service-aware networks over shared wireless access infrastructure
DESCRIPTION
II International Workshop on Challenges and Trends on Broadband Wireless Mobile Access Networks – Beyond LTE-ATRANSCRIPT
Service-Aware Networks over Shared Wireless Access Infrastructure
Luiz A. DaSilva!Professor of Telecommunications, Trinity College Dublin
Professor, Virginia Tech
II International Workshop on Challenges and Trends on Broadband Wireless Mobile Access Networks – Beyond LTE-A!
Campinas, Brasil, 5 November 2014
Sharing, Sharing, Sharing
• Infrastructure sharing among operators !
• Spectrum sharing (LSA, 3-tier, …) !
• Sharing of crowdsourced access !To improve coverage/capacity, to improve energy efficiency (or both)
Our vision: a future based on sharing and virtualisation of wireless networks
Increased efficiency and lower costs through:
1. Incentives for the deployment of localised (small cell, primarily) infrastructure by medium-sized and small operators.
2. The ability to provide service over infra‐structure that employs heterogeneous technologies, and has different properties and ownership.
3. Improved service in currently under‐served areas.
4. The ability to offer virtual wireless networks with different associated quality of experience, at different price points.
Network planning!
siloed, individual networks
limited consideration for complementary technologies (eg., WiFi hotspots)
quality of service through overprovisioning
Network planning!
private plus shared assets
other technologies are an integral part
service requirements (not coverage + capacity for all) are the main driver
Network planning and expansion
driven by over-the-top service needs
considering sharing
Significant CAPEX and energy efficiency benefits…
… but not all services benefit equally
locations of RAN: private, joint
operators
demand clusters (localized)
traffic demand, by type
each operator wishes to service a fraction of traffic thru private infrastructure
regulatory limitations
build (or update, or disable) private infrastructure
deploy shared infrastructure
Optimization problem!
minimize cost
number of private BSs deployed
proportion of shared BSs used to serve the operator’s traffic
subject to…
all the guaranteed bit rate traffic must be served
a fixed proportion of the operator’s traffic must be served by its private deployment
Algorithms!
We propose two algorithms
selection of where to deploy private infrastructure
selection of where to deploy joint infrastructure
Properties!
Both problems belong to the class of submodular set covering problems (NP-hard)
Both algorithms scale linearly with the scenario they operate on
Results!
Inaproximability: under any traffic demand, this is the best an operator can do unless it solves an NP-hard problem
Approximation ratio: can derive a bound on how far from the optimum is achieved by the algorithm
Data Set!
Call detail records from two operators in Ireland
For each transmitter: position, azimuth, sectorization, coverage area.
For voice call and data session: transmitter where it started/ended, duration, amount of data
Traffic Projection!
Cisco Virtual Network Index
Traffic projected to grow at 61% annual rate between now and 2018
Video (GBR) will account for 69% of the traffic
Identify 33,000 demand points
No point corresponds to an area larger than 6.5 sq km
No point corresponds to a population of more than 500 subscribers
Number of points can be adjusted for higher/lower granularity planning
θ = 0.5
geographic distribution of cost: θ = 0.1 (left) and θ = 0.9 (right)
geographic distribution of rate: θ = 0.1 (left) and θ = 0.9 (right)
Take-away!
By increasing θ, operators must deploy more infrastructure to serve the same amount of GBR traffic
Private BSs experience higher load
For θ = 0.5, half the traffic is served by private BSs, although these account for 25% of all infrastructure used by the operator
The vast majority of private BS deployment occurs in urban areas
Higher θ corresponds to higher costs, and more idle capacity
Most of the effects of sharing are felt by high data rate best effort traffic, especially in urban areas
Planned!
Clean-slate: network planning that accounts for shared infrastructure
Evolution: effects of sharing on upgrading and decommissioning of cellular infrastructure
Regulatory constraints: Herfindahl index
Unplanned!
Crowdsourcing of wireless infrastructure
Dense, largely unplanned deployment by operators
Potential for spectrum sharing, RAN sharing
Sharing and Small Cells!
Open subscriber groups
Small cells as infrastructure contributors to virtualised wireless access (Networks without Borders)
Small cells operating in shared spectrum
Multi-antenna use to realise spectral and energy gains
Indoor Dual stripe
Outdoor Stachus Square, Munich
MU-MIMO and Small Cells!
With MU‐MIMO, multiple UEs are spatially multiplexed on different beams within the same time/frequency resource block
Co-scheduled users must have (semi-)orthogonal precoders
With few subscribers per small cell, difficult to find UEs that can be paired for MU-MIMO
MU-MIMO across Small Cells!
Taking advantage of high cell densification, pair UEs across small cells
Emptied cells can be put into sleep mode
Reassignment!!
A. Selection of considered UEs • UEs must experience sufficiently high SINR • Expected data rate in target cell must exceed rate
in home cell by an offset t !
B. Check for target UEs !C. Selection of UEs to reassign from the set of
reassignable UEs • Maximize the number of deactivated eNBs • Version of set covering problem (NP-hard, but
polynomial-time heuristics exist)
Percentage of eNBs Deactivated
Indoor Outdoor
Power savings
Power consumption modelled according to the EARTH FP7 project
!
!
Power parameters different for micro-, pico-, femto-cells
Increase in Spectral Efficiency
Indoor Outdoor
Take-away!
MU-MIMO-based UE reassignments and centralised control of small cell sleep states can achieve simultaneous increases in spectral efficiency and reductions in energy consumption
It was possible to switch in excess of 25% (indoor) and 35% (outdoor) of small cells to a sleep state whilst still achieving considerable gains in spectral efficiency
Sharing: why stop at spectrum?
Base stations, backhaul, storage, processing, back office…
Virtualisation: why stop at SDNs?
Service-aware virtual wireless networks built on a programmable network substrate and orchestrated in response to a specific service need
Acknowledgements!!
Service-aware planning work by Jacek Kibilda, Francesco Malandrino, and Nick Kaminski
Small cell work by Danny Finn
Work funded by the Science Foundation Ireland
More info: luizdasilva.wordpress.com