instantaneous performance. These techniques have been modelled to quantify the power efficiency improvement depending on the signal load. The proposed baseband processing techniques are complementary with and facilitating some of the higher-level small-cell energy efficiency enabling techniques presented in EARTH.

3.2. ADAPTIVE SMALL-CELL RF TRANSCEIVER

Scaling down the cell deployment towards small-cells shifts the relative weight of power consumption in the base-station from the PA towards the baseband and the RF transceiver. This subsection describes how the energy efficiency is optimized in the small-cell RF transceiver on a long and short timescale.

On a long timescale of months and years, the power consumption is reduced by using energy-efficient implementation technologies and transceiver architectures. These two approaches are inspired by recent work in the area of handheld communication devices where energy efficiency is crucial to increase the battery lifetime. With respect to implementation technology, the low-cost and power-efficient CMOS technology can be used because the small-cell RF transceiver operates on small signals. This technology also benefits from technology scaling every year or two, which results in an energy efficiency improvement of about 20% per scaling step. With respect to the transceiver architecture, dense integration and reduction of the circuit complexity offer high potential to increase the cost- and power-efficiency. Changing the architectural structure is justified because the small-cell base-stations have less stringent linearity and blocking requirements with respect to large-cells because their deployment provides more flexibility to system dynamics and fewer users should be serviced. Using direct-conversion architecture instead of a heterodyne architecture shows a power reduction of about 20%. The direct-conversion architecture comes, however, with several drawbacks: increased sensitivity to even order non-linearities and to circuit imperfections in the quadrature paths. These drawbacks can be countered at design time by improving the matching and by adding complex circuits, which degrades the power reduction factor. Recent research [DeWV09] however, presents a method that deals with

0

1

2

3

4

5

6

7

20% 40% 60% 80% 100%

Po

we

r C

on

sum

pti

on

[W

]

Signal Load [%]

UL

UL EA

DL

DL EA

UL+DL

UL+DL EA

Potential for Energy Savings

20% Power reduction15%

10%

EARTH PROJECT

EARTH_WP4_D4.3 35 / 121

the architectural drawbacks in the digital domain without impacting the implementation complexity and thus maintaining the power reduction factor.

To support the energy efficiency enabling techniques depending on the traffic load proposed in the EARTH project, the small-cell RF transceiver component is made flexible. This flexibility covers different tuning knobs:

sub-component muting and deactivation: to enable micro- and deep-sleeps and for complete transceiver deactivation

bias point adaptation, gain distribution adaptation and linearity scaling: to scale the Signal to Noise and Distortion Ratio (SiNAD) performance and the transmission power

filter order and bandwidth adaptation: to change the baseband bandwidth and to adapt the interference suppression

Tuning these knobs impact the power consumption of the RF transceiver component and can be exploited to increase the energy efficiency in function of the traffic scenario. The large amount of tuning knobs requires an advanced protocol to reconfigure the RF transceiver. An on-chip serial protocol has been developed as described in [EbG08]. This generic protocol offers different instruction lengths to facilitate normal and fast addressing. The sampling speed depends on the technology and ranges up to 300 MHz for 40 nm CMOS technology.

To observe the possible cooperation between the energy efficiency enablers and the small-cell RF transceiver, FIGURE 19 indicates the optimization timescale for the transceiver sub-components. FIGURE 20 shows the custom made RF transceiver used for measurements and validation. This transceiver comprises the block diagram shown in FIGURE 19.

FIGURE 19. Small-cell RF transceiver block diagram based on a direct-conversion architecture. The energy

efficiency optimization timescale of the energy flexible sub-components are indicated on an

indicative timescale.

FIGURE 20. Small-cell RF transceiver pictogram.

Nearly all sub-components allow rapid reconfiguration as their ramp-up and/or settling effects are small and limited in time. The sub-components linked with the Voltage Controlled Oscillator (VCO) however suffer from transition effects at power-up: it takes several milliseconds to lock the Phase Lock Loop (PLL) and during this process, the LO signal does not match the wanted carrier frequency. The VCO should also remain in operation

PLL

LNA 90o

90o

PPA

Energy-efficient and -flexible

integrated RF transceiver

timescalemsec sec min hour day month year

EARTH PROJECT

EARTH_WP4_D4.3 36 / 121

for background processing such as sensing. Complete deactivation of the RF transceiver chip is only appropriate within a long optimization timescale, because of the start-up calibration and configuration procedures for the different transceiver sub-components.

Now, the energy efficiency improvement is quantified when exploiting the RF transceiver tuning knobs. In traditional base-stations (no EARTH optimisation), the RF transceiver always targets the best SiNAD performance independent of the signal load. From an energy consumption perspective (EARTH optimization), it is more advantageous to scale the transceiver to provide a ‘just good enough’ SiNAD performance and to schedule sleep states and bandwidth adaptation based on the load and the channel conditions. The energy efficiency impact of this approach is illustrated in FIGURE 21, which shows the measured power consumption for a 2x2 MIMO pico-cell base-station RF transceiver over the signal load. This figure, derived with the power model presented in [DeDG12], 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 30% (signal load <50%) and up to 55% (signal load <5%).

FIGURE 21. Power consumption of a 2x2 MIMO pico-cell base-station RF transceiver over the traffic load.

It can be concluded that the traffic load independent power efficiency improvement of the RF transceiver is about 35% due to technology scaling and architecture, and that the traffic load dependent power efficiency improvement is 30% on average due to SiNAD adaptation and time- and frequency duty-cycling (respectively sleep modes and bandwidth adaptation). The flexible RF transceiver component track is complementary with all small-cell energy efficiency enabling techniques and even facilitates some of them.

3.3. ADAPTIVE ENERGY EFFICIENT POWER AMPLIFIER

Unlike in a macro base-station, the power amplifier in a small-cell BS is not the most power consuming component. Although the relative power consumption decreases with the cell size, the energy efficiency requirement of the PA remains crucial. This requirement is given by the fact that in small-cell base-stations digital pre-distortion is rarely used and thus the PA should support the high peak to-average-power-ratio (PAPR) offered by the LTE signal. To support the high PAPR, the PA is forced to operate with a large back-off and as a result with low energy efficiency. It is therefore crucial to investigate techniques to improve the energy efficiency, as discussed next.

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

0% 20% 40% 60% 80% 100%

Po

we

r C

on

sum

pti

on

[W

]

Signal Load [%]

UL

UL EA

DL

DL EA

UL+DL

UL+DL EA

Potential for Energy Savings

58%

Power reduction

39% 32%

EARTH PROJECT

EARTH_WP4_D4.3 37 / 121

An Adaptive Energy Efficient PA (AEEPA) has been proposed to reach an energy efficiency improvement in a small-cell BS by using techniques as in a macro-cell BS presented in section 2.3. The hardware development supports two techniques: the Operating Point Adjustment (OPA) and the Component Deactivation (CD). Both techniques enable to adapt the operating point of the AEEPA to different RF output power levels depending on signal load and to deactivate some PA stages when empty LTE symbols occur.

In order to evaluate these concepts, an AEEPA has been developed for LTE signals in the frequency band from 2.11 to 2.17 GHz with 0.25 W average RF output power. A signal with around 12 dB PAPR has been considered, because digital pre-distortion is not applied.

The hardware has two stages:

The DRIVER is realized as a module with fixed bias.

The high power amplifier (HPA) is a module enabling the OPA and CD features. This stage can be adapted to different RF output power levels depending on the restricted maximum signal load and can be switched off during empty LTE symbols.

The block diagram of the AEEPA for small-cell BS is presented in the following figure:

FIGURE 22. Block diagram for Adaptive Energy Efficient PA for small-cell BS.

The hardware development for the AEEPA is implemented in an aluminium housing which integrates a RF board and a biasing and control board. Furthermore, a heat sink is incorporated to maintain the performance in temperature as appears in FIGURE 23.

A DB15 connector interconnects the baseband processor and the AEEPA. The baseband processor provides control information aligned to the signal level for adapting the AEEPA operating point according to the signal load. Furthermore, there is a dedicated line to implement the CD feature and reduce the energy consumption during empty LTE symbols. Some components in the AEEPA are deactivated when no data or signalling is transmitted. The deactivation occurs within 3µs while the activation takes about 6µs. This allows the component deactivation during empty LTE symbols and fulfils the 17 µs of maximum transient period requested by the LTE-TDD specification [LTE-SPEC].

DRIVER HPA

Variable BIAS Fixed BIAS

POWER AMPLIFIER

Supply Voltage V1

Supply Voltage V2

BIAS Control

EARTH PROJECT

EARTH_WP4_D4.3 38 / 121

FIGURE 23. Photograph of Adaptive Energy Efficient PA for small-cell BS.

The HPA is implemented using a GaN HEMT technology amplifier which operates on a supply voltage of +28 V. The OPA feature is performed modifying the gate voltage in the HPA while keeping the drain voltage. A SPI interface connects the baseband processor and the AEEPA to carry on the suitable adjustment in the RF output power level depending on the traffic load. Furthermore, there is +5 V supply voltage to supply the DRIVER and the control board.

Five operating points were defined for different signal load ranges: 100-80% for OP1, 80-60% for OP2, 60-40% for OP3, 40-20% for OP4 and 0-20% for OP5, providing significant enhancements. For each signal load range, a gate voltage is applied to the HPA using a SPI interface. The operating point adjustment occurs within 5 µs and makes possible to adapt the AEEPA in each 67 µs of LTE symbol time [LTE-SPEC].

The OPA solution applied to the AEEPA is mainly beneficial at low and medium traffic load as appears in FIGURE 24.

FIGURE 24. Power consumption of 0.25W Adaptive Energy Efficient PA for small-cell BS.

Without applying the OPA feature, the AEEPA is operation in OP1, which defines the reference for evaluating the energy efficiency performance. Using this technique, the power reductions are around 14% for OP2, 26% for OP3, 40% for OP4 and 55% for OP5. While the energy efficiency improvement for CD reaches up to 80%.

The combination of both features provides a significant energy saving and improves energy efficiency in small-cell BS, which typically are deployed in urban and dense urban scenarios.

EARTH PROJECT

EARTH_WP4_D4.3 39 / 121

The performance evaluated for the AEEPA is presented in FIGURE 25 as power efficiency for the different operating points, following the power efficiency metric defined in [EARTH-D2.4].

FIGURE 25. PAE of 0.25W Adaptive Energy Efficient PA for small-cell BS.

The power efficiency is restricted due to the 12 dB back-off considered for the small-cell application. Furthermore, as the operating point adjustment considers only the gate voltage adaptation for the HPA, additional improvement can be achieved by adjusting also to the drain voltage as already implemented for the macro-cell AEEPA.

3.4. TUNEABLE MATCHING NETWORK FOR LOAD ADAPTIVE POWER AMPLIFIER

Conventional linear PAs make use of a fixed output matching network that has been designed to maximize efficiency at peak power, and efficiency degradation occurs with decreasing output power level. With high PAPR modulated signals, the PA has to operate in back-off (BO) most of the time, resulting in relatively low average efficiency. The proposed concept of load adaptive PA is an alternative PA concept for small cells BS which aims at PA efficiency optimization at high and medium signal loads by providing dynamic impedance modulation to the PA thanks to a Tuneable Matching Network (TMN) [EARTH-D4.2]. Compared to other PA architectures (Doherty, Envelope Tracking…), the proposed architecture provides an attractive trade-off among performance, cost and integration capability and is well suited for CMOS integration with the transceiver.

A general block diagram of the load adaptive PA is shown in FIGURE 26, where the TMN provides optimum load impedance to the PA for various output power levels. Thereby the load impedance is adapted to the signal envelope. In the proposed implementation, high voltage Silicon-on-Insulator (SOI) CMOS switched capacitors have been designed and implemented to dynamically control the load impedance of a LDMOS PA.

EARTH PROJECT

EARTH_WP4_D4.3 40 / 121

FIGURE 26. Load Adaptive PA block diagram.

A prototype of the proposed load adaptive PA has been developed as shown in FIGURE 27. The TMN has been co-designed with the LDMOS PA to avoid a 50 Ω interface, enabling to optimize the bandwidth. Both devices have been integrated in a SOI CMOS process and have been designed as two separate chips which are mounted on the same board.

FIGURE 27. Photograph of the Load Adaptive PA: Board (a) and Chip-on-Board assembly (b).

The TMN makes use of fixed inductances/transmission lines for inductive sections of the matching network and tuneable high Quality Factor (Q) capacitors (Q>50) for capacitive sections. Tuneable capacitors consist of arrays of binary weighted switched capacitors using floating body SOI NMOS transistors as switch. Control logic is integrated on chip and allows the selection of appropriate capacitance values through an integrated SPI interface. The TMN circuit occupies an area of less than 1 mm2 and operates with 2.5V voltage supply. The PA consists of a SOI LDMOS power stage with on board fixed input matching and output pre-matching networks. It can operate up to 5.0 V voltage supply and provides around 34 dBm of peak power at 2.14 GHz under 4.0 V voltage supply. The TMN configuration and capacitance range has been selected through electrical circuit simulations with the goal of achieving maximum PA efficiency for different output power levels.

Expected improvements in terms of power consumption and PA power added efficiency (PAE) are given in FIGURE 28, after simulations of the load adaptive PA for different TMN states and input signal amplitudes.

TMN

PA

(a) (b)

EARTH PROJECT

EARTH_WP4_D4.3 41 / 121

FIGURE 28. Load Adaptive PA: (a) Power consumption (Pdc) and (b) PAE vs Output power (Pout) for fixed matching (blue curve) and optimized matching (red curve)

The proposed load adaptive PA shows significant potential for energy savings, with more than 30% power consumption reduction from 3 dB to 14 dB BO and PA efficiency improvement raising up to 50% from 7 dB to 12 dB BO. In the context of small-cell LTE BS, dynamic tuning of the TMN is necessary and requires properly calculated PA input and TMN control signals to avoid distortion of the modulated signal. Especially, time alignment between both input signals has to be precisely controlled because of its critical impact on linearity. Linearity and efficiency investigations using a high PAPR modulated signal have recently been carried out on such kind of PA architecture [CaNT12]. By adequate signal conditioning, it has been shown through experiments that a high average PAE can be obtained while satisfying low out-of-band distortion (-50 dB ACLR).

In conclusion, the proposed concept of tuneable matching is mainly beneficial on a long timescale and at high signal loads. If combined with proper signal conditioning, it can be used with modulated signals and it can provide significant power consumption reduction (~30%).

3.5. LOW-LOSS ANTENNA INTERFACE

The efficiency of commercial base station antennas is rarely specified, but can be estimated in the range of 80-90% which leaves low margin for improvement. The Low-loss Antenna Interface task aims at a global optimization of the Tx front-end with a focus on the opportunities offered at the antenna level to reduce the constraints on other components. The Low-loss Antenna Interface task aims at a global optimization of the Tx front-end with a focus on the opportunities offered at the antenna level to reduce the constraints on other components. More specifically, it deals with the improvement of the energy efficiency of the front-end in classical FDD architectures for LTE small-cell applications [BoDG11]. The key idea consists in eliminating the duplexer component by co-designing the duplexing function within the antenna geometry. Practically the proposed innovative architecture uses a dual polarization antenna with dedicated Tx (or DL) and Rx (or UL) accesses. The antenna itself ensures at least 30 dB isolation between the two separate ports [EARTH-D4.1], FIGURE 29. Typical state-of-the-art duplexers for small-cell BSs, which are based on SAW or BAW technology, exhibit isolation levels between Tx and Rx ports in excess of 50 dB. Maximum values of insertion losses depend on the Tx and Rx frequency-bands separation and are typically in the range 2-3 dB. A SAW duplexer from Anatech® [duplex] with 2.2 dB losses has been selected to be incorporated on the standard RF front-end (Earth_OFF)

EARTH PROJECT

EARTH_WP4_D4.3 42 / 121

which will be used as a reference for the energy efficiency validation of the task. To reach the classical required 50 dB Tx/Rx isolation level for the proposed architecture (Earth_ON), an additional Tx filter with relaxed constraints has been designed. This allows reaching excellent insertion loss lower than 0.5 dB measured over the whole DL frequency band.

FIGURE 29. Comparison between standard (a) and proposed (b) RF front-end architectures with different places of the duplexing function.

The antenna design strategies, the radiating structure description and specific simulated results (impedance matching, Tx/Rx ports isolation, etc.) are detailed in the Appendix 10.2. Concerning the energy efficiency improvement, a detailed comparison between standard and proposed front-ends provides a significant loss reduction of 1.5 to 3 dB on the Tx chain, TABLE 3. This improvement is mainly due to the low insertion loss of the relaxed Tx filter compared to the duplexer ones.

TABLE 3. Loss for the different components and energy efficiency of the proposed concept

Standard Arch. (EARTH OFF)

Proposed Arch. (EARTH ON)

Anatech Saw Duplexer/Relaxed Filter Loss 2 – 3 dB 0.5 dB

Antenna Loss 0.5 – 1 dB 0.5 dB

Total Loss 2.5 – 4 dB 1 dB

Antenna Interface energy efficiency (%) 56% – 40% 80%

This part of the RF front-end is a passive hardware solution, thus this task and its results are fully scenario-independent. This energy efficiency improvement can be applied for small-cell BSs, whatever traffic load and scenario deployment if the two following conditions about polarization channel distribution and linear polarization are met:

The proposed approach assumes that the propagation channel in the small-cell re-distributes the energy from the BS on both polarizations on the handset antenna(s). Thus a +45° orientation polarised signal radiated by the BS will be received with the both +45° and -45° orientations on the user terminal antenna. This assumption is usually used in dense multipath radio channel propagation.

The duplexing antenna radiates and receives two orthogonal (here linear) polarizations with the same co-located radiating element. This antenna concept cannot be used to implement polarisation diversity.

(a) (b)

EARTH PROJECT

EARTH_WP4_D4.3 43 / 121

The proposed duplexing antenna is also applicable in antenna array topology. Nevertheless a stronger Tx/Rx isolation specification may be one of the major limitations of the proposed solution by applying it to macro-cell BS. To conclude, the Low-loss Antenna Interface solution improves from 40/56% to 80% the global energy efficiency of the antenna interface. To provide the same radiated power in both standard and proposed architectures, this corresponds to a decreasing of the PA RF output power of 1.5 dB to 3 dB thanks to the duplexing antenna. This passive solution is fully scenario-independent and is complementary with all energy efficiency enabling techniques for small-cell BS. To reach this improvement, both the Tx/Rx isolation and Tx filtering functions are partially moved from a lossy duplexer component to an efficient duplexing antenna. Finally the proposed solution includes two duplexing antennas to demonstrate the spatial diversity (MIMO configuration) compatibility. A prototype has been realized (FIGURE 30), its experimental characterization and performance results are reported in the [EARTH-D5.3]. The out of band Tx rejection which has not been estimated by simulation, has to be evaluated by measurement on the whole RF front-end prototype. The proposed architecture has been implemented with sectorial radiation pattern antennas. However for femto-cell BS with small size constraint, it could be interesting to design omni-directional duplexing antennas. As this kind of antenna is more impacted by closed scatters, the strong Tx/Rx isolation requirement could be more delicate to obtain. This antenna interface is particularly suited for the Load Adaptive PA and its tuneable matching networks. The results show the potential on reducing the loss at the antenna interface due to the relaxed filter specification, and indicate to future developments based on efficient co-design between the PA matching network and the duplexing antenna without including a filter component.

FIGURE 30. Prototype of a low-loss antenna interface for small-cell BS.

3.6. BENEFITS ON ENERGY EFFICIENCY IN SMALL-CELL BASE STATIONS

An energy efficient “transceiver for small-cell base-stations” implies five different solutions, covering the BB processor, the RF transceiver, the PA and the antenna interface including several solutions to adapt the operation to the traffic load level. Their optimization timescale covers the complete spectrum from milliseconds towards years, where in the short timescale the operation point of the components are adapted and in the long timescale new architectures and technologies are implemented. The energy adaptation solutions on the BB engine, the RF transceiver and one PA solution are mainly beneficial at low and medium traffic load. One PA approach is also beneficial at higher traffic load.

EARTH PROJECT

EARTH_WP4_D4.3 44 / 121

For the adaptive BB processor, energy efficient techniques have been considered for hardware and software components. From the hardware side, an efficiency improvement of 20% is obtained by adopting ASIP instead of FPGA. From the software side, the proposed flexible energy aware baseband signal processing algorithms are mainly beneficial in UL at low signal loads and offer an overall power reduction of about 13% of traffic load dependent improvement.

For the adaptive RF transceiver, flexibility of different building blocks has been introduced and leads to a traffic load independent power efficiency improvement of 35% thanks to technology scaling and architecture and to a traffic load dependent power efficiency improvement of 30% on average through SiNAD adaptation and time/frequency duty-cycling.

For the PA, two different solutions have been proposed: an energy adaptive PA (AEEPA) targeting efficiency improvement at low and medium signal loads and a load adaptive PA targeting efficiency improvement at high signal loads.

The AEEPA implements the OPA and CD features described in the SLA-TRX section and is mainly beneficial at low and medium traffic loads. With the OPA feature, the power reduction reach 55% for 0-20% signal load, 40% for 20-40% signal load and 26% for 40-60% signal load. With the CD feature up to 80% efficiency improvement is obtained. Such a component can be easily connected to the RF transceiver and is controlled via a simple interface.

The load adaptive PA makes use of a tuneable matching network to optimize PA efficiency at different power levels and is mainly beneficial at high signal loads, providing 30% power consumption reduction if combined with proper signal conditioning. This concept can be used in conjunction with OPA and CD features, as implemented in the AEEPA, to provide efficiency improvement at low and medium load but it requires a more sensitive signal conditioning with proper synchronisation between RF and control signals. It shows the advantage to be integrated in CMOS technology providing a cheap and small component.

A low-loss interface making use of a duplexing antenna has been proposed as an efficient antenna interface. This fully scenario-independent passive solution improves from 56% to 80% the global energy efficiency of the antenna interface evaluated for the LTE UTRA operating band 1 and is complementary with all energy efficiency enabling techniques.

