Bearer Reservation with Preemption for Voice Call Continuity

Bearer Reservation with Preemption for

Voice Call Continuity

Yi-Bing Lin, Fellow, IEEE, Meng-Hsun Tsai, Student Member, IEEE, Hui-Wen Dai, and Yuan-Kai Chen

Abstract—In Universal Mobile Telecommunications System

(UMTS), the core network consists of two service domains: the

circuit-switched (CS) and the packet-switched (PS) domains. A

UMTS handset can initiate or receive a call in either the CS or the PS domain. During the call, the user may switch from one domain to another. The switching overhead is an important concern of domain transfer. In this paper, we propose the Bearer Reservation with Preemption (BRP) scheme to support fast domain transfer, and present both analytic model and simulation experiments to investigate the BRP performance. Our study indicates that when user behavior is irregular (i.e., either the variance of the domain residence times or the variance of the call holding times is large), the advantage of the BRP scheme becomes significant.

Index Terms—Domain transfer, IP multimedia core network

subsystem (IMS), universal mobile telecommunications system (UMTS), voice call continuity (VCC).

NOMENCLATURE

pf The probability that there is no available resource

when the UE switches back to the CS domain.

pr The probability that the reserved CS bearer has

been preempted, but the resource becomes available when the UE switches back to the CS domain.

pn The probability that the reserved CS bearer is

not preempted.

θP 2C The percentage of re-connection overhead saved

by our approach for PS-to-CS domain transfer as compared with the 3GPP approach.

θC2P The percentage of re-connection overhead saved

by our approach for CS-to-PS domain transfer as compared with the 3GPP approach.

λ Inter-VCC CS call arrival rate. tc The call holding time.

1/μ The expected value of tc.

Vc The variance for the tc distribution.

Manuscript received August 13, 2008; revised September 23, 2008 and November 16, 2008; accepted January 18, 2009. The associate editor coor-dinating the review of this paper and approving it for publication was F.-N. Pavlidou.

Y.-B. Lin is with the Department of Computer Science, National Chiao Tung University, Taiwan, ROC. He is also with the Institute of In-formation Science, Academia Sinica, Nankang, Taipei, Taiwan (e-mail: liny@csie.nctu.edu.tw).

M.-H. Tsai and H.-W. Dai are with the Department of Computer Sci-ence, National Chiao Tung University, Taiwan, ROC (e-mail: {tsaimh, hw-dai}@cs.nctu.edu.tw).

Y.-K. Chen is with Chunghwa Telecom, Taipei, Taiwan, ROC (e-mail: ykchen@cht.com.tw).

Y.-B. Lin’s work was sponsored in part by NSC 97-2221-E-009-143-MY3, NSC 97-2219-E-009-016, NSC 96-2752-E-009-005-PAE, NSC 96-2219-E-009-019, NSC 96-2221-E-009-020, Intel, Chunghwa Telecom, ITRI/NCTU Joint Research Center and MoE ATU. M.-H. Tsai’s work was supported by the MediaTek Fellowship.

Digital Object Identifier 10.1109/TWC.2009.081085

td The period that a VCC call resides at a domain

before it is switched to another domain. 1/δ The expected value of td.

Vd The variance for the td distribution.

ts The period that a high-priority (low-priority) call

utilizes (reserves) a channel at the MSC before it is switched to another domain or is completed. 1/η The expected value of ts.

C The capacity (number of channels) of the MSC. K0 The number of calls in the MSC seen by a

low-priority call arrival L.

Ki The number of high-priority calls in the MSC

when the i-th high-priority call arrives.

πk The steady-state probability that there are k calls

in the MSC.

th The inter-arrival time between the i-th and the

(i + 1)-th high-priority call arrivals.

p(m,n) The one-step transition probability from state

Ki= mto state Ki+1= n.

p(l)_(m,n) The probability that the stochastic process moves

from state m to state n with exact l steps. I. INTRODUCTION

U

circuit-switched (CS) and the packet-circuit-switched (PS) domains [1]. IP Multimedia Core Network Subsystem (IMS) is developed

in the PS domain to provide multimedia services [2]. In existing commercial operation, the CS domain provides full service coverage with limited bandwidth. On the other hand, the PS domain typically provides zonal coverage with larger bandwidth and cheaper services. Therefore, when the PS connection is available, the UE will switch to the PS domain. When the PS connection is not available, the UE switches to the CS domain. A UMTS handset can attach to the CS and the PS domains individually or simultaneously, and initiate or receive a call in either domain. The user may switch from one domain to another during the call. In order to maintain call continuation, the connection in the old domain is released, and a connection is established in the new domain. This process is called domain transfer. The technique to transfer a voice call between the CS and the PS domains is called

Voice Call Continuity (VCC) [3].

Figure 1 illustrates a simplified UMTS network architecture that accommodates VCC [3], [4]. This architecture consists of the UMTS Terrestrial Radio Access Network (UTRAN) 1536-1276/09$25.00 c 2009 IEEE

(7) MGCF (6) S-CSCF (5) GGSN (9) MGW (2) UE (3) MSC (d) PS Domain

(e) IP Multimedia Core Network Subsystem (IMS) (c) CS Domain (1) HSS (8) VCC AS (b) WiMAX / WiFi Network (10) Callee (f) Destination Network (a) UTRAN (4) PDG

Fig. 1. The UMTS network architecture (dashed lines: signaling; solid lines: signaling/data).

(Figure 1 (a)), the Worldwide Interoperability for Microwave Access (WiMAX) or the Wireless Fidelity (WiFi) network (Figure 1 (b)), the CS domain (Figure 1 (c)), the PS domain, i.e., the General Packet Radio Service (GPRS) network (Fig-ure 1 (d)), and the IMS network (Fig(Fig-ure 1 (e)). In this architec-ture, Home Subscriber Server (HSS; Figure 1 (1)) is the master database containing all user-related subscription information, which supports mobility management of the users. A mobile user utilizes a User Equipment (UE; Figure 1 (2)) to access CS and PS services. In the CS domain, the Mobile Switching

Center (MSC; Figure 1 (3)) is responsible for call control,

including the processing of user data and control signals. In the PS domain, the WiMAX/WiFi network connects to the GPRS network through the Packet Data Gateways (PDGs; Figure 1 (4)); the GPRS network connects to the IMS network through the Gateway GPRS Support Nodes (GGSNs; Figure 1 (5)). In the IMS, the transport of user data is separated from that for control signals. The IMS signaling is carried out by the Serving Call Session Control Function (S-CSCF; Figure 1 (6)), the Media Gateway Control Function (MGCF; Figure 1 (7)) and the VCC Application Server (VCC AS; Figure 1 (8)). The IMS user data traffic is transported through the Media

Gateways (MGWs; Figure 1 (9)) controlled by the MGCF.

