Improving GSM Call Completion by Call Re-Establishment

Phone Lin, Yi-Bing Lin, and J.-Y. Jeng Dept. Comp. Sci. & Info. Engr., National Chaio Tung University Hsinchu, Taiwan, R.O.C.; Fax:+886-35-724-176; Telecommunication Laboratories, Chunghwa Telecom Co., Ltd. Email: fliny,pling@csie.nctu.edu.tw

Abstract
GSM call re-establishment service allows a mobile station to resume a call in which the radio link has been temporarily interrupted due to interference or bad signal (which is referred to as an interrupted call). This service increases end user satisfaction and network quality perception. In this paper, we propose analytic models to study the performance for call re-establishment service. Our study indicates that call re-establishment can signi cantly reduces dropping for interrupted calls.

1 Introduction
Personal Communications Services (PCS) networks provide telecommunications services to moving users. During a PCS communication session, a radio link is established between the mobile station (MS) and a base station (BS) if the MS is in the cell (the radio coverage area of the BS). If the MS moves to another cell during the conversation, then the radio link to the old BS is disconnected and a radio link to the new BS is required to continue the conversation. This process is called hando 1, 2]. If the new BS does not have any idle channel, the hando call is forced to terminate. Besides forced termination due to hando , a radio link may be temporarily disconnected when propagation loss due to obstacle (e.g., bridges, tunnels) shielding. This phenomenon is referred to as \call interruption". To avoid forced termination due to call interruption, the call re-establishment service has been proposed in GSM 4, 10]. In this mechanism, if a communication channel is interrupted, the network still reserves the trunk and/or the channel for the interrupted call, and an interruption timer is triggered. If the timer expires or the remote party hangs up the phone before the interruption

2 period is over, the interrupted call is actually forced to terminate. Otherwise, the interrupted call is resumed by the call re-establishment mechanism. In this paper, we propose analytic and simulation models to evaluate the performance of GSM system with call re-establishment service.

2 Call Re-establishment Mechanisms

This section describes three algorithms to reduce forced termination caused by interruption. Consider the timing diagram in Figure 1 (a). Suppose that a call alternates between the conversation periods and the interrupted periods. De ne the ith cycle of a call as pair (xi ; yi ) where xi is the ith conversation period and yi is the ith interruption period. Every yi is associated with a periods zi that denotes the interval between when the interruption begins and when the rst of the following two events occurs: (i) the interruption timer expires, (ii) the remote party hangs up the phone. For k 0, let tk be the holding time of the rst k conversion cycles (i.e., tk = x1 + y1 + x2 + y2 + : : : + xk + yk ). By convention, t0 = 0. Let tD;k = tk + xk+1 . If the interrupted call is not resumed before the period zi expires, the interrupted call is forced to terminate. Let m be the period between the arrival of the call and when the MS enters the next cell (called cell 1), and tm;0 be the cell residence time of the MS at cell 0. Three Call Re-establishment algorithms RES1 (the radio channel is reserved during interruption; the call is not re-established if the MS moves into a new cell), RES2 (the radio channel is not reserved during interruption; the call is not re-established if the MS moves into a new cell) and RES3 (the radio channel is reserved during interruption; the call is re-established if the MS moves into a new cell) are described as follows.

Algorithm RES1: For k 0, consider the k + 1st cycle of the call. There are ve cases: CASE(I) If tk <
(Figure 1 (b)), the MS moves to cell 1 during the conversation period. The call is handed o from cell 0 to cell 1 following the standard hando procedure, and the call re-establishment mechanism is not triggered during hando . CASE (II): If tD;k +yk+1 m and yk+1 zk+1 (Figure 1 (c)), then the call is re-established at cell 0 after the MS leaves the shielding area. Figure 3 illustrates the messages exchanged between the MS and BS0 (the BS at cell 0): After interruption is over, the MS sends the call re-establishment request message to BS0 (message 1 in Figure 3). The message contains the MS identi cation (ID) and the ID of the BS at which the call is interrupted (in this case, it is BS0). When BS0 receives
m < tk + xk+1

call arrival z1 y1 z2 y2 t m,1 move to cell 1

call completion tc x2 x3 t c,i

x1

xj x*j

zj yj

t m,0 enter cell 0

t m,i leave cell 1 enter cell i leave cell i (a) x i, y i, z i, t c , t m,i ,

xk time

,t
m

c,i

, x*j

call arrival tk x1 z1 y1 x2 z2 y2 x3 xk zk yk x k+1 time

t m,0 enter cell 0

(b) CASE (I)

enter cell 1

call arrival x1 z1 y1 x2 t D,k z2 y2 x3

the call can be resumed time

x k+1

z k+1 y k+1

t m,0 enter cell 0

enter cell 1 (c) CASE (II)

Figure 1: Timing Diagram I

call arrival t D,k x1 z1 y1 x2 z2 y2 x3

zk yk

x k+1

z k+1 y k+1 time

t m,0 enter cell 0

enter cell 1 (a) CASE (III) the call cannot be resumed z1 y1 t D,k z2 y2 t c' x3

call arrival x1 x2

x k+1

z k+1 y k+1 time

t m,0 enter cell 0

enter cell 1 (b) CASE (IV)

call arrival t D,k x1 z1 y1 x2 z2 y2 x3

zk yk

x k+1

z k+1 y k+1 time

t m,0 enter cell 0

enter cell 1 (c) CASE (V)

Figure 2: Timing Diagram II

5
MS BS0
Timer starts

MSC

Call is interrupted in this period 1. call re-establishment request 2. call re-establishment request acknowledgement Call is re-established at cell 0
Timer stops

