Self-Optimising Call Admission Control for LTE … call... · FP7 ICT-SOCRATES Self-Optimising Call...

22
FP7 ICT-SOCRATES Self-Optimising Call Admission Control for LTE Downlink SOCRATES-NGMN call February 24, 2010 Work performed within the context of the SOCRATES project ~ www.fp7-socrates.eu K. Spaey, B. Sas, C. Blondia IBBT / University of Antwerp

Transcript of Self-Optimising Call Admission Control for LTE … call... · FP7 ICT-SOCRATES Self-Optimising Call...

FP7 ICT-SOCRATES

Self-Optimising Call Admission Control for LTE

Downlink

SOCRATES-NGMN call February 24, 2010

Work performed within the context of the SOCRATES project ~ www.fp7-socrates.eu

K. Spaey, B. Sas, C. Blondia IBBT / University of Antwerp

WWW.FP7-SOCRATES.EU

  (Self-optimising) call admission control

 Reference admission control algorithm

 Metrics

 Self-optimising algorithm for ThHO

 Evaluation methodology

 Simulation results

 Conclusions and future work

Outline

2

Kathleen Spaey, IBBT / University of Antwerp

More info available in TD(10)10056, joint COST2100 / SOCRATES workshop (February 5, 2010)

WWW.FP7-SOCRATES.EU

 Call admission control (AC) algorithm – Decides if a call request will be admitted or rejected – Bases its decisions on:

•  Enough resources available to guarantee QoS new call? •  If call is accepted, QoS of already accepted calls will be

maintained?

 Self-optimising call admission control – Self-optimise / auto-tune parameters of the AC algorithm –  In response to observed changes

(Self-optimising) call admission control

3

Kathleen Spaey, IBBT / University of Antwerp

WWW.FP7-SOCRATES.EU

 Development of a simple self-optimising AC algorithm  Evaluation under sudden overload

 Reference AC algorithm (static algorithm) needed – AC algorithms are vendor specific – Literature:

•  Prioritisation of acceptance of handover over fresh calls •  Recognise diverse QoS requirements for delay-sensitive (RT) and

delay-tolerant (NRT) applications

(Self-optimising) call admission control

4

Kathleen Spaey, IBBT / University of Antwerp

WWW.FP7-SOCRATES.EU

 Typical AC rule: admit call if

Reference admission control algorithm

5

Kathleen Spaey, IBBT / University of Antwerp

required capacity already accepted

calls

required capacity new call

cell capacity

Distinct margins for •  fresh / handover calls •  RT / NRT calls

Cell capacity depends on → packet scheduler decisions → channel conditions users → location users → varies over time → estimate of time-varying cell

capacity needed *

c* (t) + creq ≤ margin × C(k)

* Based on “Adaptive connection admission control scheme for high data rate mobile networks”, S.S. Jeong, J.A. Han, W.S. Jeon, VTC Fall 2005

WWW.FP7-SOCRATES.EU

Reference admission control algorithm

6

Kathleen Spaey, IBBT / University of Antwerp

  t: time of call arrival   C(k): most recent estimate of

cell capacity   creq: required capacity of arriving call   c*(t): required capacity of already

accepted active calls   c*

RT(t): required capacity of already accepted active RT calls

fresh calls are blocked

priority is given to HO calls

avoid that cell capacity is entirely filled with RT calls

WWW.FP7-SOCRATES.EU

 GoS measurements

– HO failure ratio = (# HO calls rejected by the AC algorithm) / (# generated HO calls)

– Call blocking ratio = (# fresh calls rejected by the AC algorithm) / (# generated fresh calls)

 QoS measurements (only for admitted traffic)

– Traffic loss ratio = (# lost traffic) / (# generated traffic) •  measured for RT traffic (voice / video)

– Call throughput = (# bits of call) / (call transfer time) •  measured for NRT traffic (web) •  we focus on the fraction of web calls with a call throughput smaller

than the minimum call throughput requested to the packet scheduler

Metrics to assess performance

7

Kathleen Spaey, IBBT / University of Antwerp

WWW.FP7-SOCRATES.EU

 Simulations in which the 4 performance metrics are obtained –  for various ThHO values –  in scenarios with varying call arrival rate or varying %HO calls

 Changes in the measured performance might require opposite adaptations of ThHO, depending on which performance measure is considered

– QoS degradation of ongoing calls –  Increasing HO failure ratio –  Increasing call blocking ratio → increase of ThHO

 Operator policy to decide on this trade-off

 SON algorithm which auto-tunes ThHO should take the chosen policy into account

Sensitivity analysis on ThHO

8

Kathleen Spaey, IBBT / University of Antwerp

decrease of ThHO

WWW.FP7-SOCRATES.EU

 Desired properties – ThHO should be adapted based on GoS / QoS measurements, rather

than on measurements on the system conditions – Measurements should be smoothed – Follow policy to handle contradictions in required adaptations of ThHO

 Policy considered In order of priority: – Aim to guarantee QoS of the accepted calls – Accept HO calls with priority over fresh calls – Aim to reduce call blocking ratio

Self-optimising algorithm for ThHO

9

Kathleen Spaey, IBBT / University of Antwerp

WWW.FP7-SOCRATES.EU

 At regular time instants t = kΔ, measurements are collected

 Measurements in [ kΔ ; (k+1) Δ [ , smoothed with parameter αSON – QoS_RT(k): trafic loss ratio real-time traffic – QoS_NRT(k): fraction of non-real-time calls with call throughput smaller

than amount requested to scheduler – GoS_HO(k): HO failure ratio – GoS_fresh(k): call blocking ratio

