Service Differentiation for Improved Cell Capacity in LTE Networks WoWMoM 2015 – June 2015, Boston - USA Presenter: Mattia Carpin Authors: Mattia Carpin, Andrea Zanella, Jawad Rasool, Kashif Mahmood, Ole Grondalen, Olav Osterbo University of Padova (Italy) – Telenor Research (Norway)
LTE High Level 2 Radio Access Network Core Network MME SGW HSS PGW SGi-LAN S11 S5 IP network eNodeB eNodeB, responsible for resource allocation both in downlink and uplink.
OFDMA allocation
Scheduling problem Opportunistic scheduling, for high cell spectral efficiency Fair scheduling, to provide similar service to all users
Schedulers’ metrics CQI Suitable constant for each user that depends on the average channel conditions Hybrid Opportunistic Fair Achievable bit-rate, computed according to the CQI
Previous work In a previous work we simulated the behaviour of a Fair Throughput Guarantee Scheduler (FTGS) α i is computed s.t. each UE gets in the long term the same throughput guarantee B What is the impact of cell edge users? Keep the same average cell SINR μ, position UEs so that Δ increases Avg. SINR i-th user
Cell edge users’ impact
Class Based Scheduler
CBS assumptions Assumptions: optimization interval Constant population (N users) over the optimization interval Rayleigh fading channel Average user SINR known at the eNodeB Classification according to the average SINR Same G (bit/s) to users belonging to same class We guarantee G i >G min using call-admission control mechanism G b = G min G s > h G min G g = k G s Under those assumptions we solve a system of 4N-1 equations that gives α for each user (if such α exists!) 9
Solving equations
Adaptation mechanism 11 Call admission control
Simulations Circular cell of radius r meters, s.t. γ ( d = r ) = 2 dB N users distributed over the cell area with uniform probability Results obtained comparing CBS against MTS, upper bound on spectral efficiency PFS, simple hybrid scheduling policy FTGS, equal guarantee to all users Different G min = {50, 100, 150} Kbit/s
Results Admission control
Short term analysis But what in the short term? Channel-dependent nature → short term behaviour influenced by the fading process ↔ Doppler spread Single-class of users, target throughput η We introduce the average achieved throughput ϑ over a time window of leght τ seconds Normalized gap
Short term gap PDF Long > ShortLong < Short
Variance of the PDF
K-factor for fitting
Excess probability Suppose we have an application that needs to transmit L bits in T seconds We define the excess delay probability as This is the probability of not fullfilling the request of the application Example: L/WT=0.2bit/s/Hz 18
Excess probability 19
Conclusions and future developments Addressed Issues: Dynamic algorithm for efficient resources allocation in LTE Short term behaviour of the algorithm, PDF and relation with the channel conditions Excess delay probability A look into the future Full implementation of the scheduler in NS3 Impact of the scheduler decision on the E2E delay Pre-compute and store the optimal parameters Dynamic and real time estimation of the SINR … 20
Service Differentiation for Improved Cell Capacity in LTE Networks Presenter: Mattia Carpin University of Padova (Italy) – Telenor Research (Norway) Any questions?