Figure 3: Message Flow for CASE II of RES1 the message, it checks the call record of the MS and stops the corresponding interruption timer. BS0 acknowledges the re-establishment request (message 2 in Figure 3), and the call is re-established. CASE (III): If tD;k < m < tD;k + yk+1 and yk+1 zk+1 (Figure 2 (a)), the interruption period ends before the interruption timer expires and the remote party does not hang up the phone. Since the MS enters cell 1 during the interruption period, the call is forced to terminate due to the fact that GSM follows the Mobile Assisted Hando strategy 2, 8]. In this case, BS0 will not release the reserved channel until the end of zk+1 . Note that the GSM Mobile Assisted Hando mechanism cannot perform radio link transfer if the MS fails to receive signal from the old cell (i.e., cell 0) during the hando process. Figure 4 illustrates the message ow for this case. There are two possibilities: (a) The remote party hangs up the phone rst (Figure 4 (a)), and (b) the interruption timer expires rst (Figure 4 (b)). CASE (IIIa): The remote party hangs up the phone before the interruption timer of BS0 expires. During the interruption period, the MS moves into cell 1. After the interruption, the MS sends BS1 the call re-establishment request message (see message 1 in Figure 4 (a)). Since call interruption occurs at cell 0, BS1 cannot nd the call record of the MS. BS1 replies a negative acknowledgment (see message 2 in Figure 4 (a)) that causes the call to be forced to terminate. Since yk+1 < zk+1 , BS0 still reserves the radio channel at this point. When the remote party hangs up the phone, the MSC cancels the call record of the MS, releases the trunk to the remote party, and sends a clear command message to BS0 (see message 3 in Figure 4

6 (a)). After receiving the message, BS0 cancels the call record of the MS, releases the reserved channel for the interrupted call, and sends a clear complete message to the MSC (see message 4 in Figure 4 (a)). CASE (IIIb): The interruption timer of BS0 expires before the remote party hangs up the phone. The rst two messages delivered between the MS and BS1 are the same as those in CASE (IIIa) (see messages 1 and 2 in Figure 4 (b)). BS0 then sends a radio interface failure message to the MSC (see message 3 in Figure 4 (b)). Based on the message, the MSC cancels the call record of the MS, releases the trunk to the remote party, and sends a clear command message to BS0 (see message 4 in Figure 4 (b)). When BS0 receives the message, it cancels the call record of the MS, releases the reserved channel, and sends a clear complete message (see message 5 in Figure 4 (b)) to the MSC. Note that messages 1 and 2 are not required in RES1. When the MS detects that it has moved to a new cell, it can terminate the call without exchanging these two messages. This message pair is required in RES3 to be described. CASE (IV): If tD;k + yk+1 m and yk+1 > zk+1 (see Figure 2 (b)), the user hangs up the phone before the interruption period is over. In this case, the MS does not leave cell 0 during the interruption period. The call is dropped at cell 0, and BS0 releases the reserved channel after the interruption timer expires or when the remote party hangs up the phone. The message ow for this case is similar to CASE (III), except that after the MS leaves the shielding area, it sends the call re-establishment request to BS0. Upon receipt of the message, BS0 nds that the call record of the MS does not exist. BS0 sends a negative acknowledgment to the MS, and the MS terminates the call. CASE(V): If tD;k < m < tD;k + yk+1 and yk+1 > zk+1 (Figure 2 (c)), the interruption period ends after the interruption timer expires or after the remote party hangs up the phone. After the interruption, the MS is in cell 1, and the call is dropped as in CASE (III). In this case, BS0 releases the reserved channel when the interruption timer expires or after the remote party hangs up the phone. The message ow for this case is the same as that in CASE (III). To implement RES1, we only need to make minor modi cations to the BS and MS. No changes are required in the MSC. In Figures 3 and 4, the messages delivered between BS

MS

BS0
Timer starts

MSC

BS1

The call is interrupted, and the MS enters cell 1 in this period 1. call re-establishment request 2. call re-establishment request N-ack 3. clear command message 4. clear complete message Call is forced terminated Reserved channel is released in the BS at cell 0
Timer stops

The remote party hangs on the phone

(a) The call is forced terminated at cell 1, and the reserved channel is released by the MSC BS1

MS

BS0
Timer starts

MSC

The call is interrupted, and the MS enters cell 1 in this period 1. call re-establishment request 2. call re-establishment request N-ack 3. radio interface failure Call is forced terminated 4. clear command message 5. clear complete message Reserved channel is released in the BS at cell 0
Timer expires

(b) The call is forced terminated at cell 1, and the reserved channel is released by BS0

Figure 4: Message Flow for CASE III of RES1

8 and MSC already exist in the current GSM implementation. In other word, there is no need to introduce new message types for the A interface 3] between the BS and MSC.

Algorithm RES2: RES2 is similar to RES1 except that as soon as the radio link between the MS

and BS0 is interrupted, the BS0 releases the radio link. After interruption, the MS makes a Call Re-establishment request to the BS0. If BS0 has an idle channel, the interrupted call is re-established. RES2 has been implemented in the existing Nortel GSM system 4, 10].

RES1 and RES2 fail to resume the interrupted call if the MS moves from cell 0 to cell 1 during interruption. To relax this restriction (i.e., to allow call re-establishment at cell 1), We extend RES1 as follows:

Algorithm RES3: RES3 allows a call to be re-established after the MS moves to a new cell during
interruption. For CASEs (I), (II) and (IV), the actions taken by RES3 are exactly the same as that in RES1. The actions for CASEs (III) and (V) are described as follows.

CASE (III): If tD;k

cell 1 during the interruption period, and neither the interruption timer expires nor the remote party hangs up the phone during the interruption period. The MS makes a call re-establishment request to BS1. If BS1 has an idle channel, the call is re-established. Figure 5 illustrates the message ow for CASE (III). After interruption, the MS sends the call re-establishment request message (see message 1 in Figure 5) to BS1. On receipt of the message, BS1 forwards the call re-establishment request to the MSC (see message 2 in Figure 5). The MSC checks the call record of the MS, and sends a clear command message (see message 3 in Figure 5) to BS0. BS0 stops the interruption timer, releases the reserved channel , and sends a clear complete message to the MSC (see message 4 in Figure 5). The MSC sends a cipher mode command message (that contains the cipher info for the interrupted call; see message 5 in Figure 5) to BS1. After receiving message 5, BS1 responds a cipher mode complete message (see message 6 in Figure 5) to the MSC. The MSC then sends an assignment request message (see message 7 in Figure 5) to BS1 to assign a channel to the interrupted call. BS1 queries the channel pool to nd an idle channel for the interrupted call. If BS1 has idle channels, BS1 sends a assignment complete message to the MSC (see message 8 in Figure 5) to indicate that