EARTH PROJECT

EARTH_WP4_D4.3 45 / 121

4. LOW LOSS ANTENNAS

The antenna is a key component of the transmission and reception chains of radio communication systems. In fact, as the last (first) block of the transmission (reception) chain, its performance has a major influence on the system performance. Specifically, from the EE perspective the antenna has a direct net impact on the overall EE of the system.

As discussed in [EARTH-D4.1] and [EARTH-D4.2] there is still room for improvement of the efficiency of the printed antenna structures used in radio access networks. This section is focused in proposing the use of a foam substrate to improve the antenna EE. Due to its properties it enables the realization of cheap, energy efficient and broadband antennas. Foam provides a good compromise between electromagnetic and mechanical requirements. From the electromagnetic point of view it is (macroscopically) almost air but it is mechanically enough stiff. Common foam has low losses (typically a 0.001 loss tangent) and a very low relative dielectric constant (typically 1.06).

This chapter contains the main results obtained for the application example chosen to proof that foam substrates can be used to improve the performance of BS antenna arrays. The analysis is focused in radiation efficiency. Detailed results can be obtained in Appendix 10.3.

4.1. PRINTED ARRAYS ON FOAM MATERIAL

Following the analysis carried out in [EARTH-D4.2] to proof that foam can provide printed antenna solutions for BS applications with improved EE, two antenna configurations have been designed fabricated and tested: a single patch antenna and an array of two patches. This approach will provide information on both the element efficiency and the array feed line system losses allowing extrapolating the efficiency of arrays with more elements.

LTE band 1 (frequency range 1920-1980 MHz for the UL and 2110-2170 for the DL) has been chosen. A dual-band multilayer structure of stacked patches fed by a non-resonant slot is used [CrP91, TaWP96, GaLLY02]. The multilayer profile is shown in FIGURE 31 and the array configuration in FIGURE 32. Standard thicknesses are used: 1.57 mm (0.062”) for the patch substrates, 0.79 mm (0.031”) for the feed line substrate and 10 mm for the foam.

FIGURE 31. Multilayer antenna profile.

The two stacked patches create the two required resonances (UL and DL sub-bands). A non-resonant H shaped slot is used to enhance the coupling between the bottom patch and the feeding line while keeping a low back radiation level [GaLLY02]. The thickness of the foam layer is selected according to the bandwidth specifications. Wide-band applications require a thick foam substrate layer. The antenna dimensions (patches,

Thick Foam Layer

Thin Duroid 5880 Layers

Stacked Patches

Feed Line

Feed Slot

EARTH PROJECT

EARTH_WP4_D4.3 46 / 121

slot and feed line) have been optimized to fulfil the bandwidth specifications (S11<-10 dB). The microstrip feed

line is composed of sections with 50 characteristic impedance and a quarter wavelength transformer.

FIGURE 32. The 4 metal layers of the array configuration.

4.2. EXPERIMENTAL RESULTS

The array antenna has been fabricated using conventional photolithography printing circuit technique. Photos of the array layers before and after assembling are shown in FIGURE 33.

a) b)

FIGURE 33. Array prototype before assembling (a) and after assembling (b).

EARTH PROJECT

EARTH_WP4_D4.3 47 / 121

The input reflection coefficient (S11) has been measured with a vector network analyzer. The radiation pattern (and gain) has been measured in a far field anechoic chamber. The radiation efficiency has been measured with a Wheeler cap [MeP07]. The results obtained are shown in FIGURE 34, FIGURE 35 and FIGURE 36.

FIGURE 34. Experimental input reflection coefficient of the antenna array.

FIGURE 35. Experimental radiation pattern of the antenna array (1.95GHz).

FIGURE 36. Experimental efficiency of the antenna array.

-35

-30

-25

-20

-15

-10

-5

0

1,5 1,6 1,7 1,8 1,9 2,0 2,1 2,2 2,3 2,4 2,5

S11

[d

B]

Frequency [GHz]

-20

-15

-10

-5

0

5

10

-180 -150 -120 -90 -60 -30 0 30 60 90 120 150 180

Gain

[d

Bi]

Angle [Degree]

E-Plane

H-Plane

80

82

84

86

88

90

92

94

96

98

100

1,7 1,8 1,9 2,0 2,1 2,2 2,3 2,4

Rad

iati

on

Eff

icie

ncy

[%

]

Frequency [GHz]

EARTH PROJECT

EARTH_WP4_D4.3 48 / 121

The experimental results have a good agreement with the numerical simulations. The main characteristics, at the central frequencies of LTE band 1 sub-bands, are contained in TABLE 4.

TABLE 4. Main experimental characteristics of the antenna array

Frequency 1.95 GHz 2.14 GHz

S11 [dB] -14.2 -25.1

Radiation efficiency [%] 93 93

Gain [dBi] 9.6 10.2

H-Plane Half-Power Beamwidth [Degree] 69 70

E-Plane Half-Power Beamwidth [Degree] 44 46

Front to Back Ratio [dB] 14.8 15.1

From the numerical simulations and experimental results the radiation efficiency of arrays of patches with different number of elements has been extrapolated. Comparison between foam and the commonly used Rogers’ RO 3003 substrate has been carried out. The corresponding results are contained in TABLE 5.

TABLE 5. Radiation efficiency of arrays of patches with different number of elements (N)

Substrate N=1 N=2 N=4 N=8

Foam 98.0 97.3 96.4 94.7

Rogers RO 3003 88.9 85.4 81.1 73.6

As expected, the EE improvement provided by foam increases as the number of array elements increases. The improvement ranges from 9% for the single element antenna to 21% for the eight elements array.

In the proposed antenna prototype foam is used together with a Teflon substrate (Rogers Duroid 5880) but in industrial mass production processes foam only solutions can be adopted, minimizing the cost but maintaining the energy efficiency performance. Foam with predefined 3D shapes can be moulded from liquid solutions, and metal layers can be incorporated into the structure using a multi-step fabrication procedure.

4.3. BENEFITS ON ENERGY EFFICIENCY DUE TO LOW LOSS ANTENNAS

The main conclusion of the present chapter is that foam is a dielectric material adequate for application in BS printed antennas. It is rugged and cheap and compared with other cheap common substrates can provide a very significant radiation efficiency improvement (above 15% for arrays of practical interest). As the antenna is the last (first) device of the radio communication transmission (reception) chain this EE improvement comes on top of the improvement provided by all the other devices of the chain.

The overall effect of this EE improvement on a network depends on the percentage of BSs that use printed antennas. A Portuguese operator estimates the penetration of printed antennas in its 3G network as 30%. Therefore the potential of energy saving provided by the use of foam in printed antenna BSs is significant.

It is not difficult to foresee an increase in the complexity of future BS networks using sectorised cells and advanced techniques such as beamforming (Chapter 6) and MIMO (Chapter 7) in which cases the adoption of printed antennas will most probably be increased and therefore the impact of the proposed foam substrate solution will be even more significant.

EARTH PROJECT

EARTH_WP4_D4.3 49 / 121

5. RADIO INTERFACE TECHNIQUES EXPLOITING HARDWARE FEATURES

Radio interface techniques exploit the potential of the hardware features and the interaction between higher protocol layers and the hardware to the benefit of improving the energy efficiency. Three techniques are described in this chapter: 1) Cell DTX, which is de-activating the radio parts of a transmitter when there is no data to transmit. 2) Scheduling strategies operating in frequency and time domain, which is based on the adjustment of the bandwidth and PA operating point by considering the required traffic load. 3) Adaptability to system dynamics solution adapts the resource scheduling to the channel conditions to reduce the energy consumption. Through these techniques, the system dynamics translates into optimized energy efficiency via the hardware components which means that having energy efficient and energy adaptive hardware components is essential. Three techniques are described below and further explanations are given in the following subchapters.

1. Cell DTX

Discontinues transmission in time domain in radio part of a transmitter is called cell DTX. It is an enabler for energy efficiency improvements when there is no data to transmit. We propose four modes of DTX, ranging from duration from one or a few OFDM symbols to several hours. For LTE it makes sense to relate the DTX mode to the sub-frame (1 ms) and radio frame duration (10 ms) as follows:

Micro DTX: Radio unit is turned off in time intervals less than 1 ms. No changes occur in the control signals. This type of DTX is supported in LTE Rel-8 and is depicted in FIGURE 103.

MBSFN-based DTX: Radio unit is turned off in time intervals less than 10 ms. Radio frame consists of a combination of micro DTX transmission configuration and MBSFN sub-frames. It is visualized in FIGURE 104.

Short DTX: Radio unit is turned off in time intervals less than 10ms (but at least equal to 1 ms). As can be seen in FIGURE 105, CSRSs are removed; therefore, UEs perform mobility measurements on primary and secondary synchronization signals (PSS, SSS). Only SSS, PSS and PBCH transmissions are mandatory. It is visualized in FIGURE 105.

Long DTX: Radio unit is off in 10 ms or longer durations. The long DTX duration shall however not be so long that the UEs are not able to perform initial cell search and mobility measurements. It can be seen in FIGURE 37.

2. Scheduling strategies operating in frequency and time domain

In macro base stations (BS) the power amplifier (PA) is the component with the highest energy consumption. Even when no data is transmitted the PA requires a DC voltage holding its fixed operation point. At present, the supply voltage is fixed independently of the traffic load. Power reduction can be achieved by: 1) Adapting the operating point of the PA when the required signal load is below the maximum level. 2) Deactivating the PA whenever no data or signalling is transmitted. Very promising energy savings are possible in low load situations by making the scheduler aware of power saving modes of the PA. Therefore, different approaches have been investigated:

Bandwidth Adaptation (BW), which is based on the adjustment of the bandwidth to the required traffic load. Depending on traffic load the bandwidth can be stepwise downscaled that lower numbers of physical resource blocks (PRBs) are allocated. Then the PA can adapt to lower supply voltage and less reference signals have to be sent. This can be performed either by the usage of a fast carrier aggregation (CA) procedure or by an approach based on the reconfiguration of BS system parameters. The required CA procedure is currently not standardised by 3GPP and needs specification of a fast stepwise switching, e.g. between the lowest BW usage at 1.4 MHz and up to

EARTH PROJECT

EARTH_WP4_D4.3 50 / 121

the highest BW usage at 10 MHz, which may be applied during night or in busy hours, respectively. The EARTH evaluation is limited to 10 MHz which refers to the baseline parameter, but the approach works also for higher bandwidth usages. The other solution using a reconfiguration of cell parameters is very slow and may impact the performance of cell-edge users.

Capacity Adaptation (CAP) is a method, which does not change the maximum used bandwidth and the number of reference signals. An adaptation to lower load is performed by scheduling only a part of the subcarriers, i.e. limiting the number of scheduled PRBs. This approach allows lowering the PA supply voltage, but it is transparent to the mobile terminals and maintains frequency diversity.

3. Adaptability to system dynamic

As described in [EARTH-D4.1, EARTH-D4.2], there is a large difference in energy consumption per bit when transmitting with the highest modulation and coding rate compared to the lowest one. To meet BER performance constraints, the appropriate modulation and coding rate is selected by means of a channel quality indicator (CQI) index that is an indication of the channel quality. Adaptive modulation and coding rate adapts the transmission scheme to the channel conditions, i.e., when the channel conditions are good, a high CQI index will be used since a high throughput communication is possible. In bad channel conditions, energy-inefficient modulation and coding schemes need to be selected to ensure that BER requirements can be met. However, by not transmitting under bad channel conditions in certain time slots in the hope of having better conditions in the next slots, we can trade off packet delay with energy consumption. This can be reflected in a scheduling algorithm that sends packets as long as good channel conditions are present, therefore using medium to high rate transmission schemes. When the conditions are not good enough, these packets could be buffered expecting to be sent when the conditions improve or, if too long, when a delay timer expires.

5.1. CELL DTX

In this chapter, Long DTX method and how it can be realized and operated will be explained. Details on micro DTX, MBSFN-based DTX and short DTX methods, already described in [EARTH-D4.1, EARTH-D4.2], are presented in the Appendix subsections 10.6.1, 10.6.2 and 10.6.3 respectively. Moreover, system simulations are performed and presented in the following section in order to evaluate the network energy consumption reduction by micro DTX, MBSFSN-based and short DTX modes.

5.1.1. Long DTX Network Implementation

In long DTX, the radio can be turned off for more than a radio frame, i.e. the DTX time is larger than 10 ms. In order to enable this it is needed to define a low activity mode for the base station. The low activity mode consists of an active period with normal transmission of synchronization signals and broadcast channels, as well as an idle period where no transmissions are present, see FIGURE 37.

Low activity mode

Active Period (e.g. 100 ms) Long DTX

(e.g. 900 ms)

Normal mode

(Only short DTX)

Normal mode

(Only short DTX)

FIGURE 37. Illustration of long DTX.

EARTH PROJECT

EARTH_WP4_D4.3 51 / 121

The simple idea with long DTX is thus not to transmit in the downlink in a cell with no active terminals, except for intermittent transmissions of the signals necessary for active terminals in the proximity to find the idle cell. This is especially useful in cell scenarios where macro cells overlays micro or pico cells. In case of low load idle terminals can camp on the macro cells and hence use their sync symbols to be in sync with the network. As the network knows which cells the active terminals are connected to, it is straight-forward to identify “empty” cells. Once a terminal moves into the area covered by the idle cell, the base station needs to wake up and resume transmission as well as reception of signals. To determine if the cell should resume the transmission/reception activity, it is necessary to detect if a terminal is moving into the idle cell. Such a terminal expects certain signals (synchronization signals, reference signals, broadcast information) to be transmitted; otherwise it is not able to find the cell. Therefore, despite being in low activity, the signals necessary for mobility are intermittently transmitted. The potential energy reduction at the transmitter side is approximately proportional to Ta / (Tp + Ta) where Ta is the duration of the active period and Tp is the duration of the idle period. The value of Ta should be selected large enough to allow the terminal to find synchronization with sufficiently high probability as well as being able to perform signal measurements on the cell. The time needed for this depends on the signal-to-noise ratio at the terminal, but if Ta is in the order of 100 ms to 1 second the probability of not finding the idle cell can be expected to be sufficiently low. The value Tp should be large enough to allow for a significant reduction in energy consumption. At the same time, a too large Tp means that terminals may not find the idle cell. A typical value of Tp could be in the order of 1 to 10 seconds. In order to enable long DTX also for macro cells, we need to ensure that mobile stations are able to perform initial cell search and mobility measurements on cells in low activity mode. Initial cell search can be obtained by extending the UE cell search procedure:

1. First a normal cell search is performed. The UE can in this step only detect cells that are operating in normal mode.

2. If the normal cell search procedure fails the UE performs an extended cell search in order to find cells in low activity mode. Since the eNB is only performing intermittent transmissions the UE will need to measure longer on each candidate frequency in order to detect these transmissions.

The drawback with the extended cell search procedure is that occasionally an initial cell search procedure can take considerably longer to perform compared to a normal cell search. In order to speed up initial cell search over several different frequencies we can stagger the active mode transmissions in time with a time offset that depends on the frequency, see FIGURE 38.

EARTH PROJECT

EARTH_WP4_D4.3 52 / 121

Active Period (e.g. 100 ms)

Idle Period (e.g. 900 ms)

0 1 2 3 [seconds] Network timing:

f0

f1

f2

f3

1

2

3

0 1 2 3 [seconds] UE timing:

UE

FIGURE 38. Timed UE cell search relying on a low accuracy UE time. Active periods are shifted differently for

each frequency. UE cell search windows for 2,10 ,, fff and 3f are shown in red, yellow, green, and blue.

In the example in FIGURE 38, the UE can scan four frequencies in one second and consequently an initial cell search procedure can be accomplished four times quicker compared to the extended cell search procedure without time staggering of active periods described above. If the UE does not have a sufficiently accurate time reference it is still possible to inform the UE (e.g. using 3GPP signalling procedure) about the relative order and timing of transmissions on different carriers. The UE need only to perform a long cell search procedure on the first frequency it measures on and once a synchronization signal is detected the UE has also obtained the information needed to quickly find all other transmissions on different frequencies.

In general it is also a good idea to allow the network to control when active UEs perform mobility measurements since this allows for long DTX without putting any additional search burden on the UEs. Preferably, the long-DTX pattern in the base station is selected to match the DRX pattern configured in the terminals. If the terminals wake up according to a certain pattern to perform neighbouring cell measurements, it is beneficial if the long-DTX pattern in the base station matches this pattern. An idle cell would exit the long-DTX period when it detects that a terminal is performing random access. In addition, as handover is controlled by the network by sending commands to the terminals, the network can also wake up idle base stations in conjunction with the handover procedure. For example a base station A (controlling cell A) commanding a terminal to perform handover from cell A to cell B (controlled by base station B) would, in addition to the handover command to the terminal, also send a wake up command to base station B.

5.1.2. Micro, MBSFN-based and Short DTX Evaluation Results Aggregated to Global Scale Using the E3F

To assess the gains of Micro, MBSFN and short DTX at a larger scale, aggregation to global scale according to the long term traffic model and large scale deployment model in [EARTH-D2.3] is done. The global average

EARTH PROJECT

EARTH_WP4_D4.3 53 / 121

power consumption per area unit and energy per bit is given for DTXMicroDTXShort PP __ . FIGURE 39 presents

the simulation results for power consumption per area unit and power consumption per bit in the medium traffic density scenario. As is shown in the simulation results, Micro DTX and MBSFN based DTX give a power consumption reduction of 15% and 19% respectively compared to the reference system, while for the short DTX scheme, the energy consumption drops by 28 percent of the legacy system. Evaluations are based on the SOTA power model where Earth “off” and sleep mode “on” were selected. It should be mentioned that when applying the sleep mode, the power consumption of the PA is not affected, instead it is the base band part that provides the largest part of DTX power reduction.

FIGURE 39. Global energy consumption per area-time unit and per bit in high traffic density network, using

DTXMicroDTXShort PP __ .

The highest expected overall energy reduction is 28% for the short DTX case compared to the legacy LTE without DTX.

5.2. SCHEDULING STRATEGIES OPERATING IN FREQUENCY AND TIME DOMAIN

The characteristic of a theoretically evaluated Doherty PA with energy savings features (see FIGURE 40) has been used for investigations according to approaches based on BW adaptation, CAP adaptation, micro DTX and combined approaches based on BW/CAP adaptation and micro DTX. The state-of-the-art (SOTA) characteristic describes the energy consumption with over 100 W DC input power at full load, where a 10 MHz macro cell typical radiates up to 40 W output power. The lower curves show an improved PA with adaptability of the operation point by changes in the DC voltage. The use of a lower voltage limits the maximum output

2012 ref Micro DTX MBSFN Short DTX0

0.1

0.2

0.3

0.4

0.5

P/A

[kW

/km

2]

2012 power model, medium traffic density

0.46

0%

0.39

-15%

0.37

-19%

0.33

-28%

2012 ref Micro DTX MBSFN Short DTX0

0.1

0.2

0.3

0.4

E/B

[kJ/M

bit

]

0.19

0%0.16

-15%

0.15

-19%0.14-28%

EARTH PROJECT

EARTH_WP4_D4.3 54 / 121

power due to saturation of the PA. The deactivation of the PA enables a micro sleep mode that allows very low power consumption of only 4.4 W during OFDM symbols with no transmission.

Other components, which make a significant contribution to the power consumption of a cell, are baseband processing, RF part, cooling, the limited efficiency of DCDC conversion, and ACDC power supply. A load dependent power consumption that is much lower as for SOTA has been considered for these other components [EARTH-D2.3]. For more efficient BSs, especially with the PA mounted close to the antenna (“remote radio head”, RRH), even passive cooling by air convection can be sufficient, which requires no power consumption compared to an indoor BS system with constant high power consumption. Therefore, improvements by these other components including the PA have been additionally considered.

FIGURE 40. PA characteristics over RF output, depending on Operation Point settings (OPi with i=1,…,8).

The scheduler sets the principal physical parameters for the transmission at each sub-frame and, in consequence, decides about the radiated power. For the BW and CAP adaptation modes the scheduler adapts resources to bandwidths by equal utilization of PRBs over a given time period. Depending on this load an optimised PA characteristic with limited output RF power can be used. On the other hand, a scheduler can reserve some of the time intervals (micro DTX) and set an energy saving deactivation mode of the PA during these times. The difference between the resource utilisation strategies is shown in FIGURE 41 for an example scenario which considers a BS loaded with 40% user data.

(a) (b)

FIGURE 41. Resource utilisation per symbol at 40% user load: (a) BW adaptation, (b) micro DTX.

0

10

20

30

40

50

60

70

80

90

100

110

0 5 10 15 20 25 30 35 40

Po

we

r C

on

sum

pti

on

of P

A [

W]

RF Output Power [W]

OP1=SOTA

PA Deactivation

OP2

OP3

OP4

OP5

OP6

OP7

OP8

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Load

per

Sym

bo

l

OFDM Symbols along 10 ms0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Load

per

Sym

bo

l

OFDM Symbols along 10 ms

EARTH PROJECT

EARTH_WP4_D4.3 55 / 121

The scheduled user data is shown under consideration of the same fixed modulation and coding scheme for a standard LTE radio frame as illustrated in FIGURE 103. For the used example a bandwidth of 5 MHz meets already the load requirements for the BW adaptation approach compared to 10 MHz for the micro DTX approach. Therefore, less reference signals are transmitted for the BW adaptation approach.

The evaluated daily total power consumption of a cell is illustrated for all scheduler strategies in FIGURE 42 both for a typically dense urban (a) and rural traffic scenario (b). FIGURE 43 shows the energy savings per cell for these green scheduling strategies for improvements by the PA only (a) and improvements by all other components including the PA (b).

(a) (b)

FIGURE 42. Daily power consumption of a cell for SOTA and improvements by the PA only based on different scheduling strategies in dense urban scenario (a) and rural scenario (b).

(a) (b)

FIGURE 43. Daily energy savings for different scheduling strategies by PA only (a), by PA and all other components (b).

