Upload
trantu
View
215
Download
1
Embed Size (px)
Citation preview
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 1
Revisit the Cellular: A Hyper-cellular Framework for Green and
Smart (5G) Mobile Communication Systems
Electronic Engineering, Tsinghua UniversityTsinghua National Lab for Information Science and Technology
June 14, 2014
Zhisheng Niu
Tutorial at IEEE ICC2014, Sydney
Migration of Mobile Communications
• Cooper’s Law: “The data rate available to a wireless device doubles roughly every 30 months” (Martin Cooper)– This has held for over 50 yrs, leading to 1,000,000x increase – Technology: 1G (’80s) 2G (’90s) 3G (’00s) 4G (’10s)
“People always over-estimate things for 3 years scope,
but under-estimate things for 10 year scope”
– Bill Gates
What does 5G look like?
What will the enabling technologies be for 5G?
2
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 2
What’s the technology that mostly contributed to this success?
TDMA? CDMA? AMC? Turbo? OFDM? MIMO? ……
To answer this question, we need understand
25x 5x 5x
1600x
0
500
1000
1500
2000
widerspectrum
dividing thespectrum intosmaller slices
bettermodulationscheme
reduced cellsizes
Wireless Capacity…
Source: William Webb, Ofcom
It’s Cellular!
Cellular was invented for spectrum-efficiency
But, is it really energy‐efficient? Is it smart enough to support massive M2M connections?
4
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 3
This has not been a major concern
Energy of firewood:16.2 megajoules/kgOnly one bit: invasion or no-invasion
Extremely energy inefficient, yet needed5
But, it is a big concern today
• But, energy consumption and cost increased dramatically – Globally, #BS > 5 million, #Users>5 billion, EC> 100bn KWh (2012)
– As 4G/5G deploys and IoT boosts, EC & Connections grow dramatically
– Energy cost is also increasing (price and environmental impact)
How to carry 1000X traffic and connections using limited spectrum & energy?
6
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 4
EE as a key decision-making factor
Equipment Type TEEER Formula Min. TEEER Allowable
Transport ‐log (Ptotal / Throughput) 7.54
optical and Video 7.54
P2P Microwave 5.75
Switch/Router ‐log (Ptotal / Forwarding Capacity) 7.67
Media Gateway ‐log (Ptotal / Throughput) 6.54
Access (Access Lines / Ptotal ) +1 2.50
Power (POut Total / PIn Total ) X 10 9.20
Power Amplifier (Wireless)
(Total RF Output Power / Total Input Power) X 10
1.05
Base Station ? ?
Verizon’s TEEER (Telecom Equipment Energy Efficiency Rating) since 2009
www.verizonnebs.com/TPRs/VZ-TPR-9207.pdf
Ptotal = 0.35 x Pmax + 0.40 x P50% load + 0.25 x Psleep
7/30
Smartness was also not an issue, but
Densely and randomly deployed
2G/3G/4G Coexisting (HetNet)
J. Andrews, “Seven Ways that HetNets are Cellular Paradigm Shift”, IEEE ComMag, March 2013 8
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 5
Diversified Needs for 5G
• Mobile traffic will have another exp. growth by 2020– Capacity-hungry video dominates: higher SE and EE (Green)– Control-intensive massive connections access should be Smarter
9
5G Cellular: Greener and Smarter
Capacity-oriented
2000
Coverage‐oriented
Traffic Vo
lum
e or E
nerg
y Co
nsu
mp
tion Time
2G
Energy-oriented
2010
3G3G+
4G4G
Traffic Volume
Green&
Smart
2020
Energy Consumption
5G
10
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 6
LTE‐A
LTE
GREEN
PHY approach only is no more enough
Energy-Spectrum Tradeoff in Wireless Transmissions
11
– 1G (’80s): Analog, Voice, FDMA, Macro (Coverage-oriented)
– 2G (’90s): Digital, Voice, TDMA, Macro (Coverage-oriented)
– 3G (’00s): Digital, Data, CDMA, Micro (SE-oriented)
– 4G (’10s): Digital, Video, OFDMA, Pico/Femto (SE-oriented)
– 5G (’20s): Digital, Video/M2M, BDMA?, ????? (SE/EE-oriented)
12
5G: A Paradigm Shift of Cellular Architecture
Cell densification is trying to further improve SE, but is it
also good for EE and smart enough to support M2M?
12
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 7
Where is Energy Spent in Cellular Networks?
• 70~80% energy consumed by BSs in a cellular network
– Reducing the power consumption of BSs is the key!
Source: Ericsson13
Energy Waste in Existing Cellular
Traffic data from 319 HSPA cells in a European capital city measured from Jan. 1-22 2009 (Ericsson)
3 sector HSPA Site
1 25 50 75 1000
10
20
30
40
50
60
70
80
90
100
Load [%]
DC
Pow
er C
onsu
mpt
ion
[%]
OtherFans
RU3
RU2
RU1Base band
80% of the BSs are quite lightly loaded for 80% of the time, but still consume (waste) a lot of energy
14
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 8
Energy Waste in Existing Cellular
8 pm5 am
Real measurement data from a Chinese Operator in Zhejiang Province, Feb. 2013
15
Energy Waste in Cellular Networks
E. Oh, B. Krishnamachari, X. Liu, and Z. Niu, “Towards Dynamic Energy-Efficient Operation of Cellular Network Infrastructure”, IEEE Commun. Mag., June 2011
BSs are densely deployed and overlapping, further wasting energy
BS location data from a part of Manchester, UK.
16
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 9
Energy Waste in Cellular Networks
– All BSs are ON (active) all the time (in order to keep coverage), although traffic is almost zero in many areas
– Each BS almost transmits in peak power, although peak traffic only lasts for a very short time in most cells
– Multi‐BSs (small cells, HetNet) are densely deployed in many areas without any collaboration (work almost independently)
– As cell size is getting smaller AND traffic dynamics more bursty, energy waste is getting more serious
Business
Residential
Daytime Nighttime
Residential
Business
Residential
Residential
Business
Business
17
Why so many BSs under‐utilized, while still need to be densely deployed in some area?
Existing cellular is neither smart nor green
Why lightly‐loaded BSs can’t be switched off (sleep)?
- Mobile traffic is highly dynamic!