When the UE makes a call to a call party (Figure 1 (10)) in a different network (Figure 1 (f)), the call is first routed to the MGW. Then the MGW routes the call to the callee through the destination network.

When the UE switches from one domain to another during a voice call, the call is domain-transferred by the VCC AS. A major problem of domain transfer is a large number of message exchanges and resource reservation that result in long switching latency. To support fast domain transfer, we propose the Bearer Reservation with Preemption (BRP) scheme. We present an analytic model for the BRP scheme. The proposed analytic model is used to validate against the simulation model. Then we conduct simulation experiments to investigate the BRP performance.

II. VCC CALLSETUP ANDDOMAINTRANSFER This section describes the VCC call setup and domain transfer procedures defined in 3GPP [5].

A. VCC Call Setup

To support domain transfer, the VCC AS is inserted into the signal path of the call. This is achieved by adding some VCC service triggering criteria (called initial filter criteria or iFC [6]) into the UE’s profile in the HSS. When a UE registers to the IMS, the S-CSCF downloads these iFC of the UE from the HSS. When a call arrives at the S-CSCF, the call is evaluated against the iFC. If the VCC service criteria are matched, the call is routed to the VCC AS for further processing. The VCC call path is partitioned into two segments: the UE-MGW segment and the MGW-Callee segment. When the UE moves from one domain to another during a call, the UE-MGW segment is switched, and the MGW-Callee segment remains unchanged.

For the reader’s benefit, the VCC call origination in the PS domain is described in Appendix A. The reader is referred to [5] for VCC call origination in the CS domain and VCC call termination in both domains.

B. 3GPP Domain Transfer

In the CS domain, VCC service control is provided through the Customized Applications for Mobile Network Enhanced

Logic (CAMEL) [7], where the VCC service logic is

imple-mented in the VCC AS. This subsection describes how 3GPP domain transfer works. Detailed description of 3GPP domain transfer is given in Appendix B.

When a UE decides to transfer its VCC call from the PS domain to the CS domain, a new call is initiated in the CS domain with a specific called number VCC Domain Transfer

Number (VDN). This number is then translated into a routable

number IP Multimedia Routing Number (IMRN) through the VCC service logic. The IMRN is used to route the call from the CS domain to the VCC AS in the PS domain. When the call setup signal arrives at the VCC AS, the VCC AS updates the UE-MGW segment. Details of PS-to-CS domain transfer are given in Steps B.1-B.16 in Figure 10.

After the call has been successfully switched to the CS domain, the UE may decide to switch the call back to the PS domain again. To trigger CS-to-PS domain transfer, the UE initiates a new call in the PS domain with a specific called identity VCC Domain Transfer Uniform Resource Identifier (VDI). When the new call arrives at the VCC AS, the VCC AS updates the UE-MGW segment for the UE. Details of CS-to-PS domain transfer are given in Steps C.1-C.9 in Figure 11.

III. BRP DOMAINTRANSFER

In the 3GPP CS-to-PS domain transfer procedure, the CS bearer of the UE-MGW segment is released after the IP bearer is established. If the UE moves back to the CS domain again, the released CS bearer must be re-established. Such bearer re-establishment contributes extra overload to the domain transfer. To speed up the subsequent switchings, we may not release the CS bearer at the CS-to-PS domain transfer, and

UE MSC MGCF CS IMS IP Bearer CS Bearer CS Bearer IP Bearer IP Bearer CS Bearer CS Bearer S-CSCF MGW VCC AS Callee

C*.6 ISUP FAR (low priority) C*.7 ISUP FAA

Initiation of IP bearer establishment (see Steps C.1-C.5 in 3GPP procedure)

Completion of IP bearer establishment (see Steps C.6-C.9 in 3GPP procedure)

Fig. 2. CS-to-PS domain transfer with IP bearer establishment (BRP).

postpone the bearer release until the VCC call is complete. If the user moves back from the PS domain to the CS domain, the bearer re-establishment is eliminated. Same argument applies to the IP bearer re-establishment.

Based on the above intuition, we propose the Bearer Reser-vation with Preemption (BRP) scheme that speeds up the domain transfer process. The BRP scheme utilizes enhanced

Multi-Level Precedence and Pre-emption (eMLPP) service [8]

and Multimedia Priority Service (MPS) [9] to provide reser-vation and preemption of CS and IP bearers. In BRP, two eMLPP priority levels are defined: the high priority and the

low priority. When there is no available channel at the MSC,

a call arrival with high priority can preempt a call with low priority, i.e., the high priority call is established, and the low priority call is force-terminated.

In BRP, a VCC call before domain transfer is set up with high priority. When the UE switches this call from the CS domain to the PS domain, instead of releasing the CS bearer in the UE-MGW segment, this CS bearer is reserved with low priority. When the UE switches the call back to the CS domain, the domain transfer process simply raises the priority level of the reserved CS bearer to high priority. If the reserved CS bearer with low priority is preempted (and the preempted channel is used by an incoming high-priority call), the CS bearer is released. In this case, the VCC call is not terminated because the IP bearer is used. When the call is switched back to the CS domain, the CS bearer needs to be re-established.

A. CS-to-PS Domain Transfer in the BRP Scheme

Figure 2 illustrates the BRP message flow for CS-to-PS domain transfer with IP bearer establishment with the following steps:

Steps C.1-C.5 Same steps as in the 3GPP CS-to-PS domain

transfer procedure (see Figure 11 in Appendix B) initiate the establishment of the IP bearer in the UE-MGW segment.