< m < tD;k + yk+1 and yk+1

zk+1 (Figure 2 (a)), the MS enters

9
MS BS0
Timer starts

MSC

BS1

The call is interrupted is in this period, and the MS moves to cell 1 1. call re-establishment request
Timer stops

3. clear command message 4. clear complete message

2. call re-establishment request 5. Cipher Mode Command 6. Cipher Mode Complete 7. Assignment Request 8. Assignment Complete 9. call re-establishment request acknowledgement

10. call re-establishment request acknowledgement The call is reestablished on the link at cell 1

Figure 5: Message Flow for CASE III of RES2 BS1 is ready to accommodate the call transfer. The MSC sends a call re-establishment request acknowledgment to BS1 (see message 9 in Figure 5), which is forwarded to the MS (see message 10 in Figure 5). At this point, the call is re-established at cell 1. Note that message 5-8 are standard GSM actions for link setup. For more details, the reader is referred to 4]. CASE (V): If tD;k < m < tD;k + yk+1 and yk+1 > zk+1 (see Figure 2 (c)), the call is forced to terminate at cell 1, and BS0 releases the reserved channel either when the interruption timer expires or when the remote party hangs up the phone. Figure 6 illustrates the message ow for this case. Period zk+1 expires before the interruption period ends. Thus, BS0 releases the reserved channel before the MS makes a call re-establishment request to BS1. The message ow for this action is the same as CASE (III) in RES1. When interruption is over, the MS sends the call re-establishment request message to BS1 (see message 3 in Figure 6). After receiving the message, BS1 forwards the request to the MSC (see message 4 in Figure 6). Since the MSC cannot nd the

10
MS BS0
Timer starts

MSC

BS1

The call is interrupted is in this period, and the MS moves to cell 1 3. call re-establishment request

1. clear command message 2. clear complete message

Timer stops

4. call re-establishment request 5. call re-establishment request N-ack 6. call re-establishment request N-ack

The call is forced terminated at cell 1

Figure 6: Message Flow for CASE V of RES2 call record for the MS (the MSC call record has been deleted after the BS interruption timer expired or after the remote party hung up the phone). The MSC replies a negative acknowledgment to BS1 (see message 3 in Figure 6), which is forwarded to the MS (see message 4 in Figure 6). At this point, the interrupted call is dropped in cell 1. To implement RES3, modi cations are made to the BS, MS, and MSC.

3 Analytic Models
We propose analytic models for GSM basic scheme (without call re-establishment), RES1 and RES3. The call incompletion probability Pnc is derived to investigate the performance of these algorithms. Call incompletion includes new call blocking and connected call dropping. This section describes the analytic models for RES1 and RES3. The model for GSM basic scheme is similar to that for RES1 (but is less complicated) and is omitted.

11

3.1 An Analytic Model for RES1

Consider a cell in the GSM system. Let o ( h ) be the new (hando ) call arrival rate to the cell. Let Pb be the probability that all channels are busy when a call (either a new call or a hando call) arrives. In GSM, the same channel assignment procedure is used for both the new calls and hando calls. This non-prioritized scheme is considered in this paper. By using the techniques we proposed in 9], our model can be extended to study the case where the hando calls have priority over the new calls. Let Ph;1 (Ph;2 ) be the probability that a connected new (hando ) call at the cell will hand o to the next cell. For a homogeneous cell structure (where the hando rate entering the cell is equal to the hando rate leaving the cell) , we have
h = o (1 ? Pb )Ph;1 + h (1 ? Pb )Ph;2

(1)

Let Pi;1 (Pi;2 ) be the probability that a connected new (hando ) call at the cell will be disconnected due to interruption. As we described in Section 3, a call alternates between the conversation periods and the interrupted periods. Assume that at the end of the conversation period xi , the call is complete with probability 1 ? , and with probability , the radio channel is interrupted for a period yi . If yi zi , the call is re-established and continues with the next conversation period xi+1 . Assume that xi are i.i.d. random variables with the density function x e? x xi , yi are i.i.d. random variables with the density function y e? y yi , and zi are i.i.d. random variables with the density function fz (zi ) respectively. Exponential interruption periods are used in the analytic model to provide the mean value analysis. E ect of higher moments for general distribution can be studied in our simulation. Let be the probability that a call is re-established after interruption. Then = Pr yi Z zi ] Z 1 zi = y e? y yi fz (zi )dyi dzi zi =0 yi =0 = 1 ? fz ( y )

(2)

where fz (s) is the Laplace Transform of the zi distribution. Let tc be the call holding time of a complete call without considering the hando e ect (see Figure 1 (a)). The density function fc(tc ) for tc is
fc(tc ) =
X

tc

tc ?x1

k=0 k Y i=1

x1 =0 y1 =0

:::

tc ?x1 ?y1 ?:::?xk

yk =0

k k (1 ?

) (3)

i ( x e? x xi y e? y yi ) x e? x (tc ?x1 ?y1 ?:::?xk ?yk ) dyk : : : dx1

12 From (3), the Laplace Transform of the tc distribution is (s + x)(s + y ) ) (4) = (s + (1 ? s + x (s + y ) x )( y) ? x y Let tk be the holding time for the rst k cycles of a call (see Figure 1 (b)). The density function fk (tk ) of tk (without considering hando ) is expressed as
k=0

fc (s) =

"

x y

(1 ? ) x s+ x

::: yk =0 xk =0 fk (tk ) = > x1 =0 y1 =0 k > Y > > > ( x e? x xi y e? y yi )dyk dxk : : : dy1 dx1 ; > :
i=1

8 > > > > > > <

0; Z

tk

tk ?x1

tk ?x1 ?y1 ?:::?xk?1 ?yk?1

tk ?x1 ?y1 ?:::?yk ?xk k k

if k = 0 (5) if k > 0

and its Laplace Transform is

fk (s) =

"

(6) (s + The residence time of the MS at cell i (the time interval that the MS stays in cell i) is tm;i (see Figure 1 (a)). For all i 1, tm;i are assumed to be i.i.d. random variables with the density function fm (tm;i ) = e? tm;i . Suppose that a call arrives when the MS is in cell 0. Let m be the period between the arrival of the call and when the MS moves out of cell 0. In our study, the cells are numbered 0, 1, 2, , in the order they are visited by the MS. Let rm ( m ) be the density function of m with the Laplace transform rm (s). The probability Ph;1 is derived as follows.
Ph;1 = Pr CASE (I) in RES1 occurs]

x y x )(s + y )

= = + = +

k=0 "
X

Pr tk < m < tk + xk+1 ]

Z

=1 Z

X
=1 Z

tk =0 m =tk xk+1 = m ?tk Z 1 rm ( m ) x e? x x1 dx1 d m m =0 x1 = m

fk (tk )rm ( m ) x e? x xk+1 dxk+1 d m dtk

tk =0 m =tk

fk (tk )rm ( m )e?

x md m

x ( m ?tk ) d m dt k

m =0

rm ( m )e?