- BSs need to provide data services as well as network coverage simultaneously
18
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 10
Key idea: Reduce Energy Waste by Adapting to Real-traffic Dynamics (REWARD)
• Exploiting traffic dynamics (reduce energy consumption when traffic is low)
– Targeting THROUGHPUT rather than CAPACITY per joule
• Exploit energy model (much energy is consumed at BB/PA/AC rather than RF, therefore BS sleeping is the most efficient way for energy saving)
– Targeting TOTAL ENERGY rather than RF power reduction only
• Exploit cell collaboration (cell densification and HetNet make cell collaboration possible, helping to turn more BSs off)
– Targeting NETWORK rather than LINK/CELL performance
GREEN: Globally Resource‐optimized & Energy‐Efficient Networks
CapacityEE = ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
Power Consumption
Network ThroughputNEE = ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐
Total Energy Consumption
19
Tango: Traffic-aware network planning & green operation- Adapted to traffic distribution (temporally and spatially non-uniform) - Adapted to traffic characteristics (unicast, multicast, broadcast)- Adapted to QoS requirements (realtime, nonrealtime)
5G Cellular: Adapt to Traffic Dynamics(Traffic dynamics can provide opportunities for energy saving)
0:00 12:00 24:00
Power
t
Reduced Consumption
Usual Power Consumption (non-adaptive)
Traffic
Key challenge: How to guarantee the coverage and QoS?How to model and predict traffic dynamics?
Z. Niu, “TANGO: Traffic-Aware Network Planning and Green Operation”,IEEE Wireless Commun., Oct.2011 (invited article)
BS Sleep
Power Adaptation
20
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 11
Example: Cell Zooming
Z. Niu, Y. Wu, J. Gong, Z. Yang, “Cell zooming for cost-efficient green cellular network,” IEEE ComMag, Nov. 2010 (IEEE APB Best Paper Award 2014)
• Cell Zooming for Smart Cellular Network
Central cell zooms in as traffic load increases
Central cell zooms out as traffic load decreases
Central cell sleeps as traffic load getting quite low
21
A Dynamic Programing Approach for BS Sleeping
x-axis (m)
y-a
xis
(m
)
500 1000 1500 2000 2500 3000
500
1000
1500
2000
2500
High Load
Medium
Low Load
Active cells
Sleeping cells
J. Gong, S. Zhou, Z. Niu, “A Dynamic Programming Approach for Base Station Sleeping in Cellular Networks,” IEICE Trans. Commun., Vol.E95-B, No.2, pp.551-562, Feb. 2012
22
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 12
5G Cellular: Adapt to Environment(BS collaboration can provide opportunities for energy saving)
• CHORUS: Collaborative & Harmonized Open Radio Ubiquitous System– Open Radio: spectrum in HetNet are shared by multi-modal terminals (software defined radio)
– Globally optimized: cross-layer cross-node cross-network/system (software defined network)
cross-netw
ork/systemdesign
cross-layer cross-node design
[1] S. Zhou, Z. Niu, S. Tanabe, “CHORUS: Collaborative and Harmonized Open Radio Ubiquitous Systems”, 4th Intl. Conf. Commun. Sys. & Nets. (COMSNETS), Bangalore, India, Jan. 2012 (invited)[2] S. Zhou, Z. Niu, S. Tanabe, and P. Yang, “CHORUS: Framework for Scalable Collaboration in Heterogeneous Networks with Cognitive Synergy,” IEEE Wireless Commun. Mag, accepted, 2012
Challenges: 1) How to detect the NSI? (information explosion and incompleteness?)2) How to virtualize the network resources? (self-optimizing networks)
HetNet
23
Example: BS Sharing
B. Leng, P. Mansourifard, B. Krishnamachari, Z. Niu, “Microeconomic Analysis of Base-Station Sharing in Green Cellular Networks”, IEEE INFOCOM 2014, Toronto, Canada, April 2014
24
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 13
5G Cellular: Deal with the Dilemma
Capacity‐hungry Apps (e.g., mobile videos)
Higher SE
Higher EE
Control‐intensive Apps (e.g., M2M, social networking)
Faster Connectivity
Higher Reliability
?Smaller cells Larger cells
C-plane larger
D-plane smaller
decouple
Less signaling overhead
Global optimization
SmartCoverage-on-demand
Densely deployed
Green25
Bottlenecks of Existing Cellular Architecture
• GREEN involves more controls/signaling exchanges– TANGO needs to get traffic and QoS information of neighboring nodes
– CHORUS needs to collect network-state information (NSI) of other nodes/networks in addition to channel state information (CSI)
– Overhead of signaling traffic will get higher as cell size gets smaller
• Deeply coupled structure can’t flexibly adapt to traffic – Tight coupling of the coverage for control signals and data signals makes it
less flexible to the traffic variation and frequent handovers, and less friendly to cross-network design
– Users in sleeping cells will be shadowed from the network
NSI
control
data
26
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 14
Hyper Cellular for Green and Smart
• Decouple control and traffic coverage so that data cells could be more adaptive to traffic dynamics and network state, and control cells can take global optimization
Traffic analysis
Broad
ban
dNarro
wban
d
Signalin
g
Contro
l
GSM
3G
Macro
Micro
Hyper
27
Control BSs serve as a Central Controller?
For on‐demand data services
Unified Signaling Network (Control Plane)for Software‐Define Wireless Networks?For always‐ON coverage
Traffic BS Traffic BS
28
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 15
Advantages of Hyper-Cellular Architecture
• Energy-on-Demand– Coverage on-Demand:Macro/Micro/Pico/Femto?
– Resource on-Demand:Full-Power/Half-Power/Sleep?
– Service on-Demand: Realtime/Non-realtime/Soft-realtime?
Unicast/Multicast/Broadcast?