Self-optimising algorithm for ThHO

10

Kathleen Spaey, IBBT / University of Antwerp

WWW.FP7-SOCRATES.EU

Self-optimising algorithm for ThHO

11

Kathleen Spaey, IBBT / University of Antwerp

bad QoS or bad HO failure ratio

good QoS and good HO failure ratio and bad call blocking ratio

decrease ThHO

increase ThHO

WWW.FP7-SOCRATES.EU

 Evaluation of SON algorithm under sudden overload (unpredictable event)

– Scenarios where there is a sudden increase in call arrival rate or/and %HO calls

 Comparison of performance obtained with – Self-optimising AC algorithm (reference algorithm + auto-tuning of ThHO) – Static AC algorithm (reference algorithm, fixed ThHO)

Evaluation methodology

12

Kathleen Spaey, IBBT / University of Antwerp

WWW.FP7-SOCRATES.EU

 Simulator for downlink direction developed using OPNET Modeler

 Call generation: –  fresh / HO calls – VoIP (RT) / video streaming (RT) / web browsing (NRT)

Simulation model

13

Kathleen Spaey, IBBT / University of Antwerp

WWW.FP7-SOCRATES.EU

 Set-up: 0.6 calls/s, 30% HO calls

1 call/s, 60% HO calls

 Parameters self-optimising algorithm – Δ = 1 minute –  τQoS_RT = 1e-5, τQoS_NRT = 2%, τGoS_HO = 1%, τGoS_fresh = 5%

– αSON = 0.75, 0.90

 Static AC algorithm (no-SON): ThHO = 0.3, 0.4, …, 0.9, 1  SON AC algorithm (SON): ThHO is auto-tuned

Simulation results

14

Kathleen Spaey, IBBT / University of Antwerp

after 28 minutes (± 1000 calls)

WWW.FP7-SOCRATES.EU

 QoS: fraction of web calls with call throughput ≤ 250 kbit/s

Simulation results

15

Kathleen Spaey, IBBT / University of Antwerp

SON no SON

Before change: SON performs equally well

After change: SON performs equally well

After change: SON performs better

WWW.FP7-SOCRATES.EU

 QoS: traffic loss ratio

Simulation results

16

Kathleen Spaey, IBBT / University of Antwerp

SON no SON

Before change: SON performs equally well

After change: SON performs equally well

After change: SON performs better

WWW.FP7-SOCRATES.EU

 GoS: handover failure ratio

Simulation results

17

Kathleen Spaey, IBBT / University of Antwerp

SON no SON

Before change: SON performs equally well

After change: SON performs equally well

After change: SON performs better

WWW.FP7-SOCRATES.EU

 GoS: call blocking ratio

Simulation results

18

Kathleen Spaey, IBBT / University of Antwerp

SON no SON

Before change: SON performs equally well

After change: SON performs better

After change: SON performs considerably worse

Before change: SON performs better

defined policy achieved

defined policy achieved

defined policy not achieved

WWW.FP7-SOCRATES.EU

 SON algorithm complies better to the defined policy, both before and after the change, than the static algorithm with fixed ThHO

–  In general: •  “high” ThHO before change •  “low” ThHO after change

→ SON can adapt ThHO according to the state the system is in

Conclusions and future work

19

Kathleen Spaey, IBBT / University of Antwerp

 Future work: integration of multiple SON algorithms - Admission control SON combined with handover SON

•  both algorithms are triggered if handover failure ratio is too high

•  both algorithms aim to reduce the handover failure ratio in their own way

→ they influence each others input measurements

WWW.FP7-SOCRATES.EU

Backup slides

20

Kathleen Spaey, IBBT / University of Antwerp

WWW.FP7-SOCRATES.EU

 Time-varying cell capacity –  varying radio conditions at UEs – allocation of scheduling resources by the scheduler

 Estimation of cell capacity – divide time in intervals, 1 interval = T TTIs – M = T * C scheduling resources in an interval – m(k): # scheduling resources effectively used during interval k

Time-varying cell capacity

21

Kathleen Spaey, IBBT / University of Antwerp

1 subchannel = 180 kHz

time

frequency

C subchannels

1 interval = T TTIs 1 interval 1 TTI = 1 ms

1 scheduling resource

WWW.FP7-SOCRATES.EU

 Estimation of cell capacity * – C(k): estimate of the cell capacity at the end of interval k

– µ(k): estimate of the throughput (bits/s) at the end of interval k

→ dependency between AC and PS when admission decision is based on C(k)

* Based on “Adaptive connection admission control scheme for high data rate mobile networks”, S.S. Jeong, J.A. Han, W. S. Jeon, VTC Fall 2005

Time-varying cell capacity

22

Kathleen Spaey, IBBT / University of Antwerp

C(k) = (1−α)∗C(k −1)+α∗µ(k)

bit rate (bits/s) corresponding to

scheduling resource (t,c) of interval k

correction factor for unused scheduling resources

µ(k) = r(k, t,c)c=1

C

∑t=1

T

∑⎛

⎝ ⎜

⎠ ⎟ ∗

Mm(k)