Добавил:
Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:

Advanced Wireless Networks - 4G Technologies

.pdf
Скачиваний:
60
Добавлен:
17.08.2013
Размер:
21.94 Mб
Скачать

338 MOBILITY MANAGEMENT

performance measure. This probability Pnc can be expressed as

 

P

P

P (1

P )

=

P

Pfh PN (1 PB)

(11.25)

1 PH (1 Pfh)

nc =

B +

F

B

B +

 

where the first and second terms represent the effects of blocking and handoff attempt failure, respectively. In Equation (11.25) we can guess roughly that, when cell size is large, the probabilities of cell crossing PN and PH will be small and the second term of Equation (11.25) (i.e. the effect of cell crossing) will be much smaller than the first term (i.e. effect of blocking). However, when the cell size is decreased, PN and PH will increase. The noncompleted call probability Pnc can be considered as a unified measure of both blocking and forced termination effects.

Another interesting measure of system performance is the weighted sum of PB and PF

CF = (1 α)PB + α PF

(11.26)

where α is in the interval [(0, 1)] and indicates the relative importance of the blocking and forced termination effects. For some applications PF may be more important than PB from the user’s point of view, and the relative cost α can be assigned using the system designer’s judgment.

11.2.4 Performance examples

For the calculations, the average message duration was taken as ¯M = 120s and the maximum

T

speed of a mobile of Vmax = 60 miles/h was used. The probabilities PB and PF as functions of (new) call origination rate per unit area a can be seen in Figure 11.24, with cell radius R being a parameter. A total of 20 channels per cell (C = 20) and one channel per cell for handoff priority (Ch = 1) was assumed. The CRQ scheme was used for this figure, and the

¯ ¯

mean dwell time for a handoff attempt TQ was assumed to be TH/10. As can be seen, PF is much smaller than PB and the difference between them decreases as cell size decreases. As expected, for larger R the effect of handoff attempts and forced terminations on system performance is smaller.

Probability

100

 

R =16

R =4

R =1

 

10-1

 

 

 

 

 

 

 

R =1

 

10-2

 

 

 

 

 

 

R =4

 

 

10-3

 

 

Priority scheme II

 

 

 

20 channels/cell

 

 

R =16

 

1 handoff channel/cell

 

 

 

blocking

 

 

 

 

 

 

 

 

forced termination

10-4

 

 

 

 

10-4

10-3

10-2

10-1

100

Call origination rate density (calls per s/square mile)

Figure 11.24 Blocking and forced termination probabilities for CRQ priority scheme.

CELLULAR SYSTEMS WITH PRIORITIZED HANDOFF

339

Probability

100

Blocking

Forced termination

10-1

 

 

 

 

 

 

 

 

 

 

 

Ch = 4

Ch = 0

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

10-2

 

 

2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0

 

 

 

 

 

 

 

 

2

 

 

10-3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

4

 

 

10-4

-3

 

 

 

 

 

 

 

 

10

10-2

10-1

 

Call origination rate density (calls per s/square mile)

Figure 11.25 Blocking and forced termination probabilities for CRQ systems with 20 channels/cell, R = 2 miles.

Probability

100

PB

CR

CRQ

10-1

10-2

PF

10-3

 

 

 

10-3

10-2

10-1

100

Call origination rate density (calls per s/square mile)

Figure 11.26 Blocking and forced terminations for priority CR and CRQ schemes (20 channels/cell, one handoff channel/cell, R = 2 miles).

Figure 11.25 shows PB and PF as functions of a. As the effects of increasing priority given to handoff calls over new calls by increasing Ch, PF decreases by orders of magnitude with only small to moderate increase in PB this exchange is important because (as was mentioned previously) forced terminations are usually considered much less desirable than blocked calls.

Blocking and forced termination probabilities for the two priority schemes are shown in Figure 11.26 as functions of call origination rate density a. The forced termination probability PF is smaller for th CRQ scheme, but almost no difference exists in blocking probability PB. We get this superiority of the CRQ priority scheme by queuing the delayed handoff attempts for the dwell time of the mobile in the handoff area.