A comparison of not combined solutions as BW adaptation, CAP adaptation and micro DTX shows that during night at lowest traffic load highest energy savings can be achieved by using the BW adaptation approach. Micro DTX is more beneficial at busy hours in dense urban scenarios, because due to the traffic load BW adaptation here cannot use one of the adaptation modes. The CAP adaptation approach yields the lowest energy savings along a day.

0

20

40

60

80

100

120

140

160

0 4 8 12 16 20 24

Co

nsu

me

d P

ow

er

pe

r C

ell

[W]

Time [h]

SOTA

BW Adaptation

CAP Adaptation

Micro DTX

BW Adaptation and Micro DTX

CAP Adaptation and Micro DTX

Dense Urban

0

20

40

60

80

100

120

140

0 4 8 12 16 20 24

Co

nsu

me

d P

ow

er

pe

r C

ell

[W]

Time [h]

SOTA

BW Adaptation

CAP Adaptation

Micro DTX

BW Adaptation and Micro DTX


Rural

14.813.1

16.7

19.7

17.3

23.9

17.9

21.3

27.4

23.1

0

5

10

15

20

25

30

BW Adaptation CAP Adaptation Micro DTX BW Adaptation and Micro DTX


Ene

rgy

Savi

ngs

[%]

Dense Urban

Rural

Improvements by PA only

44.642.9

46.749.8

47.5

57.4

51.4

56.0

61.857.9

0

10

20

30

40

50

60

70

BW Adaptation CAP Adaptation Micro DTX BW Adaptation and Micro DTX


Ene

rgy

Savi

ngs

[%]

Dense Urban

Rural

Improvements by PA and all other components

EARTH PROJECT

EARTH_WP4_D4.3 56 / 121

Highest energy savings up to 19.7% and 27.4% can be achieved by using BW adaptation combined with micro DTX in a dense urban scenario and rural scenario, respectively, if only the improvements of the new hardware features of the PA are considered. Much higher energy savings up to 49.8% and 61.8% can be achieved in dense urban and rural scenarios, respectively, when considering not only the improvements by the PA but also the improvements of all other components of a cell, especially by using passive cooling. The second best result is based on a combination of CAP adaptation and micro DTX. In this case slightly lower energy savings can be achieved.

The EARTH power model and traffic profile have been applied to compute energy consumptions of LTE deployments. Different integrated energy saving approaches have been compared by applying either a scheduling policy with adapted BW or CAP using a PA with adaptive operation point or a micro DTX scheduler using a PA with a deactivation mode or combinations of both approaches.

At rather high resource utilisation micro DTX can more flexibly utilise the void resources, where the fixed steps of BW or CAP adaptation limits the saving potential. For low utilisation BW adaptation has the highest saving because rather low efficiency for sending reference signals limits the saving for micro DTX and CAP adaptation. Applying a combination of BW adaptation and micro DTX is the best solution according to energy savings, because this approach takes always benefits from low and high load conditions. For all strategies higher savings can be achieved in rural cells compared to dense urban cells, because the lower resource utilisation leaves more scheduling potential for energy savings.

5.3. ADAPTIVE ENERGY EFFICIENT SCHEDULING ALGORITHM FOR LTE PICO BASE STATIONS

In this approach, we try to reduce the energy per information bit of a pico base station in the downlink by transmitting to a UE only when the channel conditions in a resource block are good enough to support a high MCS and sleeping the rest of the time. We then buffer packets that are not delay-sensitive until there are good channel conditions defined by a CQI-threshold. If the packet is delay-sensitive, it is transmitted even with unfavourable channel conditions. The advantage is twofold. First the energy per information bit is lower with higher modulation and coding schemes (MCS). Second, by transmitting at a higher MCS, we can benefit from more sleeping periods. In a multi-user environment, it is unlikely that all the users experience the same channel quality. Therefore, we avoid the transmission of data for the users experiencing a bad channel at a certain moment. Even more, in a small cell scenario, given the reduced number of users, the opportunities to put the system into full sleep mode are higher compared to a large cell. The impact on average and maximum packet delay is analyzed with system-level simulations. This approach uses the 3GPP standard signalling and is hence standard-compliant. This approach can be shown descriptively in FIGURE 44.

FIGURE 44. Energy efficient scheduling algorithm: send packets when high MCS can be used and sleep the rest of the time.

EARTH PROJECT

EARTH_WP4_D4.3 57 / 121

Using the EARTH power models, we can compute the power consumed by a pico base station working full load using each of the MCS defined by the LTE standard. Even though higher loads consume more power, the corresponding information bit rate is also larger. Hence, we propose to use the energy consumption per information bit as the metric to evaluate energy efficiency. This can be seen in FIGURE 45, where higher orders of modulation and coding rate are much more energy efficient since they can process a larger amount of information bits for a marginal increase in power consumption. In fact, the highest transmission mode is around 25 times more energy efficient than the lowest transmission mode.

FIGURE 45. Energy per information bit for all the CQI values.

The drawback of not transmitting during periods with bad channel conditions is the increase in packet delay. However, we define a maximum packet delay (delaymax) as the maximum time a packet can be waiting in a queue before it is sent. If this limit is reached, the packet is transmitted even with bad channel conditions to satisfy the required quality of service (QoS). For high priority packets a lower delaymax can be implemented. The threshold that we use to define whether channel conditions are good enough to send a packet in a resource block is based on the CQI reported by the UEs. If the CQI for a resource block is larger than a certain threshold, we transmit a packet with the modulation and coding rate indicated by the CQI. Otherwise we buffer it until the channel conditions improve, i.e. until the reported CQI is higher than the threshold or the maximum packet delay is reached. In these simulations we assume that one packet fits in one resource block.

The base station can adaptively tune the CQI threshold per user, increasing it if the average packet delay does not exceed a certain limit (avg_delaymax) and decreasing it in the opposite case. The initial CQI threshold could be set based on previous measures of average CQI per user. In this way we try to minimize the average energy per information bit of the system in a dynamic way. The current packet delay can be computed as the time a packet has spent waiting in a queue without being transmitted. The average packet delay computes an average over the previous packets waiting in a queue.

To validate this approach with realistic channel and network conditions we use the following scenario. We simulate the downlink scheduling behaviour of a pico base station with 10MHz bandwidth, 2 antennas, and, for simplicity purposes, 3 users randomly located in a radius of 60 m in a full LTE simulation framework. In this approach we assign resources to the UEs every transmission time interval (TTI) of 1 ms. We consider a constant application throughput that sends packets to the queue of each user with a Poison distribution with 2.5 ms in average. We also consider a maximum packet delay of 20 ms per packet to satisfy the required QoS. If a packet is buffered longer than 20 ms, it is sent in the next TTI even if the reported CQI is lower than the threshold. In FIGURE 46 we observe a snapshot of the variation of a resource block CQI reported by 3 UEs. It was obtained using a system-level single-cell multi-user LTE simulation framework. We observe that the dynamics

EARTH PROJECT

EARTH_WP4_D4.3 58 / 121

of the channel conditions can vary the CQI at every TTI. This approach benefits from this dynamism by scaling down the system power to serve only the UEs with best channel conditions at each moment in time.

FIGURE 46. CQI reporting for a resource block for three UEs.

Under these channel conditions, we evaluate the energy per information bit using the power models of EARTH project. FIGURE 47 shows the instantaneous (inst) and the average (avg) energy per information bit with different CQI thresholds compared to the case with no threshold and the case with the low-speed low-power transmission scheme reported in [CaG10]. The energy difference between the no-threshold and the low-power scheme resides in the fact that the first one uses sleeping periods when no transmission is required. Therefore, the low-power transmission scheme has a large penalty due to the long-lasting active state. This can be also explained by the fact that Dct is more energy efficient than Dcf. With the proposed threshold approach, the average energy per bit can be further reduced by 8.5%, 25% and 61% compared to the no-threshold approach as we increase the CQI threshold by 9, 11, and 13 at the cost of increasing the average delay by 0.05 ms, 0.15 ms, and 6 ms, which is acceptable even for VoIP communication which has an average packet delay of 20 ms and a maximum of 150 ms (TABLE 6).

FIGURE 47. Instantaneous and average energy consumption per information bit for the proposed approach with different CQI thresholds compared to the no-threshold and the low-power approaches.

EARTH PROJECT

EARTH_WP4_D4.3 59 / 121

TABLE 6. Energy per bit reduction and average packet delay for different CQI thresholds.

CQI threshold Energy per bit reduction Average Delay (ms)

9 8.5% 0.05

11 25% 0.15

13 61% 6

In the case of delay-sensitive applications, the energy gains are reduced since the margin left to buffer packets is reduced. Also, the more variable the channel conditions the more opportunities to exploit. This allows the base station to scale down the energy consumption of the system more often and to serve only those users that have good channel conditions at each moment in time.

The main area of application of this solution is the scenario where the network load can be considered low or medium and there are hence margins in the application layer requirements to buffer or delay packets. More specifically this solution finds the largest potential when user applications are less delay or throughput sensitive and can allow some buffering of the packets until good channel conditions arrive. In a small cell environment, dips in the channel frequency response can be common but usually not long lasting. This allows the base station to scale down the system and to serve only those users that have good channel conditions reported at a certain time slot. As such, the optimization time scale of this approach is in the range of milliseconds.

This approach can fit perfectly in the rest of the solutions at network, system, and node level. It is based on buffering packets at millisecond level based on the channel conditions reported by the users; therefore all the techniques considering MIMO techniques, beamforming, and energy efficiency transceivers and antennas are independent from the proposed approach, so they can be perfectly coupled because their area of application is different. Particular attention is needed with solutions such as bandwidth adaptation (fast) and Cell DTX (micro) since they have the same area of application. Since the buffering of packets might slightly increase the network load in future time slots, these might impact the sleeping periods or the adaptation of the bandwidth in such periods. However, we expect this impact to be marginal in the case of bandwidth adaptation and beneficial in the case of sleeping periods.

5.4. BENEFIT ON ENERGY EFFICIENCY DUE TO RADIO INTERFACE TECHNIQUES

Three energy saving approaches based on scheduling solutions operating in time and frequency domain have been investigated, taking benefit from energy efficiency enabling hardware features and leading to a maximum of power reduction.

Simulations of cell DTX solutions, which are based on duty-cycling in time combined with sleep modes in BS components show up to 28% of daily energy saving when applying the short DTX scheme. This average value is weighted over different deployment scenarios describing the performance of a complete radio network. Only the deactivation of the PA is considered for this evaluation as energy efficiency enabling hardware feature. As short DTX cannot be applied with current mobile standards, appropriate modifications have to be done in new standard releases to enable this high amount of energy saving. These modifications shall be specified by 3GPP within Release 11.

A standard conform application of the presented scheduling solutions is the combination of slow BW adaptation, which may impact the performance of cell-edge users, with micro-DTX, a combination of duty-cycling in time and in frequency approaches. When considering the benefit only due to new hardware features of the PA, up to 19.7% and 27.4% energy savings can be achieved in dense urban and rural scenarios.

EARTH PROJECT

EARTH_WP4_D4.3 60 / 121

Considering the improvements by the PA together with all other components of the BS, higher energy savings up to 49.8% and 61.8% can be achieved in dense urban and rural scenarios, respectively. By CAP adaptation instead of BW adaptation in combination with micro DTX, slightly lower energy savings (47.5% respectively 57.9%) can be achieved. Therefore, the preferred solution is a combination of BW adaptation and micro DTX even if it requires some changes in the LTE standard for applying a fast CA procedure under consideration of small channel bandwidths.

A proposed energy opportunistic scheduling algorithm leads to decreased energy per bit values by making use from good channel conditions and a threshold for the introduced transmission delay. By applying this solution for small-cell scenarios in combination with micro DTX, the average energy per bit can be reduced by 8.5%, 25% or 61% compared to the no-threshold approach at the cost of increasing the average delay by 0.05 ms, 0.15 ms, or 6 ms, which is considered as acceptable even for voice communication.

The presented energy saving approaches are beneficial in different deployment scenarios and traffic conditions. At low resource block utilization (e.g. during the night) CAP and BW adaptation are efficient techniques while at high resource block utilization (e.g. during busy hours in dense urban scenarios), micro DTX is more favourable. A combination of slow BW adaptation and micro DTX is the best energy saving solution in variable load conditions. Although not currently supported in LTE, short and long cell DTX can bring further gains if the LTE standard is modified accordingly as expected.

The energy opportunistic scheduling algorithm applied in small cells lead to additional energy saving. This solution shows the highest benefit for user applications which are less delay or throughput sensitive and if additional energy saving techniques such as MIMO, beam forming, energy efficient transceivers and antennas are used in combination.

EARTH PROJECT

EARTH_WP4_D4.3 61 / 121

6. BEAMFORMING IN RECONFIGURABLE AND ADAPTIVE ANTENNA SYSTEMS

Multi-antenna techniques are usually seen in radio access networks as a way to provide increased system capacity and beamforming is one of the key techniques to achieve this goal. Anyway, it is now generally known that beamforming could be analyzed not only from the capacity increase point of view, but also focusing on the potential to reduce power consumption, and, therefore, improving energy efficiency in LTE.

There are different applications of the beamforming technique, with different impacts in terms of achievable improvements in energy efficiency, FIGURE 48.

In a first application potential gains from "reconfigurable antennas" are evaluated in a scenario where antenna parameters are adapted specifically according to spatial changes in traffic situation but also to changes in traffic load. Adaptation is done on a fairly slow time frame, in the order of minutes or longer, and it is typically applied to all antenna ports in a cell. This is sometimes denoted as cell-specific beamforming.

The second application, about "adaptive antennas", addresses the gains from applying user specific beamforming, where each user, or group of users, are served by a beam uniquely designed for them on a very short time frame. The user specific beam is generated by applying appropriate weighted signals to the multiple antenna ports in the cell.

FIGURE 48. Reconfigurable and Adaptive Beamforming.

For both solutions it is possible to integrate the expected gains with the usage of “active antennas”, where the losses due to connecting cables in the transmitters are completely nulled, hence, ensuring a reduction in the consumed power with equal performances.

The analysis performed in this section have been led in compliance with the simulation set-up described in [EARTH-D2.2] and according to the E3F model introduced in [EARTH-D2.3]. In particular, for the adaptive case, the simulations have been done in a dense urban scenario only, either for full load case and for lower traffic levels. The power model for active antennas has been obtained as an application of the power model examples described in [EARTH-D2.3].

6.1. RECONFIGURABLE ANTENNA SYSTEMS

The concept of the Reconfigurable Antenna system (RAS) addressed herein is to adapt antenna parameters to traffic conditions in contrast to a conventional, reference, system (Ref) where antenna parameters typically are static over very a long time scale. The three antenna parameters considered are the ones defining the cell, beam tilt in elevation and beam pointing and beam width in azimuth, and for that reason the technique is sometimes called “cell specific beamforming”. The time constant for the adaption in a real system would be fairly long, hours or longer, in order to capture “average” conditions both with respect to spatial distribution of the traffic as well as the load in the system. In this evaluation antenna parameters are found basically by

EARTH PROJECT

EARTH_WP4_D4.3 62 / 121

means of exhaustive search, i.e. testing all parameter combinations. This is of course not applicable in a live system, so some performance degradation, the magnitude has not been analyzed, can be expected when a realistic algorithm is applied, compared to the savings presented here.

Only three sector sites have been considered but the concept is closely related to “Dynamic sectorization”, where the number of sectors is changed depending on system load, typically requiring antenna parameters in active sectors to be changed [EARTH-D3.2].

The fundamental principle of the concept is to adapt antenna parameters such that link budget in general is improved leading to the average resource element utilization being decreased, and by that also the energy consumption, while still serving the traffic offered to the system. The stronger the relation is between power consumption and resource element utilization the larger energy savings can be done. For this reason three power models are evaluated. A first power model is the baseline power model, year 2012, as defined in [EARTH-D2.3]. The second model reflects power consumption for improved hardware as proposed in chapter 2 and finally the third model is identical to the second one except that sleep mode (sm) is ON. It shall be noted that both RAS and Ref systems are evaluated using the same power model so the savings identified comes in addition to the savings from improving the hardware itself as reflected in the power model. More details about the simulator can be found in appendix 10.5.1.

The deployment scenario studied here is the dense urban scenario and traffic scenarios 2 and 3, [EARTH-D2.2], plus scenarios with even higher load. The traffic is assumed to be spatially non-uniform which is modelled by means of hotspots with an increased offered traffic density.

An example of cell plans for RAS and Ref systems is shown in FIGURE 49. The system consists of 12 unique cells and wrap around is applied to mimic a large system. The areas covered by each of the 12 unique cells are coded in colour whereas antenna pointing directions and azimuth beamwidths are indicated via legends. The brightness of the legend indicates antenna tilt where darker means larger tilt. The grey areas show in this example 10 randomly dropped hotspots within the area covered by the 12 cells. FIGURE 49 clearly shows that the, for RAS, cell unique antenna parameters impacts the cell coverage which is no longer identical for all cells as is the case for the Ref system.

(a) (b)

FIGURE 49. Example of system layout with reconfigurable antennas (a) and conventional antennas (b).

Three different parameters are used for describing the traffic situation; the offered traffic; the number of hotspots (NHS) and the probability for hotspot traffic (PHS). Power savings by using reconfigurable antenna system are evaluated along each of these three dimensions.

EARTH PROJECT

EARTH_WP4_D4.3 63 / 121

6.1.1. Offered traffic

The power consumption per area unit as a function of offered traffic is shown in FIGURE 50 for a scenario with 10 hotspots and hotspot probability of 39% for the three power models being considered. The power consumption is averaged over 30 random drops of hotspots where, for each drop, antenna parameters for the RAS system are selected to minimize the power consumption. Markers, stars (*) for RAS and circles (o) for Ref, indicate “outage” which here means that the offered traffic is not served in all cells for all drops. As can be seen the RAS can serve a much higher offered traffic than the Ref system. Comparing power consumptions when one system serves the offered traffic whereas the others do not (in outage) is unfair. For this reason power consumptions are primarily estimated for the actually served traffic and then up scaled to correspond to the offered traffic.

FIGURE 50. Power consumption for RAS and Ref systems as a function of offered traffic.

Relative power consumption is shown in FIGURE 51 where it is clearly seen that savings increase with offered traffic giving that the major savings appear when the offered traffic is high.

FIGURE 51. Relative power consumption for RAS versus Ref systems as a function of offered traffic.

EARTH PROJECT

EARTH_WP4_D4.3 64 / 121

In accordance with [EARTH-D2.3] power savings are estimated over daily variations in traffic load. The peak offered traffic for deployment scenario 1 shall be 52 Mbps/km2 for traffic scenario 2 and 92 Mbps/km2 for traffic scenario 3. Relative power consumptions per hour for an offered traffic of 92 Mbps/km2 and 194 Mbps/km2 are shown in FIGURE 52. As can be seen from FIGURE 52 the savings for 92 Mbps/km2 are quite small and the relative power consumptions averaged over 24 h are only a few percent. Savings for 52 Mbps/km2 will be even lower and have not been estimated. Savings for a peak offered traffic of 194 Mbps/km2, the value where outage starts to appear for the Ref system, is estimated in order to show the potential of the RAS system to handle higher loads. The savings for 194 Mbps/km2, averaged over 24 h, range from approximately 2 to 8% depending on the power model whereas savings during high traffic hours can be up to 15%.

FIGURE 52. Relative power consumption for RAS versus Ref systems over 24h for offered traffic of 92 Mbps/km2 (upper) and 194 Mbps/km2 (lower).

6.1.2. Number of hotspots and hotspot probability

The number of hotspots has been varied, for a given hotspot probability of 39%, from 1, which models an extreme clustering of the offered traffic, up to 67 for which the traffic becomes spatially uniform for the given hotspot probability. FIGURE 53 shows that the power saving, here exemplified for power model Earth ON (EARTH improved hardware used) and sm ON (micro DTX used), depend slightly on the number of hotspots but the more interesting result is that the RAS system is capable of handling a much wider range of hotspots than the Ref system. This can be seen from the Ref system being in outage for multiple settings of number of hotspots (indicated by circles).

EARTH PROJECT

EARTH_WP4_D4.3 65 / 121

FIGURE 53. Power consumption per cell for RAS and Ref systems as a function of number of hot spots.

Power consumption per cell as a function of hotspot probability is shown in FIGURE 54. Mean power consumption is lower for the RAS compared to the Ref system. It is also clear that the RAS system has lower maximal resource element utilization than the Ref system. For this power model 100% resource utilization corresponds to 344 W so for the highest hotspot probability the Ref system serves the offered traffic with no margins.

FIGURE 54. Power consumption per cell for RAS and Ref systems as a function of hotspot probability.

6.2. ADAPTIVE BEAMFORMING ON ACTIVE ANTENNAS

From a general point of view the principle of adaptive beamforming is that the signal arriving at each antenna element is individually weighted to form a beam into the direction of interest; the signals of all elements of the antenna are then combined in order to create a single signal that will be captured by the receiver. At the same time, a feedback process adjusts the weights of each antenna element.

There are several algorithms able to implement optimal weighting, differing fundamentally on the speed of convergence. Although the adaptive beamforming technique is applied at the BS on DL, the algorithms for optimum weighting are generally based on a UL reference signal that is used to track the user’s location and set up the weighting vector. One example of these algorithms is the minimum mean-square error (MMSE), with details in Appendix 10.5.3.

In this framework the algorithm adopted in EARTH has been based on the so-called DoA method, estimating in an ideal way the direction of arrival of the UL signal on a per user basis (see [EARTH-D4.1] for details on the DoA method). Moreover, EARTH has considered the application of such beamforming algorithms on a per user basis together with Active Antennas, adding gains in terms of energy efficiency.

EARTH PROJECT

EARTH_WP4_D4.3 66 / 121

6.2.1. Active Antennas

As a general definition, an Active Antenna System (AAS) includes a certain number of active antenna elements arranged in an array and allowing the radiation pattern shaping to be remotely controlled. The array could be developed either in the vertical direction only or in a planar arrangement or in other geometries, depending on the beam shaping needs. Each active antenna elements contain a number of traditional antenna elements (dipoles, patches or other geometries) and a complete radio transceiver digitally connected to a Beamforming Unit able to perform digital beamforming processing.