Step C*.6 The MGCF lowers the priority level for the CS

bearer, and sends an ISDN User Part (ISUP) Facility Re-quest (FAR) message with the parameter “low priority” to the MSC. UE MGW CS MGCF IP Bearer IP Bearer B*.3 H.248 Move CS Bearer CS Bearer S-CSCF

B*.1 CM SERVICE REQUEST (high priority) B*.2 ISUP FAR (high priority)

B*.5 ISUP FAA B*.6 CM SERVICE ACCEPT IP Bearer CS Bearer CS Bearer IP Bearer MSC IMS VCC AS Callee B*.4 H.248 Reply

Fig. 3. PS-to-CS domain transfer without CS bearer establishment (BRP).

Step C*.7 According to the priority level indicated in the

received ISUP FAR message, the MSC lowers the priority level for the CS bearer, and sends an ISUP Facility Accepted (FAA) message to the MGCF.

Steps C.6-C.9 Same steps as in the 3GPP CS-to-PS domain

transfer procedure (see Figure 11 in Appendix B) ex-change the Session Initiation Protocol (SIP) 200 OK and ACK messages to complete the IP bearer establishment in the UE-MGW segment.

By adding two messages (Steps C*.6 and C*.7), the BRP scheme eliminates eleven messages (Steps C.10-C.18) in Fig-ure 11. Therefore, the message exchange cost is reduced by 36%. If the IP bearer has been reserved and not preempted before the domain transfer occurs (not shown in this paper), the message exchange cost is reduced by 68%. Also note that after the transfer, the CS radio link to the UE may be disconnected, but the CS bearer at the MSC is still maintained. This idea is similar to the “always on” concept of GPRS [4].

B. PS-to-CS Domain Transfer in the BRP Scheme

After the call has been successfully switched to the PS domain, the UE may decide to switch the call back to the CS domain again. If the reserved CS bearer has not been preempted, the UE does not need to initiate a new call for establishing the CS bearer in the UE-MGW segment. Instead, the UE only needs to raise the priority level of the reserved CS bearer to high priority. Also, unlike the procedure in Figure 10, the IP bearer is not released. Therefore, IP bearer needs not be re-established when the call switches back to the PS domain. Figure 3 illustrates the BRP message flow for PS-to-CS domain transfer without CS bearer establishment with the following steps:

Step B*.1 The UE sends a Call Management (CM)

SER-VICE REQUEST message to the MSC to raise the priority level of the CS bearer in the UE-MGW segment.

Step B*.2 The MSC raises the priority level for the CS bearer.

Then the MSC sends an ISUP FAR message with the parameter “high priority” to the MGCF.

Steps B*.3 and B*.4 The MGCF raises the CS bearer’s

exchanges H.248 Move and Reply messages with the MGW to switch the UE-MGW segment from the PS bearer to the reserved CS bearer.

Steps B*.5 and B*.6 To complete this priority update, the

MGCF sends an ISUP FAA message to the MSC. Then the MSC sends a CM SERVICE ACCEPT message to the UE to indicate successful priority update of the CS bearer. At this point, the UE-MGW segment is switched from the IP bearer to the CS bearer.

In the BRP scheme, six messages (Steps B*.1-B*.6) modify the priorities of the CS and the PS bearers. On the other hand, the 3GPP procedure in Figure 10 exchanges twenty-six messages (Steps B.1-B.18) to establish a new CS bearer, and release the old IP bearer. Therefore, the message exchange overhead is reduced by 77%. If the CS bearer has been preempted before the call is switched back to the CS domain (not shown in this paper), the message exchange cost is reduced by 15.4%.

IV. ANALYTICMODELING OFBRP

This section proposes an analytic model to study the per-formance of the BRP scheme. Without loss of generality, we investigate the BRP performance in the CS domain when the new calls arrive at the MSC are VCC calls (i.e., we do not consider non-VCC calls). Similar conclusions also apply to the PS domain, and the details are omitted. Suppose that a UE has switched its VCC call from the CS to the PS domain. In the BRP scheme, the CS bearer is reserved with low priority. When the UE switches from the PS domain back to the CS domain at time τ, there are three possibilities:

Case I) Before the UE switches back to the CS domain, the reserved CS bearer has been preempted, and there is no available resource (i.e., no channel in the MSC) at time τ. The call is force-terminated. Let pf be the

probability that this case occurs.

Case II) The reserved CS bearer has been preempted before

τ, but the resource becomes available when the call is

switched back to the CS domain at τ. The CS bearer is re-established at the PS-to-CS domain transfer. In this case, only 15.4% message overhead is saved by our approach. Let prbe the probability of this case.

Case III) The reserved CS bearer is not preempted. The UE only needs to raise the priority level of the reserved CS bearer to high priority by executing the procedure in Figure 3. In this case, our approach saves 77% message overhead as compared with the 3GPP approach. The probability of this case is pn.

It is clear that pf is the same for both the 3GPP and the BRP

schemes. Probabilities prand pnare used to actually compute

the overhead saved by our approach. We use θP 2C (or θC2P)

to represent the percentage of re-connection overhead saved by our approach as compared with the 3GPP approach for PS-to-CS (or CS-to-PS) domain transfer. Specifically,

θP 2C = 15.4%× pr pn+ pr + 77%× pn pn+ pr , θC2P = 36%× pr pn+ pr + 68%× pn pn+ pr (1) 0 1 ... k-1 k k+1 ... C-1 C k k k k C C

Fig. 4. State transition rate diagram for the BRP scheme.

For the illustration purpose, we only consider θP 2C in this

paper. Similar conclusions also apply to θC2P. The following

input parameters are considered in this study:

• The arrivals of new VCC CS calls are a Poisson stream

with rate λ.

• The call holding time is a random variable tc with mean

1/μand variance Vc.

• A VCC call resides at a domain for a period td before

it is switched to another domain. Let td be a random

variable with mean 1/δ and variance Vd.

• A high-priority (low-priority) call utilizes (reserves) a

channel at the MSC for a sojourn time ts before it is

switched to another domain or is completed. It is clear that ts= min(tc, td). Let ts be a random variable with

mean 1/η.