(7)

13 Since fm (tm;i ) = e? tm;i , from the excess life theorem 6], rm ( m ) = e? as

Ph;1 =
X
=1 Z

and (7) is re-written

+ =

1
=0

tk =0 m =tk

fk (tk ) e?
#

m e? x ( m ?tk ) d m dtk

e?(

x) md m

"

) = ( + )( (++ )y? (8) x y x y The probability Pi;1 is derived as follows. Consider CASE (IV) in RES1. Let tc0 = tD;k + yk+1 and yk+1 > zk+1 (see Figure 2 (b)). Then the density function fc0 (tc0 ) is
fc0 (tc0 ) =
X

+ x

k=0

fk ( )

"

tc0

tc0 ?x1

k=0 k Y

x1 =0 y1 =0

:::

yk?1 =0

tc0 ?x1 ?y1 ?:::?xk+1 k+1 k (1 ?

i=1 i x e? x xk+1 y e? y (tc0 ?x1 ?y1 ?:::?xk+1 ) dxk+1 : : : dx1

( x e? x xi y e? y yi )

(9)

and its Laplace Transform is

x (1 ? ) y (s + x )(s + y ) ? x y From (10), the probability of CASE (IV) in RES1 is derived as

fc0 (s) =

(10)

Pr CASE (IV) in RES1 occurs] = Pr tc0 Z m ] Z 1 m = e?

m =0 tc0 =0

m fc0 (tc0 )dtc0 d m

(11) = ( + )( x (1 ? ) ) y x + y ? x y Consider CASE (III) and CASE (V) in RES1 (see Figure 2 (a) and (c)). The density function fD;k (tD;k ) of tD;k is expressed as
fD;k (tD;k ) = > x1 =0
> > > > > : Y 8 > > > > > > < Z

t xe? xZD;k ; tD;k tD;k ?x1 k

i=1

tD;k ?x1 ?y1 ?:::?xk?1 ?yk?1 k+1 k ::: xk+1 =0 y1 =0 x e? x xi y e? y yi ) x e? x xk+1 dxk+1 dyk dxk : : : dy1 dx1 ;
Z

if k = 0 (12) if k > 0

14 and its Laplace Transform is

"

fD;k (s) =

(s + x)(s + y )

x y

s+ x

(13)

We have Pr CASE (III) or (V) in RES1 occur] = Pr tD;k < m < tD;k + yk+1 ] " = = = =
X

k=0 " tD;k =0 m =tD;k yk+1 = m ?tD;k

X

fD;k (tD;k )rm ( m ) y e? y yk+1 dyk+1 d m dtD;k

y ( m ?tD;k ) d m dtD;k
# #

k=0 " tD;k =0 m =tD;k

X

fD;k (tD;k )rm ( m )e

!

k=0

tD;k =0

fD;k (tD;k )
X

!"

? tD;k dtD;k + y e

+ y

= ( + )( x x y Since CASE (IV), (III) and (V) in RES1 will drop the call, from (11) and (14), we have
Pi;1 = Pr CASE (IV) in RES1 occurs] + Pr CASE (III) or (V) in RES1 occur]

k=0 x + y) ?

fD;k ( )

(14)

= ( + )( x (1 ? ) ) y x + y ?

+ ( + )( + x ) ? x y x y

x y

(15)

Similarly ,
Ph;2 =

( + y) ( + x )( + y ) ?

x y

(16)

and
Pi;2 =

x + (17) ( + x y ( + x )( + y ) ? x y Consider an observation period 4t. During this period, there are o 4 t new call arrivals to a cell. These new calls generate h 4 t hando calls. From the homogeneous cell structure, the rate of hando calls leaving this cell equals the hando rate owing into the cell. Among the new and the hando call arrivals, (1 ? Pb )Pi;1 o 4 t new calls and (1 ? Pb )Pi;2 h 4 t hando calls will be forced to terminate due to interruption. Thus, the number of blocked calls at the cell is

x (1 ? ) y x )( + y ) ?

15
Pb o 4 t + Pb h 4 t + (1 ? Pb )Pi;1 o 4 t + (1 ? Pb )Pi;2 h 4 t, and the incompletion probability Pnc

is expressed as

Pnc =

Let tcn be the channel occupation time of a new call that is either complete in a cell or handed o to the next cell. The expected value E tcn ] is derived as follows.
E tcn ] = E tc where m > tc] + E m where for all k

= Pb + h Pb + (1 ? Pb )Pi;1 + h (1 ? Pb )Pi;2 o o

Pb o 4 t + Pb h 4 t + (1 ? Pb )Pi;1 o 4 t + (1 ? Pb )Pi;2 h 4 t o4t

(18)

0, tk < m < tk + xk+1 ] (19)

Where
E tc where m > tc] =
Z

1 1

tc =0 m =tc Z tc =0

tc e ?

m fc (tc )d m dtc

tc fc(tc )e? tc dtc

d = ? ds fc (s) s= (1 ? ) x ( + x)(2 + x + y ) = ( + )( + ) ? x y x y ]2 ? x ? ( + )((1 + )) ? x y x y

(20)

and
E m where for all k, tk < m < tk + xk+1] "

= = =

k=0
X

tk =0 m =tk xk+1 = m ?tk

Z

m fk (tk ) e? m x e? x xk+1 dxk+1 d m dtk m e?(

+

k=0
X

tk =0

fk (tk ) e x tk
Z

k=0

+ x

m =tk

x) m d m

dtk

+ ( + )2 x 1 " X = +
k=0

tk =0

tk fk (tk )e? tk dtk

! #

tk =0 x

fk (tk )e? tk dtk

k( ? dfdss) s=

+ ( + )2 x

fk ( )