The system introduced in EARTH is able to perform a “per user” beamforming in a digital way directly in the AAS itself, representing a huge step forward with respect to traditional solutions, where the digital processing for the beamforming is managed centrally and not in the antenna, causing an often intolerably high amount of data to be transmitted to the antenna elements (see Appendix 10.4 for details on the AAS architecture), [EARTH-D4.1].

As well as ensuring a more convenient application of beamforming algorithms, especially the adaptive ones, AAS by themselves are an interesting energy efficiency enabler, due to the fact that no losses are any more to be included in the link budget with respect to traditional layouts. A representation of the gains due to AAS in the scenario of DoA adaptive beamforming is reported in FIGURE 55, showing that they are in the range of 40% with respect to the baseline EARTH scenario. The gains are reported in terms of Energy Consumption Indicator (ECI). In FIGURE 56 the ECI is reported in the daily representation. Results have been computed for the dense urban scenario as defined in [EARTH-D2.3].

FIGURE 55. Energy Consumption Indicator of AAS with reference to the baseline.

00,10,20,30,40,50,60,70,80,9

1

0 0,02 0,04 0,06 0,08 0,1 0,12 0,14

ECI [kJ/Mbit]

Energy Consumption Indicator CDF

ECI AAS

ECI Baseline

EARTH PROJECT

EARTH_WP4_D4.3 67 / 121

FIGURE 56. Daily ECI of AAS with reference to the baseline.

6.2.2. Adaptive Beamforming

The DoA algorithm has been described in previous EARTH deliverables, e.g., [EARTH-D4.1]. Some more details are reported for convenience in Appendix nn, together with other options that can be used for adaptive beamforming. Some results were presented in [EARETH-D4.1] and [EARTJ-D4.2]; in the first deliverable results were based on the full load case within E3F, in the second they were about lower levels of load but with an ideal and simplified power model.

In the following some results are shown regarding the variable load cases with a dedicated power model, either for the baseline in E3F and for the AAS case. In the AAS power model the EARTH power model has been applied to a different architecture, an active antenna with 4 active elements and a PA tailored to the lower needed power levels. The different load levels were simulated by varying the frequency resource utilization variable in the EARTH power model script.

In the following figures (FIGURE 57 and FIGURE 58) the case with DoA beamforming only is considered on top of the baseline scenario; the estimated benefits in terms of energy efficiency are in the order of 9%.

FIGURE 57. Energy Consumption Indicator of DoA BF with reference to the baseline.

0

0,02

0,04

0,06

0,08

0,1

0,12

0,14

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

EC

I [k

J/M

bit]

Time

Energy Consumption [dense urban]

Day ECI BF Day ECI BASELINE

00,10,20,30,40,50,60,70,80,9

1

0 0,02 0,04 0,06 0,08 0,1 0,12 0,14 0,16

ECI [kJ/Mbit]


ECI BF

ECI Baseline

EARTH PROJECT

EARTH_WP4_D4.3 68 / 121

FIGURE 58. Daily ECI of DoA BF with reference to the baseline.

In the second set of results (FIGURE 59 and FIGURE 60) the DoA beamforming has been applied on top of a system equipped by AAS, the ones described above. A superimposition of the effects takes place and the overall gains ramp up to about 46%.

FIGURE 59. Energy Consumption Indicator of DoA BF and AAS with reference to the baseline.

0

0,02

0,04

0,06

0,08

0,1

0,12

0,14

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

EC

I [k

J/M

bit]

Time



00,10,20,30,40,50,60,70,80,9

1

0 0,02 0,04 0,06 0,08 0,1 0,12 0,14

ECI [kJ/Mbit]


ECI BF

ECI Baseline

EARTH PROJECT

EARTH_WP4_D4.3 69 / 121

FIGURE 60. Daily ECI of DoA BF and AAS with reference to the baseline.

The results shown here are for the dense urban scenario, as defined in [EARTH-D2.3]. Further evaluations for other scenarios suggest that the single DoA approach could be very beneficial also for the Rural Macro scenario, as always defined in [EARTH-D2.3], where the above mentioned gains could be higher of about a further 5%.

6.3. BENEFIT ON ENERGY EFFICIENCY DUE TO BEAMFORMING AND ACTIVE ANTENNAS

Beamforming techniques have been analysed in this chapter, also in possible coexistence with active antenna systems. Even if the methods analysed here have been dealt with as a single innovation, it has to be pointed out that each of them, i.e. reconfigurable beamforming, adaptive beamforming and active antennas, are valuable actuators of improved energy efficiency.

In particular active antennas ensure an improvement simply adopting them instead of traditional antennas, due to the reduction of power losses in the cables connecting radio frequency modules to passive antennas. In this sense, then, active antennas can be used in all the scenarios identified in [EARTH-WP6] (dense urban, urban, rural…), even if mainly for macro-cellular solutions. In the simulations performed for adaptive beamforming in the case of dense urban environment the simple usage of active antennas enabled a gain estimated in the range of 40% with respect to the EARTH baseline.

Regarding the beamforming techniques, they have been studied in EARTH in the framework of dense urban scenarios, where most of the traffic is generally considered. In such a scenario the cell specific, reconfigurable beamforming method has shown a possible gain estimated in about 8% with respect to the baseline. It’s to be noted that this gain should be exploited in medium/long term variations of traffic, for the inherent features of the reconfiguration, slowly following the users in the coverage area.

As for the adaptive approach, it is by nature a fast follower of the traffic and hence is categorized in the short term innovations. This user specific solution shows 9% of power reduction in the urban scenario. When using this adaptive beamforming with active antennas, a total gain of energy efficiency improvement of around 46% with respect to the baseline is expected. When applying these combined solutions for rural macro environment, a further 5% of gain is obtained.

0

0,02

0,04

0,06

0,08

0,1

0,12

0,14

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

EC

I [k

J/M

bit]

Time



EARTH PROJECT

EARTH_WP4_D4.3 70 / 121

7. SMART MIMO OPERATION

Multiple-input multiple-output (MIMO) techniques are essential to achieve high data rates in wireless systems. Today they are used in all wireless standards, and in LTE and LTE-advanced up to 8 antennas are supported. However, having multiple antennas might not be the best choice when dealing with energy efficiency. Therefore, in this chapter we analyze different MIMO configurations under an energy-efficient point of view. In scenarios with low network loads, the MIMO configuration could be switched to reduce the energy consumption while meeting the data rate constraints.

The main concepts that we explain in these chapters deal with 3 main topics. In section 7.1, an energy efficient usage of transmit antennas is explained. This is combined with the modulation and coding rate necessary to achieve a target data rate in a low loaded cell. Section 7.1.1 focuses on picocells while section 7.1.2, on macrocells. In order to benefit from an energy efficient resource allocation both duty-cycling in time and in frequency are explored.

Antenna muting is also related to the adaptation of the number of transmit antennas by activating antennas only when motivated by traffic load and/or data rate but not reconfiguring the number of antennas in the cell. The implementation of the antenna muting in LTE is discussed in section 7.2.

In the case of a high network load, precoding schemes can help to direct the transmitted power on each antenna through the choice of precoder coefficients. This is done in section 0 where an energy efficient LTE-compliant precoding for multi-user MIMO (MU-MIMO) achieving a low PAPR is proposed.

Finally, general conclusions on the benefit of MIMO usage from the energy efficiency point of view are drawn in Section 7.4.

7.1. ENERGY EFFICIENT RESOURCE ALLOCATION USING MULTIPLE TRANSMIT ANTENNAS

MIMO schemes are now used in most wireless systems, including LTE and LTE-advanced. Selecting the best number of active transmit antennas at the base station as function of the channel and network conditions is crucial for maximum energy efficiency. Evidently, this selection, together with link parameters such as the modulation and coding rate, can also affect the peak throughput and change the system frequency and time duty-cycling based on the required data rate. We call both the selection of active transmit antennas and the link-level parameters the MIMO mode. MIMO mode selection could bring energy savings by scaling down the MIMO processing and the number of active antenna chains. Even though the use of MIMO is necessary for high throughput scenarios, important energy gains can be achieved through the use of a single antenna whenever the network load requirements are low. In the following sections we present the downscaling approach for a picocell and a macrocell, respectively.

7.1.1. MIMO Power-Performance Optimization in Picocells

In this section, we analyse the impact on the energy consumption of selecting the best MIMO mode that satisfies the required throughput in a picocell.

For this purpose, we carried out link-level simulations based on channel, performance, and power models. The goal of the channel and performance modelling is the derivation of the link SNR-PER (Signal to Noise Ratio – Packet Error Rate) characteristics that will help us to select the best number of transmit antennas and link parameters that achieve a target throughput. Such modelling builds on a single-cell standard-compliant chain with all realistic effects including non-idealities. By combining PER values with the throughput derived from the selected link parameters (modulation and coding scheme or MCS), we can determine the goodput (useful data rate from correct packets) of the system, which is proportional to the net spectral efficiency. The work on

EARTH PROJECT

EARTH_WP4_D4.3 71 / 121

power modelling that is reported in [EARTH-D2.3] is used in order to provide consistent and comparable power consumption values.

In order to allow energy saving through duty-cycling, two ways are considered to down-scale the system power when the total available goodput is higher than the required application data rate. The first one is based on duty-cycling in frequency (P_Df), which means not loading all the subcarriers (PRBs) of the system bandwidth when not needed. The second one uses duty-cycling in time (P_Dt), which means to load all subcarriers during data transmission and applying sleep mode in time periods without data transmission, as long as the required throughput can be achieved on average.

The targeted throughput was selected corresponding to what the system would provide working at 100% duty-cycling when using a spectral efficiency (Eff) of 1 bit/s/Hz. Hence, for example, using MIMO 2x2 with QPSK and coding rate 1/2 leads a spectral efficiency of 2, meaning that a duty-cycling of 50% is sufficient for the targeted throughput, and enables some savings by reducing the system activation.

TABLE 7. Results of SISO and MIMO 2x2 and4x4 modes for minimal power consumption as function of SNR and duty-cycling for a given throughput.

SISO MIMO 2x2 MIMO 4x4

SNR MCS Eff. P_Df (W)

P_Dt (W)

MCS Eff. P_Df (W)

P_Dt (W)

MCS Eff. P_Df (W)

P_Dt (W)

5 dB / 0 / 0 / 0

10 dB

QPSK, 1/3

0.7 3.9 3.9 QPSK, 1/2 2.0 7.0 4.1 / 0

15 dB

16-QAM, 1/3

1.3 2.8 2.1 QPSK, 2/3 2.7 6.1 3.1 / 0

20 dB

16-QAM, 1/2

2.0 2.4 1.4 64-QAM, 1/3

4.0 6.0 2.1 / 0

25 dB

16-QAM, 7/8

3.5 2.2 0.9 64-QAM, 1/2

6.0 5. 6 1.4 / 0

30 dB

64-QAM, 2/3

4.0 2.1 0.8 16-QAM, 5/6

6.7 5. 6 1.2 QPSK, 1/3

2.7 11.0 5.9

35 dB

64-QAM, 3/4

4.5 2.0 0.7 64-QAM, 2/3

8.0 5.5 1.0 16-QAM, 1/3

5.3 10.2 3.0

40 dB

64-QAM, 3/4

4.5 2.0 0.7 64-QAM, 2/3

8.0 5.5 1.0 64-QAM, 1/3

8.0 10.1 2.0

45 dB

64-QAM, 3/4

4.5 2.0 0.7 64-QAM, 3/4

9.0 5.5 0.9 64-QAM, 1/3

8.0 10.1 2.0

EARTH PROJECT

EARTH_WP4_D4.3 72 / 121

The power numbers are presented in TABLE 7. The number of active receive antennas at the UE are assumed to be the same as the base station to achieve a proportional data-rate increase with the number of active transmit antennas.

From TABLE 7 one can observe that time-domain duty-cycling is much better for energy savings than frequency-domain duty-cycling. The reason is that frequency-domain duty-cycling does not enable to have many components sleeping. In fact, the full analogue front-end, PA, and time-domain digital signal processing are continuously active, only the frequency-domain processing of symbols benefits from this duty-cycling. On the other hand, time-domain duty-cycling allows all components to sleep during the symbols without transmission. This is independent of the number of transmit antennas.

Furthermore, it is observed that the use of multiple antennas, does not lead to energy savings, FIGURE 61. The primary goal of MIMO is indeed to increase the spectral efficiency, which is the main reason why it is deployed in cellular base stations. The energy saving mode actually comes from down-scaling to one antenna when the full system capacity is not required, letting one or more chains completely in sleep. From simulations a factor of 15 times(11 W for MIMO 4x4 P_Df and 0.7 W for SISO P_Dt) variation between the worst-case power consumption and the optimally selected MIMO mode and a factor of 10 between the reference mode and the optimally selected MIMO mode (7 W for MIMO 2x2 P_Df and 0.7 W for SISO P_Dt) is obtained. This is assuming the constraint of a fixed output power spectral density as in LTE. Relaxing this constraint and focusing only on the output power could open the door to additional savings.

FIGURE 61. Power consumption as a function of SNR by using the most energy-efficient link adaptation.

7.1.2. MIMO and Resource Allocation Optimization in Macrocells

Energy efficiency of MIMO schemes with different channel knowledge and the adaptation of the number of used transmit antennas are considered in this section. The purpose of this section is also to study potential energy savings of MIMO schemes in LTE compliant system level simulations with different resource allocation strategies. One resource allocation strategy is aiming at energy savings minimizing the number of used PRBs in frequency domain whereas another method is based on discontinuous transmission (DTX).

Fixed codebook based MIMO precoding or open loop MIMO methods and sub-block based channel quality indicator (CQI) are usually utilized in LTE systems. The spectral efficiency and energy efficiency can be improved by using further available channel knowledge. For example, the channel reciprocity can be utilized in TDD systems to obtain channel state information (CSI). The availability of CSI at the base station enables good beamforming and PRB-wise channel frequency response. Due to the improved spectral efficiency, less PRBs

EARTH PROJECT

EARTH_WP4_D4.3 73 / 121

are needed to serve users when CSI is utilized. Full CSI also enables the so called multi-user MIMO (MU-MIMO) techniques which improves the energy and spectral efficiency of the system.

PRBs are allocated according to a proportional fair (PF) metric using either capacity adaptation (CAP) or micro DTX based energy saving method. Continuous frequency domain PRB allocation is performed in CAP method. The principle of CAP based resource allocation method is described in Section 5.3 and further details for micro DTX method can be found in Section 5.1. In DTX based scheduling, hardware features with component deactivation is utilized to maximize the energy efficiency. In CAP based resource allocation strategy, PA operation point and other components adjustments has been applied according to the number of used PRBs to take benefit of hardware features.

The considered energy saving solutions are LTE compatible. LTE specific reference signals have been assumed to be transmitted in every sub frame in both the considered resource allocation strategies. The adaptation of the number of transmit antennas and resource allocation methods can be applied to all network traffic loads in each deployment scenario at fast time scale. As usual trend, most of the gain is obtained at the low traffic load case. The advanced MU-MIMO methods based on an instantaneous CSI are mostly useful at dense urban macro networks with high traffic load.

System level simulation results are presented in the following.

It is assumed that the base station is equipped with four transmit antennas and the receiver uses two transmit antennas to detect the received signal. BS may use one, two or all the four antennas for transmission. The air interface assumptions are LTE specific and the system simulation assumptions can be found in [EARTH-D2.2]. 10MHz transmission bandwidth with TDD is considered in the network consisting of 19 three sector sites. Dense urban scenario with inter site distance of 500 meters is considered. The power model described in [EARTH-D2.3] is assumed with EARTH component improvements in all studied MIMO schemes. The total output power of 40 W is assumed in 2x2 and 4x2 MIMO schemes whereas output power is 20 W in the case of single antenna transmission. TDD enables also the assumption of the full CSI at the BS due to the channel reciprocity in UL and DL. The CSI estimate is periodic and delayed due to the assumption of periodic UL sounding reference signal. This study focuses on the traffic model, where input data for each user comes continuously with rate 1 Mbps. This traffic model is valid, e.g. for video streaming or for the case that an operator offers mobile data services with a limited rate. Users are uniformly distributed over the network area. For example, the number of the users is 124 with 30 Mbps/km2 load.

FIGURE 62 illustrates BS’s mean power level per area versus traffic load. Single antenna transmission is able to serve cell edge users at maximum load of 60 Mbps/km2, two transmit antennas are needed from 70 to 90 Mbps/km2 traffic load whereas four antennas are needed when traffic load is higher than 90 Mbps/km2. If CSI is available at the transmitter, MU-MIMO can be applied provided that it increases energy efficiency. One can see from FIGURE 62 that CSI and MU-MIMO does not provide power savings in 2x2 MIMO system compared to the baseline SU-MIMO with LTE compliant CQI and PMI. In the case of high traffic load, performance improvement attained via CSI is two fold since energy savings and better cell edge performance is obtained compared to the SU-MIMO with CQI and PMI. Actually CSI is needed to serve cell edge users close to 1 Mbps when traffic load is higher than 120 Mbps/km2.

EARTH PROJECT

EARTH_WP4_D4.3 74 / 121

FIGURE 62. BS’s mean power level versus system traffic load with CAP.

Daily power consumption for the dense urban network is shown in FIGURE 63. The used traffic profile scenario 2 with 155 Mbps/km2 peak data rate has been defined in [EARTH-D2.3]. Daily energy savings are shown in TABLE 8 with respect to a 2x2 baseline scheme and fixed 4x2 CQI/PMI based scheme. CAP scheduling and hardware features reducing the power consumption of components according to cell load are used also in reference schemes. The number of used transmit antennas has been selected adaptively in order to save energy at low loads and to achieve desired data rates at the high loads. Note that the fixed 2x2 baseline scheme is not able to serve desired data rates to cell edge users and not fully fair reference for 4x2 systems when peak data rate exceeds 90 Mbps/km2. The combination of DTX, the adaptation of the number of used transmit antennas as well as MU-MIMO attains notable energy savings compared to fixed baseline scheme while maintain desired data rates also for cell edge users. MU-MIMO provides energy savings only at the high load and four transmit antenna case. For example, 4x2 CSI based MU-MIMO gives 6.5% energy saving compared to 4x2 SU-MIMO with CQI and PMI in the case of DTX scheduling and 155 Mbps/km2 traffic load. In the case of mixed traffic services, e.g., consisting of burst ftp download and constant rate services (video and VoIP), the optimal use of antenna muting and scheduling strategies may be different than considered. A cell traffic model is like full buffer model for a while when ftp download is applied and therefore the maximum number of transmit antennas as well as full bandwidth could be used to serve user with available throughput. Antenna muting and DTX or CAP scheduling could be applied when there are no ftp packets to deliver. However, big difference for daily energy savings are not expected for mixed traffic case when the DTX is applied.

0 20 40 60 80 100 120 140 160

1

1.5

2

2.5

3

3.5

4

4.5

Network load [Mbps/km2]

Mean p

ow

er

per

are

a [

kW

/km

2]

CQI

CSI

NTx=1

NTx=2

Baseline

NTx=4

EARTH PROJECT

EARTH_WP4_D4.3 75 / 121

FIGURE 63. Daily BS’s mean power consumption with peak data rate 155 Mbps/km2.

TABLE 8. Daily energy savings of adaptive MIMO schemes when peak data rate is 155 Mbps/km2

Reference CQI/PMI, CAP [%] CSI, CAP [%] CQI/PMI,DTX [%] Opt,CSI,DTX [%]

Baseline 2x2, CAP 1.6 3.9 19.3 22.0

Fixed 4x2 CQI,PMI, CAP 20.4 22.3 34.7 36.9

In this section energy efficiency of different MIMO modes in dense urban macro environment when each user was served with 1 Mbps throughput is considered. Based on the numerical examples, it can be concluded that the BS should adapt the number of used antennas in order to obtain energy savings. This means that the antenna muting for which further results and implementation issues are considered in Section 7.2 is a promising method to be implemented. Furthermore, the DTX scheduling provided better power savings than that the capacity adaptation based resource allocation. In the case of high network load, the use of MU-MIMO can also decrease the BS power level when CSI is available. The results indicate that the 1x2 single antenna mode is suitable for low network load scenario. For medium network loads the 2x2 MIMO is the most suitable whereas at the high load case, 4x2 MIMO is needed to guarantee the desired user throughput.

7.2. ANTENNA MUTING IN LTE

As seen in the previous section, fast adaptation of the number of transmit antennas is a good strategy to save energy. In this section it will be shown how this can be implemented in an LTE compliant manner. The idea of antenna muting is to activate an antenna only when motivated by traffic load and/or data rate but not reconfiguring the number of antennas in the cell. From a power-savings perspective it is desirable to activate multiple power amplifiers only when motivated by traffic load and/or data rate. This is illustrated in FIGURE 64.

0 5 10 15 20 25

0.5

1

1.5

2

2.5

3

3.5

4

4.5

Time

Mean p

ow

er

per

are

a [

kW

/km

2]

Fixed 2x2 baseline,CAP

Fixed 4x2 CQI/PMI,CAP

CQI/PMI,CAP

CSI,CAP

CQI/PMI,DTX

Opt,CSI,DTX

EARTH PROJECT

EARTH_WP4_D4.3 76 / 121

High traffic Low traffic

FIGURE 64. Adaptive multi-antenna transmission.

An LTE base station that supports MIMO can have 2 or 4 transmit antennas per sector. When a cell has no traffic, even for as short duration as a few milliseconds, one would like to de-activate radio chains in order to reduce energy consumption. During low traffic there are also a lot of low rate services that can be handled perfectly well with only one radio chain (status updates, checking for emails, etc).