• Global resource optimization– Easily match users to the best BSs depending on the user requests
– Data cells can easily adapt to the traffic and network state changes
– Friendly to management of heterogeneous networks
control
micro
macro
29
Cellular Network Architecture Migration
• R99R5R8 (3GPP):from tree to mesh, decoupling to cooperate and reconfigure – Decouple of RNC and BTS
– Decoupling of BBU and RRU
30
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 16
Hyper-cellular: Virtual (Elastic) Coverage
Decouple of signaling coverage and traffic coverage
Signaling coverage is seamless and traffic coverage is reconfigurable
Challenge: unified signaling for virtualized cellular NWs
U‐Plane
C‐Plane
Seamless
Elastic
31
Hyper-cellular: Virtual Cells
Decouple of antennas and AI processing
No. of antenna is independent of computation units
Node C3 Node C4
Node C1 Node C2
Cable/Fiber
Node A
V-NodeB1
V-NodeB2 MT1
MT2
V‐Cell
32
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 17
Hyper-cellular: Virtual (Cloud) Computing
• Decouple of AI functions and computational resources
• The AI functions are processed in any unit, and computed offloading
33
Combine Cellular with Cloud
J. Liu, T. Zhao, Y. Chen, S. Zhou, Z. Niu, “CONCERT: A Cloud-Based Architecture for Next-generation Cellular Systems”, submitted to IEEE Wireless ComMag, 2014
C‐RAN
CONCERTS: CONvergence of Cloud and cEllulaR sysTems
34
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 18
C-RAN: It’s just the first step to SDN
35
From DWCS to CRAN
Traditional BTS Distributed BTS C-RAN
• Traditional BTS system– (Huge) integrated system– BTS & supporting facility
require indoor protection– Long RF cable to antenna
• Distributed BTS– Outdoor RU, Indoor BBU– DU to multi-RU– DU-RU connected via
point-to-point dark fiber
• C-RAN– Centralized processing– Collaborative Radio– Open platform towards
Cloud computing
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 19
CM’s Vision on NG Wireless Network
37
IT based core network
Anchor BS
Nano AP
Virtual BB pool
Content Pool
Anchor BS
骨干站点
RRU
Relay D2D
relay
D2D
Indoor Coverage
User Centric Access Network Supporting exclusive usage of available spectrum of each user
Branch BS
LSAS
Technical Challenges
• How to decouple signaling from data coverage? How to integrate the signaling functions of HetNets? – Complete decoupling may lead to new bottleneck and delays due to frequent
visits to signaling-BSs (main difference from BCG2), but which functions should be left into the data-BSs?
• How to guarantee signaling coverage highly reliable? – Need new protocol for S-BSs. Also, tradeoff between reliability and delay
• How to detect user behaviors, QoS requests, terminal capability, and provide services in an EE manner? – Data mining, cognitive radio, on-line learning, …
• How to locate users and associate them to the best D-BS? – The best cells may be in sleeping state, activate or not?
• How to balance the EC of network parts and user terminals? – User terminals need to keep associations with S-BS in a wider scope
• ……Z. Niu, S. Zhou, S. Zhou, X. Zhong, J. Wang, “A Hyper-Cellular Paradigm for Globally Resource-optimized and Energy-Efficient Networks (GREEN)”, Science in China, Sep. 2012 (in Chinese) 38
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 20
Enabling Technologies for 5G
• Massive MIMO and dense small cell networks (for throughput improvement)
• Highly flexible/reliable and realtime MAC protocol (for efficient support of IoT applications)
• Advanced interference and mobility management
• Cognitive or smart radio technologies (for spectrum efficiency)
• Single frequency full duplex radio technologies
• mmWave (for wireless backhaul and/or access)
• Pervasive networks (for multihoming or multiple concurrent data transmission)
• Multi-hop networks and D2D communications (for coverage extension)
• IPv6 (for seamless handover and roaming)
• Virtualized and cloud-based radio access infrastructure (for network flexibility:
different slices of the network with different technologies for different applications)
• World wide wireless web (WWWW) (for comprehensive wireless-based web applications that include full multimedia capability beyond 4G speeds)
• Wearable devices with AI capabilities (for augmented reality)39
Global Research Activities on “5G”
• “BDMA and Relay with group cooperation” (Korea, 2008)
• “5G Communications Research Lab” (Univ. of Dresden, 2012.5)
– Jointly funded by National Instruments
• “₤35m for 5G Research Centre” (Univ. of Surrey, 2012.10)
– jointly funded by UK Research Partnership Investment Fund (UKRPIF) and a consortium of Huawei, Samsung, Telefonica Europe, Fujitsu Laboratories Europe, Rohde & Schwarz, and Aircom International
• “China launched a WG on 5G” (China Academy of Telecom Research, 2012.11)
• “Huawei invests $600m for 10Gbps 5G network” (2013.11)
• “Korea to spend $1.5 billion on 5G mobile network” (2014.1)
• “China Mobile joined NGMN 5G Alliance” (MWC2014, 2014.2)
40
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 21
Global Research Activities on “5G”
• “€50m EU research grants to develop '5G' technology” (EC, 2013.2)– METIS: Mobile and wireless communications Enablers for Twenty-twenty (2020)
Information Society (SWE Ericsson, 29 partners) specifies 5G should provide 1000X higher mobile data volume per area
10‐100X higher No. connected devices for Internet of Things 10‐100X higher typical user data rate
10X longer battery life for low power M2M Communications 5X reduced e2e latency
– 5GNOW: 5th Generation Non-Orthogonal Waveforms for Asynchronous Signalling (GER)
– iJOIN: Interworking and JOINt Design of an Open Access and Backhaul Network Architecture for Small Cells based on Cloud Networks (ESP)
– TROPIC: Distributed computing, storage and radio resource allocation over cooperative femtocells (ESP)
– COMBO: joint optimisation of fixed and mobile access (GER)
– MOTO: Mobile OpportunisTic Traffic Offloading (FRA)
– PHYLAWS: PHYsical LAyer Wireless Security
• "5GrEEn - Towards Green 5G Mobile Networks“ (EIT ICT Labs. 2013.9) 41
Green Activities in China
End-to-End Energy Efficient Networks (National 863 Program, 2012~2015)
Green Radio Excellent in Arch. and Tech. (Huawei Program, 2010 ~ )
Globally Resource-optimized and Energy-Efficient Networks (National 973 Program, 2012~2016)
2014/06/14 42
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 22
5G Study in China (“863 Program”, $26m in 2014, 2nd phase in 2015)
5G NET(Hyper, Cognitive, Dense)
(THU, BUPT, Huawei)
5G PHY(Massive Distributed MIMO)
(Southeast University)
5G Testing and Verification (Wireless Communication Research Institute, Shanghai)
5G Overall Description & Standardization
(China Academy of Telecom Research, MII)
43
5G: Key Questions
• What does the network architecture of 5G look like?
How can make cellular architecture more smart and green?
Should we also include WiFi into 5G family? How to offloading?
How does CLOUD and/or SDN help & how to combine with them?
• What should be the fundamental components and the enabling technologies in order to make it happen?
Cognitive radio and networking
Software defined radio and networking
Content delivery network
WiFi offloading
• How to migrate to the new architecture from today?44
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 23
5G: Major Design Objectives
• Implementation of massive capacity and massiveconnectivity
• Support for an increasingly diverse set of services, applications and users – all with extremely diverging requirements
• Flexible and efficient use of all available non-contiguous spectrum for wildly different network deployment scenarios
5G is the next frontier of innovation for entire mobile industry (paradigm shift)
45
5G: Timeliness of the Topic
IEEE Communications Magazine Special Issue on 5G Wireless
Communication Systems: Prospects and Challenges (Deadline 15 Sept.