16 ( = ( + +)( y )(2 +? x + y ) ]2 ? ( + )( + ) ? x + y) x y x y x y From (20) and (21), (19) is re-written as (1 ? ) x (1 ? ) x ( + x )(2 + x + y ) ? E tcn ] = 2 ( + x)( + y ) ? x y ] ( + x)( + y ) ? ( + ( + +)( y )(2 +?x + y ) ]2 ? ( + )( + ) ? + )
x y x y x y

(21)

x y x y

(22)

Suppose that a call successfully hands over i times. Let tc;i be the period between when the MS moves into cell i and when the call is complete (see Figure 1 (a)). The period tc;i is called the excess life of tc , which has the density function fc;i(tc;i) for all i 1. Let xj denote the period between when the call is handed o to cell i and when xj ends (see Figure 1 (a)). If tm;i is exponentially distributed, then from the excess life theorem, xj and xj have the same density function, and fc;i(tc;i) = fc(tc ). Let tch denote the channel occupation time of a hando call. It is apparent that
E tch ] = E tcn ]

(23)

(24) + x )( + x y] Let tcih denote the channel occupation time of a hando call which is disconnected due to the interruption. Similar to (23)
y(

Let tcin denote the channel occupation time of a new call which is either forced to terminate due to an interruption at cell 0 or is handed o to the next cell during the interruption period. The expected value E tcin ] is derived from Appendix B, which is expressed as 1 (1 ? ) x y (2 + x + y ) E tcin ] = ( + x)( + y ) ? x y ]2 ? ( + x )( + y ) ? x y x + ( + )( (2+ + ) x + y ) ]2 ?

x y) ?

x y

E tcih] = E tcin]

(25)

The net tra c to a cell consists of four parts: (i) The tra c of o (1 ? Pi;1 )E tcn] generated by non-forced-terminated new calls, (ii) the tra c of o Pi;1 E tcin] generated by forced-terminated new calls, (iii) the tra c of h (1 ? Pi;2 )E tch] generated by non-forced-terminated hando calls, (iv) the tra c of h Pi;2 E tcih] generated by forced-terminated hando calls, and = o (1 ? Pi;1 )E tcn ] + h (1 ? Pi;2 )E tch ] + o Pi;1 E tcin ] + h Pi;2 E tcih] (26)

17 With net tra c , the channel allocation for RES1 can be modeled by an M=G=C=C queue 6], and the blocking probability Pb is expressed as ( c =c!) (27) Pb = Pc i i=0 ( =i!) where c is the number of channels in a cell. The probability Pb can be obtained by assigning an initial value for h, and then iterating (1) and (27) until the h value converges. Finally, the call incompletion probability can be obtained from (18).

3.2 An Analytic Model for RES3

The analytic model for RES3 is similar to that for RES1 except for CASE(III). In this case, MS makes a call re-establishment request to cell 1. If cell 1 has an idle channel, then the call can be re-established in RES3. We have Pr CASE (III) in RES3 occurs] = Pr tD;k < m < tD;k + yk+1 and yk+1 zk+1 ] = Pr CASE (III) or (V) in RES1 occur] From (14), (28) is re-written as Pr CASE (III) in RES3 occurs] = ( + )( + x ) ? x y From (8) and (29), Ph;1 in RES3 is expressed as
x y

(28) (29)

) + ( + )( + x ) ? (30) = ( + )( (++ )y? x y x y x y x y Since the call is forced to terminate for CASEs (IV) and (V) in RES3, Pi;1 is expressed as follows.
Pi;1 = Pr CASE (IV) in RES3 occurs]

Ph;1 = Pr CASE (I) in RES3 occurs] + Pr CASE (III) in RES3 occurs]

+ Pr CASE (V) in RES3 occurs] From (14), Pr CASE (V) in RES3 occurs] = Pr tD;k < m < tD;k + yk+1 and yk+1 > zk+1 ] = (1 ? ) ( + )( + x ) ?
x y x y

(31)

(32)

18 From (11) and (32), (31) is re-written as x (1 ? ) y Pi;1 = ( + x)( + y ) ? x y + ( + Similarly, ( + y) Ph;2 = ( + x)( + y ) ? x y + ( and x (1 ? ) y Pi;2 = ( + x )( + y ) ? x y + ( Following the same derivation for (18), we obtain The expected value E tcn ] for RES3 is derived as follows.
o

(1 ? ) x )( + y ) ? x + x)( + y ) ? (1 ? ) x + x )( + y ) ?
o x

x y

(33) (34) (35) (36)

x y

x y

Pnc = Pb + h Pb + (1 ? Pb )Pi;1 + h (1 ? Pb )Pi;2 E tcn ] = E tc , where m > tc ] + E m , where for all k, tk < m < tk + xk+1 ] + E tD;k + yk+1, where for all k, tD;k < m < tD;k + yk+1

and yk+1 zk+1 ] Since = Pr yk+1 zk+1 ], we have

E tD;k + yk+1 where for all k, tD;k < m < tD;k + yk+1 and yk+1 = E tD;k + yk+1 where for all k, tD;k < m < tD;k + yk+1] zk+1 ]

(37)

(38)
zk+1 ]

From (48) and (52), we have

x = ( + )(x (2 +) ? + y ) ]2 ? ( + )( + x ) ? x + y x y y x y From (20), (21) and (39), we have

E tD;k + yk+1 where for all k, tD;k < m < tD;k + yk+1 and yk+1
x y]

(39)

( ? x = (1 (? +) x)( + x )(2 + x + ]2y ) ? ( + )((1 + )) ? x + y) ? x y x y x y ( + y )(2 + x + y ) ? + ( + )( + ) ? x y x y ]2 ( + x)( + y ) ? x y x (2 + x + y ) x + ( + )( + ) ? ? ( + )( + ) ? x y x y ]2 y x y x y]

E tcn ]

(40)

19 Similar to the derivation for (23), E tch ] = E tcn ]. E tcin ] is derived as follows.
E tcin ] = E tc0 where tc0 < m ] + E tD;k + yk+1, where for all k, tD;k < m < tD;k + yk+1 and yk+1 > zk+1 ]

(41)

Where
E tD;k + yk+1, where for all k, tD;k < m < tD;k + yk+1 and yk+1 > zk+1 ] = E tD;k + yk+1 where for all k, tD;k < m < tD;k + yk+1 ] (1 ? )

(42)

From (47), (48) and (52), (41) is re-written as (1 ? ) x y (2 + x + E tcin ] = ( + x)( + y ) ? x (1 x + ( + ?)( ) + (2 ) + x + ?