In order to de-activate a radio chain one can either simply de-activate radio chains as shown in FIGURE 65a. This is equivalent to a constant fading on the muted physical antennas. Alternatively we can add (merge) the signals corresponding to the different logical antennas first and then mute all physical antennas except one, as shown in FIGURE 65b. This is equivalent to creating a channel where the correlation between the antennas equals 1. Combinations are also possible by merging signals on antennas 0 and 1 while muting signals on antennas 2 and 3.

FIGURE 65. a) Logical and physical antenna muting, b) Logical antenna port merging and physical antenna port muting.

The base-band processing required for muting and merging antenna ports can be expressed as an antenna pre-coder matrix operation. The normal operation mode is to use an identity matrix, i.e. [1,0,0,0; 0,1,0,0; 0,0,1,0; 0,0,0,1] as pre-coder matrix where each row corresponds to each physical antenna port while each column corresponds to each logical antenna port, or transmission layer. The antenna port muting operation in FIGURE 65a then corresponds to the pre-coder matrix [1,0,0,0; 0,0,0,0; 0,0,0,0; 0,0,0,0] while the antenna port merging operation in FIGURE 65b corresponds to [1,1,1,1; 0,0,0,0; 0,0,0,0; 0,0,0,0]. We also consider a muting operation with pre-coder matrix [sqrt(2),sqrt(2),0,0; 0,0, 0,0; 0,0,0,0; 0,0,0,0]), to maintain the total transmit power on the remained physical antenna ports.

Port 1

Port 2

Port 3

EARTH PROJECT

EARTH_WP4_D4.3 77 / 121

7.2.1. Link Level Evaluation Results

The key LTE control channels evaluated in this section with radio chain deactivation are the Physical Downlink Control Channel (PDCCH) and the Physical Broadcast Channel (PBCH). For the link level evaluations an LTE system with two or four transmit antennas is considered. The system is normally calibrated for transmitting with all available antenna ports, which is also the configuration used as reference when we examine different antenna muting alternatives. Some main parameters used in the simulator are listed in TABLE 9.

TABLE 9. Simulation parameters

Parameter Value

Carrier Frequency [GHz] 2.6

System bandwidth [MHz] 5

Duplex FDD

PDCCH coding (uncoded bits / coded bits) 20 / 72

PDSCH modulation 16-QAM (2×2), QPSK (4×2)

PDSCH coding rate 0.4385

Doppler spread [Hz] 5

Transmission mode LTE Tx-diversity

Channel model Vehicular A

PDCCH performance for 4 Tx and 2 Rx configuration is shown in FIGURE 66.

FIGURE 66. PDCCH BLER versus SINR, 4 Tx and 2 Rx antenna configuration.

-10 -5 0 5 10 15

10-4

10-3

10-2

10-1

100

SINR (dB)

BL

ER

PDCCH, EPA MEDIUM, 4x2

system.port_to_tx_weights = [1 0 0 0;0 1 0 0;0 0 1 0;0 0 0 1]



system.port_to_tx_weights = [1.4 1.4 0 0;0 0 0 0;0 0 0 0;0 0 0 0]




EARTH PROJECT

EARTH_WP4_D4.3 78 / 121

The performance for two transmit and two receive antennas is shown in FIGURE 67 and FIGURE 68. In FIGURE 67 the PBCH performance for the 2×2 antenna configuration is examined. It is observed that the merging and muting configuration (crosses) outperforms all the other tested muting schemes. The performance degradation at BLER 10-2 is in the order of 2 dB.

Results of BLER versus SINR for PDCCH are provided in FIGURE 68. The results for PDCCH are consistent with the 4×2 results. The main difference is that the results are more stable for the muting only configurations ([1,0;0 0] and [sqrt(2),0;0,0]) compared to the corresponding 4 Tx muting only configurations. Also for the 2×2 cases the merging and muting configuration performs well.

FIGURE 67. PBCH BLER versus SINR, 2 Tx and 2 Rx antenna configuration.

FIGURE 68. PDCCH BLER versus SINR, 2 Tx and 2 Rx antenna configuration.

-10 -8 -6 -4 -2 0 2 410

-3

10-2

10-1

100

SINR (dB)

BL

ER

PBCH, EVA MEDIUM, 2x2

system.port_to_tx_weights = [1 0;0 1]


system.port_to_tx_weights = [1.4 0;0 0]


-4 -2 0 2 4 6 8 10 1210

-5

10-4

10-3

10-2

10-1

100

SINR (dB)

BL

ER

PDCCH, EVA MEDIUM, 2x2



system.port_to_tx_weights = [1.4 0;0 0]


EARTH PROJECT

EARTH_WP4_D4.3 79 / 121

It is observed, that by merging the different layers (or logical antenna ports) and then transmitting the composite signal from a single (physical) antenna port is sufficient for situations when the traffic in the cell is close to zero. By doing this significant energy savings (up to 45%) in the radio base station can be obtained since this enables to de-activate radio chains that are not needed.

7.2.2. System Level Evaluation Results

The concept has been evaluated according to scenario 1 in [EARTH-D2.2], for both 2 and 4 antenna ports and energy consumption per area unit calculated as a function of system throughput as shown in FIGURE 69. The legend “all active” is the reference scenario where all antenna ports are always active whereas “adaptive” is the scenario with antenna muting. As can be observed in FIGURE 69, absolute savings are significant and largest savings of course appear at low load. As energy consumption increases with system throughput this will be even more significant for relative savings where up to 45% energy can be saved.

FIGURE 69. Energy consumption per area unit.

FIGURE 70. Resource utilization as function of system throughput.

0 20 40 60 80 100 1202

3

4

5

6

7

8

9

System throughput [Mbps/km2]

EC

I [k

W/k

m2]

4 Tx: all active

4 Tx: adaptive

2 Tx: all active

2 Tx: adaptive

1 Tx

0 20 40 60 80 100 1200

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

System throughput [Mbps/km2]

Re

so

urc

e u

tiliza

tio

n

4 Tx: all active

4 Tx: adaptive

2 Tx: all active

2 Tx: adaptive

1 Tx

EARTH PROJECT

EARTH_WP4_D4.3 80 / 121

FIGURE 70 shows the resource utilization versus system throughput for different configurations. As can be seen, average physical resource block utilization increases with the offered load but we never reach 100%, not even for the one antenna case, but it is less for the 4 Tx adaptive configuration.

FIGURE 71 presents the simulation results for power consumption per area unit and power consumption per bit using 2 Tx antenna muting scheme in the medium traffic density scenario. Daily traffic distribution is utilized in this simulation and it is assumed that both the PA and the RF part of the muted antenna consume 0 W, whereas the BB is not affected by antenna muting. As it is observed in the figure, the estimated power gain is 24%.

FIGURE 71. Power consumption per area unit and power consumption per bit.

7.3. ZERO-FORCING LIKE MULTIUSER MIMO PRECODING SUPPORTING THE CONSTANT MODULUS PROPERTY

In the downlink of multiuser MIMO (MU-MIMO) communications systems, precoding design and scheduling of users are the key to achieve high capacity and low inter-user interference. Moreover, precoding design affects the transmitted power at each antenna through the choice of precoder coefficients. This is because the precoder generally weighs the input signals with different amplitudes and therefore, loads each antenna with a different transmit power. Consequently, the PA at each antenna need to be designed to support higher transmit output powers. In order to circumvent this problem, the novel MU-MIMO precoder is chosen to fulfil the constant modulus property which has been already proposed in [EARTH-D4.2]. According to capacity-level simulations in [EARTH-D4.2], results showed that the proposed method performs similarly to the conventional zero-forcing (ZF) MU-MIMO technique in terms of sum rate. In this section, we clarify how much power can be saved compared to the conventional ZF method by applying the proposed method. All investigations are based on the EARTH power model, [EARH-D2.3].

In the previous deliverable, we estimated the transmit power of the proposed method and the one of the channel vector quantization (CVQ) based ZF method using the coefficients of the precoding matrix P. If the average transmit power of all scheduled symbols is assumed to be the same and unit, the transmit power at each antenna can be written as the norm of the corresponding row vector of the precoding matrix as follows:

B BA0

0.1

0.2

0.3

0.4

0.5

P/A

[kW

/km

2]

Medium traffic density

0.46

0%

0.35

-24%

B BA0

0.1

0.2

0.3

0.4

E/B

[kJ/M

bit

]

0.19

0%0.14

-24%

EARTH PROJECT

EARTH_WP4_D4.3 81 / 121

MiPD

j

jii ,,1,1

2

,TX, P . (7-1)

Here, PTX,i is the transmit power at the i-th transmit antenna and [P]i,j denotes the element of the i-th row and j-th column of the precoding matrix P. M and D are the number of transmit antennas and the scheduled users, respectively.

FIGURE 72 shows the transmit powers at the individual antennas for the CVQ based ZF and the proposed method. Here, the transmit powers have been computed for several channel realizations (trials) based on (7-1) at an SNR of 25 dB and assuming an uncorrelated channel scenario. The transmit powers have been normalized to the values of the proposed method. The colored lines show the transmit powers of the CVQ based ZF method and the black line the ones of the proposed method, both for different transmit antennas. Note that whereas the results in FIGURE 72 depict only 100 trails, the maximum difference in transmit powers between the two methods, which is 0.94 dB in the uncorrelated and 1.22 dB in the highly correlated channel case (not shown in FIGURE 72), has been determined based on a much higher number of randomly generated channel realizations. Since the CVQ based ZF method does not satisfy the constant modulus property, the transmit power at each antenna changes and generates a higher peak power than the one of the proposed method. On the other hand, the proposed method results in constant and equal transmit powers at each antenna, independent from the channel realization. Although the instantaneous transmit power of the CVQ based ZF method changes depending on the specific channel, the average powers of the CVQ based ZF and the proposed method are almost the same. Therefore, the proposed method has a smaller PAPR compared to the one of the CVQ based ZF method.

FIGURE 72. Normalized transmit powers for the CVQ ZF and the proposed method in the uncorrelated channel case.

Based on the above consideration and the power model proposed in [EARTH-D2.3], the power savings obtained by applying the proposed method with respect to the PAPR is estimated. Main focus is given to the reduction of the DC power consumption of PAs. Please note that MU-MIMO will be mainly used for macro/micro cells and high traffic load because multiuser MIMO provides larger sector throughputs in areas experiencing heavy data traffic. Here, the power consumption is computed only for micro base stations (BSs). In addition, the DC power consumption at the maximum radio frequency (RF) output power is evaluated. Mathematically speaking, the DC power consumption (Pdc) of the PA is obtained from the power added efficiency (PAE) as follows:

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

0 20 40 60 80 100

No

rma

lize

d t

ran

sm

it p

ow

er

[dB

]

Number of Trial

TX1 TX2

TX3 TX4

CVQ-based ZF MU-MIMO

0.94dB

Proposed method

EARTH PROJECT

EARTH_WP4_D4.3 82 / 121

PAEinout

dc

PPP , (7-2)

where Pout and Pin are the RF output power and RF input power of the PA, respectively. Here, since the gain of the PA is high, Pout >> Pin is assumed. FIGURE 73 shows total PA DC power consumption versus PAPR in the case of four transmit antennas and a total average transmit power of 38 dBm. The PAE curves are obtained from the EARTH power model. According to the EARTH power model, the PAPR is set to 11.0 dB on the micro BS as a default value. The suppression of PAPR is directly related to the saving in PA DC power consumption.

FIGURE 73. Total PA DC power consumption versus PAPR based on the EARTH PA model.

Previous simulation results showed that the proposed method has the potential to reduce the transmit power of the PA by around 0.94 dB or 1.22 dB in the uncorrelated or highly correlated channel case, respectively, both at an SNR of 25 dB. Here, although it may be optimistic, we assume that the reduction in transmit powers between the two schemes is equivalent to the reduction of the corresponding PAPRs. If we additionally assume that the proposed method achieves a PAPR of 11.0 dB, the PAPR of the ZF method is then 11.94 dB or 12.22 dB in the uncorrelated or correlated channel case, respectively. Every antenna needs to have the capability to amplify a signal with a maximal PAPR of 11.94 dB if the CVQ-based ZF method is used. In other words, the CVQ-based method has to operate the PAs at a lower power efficiency region.

FIGURE 74. Performance comparison between the conventional and the proposed method.

30

40

50

60

10 11 12

Tota

l PA

DC

Po

we

r C

on

sum

pti

on

[W

]

PAPR [dB]

2010 …

EARTH PROJECT

EARTH_WP4_D4.3 83 / 121

FIGURE 74 shows the results in case of four transmit antennas and a total average transmit power of 38 dBm. Again, the results are only shown for the uncorrelated channel case. As one can see, the proposed method saves 6.9 W (15.2%) in total DC power consumption. TABLE 10 shows the results for both, the uncorrelated as well as the highly correlated channel case.

TABLE 10. Comparison of methods in terms of total PA DC Power Consumption.

CVQ based ZF method [W] Proposed method [W] Energy Savings [W]

Uncorrelated 45.4 38.5 6.9 (15.2%)

High correlated 47.7 38.5 9.2 (19.3%)

In conclusion, it is shown that the proposed method saves the PAs DC power consumption because of its lower PAPR value. Moreover, based on the EARTH power model, the benefits when the proposed method is applied are clarified. MU-MIMO will be mainly used for macro/micro sites and high traffic loads because MU-MIMO provides a larger sector throughput in areas experiencing heavy data traffic. Therefore, the proposed method will be mainly beneficial for such scenarios.

7.4. BENEFIT ON ENERGY EFFICIENCY DUE TO SMART MIMO USAGE

Different MIMO configurations have been analyzed under an energy-efficient point of view. MIMO is essential to achieve high data rates in LTE, but not always optimal in terms of energy consumption. Solutions for energy efficient MIMO operation have been evaluated for enabling both, high throughput in high traffic load scenarios and a maximum of energy efficiency in medium and low load situations. The evaluations show that in scenarios with medium and low traffic load, the MIMO configuration has to be modified for improving the energy efficiency by making use from energy efficient resource allocation and antenna muting combined with sleep modes in transceivers.

According to simulations, the flexibility of MIMO configuration, resource allocation, and duty-cycling lead to a large variation in power consumption. In the case of pico-cells, we observed a span up to factor of 15 times variation between the worst-case power consumption and the optimally selected MIMO mode and 10 times variation between the reference mode and the optimally selected MIMO mode. This range shows the potential on adapting the MIMO configuration for energy efficient operation. For macro-cells, the results indicate that the 1x2 single antenna mode is suitable for low network load scenario. For medium network load the 2x2 MIMO is the most suitable whereas at high load, 4x2 MIMO is needed to guarantee the desired user throughput. This adaptation of the MIMO configuration results in a maximum of daily energy saving of 36.9% in dense urban scenario by applying sleep modes or optimized component operation.

The optimization speed of MIMO adaptation is slow since the base station can typically vary the MIMO transmission based on relatively long-term scale (through broadcast channel messages). Only when the total traffic within the cell increases, the base station will switch to a more spectrally efficient mode at the cost of some energy efficiency.

A fast muting of antennas in medium and low traffic situations shows a significant potential to reduce the power consumption in dense urban scenarios. The optimization timescale of antenna muting is in the range of milliseconds, as the power amplifiers can be deactivated very fast. This energy reduction can be achieved without the reconfiguring the number of transmit antennas in the cell. Through 2 Tx antenna muting 24% daily energy savings are expected in small cell radius. It is expected to achieve more power saving when applying antenna muting on 4 Tx antenna ports.

EARTH PROJECT

EARTH_WP4_D4.3 84 / 121

Not only in the low and medium network load can energy gains be achieved with MIMO schemes. In high load network scenarios, the use of energy efficient MU-MIMO precoding scheme can bring energy gains due to its reduced PAPR value. This solution can be mainly used in macro/micro base stations because MU-MIMO provides a larger sector throughput in areas experiencing heavy data traffic. The proposed precoding scheme, evaluated for a macro BS scenario, saves 6.9 W (15.2%) in total DC power consumption for the uncorrelated channel case and 9.2 W (19.3%) for the highly correlated channel case.

Considering different energy efficiency enabling solutions, it has been shown that time-domain duty-cycling has a higher impact on energy savings, compared with frequency-domain duty-cycling, independent of the MIMO configuration. This significant difference is achieved for small-cell scenarios where the relative impact of energy saving features in the BB is the highest, and when applying the short DTX approach, which is asking for modifications in the mobile standard to enable its operation.

EARTH PROJECT

EARTH_WP4_D4.3 85 / 121

8. CONCLUSIONS

This document presents the Green Radio Technologies proposed for energy efficient base stations and networks. It provides energy efficiency enablers on base station and radio link level. The analysed concepts and solutions are presented and their areas of application or application conditions are given using the categories deployment scenarios, optimization time scale and traffic load. The energy efficiency improvements yielded by the concepts and solutions have been evaluated based on simulations and by experimental investigations. The results are presented together with summaries on main findings and recommendations for exploiting the identified potential.

Several hardware solutions have been designed for energy efficiency improvements in macro- and small-cell base stations. The provided power performance characteristics are applied in the project for defining the BS power models used for energy performance evaluations on network level in the context of “Green Networks “ [EARTH-D3.3] and of the “Integrated Solutions” [EARTH-D6.4].

A signal load adaptive transceiver (SLA-TRX) has been defined for macro-cell BS, which exploits the power saving potential in medium and low load situations. It is based on the optimization of operating points (OPA) at reduced signal level and the deactivation of certain components (CD) in time periods of no signal transmission. The SLA-TRX integrates individual component solutions like an adaptive PA, an adaptive power supply, a small signal RF transceiver and a digital transceiver block which operate in a concerted way. The power characteristics provided for 8 different operating points show a power reduction of up to 23% at low signal level, while the deactivation of components provides 55% instantaneous power savings. The determined deactivation and reactivation transition times of 3 µs respectively 10 µs are short enough to allow for applying the CD feature in time slots of 2 or 3 successive LTE symbols allowing to apply the micro-DTX feature discussed below.

All the small-cell BS components, from baseband engine to antenna interface, have been designed and optimized to be energy efficient or supporting power management. An adaptive BB processor provides 20% of efficiency improvement by using ASIP instead of FPGA. Flexible energy aware BB signal processing algorithms are mainly beneficial in UL at low signal loads and offer an overall power reduction of about 13%. An adaptive RF transceiver introduces power saving features in different building blocks leading to a traffic load independent power efficiency improvement of 35% thanks to technology scaling and architecture, while through SiNAD adaptation and time/frequency duty-cycling a traffic load dependent power efficiency improvement of 30% on average can be achieved.

For the small-cell PA, two different solutions are proposed. An adaptive energy efficient PA targeting energy efficiency improvement at low and medium signal loads follows the same principles as presented with the SLA-TRX providing up to 55% of power savings at low signal level and 80% during deactivation. A load modulated PA makes use of a tuneable matching network to optimize PA efficiency at different power levels and is mainly beneficial at high signal loads, providing 30% power consumption reduction.

A low-loss antenna interface is based on a duplexing antenna and allows to relax the transmit filter requirements, lowering the insertion loss. This fully scenario-independent passive solution increases the energy efficiency of the antenna interface from 56% to 80% and is complementary with all the aforementioned energy efficiency enabling techniques.

On the area of antenna components, printed antennas based on foam substrate have been proposed to improve their energy efficiency. The usage of this rugged and cheap foam substrate enables an energy efficiency improvement of about 15% compared with more conventional substrates.

Three energy saving approaches are investigated in the area of radio interface technologies, that exploit the interaction between higher protocol layers and the BS equipment. These techniques allow for maximizing the benefit on energy efficiency improvement provided by energy efficiency enabling hardware features (sleep modes and adaptive component operation). Scheduling solutions operating in time and frequency domain like

EARTH PROJECT

EARTH_WP4_D4.3 86 / 121

discontinues transmission or bandwidth adaptation provide a huge potential on energy saving. Cell DTX solutions provide up to 28% of daily power reduction, which is achievable when applying the short DTX scheme, assuming modifications on the radio standard. The evaluations are based on the SOTA power model where Earth “off” and sleep mode “on” are considered. By using slow bandwidth adaptation combined with micro DTX in dense urban and rural scenarios, energy savings up to 49.8% and 61.8% can be achieved by considering a signal load adaptive transceiver and the improvements of all other components included in the EARTH power model. The combination of BW adaptation with micro DTX is beneficial, as the first technique shows advantages at low resource block utilization (e.g. during the night) while the second one is more favourable at high resource block utilization (e.g. during busy hours in dense urban scenarios). An operation compliant with currently applied radio standards is possible, but shows impact on cell edge performance. For taking benefit from this approach without any drawback, appropriate standard modifications are required.

An energy opportunistic scheduling for taking benefit from good channel conditions allows selecting the least power consuming modulation with regard to the fast changing channel quality. It can be applied in low or medium network load scenarios to reduce the energy consumption at the network side by allowing a transmission delay up to a certain threshold. The average energy per bit can be reduced by 8.5%, 25% or 61% compared to the no-threshold approach at the cost of increasing the average delay by 0.05 ms, 0.15 ms or 6 ms, which is acceptable even for voice communication. This solution is more effective when user applications are less delay or throughput sensitive and all the other energy saving techniques such as MIMO techniques, beam forming, and energy efficiency transceivers and antennas can be perfectly coupled since their area of application is different.

Beamforming techniques using active antenna systems have been analysed. Applying cell specific beamforming based on reconfigurable antennas, the daily power consumption can be reduced by about 8% with respect to the baseline dense urban scenario. This gain should be exploited in medium/long term variations of traffic, for the inherent features of the reconfiguration, slowly following the user distribution in the coverage area. User specific beamforming immediately following the traffic distribution leads to daily power reduction of about 9%. Applied with active antennas, a total power reduction of around 46% has been estimated for an urban scenario including the benefit due to beamforming and due to the avoidance of cable losses for connecting the antennas as a naturally effect when using active antennas.

As MIMO techniques are developed for high throughput scenarios, leading to higher power consumption due to its higher complexity, it has been analysed how to use MIMO systems in the most energy efficient way in medium and low traffic situations. Therefore, the usages of appropriate MIMO mode selection schemes and resource allocation strategies have been considered, as well as downscaling to SISO/SIMO when possible.