2013)
IEEE JSAC Special Issue on 5G Wireless Communication Systems (Deadline
4 Dec. 2013)
IEEE Communications Magazine Special Issue on Millimeter Wave
Communications for 5G (Deadline 1 Feb. 2014)
IEEE Communications Magazine Special Issue on 5G Networks: End‐to‐
end Architecture and Infrastructure (Deadline 1 Feb. 2014)
• CFPs from leading COMSOC journal/magazine
46
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 24
5G: Keywords from the above CFPs
Definition of 5G
Deployment requirements
Heterogeneous and small cell networks (HetSNets)
Cloud-based RAN
Massive MIMO
Energy‐efficiency
Cognitive and reconfigurable
Heterogeneous architecture Interference
coordination Multi-media traffic "no cell" concept
CoMP
Cloud data centers
Collaborative communications
Programmable optical backbone
3D Audio and Video
5G evaluation tools and testbeds
mmWavecommunications
Beyond OFDMA
Full Dimension MIMO
Channel aggregation 5G PHY 5G NET
47
Multi-antenna transmission/reception
5G: PHY vs NET Solutions
Full duplex, network coding, …
Multi-layer coordinationCoordination
Multi-site transmission/reception
Interference suppression
Physical‐layer evolution will remain important
But the main aim for the PHY evolution will be to enable more advanced system‐level features
Courtesy: Erik Dahlman (Ericsson) 48
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 25
Some Research Progress in 2012/13
How much energy can be saved by Separation?
Traditional Cell Hyper Cell
in outP k P b
50
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 26
How much energy can be saved by separation?
Total power consumption of a cell: more than 50% saving
Average power consumption of a cell: robust to cell size
Z. Wang, W. Zhang, “The Capability of A Separation Architecture for Achieving Energy-efficient Cellular Networking“, IEEE TWC, 2013 (accepted)
51
Elastic Coverage for Non-Uniform Mobile Traffic
1 1m m m
m m m
Ah
A B
Ratio of traffic in hot areas over all traffic
Ratio of hot area over total area
Hyper-Cellular
Always-on Control BS for
traffic in cool area
On-demand Traffic BS for
traffic in hot area
Stochastic Geometry Theory
Data rate in hot area vs data rate in cool area (bps/Hz/m2)
Hot area vs cool area
Grouping degree
52
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 27
Energy Saving Gain by Elastic Coverage
If (ratio of basic power of micro and Macro BSs)ESG increases as h increases
/c cm m MP P
IfESG decreases as h increases
/c cm m MP P
0 0.1 0.2 0.3 0.4 0.5 0.60
0.5
1
1.5
2
2.5
3
3.5
m
= 0.02
m = 0.07
m = 0.12
m = 0.17
m = 0.22
m = 0.27
m = 0.32
m = 0.37
m = 0.42
m = 0.47
h
Netw
ork En
ergy Efficiency
53
Energy Saving Gain by Elastic Coverage
Parameter Value
CBS Power 373W/sector
TBS Power 11W
No. of hot spots in CBS
4, 10
Radius of TBS
40m, 50m
Radius of CBS
288m
No. of users in TBS
4, 6
No. of users in CBS
60
Up to 42%~300% ESG by elastic coverage in 3GPP typical scenario
For h=80%, γm=23%, ESG can go to 340%
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.80
50
100
150
200
250
300
350
rm =23%
h
3GPP Typical Parameters
Energy Savin
g Gain
(%)
rm =14%
rm =9%
54
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 28
• Separation is not easy!– Challenge 1: difficult to categorize (millions of signal types)
– Challenge 2: Difficult to separate (complicated signal interactions)
– Challenge 3: difficult to manage (synchronization)
Standard Signal TypesCategorize & SeparateCategorize & Separate
Signal Types
Standard
State
FunctionalityFunctionality Separation?
How to Separate? - Principle
2014/06/14 55
How to Separate? – State Definition
X. Xu, G. He, S. Zhang, Y. Chen and S. Xu, “On Functionality Separation for Future Green Mobile Network: Concept Study over LTE”, IEEE ComMag, May 2013
56
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 29
How to Separate? - State–Functionality Mapping
State UE Activities
Network Functionalities
Syn.Broadcast of
System Information
Paging Multicast Unicast
Detached Cell Selection √ √
Idle
Acquisition and Update of System Configuration
√ √
Monitoring of Upcoming Transmission Notification
√
Cell Reselection √ √Receiving of MBMS √
Active
Acquisition and Update of System Configuration.
√ √
Monitoring of Upcoming Transmission Notification
√
Cell Handover √ √ √Receiving of MBMS √
Transmission of UE-Specific Data
√
2014/06/14
How to Separate? - Functionality–Signal Mapping
Network
Functionality
Signal Types
Syn. PilotFrame
Control
System
Info.
Bearer
Paging
Info.
Bearer
Multicast
Info.
Bearer
Unicast
Info.
Bearer
Syn. √ △
Broadcast
of System
Information
√ △ √
Paging √ √ √
Multicast √ √ √
Unicast √ √ √
△ means this relationship may change among different standards. For example, in GSM/UMTS system, the location of system information bearer is pre-defined and the frame control signal is omitted. However, in LTE systems, the location of system information bearer will be dynamic and the frame control signal is mandatory.
2014/06/14
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 30
How to Separate? - Mapping to 3GPP Standard
Signal Types3GPP Standard
GSM UMTS LTE
SynchronizationFCCHSCH
SCH PSS/SSS
Pilot TSCCPICHDPCCH
S-CCPCHRS
Frame ControlAGCHSACCH
PICHMICHAICH
DPCCHS-CCPCH
PHICHPCFICHPDCCHPMCH
Paging Inform. Bearer
PCH S-CCPCH PDSCH
System Inform. Bearer
BCCHSACCH
P-CCPCHPBCH
PDSCH
Multicast Inform. Bearer
CBCH S-CCPCH PMCH
Unicast Inform. Bearer
SDCCHSACCHFACCH
TCH
S-CCPCHDPDCH
PDSCH
2014/06/14
Channel Separation
Channel separation scheme for 3GPP standards
Implementation for GSM/GPRS protocol
60
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 31
A Lab Demo using USRP and OpenBTS
T. Zhao, P. Yang, H. Pan, R. Deng, S. Zhou, and Z. Niu, “Software Defined Radio Implementation of Signaling Splitting in Hyper-Cellular Network,” ACM SIGCOMM Workshop of Software Radio Implementation Forum (SRIF 2013), Hong Kong, Aug. 2013.