?
Similar to (23),

(1 ? ) x y ( + x )( + y ) ?

y) y ]2 y) x y ]2

? ( + )( + 1 ) ? x y

x y

x y]

(43)

E tcih ] = E tcin ]

(44) (45)

Following the same reasoning for (28), the net tra c for RES2 is = o (1 ? Pi;1 )E tcn ] + h (1 ? Pi;2 )E tch ] + o Pi;1 E tcin] + h Pi;2 E tcin ]

The probability Pb for RES3 can be obtained by the same interactive algorithm for RES1, and the call incompletion probability can obtained from (36).

3.3 Simulation Validation

The analytic models were validated by simulation experiments. In the simulation experiments, we considered 6 6 wrapped mesh cell structure. The simulation model follows the discrete event approach as in 7, 5]. RES2 is evaluated by simulation experiments without analytic modeling. In Figures 7 9, the solid curves represent the Pnc curves based on simulation, and the dashed curves are based on analysis. These gures indicate that both analysis and simulation are consistent.

20

12 11 10 9 8 7 P (%) 6 5 4 3 2 1 0
nc

... .. .. ... . . .. ...... . .. ...... .. ........ .. ......... .. ....... . .. ........ .. . .. . ......... ... .. . ....... . . . .. ..... ...... .. ..... .. ..... . ..... . ..... . . . .. ... .. ... . ... ...... .... . .. ..... .. ..... ..... ... .. .. ..... . .. . .. ..... .. .. . .. ... .. .... . .. ... ... .... ... .. ... . . .... .. ..... .. ... .. ..... ... . . . ...... .... .. ..... ..... . . . .. . ... ... . .. .... ..... .. ... ... .. .... .. ... . ... ... .. .. ... ... . ... . .... . .... ... .. ... .... .. . ... . . . .. .. .. .. ... ..... .. .... ... ... . . .... . .... ..... . .. . . . . . .. .. .... .... . .. ... .. ... .. ... ..... ... .. . .. .. . ... . ... ... .... .. . .. .. . . .... . .... .... .. .... .. ... .... ... . . ... . . . .. .. ... .. .... ... . ... . ... . . .. .. .. ... .. . .. ... .. . ... .. .... . .... . ... .. ... .. . .. .. . .... . .. .. ... .... .. ... . .. .. .. . . .. ... .. .... .. ... .. .. . ... . .. .. ... .. ... .. .. .. . ... .. .. ... ... .. ... .. .. .. . ... .. ... .. . .. ... .. .. .. . ... .. ... .. ... .. .. .. . . . . .. .. ... .. .. .. .. .. . .. .. .. ... .. .. ... .. .. .. . ... .. . ... .. ... .. .. . ... .. . ... .. .. ..... . . .. .. . .. . .. . . .. ..... .. .. .. .. . .. .. . ... .. ..... .. . .. .. . .. ..... .... .. .. .. . . .. . .. .. . .. .. .. . .. .. .. . .. . ... . .. .. .. .. . .. . . .. .. .. .. .. . .. . . . ... .. .. .. . .. . .. .. . . . .. .. . .. .. . .. .. . .. .. .. . .. . ... .. .. . .. . .. . .. ... .. .. . .. .. . .. ... . ... ... . .. .. ... . ... .. ... .. . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

: RES 1 : RES 2 : RES 3

1 2 3 4 5 6 7 8 9 10
z

Figure 7: The E ect of z (Solid: simulation; Dashed: analysis)

4 Performance Evaluation
This section compares the performance (speci cally Pnc ) of the three call re-establishment algorithms. In this comparison, input parameters such as , o and y are normalized by x . For example, if the expected time of a conversation period 1x = 30 seconds, then y = 10 x means that the expected interruption time is 3 seconds.

E ect of z : Figure 7 plots Pnc as a function of z , where = 0:2, = 0:7 x ,

= 2 x , and y = 10 x . Note that Pnc for the basic scheme (GSM without call re-establishment) is about 20% for all z values, which is not shown in Figure 7. The gure indicates that Pnc increases as z increases for all three RES algorithms (because the larger the z , the higher the probability that yi > zi , and thus the probability of call dropping). We observe that the Pnc for both RES1 and RES2 are almost identical. When z = x , RES3 results in 35% improvement over RES1 and RES2. When z = 10 x , the improvement of RES3 becomes insigni cant. For a
o

21
5.0 4.5 4.0 3.5
. .... .. .. .... ..... ..... ..... ...... ...... ...... ...... ........ ....... ...... . ...... ............ .. ........ .. ......... ............. . ...................................... ................................... .

(%)

Pnc

3.0 2.5 2.0 1.5 1.0 0.5 0.0 2.0 : RES 1 : RES 2 : RES 3 2.5 (a) The E ect of
o
.. .... . ... .. .. ..... ..... ..... ..... ..... ...... ..... ....... ...... ....... ....... . ...... ...... ........... ........... . ....................................... ...................................... .

3.0
x o

32 30 28 26 24 22 20 Pnc 18 (%) 16 14 12 10 8 6 4 2 0

. . .. .. . . .. .. .. . .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. . .. . .. . .. .. .. . . .. .. .. . .. .. .. .. .. .. .. .. .. .. .. .. .. . . . .. .. .. .. .. .. .. .. . .. .. . .. . .. .. .. .. .. .. .. .. .. .. . . .. .. .. . .. .. .. . .. .. .. .. . . .. .. .. . .. .. .. .. .. .. . . . . .. .. .. .. .. .. . . .. .. .. . .. .. .. .. .. .. .. .. .. .. .. .. ..