The ranges of MIMO configuration, resource allocation, and duty-cycling lead to a large variation in power consumption. For pico-cells, a range of factor 15 times between the worst-case power consumption and the optimally selected MIMO mode and 10 with respect to the reference mode has been estimated, showing the potential on adapting the MIMO configuration for energy efficient operation. In the case of macro-cells, the results indicate that the 1x2 single antenna mode is suitable for low network load scenario. For medium network loads the 2x2 MIMO is the most suitable whereas at the high load case, 4x2 MIMO is needed to guarantee the desired user throughput and high capacity. This adaptation of the MIMO configuration results in a maximum of daily energy saving of 36.9% in dense urban scenario by applying sleep modes or optimized component operation. These results show the maximum amount of power saving when operating MIMO following energy efficiency criterions, respectively the maximum amount of power wastage, if no energy efficiency criterions are considered. The optimization speed of MIMO adaptation is slow since the base station can typically vary the MIMO transmission based on relatively long-term scale (through broadcast channel messages).

Muting and activating of antennas in multi antenna systems following the required throughput without reconfiguring the number of antennas in the cell, enable power savings in low and medium traffic load situations. The optimization timescale of antenna muting is in the range of milliseconds, as the power

EARTH PROJECT

EARTH_WP4_D4.3 87 / 121

amplifiers can be deactivated very fast. Antenna muting leads to 24% of daily energy savings in medium traffic load scenarios allowing the operation of base stations designed for highest throughput in an energy efficient way during non-busy hours.

Energy gains with MIMO schemes can be achieved not only in the low and medium traffic load situations. The use of energy efficient MU-MIMO precoding scheme supporting the constant modulus property leads to equal transmission power for each antenna and thus lower maximum output power for each PA of 0.94 dB for the uncorrelated channel and 1.22 dB for the correlated channel. Thereby the PAs can be designed for lower output power leading to a reduction of the consumed power of 15.2% respectively 19.3% even in high load situations, evaluated for micro-cell base stations.

The presented energy efficiency enablers proposed for radio nodes are key components of the integrated solutions as defined in WP6. Further the analysis and modelling of the power consumption behaviour is the basis on which the performance evaluations on network level are calculated in WP3 and WP6. The computed overall gains of individual enablers and of the integrated solutions are reported in [EARTH-D6.4].

EARTH PROJECT

EARTH_WP4_D4.3 88 / 121

9. REFERENCES

[BoDG11] Bories, S.; Dussopt, L.; Giry, A.; Delaveaud, C.; "Duplexer-less RF Front-end for LTE Pico-Cell using a Dual Polarization Antenna," Wireless Conference 2011 - Sustainable Wireless Technologies (European Wireless), 11th European , vol., no., pp.1-3, 27-29 April

[CaG10] E.Calvanese and P. Greco. "Green Resource Allocation for OFDMA Wireless Cellular Networks",

International Symposium on Personal Indoor and Mobile Radio Communications, 2010.

[CaNT12] H. Cao, H.M. Nemati, A.S. Tehrani, T. Eriksson, C. Fager, "Digital Predistortion for High Efficiency Power Amplifier Architectures Using a Dual-Input Modelling Approach", IEEE Transactions on Microwave Theory and Techniques, vol. 60, no.2, pp.361-369, Feb. 2012

[CST11] www.cst.com

[CrP91] F. Croq and A. Papiernik, “Stacked Slot-Coupled Printed Antenna”, Microwave and Guided Wave Letters, IEEE, vol. 1, no. 10, pp. 288-290, Oct. 1991.

[DeWV09] B. Debaillie, P. Van Wesemael, G. Vandersteen, J. Craninckx, "Calibration of Direct-Conversion Transceivers," IEEE Journal of Selected Topics in Signal Processing, Vol. 3, pp. 488-498, 2009.

[DeDG12] C. Desset, B. Debaillie, V. Giannini, A. Fehske, G. Auer, H. Holtkamp, W. Wajda, D. Sabella, F. Richter, M.J. Gonzalez, I. Godor, P. Skillermark, M. Olsson, M. Imran, A. Ambrosy, O. Blume, “Flexible Power Modeling of LTE base stations," submitted to IEEE Wireless Communications and Networking Conference, 2012.

[duplex] Anatech Electronics Inc. datasheet of the duplexer component Part Number: AM1900D325 available at: http://www.amcrf.com/products/large/datasheet/AM1900D325.pdf

[EARTH-D2.1] INFSO-ICT-247733 EARTH-Report D2.1, “Economic and Ecological Impact of ICT”

[EARTH-D2.2] INFSO-ICT-247733 EARTH-Report D2.2, “Definition and parameterization of reference system and scenarios”

[EARTH-D2.3] INFSO-ICT-247733 EARTH-Report D2.3, “Energy efficiency analysis of the reference systems, areas of improvements and target breakdown”

[EARTH-D2.4] INFSO-ICT-247733 EARTH-Report D2.4, “Most suitable efficiency metrics and utility functions”

[EARTH-D3.3] INFSO-ICT-247733 EARTH-Report D3.3, “Final Report on Green Network Technologies”

[EARTH-D4.1] INFSO-ICT-247733 EARTH-Report D4.1, “Most Promising Tracks of Green Radio Technologies”

[EARTH-D4.2] INFSO-ICT-247733 EARTH-Report D4.2, “Green Radio Technologies”

[EARTH-D5.3] INFSO-ICT-247733 EARTH-Report D5.3, “Report on Validation”[EARTH D6.4] INFSO-ICT-247733 EARTH-Report D6.4, “Final integrated concept”

[EARTH-Leaf] EARTH project Summary leaflet, https://www.ict-earth.eu/

[EbG08] W. Eberle and M. Goffioul, "A scalable low-power digital communication network architecture and an automated design path for controlling the analog/RF part of SDR transceivers," Conference on Design, Automation and Test in Europe, pp. 710-715, Nov. 2008.

[ENS98] ENSEMBLE. “Design Review & 1D Array Synthesis”, version 5.1, User´s Guide, Ansoft Corp., January 1998. [FaBNC06] C. Suárez-Fajardo, M. Ferrando-Bataller, A. Valero-Nogueira, and M. Cabedo, “Dual Polarized Patch-Antenna for Base Station Applications, Antennas and Propagation, IEEE Int. Symposium, Albuquerque, NM, USA, pp. 3655-3658, July 2006.

http://www.cst.com/

EARTH PROJECT

EARTH_WP4_D4.3 89 / 121

[FeBBWP10] D. Ferling, T. Bitzer, T. Bohn, D. Wiegner, and A. Pascht, “Power Efficient Transceivers to Enable Energy-Efficient Mobile Radio Systems” Bell Labs Tech. J., 15:2 (2010) pp. 59.

[GaLLY02] S.-C. Gao, L.-W. Li, M.-S. Leong, and T.-S. Yeo, “Wide-Band Microstrip Antenna with an H-Shaped Coupling Aperture”, Vehicular Technology, IEEE Transactions on, vol. 52, no. 1, pp. 17-27, Jan. 02.

[GaLLY03] S.-C. Gao, L.-W. Li, M.-S. Leong, and T.-S. Yeo, “Dual-Polarized Slot-Coupled Planar Antenna with Wide Bandwidth”, Antennas and Propagation, IEEE Transactions on, vol. 51, no. 3, pp. 441-448, Mar. 2003.

[LTE-SPEC] 3GPP TS 36.104 V10.2.0 (2011-04); LTE Technical Specification (Release 10)

[MeP07] C. Mendes and C. Peixeiro, “Theoretical and Experimental Validation of a Generalized Wheeler Cap Method”, Proc European Conf. on Antennas & Propagation - EUCAP, Edinburgh, UK, November 2007.

[Mont47] C. G. Montgomery, Technique of Microwave Measurements, McGraw-Hill, New York, 1947.

[TaWP96] S. Targonski, R. Waterhouse, and D. Pozar, “Wideband Aperture Coupled Stacked Patch Antenna Using Thick Substrates”, Electronics Letters, vol. 32, no. 21, pp. 1941-1942, Oct. 1996.

EARTH PROJECT

EARTH_WP4_D4.3 90 / 121

10. APPENDIX

10.1. SIGNAL LOAD ADAPTIVE TRANSCEIVER

10.1.1. Impact of gain variation during component reconfiguration

When applying OPA to the AEEPA during signal transmission, the gain of the PA is varying within single symbols leading to distortions in the spectrum of these symbols and to increased noise outside the signal band with the risk of violating the spectrum emission mask. This effect is shown in FIGURE 75 based on simulations with 40 ms LTE signals of 30% traffic load performed for baseband signals. The increase of spectral noise is significant and is violating the spectrum emission mask close to the signal band. For complying with the LTE specifications, a powerful compensation of the gain variation is to be implemented in the DSPC. The simulation results show the significant reduction of the noise in the adjacent channels for 90% reduction of gain variation. A maximum of 10% residual gain variation is considered as the performance requested for gain variation suppression.

FIGURE 75. Impact on LTE signal spectrum (40 ms; 30% load) after passing the AEEPA.

When modifying the PA operating point individually for successive LTE symbols, the gain variation in the transceiver implies a path loss variation for successive symbols. This would lead to distortions in the path estimation in the user equipment and would translate into increased Error Vector Magnitude (EVM) values which can exceed about 10% for the signal received by the mobile equipment. The compensation of these gain variation to be implemented in the DSPC as proposed above, will minimize the impact on the EVM in the UE-receiver to acceptable values. For 90% reduction of the gain variation the increase of the EVM would be reduced below 1 or 2%.

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5

x 107

-90

-80

-70

-60

-50

-40

-30

-20

-10

0

[dB

]

frequency [Hz]

Power Spectral Density

Clipped Input Signal

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5

x 107

-90

-80

-70

-60

-50

-40

-30

-20

-10

0

[dB

]

frequency [Hz]

Power Spectral Density

Output SignalSuppressionof gain

variation

0%

90%

100%

EARTH PROJECT

EARTH_WP4_D4.3 91 / 121

10.1.2. Circuitries for the small signal RF transceiver

The small signal RF transceiver is composed of two main parts: The dual mode small signal RF transmitter and the receiver.

10.1.2.1. Receiver

The receiver chain amplifies and down converts the aerial signals to the baseband level in order to be processed by the DSPC properly. The most critical design parameters for a receiver, in order to guarantee the quality of service, are the followings:

Input referred noise figure (NF) the complete receive chain.

Linearity of the complete chain: Input 2nd and 3rd order intercept points (IIP2, IIP3) as well as the input 1dB compression point (ICP1).

Receiver gain from antenna to the DSPC.

The spectral purity of the local oscillator used for the aerial signal down conversion. A simplified block diagram of the RF receiver designed for the SLA-TRX is depicted below (FIGURE 76). The receive chain has two channel receiver, and each receiver is constituted by a low noise amplifier (LNA) followed by a medium power amplifier (MPA). The amplified signal is filtered out by a band select (system select) filter and then down converted by a high isolation dual mixer having certain level of isolation between the two channels. The required local oscillator (LO) signal for the down conversion mechanism is generated by an onboard LO generator. Once the RF signal is down converted into an intermediate frequency (IF), it is further filtered out by a relatively narrower filter (IF filter) having a pass-band of the baseband signal. The same signal is amplified again by a gain controlled IF amplifier (VGA) followed by the final filtering before transforming the analogue IF signal to the digital domain by a dual ADC.

FIGURE 76. Block diagram of the heterodyne receiver for the SLA-TRX in macro BS.

Having an RF interface via SMA connectors at the LNA inputs, the receiver referred in FIGURE 77 can easily be cascaded with additional front-end blocks like additional gain blocks or filters in front in order to get further amplification and/or filtering.

DSPC

Clock

Regeneration

&PLL

ADC

ADCPost-

amp

Post-

amp

Dual IF VGA:

Vcc_5V

Vcc_BB_dig.

(1V8)Vcc_BB_an.

(3V3)Vcc_5V

Dual Mixer:

MN

MN

MN

MN

Vcc_5V

MNMN

LNA2

MNMN

LNA1

Small Signal Transceiver: Receive Chain

Loop

filter

VCO&PLL

Rref

Vcc_3V3

Clk_Ref_RF

Vcc_3V / 3V3

Vcc_5V

MNMN

LNA2

MNMN

LNA1 RX_1

Vcc_3V/3V3

Clk Ref.

Vcc_3V3

Clk_Ref_RF

Clk_Ref_Dig

Vcc_3V3

Shared with TX

and other blocks

Vcc_RF

(5V)

Vcc_RF

(3V3)

Vcc_BB_an.

(3V3)

Vcc_BB_dig.

(1V8)

Dc-Dc /

LDO

Power

Management

48V

RX_2

RX_N

EARTH PROJECT

EARTH_WP4_D4.3 92 / 121

FIGURE 77. Prototype of the dual chain receiver as a part of the SLA-TRX for macro BS.

The breakdown of the different parameters of the receiver referred above is given in FIGURE 78. The lab evaluations show similar results compared to the simplified link budget referred in this figure.

EARTH PROJECT

EARTH_WP4_D4.3 93 / 121

FIGURE 78. Performance and DC power consumption partitioning of the receiver.

EARTH PROJECT

EARTH_WP4_D4.3 94 / 121

10.1.2.2. Dual Mode Transmitter

The main functionality of the small signal transmitter (TX) is to convert the digital baseband signal provided by the DSPC into the modulated RF signal, then to amplify and filter-out the modulated analogue signal in order to drive the power amplifier with an appropriate signal in terms of power level and spectral purity.

Depending on the complete transmit architecture mainly related to the linearity versus energy efficiency constraints the small signal implementation varies in order to be compatible with the architecture chosen. For instance, the architecture of the small signal transmitter designed for the SLA-TRX is a part of the digital pre-distortion (DPD) architecture as referred in the DSPC section. The forward path is designed for direct up conversion (i.e., no intermediate frequency used neither in digital domain nor in analogue domain) which consumes relatively less power compared to the high IF counterparts (see FIGURE 79). The feedback path is designed for down-converting and digitising the output signal monitored at the output of the AEEPA module. This signal is then used by the signal conditioning algorithm related the DPD architecture.

FIGURE 79. Block diagram of the small signal SLA_TX for macro BS.

Moreover, the small signal transmitter can be configured by the DSPC according to the component control algorithms which enable additional energy saving by deactivating some of the transmitter blocks when there is no symbol to be transmitted.

Small Signal Transceiver:

Transmit Chain

AUX_DAC

I_DAC

Q_DAC

ADC

Vcc_BB_an.

AUX_DAC

+

+

+

+

R,C network:

-DC_cm,

-Diff&quad

compens.

Vcc_RF

TX_VGA

Loop

filterVCO&PLL

Vcc_VCO_RF

Clk_Ref_RF

Feedback

Mixer:

Vcc_BB_dig.

Energy Efficient

Adaptive PA

Loop

filter

VCO&PLL

Rref

Vcc_VCO_RF

Clk_Ref_RF

Clk. Ref.

Clock Distribution

Clk_BB1, 2

Clk_RF1/N 1/N

Vcc_Clk

Vcc_Clk

IQ_modulator

0°90°

Vcc_RF

Clk_BB1

DS

PC

Vcc_RF

Vcc_Clk

Vcc_VCO_RF

Vcc_BB_an.

Vcc_BB_dig.

Dc-Dc /

LDO

Vd

d_

PA

1

Power

Management

Vg

s_P

A (

+/-

)

48V

Clk_BB2

SPI & GPIO (Power on/off)

Vd

d_P

A2

Vd

d_P

A3

Vcc_RF

Power_on/off Power_on/off

Power_on/off

Power_on/off

Vcc_5V_RF

Vcc_3V3_RF

EARTH PROJECT

EARTH_WP4_D4.3 95 / 121

FIGURE 80. Prototype of the small signal SLA_TX for macro BS.

EARTH PROJECT

EARTH_WP4_D4.3 96 / 121

10.1.3. AEEPA module

The hardware prototype of the AEEPA proposed for macro-cell BS is composed of different blocks as shown in FIGURE 81. The components blocks are mounting on top of a heatsink, which is cooled by three fans to maintain the performance in temperature. There is a housing that integrates the Pre-DRIVER and the DRIVER. It contains a RF board and a biasing and control board. The HPA is realized in a Doherty configuration and is mounted into an individual housing integrating a RF board and two biasing and control boards, one for the carrier amplifier and one for the peak amplifier. The third housing is used for interfacing the AEEPA with the DSPC and the APSU.

FIGURE 81. Photograph of the Adaptive Energy Efficient PA prototype for macro BS.

The hardware prototype has RF and bias connections with the different modules in the SLA-TRX: DSPC, SS RF-TRX and APSU.

A DB15 connector interconnects the DSPC and the AEEPA. The DSPC provides control information aligned to the signal level for adapting the AEEPA operating point according to the signal load. Furthermore, there are two dedicated lines to implement the CD feature and reduce the energy consumption during empty LTE symbols. Some components in the AEEPA can be deactivated when no data or signalling is transmitted.

The OPA feature in the AEEPA is performed by adjusting the supply voltages in the Doherty amplifier and the driver to the maximum required RF output power level, leading to the optimal power efficiency value for the related signal load. This adjustment is controlled by the DSPC, initiating changes in supply voltages of AEEPA and APSU via SPI and GPIO interfaces. This supply voltage adaptation limits the maximum output power level and leads to increased power efficiency for the related signal load.

The AEEPA is connected to the SS RF-TRX using two SMA connectors. The RF output of the SS RF-TRX is connected to the RF input of the AEEPA, while a feedback signal is provided from the AEEPA output via the SS RF-TRX to the DSPC as monitor signal for applying the digital predistortion. This technique provides the minimum requirements defined in the LTE standard specifications [LTE-SPEC].

EARTH PROJECT

EARTH_WP4_D4.3 97 / 121

Finally, there are two connectors between the AEEPA and the APSU to provide supply voltages. In the AEEPA, some control signals have been implemented to deactivate the drain voltages for protecting the high power amplifiers if some deviations are detected in supply voltages and temperature range. A sensor is integrated in the driver and the Doherty amplifier to control the temperature of the semiconductor components.

10.1.4. Power characteristic of the APSU

The power characteristic of the APSU is presented in FIGURE 82 showing the DC input power versus the DC power supplied to the AEEPA for eight different operating points and for the situation when the PA stages are deactivated (CD point). These curves serve to define the base station power model.

FIGURE 82. APSU power characteristic.

0

10

20

30

40

50

60

70

0 20 40 60

Inp

ut

Po

we

r (W

)

Output Power (W)

APSU Power performance

OP1 OP2 OP3 OP4 OP5 OP6 OP7 OP8 CD

EARTH PROJECT

EARTH_WP4_D4.3 98 / 121

10.2. LOW-LOSS ANTENNA INTERFACE

The concept of the low-loss antenna interface is investigated through electromagnetic (ANSYS HFSS®) and circuit (ANSYS Designer®) simulations for the antenna and circuit components respectively. Two different antennas have been designed: the Full Band1 antenna and the Dedicated UP/DL antenna. Both have two accesses, one for the Tx chain and the second for the Rx chain. Both accesses of the Full Band1 antenna cover the whole LTE band 1 frequency range (1920 – 2170 MHz); whereas the Tx access of the Dedicated UP/DL antenna covers only the DL frequency band (2110 – 2170 MHz) and the Rx access covers only the UL frequency band (1920 – 1980 MHz). The FIGURE 83 illustrates the return loss and isolation between the Tx and Rx accesses for both simulated antennas. Compared to the Full Band1 antenna, the Dedicated UP/DL antenna presents a better impedance matching (|S11|<-15 dB) and a simulated isolation better than 45 dB on the more critical Rx frequency band. Notice that in the final MIMO version (FIGURE 85), the simulations of real-life details (coupling by the cavity, feeding lines and components proximity, second antenna at one wavelength for spatial diversity and non square substrate and box) shows up to 10 dB penalty on isolation level. However the relaxed Tx filter and Rx filter provide an additional isolation (about 20 dB more) between Tx and Rx ports.

FIGURE 83. Return Loss and Tx/Rx isolation for the Full Band1 (a) and Dedicated UP/DL antennas (b).

The FIGURE 84 illustrates the dual-port slot-coupled patch antenna which is the main structure of both presented antennas. A back cavity has been added both to ensure better front to back radiation ratio and also to improve the Tx/Rx isolation. A thin (0.8 mm) dielectric substrate is used for the patch; which minimizes dielectric losses and so ensures radiation efficiency better than 90% for the patch antenna. Its radiation pattern presents a 90° lobe aperture with a maximum gain of 8 dBi over the whole frequency band. A plastic radome can be positioned 2 cm above the patch. Its impact on Tx/Rx isolation level is kept non-significant.

FIGURE 84. Antenna structure: layers description with material (a), 3D view of the simulated structure (b).

1.85 1.95 2.05 2.15 2.25Freq [GHz]

-50

-40

-30

-20

-10

0

dB

Dedicated UP/DL antenna ANSOFT

Curve Info

Return Loss Rx

Tx/Rx Isolation

Return Loss Tx

(b)(a)

1.85 1.95 2.05 2.15 2.25Freq [GHz]

-50

-40

-30

-20

-10

0

dB

Full Band 1 antenna ANSOFT

Curve Info

Return Loss Rx

Return Loss Tx

Tx/Rx Isolation

1920 – 2170 MHz1920 – 1980 MHz

2110 – 2170 MHz

1.85 1.95 2.05 2.15 2.25Freq [GHz]

-50

-40

-30

-20

-10

0

dB

Dedicated UP/DL antenna ANSOFT

Curve Info

Return Loss Rx

Tx/Rx Isolation

Return Loss Tx

Aluminium (cavity)

Copper (slots in ground plane)

RO4003 substrateCopper (feeding lines)

Copper (radiating patch)

Plastic ABS (radome)

Rx (UL) port Tx (DL) port

10 cm

10 cm

3 cm

RO4003 substrate (optional)

Nylon screw

(b)(a)

EARTH PROJECT

EARTH_WP4_D4.3 99 / 121

FIGURE 85. Two duplexing antennas structures with spatial diversity (MIMO configuration).

