Post on 24-Jul-2018
21. VDE/ITG Fachtagung Mobilkommunikation11.-12. Mai 2016, Osnabrueck
Camilo Solano, Ericsson Radio Research Germany
End-user Benefits of LTE Dual Connectivity in Heterogeneous Networks
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 2/17 |
› New LTE feature Dual Connectivity (DC) standardized in 3GPP Rel-12 (Nov 2014)
• Allowing resource aggregation from two eNBs to increase the user throughput
› Core question: How much does the end-user gain from LTE networks with the DC feature?
• Typical internet traffic type of today: web page downloads
› Focus of this research work:
• Analysis and optimization of algorithms for traffic offloading with DC
• DC performance evaluation in a realistic heterogeneous network deployment
Introduction
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 3/17 |
› Dual Connectivity Background
› Traffic offloading with DC
› Performance evaluation
› Conclusions
Outline
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 5/17 |
› Rel-8 LTE Baseline (Single Connectivity)
Internet
Scheduler
eNB
› Rel-10/11 LTE Carrier Aggregation (CA)
Internet
Scheduler
eNB
LTE Architecture Options
Internet
Scheduler
eNB
ideal backhaul
Internet
Splitter
eNB 1
non-ideal backhaul (>10ms)
› Rel-12 LTE Dual Connectivity (Split Bearer)› Rel-11 LTE Coordinated Muti Point operation (CoMP)
Internet
Scheduler
eNBhandover
eNB 2
Scheduler
Scheduler
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 6/17 |
Internet
› UE aggregates radio resources from two eNBs
• Different carrier frequencies
• Downlink user traffic
• Aggregation point at Packet Data Convergence Protocol (PDCP)
• Any backhaul channel between eNBs
Rel-12 Dual Connectivity
UE
Secondary eNB
(SeNB)Master eNB
(MeNB)
non-ideal backhaul (> 10ms)
f1f2
Scheduler
Scheduler
PDCP(Splitter)
Web Server
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 7/17 |
› Data buffering depends on traffic type
Buffering at the eNB
Full buffer traffic
(e.g. FTP large file size
transmissions)
Bursty traffic
(e.g. webpages
downloads)
Bu
ffer
Level
Bu
ffer
Level
time
time
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 9/17 |
› Transmit packets via the MeNB or via
the SeNB?
PDCP Buffer Splitting
• Answer: via the fastest path to the
UE
Transmit packet via MeNB if
otherwise via SeNB.
)(
)()(
DLS
DL
Q
SDLS
Tt
TtqTtD
)(
)()(
t
tqtD
M
Q
MM
,, DLDLMSM T) T(tD (t) D
throughput MeNB Current
level buffer MeNB Current
delay buffering MeNB Current
:)(
:)(
:)(
t
tq
tD
M
Q
M
M
DM(t) DS(t+TDL)
(TDL)
delay downlink Backhaul
at throughput SeNB
at level buffer SeNB
atdelay buffering SeNB
:
:)(
:)(
:)(
DL
DLDLS
DLDL
Q
S
DLDLS
T
TtTt
TtTtq
TtTtD
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 10/17 |
› Estimation of the SeNB buffer level at the MeNB based on a fluid approximation and a recursive
feedback loop
Delay Compensated Buffer Splitting
t
tSDLMDL
Q
S
Q
S dtTrtqtq0
))()(()()( ,0
MeNB Buffer Splitter
SeNB Buffer
Packets to be transmitted to the UE
)()(, tTtr SDLMDL
)(, tr MDL
)(, DLMDL Ttr
Feedback
EstimationULULS
t
TtDLMDLUL
Q
S
Q
MSTTtdTrTtqtq
UL
*)()()()( ,,
)(, DL
Q
MSTtq
)(tS
delay uplink Backhaul :ULT
?)( DL
Q
S Ttq
Feedback(t-TUL)
)(tqQ
S
MeNB the from outgoingcurrently rate, data Downlink :)(, tr MDL
delay downlink Backhaul
throughput SeNB Current
level buffer SeNB Initial
level buffer SeNB Current
:
:)(
:)(
:)(
0
DL
S
Q
S
Q
S
T
t
tq
tq
DLS
Q
S Tttq ),(),(
MeNB the at level buffer SeNB Estimated :,
Q
MSq
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 12/17 |
› Asian city inspired by Tokyo/Seoul
• Population: 10000 users/km2
• Indoor user probability: 90%
› Heterogeneous network deployment
• Two 10 MHz carriers
• Macro eNB @900MHz
• Pico eNB @2GHz
› Propagation model well tuned from real
measurements
› LTE protocol stack explicitly modeled
• DC PDCP buffer splitting
LTE System Simulator
Macro eNB
Pico eNB
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 13/17 |
Performance Comparison
UE i
f1 f2
UE j
eNB i
eNB j
› LTE architectures: LTE Baseline and LTE Dual Connectivity
UE ij
f1 f2
UE ji
eNB i
(MeNB/SeNB)
eNB j
(MeNB/SeNB)
Rel-12 LTE Dual Connectivity (DC)
• Backhaul delays: 0ms (ideal), 10ms, 30ms
Rel-8 LTE Baseline: Single Connectivity
non-ideal backhaul
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 14/17 |
Performance Evaluation› Full buffer traffic High buffering High potential benefits of DC
› Bursty traffic less buffering lower DC benefits
Top25 web sites (bursty traffic)
Mean
user
perc
eiv
ed
th
rou
gh
pu
t[M
bit
/s]
Mean
user
perc
eiv
ed
th
rou
gh
pu
t [
Mb
it/s
]
FTP 16MB file size (full buffer traffic)
~0% gain
60% gain
45% gain
< 8% gain
< 12% gain
< 8% gain
Low load Medium load High load
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 15/17 |
Cell Edge Users (loading YouTube main page)
Performance Evaluation› The cell edge user experience is significantly improved with DC for bursty traffic
› Low Loads low buffering low DC benefits
› Medium Loads more buffering gains by aggregating resources
› High Loads lower throughput, more buffering No gains with DC, all resources being used in both eNBs
Me
an
use
r p
erc
eiv
ed
th
rou
gh
pu
t[M
bit
/s]
Gain
[%
]
Cell Edge Users (loading YouTube main page)
Commercial in confidence | © Ericsson | 2016-05-11 | Slide 16/17 |
› The end-user performance improves with LTE DC by offloading traffic from loaded eNBs
buffers.
› In full buffer traffic conditions (large file transfers), DC provides high gains for the end-
users.
› In bursty traffic, especially for the worst users in the system, DC enables to significantly
reduce the webpage download times.
› DC can be considered as a valuable feature to achieve the system capacity that will be
demanded within the next years (2021 forecast).
CONCLUSIONS