340 MOBILITY MANAGEMENT

11.3 CELL RESIDING TIME DISTRIBUTION

In this section we discuss the probability distributions of the residing times Tn and Th. The random variable Tn is defined as the time (duration) that a mobile resides in the cell in which its call originated. Also Th is defined as the time a mobile resides in a cell to which its call is handed off. To simplify analysis we approximate the hexagonal cell shape as a circle. For a hexagonal cell having radius R, the approximating circle with the same area has a radius, Req, which is given by Req = (33/2π ) R 0.91R and illustrated in Figure 11.27. The base station is assumed to be at the center of a cell and is indicated by a letter B in the figure. The location of a mobile in a cell, which is indicated by a letter A in the figure, is represented by its distance r and direction φ from the base station as shown. To find the distributions of Tn and Th, we assume that the mobiles are spread evenly over the area of the cell. Then r and φ are random variables with PDFs

 

 

2r

0 r Req

 

 

1

,

0 φ 2π

 

 

 

,

 

 

 

 

fr (r) =

 

R2

,

fφ (φ) =

2π

 

eq

 

 

 

 

(11.27)

 

0,

 

elsewhere

 

 

0,

 

elsewhere

Next it is assumed that a mobile travels in any direction with equal probability and its direction remains constant during its travel in the cell. If we define the direction of mobile travel by the angle θ (with respect to a vector from the base station to the mobile), as

shown in the figure, the distance Z from the mobile to the boundary of approximating

circle is Z = [Req2 (r sin θ )2] r cos θ . Because φ is evenly distributed in a circle, Z is independent of φ and from the symmetry we can consider the random variable θ is in interval [0, π ] with PDF

 

1

,

0 θ π

fθ (θ ) =

 

π

 

 

 

(11.28)

 

0,

elsewhere

 

 

R

C

 

 

Req

 

φ

 

 

Z

r

B

q

 

 

y = rsinq

A

 

x = rcosq

Figure 11.27 Illustration of distance from point A in cell (where call is originated), to point C on cell boundary (where mobile exits from cell).

CELL RESIDING TIME DISTRIBUTION

341

If we define new random variables x, y as x = r cos θ , y = r sin θ , then Z = (Req2 y2) x and W = x. Since the mobile is assumed to be equally likely to be located anywhere in the

approximating circle

 

 

 

2

, Req x Req,

0 x2 + y2 Req2 , 0 y Req

fX Y (x, y) =

 

π R2

 

 

eq

 

 

0,

elsewhere

 

From Equations (11.27)–(11.28), the joint density function of Z and W can be found by standard methods

f

Z W

(z, w)

=

 

 

 

|z + w|

 

 

 

 

f

X Y

(x, y)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Req2 (z + w)2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

=

 

 

2

 

 

|z + w|

 

 

 

 

 

 

, 0

z

2R

,

1

z

w

≤ −

z

+

R

 

 

 

 

 

π Req2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2

 

 

 

 

 

 

 

 

2

 

 

 

 

 

 

 

eq

2

 

 

 

eq

 

 

 

 

 

 

 

 

 

 

Req (z + w)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

The PDF of the distance Z is then becomes

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Reqz

 

 

 

 

 

(z + w)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

f

Z

(z)

=

 

2

 

 

 

 

 

 

dw, 0

z

2R

eq

 

 

 

 

 

 

 

 

 

 

π Req2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2

 

(z + w)

2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

z/2

 

 

 

 

Req

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2

 

Req2

 

 

 

 

z

2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

, 0 z 2Req

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

=

 

 

π R2

 

2

 

 

 

 

 

 

 

(11.29)

 

 

 

 

 

 

 

 

 

 

eq

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0, elsewhere

If the speed V of a mobile is constant during its travel in the cell and random variable which is uniformly distributed on the interval [0, Vmax] with PDF

 

 

 

1

, 0

v

Vmax

 

 

 

 

 

fV (v) =

Vmax

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0, elsewhere

 

