Global versus Local Call Admission Global versus Local Call Admission Control in CDMA Cellular NetworksControl in CDMA Cellular Networks

Robert Akl, D.Sc.Robert Akl, D.Sc.

Asad ParvezAsad Parvez

University of North TexasUniversity of North Texas

OutlineOutline

• Interference model impact on capacityInterference model impact on capacity• Global call admission controlGlobal call admission control• Local call admission controlLocal call admission control• Global vs localGlobal vs local• ConclusionsConclusions

Code Division Multiple Access Code Division Multiple Access (CDMA) Overview(CDMA) Overview

• Multiple access schemesMultiple access schemes

Call 1

Call 3

Call 2

Call 4

Freq

uenc

y

Time

FDMAFr

eque

ncy

Time

Call 1 Call 2 Call 3

Call 4 Call 5 Call 6

Call 7 Call 8 Call 9

Call 10Call 11Call 12

TDMA CDMA

Frequency

Time

Code

Call 1Call 2

Call 3Call 4

Relative Average Inter-cell Relative Average Inter-cell Interference ModelInterference Model

dA

Cell j

Cell i

jr

ir

yxdAA

n

yxr

yxrji

j

j

Cji

mi

jmjI ,

/,

10,E

2

10

),( ),(

),(2

yxdAyxr

yxrI Cj m

i

mj

A

neji

j

js

jiI Relative average interference at cell i caused by nj users in cell j

A

B

Back

1111 1212 1313 …… …… 1M1M

2121 2222

3131 3232

…… ……

…… ……

M1M1 M2M2 MMMM

ijF ,

Interference Matrix

jn

MjinIijF

j

jji

cellin users ofnumber theis and

,,...,1,for /],[ where

Hence, the total relative average inter-cell interference experienced by cell i is

ijFnIM

j

ji ,1

1111 1212 1313 …… …… 1M1M

2121 2222

3131 3232

…… ……

…… ……

M1M1 M2M2 MMMM

12I

C

]2,1[12 FI

Relative Actual Inter-cell Relative Actual Inter-cell Interference ModelInterference Model

• Interference matrix F cannot be calculated in advance• Instead, a new interference matrix U is computed as follows• For a user k in cell j, the relative actual interference offered by this user to cell i is

m

i

jsk r

rejiU

2

jiUikji

jn

k

M

j

I

for ,11

• Hence, the total relative actual inter-cell interference at cell i caused by every user in the network is

D

E

k users in cell j

Actual Interference Matrix Actual Interference Matrix UU

•Example: for a new call in cell 2, compute row matrix U[2,i] for i = 1,…,M using equation D

]2M ...... 23 22 21[2 iU

• Update 2nd row of interference matrix U by adding the above row matrix to it.

1111 1212 1313 …… …… 1M1M

2121 2222

3131 3232

…… ……

…… ……

M1M1 M2M2 MMMM

iUijU 2],[

CapacityCapacity

• The capacity of a CDMA network is determined by maintaining a lower bound on the bit energy to interference density ratio, given by

Mi

NWInRE

E

I

E

iib

b

i

b

,...,1for

/1 00

W = Spread signal bandwidth

R = bits/sec (information rate)

α = voice activity factor

ni = users in cell i

N0 = background noise spectral

density

F

,...,1for

1/

11/

0

Mi

cNE

RWIn eff

bii

• Let τ be that threshold above which the bit error rate must be maintained, then by rewriting Eq. F

G

Back

Global Call Admission Control (CAC)Global Call Admission Control (CAC)

• A CAC algorithm decides whether or not a A CAC algorithm decides whether or not a network shall accept a call.network shall accept a call.

• Designing a CAC algorithm for CDMA is harder Designing a CAC algorithm for CDMA is harder than designing for TDMA or FDMA.than designing for TDMA or FDMA. Self interference.Self interference. Affects the entire network.Affects the entire network.

• A A globalglobal CAC algorithm takes the entire network CAC algorithm takes the entire network in account for every call making decision.in account for every call making decision.

Mobility ModelMobility Model

• Call arrival process is a Poisson process with rate: Call arrival process is a Poisson process with rate: λλ• Call dwell time is a random variable with exponential Call dwell time is a random variable with exponential

distribution having mean: distribution having mean: 1/μ1/μ• Probability that a call in cell Probability that a call in cell ii goes to cell goes to cell jj after after

completing its dwell time: completing its dwell time: qqijij• Probability that a call in progress in cell Probability that a call in progress in cell ii remains in cell remains in cell ii

after completing its dwell time: after completing its dwell time: qqiiii• Probability that a call will leave the network after Probability that a call will leave the network after

completing its dwell time: completing its dwell time: qqii

Mobility Model – Handoff CallsMobility Model – Handoff Calls

• Handoff calls (Handoff calls (vvjiji): calls that have moved from cell ): calls that have moved from cell

jj to an adjacent cell to an adjacent cell ii..

jAx

xjjijjijjji qBqB 11

jjijji qB 1

• Bj : Call blocking probability for cell j• Aj : Set of cells adjacent to cell i• ρj : Total offered traffic to cell j

j

jAx

jjxjjj

New arriving calls

Handoff calls

i

Global CAC AlgorithmGlobal CAC Algorithm

• A new call is accepted if the following set of equations A new call is accepted if the following set of equations still hold upon acceptance.still hold upon acceptance.

