from always‐on to always‐available to always‐optimized! · nb2. assumed device mix: 60%...
TRANSCRIPT
From Always‐on to Always‐available to Always‐optimized!
Presented By –Dr. Ylva Jading, Senior Specialist
Ericsson Researchylva.jading[at]ericsson.com
1st IEEE Energy Efficiency Tutorial:
Wednesday, September 19, 2018
ALL INFORMATION SHALL BE CONSIDERED SPEAKER PROPERTY UNLESS OTHERWISE SUPERSEDED BY ANOTHER DOCUMENT.
—An introduction and some background—Economy, Ecology and Engineering
—Putting energy performance on the agenda with EARTH—Metrics, Models, Mindset and some concepts
—From always‐on to always‐available with 4G/LTE—Cred, Trust and perhaps some Brilliant Mistakes…
—Operating closer to our highest potential with 5G/NR—Always‐optimized with ultra‐lean design and beamforming
2
Energy Performance – a story of change
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
ALL INFORMATION SHALL BE CONSIDERED SPEAKER PROPERTY UNLESS OTHERWISE SUPERSEDED BY ANOTHER DOCUMENT.
Introduction and Background
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
ALL INFORMATION SHALL BE CONSIDERED SPEAKER PROPERTY UNLESS OTHERWISE SUPERSEDED BY ANOTHER DOCUMENT.
—Energy efficiency has continuously improved over the years!
—So why do we still care about network energy consumption?
4
Why network energy performance?
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
ALL INFORMATION SHALL BE CONSIDERED SPEAKER PROPERTY UNLESS OTHERWISE SUPERSEDED BY ANOTHER DOCUMENT.
The big picture
4G
3G
2G1G~20% ~80%
RANCore / IP
The big picture
2G3G
4G5G
…Or can we do better?
BusinessAs Usual
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
Network energy performance
Minimizing total network energy consumption, despite increased traffic and service expansion
Economy Ecology Engineering
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
Perspectives and players
Energy performance – Same answer, different questions
Economy
Operators Vendors Governments
Ecology
Engineering
Reduce OPEXBranding
Capture Energy SpendPremium Brand
Reach CO2reduction targets
Technology for good“Walk the talk”
Drive energy and climate transition
“Less is more”
Size, weight, enable new deployments etc.
Subscriber growth in off‐grid areas
Enable new power solutions
Sustainable growth
Drive innovation and job creation
ALL INFORMATION SHALL BE CONSIDERED SPEAKER PROPERTY UNLESS OTHERWISE SUPERSEDED BY ANOTHER DOCUMENT.
Putting Energy Performance on the agenda
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
ALL INFORMATION SHALL BE CONSIDERED SPEAKER PROPERTY UNLESS OTHERWISE SUPERSEDED BY ANOTHER DOCUMENT.
LTE Rel‐8 (2008‐2009)― “CRS all the time over all the bandwidth” was identified as a problem very late
― No recognized models or methodology for evaluating NW energy consumption
No agreement to change Rel‐8 specifications late in the 3GPP process!
― “Backward compatibility” prevented us from doing it for later LTE releases
― Investigated and formulated the problem and started to build consensus…
Specifying 4G/LTE in 3GPP
Dense UrbanUrban
Suburban
Rural
SuperDenseUrban
Energy Efficiency Evaluation Framework (E3F)Deployment models Traffic scenarios
Power models 24h traffic profile
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
Dense UrbanUrban
Suburban
Rural
SuperDenseUrban
―Typical European country (excluding Nordic countries and Russia)
―NB: Extremely small share of Super Dense Urban
―Extremely large share of “wilderness”―43% area coverage, 89% population coverage
Large scale deployment model
Area type Pop. density[citizens/km2]
Area share [%]
Super dense urban 20000 0,05
Dense urban 3000 0,95
Urban 1000 2
Sub‐urban 500 4
Rural 100 36
Wilderness 25 57Source: The world bank, “population density”
61% of the population lives in the cities (SDU, DU, U, SU) in 7% of the area.
Data traffic demand per areaTypical European, 2014‐2020
Area 2014 2017 2020
Super Dense Urban 200 400 750
Dense Urban 30 60 110
Urban 10 20 40
Suburban 5 10 20
Rural 1 2 4
Wilderness 0,25 0,5 1
NB1. Figures show total demand. Should be multiplied by operator market share (default 1/3).NB2. Assumed device mix: 60% smartphone, 30% mobile PC, 10% tabletNB3. Figures are peak busy hour demand.
Device Type 2014 2017 2020
Mobile PC 4300 8031 15000
Smartphone 900 1775 3500
Tablet 1900 3800 7600
MB/month for different device types Peak traffic demand per area type [Mbps/km2]
―Traffic is un‐evenly distributed ―Many sites with low traffic
―Low and high traffic is everywhere
―On a millisecond time‐scale most cells are empty―Networks dimensioned for future peak‐hour traffic ―High traffic is rare and average traffic load is fairly low
Data Traffic
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
Low traffic dominates
Cells sorted in increasing average throughput [%]
Tim
e so
rted
in in
crea
sing
thro
ughp
ut [%
]
0 10 20 30 40 50 60 70 80 90 100
0
10
20
30
40
50
60
70
80
90
100 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Cells sorted w.r.t. increasing average throughput [% of cells]
Two weeks of traffic measurements from 670 cells in a metropolitan area
Sort
ed s
ampl
es p
er c
ell [
% o
f tim
e] Load relative to cell capacity
› Low average traffic
› Large variations
› Peak dimensioning
Characteristics
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
24h traffic model
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
―EARTH developed common power models
―We found that power consumption was approximately linear with load
―For Macro Base Stations ―Dominant consumer = Power amplifier (PA)
The Macro base station power model
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
Traffic vs energy consumption
Networks dimensioned for peak traffic demand
Network Traffic Load50%40%30%20%10%0%
Network Energy Consumption
Normal traffic Very high traffic Extreme traffic
Low average resourceutilization
Considerable static energyconsumption in networks
Improve load dependence!