In the analytic model, we assume that tc and td are

exponentially distributed. Therefore, ts is also exponentially

distributed. The above exponential assumptions result in mean value analysis [10] (this exponential assumption will be re-laxed in simulation experiments). We conduct the mean value analysis to provide understanding on the “trend” of perfor-mance. Furthermore, this exponential-based analytic model is used to validate the simulation model. Then the validated simulation model will relax the exponential assumptions to accommodate more general (and therefore more practical) scenarios.

The BRP scheme is modeled by a stochastic process. We first derive the number of channels occupied (either used or reserved) by the calls at the MSC. Let C be the capacity (number of channels) of the MSC. Figure 4 illustrates the state transition rate diagram of the stochastic process where state k denotes that there are k calls (either high-priority or low-priority) in the MSC. We note that during the call holding time tc of a VCC call, the call may be switched between the

CS and the PS domains, and the channel at the MSC is always occupied by the call. Therefore, the stochastic process can be modeled by a simple M/M/C/C queue with the parameters λ and μ. Let πk denote the steady-state probability that there

are k calls in the MSC. From the standard technique [11], we have πk= π0  λk (k!)μk  , π0=  1 + C  j=1 λj (j!)μj −1 for0 ≤ k ≤ C (2)

After a VCC call L switches from the CS to the PS domain, it becomes a low-priority call at the MSC. Figure 5 illustrates the timing diagram during L’s sojourn time ts, where L arrives

at the MSC at τ0 (i.e., it transfers from the high to the low priority at τ0), and leaves the MSC at τ7(i.e., it completes or transfers back with high priority). There are two high-priority call arrivals at the MSC at τ2and τ5, and there are four

high-time 0 1 2 L’s Arrival t 4 t L’s Departure 3 K0= 3 t 5 6 7 High-priority Call Departure High-priority Call Arrival High-priority Call Arrival K1 = 3 K2 = 2 ts* High-priority Call Departure High-priority

Call Departure Call DepartureHigh-priority

Fig. 5. Events that may occur inL’s sojourn time.

priority call departures at τ1, τ3, τ4 and τ6. When k = C, a high-priority call arrival will preempt an existing low-priority call. The order of preemption is based on the

Last-Come-First-Preempted scheme (i.e., the last call arrival will be preempted

first) [12]. Let ¯pn = 1− pn be the probability that a

low-priority call L is preempted during its sojourn time. Let K0 be the number of calls in the MSC seen by L at domain transfer (i.e., at τ0), where L is not included in K0. Note that from L’s viewpoint, these K0 calls are “high-priority” (i.e., none of them will be preempted before L is preempted). Since

td is exponentially distributed, Pr[K0 = m] can be derived based on the “flow rate” concept [13]. Under this concept, (m + 1)δπm+1 represents the number of calls that leave the

MSC through domain transfers in a time unit when the system is at state m + 1 (where δ = 1/E[td]is the domain transfer

rate for a call; see page 59 in [13]), and

C_−1

j=0

(j + 1)δπj+1

represents the total number of domain-transferred calls that leave the MSC in a time unit. From (2) and the “flow rate” concept, Pr[K0= m]is derived as Pr[K0= m] = (m + 1)δπm+1 C−1 j=0(j + 1)δπj+1 =  λm+1 (m!)μm+1  ⎡ ⎣C−1 j=0 λj+1 (j!)μj+1 ⎤ ⎦ −1 (3) After τ0, L can only be preempted by a high-priority call arrival or by a PS-to-CS domain transfer. For simplicity, we do not consider PS-to-CS domain transfers, and we simply observe the moments when a high-priority call arrives. After

τ0, for i≥ 1, let Kibe the number of high-priority calls in the

MSC (from L’s viewpoint, the low-priority calls counted in

K0 are also included in these “high-priority” calls) when the

i-th high-priority call arrives, where this high-priority call is

included in Ki. In Figure 5, if K0= 3at τ0, then K1= 3at τ2 (because there is one high-priority call departure in [τ0, τ2]), and K2 = 2 at τ5 (because there are two high-priority call departures in [τ2, τ5]).

For the i-th high-priority call arrival (i≥ 0; by convention, Lrepresents the 0-th call arrival), let p(m,n) be the one-step transition probability from state Ki = mto state Ki+1 = n.

That is, p(m,n)is the probability that there are m−n+1 high-priority call departures during the inter-arrival time thbetween

the i-th and the (i+1)-th high-priority call arrivals. Therefore,

p(C−1,C) is the probability that when Ki= C− 1, L will be

preempted by the (i + 1)-th high-priority call arrival. Note

that for 0≤ n ≤ C, p(C,n)= 0 because L has already been preempted by the i-th high-priority call arrival. In addition, for 0 ≤ m ≤ C, p(m,0) = 0 because the (i + 1)-th high-priority call arrival is included in Ki+1, and Ki+1 is always

larger than 0. Also, p(m,n) = 0 if n > m + 1 (because the (i+1)-th high-priority call is the only new call that contributes to Ki+1). In Figure 5, let t∗s be the excess life (residual life)

of ts upon a high-priority call arrival, which has the density

function f∗_(t∗

s)and the distribution function F∗(t∗s). Since ts

is exponentially distributed, t∗

s has the same distribution as tsdue to the memoryless property. Therefore, when m= C, n= 0 and n ≤ m+1, p(m,n)can be derived by considering the relationship between t∗

s(for L and the m existing high-priority

calls) and th(the inter-arrival times of two high-priority calls):

p(m,n) =  _∞ t∗_s=0  t∗_s th=0 m n − 1  F∗(th)m−n+1 [1 − F∗_(t h)]n−1λe−λthf∗(t∗s)dthdt∗s (4) =  _∞ t∗_s=0  t∗_s th=0 m n − 1 _m−n+1  j=0 m − n + 1 j  (−1)j λe[−(n+j−1)η−λ]th_ηe−ηt∗s_dthdt∗ s = m−n+1 j=0 m n − 1, j   (−1)j λ λ + (n + j)η  (5)