 

 

 

 

 

 

then the time Tn is expressed by Tn = Z / V with PDF

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

fTn (t) = |w| fZ (tw) fV (w)dw

 

 

 

 

 

 

 

 

 

−∞

Vmax

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2

 

 

 

 

tw

2

 

 

2Req

 

 

 

 

 

 

 

 

 

Req2

 

 

 

 

 

w

 

 

 

dw, 0 t

 

 

 

 

Vmaxπ R2

2

 

Vmax

 

eq

0

 

 

 

 

 

 

 

 

 

 

 

 

 

=

 

 

 

 

 

 

 

 

 

 

(11.30)

 

2Req t

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2

 

 

2Req

 

 

2

 

 

 

 

 

tw

 

 

w

 

 

Req2

dw, t

 

Vmaxπ R2

 

 

2

Vmax

 

eq

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0

342 MOBILITY MANAGEMENT

C

Z B

2Req

q

A

Figure 11.28 Illustration of distance from cell entering point (A on cell boundary), to cell exiting point (C on cell boundary).

 

8Req

 

 

 

t Vmax

2

3

2Req

 

 

 

 

 

 

1

1

 

, 0 t

=

3Vmaxπ t2

2Req

 

Vmax

8Req

 

2Req

 

 

 

 

 

, t

 

 

 

 

 

 

 

 

 

 

 

 

 

3Vmaxπ t2

Vmax

 

 

 

 

and the cdf of Tn is

t

FTN (t) = fTn (x)dx FTn

−∞

 

 

 

 

 

 

 

 

 

 

4

 

 

 

1

 

 

 

 

2

arcsin

 

Vmaxt

 

 

tan

arcsin

Vmaxt

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

π

 

2Req

3π

2

2Req

 

= +

1

 

sin

2 arcsin

 

Vmaxt

 

, 0 t

2Req

 

(11.31)

3π

 

2Req

Vmax

1

 

8Req 1

 

2Req

 

 

 

 

 

 

 

 

 

, t

 

 

 

 

 

 

 

 

3π Vmax

t

Vmax

 

 

 

 

 

 

To find the distribution of Th, in the next step we note that, when a handoff call is attempted, it is always generated at the cell boundary, which is taken as the boundary of the approximating circle. Therefore, to find Th one must recognize that the mobile will move from one point on the boundary to another. The direction of a mobile when it crosses the boundary is indicated by the angle θ between the direction of the mobile and the direction from the mobile to the center of a cell as shown in Figure 11.28 [75, 76].

If the mobile moves in any direction with equal probability, the random variable θ has PDF given by

 

1

,

π

θ

π

fθ (θ ) =

 

π

2

2

0,

elsewhere

 

 

 

 

 

CELL RESIDING TIME DISTRIBUTION

343

The distance Z as shown in Figure 11.28 is Z = 2Req cos θ , which has a CDF given by

 

 

 

 

0, z < 0

 

 

 

 

 

 

 

FZ (z) = Pr {Z z} =

 

1

2

arccos

 

z

, 0 z 2Req

(11.32)

 

 

 

 

 

 

 

 

 

π

 

2Req

 

 

 

 

1, z > 2Req

 

 

 

 

 

 

 

The PDF of Z is

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1

 

 

1

 

 

 

, 0

z 2Req

 

 

d

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

fZ (z) =

FZ (z)

=

 

π

 

 

 

Req2

z 2

 

(11.33)

dz

 

 

 

 

 

2

 

 

 

 

0, elsewhere

The time in the cell Th is the time that a mobile travels the distance Z with speed V , then Th = Z / V. With the same assumption about V , the PDF of Th is

 

 

 

 

 

 

 

 

 

 

 

 

1

 

 

 

Vmax

 

 

 

 

 

 

w

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2Req

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

dw, 0

t

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

π Vmax

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Vmax

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

tw

2

 

 

 

 

 

 

 

 

 

 

0

 

 

 

Req2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

fTh (t) = |w| fZ (tw) fV (w)dw =

 

 

1

 