61
Lab Demo of Hyper Cellular Concept
GPRS data separation
Simple DBS scheduling scheme
SBS log
DBS logPlatform setup
62
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 32
Service Flow
UE associates to Signaling BS.
UE requests for traffic channel,Signaling BS assigns a Data BSand informs UE.
Traffic data transmission is established between UE and Data BS.
63
Establishment of Transmission
DBS DBS
Channel Request
UE SBS
Request Message Request Message
Assignment Message Assignment Message
Immediate Assignment
RLC/MAC BlockRLC/MAC Block
RLC/MAC Data Block RLC/MAC Data Block
Packet Uplink ACK Packet Uplink ACK
Data Base Station Scheduling
64
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 33
Traffic-Aware Dynamic BS Sleeping
• Settings
– 10x10 hexagon cells
– Cell Radius 200m
– Binary BS power
– Link parameters according to
ITU micro-cell test environment
– Traffic:
3 hotspots in the network – space
– Hotspot center traffic , 1st tier traffic , 2nd tier traffic
, others ,
Average intensity varies ‐ time
( )h t 1 ( )h t
2 ( )h t 3 ( )h t 3 2 10 1
Jie Gong, Sheng Zhou, Zhisheng Niu, “A Dynamic Programming Approach for Base Station Sleeping in Cellular Networks,” IEICE Trans. Commun., Feb. 2012
65
Traffic-Aware Dynamic BS Sleeping
• Compare with uniform sleeping alg. [Marsan’09]– Multiple sleeping pattern– Active BSs uniformly located
(0.88 0.63 0.50) (0.83 0.50 0.33) (0.81 0.44 0.25) (0.80 0.40 0.20) (0.79 0.38 0.17)60
61
62
63
64
65
66
67
68
69
70
Ave
rag
e N
o. o
f Act
ive
BS
s
(0.88 0.63 0.50) (0.83 0.50 0.33) (0.81 0.44 0.25) (0.80 0.40 0.20) (0.79 0.38 0.17)10
-4
10-3
10-2
10-1
Ave
rag
e B
lock
ing
Pro
ba
bili
ty
(1 2
3)
Uniform alg. ave. active BSs
DP alg. ave. active BSs
DP alg. ave. blocking
Uniform alg. ave. blocking
DP algorithm performs better as the hotspots become hotter
Non‐uniformity increases
66
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 34
How densely should D-BSs be deployed?
• Problem: For given QoS, how densely should the DBs be deployed for a given coverage and QoS guarantee? – BS density should adapt to traffic dynamics (e.g., cell zooming, BS sleeping)– Deploying more smaller BSs may save energy ?!(increasing sleeping
opportunity)
[1] Z. Niu, Y. Wu, J. Gong, Z. Yang, “Cell zooming for Green cellular networks”, IEEE Com Mag, Nov. 2010 [2] X. Weng, D. Cao, Z. Niu, “Energy-Efficient Cellular Network Planning under Insufficient Cell Zooming”, IEEE VTC2011-spring, Budapest, Hungary, May 2011
0 1 2 3 4 5 60
1
2
3
4
5
6
0.5
1
1.5
2
2.5
3
3.5
4x 10
-4
0 10 20 30 400
0.05
0.1
0.15
0.2
0.25
到率
Temporal Dynamics Spatial Dynamics Insufficient Zooming Sufficient Zooming
67
Optimal BS Density for Green(Regular Deployment Case)
• Normalized EC vs Inter-BS Distance (PB<2%)
Deploying more smaller BSs can save energy!!!
1. Z. Niu, S. Zhou, Y. Hua, Q. Zhang, and D. Cao, “Energy-aware network planning for wireless cellular systems with inter-cell cooperation”, IEEE TWC., vol.11, no.4, pp.1412-1423, 2012
2. Y. Wu, Z. Niu, “Energy Efficient Base Station Deployment in Green Cellular Networks with Traffic Variations”, IEEE ICCC2012, Beijing, China, Aug. 2012
0 5 10 15 20 250
0.05
0.1
0.15
0.2
0.25
时间(小时)
到达
率(/
Km
)
业务模型 1业务模型 2
Time (h)
Arrival rate (/km
)
300 400 500 600 700 8000.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
基站部署间距 (m)
归一
化网
络能
耗
业务模型 1业务模型 2
Traditional planning
EE planning
Inter‐BS Distance (m)
Norm
alized EC
68
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 35
Optimal BS Density for Green(Heterogeneous & Stochastic Deployment Case)
1. Two‐tier PPP models with BS density ρM and ρm
2. Always connect to the BS with highest SNR (not necessarily the nearest)
Weighted Poisson‐Voronoi Tessellation:
f(A) follows ‐distribution with density
where:Stochastic Geometry Modeling
0 10 20 30 40 50 60 70 80 90 1000
10
20
30
40
50
60
70
80
90
100
X coordinate
Y c
oo
rdin
ate
D. Cao, S. Zhou, Z. Niu, “Optimal Combination of Base Station Densities for Cost-Efficient Two-tier Heterogeneous Cellular Networks”, IEEE TWireless, Sep. 2013
69
B. Rengarajan, G. Rizzo, and M. A. Marsan, ``Bounds on QoS-Constrained Energy Savings in Cellular Access Networks with Sleep Modes’’, ITC 2011, pp. 47-54, San Francisco, USA, Sep. 2011.