Equation (4) says that if Ki = mand Ki+1 = n, then among

these m calls, the residual sojourn times of n− 1 calls are

larger than th (and therefore remain in the MSC at the end of th). The other m− n + 1 calls have shorter residual sojourn

times than th(and leave the MSC before the end of th). For l≥ 2, let p(l)_(m,n)= C  j=0 p(l_(m,j)−1)p(j,n) (6) In (6), p(l)

(m,n) is the probability that the stochastic process moves from state m to state n with exact l steps (i.e., there are l subsequent high-priority call arrivals). By convention,

p(1)_(m,n)= p(m,n). Then for i≥ 1, Pr[Ki= n] is expressed as

Pr[Ki= n] = C_−1 m=0 Pr[K0= m]p (i) (m,n) (7)

For i≥ 2, (7) can be recursively computed by using (6), and

we have Pr[Ki= n] = C_−1 m=0 Pr[K0= m] ⎡ ⎣C j=0 p(i_(m,j)−1)p(j,n) ⎤ ⎦ From (3), (5) and (7), the preemption probability ¯pnis derived

as ¯ pn= ∞  i=0 Pr[Ki= C− 1]p(C−1,C) (8) Note that we typically do not see infinite high-priority call arrivals during L’s sojourn time. From (7), it is clear that

lim

i→∞Pr[Ki = C − 1] = 0. Therefore, it suffices to consider i ≤ 50 in (8). In this analytic model, pf can be analytically

derived using the technique in [14], and pr is then computed

0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 pr 10−2 10−1 100 ₁₀1 ₁₀2 Vd(unit: 1/δ2) • : Vc= 10/μ2  : Vc= 100/μ2  : Vc= 1000/μ2 ..._... ..._... ..._... ..._... ..._... ..._... ... ..._... • • _• • • • • _{• •} ..._... ..._... ..._... ..._... ..._... ...    _    _{ } ..._... ..._... ..._... ..._...        _{ } (a) pr 0.55 0.60 0.65 0.70 0.75 0.80 θP 2C 10−2 10−1 100 ₁₀1 ₁₀2 Vd(unit: 1/δ2₎ • : Vc= 10/μ2  : Vc= 100/μ2  : Vc= 1000/μ2 ... ... ... ... ... ... ... ... • • • • • • • • • ... ... ... ... ... ...          ... ... ... ...          (b) θP 2C

Fig. 6. Effects ofVcandVdonprandθP 2C(C = 5, δ = μ/5, λ = 2μ). The above analytic model is used to validate against the discrete event simulation experiments described in [15]. Our study indicates that the analytic results are consistent with the simulation results (see Table 1 in [15]; the differences are within 1.7%).

V. NUMERICALEXAMPLES

Based on the simulation experiments, this section inves-tigates the performance of the BRP scheme. Suppose that

tc has Lognormal distribution with mean 1/μ and variance Vc. The Lognormal distribution is selected because it has

been shown that the call holding time distribution can be accurately approximated by a mix of two or more Lognor-mal distributions [14]. Similarly, we assume that td has the

Gamma distribution with mean 1/δ and variance Vd. The

Gamma distribution is considered because the distribution of any positive random variable can be approximated by a mixture of Gamma distributions (see Lemma 3.9 in [16]), and is often used to represent the location residence times (inter-moving times) [10], [17], [18]. We can measure VCC call holding times and domain residence times from the commercial operation and then generate the Lognormal and Gamma distributions from the measured data. Experience from commercial operation shows that δ = μ/10 ∼ 10μ is

reasonable. In our numerical examples, we set δ = μ/5. The results for other δ values are similar, and are not presented. The effects of the input parameters are investigated as follows:

Effects of Vc and Vd on pr and θP 2C : Figure 6 (a) plots pr against Vc and Vd, which indicates that pr decreases

as Vd increases. This phenomenon is explained as

fol-lows. When the domain residence times become more irregular (i.e., Vdincreases), more short domain residence

times are observed. Since ts = min (tc, td), more short

sojourn times ts are also observed. For a CS-to-PS

domain transfer, the reserved CS bearer is less likely to be preempted if the call is more quickly switched

0.000 0.005 0.010 0.015 0.020 0.025 pf 10−2 10−1 100 ₁₀1 ₁₀2 Vd(unit: 1/δ2) • : Vc= 10/μ2  : Vc= 100/μ2  : Vc= 1000/μ2 ..._... ..._... ..._... ..._... ..._... ..._... ..._... • _{• •} • • • • _{• •} ..._... ..._... ..._... ...    _      ..._... ..._...    _     

Fig. 7. Effects ofVcandVdonpf (C = 5, δ = μ/5, λ = 2μ).

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 pr 3 4 5 6 7 8 9 C • : Vd= 0.1/δ2  : Vd= 1/δ2  : Vd= 10/δ2 ..._... ..._... ..._... ..._... ..._... ..._... ..._... ..._... • • • • • _• • ..._... ..._... ...    _    ...        (a) pr 0.2 0.3 0.4 0.5 0.6 0.7 0.8 θP 2C 3 4 5 6 7 8 9 C • : Vd= 0.1/δ2  : Vd= 1/δ2  : Vd= 10/δ2 ... ... ... ... ... ... ... ... ... ... • • • • • • • ... ... ...        ...        (b) θP 2C

Fig. 8. Effect ofC on prandθP 2C(δ = μ/5, λ = 2μ, Vc= 1/μ2).

back to the CS domain (i.e., td is shorter). Therefore, pr decreases as Vd increases. For the same reason, pr

decreases as Vc increases. The figure also indicates that

when Vd> 10/δ2, pris not significantly affected by Vc.

Since θP 2Cis a decreasing function of prin Equation (1), θP 2C is an increasing function of Vdand Vc as illustrated

in Figure 6 (b).

Effects of Vc and Vd on pf : Figure 7 plots pf against Vc

and Vd. This figure shows that pf decreases as Vc or Vd

increases. This phenomenon is similar to that of Vd and Vc on pr, and is consistent with that observed in [17].

When Vd> 30/δ2, pf is small and is not sensitive to the

change of Vc.

