EU FP7 Project iJOIN - TUM Challenge: computational complexity Main contributor to computational...
Transcript of EU FP7 Project iJOIN - TUM Challenge: computational complexity Main contributor to computational...
http://www.ict-ijoin.eu/ @ict_ijoin
EU FP7 Project iJOIN
Functional Split Options for Centralized RAN with Imperfect
Backhaul
46. Treffen der VDE/ITG-Fachgruppe 5.2.4
Mobilität in IP-basierten Netzen Dec. 4th 2014
Andreas Maeder, Peter Rost (NEC Laboratories Europe)
Contact: {andreas.maeder, peter.rost}@neclab.eu
iJOIN: Interworking and JOINt Design of an Open Access and Backhaul Network Architecture
for Small Cells based on Cloud Networks
http://www.ict-ijoin.eu/ @ict_ijoin (( 2 ))
Key enablers to satisfy data demands
0
200
400
600
800
1000
1200
1400
1600
MoreSpectrum
FrequencyDivision
Modulationand Coding
SpectrumReuse
25 5 5
1600
Centralised Processing
• C-RAN handles inter-cell interference, allows for higher
utilisation and to avoid peak-provisioning
• Up to 50% energy-saving
• 20%-50% OPEX reduction, 15% CAPEX reduction
• Requires high capacity and low delay backhaul
Centralisation is an option to implement the network but
requires more flexibility than today
Small Cells
• 50% Total cost of ownership (TCO) savings
• Four-fold increase in density until 2014
• Worth about 6.1bln USD until 2014
Small-cells are the option to handle higher rates and to
improve energy/cost-efficiency Facto
r o
f im
pro
ve
me
nt
http://www.ict-ijoin.eu/ @ict_ijoin (( 3 ))
• SaaS
• PaaS
• IaaS
• On-demand
• Broad access
• Pooling
• Elasticity
• Measured
• C-RAN
• RAN-Sharing
• SDN
• SDR
How the “Cloud” changes the picture …
SQL ISP
• NFV
• vEPC
• SDN
Core Network HetNet RAN Heterogeneous Backhaul
60GHz
Relay/Microwave
Copper
Optical
Mobile Network
Virtualization / “Cloudification”
The communication realm The IT world
Blurring borders between Communication and Information Technology
http://www.ict-ijoin.eu/ @ict_ijoin (( 4 ))
Key Concepts
Flexible centralisation through RANaaS
(RAN-as-a-Service) Push RAN functionality into cloud-like platforms
Simplified RAN management and flexible small-
cell solutions
Reduce complexity & cost through elastic &
flexible function assignment
Higher energy-efficiency through computational
diversity and higher utilisation
Joint design and optimisation of RAN and
backhaul Interworking of access and backhaul network
Optimise for flexible centralisation
Optimise backhaul for small cells
Consider heterogeneous backhaul network
(fibre and wireless)
Relax backhaul requirements through dynamic
provisioning (“on-demand”)
Cloud Platform Core
Network
eNB
iSC iSC
iSC
iSC
iSC
eNB
RANaaS
Join
t O
ptim
isation
of
RA
N a
nd
Backhaul
iTN
RANaaS
LTE Air I/F
mmWave
Fibre Link
Ve
ry d
en
se s
ma
ll
ce
ll n
etw
ork
He
tero
ge
ne
ou
s
backh
au
l la
ye
r
Fle
xib
le R
AN
fu
nct.
assig
n. to
clo
ud
iSC: iJOIN
Small Cell
http://www.ict-ijoin.eu/ @ict_ijoin (( 5 ))
Key concepts: flexible functional split
RF
PHY
MAC
RRM
Netw.
Mgmt.
Adm./Cong.
Control
“Conventional”
implementation of LTE
RF
PHY
MAC
RRM
Netw.
Mgmt.
Adm./Cong.
Control
C-RAN Implementation
(BB-pooling)
Execute
d a
t B
S
Centra
lly e
xecute
d C
en
trally
exe
cu
ted
Exe
cu
ted
at R
RH
Flexible Functional Split
Example: Partly centralised
(inter-cell) RRM
Example: Joint
Decoding
http://www.ict-ijoin.eu/ @ict_ijoin (( 6 ))
Key concepts: exploit centralization gains
Coordination gains: e.g. ICIC, joint precoding, joint decoding, … Computational diversity
Exploitation of temporal and spatial traffic fluctuations Efficiently use available resources, scale resource according to needs
(resource pooling, elasticity)
RAN
RANaaS
RANaaS Interface RANaaS Hypervisor
http://www.ict-ijoin.eu/ @ict_ijoin (( 7 ))
Key concepts: service and function agility
Localized optimization Optimization based on purpose, deployment, …
Using software implementation rather than configuration (SON)
Fast deployment of tailored software depending on location and use case
RAN
RANaaS
RANaaS Interface RANaaS Hypervisor
http://www.ict-ijoin.eu/ @ict_ijoin (( 8 ))
Challenges
Functional split of RAN protocol stack
Computational resource provisioning
“cloudified” RAN architecture
Utilization efficiency
Cloud platform latency/U-plane capabilities
http://www.ict-ijoin.eu/ @ict_ijoin (( 9 ))
Challenge: Where to split?
- higher requirements
(more bandwidth, lower
latencies, higher comp.
effort)
+ gains from
coordination, diversity,
joint processing
Constraints and requirements:
Backhaul characteristics (latency,
bandwidth)
3GPP requirements (latency,
bandwidth, protocol stack)
Computational resources
Benefits:
Multiplexing gains (computational
resources, user traffic)
Coordination gains (interference
mitigation, cancellation, load
balancing, ...)