EARTH PROJECT

EARTH_WP4_D4.3 100 / 121

10.3. LOW LOSS FOAM PRINTED ANTENNAS

This appendix is focused in describing the proposed solution consisting in the use of a cheap foam substrate to improve the antenna EE. For the sake of completeness some results already contained in [EARTH-D4.2] are also included.

10.3.1. Foam Printed Antennas

Initially [EARTH-D4.1] the efficiency of a single microstrip antenna element and of an array of 4 elements printed on a foam substrate was compared with solutions that use other common cheap substrates. The RF Rogers RO 3003 and the low frequency FR4 have been used. The obtained results are included in TABLE 11.

TABLE 11. Cheap substrate characteristics and antenna efficiency (at 2 GHz).

Substrate εr tan δ Antenna Efficiency [%]

Single Element 4x1 Array Foam 1.06 0.001 98.0 96.4

RO 3003 3.0 0.005 88.9 81.1

FR4 4.0 0.03 55.8 38.2

For the array the foam substrate provides a radiation efficiency improvement of about 15% and 58% over the RO 3003 and the FR4 substrates, respectively.

10.3.2. Foam Characterization

As described in [EARTH-D4.1], the relative dielectric constant ( r) and loss tangent (tanδ) of Styrofoam™ has been measured, by the free space method [Mont47], in the frequency range 1.5 – 12 GHz. The experimental

results obtained are shown in FIGURE 86 and FIGURE 87. The relative dielectric constant is 1.06 0.02. The loss

tangent results are not accurate but can be estimated as 0.001 0.0005.

FIGURE 86. Experimental relative dielectric constant (εr) of Styrofoam™.

1

1,02

1,04

1,06

1,08

1,1

1,12

1 2 3 4 5 6 7 8 9 10 11 12

Re

lati

ve D

iele

ctri

c C

on

stan

t

Frequency [GHz]

EARTH PROJECT

EARTH_WP4_D4.3 101 / 121

FIGURE 87. Experimental loss tangent (tan δ) of Styrofoam™.

10.3.3. Software Validation

Software tools are essential in the design and optimization procedures of modern antenna systems. However, even nowadays powerful software tools need experimental validation. As described in [EARTH-D4.1], the microstrip patch antenna configurations shown in FIGURE 88 have been used to validate several software packages. A good agreement between ENSEMBLE [ENS98] and CST Microwave Studio [CST11] input reflection coefficient and radiation pattern numerical simulations and experimental results has been obtained. This experimental validation has strengthened the confidence in the used software tools.

FIGURE 88. Antenna prototypes used to validate software tools.

10.3.4. Antenna Application Examples

To confirm experimentally the main conclusions concerning microstrip antennas printed on foam substrates two antenna prototypes have been designed fabricated and tested: a single element configuration and an array of two elements. With this approach it will be possible to obtain the element efficiency and the losses associated with the microstrip line feeding system. The efficiency of uniform linear arrays with more elements can then be extrapolated.

0

0,005

0,01

0,015

0,02

1 2 3 4 5 6 7 8 9 10 11 12

Lo

ss T

an

gen

t

Frequency [GHz]

EARTH PROJECT

EARTH_WP4_D4.3 102 / 121

LTE band 1 (frequency range 1920-1980 MHz for the UL and 2110-2170 for the DL) has been chosen. A dual-band multilayer structure of stacked patches fed by a non-resonant slot is used [CrP91, TaWP96, GaLLY02]. The multilayer profile is shown in FIGURE 31 and the array configuration is shown in FIGURE 32. Standard thicknesses are used: 1.57 mm (0.062”) for the patch substrates, 0.79 mm (0.031”) for the feed line substrate and 10 mm for the foam. The thickness of the foam layer has been selected according to the bandwidth specifications. Wide-band applications require a thick foam substrate layer. The antenna dimensions (patches, slot and feed line) have been optimized to fulfil the bandwidth specifications (S11<-10 dB).

10.3.4.1. Dual-Band Element

The two stacked patches create the two required resonances (UL and DL sub-bands). A non-resonant H shaped slot is used to enhance the coupling between the bottom patch and the feeding line while keeping a low back radiation level [GaLLY02]. The element dimensions (top square patch length, bottom square patch length, H-shaped slot dimensions and position, and feeding line stub length) have been optimized to provide S11 < -10 dB in the LTE band 1 UL and DL sub-bands (1920-1980 MHz and 2110-2170 MHz).

The antenna element has been fabricated using a conventional photolithography printing circuit technique. All the substrate layers are square with 120 mm length. The two stacked patches are centered on the substrate layers. Photographs of the element layers before and after assembling are shown in FIGURE 89. The most critical parameter of the antenna, the input reflection coefficient, has been measured with a vector network analyzer. The results obtained are shown in FIGURE 90.

a) b)

FIGURE 89. Single element prototype before assembling (a) and after assembling (b).

EARTH PROJECT

EARTH_WP4_D4.3 103 / 121

FIGURE 90. Experimental input reflection coefficient of the single element antenna.

10.3.4.2. Dual-Band Array

The geometry of the two elements array is shown in FIGURE 32. All the four substrate layers are 120 mm wide and 180 mm long. Photographs of the array layers before and after assembling are shown in FIGURE 33. The following experimental results have been obtained;

Input reflection coefficient (with vector network analyzer);

Far field radiation pattern (in far field anechoic chamber);

Radiation efficiency (with Wheeler cap).

The measurement setups are shown in FIGURE 91. The experimental results obtained are shown in FIGURE 34, FIGURE 92, FIGURE 93 and FIGURE 36.

FIGURE 91. Measurement setups of S11, radiation pattern and efficiency.

-30

-25

-20

-15

-10

-5

0

1,5 1,6 1,7 1,8 1,9 2,0 2,1 2,2 2,3 2,4 2,5

S11

[d

B]

Frequency [GHz]

EARTH PROJECT

EARTH_WP4_D4.3 104 / 121

FIGURE 92. Experimental far field radiation pattern of the antenna array (1.95 GHz).

FIGURE 93. Experimental far field radiation pattern of the antenna array (2.14 GHz).

There is a general good agreement between numerical simulations (ENSEMBLE and CST) and the experimental results. However there is a difference of about 5 dB in the front-to-back ratio. This difference is originated in the radiation of the coaxial cable used to feed the array in the experimental setup. Although the cable has a weak radiation it is very significant compared with the antenna radiation in the back side. There is also a small discrepancy (about 4%) between the foreseen radiation efficiency (97.3%) and the results obtained experimentally (93%). This is due to fabrication and assembling imperfections not accounted for in the theoretical models.

10.3.4.3. Final Remarks

The choice of the two element array as application example was determined mainly by the size of the Wheeler cap (140 mm diameter by 200 m high). The efficiency testing has also influenced the shape of the feeding line system. The losses of the feeding line actually used must be very similar to the losses of a 4 element array.

Based on the geometry proposed other configuration can be used, such as dual-polarization [GaLLY03, FaBNC06] or even circular polarization.

-30

-25

-20

-15

-10

-5

0

-180 -150 -120 -90 -60 -30 0 30 60 90 120 150 180

Rad

iati

on

Pat

tern

[d

B]

Angle [Degree]

E-Plane Co

E-Plane Cross

H-Plane Co

H-Plane Cross

-30

-25

-20

-15

-10

-5

0

-180 -150 -120 -90 -60 -30 0 30 60 90 120 150 180

Rad

iati

on

Pat

tern

[d

B]

Angle [Degree]

E-Plane Co

E-Plane Cross

H-Plane Co

H-Plane Cross

EARTH PROJECT

EARTH_WP4_D4.3 105 / 121

10.4. ACTIVE ANTENNAS

The adoption of active antennas could enable significant power savings in the BS and also in the UEs. In fact: the RF transceivers are integrated into the antenna so that feeder losses are removed: DL power needs are reduced and a better sensitivity of the BS reception chain can be achieved; the beamforming algorithms should allow to concentrate the DL radiated power on a per-user basis and to select the received signal from each user in UL, so both UL and DL interference should be reduced and the overall UL (at UE level) and DL (at BS level) power requirements dropped.

FIGURE 94. Schematic concept of Active Antenna system.

A particular active antenna array is studied in EARTH (see FIGURE 94) describing the architecture and the overall layout. As a general definition, an Active Antenna System (AAS) includes a certain number of active antenna elements arranged in an array and allowing the radiation pattern shaping to be remotely controlled. Each active antenna element contains a number of traditional antenna elements (dipoles, patches or other geometries) and a complete radio transceiver digitally connected to a Beamforming Unit able to perform digital beamforming processing. Then active antenna is digitally connected to a Central Unit (where the base-band processing is performed) via optical fiber. The active antenna could be considered as an example of RRH including the antenna, and the antenna is an antenna array in order to exploit its beamforming capabilities.

In such configuration based on a “per-cell” beamforming approach, the optical connection between the central unit and the remote unit with active antennas could be managed by standard CPRI/OBSAI interface or similar transmission mode. In different applications, the AAS enables advanced beamforming techniques based on a “per-user” approach: Grid of Fixed Beams is widely accepted as one of the simplest (but anyway able to improve the performances of the cell) forms of beamforming; fully adaptive beamforming approaches with per-user tracking can also be implemented. For an adaptive antenna to become a truly EE enabler by taking advantage of its per-user beamforming functionalities, it’s important to design its architecture such that it’s really feasible from both a technical and a cost-efficiency point of view.

In EARTH deliverable D4.1 a full description of the “per user” based Active Antennas was reported, assessing the better performances it could ensure with reference to the traditional “per cell” architectures. The suggested new architecture is the one reported in FIGURE 95 for the transmitter part and FIGURE 96 for the reception part. Such new architecture should enable to perform the signal processing on a per-user basis either within the antenna or in a cooperative form involving multiple antennas thanks to a particular partitioning of the baseband (L1) modem functionalities between the CU and the AAs. In particular, coding, H-ARQ, interleaving, modulation, MIMO processing and resource mapping are still performed in the central unit, while the adaptive beamforming is performed in an enhanced BFU within the antenna. So, the signal transmitted over the optical fiber is the signal at the output/input of the resource mapping where users are

BasebandProcessing

O/EOptical link

CPRI/OBSAI

IP network

RF RF

RFRF

Active Antenna

• RF: TX/RX front-end

• BFU: Beam Forming

Unit

• E/O: Electrical/Optical

transceiver

E/O

O&M

BFU

BasebandProcessing

O/EOptical link

CPRI/OBSAI

IP network

RFRF RFRF

RFRFRFRF

Active Antenna



Unit


transceiver

E/O

O&M

BFUCU

BasebandProcessing

O/EOptical link

CPRI/OBSAI

IP network

RF RF

RFRF

Active Antenna



Unit


transceiver

E/O

O&M

BFU

BasebandProcessing

O/EOptical link

CPRI/OBSAI

IP network

RFRF RFRF

RFRFRFRF

Active Antenna



Unit


transceiver

E/O

O&M

BFUCU

EARTH PROJECT

EARTH_WP4_D4.3 106 / 121

separated in the frequency domain and the overall throughput with respect to the traditional AAS architecture could remain the same or even be reduced.

FIGURE 95. AAS architecture for enabling a per-user basis beamforming: transmitter part.

FIGURE 96. AAS architecture for enabling a per-user basis beamforming: receiver part.

The proposed new architecture would allow user-based signal processing operations such as adaptive beamforming to be performed in the BFU for each user separately while the calculation of the adaptive weighting coefficients could be performed either in the BFU or in the CU if network coordination algorithms were required.

Resource

mappingE/O

L2/L3

RRC/RLC/MAC

protocols

S1 interface

Central Unit

freq

Resource grid

u1

u3u2

time

Remote Unit

O/E

Use

r W

eig

hti

ng

D/A RFFront

end

D/A RFFront

endIFFT

CP

insertIFFT

CP

insert

1

Km

Coding

H-ARQ

Interleaving

Modulation

MIMO proc.

Resource

demapping

L2/L3

RRC/RLC/MAC

protocols

S1 interface

Central Unit

freq

Resource grid

u1

u3u2

time

Remote UnitO/E

Use

r W

eig

hti

ng

A/D RFFront

end

A/D RFFront

endFFT

CP

rem.FFT

1

Km

CP

rem.

E/O

MIMO proc.

Demodulation

Deinterleaving

H-ARQ

Decoding

EARTH PROJECT

EARTH_WP4_D4.3 107 / 121

10.5. BEAMFORMING METHODS

10.5.1. Reconfigurable “per cell”

10.5.1.1. Introduction.

The objective of this work is to analyze to what extent adaption of spatial filtering in the antennas to the traffic situation can be used to reduce power consumption. Of specific interest are situations where the offered traffic is spatially non-uniform, either with locations being static or varying over time.

The goal is to reduce energy consumption in relation to a reference system so for that reason energy consumptions are calculated for two different systems. One system is a reference system (Ref) according to D2.2 whereas the other system is equipped with reconfigurable antennas (RAS).

For evaluation purposes a static system simulator is used with four, three-sector sites, each sector being equipped with either reconfigurable or conventional antennas. Wrap around is applied to mimic a larger system, especially from an interference point of view.

10.5.1.2. Simulator details

Selecting antenna parameters:

Antenna parameters resulting in as low power consumption as possible is searched for in a process involving several steps.

1. For a given setup of 36 antenna parameters, 3 per cell and 12 cells, activity factors is primarily estimated on bin level by means of fixed point iteration and aggregated to cell level. The fixed point iteration results in a vector of activity factors, one per cell, such that the offered traffic per bin (surface element) in the system is actually served. In case the system is not capable of serving the offered traffic, requiring a nonphysical activity factor > 1, the activity factor is limited to 1, the maximum, and the cell is marked as not serving the offered traffic.

2. Power consumption is estimated based on the activity factors.

3. Antenna parameters are updated and steps 1 and 2 are repeated till no better antenna parameter setups, i.e. resulting in lower power consumption, are found

The activity factor (af), mentioned above, is a number within the range [0 1] and defines the fraction of time a cell is transmitting user plus control data during which 14% of the resource elements are used for control data. For the rest of the time, 1-af, only control channel data is transmitted.

It shall be stressed that transmission of user and control data are assumed to be orthogonal to each other meaning that control channel transmission does not interfere with user data and vice versa. This means that only user data (offered traffic) need to be involved in the process of finding activity factors but control channel transmission is of course taken into account when it comes to estimating power consumption.

Power model:

The evaluation involves four different, linearized power models according to FIGURE 97.

EARTH PROJECT

EARTH_WP4_D4.3 108 / 121

FIGURE 97. Linearized Earth power model, year 2012. The circles at resource utilization = 0 indicate power consumption when sleep mode is ON.

Path loss model, SNR settings:

As parameters according to D2.2, for site-to-site distance of 500 m, results in a severely interference limited system (i.e. waste of power) a modified path loss model is used herein. First the optimal tilt is found, in terms of system throughput, for the reference system assuming full buffers (all BS’s transmitting at full output power) and interference limitation (i.e., high power or similarly low path loss). Then the path loss is adjusted, by adjusting the constant, such that the SNR at 10%-level equals 15 dB. Lognormal fading is not taken into account in this evaluation.

Antenna model:

The antenna model used is basically according to D2.2 but beam tilt as well as the beam pointing direction and beamwidth the azimuth can be changed according to TABLE 12. The shape of the beam, Gaussian in power, is maintained whereas the gain is scaled inversely proportional to the beamwidth to reflect the physical properties of the antenna.

TABLE 12. Antenna parameter ranges.

Description Range

RAS system Reference system

Antenna tilt [5 – 20] deg 15 deg

Azimuth beampointing [-45 – 45 ] deg 0 deg

Azimuth beamwidth [30 – 110] deg 70 deg

Clustered offered traffic:

Traffic is being offered to the system as a certain average data traffic given in Mbps/km2. For a given a hotspot scenario the average offered traffic is modified to an offered traffic per area unit within hotspots as well as outside hotspots while the spatial average over the entire system is according to the primarily offered average traffic.

The non homogeneous spatial traffic distribution used in this study is influenced by the model specified for HetNets in D2.2 but doesn’t follow the exact parameters. A default value of 10 hotspots is used and the hotspots are dropped randomly over the system of 12 cells. To prevent overlap of hotspot areas it is assured that there is a minimal distance of 90 m between hotspot centres, corresponding to two times the hotspot

EARTH PROJECT

EARTH_WP4_D4.3 109 / 121

radius + 10 m. The degree of clustering is given by the hotspot probabilities, or rather the fraction of the offered traffic appearing in a hotspot, and ranges from uniform spatial traffic distribution, for 10 hotspots, to 56% of the traffic appearing in hotspots corresponding to 5.8% of the area. To get statistics a total of 30 drops of the actual number of hotspots is generated. The same drops, or a subset of in some cases, are used for all evaluations.

TABLE 13. Modeling of, clustered, offered traffic.

Description Range

RAS system

Hotspot radii 40 m

Number of hotspots [1, 2, 4, 5, 8, 10, 12, 15, 20, 30, 40, 67]

Hotspot traffic probability [0.058, 0.11, 0.24, 0.39, 0.56]

Average offered traffic [55, 111, 139, 166, 194, 222, 249, 277]

Mbps/km2

10.5.2. Single DoA, Eigen BF

In downlink, beamforming is implemented at the base station, and the beamforming weights should be predetermined. Generally, downlink weights design is more complicated with respect to the uplink one. Only if changes in the spatial channel are small over the time from start of the uplink frame to the end of the downlink frame, and over the uplink-downlink frequency difference, the downlink and uplink channel will be approximately the same. In practice, uplink and downlink channel differences are often so large in FDD that the transmissions in these two links are independent. As a consequence, neither the exploitation of the uplink CSI nor the reuse of the uplink weights can be applied directly in the downlink, but other kind of processing has to be introduced.

A possible solution is a closed loop method, accomplished by introducing in the uplink transmission some feedback information related to the downlink channel, known at the user terminal. Another approach to estimate the beamforming weights for downlink beamforming transmission is to use an open loop method. In this case, the feedback of CSI information from the terminal is not required for the evaluation of the transmitter weights and, hence, it is not transmitted to the base station. The beamforming weights are estimated at the base station by only taking into account the uplink signals received by the array of antenna elements. In FDD systems, where different carrier frequency for uplink and downlink can lead to weights estimation error, this action can be done by exploiting techniques based on slow variation in the channel such as the Direction of Arrival (DoA) and the average path loss of useful signals, since they can be evaluated in uplink and maintained for downlink as well.

In Release 8 of LTE system, the transmission beamforming at the eNB for FDD is foreseen with open loop approach, applied by considering uplink long term information, since no specific feedback for beamforming has been introduced in the specifications.

10.5.2.1. DoA based beamforming

A conventional beamformer is sometimes known as the delay-and-sum one, with all its weights of equal magnitudes. For the computation of the proper weights to be applied in a downlink beamformer, a simple steering of the antenna beam in the desired direction can be implemented. This approach can be worthwhile

EARTH PROJECT

EARTH_WP4_D4.3 110 / 121

in the transmitting beamforming at eNB, since it does not require downlink channel information, but it just exploits the uplink channel information.

The incoming uplink signals to a receiver may consist of desired energy and interference energy, for example from other users or from multipath reflections. The various signals can be characterized in terms of the angle relative to the DoA of each received signal. The DoA beamforming method estimate the dominant DoA on the uplink and then uses that angle to determine the beamforming on the downlink, by forming the array manifold vector on the downlink that is associated with the DoA. The exploitation of a DoA method in the LTE FDD system is based on the assumption that, in theory, the dominant DoA, as a long term channel statistic (like the path loss), should be the same at both uplink and downlink frequencies.

For the performance evaluation through system level simulations, different approaches on DoA beamforming can be exploited. The complete solution requires using specific signal processing techniques in order to estimate the DoA from the uplink received signal. Among DoA estimation algorithms, the Multiple Signal Classification (MUSIC) and the Estimation of Signal Parameters via Rotational Invariance Technique (ESPRIT) can be mentioned and a detailed description of these types of beamforming algorithms can be found in 10.5.3.

For particular scenarios also an easier approach can be taken into account, based on the geometric direction instead of the estimated DoA. Whilst the latter depends on the spatial characteristics of the downlink multipath channel model, the geometric direction takes the angle between the eNB transmitter and the terminal receiver as the estimated DoA. i.e. it exploits the Line of Sight (LoS) as a dominant angle of arrival. This method significantly reduce the complexity in DoA estimation processing by still providing a good approximation of the real DoA and, hence, reliable performance.

10.5.2.2. Eigen Beamforming algorithm

Eigen-beamforming transmission for Multiple-Input Multiple-Output (MIMO) systems is an effective means to maximize the receiver Signal to Noise Ratio (SNR) especially in a noise limited environment, but may suffer some performance degradation when strong interference signals exist. While DoA beamforming described in previous Section 2.2 physically represents the steering of the beam in the desired direction, the eigen-beamforming mechanism does not directly have a similar physical interpretation. Instead of using the array-response vectors from the DoAs relative to the angle of all different users, downlink eigen-beamforming works with the downlink channel impulse response of each antenna element to find array weights that satisfy a desired criterion, such as SNR maximization or Mean Square Error (MSE) minimization (see also Section 10.5.3).

By using channel knowledge at the transmitter, this approach exploits the eigen-decomposed channel response in order to focus the transmit signal to the desired user, even if there are co-channel interfering signals with different DoAs. Since eigen-beamforming is a mathematical technique, rather than a physical technique, for increasing the desired power and suppressing the interference signals, it is more viable in realistic wireless broadband environments, which are expected to have significant local scattering.

In order to have a comparison between the single DoA method and the Eigen Beamforming algorithm a radiation diagram has been computed by means of both methods for an Urban Macro scenario and the achieved results are reported in FIGURE 98.

EARTH PROJECT

EARTH_WP4_D4.3 111 / 121

FIGURE 98. Radiation pattern comparison between Single DoA and Eigen BF algorithms.