: RES 1 : RES 2 : RES 3 : basic

.. .. .. . .. . ....... .. ........ . .. .. ...... .. .. . .. ..... .. . ... . . ..... .. .. ..... .. ........ .. ........ .. . ... . ... . ........... ........... . .. . ....... .... .. ...... ....... .. .. ........ .. .. .... ... ...... ....... .. ...... ........ . ........ . ...... ........ . . .. ... ........ . ...... .. ..... ......... . ....... ......... . ...... .. ...... ......... ......... ....... ....... ........... .......... ......... ......... . ... .. ........ ....... ....... ...... ........... .......... .......... ........... ............ ...........

0.1

0.2 (b) The E ect of

0.3

Figure 8: The E ects of o and (Solid: simulation; Dashed: analysis) large z , an interrupted call is likely to be forced to terminate due to zi expiration, no matter the MS moves into a new cell or not. In this case, three call re-establishment algorithms have the similar performance. Thus, we conclude that RES3 signi cantly outperforms RES1 and RES2 when z is small.

E ect of o: Figure 8 (a) plots Pnc as a function of o , where = 0:2, = 0:7 x ,

= 10 x , and z = 1:0 x . It is apparent that Pnc increases as o increases. The gure indicates that RES1 and RES2 have the same performance. When o = 2:0 x , RES3 has 35% improvement over RES1 and RES2. When o = 3:0 x , the improvement is 30:3%. Thus, the improvement of RES3 over RES1 and RES2 becomes more signi cant as o decreases.
y

E ect of : Figure 8 (b) plots Pnc against , where = 0:7 x ,

= 2 x , y = 10 x , and z = 1:0 x . As increases, a call is more likely to be interrupted, and Pnc is larger. When = 0:1, RES1 and RES2 outperform the GSM basic scheme by 84%. When = 0:3, the
o

22
4.0 3.5 3.0 2.5 (%) 2.0 1.5 1.0 0.5 0.0 0.3 : RES 1 : RES 2 : RES 3 0.5 (a) The E ect of
=
x
. . .. . . . . . . .. . . . .. . .. . .. . .. . ... .. ... .. .... .. ..... .... . . .. .. . ...... .. . ....... .. . .. .... ....... .. ........ .. .. ........ ........ . . .. .. . . ......... ......... .. ... .......... .......... ... .. ..... ... ..... ... .. ........... ... ........... ... . .... .. .. .. ..... ..... ..... ...... .. . .. ..... .... .. . ............. .. . ... ... .. ............... .. ............... .. . .. . .. . .. . .. ........ ..... .. .............. . . .. .............. . . ................. .. ....... . ............ . . .. .... . .....

Pnc

. . . .................................................................................................... ................................................................................................... ... ... ... ... ... ... ... ... ... ... ... ... .. .. .. .. .. .. .. ...

0.7

6.5 6.0 5.5 5.0 4.5 4.0 Pnc 3.5 (%) 3.0 2.5 2.0 1.5 1.0 0.5 0.0

. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . .. . .. . . . . . . .. . .. . . . .. . .. .. . . .. .. . . .. . . .. . . .. . . . . .. . . . . .. . . .. . . .. . . . . .. . .. . . .. .. .. . .. .. .. . . .. . . .. . .. . . .. . .. .. .. .. .. .. . .. . . .. . .. . . .. .. . .. .. .. . . .. . .. . . .. . .. .. .. .. .. .. .. .. .. .. .. ... . .. .. .. .. .. .. . .. .. . .. . .. ... . .. ... . ... . .. .. . .. .. ... .. ... .. ... . .. .. . ... .. ... .. .... . ... . .. .. . . ... .. .. .. .... . .. ... .. .. .... .. .. .... . .. ..... . ... ... .. ..... .. .. .. ..... ... .. ..... ... ... ... .. ..... ..... .. .. .... .. ... ... .. ... .. ... ... ... .. .. ... ... ... ... .... ... ... .... . .. ..... .... .. ..... .. ..... ..... .... . ... ..... ..... ...... ..... . ... .. .

: RES 1

: RES 2 : RES 3 10
y

15
=
x y

20

(b) The E ect of

Figure 9: The E ects of and y (Solid: simulation; Dashed: analysis) improvement is 81%. Thus, RES1 and RES2 signi cantly outperform the basic scheme for all values. When = 0:1, RES3 outperforms RES1 and RES2 by 37:2%. When = 0:3, RES3 outperforms RES1 and RES2 by 33:8%.

E ect of : Figure 9 (a) illustrates the Pnc performance for various mobility rate , where = 0:2,
o = 2 x, y

= 10 x , and z = 1:0 x . Figure 9 (a) shows that Pnc increases as increases for RES1 and RES2. For a large , an MS is likely to move to a new cell during an interruption period. For RES3, Pnc is not e ected by because the interrupted calls can be re-established when the MS moves to a new cell. With = 0:3 x , RES3 outperforms RES1 and RES2 by 20%, and with = 0:7 x , RES3 outperforms RES1 and RES2 by 35%. Thus, the improvement of RES3 over RES1 and RES2 becomes signi cant as increases. = 2 x, and z = 1 x . This gure indicates that Pnc decreases as y increases. Note that increasing
o

E ect of y : Figure 9 (b) plots Pnc as a function of y , where = 0:2, = 0:7 x ,

23 has the same e ect as decreasing z . When y = 5 x , RES3 outperforms RES1 and RES2 by 31:8%. When y = 20 x , RES3 outperforms RES1 and RES2 by 37:2%. Thus, the improvement of RES3 over RES1 and RES2 becomes signi cant as y increases.
y

5 Conclusion
We proposed analytic models to investigate the performance for GSM call re-establishment service. The call re-establishment algorithms under evaluation are RES1 (the radio channel is reserved during interruption; the call is not re-established if the MS moves into a new cell), RES2 (the radio channel is not reserved during interruption; the call is not re-established if the MS moves into a new cell) and RES3 (the radio channel is reserved during interruption; the call is re-established if the MS moves into a new cell). The analytic models are validated by simulation experiments. Our study indicated that call re-establishment can signi cant reduce the call incompletion probability for interrupted calls (more than 80% improvement was observed in most cases of this paper). Furthermore, we observed that both RES1 and RES2 have the same performance, and RES3 may signi cantly outperform RES1 and RES2, especially for long zi , small call arrival rate o and large mobility rate .