 

 

2Req t

 

 

 

 

 

w

 

 

 

 

 

 

 

 

2Req

 

0

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

dw,

t

 

 

 

 

 

 

 

 

 

 

 

 

π Vmax

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Vmax

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

tw

2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0

 

 

 

Req2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

4Req

1

 

 

 

 

 

 

 

 

 

 

 

Vmaxt

2

 

 

 

 

 

 

2Req

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1

 

 

 

1

 

 

 

 

 

 

, 0 t

 

 

 

 

 

 

 

 

 

 

=

 

 

π Vmax t2

 

 

 

2Req

Vmax

 

 

 

 

 

 

 

 

 

 

 

4Req

 

 

1

 

 

 

 

 

2Req

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

,

t

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

π Vmax

 

 

t2

 

Vmax

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(11.34)

and the CDF of Th is

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

t

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

FTh (t) = fTh (x) dx

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

−∞

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0, t < 0

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2

arcsin

Vmaxt

2

 

tan

1

arcsin

 

 

Vmaxt

 

 

 

 

, 0 t

2Req

(11.35)

=

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

π

2Req

π

2

 

 

2Req

 

 

 

 

 

Vmax

1

4Req

1

, t

2Req

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

>

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

π Vmax

t

Vmax

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 11.29 shows the mean channel holding time in a cell TH. Notice that TH becomes smaller with smaller cell size, but sensitivity to change in cell size is smaller for larger cells. Finally, earlier in Section 11.2 we approximated the cumulative distribution function of the channel holding time in a cell as suggested in References [75, 76]. The goodness-of-fit

344 MOBILITY MANAGEMENT

130

 

 

 

 

 

 

 

120

Maximum velocity of mobile = 30 miles/h

 

 

 

 

 

 

 

 

 

 

110

 

 

 

 

 

 

 

100

 

 

 

40

 

 

 

90

 

 

 

60

 

 

 

 

 

 

 

 

 

 

80

 

 

 

 

 

 

 

70

 

 

 

 

 

 

 

60

 

 

 

 

 

 

 

50

 

 

 

 

 

 

 

0

2

4

6

8

10

12

14

 

 

 

Cell radius (miles)

 

 

 

Figure 11.29 Mean channel holding time (s) in cell vs R (average call duration = 120 s).

Table 11.1 The goodness-of-fit G of the approximation

Cell radius, R

G

 

 

1.0

0.020220

2.0

0.000120

4.0

0.000003

6.0

0.000094

8.0

0.000121

10.0

0.000107

12.0

0.000086

14.0

0.000066

16.0

0.000053

 

 

G of this approximation, defined in Equation (11.8), is shown in Table 11.1 for various cell sizes. We see that G is very small for a wide ranges of cell radius R. These values support the use of the approximation in the calculations.

11.4 MOBILITY PREDICTION IN PICOAND MICROCELLULAR NETWORKS

It should be expected that 4G networks will further reduce the size of the cells. In a microand picocellular network, resources availability varies frequently as users move from one access point to another. In order to deterministically guarantee QoS support for a mobile unit, the network must have prior exact knowledge of the mobile’s path, along with arrival and departure times to every cell along the path. With this knowledge, the network can verify the feasibility of supporting the call during its lifetime, as the mobile moves across the network. It is impractical, however, to require the users to inform the network of their exact movement, since they may not know this information a priori. Even if the path is

MOBILITY PREDICTION IN PICOAND MICROCELLULAR NETWORKS

345

known, the exact arrival and departure times to the cells along the path are still hard to determine in advance. Therefore, it becomes crucial to have an accurate mechanism to predict the trajectory of the mobile user.

As an example, the virtual connection tree is designed to support QoS guarantees for mobile units [77]. In this scheme, a number of adjacent cells are grouped into a cell cluster in a static fashion. Upon the admission of a call, the scheme pre-establishes a connection between a root switch and each base station in the cell cluster. The scheme does not take user mobility into consideration to predict the set of base stations which may potentially be visited by the mobile unit. This may result in an unnecessary resource overloading that may underutilize the network resources and cause severe congestion.