Mi

cInC effiii

,...,1for

,

• Actual Interference case:Actual Interference case:

MiijUtntCM

jii ,...,1for ],[)()(

1

• Average Interference case:Average Interference case:

MiijFntntCM

jjii ,...,1for ],[)()(

1

Simulator – Call Arrival and Admission Module Simulator – Call Arrival and Admission Module (Global CAC)(Global CAC)

For Cells(i) = 1 to M

No

Yes

For Cell(i) = 1 to M Calculate )(ti

)()()( ttt iii

for Cells(i)=1 to M Is ffi CeC )()()( tAttR iii

Calculate

1)()(

1)()(

tntn

tAtA

ii

ii

i

For new call arriving in cell i Calculate Calculate

UiC

From call removal module

To call removal module

Simulator – Call Removal Module (Global CAC)Simulator – Call Removal Module (Global CAC)

if

Yes

1i

No

Yes

No

Is callstaying innetwork

Is call moving toadjacent cell

Yes

No

For Cells(i) = 1 to M

1 ii

1)()( tntn ii

Update

Calculate new i

1)()(

1)()(

tntn

tt

ii

ijij

U

Calculate new j

j

To call arrival and admission module

From call arrival and admission module

Performance MeasurementsPerformance Measurements

• Network throughput:Network throughput: Number of calls per unit time that are Number of calls per unit time that are admitted and stay in the network till termination.admitted and stay in the network till termination.

• Blocking probability:Blocking probability: For a cell, it is the ratio of rejected For a cell, it is the ratio of rejected calls to total offered traffic to that cell.calls to total offered traffic to that cell.

T

t

M

iiiT ttA

TH

1 1

)()(1

T

t i

iiT t

tR

TB

1 )(

)(1)(

Local Call Admission ControlLocal Call Admission Control

• A A local local CAC algorithm considers only a single CAC algorithm considers only a single cell for making a call admittance decision even cell for making a call admittance decision even though its design may look at the network as a though its design may look at the network as a whole.whole.

• A simple approach: Find N, the maximum number A simple approach: Find N, the maximum number of users that are allowed in a cell, which is the of users that are allowed in a cell, which is the same for all the cells in the network.same for all the cells in the network.• Disadvantage: InefficientDisadvantage: Inefficient

Traditional CAC AlgorithmTraditional CAC Algorithm

Mi

cijNFN

H

M

jeff

N

,...,1for

,],[ subject to

, max

1

)(

• A traditional CAC algorithm is formulated that calculates N, the maximum number of calls allowed in each cell. The optimization problem is given by

• Define network throughputDefine network throughput

M

iiiiii BBH

1

1

Our Optimized Local CAC Our Optimized Local CAC AlgorithmAlgorithm

Mi

cijFNN

H

M

jeffji

NN M

,...,1for

,],[ subject to

, max

1

),...,(

• Solve a constrained optimization problem that maximizes Solve a constrained optimization problem that maximizes the network throughput with signal-to-interference ratio the network throughput with signal-to-interference ratio constraints as lower bounds.constraints as lower bounds.

Simulator – Call Arrival and Admission Module Simulator – Call Arrival and Admission Module (Local CAC)(Local CAC)

For Cells(i) = 1 to M

For Cell(i) = 1 to M Calculate )(ti

)()()( ttt iii

)()( tnNitA ii

)()()( ttntn iii No

Yes

)()()(

)()(

tAttR

tNtn

iii

ii

)()( tAt ii if

From call removal module

To call removal module

Simulator – Call Removal Module (Local CAC)Simulator – Call Removal Module (Local CAC)

if

Yes

1i

No

Yes

No

Is callstaying innetwork

Is call moving toadjacent cell

Yes

No

For Cells(i) = 1 to M

1 ii

1)()( tntn iiCalculate new i

1)()(

1)()(

tntn

tt

ii

ijij

Calculate new j

j

To call arrival and admission module

From call arrival and admission module

Global CAC vs Local CACGlobal CAC vs Local CAC

GlobalGlobal• Call admission based on Call admission based on

all the calls present in the all the calls present in the network.network.

• Slower.Slower.• Inherently optimized.Inherently optimized.• Adaptable.Adaptable.• Complexity: O(M).Complexity: O(M).

LocalLocal• Call admission based on Call admission based on

calls present in the cell calls present in the cell under consideration only.under consideration only.

• FasterFaster• Optimized only for a Optimized only for a

given traffic distribution given traffic distribution profile.profile.