EARTH PROJECT

EARTH_WP4_D4.3 112 / 121

10.5.3. MMSE

10.5.3.1. Introduction

In this section Beamforming is analysed from the view point of achieved energy improvement, compared with ordinary static sector antennas, providing the same service. Several scenarios, cell ranges, antenna configurations and cell’s load are evaluated in order to find the portion of energy that beamforming is able to save. Instead of long-term radiation patterns dependent on statistics for cell’s load and traffic, the goal is to generate individual radiation patterns to serve each user in the most efficient way, expecting to reduce the overall radiated power. The algorithm that is used in this work is MMSE. This section is organised as follows. The multi-user model and simulator implementation is described in the next section. Simulated scenarios and results are discussed in the following one. Conclusions are drawn at the end.

10.5.3.2. Model and simulator implementation

The main blocks of the simulation model used to evaluate the energy efficiency that can be achieved when beamforming is applied is illustrated in FIGURE 99. The adaptive antenna performance is measured in a single cell. The user that is at the cell of interest is the desired user and the users at adjacent cells are interferers, non-desired users for the BS of interest.

Disposition of users according to the scenario.Generation of UL control channel messages for each user Determination of pathlosses for each user

Users’ Generation

Simulation of array’s output, according to users’ direction and pathloss.

Received Signals

Weighting of antenna array to serve the desire user(s).Generation of radiation pattern for the desired user(s).

Weighting and radiation patterns

Computation of the power improvement achieved to serve the specified number of users

Metrics

Input Parameters

Output File

(…)

FIGURE 99. Simulation model.

The process of antenna weighting is based on UL control channel information. The signal sent by the desired user is used to weight individually each element of the antenna array. The objective is to focus the main antenna lobe into the direction of the desired user, and to null the direction of interferers. From the stage where the location of all intervening users is known, the path-loss for all users is determined depending on the scenario being considered.

The performance of the adaptive antenna is evaluated through the signal’s power improvement, referred to the reference antenna, . The performance achieved by the adaptive antennas. is evaluated as

the variation of radiated power to serve the desired user, relative to the reference static antenna, the metric being:

EARTH PROJECT

EARTH_WP4_D4.3 113 / 121

(1)

where is the adaptive antenna gain in the direction of the desired user.

The functions to generate users comprise aspects related to positioning, UL messages, and signal propagation, namely path-loss. In the Received Signals area, aspects about the reception of signals and the way how an antenna array receives a signal is addressed. Weighting and radiation patterns generation are responsible for correctly set the antenna array according to user’s position, as well as for generate the correspondent radiation pattern, to be assessed in Metrics.

The assessment of the simulator is a crucial step in order to perform simulations with confidence. The model developed and implemented in the simulator has its main function assigned for the antenna array weighting, which provides the best radiation pattern to ensure an improvement in the signal-to-interference ratio. The first assessment is the evaluation of antenna array weighting under some users’ locations situation, where the behaviour of the radiation patterns can be expected.

An example of a radiation pattern obtained during the assessment phase is presented in FIGURE 100. It can be observed that main beams are steered to the direction of the desired user, even if in the same direction there are several other non-desired users that will suffer extra interference. The awareness for interference suppression for the direction of non-desired users is visible through the location of secondary lobes. Secondary lobes are, as much as possible, steered into the directions where there are not any non-desired users. When there are no free directions from non-desired users, the radiation pattern directs the secondary lobes always outside the direction of major groups of non-desired users or steers the secondary lobes into the direction of the non-desired user that will not suffer so much severely from interference caused by others, usually the farthest.

(a): Radiation pattern no. 1.

(b): Radiation pattern no. 2.

FIGURE 100. Radiation patterns, with 8 antenna elements.

10.5.3.3. Simulation scenarios and results

5 10

15

30

210

60

240

90

270

120

300

150

330

180

0

Ante

nna G

ain

with r

efe

rence t

o t

he isotr

opic

ante

nna

2 4 6 8

30

210

60

240

90

270

120

300

150

330

180

0

Ante

nna G

ain

with r

efe

rence t

o t

he isotr

opic

ante

nna

EARTH PROJECT

EARTH_WP4_D4.3 114 / 121

Beamforming energy saving performance is evaluated for several general scenarios, where users are placed uniformly within the cell’s sector area and there are no constraints on urbanistic or demographic issues, in order to assess the energy gain when the scenario is generically classified as dense urban, suburban or rural/high speed, as defined in EARTH [EARTH-D2.2].

The dense urban scenario comprises the centre and main arteries of a city, with a strong commercial and services traffic, a substantial component of occasional users at the main public transports and streets (tourists, moving professionals, students) and a component of residential load. The dense urban scenario usually has a great concentration of BSs due to the demand for capacity, where cells’ areas just assume some hundreds of metres. The suburban scenario is modelled to correspond to typical city’s periphery, with major habitation blocks with several floors, that absorbs by night an important fraction of users that were at the city’s centre by day. Suburban cells do not need to be as near as at city’s centre, since users’ location and demand for capacity is usually sparser. The remaining territory corresponds to rural low dense populated scenarios that can be crossed by important highways, and covers areas with a few kilometres radius, where sparsely populated villages are located. The path-loss between the users and the BS is computed as proposed in [EARTH-D2.2].

Simulations for general scenarios are defined to evaluate the behaviour of power improvement when some parameters vary independently, i.e., number of antennas, cell’s radius and number of users. TABLE 14 shows simulation parameters being considered; for each antenna configuration (four or eight elements, half a wavelength inter-element spacing), the three scenarios were simulated with four cell’s radius ranges each, which were tested from 0 to 8 interferers in neighbouring cells. Eight interferers correspond to the eight nearest neighbour cells in front of the antenna, forming the so called first and second rings of cells of a cellular deployment. For each combination of scenario/interferers, 500 simulations were run in order to establish a mean value and a standard deviation.

TABLE 14. Simulation set for general scenarios.

Antennas 4 elements 8 elements Cell's Radius [km] Interferers Interferers

Dense urban

0.2 0.4 0.6 0.8

Suburban

1.0 1.5 0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8 2.0 2.5

Rural/High Speed

3.0 4.0 5.0 6.0

Fehler! Verweisquelle konnte nicht gefunden werden. FIGURE 101 shows the power improvement when cell radius varies for the same number of interferers in neighbouring cells. The first aspect is the constant behaviour (within one dB interval) of power improvement when the radius of cells changes, which can be explained with the attempt by adaptive antenna in positioning the main lobe into the direction of the desired user. When the distance of the desired user changes, the adaptive antenna also locks the main beam of the radiation pattern in the direction of the desired user and the power improvement is given by the extra gain of the main lobe, compared with the reference antenna. The gain of the main lobe of the antenna does not depend on the cells radius, hence the constant power improvement behaviour.

EARTH PROJECT

EARTH_WP4_D4.3 115 / 121

(a): 0 interferers, antenna with 4 elements.

(b): 4 interferers, antenna with 4 elements.

(c): 0 interferers, antenna with 8 elements.

(d): 4 interferers, antenna with 8 elements.

FIGURE 101. Variation of power improvement as a function of cells’ radius.

Another aspect that must be noticed is the change in standard deviation with the variation of cells’ radius. For the situation where no interferers are present, the standard deviation remains controlled, since it is just conditioned by the capacity of the adaptive antenna in positioning the main lobe in the direction of the desired user. When more interferer users are placed in neighbouring cells, besides the ability to direct the main beam to the desired user’s direction, there are radiation constraints on the direction of non-desired users, not being possible to position of high gain beams to serve the desired user. In this way, the standard deviation grows (up to 2 dB), which can make the operation of adaptive antennas even more expensive in energy than ordinary sector antennas, for four elements.

FIGURE 102 shows the variation of power improvement as a function of the number of interferers in the neighbouring cells.

-2

-1

0

1

2

3

4

5

6

7

0,1

0,6

1,1

1,6

2,1

2,6

3,1

3,6

4,1

4,6

5,1

5,6

Po

we

r Im

pro

vem

en

t [d

B]

Cell Radius [km] -2

-1

0

1

2

3

4

5

6

7

0,1

0,6

1,1

1,6

2,1

2,6

3,1

3,6

4,1

4,6

5,1

5,6

Po

we

r Im

pro

vem

en

t [d

B]

Cell Radius [km]

-2

-1

0

1

2

3

4

5

6

7

0,1

0,6

1,1

1,6

2,1

2,6

3,1

3,6

4,1

4,6

5,1

5,6

Po

we

r Im

pro

vem

en

t [d

B]

Cell Radius [km] -2

-1

0

1

2

3

4

5

6

7

0,1

0,6

1,1

1,6

2,1

2,6

3,1

3,6

4,1

4,6

5,1

5,6

Po

we

r Im

pro

vem

en

t [d

B]

Cell Radius [km]

EARTH PROJECT

EARTH_WP4_D4.3 116 / 121

(a): Dense Urban scenario (0.4km radius), antennas with 4 elements.

(b): Dense Urban scenario (0.4km radius), antennas with 8 elements.

(c): Suburban scenario (1.5km radius), antennas with 4 elements.

(d): Suburban scenario (1.5km radius), antennas with 8 elements.

(e): Rural/High Speed scenario (4km radius), antennas with 4 elements.

(f): Rural/High Speed scenario (4 km radius), antennas with 8 elements.

FIGURE 102. Variation of power improvement as a function of number of users for LTE.

-2

-1

0

1

2

3

4

5

6

7

0 1 2 3 4 5 6 7 8 Po

we

r im

pro

vem

en

t [d

B]

Interferers -2

-1

0

1

2

3

4

5

6

7

0 1 2 3 4 5 6 7 8 Po

we

r im

pro

vem

en

t [d

B]

Interferers

-2

-1

0

1

2

3

4

5

6

7

0 1 2 3 4 5 6 7 8 Po

we

r im

pro

vem

en

t [d

B]

Interferers -2

-1

0

1

2

3

4

5

6

7

0 1 2 3 4 5 6 7 8 Po

we

r im

pro

vem

en

t [d

B]

Interferers

-2

-1

0

1

2

3

4

5

6

7

0 1 2 3 4 5 6 7 8 Po

we

r im

pro

vem

en

t [d

B]

Interferers -2

-1

0

1

2

3

4

5

6

7

0 1 2 3 4 5 6 7 8 Po

we

r im

pro

vem

en

t [d

B]

Interferers

EARTH PROJECT

EARTH_WP4_D4.3 117 / 121

With the increment of interfering users, the power improvement of adaptive antennas decreases. However the difference between the maximum and the minimum improvement rounds only 1 dB, for all analysed scenarios and ranges and the trend for improvement decreasing is always present. When the number of interferers grows, also the constraints in interference limitation assumes more relevance, since the radiation pattern must be able to direct a high gain beam in the direction of the desired users, and at the same time to limit as much as possible the gain in the direction of the interferers. With more interferers, the requirements for interference limitation can imply that the gain of the desired user’s beam becomes lower than it would be if no interferers were present.

Another aspect is the increment of the standard deviation with the rise of interferers. When no interferers are active, power improvement does not float considerably. However, caused again by the growth of interference constraints, when the number of interference grows the standard deviation also becomes larger, and the capacity to predict power improvement of a certain scenario is reduced. In fact, the absolute value of the standard deviation does not constitute a problem, but its value, compared with the average one, is so high that can caused a situation where it is not possible to predict if adaptive antennas will save or waste energy.

Adaptive antennas with four elements are able to improve the propagation’s efficiency on average between 1.5 and 2.5 dB. Under certain conditions of users’ alignment, when it is possible to focus the main high gain beam in the direction of the desired user without creating extra interference in the non-desired ones, power improvements can reach up to 4.5 dB, corresponding to a power saving of 65%. Concerning antennas with eight elements, on average they achieve 2.5 dB, more than the one with just four elements. The average power improvement reached by antenna with eight elements is 4.5 dB, which is the maximum for antennas with four elements under the best conditions. Antennas with eight elements can reach a maximum of 6.5 dB, under certain conditions of users’ alignment, which corresponds to 77% of energy saved.

10.5.3.4. Conclusions

From the conceptual viewpoint, the introduction of adaptive antennas can be done at the BS, since some timing constraints related with the radio interface are respected. Timing is crucial to obtain good performance. Since an UL control channel is used, the periodicity on which the UE sends information cannot be too long, in order to be possible to correct update of users’ radiation patterns. LTE has a TTI of 1 ms. 10 ms between radiation pattern update would not be a problem for pedestrians, which move less than 1 m in this period, at a maximum of 30 km/h. However, considering a fast train traveling at 300 km/h, in 10 ms it is able to move more than 80 m, which can be a problem at short distances from the BS due to highly directive radiation patterns. The choice of the update period for the adaptive antennas must be, first of all, a routine that already is implemented into the radio interface, to avoid extra complexity. Other aspect that should be measured into the choice of the update period is the type of traffic. For high dense populated areas, with traffic generated mainly by pedestrian, there is no need to implement short update periods if people only move some metres in between updates, and a cell cover several hundreds of metres.

Concerning cell radius, LTE power improvement, compared with current static sector antennas, does not change with the variation on the cell’s radius, for the same number of interferers in neighbouring cells. When the number of interferers in neighbouring cells varies, the radiated power that is possible to save does not vary substantially, with the average improvement being near 2.0 dB for antennas with four elements and 4.5 dB for eight elements, which corresponds roughly to an average power reduction of 35% and 65%, respectively. The variation on the number of neighbour interferes affects mainly the standard deviation, which rises with the increment of the number of interferers. This behaviour is explained by the fact that the interferers are active in neighbour cells, more interference constraints adaptive antennas faces, which may imply that the main beam of the radiation pattern cannot be steered in the direction of the desired user due to the increment of interference that the radiation pattern creates. Adaptive antennas in LTE have a significant power efficiency, especially for arrays with eight elements where power saving of 65% is achieved on average.

EARTH PROJECT

EARTH_WP4_D4.3 118 / 121

The power efficiency that is achieved for arrays with eight elements is appealing, for example, to increase the coverage of a cell without spending extra power, since the gain that is introduced in the link between BS and users can make the difference between being covered or not, with positive consequences that new covered customers profits can bring for operators.

As reported in D4.2, the amount of radiated power corresponds roughly to 7 to 12% of the total consumed power of a BS, for the case of macro and micro cells, respectively. Hence, the expected global energy efficiency corresponds to a reduction of 2.5 to 7.5% of the total consumed power. These results are in agreement with the ones already reported in D4.2 for the case of simple beamforming (active antennas not being considered here) [EARTH-D4.2]. It should be stressed that the increased consumption due to extra signal processing needs are not taken into account in these figures.

EARTH PROJECT

EARTH_WP4_D4.3 119 / 121

10.6. CELL DTX

When de-activating the radio parts of a transmitter the term discontinuous transmission (DTX) is commonly used. DTX has for a long time been used as an important tool for achieving long battery life of mobile terminals. That is, in order to save energy the terminal only transmits when it needs to.

In order to save energy also at the network side, the proposal is to apply DTX also there. Examining the traffic in a cellular network with a short enough time resolution, e.g. on millisecond level, it is found that no user is served in most of the cells most of the time. On the other hand, on a longer time scale, e.g. 15 minutes, only very few cells carry no traffic. Consequently, there is a large energy saving potential for fast dynamic mechanisms that allow for a cell to put the radio parts into DTX. However, one possible obstacle for cell DTX is that even when there are no active users in the cell most current cellular systems, e.g. GSM, HSPA, LTE, 802.16m, mandate that the base stations transmit certain signals. Examples of such signals are reference signals (pilots), synchronization signals and broadcast information. Such idle-mode transmissions can significantly limit the feasibility of cell DTX. Here we will focus on LTE and analyze how these signals are designed and propose possible cell DTX modes. Also possible modifications of the LTE standard that will allow more efficient cell DTX modes will be discussed and evaluated.

The transmission format specified for the LTE downlink is orthogonal frequency division multiplexing (OFDM). Since OFDM is a multi-carrier transmission technique the physical radio resource can conceptually be described as a time-frequency grid where each resource element consists of one sub-carrier during one symbol. FIGURE 103 shows a LTE downlink radio frame that consists of 10 sub-frames of 1 ms duration each, numbered from 0 to 9. Consequently, the duration of the entire radio frame is 10 ms. At the beginning of each sub-frame there is a downlink control region (not shown in the figure) spanning one to three OFDM symbols. The primary and secondary synchronization signals (PSS and SSS, respectively) are transmitted in the last two OFDM symbols of the first slot in sub-frames 0 and 5 and the physical broadcast channel (PBCH) is transmitted over three OFDM symbols in the second slot of sub-frame 0. However, the most abundant signals are the cell-specific reference signals (CSRS) which need to be transmitted up to four times per sub-frame, and are used as demodulation reference, for channel quality estimates, and for mobility measurements as discussed in [EARTH-D4.1]. Finally, the white resource elements are intended for user specific data. There are four modes of DTX, ranging from duration from one or a few OFDM symbols to several hours. During the following sections, focus will be on three Cell DTX methods, and details on how they can be realized and operated will be given.

10.6.1. Micro DTX

An example of a downlink radio-frame in LTE is presented in FIGURE 103. It consists of 10 sub-frames of 1 ms duration each, numbered from 0 to 9. In FIGURE 103, one radio resource block element is visualized. It consists of 14 OFDM symbols in time domain and 12 sub-carriers in frequency domain.

FIGURE 103. LTE downlink radio frame with 10 sub-frames showing cell specific reference symbols (CSRS) for one antenna port, downlink control region (PDCCH) with a size of one OFDM symbol, primary and secondary

synchronization signals (PSS and SSS), and broadcast channel (BCH). Only micro DTX is possible.

EARTH PROJECT

EARTH_WP4_D4.3 120 / 121

The duration of the entire radio frame is 10 ms. At the beginning of each sub-frame there is a downlink control region (not shown in the figure) spanning one to three OFDM symbols. The primary and secondary synchronization signals (PSS and SSS, respectively) are transmitted in the last two OFDM symbols of the first slot in sub-frames 0 and 5 and the physical broadcast channel (PBCH) is transmitted over three OFDM symbols in the second slot of sub-frame 0. CSRS are used as demodulation reference, for channel quality estimates, and for mobility measurements as discussed in [EARTH-D4.1]. The white resource elements are intended for user specific data. By switching off the radio during the OFDM symbols in between the CSRS transmission, micro DTX can be realized in LTE.

10.6.2. MBSFN-based DTX

Beside the normal (unicast) data sub-frames, LTE Rel-8 also provides a special type of sub-frames, i.e. the multi-cast and broadcast single frequency network (MBSFN) sub-frame (which is standardized for broadcasting data from a single BS or couple of BSs to the UEs), as illustrated in FIGURE 104. This type of sub-frame consists of a short control region at the beginning of the sub-frame, similar to the normal sub-frames, while the rest of the sub-frame may be empty. Six (1, 2, 3, 6, 7, and 8) out of the ten sub-frames in a radio frame can be configured as such MBSFN sub-frames (LTE standard does not support configuration of 8 MBSFN sub-frames currently).

FIGURE 104. Example of a downlink radio frame in LTE with 6 MBSFN sub-frames showing downlink control region (PDCCH) with a size of one OFDM symbol, primary and secondary synchronization signals (PSS and SSS),

and broadcast channel (BCH).

One Rel-8 compatible method to improve the cell micro DTX scheme is to introduce one or several MBSFN sub-frames. This can reduce the average cell power use at low load since fewer cell-specific reference symbols need to be transmitted. Thus when the traffic load is low we can, in order to reduce energy consumption, configure a cell with up to 6 MBSFN sub-frames that are not used. Hence, in FIGURE 104 we see the minimum amount of signals that need to be transmitted in a cell with no traffic at all (in addition higher layer system information is required on a less frequent time scale than a radio frame, not shown in FIGURE 104). There is a significant possibility to reduce energy consumption with cell DTX in LTE Rel-8. When counting the actually transmitted OFDM symbols in a radio frame with 6 MBSFN sub-frames and assuming a PA wake up time of 30 μs it becomes apparent that the PA can be in low power DTX mode in 74% of the time when there is no traffic in the cell. This simple calculation alone gives some indication that cell DTX is an effective method for reducing energy consumption in LTE when there is no or very little traffic. It should be clear that introducing MBSFN sub-frames can lower the power use by allowing for a more efficient DTX scheme. There are, however, also negative consequences, both in terms of reduced cell capacity and degraded user throughput. Hence, it is of importance, if not inevitable, that fast switching between MBSFN configurations can be performed in an adaptive manner. The adaptive mechanism would switch from many

EARTH PROJECT

EARTH_WP4_D4.3 121 / 121

MBSFN sub-frames at low activity and react to changes in the traffic situation by removing the MBSFN sub-frames when traffic increases. The current mechanism for switching between MBSFN configuration is via system information and hence not a particularly fast method; in LTE Rel-8 we may only change the information on the system broadcast channel once every 40 ms according to specifications. Therefore, in future LTE releases it is important to allow for unicast data transmission also in sub-frames that are configured for MBSFN use.

10.6.3. Short DTX

By examining FIGURE 103 precisely, one realizes that the CSRS are the most abundant signals which need to be transmitted up to four times per sub-frame. CSRSs are used as demodulation reference, for channel quality estimates, and for mobility measurements. To put the radio in DTX mode, one logical idea is not to transmit the CSRSs all the time. CSRSs are used by the idle and active UEs to identify the cell during mobility measurements. If it is possible for the UE to perform the mobility measurements on the secondary synchronization signal (SSS) instead of CSRS, Although not included in the LTE standard today, this would enable turning off the radio chains in more than four sub-frames (between the transmissions in sub-frame 0 and 5). This mechanism is called short DTX and is depicted in FIGURE 105.

FIGURE 105. Proposed LTE downlink radio frame without CSRS. Short DTX can be applied.

earth_wp4_ead4.3

Documents

x earth project earth

earth deliverable d4

research project earth

energy efficiency improvements

energy efficiency potential

radio interface technologies

radio node

radio networks