Effect of C on pr and θP 2C : Figure 8 plots pr and θP 2C

against Vd and C. The figure illustrates the trivial result

that pr is a decreasing function of C, and θP 2C is

an increasing function of C. The non-trivial result is that we quantitatively show that when C < 7, adding more channels at MSC significantly reduces pr (and

therefore significantly increases θP 2C). When C ≥ 7, pr is sufficiently small (θP 2C is sufficiently large), and

shows that the user behavior (i.e., Vd) significantly affects

the resource allocated at the MSC (i.e., C) to achieve the same prand θP 2C performances. For example, if the

mobile operator wants to limit prto 15% (which ensures

that θP 2C ≥ 67%) under the condition δ = μ/5, λ = 2μ

and Vc = 1/μ2, then only 4 channels are required at

the MSC when Vd = 1/δ2, while 6 channels should be

supported when Vd = 0.1/δ2 (when user behavior is

regular). In addition, when Vd > 10/δ2 (user behavior

becomes more irregular), pr is sufficiently small (and θP 2C is sufficiently large), and there is no need to add

extra resources (i.e., to increase C) at the MSC. Note that

Vc has the same effect on pr and θP 2C as Vd does, and

the details are omitted.

VI. CONCLUSIONS

This paper investigated Voice Call Continuity (VCC) tech-nique that transfers a voice call between the CS and the PS domains. When a UE switches from one domain to another during a VCC call, the bearer in the old domain is released, and a bearer is established in the new domain. This paper proposed the Bearer Reservation with Preemption (BRP) scheme to support fast and seamless domain transfer. When the UE switches the call from the CS domain to the PS domain, instead of releasing the CS bearer, this CS bearer is reserved with low priority. When the UE switches the call back to the CS domain, the domain transfer process simply raises the priority level of the reserved CS bearer to high priority. Through the preemption mechanism, the reserved bearers in the BRP scheme do not occupy the resources in the MSC for other normal calls. The percentage of re-connection overhead saved by BRP over the 3GPP procedures is denoted by θP 2C

for domain transfer from the PS domain to the CS doamin. From the BRP performance study, we observe the following:

• As Vdor Vcincreases, θP 2Cincreases. When Vdis large, θP 2C is not sensitive to the change of Vc.

• Vd and Vc significantly affects the resource (i.e., C)

allocated to achieve the same θP 2C performance. When C, Vd or Vc is large, θP 2C is sufficiently large, and

increasing C simply wastes the resources at the MSC. The above observations are also true for domain transfer from the CS domain to the PS domain, which indicate that when the user behavior (either in terms of call holding time or movement pattern) is more irregular, the advantage of the BRP scheme becomes more significant.

APPENDIXA VCC CALLSETUP

Suppose that a UE is attached to both the CS and the PS domains, and has performed the IMS registration. This UE can initiate or receive a call in either domain. Figure 9 illustrates the message flow for VCC call origination in the PS domain with the following steps:

Step A.1 The UE sends the Session Initiation Protocol

(SIP) [4], [19] INVITE message to the S-CSCF through the PS domain. This message contains the media infor-mation (e.g., IP address, port number and codec) for user data connection.

A.1 SIP INVITE

A.3 SIP INVITE

A.7 200 OK

A.12 SIP ACK A.4 H.248 Add UE Callee MSC VCC AS MGW CS IMS MGCF

A.2 SIP INVITE

A.13 SIP ACK

IP Bearer IP Bearer

A.3 SIP INVITE

A.6 SIP INVITE

A.10 SIP 200 OK

A.13 SIP ACK

A.12 SIP ACK A.10 SIP 200 OK

A.11 SIP 200 OK A.11 SIP 200 OK

A.14 SIP ACK A.8 H.248 Add

S-CSCF

A.9 H.248 Reply A.5 H.248 Reply

Fig. 9. VCC call origination in the PS domain.

Step A.2 The S-CSCF evaluates the SIP INVITE message

against the iFC of the UE. If the VCC service criteria are matched, the S-CSCF forwards the message to the VCC AS.

Step A.3 Based on the received SIP INVITE message, the

VCC AS records the call information (e.g., From, To and Call-ID headers), and then forwards the SIP INVITE message to the MGCF through the S-CSCF.

Steps A.4 and A.5 Based on the media information retrieved

from the SIP INVITE message, the MGCF exchanges the H.248 [20] Add and Reply messages with the MGW to allocate media resources for this call.

Steps A.6 and A.7 The MGCF modifies the media

informa-tion contained in the SIP INVITE message and forwards the modified message to the callee. Then the callee replies a SIP 200 OK with its media information to the MGCF.

Steps A.8 and A.9 The MGCF retrieves media information

from the SIP 200 OK message, and finalizes the MGW media resources for this call by exchanging the H.248 Add and Reply messages with the MGW.

Steps A.10 and A.11 The MGCF provides the final media

information and forwards the SIP 200 OK message to the VCC AS. Then the VCC AS forwards this message to the UE. The UE retrieves media information from this message, and the call path in the UE-MGW segment is established.

Steps A.12-A.14 The UE sends a SIP ACK message to the

callee through the S-CSCF, the VCC AS and the MGCF. After the callee has received the acknowledgment, the VCC call is established.

APPENDIXB 3GPP DOMAINTRANSFER

The message flow for PS-to-CS domain transfer is illus-trated in Figure 10 with the following steps:

UE

Callee

MSC VCC _AS

B.1 CC SETUP (VDN)

B.2 CAP IDP (VDN) B.4 ISUP IAM (IMRN)

B.7 SIP INVITE (IMRN) MGW

B.3 CAP CONNECT (IMRN)

CS IMS MGCF B.8 SIP re-INVITE IP Bearer IP Bearer B.15 ISUP ANM B.16 CC CONNECT IP Bearer CS Bearer B.13 SIP 200 OK B.14 SIP ACK CS Bearer B.17 SIP BYE B.18 SIP 200 OK CS Bearer CS Bearer B.11 SIP 200 OK IP Bearer IP Bearer

B.7 SIP INVITE (IMRN)

B.12 SIP ACK B.12 SIP ACK B.11 SIP 200 OK B.13 SIP 200 OK B.14 SIP ACK B.17 SIP BYE B.18 SIP 200 OK S-CSCF B.8 SIP re-INVITE B.9 H.248 Move B.10 H.248 Reply B.5 H.248 Add B.6 H.248 Reply

Fig. 10. PS-to-CS domain transfer (3GPP TS 24.206).