The shadow cluster (SC) scheme [78] provides a distributed call-admission control (CAC) based on the estimated bandwidth requirements in the SC. An SC is a collection of base stations to which a mobile unit is likely to attach in the future. The admission decision is made in a distributed fashion by all the base stations within the SC. The scheme partitions the time into predefined intervals and verifies the feasibility of supporting calls over those intervals. This requires the communication of large number of messages between base stations during every time interval. Moreover, since bandwidth estimates are calculated at the beginning of each time interval and the admission decisions are made at the end of each time interval, admission of new calls is delayed for at least a time equal to the length of these predefined time intervals.

Both of the above two schemes lack the mechanism to predict the mobile’s trajectory and determine the future cells to which the mobile may hand off. Several techniques have been proposed in the literature to address this issue. In Bharghavan and Mysore [79], a profile-based algorithm is proposed to predict the next cell that the mobile unit will hand off, using a user profile and a cell profile, which are simply the aggregate values of the handoff’s history. In Liu and Maguire [80], a mobile motion prediction (MMP) algorithm is proposed to predict the future locations of the mobile unit. This algorithm is based on a pattern matching technique that exploits the regularity of the users’ movement patterns. The MMP algorithm was further expanded to a two-tier hierarchical location prediction (HLP) algorithm [81]. In the latter case, the two-tiered prediction scheme involves both an intercell and an intracell tracking and prediction component. The first tier uses an approximate pattern matching technique to predict the global intercell direction and the second tier uses an extended self-learning Kalman filter to predict the trajectory within the cell using the measurements received by the mobile unit.

In order to support QoS guarantees of multiple classes of services, the scheme must integrate call and admission control with the mobility profile of the mobile user. The integration of these two components makes it possible to use mobility prediction to verify the feasibility of admitting a new call and make sure that the required QoS can be supported during the lifetime of the call. In other words we should be able to predict location (space) and time when a certain resources well be needed in the network. This concept will be referred to as space time predictive QoS or STP QoS. The mobility prediction algorithm must be easy to implement and maintain, since it will be invoked on a per-user basis. Furthermore, the admission control procedure must be invoked only when needed with minimum overhead and in a distributed fashion, where each network cell, potentially involved in supporting the QoS requirements of the call, participates in the decision process [82, 83].

In this section, we present such a framework, which efficiently integrates mobility prediction and CAC, to provide support for PST-QoS guarantees, where each call is guaranteed

346 MOBILITY MANAGEMENT

its QoS requirements for the time interval that the mobile unit is expected to spend within each cell it is likely to visit during the lifetime of the call.

In this framework, efficient support of PST-QoS guarantees is achieved based on an accurate estimate of mobile’s trajectory as well as the arrival and departure times for each cell along the path. Using these estimates, the network can determine if enough resources are available in each cell along the mobile’s path to support the QoS requirements of the call. The framework is designed to easily accommodate dynamic variations in network resources. The basic components of this framework are:

(1)a predictive service model to support timed-QoS guarantees;

(2)a mobility model to determine the mobile’s most likely cluster (MLC); the MLC represents a set of cells that are most likely to be visited by a mobile unit during its itinerary; and

(3)a CAC model to verify the feasibility of supporting a call within the MLC.

The service model accommodates different types of applications by supporting integral and fractional predictive QoS guarantees over a predefined time-guarantee period. The MLC model is used to actively predict the set of cells that are most likely to be visited by the mobile unit. For each MLC cell, the mobile’s earliest arrival time latest arrival time and latest departure time are estimated. These estimates are then used by the CAC to determine the feasibility of admitting a call by verifying that enough resources are available in each of the MLC cells during the time interval between the mobile’s earliest arrival time and its latest departure time. If available, resources are then reserved for the time interval between the mobile’s earliest arrival time and latest departure time and leased for the time interval between the mobile’s earliest and latest arrival times. If the mobile unit does not arrive before the lease expires, the reservation is canceled and the resources are returned to the pool of available resources. The unique feature of the this frame-work is the ability to combine the mobility model with the CAC model to determine the level of PST-QoS guarantees that the network can provide to a call and dynamically adjust these guarantees as the mobile unit moves across the network.