Bay area of Sydney, Australia.Dense deployment: 81.64 per Km^2
Verification of PPP Models
Australian Geographical Radio Frequency Map (http://spench.net/)
70
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 36
Verification of PPP Models
Distribution of BSs in a Square Area of a Chinese Operator
Rural Dense Urban
No. of BSs No. of BSs
71
Verification of Gamma Distribution
Distribution of Cell Areas of a Chinese Operator
72
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 37
QoS Metrics
• Coverage Probability
If α=4,
• Service Outage Probability
73
Optimal BS Density - Formulation(Homogeneous Case)
74
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 38
Optimal BS Density – Upper Bound(Homogeneous Case)
75
Optimal BS Density – Lower Bound(Homogeneous Case)
76
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 39
Optimal BS Density - Performance(Homogeneous Case)
Table: Optimal BS density for 3 typical scenarios (in BSs/Km2, EARTH model)
Table: Optimal BS density with transmit power adaption
Conclusion: noiseless assumption is acceptable for suburban and dense urban scenarios, but not in rural scenario
Conclusion: Joint BS density adjustment and transmit power adaption can help save more energy
D. Cao, S. Zhou, Z. Niu, “Optimal Combination of Base Station Densities for Cost-Efficient Two-tier Heterogeneous Cellular Networks”, IEEE TWC, Sep. 2013 77
Optimal BS Density and Tx Power(Heterogeneous Case)
1. Two‐tier PPP models with BS density ρM and ρm
2. Always connect to the BS with highest SNR (not necessarily the nearest)
Weighted Poisson‐Voronoi Tessellation:
f(A) follows Gamma distribution with density
where:Stochastic Geometry Modeling
0 10 20 30 40 50 60 70 80 90 1000
10
20
30
40
50
60
70
80
90
100
X coordinate
Y c
oord
inat
e
where {CM , Cm} are deployment (energy) cost
Coverageguarantee
78
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 40
Optimal BS Density – Near-optimal Solution(Heterogeneous Case)
;
79
Optimal BS Density and Tx Power(Heterogeneous Case)
• Dynamic BS Sleeping in Dense Urban Scenario (EARTH Model)
– CM = 780 + 28.2PM , Cm = 112 + 5.2Pm
– PM = 20W, Pm =2.42W = 0.0927 < c-1=0.3162
– Reference model: macro-only homogeneous network with no BS sleeping: total energy consumption=3.26 KW/Km2
0.82 (average)(75% saving)
Conclusion: Joint optimization of Macro/Micro‐BS densities can help to save more energy!
80
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 41
Optimal BS Density – Optimal Policy(Heterogeneous Case)
;
If <c=0.3162, preferentially add micro BSs or sleep macro BSsIf >c=0.3162, preferentially add macro BSs or sleep micro BSs
Ratio of Micro‐BS density and Macro‐BS density ()
Total Energy Density
81
Heterogeneous Networks with PSR
• PSR (Partial Spectrum Reuse) to reduce over-provisioningand potential interference (to macro BSs and among micro BSs)
Total Spectrum
Macro BS
Micro BS1
Micro BS2
Micro BS3
D. Cao, S. Zhou, Z. Niu, “Improving the Energy Efficiency of Two-Tier HetwrogeneousCellular Networks through Partial Spectrum Reuse”, IEEE TWC, Aug. 2013
Optimal β=Wm/WM?
If β<1, allocate FULL spectrum to macro BSs and PARTIAL spectrum to micro BSs; If β>1, vice versa.
2
; ( )m M
M m
C Pe c
C P
82
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 42
Energy Saving Gain by PSR
PSR achieves the near-optimal performance
PSR can save up to 50% of energy consumption
83
Application to Network Planning – Capacity Extension (EARTH Model: Dense Urban, Peak Traffic increases up to 74.3/Km2)
ρM=1 BS/km2 (¾ used for coverage), EC=5.9kW/km2
Macro BSs for coverage guarantee
Other macro BSs (could be switched off)
Newly added BSs for capacity extension
Network Topology before capacity extension
Km
Km
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 20
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Adding macro BSs: ρM 1.75 BSs/km2
EC 3.56 kW/km2 (40% saving)
Capacity extension by homogeneous BSs
Km
Km
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 20
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Km
Adding micro-BSs: ρm 4.25/Km2
EC 1.87 kW/km2 (48% further savings)
Km0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 20
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Capacity extension by heterogeneous BSs
84
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 43
Application to Energy Saving – BS Sleeping(EARTH Model, Dense Urban)
Network Topology during Peak Traffic (75/Km2)
ρM=1 BS/km2 , ρm=4.25 BS/km2, EC=1.87 KW/Km2
Macro BSs for coverage guarantee
Other macro BSs (could be switched off)Micro BSs
Km
Km
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 20
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Awake 35% micro BSs: ρm=1.5 /km2, EC=1.18KW/Km2 (↓37%)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 20
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Traffic Load up to 50% (37/Km2)
All unnecessary BSs going to sleep, ρM=0.75/km2, EC=0.97KW/Km2 (↓50%)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 20
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Traffic load down to 20% (15/Km2)
85
How long should a BS sleep?
• Energy-Delay Tradeoff (EDT) in BS Sleeping Control– Longer sleep can save energy, but bring delays to customers
• Wake-up Policies– N-Policy: wake up whenever N new requests come during sleep– SV-Policy: wake up after a random sleep and then keep awake– MV-Policy: wake up after a random sleep and sleep again if find no requests
• Challenge #1: Both energy and delay concepts need to be extended– Energy = transmitting power + circuit (processing) power + basic power– Delay = transmitting delay + queueing delay + sleeping period
• Challenge #2: EDT should be evaluated in the whole network wide– EDT on link-level single-cell level multi-cell level
1. Z. Niu, Jianan Zhang, Xueying Guo, Sheng Zhou, “On the Energy-Delay Tradeoff in Base State Sleep Mode Operation”, IEEE ICCS2012, Singapore, 21-23 Nov., 2012 (invited)2. X. Guo, S. Zhou, P. R. Kumar, Z. Niu, “Optimal Wake-up Mechanism for Single Base Station with Sleep Mode”, 25th International Teletraffic Congress (ITC25), Shanghai, China, Sep. 2013. (Best Paper Award)
Energy‐Delay Tradeoff is getting much more complicated
86
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 44
Ex:Transmit Power vs Transmit Delay
AWGN Channel Upper limit of transmit rate
Delay per bit
Energy per bit
20
log (1 )P
R WWN
1/bt R1
0 (2 1)bt Wb bE WN t
Power can be traded off by delay!