Step B.1 Through the CS domain, the UE sends a Call

Control (CC) SETUP message with the specific called VDN to the MSC.

Step B.2 The MSC sends a CAMEL Application Part (CAP)

Initial Detection Point (IDP) message to the VCC AS. This message contains the calling number of the UE and the called VDN.

Step B.3 Based on the calling number in the CAP IDP

message, the VCC AS identifies the ongoing call of the UE and allocates an IMRN for this call. Then the VCC AS replies a CAP CONNECT message with the IMRN to the MSC.

Step B.4 The MSC sends an ISDN User Part (ISUP) Initial

Address Message (IAM) to the MGCF to set up the CS bearer. This message includes the IMRN received at Step B.3 as the called party number.

Steps B.5 and B.6 Upon receipt of the ISUP IAM message,

the MGCF retrieves media information, and exchanges the H.248 Add and Reply messages with the MGW to allocate media resources for CS bearer between the UE and the MGW.

Step B.7 The MGCF sends a SIP INVITE message with the

called IMRN to the VCC AS through the S-CSCF.

Step B.8 Based on the calling party’s identity in the received

SIP INVITE message, the VCC AS retrieves the ongoing call information (i.e., the call information recorded at

Step A.3) of the UE, and then sends a SIP re-INVITE message to the MGCF to modify the call path in the UE-MGW segment.

Steps B.9 and B.10 Upon receipt of the SIP re-INVITE

mes-sage, the MGCF retrieves media information, and ex-changes the H.248 Move and Reply messages with the MGW to switch the ongoing call in the PS domain to the new call in the CS domain.

Steps B.11 and B.12 The MGCF exchanges the SIP 200 OK

and the SIP ACK messages with the VCC AS to indicate successful switching of the call path in the UE-MGW segment (corresponding to the re-INVITE message at Step B.8).

Steps B.13 and B.14 To complete the establishment of the

CS bearer, the VCC AS exchanges the SIP 200 OK and the SIP ACK messages with the MGCF (corresponding to the INVITE message at Step B.7).

Steps B.15 and B.16 The MGCF sends an ISUP Answer

Message (ANM) to the MSC. Then the MSC sends the CC CONNECT message to the UE. At this moment, the CS bearer for the UE-MGW segment is established.

Steps B.17 and B.18 When the SIP ACK message arrives,

the VCC AS exchanges the SIP BYE and the SIP 200 OK messages with the UE to release the previously-established IP bearer in the UE-MGW segment.

Figure 11 illustrates the message flow for CS-to-PS domain transfer with the following steps:

Step C.1 The UE sends a SIP INVITE message with the

called VDI to the S-CSCF.

Step C.2 The S-CSCF evaluates the SIP INVITE message

against the iFC of the UE. If the VCC service criteria are matched, the S-CSCF routes the call to the VCC AS.

Step C.3 Based on the calling party’s identity in the received

SIP INVITE message, the VCC AS retrieves the ongoing call information (i.e., the call information recorded at Step A.3) of the UE, and then sends a SIP re-INVITE message to the MGCF through the S-CSCF to switch the call path in the UE-MGW segment.

Steps C.4 and C.5 Upon receipt of the SIP re-INVITE

mes-sage, the MGCF retrieves media information, and ex-changes the H.248 Move and Reply messages with the MGW to switch the ongoing call in the CS domain to the new call in the PS domain.

Steps C.6 and C.7 The MGCF exchanges the SIP 200 OK

and the SIP ACK messages with the VCC AS to indicate successful switching of the bearer in the UE-MGW segment (corresponding to the re-INVITE message at Step C.3).

Steps C.8 and C.9 To complete the IP bearer establishment,

the VCC AS exchanges the SIP 200 OK and the SIP ACK messages with the UE (corresponding to the INVITE message at Steps C.1 and C.2). At this point, the IP bearer for the UE-MGW segment is established.

Step C.10 To release the previously-established CS bearer,

the VCC AS sends the SIP BYE message to the MGCF.

Steps C.11 and C.12 The MGCF exchanges the H.248

UE

Callee

MSC MGCF VCC

AS

C.1 SIP INVITE (VDI)

C.2 SIP INVITE (VDI) C.3 SIP re-INVITE

C.6 SIP 200 OK

C.8 SIP 200 OK

C.9 SIP ACK _{C.9 SIP ACK} CS IMS IP Bearer IP Bearer CS Bearer C.3 SIP re-INVITE C.4 H.248 Move C.6 SIP 200 OK C.8 SIP 200 OK IP Bearer C.10 SIP BYE C.10 SIP BYE C.13 ISUP REL C.15 CC DISCONNECT C.16 CC RELEASE C.14 ISUP RLC C.18 SIP 200 OK _{C.18 SIP 200 OK} C.17 CC RELEASE COMPLETE CS Bearer IP Bearer IP Bearer CS Bearer CS Bearer S-CSCF

C.7 SIP ACK C.7 SIP ACK MGW

C.11 H.248 Subtract C.12 H.248 Reply C.5 H.248 Reply

Fig. 11. CS-to-PS domain transfer (3GPP TS 24.206).

CS bearer between the MSC and the MGW.

Steps C.13 and C.14 To complete the CS bearer release

be-tween the MSC and the MGW, the MGCF exchanges the ISUP RELEASE (REL) and the RELEASE COM-PLETE (RLC) messages with the MSC.

Steps C.15-C.17 The MSC exchanges the CC

DISCON-NECT, the CC RELEASE and the CC RELEASE COMPLETE messages with the UE to disconnect the CS bearer between the MSC and the UE.

Step C.18 Upon receipt of the ISUP RLC message at Step

C.14, the MGCF sends a SIP 200 OK message to the VCC AS to indicate successful release of the CS bearer.