• Cannot compensate for Cannot compensate for big fluctuation in traffic.big fluctuation in traffic.

• Complexity: O(1)Complexity: O(1)

SimulationsSimulations

• Network configurationNetwork configuration• COST-231 propagation modelCOST-231 propagation model

• Carrier frequency = 1800 MHzCarrier frequency = 1800 MHz

• Average base station height = 30 metersAverage base station height = 30 meters

• Average mobile height = 1.5 metersAverage mobile height = 1.5 meters

• Path loss coefficient, m = 4Path loss coefficient, m = 4

• Shadow fading standard deviation, Shadow fading standard deviation, σσss = 6 dB = 6 dB

• Processing gain, W/R = 21.1 dBProcessing gain, W/R = 21.1 dB

• Bit energy to interference ratio threshold, τ = 9.2 dBBit energy to interference ratio threshold, τ = 9.2 dB

• Interference to background noise ratio, IInterference to background noise ratio, I00/N/N00 = 10 dB = 10 dB

• Voice activity factor, α = 0.375Voice activity factor, α = 0.375

Simulations – Network ParametersSimulations – Network Parameters

• Non-uniform traffic distribution• Group A (cells 5, 13, 14, 23) : 14 calls/time• Group B (cells 2, 8, 9, 19) : 14 calls/time• Rest of the cells : 3 calls/time

• Ceff = 38.25• No mobility probabilities

• qij = 0• qii = 0.3• qi = 0.7

AAii qijqij qiiqii qiqi

33 0.0200.020 0.2400.240 0.7000.700

44 0.0150.015 0.2400.240 0.7000.700

55 0.0120.012 0.2400.240 0.7000.700

66 0.0100.010 0.2400.240 0.7000.700

AAii qijqij qiiqii qiqi

33 0.1000.100 0.0000.000 0.7000.700

44 0.0750.075 0.0000.000 0.7000.700

55 0.0600.060 0.0000.000 0.7000.700

66 0.0500.050 0.0000.000 0.7000.700

Low mobility probabilities High mobility probabilities

Maximum users admitted per cell for average and actual interference for the three mobility cases.

Network throughput for average and Network throughput for average and actual interference for the three actual interference for the three mobility cases.mobility cases.

Blocking probability for Blocking probability for average and actual interference average and actual interference for the three mobility cases.for the three mobility cases.

Results Global CACResults Global CAC

• Network throughput is always a little higher for average Network throughput is always a little higher for average interference in all the three mobility cases.interference in all the three mobility cases.

• Blocking probabilities are a little higher for actual interference Blocking probabilities are a little higher for actual interference for all three mobility cases. for all three mobility cases.

• Blocking probability is around 10% in all the three mobility Blocking probability is around 10% in all the three mobility cases for the cells with high demand.cases for the cells with high demand.

• Throughput is highest and blocking probability is lowest for the Throughput is highest and blocking probability is lowest for the high mobility case.high mobility case.

Erlang traffic and maximum number of Erlang traffic and maximum number of calls allowed to be admitted per cell for calls allowed to be admitted per cell for all three mobility cases.all three mobility cases.

High mobility has an equalizing effect on non-uniform traffic distribution.

Network throughput for our Network throughput for our optimized local CAC for all optimized local CAC for all three mobility cases.three mobility cases.

Theoretical and simulated network Theoretical and simulated network throughput for our optimized local throughput for our optimized local CAC and traditional CAC for all three CAC and traditional CAC for all three mobility cases.mobility cases.

Theoretical and simulated Theoretical and simulated blocking probability for our blocking probability for our optimized local CAC and optimized local CAC and traditional CAC for all three traditional CAC for all three mobility cases.mobility cases.

Results Local CACResults Local CAC

• Our optimized local CAC algorithm adapts in Our optimized local CAC algorithm adapts in response to the traffic demand due to users’ response to the traffic demand due to users’ mobility.mobility.

• Our local CAC network throughput is higher than Our local CAC network throughput is higher than traditional CAC throughput by nearly 13%.traditional CAC throughput by nearly 13%.

• Our local CAC algorithm strikes a good balance Our local CAC algorithm strikes a good balance between the blocking probabilities of the low and between the blocking probabilities of the low and high traffic cells.high traffic cells.

Network throughput for our Network throughput for our optimized local and global optimized local and global CAC algorithms.CAC algorithms.

Blocking probability for our Blocking probability for our optimized local and global optimized local and global CAC algorithms.CAC algorithms.

SummarySummary

• High mobility results in highest throughput High mobility results in highest throughput because it equalizes non-uniform traffic.because it equalizes non-uniform traffic.

• Our optimized local CAC algorithm performance Our optimized local CAC algorithm performance is better than traditional CAC algorithm.is better than traditional CAC algorithm.

• Our optimized local CAC algorithm performance Our optimized local CAC algorithm performance is just as good as a global for a given traffic is just as good as a global for a given traffic distribution.distribution.