87
Ex:Total Power vs Transmission Delay
Transmission delay
Cir
cuit
Po
wer
RF
Po
wer
Tota
l P
ow
er RF Power Dominant
Circuit Power Dominant
Channel(transmission delay)
TxGreedyTraffic Rx Throughput
RF
Circuit
RF
Circuit
Transmission delay Transmission delay
)log(N
PWC r 1
G. Miao, G. Y. Li, “Cross-Layer Energy-Efficient Wireless Communications: A survey,” Wireless Com & Mobile Comp, 2009
Energy and Delay is not always a tradeoff!But, traffic dynamics was not considered (no queueing delay, etc)
88
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 45
Ex: Impact of Queueing Delay on EDT
Server(capacity)
Poisson (λ)
M/M/1(k) – non realtime service system
Exp (μ)
Buffer (k)
PB(2)=ρ2
k+1(1-ρ2)/(1-ρ2k+2)
where ρ2=λ/μ
For k=2 and ρ2=0.5, then s=7, while W2=1.7μ2-1
For k=8 and ρ2=0.5, then s=511, while W2=1.9μ2-1
PB(1)=PB
(2)
s=(1-ρ2k+1)/[(1-ρ2)ρ2
k ]
Server(capacity)
M/M/1(0) – realtime service system
PB(1)=ρ1/(1+ρ1)
where ρ1=λ/sμ
Poisson (λ)
Exp (sμ)
Considering , we know a small delay can trade for a great amount of energy savings
)log(N
PWC r 1
89
Characterizing EDT need to combine IT and QT
• Information Theory (IT) focuses mainly on link capacity (i.e., capability) over noisy channels, but not traffic dynamics
• Queueing Theory (QT) focuses mainly on traffic dynamics and systemperformance, but not channel unreliability
μRandom
arrival (λ)Random departure
(=λ bits/s)
noiseGreedy Source
Random departure
(≤ C bits/s)
noise
An Unconsummated Union
A. Ephremides, B. Hajek, “Information Theory and Communication Networks: an Unconsummated Union”, IEEE Trans. IT, Oct. 1998
Effective Capacity(Rate‐Accuracy Tradeoff)
Efficient Bandwidth(QoS‐Load/Randomness Tradeoff)
90
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 46
Optimal Condition of a Queueing System
• For a given traffic load, what’s the best service rate? – Higher service rate (capacity) lower queueing delay Wq (higher throughput)
– Lower service rate (capacity) higher server utilization ρ (less energy waste)
• Considering a queue as a blackbox with equivalent service rate μ’=1/(Wq+1/μ), maximize ρμ’ – effective power– For M/G/1, ρμ’=2 μqρ(1- ρ)/[ρμ2b2+2(1- ρ)], which leads to
ρopt=1/[1+sqrt(μ2b2/2)]
Wqopt=sqrt(b2/2)
W opt=sqrt(b2/2)+1/μ
Lqopt= ρopt sqrt(μ2b2/2)
L opt= ρopt [sqrt(μ2b2/2)+1]=1 (!)
Y. Yoshioka, “on the Optimization of Queueing Systems”, IEICE Trans. Commun., Aug. 1977. (in Japanese)
M/Ek/1
How does sleep operation change this EDT? (When and how long should the BS sleep?)
91
Dynamic Control of a Queue
• Considers a M[x]/M/2 queue. The faster server is always on, but the slower server is only used when the queue length exceeds a certain level and switched off when it completes a service.
• Activating the slower server involves fixed set-up costs. Also there are linear operating costs and linear holding costs.
• Conclusion: the two-level hysteretic switching rule that turns the slower server on when the number of jobs in the system exceeds some pre-specified upper level and turns the slower server off when upon service completion by the slower server the number of jobs left behind is below some pre-specified lower level.
Rein D. Nobel and Henk C. Tijms, “Optimal Control of a Queueing System with Heterogeneous Servers and Setup Costs”, IEEE Trans. Automatic Control, April 2000
f
sL2
L1
92
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 47
How does sleep operation change the tradeoff?
• For a typical server under dynamic load, where entering and leaving sleeping mode incurs an energy and a response time penalty, the optimal sleeping policy has a simple hystereticstructure– When asleep, the server stay in OFF mode until the queue builds up to
the point where the ON threshold is met
– After waking up, the server stays awake until all jobs in the queue are processed and then is turned OFF
• Compared with a baseline policy that never puts the server to sleep, it was shown that – low utilization can result in almost 87% energy saved
– high utilization results in only 7.4% energy saved
Ioannis Kamitsos, Lachlan Andrew, Hongseok Kim, Mung Chiang, “Optimal Sleep Patterns for Serving Delay-Tolerant Jobs”, Proceedings of the 1st International Conference on Energy-Efficient Computing and Networking, pp.31-40, 2010
93
Other Related Work
Delay
Transmitting Delay
Processing Delay
QueueingDelay
Energy
Transmit Power
Circuit Power
Basic Power
[Anatharam96][Morabito07]
[Miao09]
[Anatharam96] V. Anantharam, S. Verdu, “Bits Through Queues”, IEEE Trans. on Info. Theory, 1996
[Morabito07] G. Morabito, “Increasing capacity through the use of the timing channel in power‐constrained satellite network”, INFOCOM, 2007
[Miao09] G. Miao, G. Y. Li, “Cross‐Layer Energy‐Efficient Wireless Communications: A survey,” Wireless Com & Mobile Comp, 2009
[Berry02] R. A. Berry, R. G. Gallager, “Communication over Fading Channels with Delay Constraints”, IEEE Trans. IT, May. 2002
[Neely09] M.J. Neely, “ Intelligent Packet Dropping for Optimal Energy‐Delay Tradeoffs in Wireless Downlinks”, IEEE Trans. Automatic Control, March 2009
[Chen11] Y. Chen , G. Y. Li, et al, “Fundamental trade‐offs on green wireless networks”, IEEE Comm. Mag., June 2011
Single-Link[Berry02] [Neely09]
94
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 48
M/G/1 Vacation queue with setup and close-down time
• Rationale of single-server queue– Consider whole BS as a single server,
which is either serving customers (busy) or idling (close-down, idle, setup)
–
95
Some Preliminary Studies
1. Z. NIU and Y. TAKAHASHI, “A Finite-Capacity Queue with Exhaustive vacation/close-down/setup times and Markovian Arrival Processes”, QUESTA, vol.31, no.1, pp.1-23, 1999.
2. Z. Niu, T. Shu, Y. Takahashi, “A Vacation Queue with Setup and Close-down Times and Batch Markovian Arrival Processes”, Performance Evaluation, vol. 54, pp.225-248, 2003.
3. F. Zhu, Y. Wu and Z. Niu, “Delay analysis for sleep-based power saving mechanisms with downlink and uplink traffic,” IEEE Com Lett, 2009.