11.4.1 PST-QoS guarantees framework

The first approach, to achieve a high level of QoS support guarantees in mobile environments, is to allocate resources for the duration of the call in all future cells that the mobile unit will visit. This means that the resources within each cell that is to be visited will be held, possibly for the duration of the call, even if the mobile never moves into the cell. This approach is similar to the one proposed in Talukdar et al. [85] and will be referred to as a predictive space or PS QoS model. Clearly, such an approach will result in underutilization of the network resources as resources are being held, but not used by any call.

The second approach is to only reserve resources in all future cells that the mobile unit may visit for the time interval during which the mobile will reside in each cell. If ti and ti+1 represent the expected arrival and departure times of the mobile unit to cell i along the path, respectively, resources in cell i will only be reserved for the time interval [ti , ti+1] [84]. Unlike the first approach, this approach is likely to increase resource utilization, since resources in every cell remain available to other calls outside the reservation intervals. This approach, however, is only feasible if exact knowledge of the mobile path and arrival

MOBILITY PREDICTION IN PICOAND MICROCELLULAR NETWORKS

347

and departure times to every cell along the path is available. Obtaining exact knowledge of mobile mobility is not possible in most cases, due to the uncertainty of the mobile environments and the difficulty in specifying the mobility profiles of mobile units. An acceptable level of service guarantees, however, can be achieved if the path of the mobile can be predicted accurately. This approach is discussed in this section and will be referred to as predictive space and time or the PST QoS model. The model attempts to achieve a balance between an acceptable level of service guarantees and a high level of network resource utilization. Based on this model, the required QoS support level is guaranteed by reserving resources in advance in each cell that is most likely to be visited by the mobile unit. These reservations only extend for a time duration equal to the time interval the mobile unit is expected to spend within a cell, starting from the time of its arrival time to the cell until its departure time from the cell. In order to characterize the set of ‘most likely’ cells and capture the level of QoS guarantees requested by the application, the service model uses the following parameters:

(1)the time guarantee period TG ;

(2)a cluster-reservation threshold τ , and a

(3)bandwidth-reservation threshold, γ .

All of these parameters are application-dependent. The parameter TG specifies the time duration for which the required QoS level is guaranteed; τ defines the minimum percentage of the most likely cells to be visited by the mobile unit that must support the required QoS level for the guarantee period TG . The parameter γ represents the minimum percentage of the required bandwidth that must be reserved in every cell that is most likely to be visited.

To accommodate different types of applications, the service model provides two types of predictive service guarantees, namely, integral guaranteed service and fractional guaranteed service. The integral guaranteed service ensures that all cells, which are most likely to be visited by the mobile unit, can support the requested bandwidth requirements for the lifetime of the call. In this case, TG must be equal to the call duration and τ and γ are both equal to 100 %. The fractional guaranteed service, on the other hand, guarantees that at least τ % of these cells can support at least the γ % of the requested bandwidth requirements for the next TG interval. A special case arises when either the value of τ or γ is zero. In this case, the service is referred to as best effort.

11.4.2 Most likely cluster model

The MLC model considers that the ‘most likely to be visited’ property of a cell is directly related to the position of the cell with respect to the estimated direction of the mobile unit. This likelihood is referred to as directional probability. Based on this metric, cells that are situated along the mobile unit’s direction have higher directional probabilities and are more likely to be visited than those that are situated outside of this direction.

Based on the above, the MLC at any point in time during the lifetime of a call is defined as a collection of contiguous cells, each of which is characterized by a directional probability that exceeds a certain threshold. For each MLC cell, the expected arrival and departure times of the mobile are estimated. Using these estimates, the feasibility of supporting the requested level of timed-QoS guarantees during the mobile’s residence time within each cell along path is verified. In the following, we present the method used to predict the direction