References
1] EIA/TIA. Cellular Radio-telecommunications Intersystem operations: Intersystem hando . Technical Report IS-41.2-B, EIA/TIA, 1991. 2] ETSI/TC. Handover Procedures. Technical Report Recommendation GSM 03.09, ETSI, 1993. 3] ETSI/TC. Mobile-services Switching Centre-Base Station System (MSC-BSS) interface Layer 3 speci cation. Technical Report Recommendation GSM 08.08, ETSI, 1994. 4] ETSI/TC. DMS-MSC (GSM08P) Product Computing Module Load Release Vol 1 of 2. Technical Report Recommendation GSM 04.02, ETSI, 1995. 5] J.-Y. Jeng and Lin, Y.-B. Equal Resource Sharing Scheduling for PCS Data Services. ACM Wireless Networks. 6] Kleinrock, L. Queueing Systems: Volume I { Theory. New York: Wiley, 1976.

24 7] Lai, W.R., And Lin, Y.-B. Resource Planning for Wireless PBX Systems. Intl. J. Wireless Information Networks. . 8] Lin, Y.-B. Mobility Management for Cellular Telephony Networks. IEEE Parallel & Distributed Technology, 4(4):65{73, November 1996. 9] Lin, Y.-B., Mohan, S., and Noerpel, A. Queueing Priority Channel Assignment Strategies for Hando and Initial Access for a PCS Network. IEEE Trans. Veh. Technol., 43(3):704{712, 1994. 10] Mouly, M. And Pautet, M.-B. The GSM System for Mobile Communications. M.Mouly, 49 rue Louise Bruneau, Palaiseau, France, 1992.

A Input Parameters and Output Measures

The input parameters and output measures used in this paper are listed as follows. The input parameters include
o:

the new call arrival rate to a cell the mean conversation period time the mean interruption period time

xi : the ith conversation period

1

x:

yi : the ith interruption period

1

y:

: the probability that at the end of a conversation period, the radio channel is interrupted two events occurs: (i) the interruption timer expires, (ii) the remote party hangs up the phone
tm;i : the residence time of the MS at cell i
1

zi : the ith interval between when the interruption begins and when the rst of the following

: the mean MS residence time the period between the arrival of the call and when the MS moves out of cell 0

m:

25 The output measures include

h:

the hando call arrival rate to the cell

Pb : the probability that all channels are busy when a call (either a new call or a hando call)

arrives

Ph;1 (Ph;2 ): the probability that a connected new (hando ) call at the cell will hando to the

next cell

Pi;1 (Pi;2 ): the probability that a connected new (hando ) call at the cell will be disconnected

due to interruption

Pnc : the probability that a call is not completed

: the probability that a call is re-established after interruption

tc : the call holding time of a complete call without considering the hando e ect tk : the holding time for the rst k cycles of a call tcn : the channel occupation time of a new call that is either complete in a cell or handed o

to the next cell

tch: the channel occupation time of a hando call tcin : the channel occupation time of a new call which is either forced to terminate due to an

interruption at cell 0 or is handed o to the next cell during the interruption period tion

tcih: the channel occupation time of a hando call which is disconnected due to the interruptc;i: the period between when the MS moves into cell i and when the call is complete

: the net tra c to a cell

26

B Derivation of E t ]
cin

This appendix derives the expected value E tcin ] as follows.

E tcin] = E tc0 where tc0 < m ] + E tD;k + yk+1 where for all k, tD;k < m < tD;k + yk+1] = E tc0 where tc0 < m ] + E tD;k where for all k, tD;k < m < tD;k + yk+1 ]

+ E yk+1 where for all k, tD;k < m < tD;k + yk+1] (46) Following the same derivation for (20), we have
E tc0 where tc0 < m ] =

and = =
X

? dfc0 (s) ds s ) + = ( (1 ? )( x+y (2 ) ? x + y]) + x y x y 1 ? ( + )( + ) ?

=

x y

(47)

E tD;k where for all k, tD;k < m < tD;k + yk+1 ] "
1
Z

k=0

tD;k =0 m =tD;k yk+1 = m ?tD;k

!( X

tD;k fD;k (tD;k ) e? m y e? y yk+1 dyk+1 d m dtD;k

#)

"

Since

k=0 " tD;k =0 #) 1 X fD;k (s) = + ? ds y s= k=0 x = ( + )( (2+ + ) x + y ) ]2 x y ? x y x ? ( + ) ( + )( + ) ? y x y x y]

!( Z

+ y

tD;k fD;k (tD;k )e? tD;k dtD;k

(48)

1
yk+1 = m ?tD;k

yk+1 y e? y yk+1 dyk+1

y ( m ?tD;k ) ?

= ( m ? tD;k )e?

1 e?
y

y ( m ?tD;k )

27 We have
E yk+1 where for all k, tD;k < m < tD;k + yk+1] "

?
Where
D =

k=0 " tD;k =0 m =tD;k

fD;k (tD;k ) e? m ( m ? tD;k )e? y ( m ?tD;k ) d m dtD;k

#

=D?E
X

k=0

tD;k =0 m =tD;k

1 fD;k (tD;k ) e? m e? y ( m ?tD;k ) d m dtD;k

y

(49)
#

" Z

1 1

k=0 ( tD;k =0 m =tD;k

X

fD;k (tD;k ) e? m ( m ? tD;k )e? y ( m ?tD;k ) d m dtD;k

" #

= =

k=0 "

tD;k =0

fD;k (tD;k )
X

#"

( + y )2
#

e?(

y )tD;k dtD;k

x = ( + ) ( + )( + ) ? y x y

( + y )2

k=0

fD;k ( )
x y]

(50) 1 e?
y
!

and
E =

X
=0 "

"

1
X

tD;k =0 m =tD;k
#(

fD;k (tD;k ) e?
Z

y ( m ?tD;k ) d m dt D;k
#)

=
"

"

y( y( y(

+ y)

1
#

#"

= =

+ y ) k=0 fD;k ( )
x y) ? x y]

1 X

k=0

tD;k =0

fD;k (tD;k )e? tD;k dtD;k

+ x )( + From (50) and (51), (49) is re-written as

(51)

E yk+1 where for all k, tD;k < m < tD;k + yk+1 ]

? ( + )( + y x

x = ( + ) ( + )( + ) ? y x y x y) ?

x y]

x y]

(52)