4. F. Zhu, Z. Niu, “Queueing Delay and Energy Efficiency Analyses of Sleep Based Power Saving Mechanism”, IEICE Tran Com,2010.4
Z. Niu, Jianan Zhang, Xueying Guo, Sheng Zhou, “On the Energy-Delay Tradeoff in Base State Sleep Mode Operation”, The 13th International Conference on Communication Systems, Singapore, 21-23 Nov., 2012 (invited)
Xueying Guo, Sheng Zhou, P. R. Kumar, Zhisheng Niu, “Optimal Wake-up Mechanism for Single Base Station with Sleep Mode”, 25th International Teletraffic Congress (ITC25), Shanghai, China, Sep. 2013. (Best Student Paper Award)
96
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 49
EDT in terms of Close-down Time
• M/G/1 N-policy vacation queue with setup and close-down time– Impact of close-down time
Mean Sojourn Time
Average Power
Sleep mode brings more benefits on energy saving when traffic load is light
PST = PCD = 0.9PON, PSL = 0.2PON
97
EDT in terms of Close-down Time and N
Power –Delay Tradeoff
Without sleep operation, EDT always exists
With sleep operation, energy can be further
reduced by
with same delay
It’s Linear!
98
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 50
EDT in terms of N and Setup Time
• Impact of N (for simplicity, assume close-down time equals to zero)
99
Optimization of N
Monotone if setup time is fixed and therefore N=1 is optimal
Convex if setup time is burstyand therefore there is an optimal N>1
100
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 51
EDT in terms of N and Setup Time
Deviation of setup time does not affect on mean power, while only a little on mean time while λE[S] is high
If traffic load is low, accumulating N customers take a long time, and therefore N should be small
If traffic load is high, accumulating N customers does not take a long time, and therefore N could be larger
101
EDT for Mean Delay vs Delay Bound
( ) 1 ( ) (1 ( )) [ ] (1 )(1 ( ))( ) {
[ ] 1 [ ][ ( )]
( ) [ / ( )] [ ( )] 1 ( )( ( )) ( ) ( ) }.
/ ( ) ( ) ( )
N N N
B s D D E B B sT s
E C E B s B s
N S s s B s S s B sD D
N s B s s B s
Mean delay and delay bound are almost linear, therefore similar tradeoffs hold for energy and delay bound
Large deviations in service times lead to significantly large delay bound
102
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 52
Traffic-aware Dynamic Sleeping Control and
Power Adaptation in GREEN Communication
J. Wu, Z. Niu, S. Zhou, "Traffic-Aware Base Station Sleeping Control and Power Matching for Energy-Delay Tradeoffs in Green Cellular Networks“, IEEE Trans. Wireless Commun., Vol.12, no.8, Aug. 2013
• System Modeling
– Single Cell with n users, Processor-sharing model[1]
– Traffic dynamics:
Large‐scale: load-aware
Small‐scale: queue-aware
– Optimization
where
– Lambert function: W(z)
Dynamic Power Adaptation without Sleep
[1] S.C. Borst, “User‐level performance of channel‐aware scheduling algorithms in wireless data networks”. Proc. Infocom 2003 Conference.
delay power
104
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 53
• Upper and lower bounds
Dynamic Power Adaptation without Sleep
Upper bound:
Lower bound:
λ=0.5,l=2MB105
• Objective function: weighted sum of energy consumption and mean delay
where represents the cycle time and is the switching energy
• Sleeping mechanism – N-based: go to sleep whenever queue becomes empty and wake up if N
requests accumulated during sleeping period
– V-based: go to sleep whenever queue becomes empty and wake up if V time expires
Dynamic Power Adaptation with Sleep
But, setup and close‐down times and their impact on energy consumption not considered
106
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 54
M/G/1 Processor Sharing Modeling
107
N-Policy: With or Without Sleep
Sleep operation can only help to save energy when traffic load is lower than a specific value
108
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 55
N-Policy: Region for EDT
Energy can only be traded off by delay if traffic load is higher than a specific value in N‐policy sleeping operation
109
N-Policy: Region for Energy Saving Gain
Sleeping control helps to save energy only when 1) Traffic load is light, 2) (P0-Psleep)/Esw is large, or3) N is small
110
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 56
N-Policy: Optimality of x* and Lower Bound of Ptotal
x*en is an increasing function of γ, so transmitting faster
when channels are good indeed saves energy. In addition, fast transmission is beneficial when Po - Psleep is larger
If traffic load is further lower than a specific value (?), EDT doesn’t exist all the time
111
N* is related to the switching cost in a square root form!
112
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 57
N-Policy: Pareto-Optimal (N*, X*)
113
V-Policy: EDT with Different Weight
114
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 58
V-Policy: “PA+Sleep” vs “PA only”
Sleeping Control helps only in lightly‐loaded systems
115
How long should a BS sleep?
‐ Small delay can help to save energy if well designed‐ N* is related to the switching cost in a square root form‐ Sleeping Control should be used with power adaptation
J. Wu, Z. Niu, S. Zhou, "Traffic-Aware Base Station Sleeping Control and Power Matching for Energy-Delay Tradeoffs in Green Cellular Networks“, IEEE TWC, Vol.12, no.8, Aug. 2013
116
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 59
Impact of Bursty (IPP) Arrivals
The higher the burstness, the larger the ES region117
Impact of Bursty (IPP) Arrivals
Total power consumption always decreases as the burstiness increases and optimal Pt may exist. But, the benefit for average delay is available only when N is large.
Smaller Higher burstiness
118
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 60
Impact of Bursty (IPP) Arrivals
ES gain highly depends on the system parameters 119
Impact of Bursty (IPP) Arrivals
Energy-Delay Tradeoff can be optimized by jointly adjusting N and Pt
Lower bound of total power consumption
120
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 61
Summary
• What’s 5G?– 5G should be a paradigm shift of cellular architecture for Green and Smart
• Major approaches towards 5G – Reduce Energy Waste by Adapting to Real-traffic Dynamics (REWARD)– Traffic-Aware Network Planning and Green Operation (TANGO)– Collaborative and Harmonized Open Radio Ubiquitous Systems (CHORUS)
• A novel Hyper Cellular architecture for 5G– Decoupling signaling functions from data services to make cellular more
adaptive and intelligent– Always-on hyper cells for coverage guarantee and on-demand data cells
• Enabling technologies for 5G– Separation of control and data coverage – Resource/network virtualization and network dimensioning– Traffic adaptation technologies, including cell zooming, BS sleeping,
coverage extension, ……– Energy-delay tradeoff can help to shift the peak and therefore save energy 121
Concluding Remark
• from World-Wide-Web to World-Wide-Wireless
• for World-Wide-Watch & World-Wide-Wisdom
but definitely should not World-Wide-Wait
and World-Wide-Waste!
: Smart IT for Low-carbon Environment
122
IEEE ICC2014 Tutorial "Revisit the Cellular" 2014/06/14
Zhisheng Niu @ Tsinghua University 62
For more information, visit http://network.ee.tsinghua.edu.cn/niulab/?category_name=publications