..are an operator-defined
optimization target! RF processing
A/D conversion / pre-processing
Lower PHY layer (incl. FFT)
Upper PHY layer (incl. FEC)
MAC layer (incl. MUX, HARQ)
RLC Layer (RB buffers, ARQ)
PDCP Layer (ciphering)
RRC/C-Plane
Sch
edu
ling
Radio Access Point
EPC
http://www.ict-ijoin.eu/ @ict_ijoin (( 10 ))
Functional Split Options
RF processing
A/D conversion / pre-processing
Lower PHY layer (incl. FFT)
Upper PHY layer (incl. FEC)
MAC layer (incl. MUX, HARQ)
RLC Layer (RB buffers, ARQ)
PDCP Layer (ciphering)
RRC/C-Plane
Sch
edu
ling
Radio Access Point
EPC
Split A
Split B
Split D
Split C
Split option
Lowest layer centralized
RTT requirements
Bandwidth requirements
Centralization scheme
Main centralization gains
D PDCP > 50ms U-Plane + C-Plane overhead
Centralized connection control
Load balancing, energy efficiency
C MAC <3 ms (HARQ sched.)
U-Plane + C-Plane overhead
Coordinated resource allocation
Interference mitigation, cooperative schemes
B FEC ~ 1ms (HARQ + FEC processing time)
U-plane + redundancy (~ +30%)
Coordinated resource allocation, central FEC
Computational diversity
A FFT/IFFT 5 µs (CIPRI)
Quantized symbols (several Gbps)
Full PHY centraliz-ation
Centralized precoding, joint decoding
For more information see iJOIN deliverables D2.2, D3.2
http://www.ict-ijoin.eu/ @ict_ijoin (( 11 ))
Bandwidth requirements (DL)
user diversity gains
20MHz BW
2x2 MIMO
http://www.ict-ijoin.eu/ @ict_ijoin (( 12 ))
Latency requirements: HARQ
HA
RQ
RTT
tim
er
sub
fram
e
ind
ex
UE iSC RANaaS
BH latency
Air int.
RANaaS processing
+ frame building
Air int.
BH latency
UL DL
UL data
UL data
Latency breakdown
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
0
1
2
3
4
5
6
7
8
0
1
2
3
4
5
6
7
8
UL DL
9
0
1
2
3
4
5
6
7
1
2
3
4
5
6
7
8
9
UL DL
ACK/NACK
AC
K/N
AC
K e
xpec
ted
in S
F n
+4
ACK/NACK
“Latency bottleneck”, applies
to all splits with centralized MAC
http://www.ict-ijoin.eu/ @ict_ijoin (( 13 ))
Mapping functional splits to backhaul types
Fiber (CWDM/dark)
mmWave
μ-Wave
Sub-6 GHz
PON
xDSL
Metro-optical
Split A
Split B
Split C
Split D
10 1
10 2
10 3
10 4
10 5
10 -4
10 -3
10 -2
10 -1
10 0
Throughput [Mbps]
La
ten
cy (
pe
r h
op
, R
TT
)[s
]
http://www.ict-ijoin.eu/ @ict_ijoin (( 14 ))
Challenge: computational complexity
Main contributor to computational complexity is the turbo decoding (UL)
Strongly non-linear complexity in SNR, depends on modulation and coding scheme
Computational outage: TB cannot be decoded within time budget (<3 ms)
P. Rost, S. Talarico, M. Valenti, The Complexity-Rate Tradeoff of
Centralized Radio Access Networks, submitted to IEEE TWC
http://www.ict-ijoin.eu/ @ict_ijoin (( 15 ))
Computational complexity: cell level
System-level results for per-cell CC Correlation between #UEs and CC?
Impact of scheduling, resource allocation?
3.6 3.65 3.7 3.75 3.8 3.85 3.9 3.95 4 4.05
x 104
0
2
x 104
subframe index
tota
l co
mp
. co
mp
lexity (
bits
it)
3.6 3.65 3.7 3.75 3.8 3.85 3.9 3.95 4 4.05
x 104
0
5
nu
mb
er
of U
Es p
er
ce
ll
Comp. compl.
#UEs
http://www.ict-ijoin.eu/ @ict_ijoin (( 16 ))
Joint RAN/cloud resource provisioning
0 100 200 300 400 500 600116
118
120
122
124
126
128
130
132
134
Frame number
Com
puta
tional dem
and (
CP
Us)
Provided CPUs
Occupied CPUs
Comp.-aware LA
Overall throughput loss: 0.04%
Assumptions: 456 cells/142 base stations (macro+small cell), 1200 users
96 GFlops per CPU
1. Estimate outage capacity based on
per-cell CC outage rate (e.g. 1%)
2. Provide resources (e.g. CPUs)
3. Adjust LA when necessary
http://www.ict-ijoin.eu/ @ict_ijoin (( 17 ))
Conclusions
New paradigms RANaaS: Apply Cloud principles to RAN operation
Flexible functional split: adapt level of centralization to backhaul capabilities
Increase utilization efficiency by joint cloud/RAN resource management
New insights Feasible functional splits of the LTE protocol stack
“Cloud-aware” link adaptation: first step towards RAN operation on commodity
hardware
Many more results available on www.ict-ijoin.eu, e.g. feasible splits, BH
dimensioning, data center placement, CoMP, commodity HW implementation, …
Outlook: RAN cloudification will be integral aspect of 5G
IT companies will continue to push into telco space
Technical challenges remain: RAN-capable cloud platforms
http://www.ict-ijoin.eu/ @ict_ijoin (( 18 ))
Thank you for your attention!