1

Mobility-Based Predictive Call Admission Control and

Bandwidth Reservation in Wireless Cellular Networks

Fei Yu and Victor C.M. Leung

INFOCOM 2001

2

OUTLINE

• Introduction

• Model Description

• Mobility Prediction

• CAC and Bandwidth Reservation

• Simulation Results

• Conclusions

3

Introduction 1/5

• Future mobile communication system To support broadband multimedia With diverse QoS r

equirements

• Handoff resource not guarantee Performance degradations Magnified in future micro/pico-cellular network

• Call admission control and bandwidth reservation scheme are required.

4

Introduction 2/5

• Handoff blocking are more objectionable than new call blocking.

• To keep handoff dropping rate below a target level.

5

Introduction 3/5

• Popular CAC Guard channel policy Fractional guard channel policy Distributed call admission control schemes

• Questions of the above assumption Exponentially-distributed channel holding time Perfect knowledge of the rate of handoff

6

Introduction 4/5

1. Research efforts to predict user mobility => don’t estimate channel holding time and theref

ore cannot be directly applied for efficient bandwidth reservation.

2. Each mobile will handoff to neighboring cells with equal probability.

=> This assumption may not be accurate in general

7

Introduction 5/5

• CAC and bandwidth reservation schemes based on the probabilistic prediction of user mobility.

• The Mobility prediction approach is derived from data compression techniques.

• Novel prediction approach => predict not only where the mobile users will

handoff but also when it will handoff.

8

Model Description

• The paper don’t consider Soft handoff in CDMA Delay-insensitive applications

• Subsections Network Topology Channel Holding Time User Mobility Pattern

9

Network Topology

Use a generalized graph model to represent the actual cellular network.

Modeled as a connected graph G = (V, E)

V={a,b,c,…..,n}

E={(a,b), (a,c),……(n,l)}

10

Channel Holding Time

The paper assumes that the channel holding time follows a general distribution, which allows the i.i.d. exponential channel holding time assumption to be relaxed.

11

User Mobility Pattern 1/3 Symmetric random walk model not take int

o account the trajectory and channel holding time of a mobile.

• Mobility of a user during a call can be represented by a sequence of events,

( N, H1, H2, H3, …. Hn,.. E )

12

User Mobility Pattern 2/3

sequence of events ( N, H1, H2, H3, …. Hn,. E )• N = (m, i, t) m, represents the mobile i, represents the original cell t, represents the time when the call arrives

• Hn = (Tk, i) Tk, the relative time elapsed since the beginning of the call

i, the cell to which the mobile will handoff

• E = ( Tk )• We quantize the relative time into slots of equal duration T, a design

parameter. So, Tk is the kth time slot since the beginning of the call.

13

User Mobility Pattern 3/3

• (N, H1, H2, H3, …. Hn,.. E ) is assumed to b

e generated by a mth order Markov source.

• Most mobile users have favorite routes and habitual movement patterns.

14

Mobility Prediction

• Motivated from optimal data compression methods (Ziv-Lempel algorithms )

• Compression Rationale: More probable event => short codewords Less probable event => longer codewords

• A good data compressor should also be a good predictor.

15

Optimal Data Compression

• Based on the Ziv-Lempel algorithms for data compression.

1. Parse each block of size n in a greedy manner into distinct substrings

X1, X2, ….., Xn

2. For each j ≧ 1, substring Xj without its last character is equal to some previous substring Xi ,where 0≦ i < j. X

j is encoded by the value i, using ┌ lg (j - 1) ┐ bits

3. Last character of Xj encode as ASCII using ┌ lgα┐ bits. α is the size of the input alphabet set.

16

Example 1

• Alphabet = {a, b, c}

• Input string = “aababcbaccababc…”

(a)(ab)(abc)(b)(ac)(c)(aba)(bc)

1 2 3 4 5 6 7 8

The seventh substring “aba”

“ab” match X2 , using ┌ lg (7 - 1) ┐ bits

“a” using ┌ lg 3 ┐ bits

17

Input string = “aababcbaccababc…” (a)(ab)(abc)(b)(ac)(c)(aba)(bc)

18

Pseudocode of Mobility Prediction

19

A Mobility Trie used mobility rediction

20

1. Modeling the sequence of events generated by a stationary mth order Markov source

2. Predict next events using the mobility prediction scheme derived from the

Ziv-Lempel algorithm.

=> Predict not only to which cell a mobile will handoff but also when the handoff will occur.

21

Implementation Considerations of the Mobility Prediction Scheme

• Maintain the statistics in a trie Create an array of pointers for each node Use a linked list at each node

• Use memory economically , but can be more processing

• A sliding windows may be used.

22

Call Admission Control and Bandwidth Reservation

A. Calculation of Pij(Tk)B. The Most Likely Cell-Time (MLCT)C. CAC and Bandwidth Reservation for New c

alls D. Adaptive Control of Admission Threshold

αE. CAC and Bandwidth Reservation for Hand

off Calls

23

Calculation of Pij(Tk)

• the probability that a mobile in cell i will visit cell j during time slot Tk

564

569

565

564

569

565

564

5618

5615

563

563

)5(,

)4(,

)3(,

)2(,

)1(,

TbPa

TbPa

TbPa

TbPa

TbPa

Example 2:

24

The Most Likely Cell-Time (MLCT)

• We select cells and time slots with Pij(Tk) greater than MLCT threshold , a design parameter, to form the MLCT of this mobile.

25

CAC and Bandwidth Reservation for New Calls

26

Adaptive Control of Admission Threshold

is too small, the handoff dropping probability arises.

is too large, the resource utilization will be decreased.

If Phd(m) < Phd, target (m) , is decreased by

Otherwise , is increased by

is a design parameter

27

CAC and Bandwidth Reservation for Handoff Calls

• When mobile node handoff to cell i, the CAC algorithm will admit it if the current free bandwidth of cell i can support the call.

• Bandwidth is reserved for mobile node in its MLCT accordingly.

28

Simulation Results

1. Each cell has a fixed link capacity of 40 bandwidth units (BUs)2. Time is quantized into units of T= 30s3. Voice => 1BU, Video => 4BUs4. Call durations are the same for all calls and exponentially distributed wi

th mean value of 120s5. Call requests are generated according to a Poisson process with rate 6. Two cases: low user mobility, 0-40 miles/hour higher user mobility, 40-70 miles/hour7. Target handoff dropping rate Phd is 0.018. MLCT threshold =0.08 , admission threshold =1 adaptive factor =0.02

OfferedLoad = 120 * * (( 1 – Pvoice) * 4 + Pvoice))

Assumptions:

29

Pvoice: 0.8 and 1

in the low and high mobility case

30

Comparison with static-reservation

31

Comparison with cell-reservation

32

Conclusions1. Events generated by a stationary mth order Markov sou

rce

2. Predict next events using the mobility prediction scheme derived from the Ziv-Lempel algorithm.

Predict not only to which cell a mobile will handoff but also when the handoff will occur.

• Based on assumptions more realistic than existing proposals.

• better balance of guaranteeing handoff dropping probability while maximizing resource utilization.