REFERENCES

[1] 3GPP, “3rd Generation Partnership Project; Technical Specification Group Services and System Aspects; General Packet Radio Service (GPRS); Service Description; Stage 2," Technical Specification 3GPP TS 23.060 version 7.6.0 (2007-12), 2007.

[2] J.-R. Lin, A.-C. Pang, and Y.-C. Wang, “iPTT: Peer-to-Peer Push-to-Talk for VoIP," Wireless Communications and Mobile Computing (WCMC), published online, 2008.

[3] 3GPP, “3rd Generation Partnership Project; Technical Specification Group Services and System Aspects; Voice Call Continuity (VCC) between Circuit Switched (CS) and IP Multimedia Subsystem (IMS); Stage 2," Technical Specification 3GPP TS 23.206 version 7.5.0 (2007-12), 2007.

[4] Y.-B. Lin and A.-C. Pang, Wireless and Mobile All-IP Networks. John Wiley & Sons, Inc., 2005.

[5] 3GPP, “3rd Generation Partnership Project; Technical Specification Group Services and System Aspects; Voice Call Continuity (VCC) between Circuit Switched (CS) and IP Multimedia Subsystem (IMS); Stage 3," Technical Specification 3GPP TS 24.206 version 7.4.0 (2007-12), 2007.

[6] 3GPP, “3rd Generation Partnership Project; Technical Specification Core Network; IP Multimedia Subsystem Cx and Dx Interfaces; Signaling Flows and Message Contents," Technical Specification 3GPP TS 29.228 version 7.8.0 (2007-12), 2007.

[7] 3GPP, “3rd Generation Partnership Project; Technical Specification Group Core Network and Terminals; Customised Applications for Mobile network Enhanced Logic (CAMEL) Phase 4; Stage 2," Technical Specification 3GPP TS 23.078 version 7.9.0 (2007-09), 2007. [8] 3GPP, “3rd Generation Partnership Project; Technical Specification

Group Core Network and Terminals; enhanced Multi-Level Precedence and Pre-emption service (eMLPP); Stage 1," Technical Specification 3GPP TS 22.067 version 8.0.0 (2006-12), 2006.

[9] 3GPP. “3rd Generation Partnership Project; Technical Specification Group Services and System Aspects; Multimedia priority service," Technical Specification 3GPP TS 22.153 version 8.0.0 (2007-09), 2007. [10] Y.-B. Lin, “Performance modeling for mobile telephone networks,"

IEEE Network Mag., vol. 11, no. 6, pp. 63-68 Nov./Dec. 1997. [11] D. Gross and C. M. Harris, Fundamentals of Queueing Theory, 3rd ed.

John Wiley & Sons, 1998.

[12] W. Shen and Q.-A. Zeng, “Two novel resource management schemes for integrated wireless networks," in Proc. International Conf. Inform. Technol., Las Vegas, Apr. 2007.

[13] L. Kleinrock, Queueing Systems: Volume I – Theory. New York: Wiley, 1976.

[14] W.-E. Chen, H.-N. Hung and Y.-B. Lin, “Modeling VoIP call holding times for telecommunications," IEEE Network Mag., vol. 21, no. 6, pp. 22-28, Dec. 2007.

[15] M.-H. Tsai and H.-W. Dai, “Supplementary Report for Bearer Reser-vation with Preemption for Voice Call Continuity," Technical report, ChungHwa Telecom (CHT), Taiwan, 2008. Technical Report; [Online]. Available: http://pcs.csie.nctu.edu.tw/papers/tech_report_brp.pdf. [16] F. P. Kelly, Reversibility and Stochastic Networks. John Wiley & Sons,

1979.

[17] Y.-B. Lin, S. Mohan, and A. Noerpel, “Queueing priority channel assignment strategies for handoff and initial access for a PCS network," IEEE Trans. Veh. Technol., vol. 43, no. 3, pp. 704-712, 1994. [18] S.-R. Yang, “Dynamic power saving mechanism for 3G UMTS system,"

ACM/Springer Mobile Networks Applications, vol. 12, no. 1, pp. 5-14, 2007.

[19] IETF, SIP: Session Initiation Protocol, ETF RFC 3261, 2002. [20] ITU-T, Gateway Control Protocol: Version 3, Technical Report

Recom-mendation H.248.1, ITU-T, 2005.

Yi-Bing Lin (M’95-SM’95-F’03) is Chair Professor of Computer Science, National Chiao Tung Univer-sity. His current research interests include wireless communications and mobile computing. Dr. Lin has published over 220 journal articles and more than 200 conference papers. Lin is the co-author of the books Wireless and Mobile Network Architecture (co-author with Imrich Chlamtac; published by John Wiley & Sons), Wireless and Mobile All-IP Net-works (co-author with Ai-Chun Pang; published by John Wiley & Sons), and Charging for Mobile All-IP Telecommunications (co-author with Sok-Ian Sou; published by John Wiley & Sons). Lin is an IEEE Fellow, an ACM Fellow, an AAAS Fellow, and an IET(IEE) Fellow.

Meng-Hsun Tsai (S’04) received the B.S. and the M.S. degrees from National Chiao Tung University (NCTU), Hsinchu, Taiwan, R.O.C., in 2002 and 2004, respectively. He is currently working toward the Ph.D. degree at NCTU.

His current research interests include design and analysis of personal communications services net-works, mobile computing and performance model-ing.

Hui-Wen Dai was born in Hualien, Taiwan, R.O.C., in 1984. She received the M.S. degree in Computer Science and Engineering from National Chiao Tung University (NCTU), Hsinchu, Taiwan, R.O.C., in 2008. Her current research interests include design and analysis of personal communications services networks, mobile computing and performance mod-eling.

Yuan-Kai Chen received the B.S.C.S.I.E., M.S.C.S.I.E. and Ph.D. degrees from National Chiao Tung University, Hsinchu, Taiwan, R.O.C., in 1989, 1991 and 2002, respectively. In 1991, he joined Chunghwa Telecom Co., Ltd., Taiwan, R.O.C. He has been involved in design of 2G/3G/WBA network, mobile value-added services, handset software development, and mobile network evolution. His current research interests include design and analysis of personal communications services network, mobile computing, and UMTS.