Normal traffic Very high traffic Extreme traffic
ALL INFORMATION SHALL BE CONSIDERED SPEAKER PROPERTY UNLESS OTHERWISE SUPERSEDED BY ANOTHER DOCUMENT.
from Always‐onto Always‐available
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
Improving energy performance
“Reduce overall energy consumption in case of excess capacity”
Network management power
load
“Design energy efficient systems from the start”
System design and standardization power
load
“State of the art energy lean hardware and software”
Products and solutions power
load
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
Why does idle mode network energy consumption dominate in LTE?
“Empty” LTE radio‐frame
RF output power
Full load
Active
RBS power usage
Ref. Symbols, Sync, Sys Info
Sleep mode
Time‐scale matters
Antenna muting
Dynamic psi‐omni
Cell DTX
Ref. Cell DTX
Antenna muting
DynamicPsi‐omni
0%‐2%
‐6%
User p
erform
ance
s
10 ms
>100 ms
Source: EARTH
How quickly can this be turned on? Tr
affic
Low
Hig
h
freq.
freq.
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
―Circuits should be prepared for no or minimum leakage when not used
―Multiple power‐supply domains to allow parts of the board to be powered off
―Clock domains to allow parts of the board to be halted
―Intra‐connect with bypass possibilities allowing processing nodes to sleep while maintaining communication between remaining processing nodes
Proper HW Architecture
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
Proper SW Architecture
―SW architecture for need to support energy efficiency in an HW agnostic way―Differences in HW generations and HW version must be possible to hide from the SW
― Dynamic baseband resource allocation should be default―SW needs to ensure that it does not prevent idling HW from entering sleep state
―SW functions shall not spread out over more cores than needed
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
ALL INFORMATION SHALL BE CONSIDERED SPEAKER PROPERTY UNLESS OTHERWISE SUPERSEDED BY ANOTHER DOCUMENT.
Operating closer to our highest potential
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
NR: Energy performance embedded in standard
Load adaptiveenergy consumption
Only transmit whenand where needed
From always available to always optimized!
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
5G: One network – multiple use cases
@
A common network platform with dynamic and secure network slices
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
ALL INFORMATION SHALL BE CONSIDERED SPEAKER PROPERTY UNLESS OTHERWISE SUPERSEDED BY ANOTHER DOCUMENT.
―Specifying 3GPP New Radio, 2016‐2017
―Network energy consumption established as: “key performance criterion for IMT 2020”
―Ericsson successfully pushed for long DTX in 5G NR standard
―Now 20ms maximum SSB periodicity for standalone operation …and 160ms for non‐standalone!
Specifying 5G/NR in 3GPP
High level NR EE principles supported in Rel‐15―Time‐domain
―DTX ―stand‐alone operation with 20 ms DTX―non‐standalone operation with 160 ms DTX
―high rate and capacity enable “rush to sleep”―Frequency domain
―BWP adaptation―fast carrier aggregation activation―sparse grid for initial cell search
―Antenna domain―MIMO sleep (single antenna port for idle mode signals and channels)―support for wide‐beams (less beam‐sweeping)―beamforming to increase ISD
―Network domain ―network slices: one network multiple services
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
What to expect in short term
~ 2018:LTE +
time‐to‐market optimized NR
Early NR product consume more than mature LTE productsNR bandwidth significantly wider (up to 25 times more BW)Many more radio chains (due to massive MIMO)
~2020:LTE + energy performanceoptimized NR
Future:Energy optimized site
Today:LTE‐only
NR utilizing “ultra lean design” possibilities
Site ene
rgy consum
ption
ALL INFORMATION SHALL BE CONSIDERED SPEAKERS PROPERTY UNLESS OTHERWISE SUEPRSEEDED BY ANOTHER DOCUMENT
ALL INFORMATION SHALL BE CONSIDERED SPEAKER PROPERTY UNLESS OTHERWISE SUPERSEDED BY ANOTHER DOCUMENT.32
Conclusions― Typically low load dependence of
network‐energy consumption today
― Energy performance requires addressing low traffic cases
― Large energy‐saving potential in legacy NW parts― Dynamic on‐demand activation― The faster the better
― Key technical enablers for enhanced network‐energy performance in 5G― Ultra‐lean design: Longer and more DTX― Massive MIMO Beamforming: Increased ISD
ALL INFORMATION SHALL BE CONSIDERED SPEAKER PROPERTY UNLESS OTHERWISE SUPERSEDED BY ANOTHER DOCUMENT.
Thanks a lot for your time and attention!
Any questions and/or comments?
33
Q & A