You are on page 1of 9

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO.

11, NOVEMBER 2008

4375

Channel Allocation for UMTS Multimedia Broadcasting and Multicasting
Yen-Cheng Lai, Student Member, IEEE, Phone Lin, Senior Member, IEEE, Yuguang Fang, Fellow, IEEE, and Wei-Hao Chen
Abstract—3GPP 23.246 proposed the Multimedia Broadcast Multicast Service (MBMS) to deliver multicasting content in the Universal Mobile Telecommunications System (UMTS). In MBMS, the common logical channel is enabled to serve multiple MBMS calls at the same time. Use of the common logical channel may cause interference to the dedicated logical channels serving the traditional calls. To more efficiently utilize the radio resource to serve both traditional and MBMS calls, this paper proposes two channel allocation algorithms: Reserved Resource for Multicasting (RRM) and Unreserved Resource for Multicasting (URM). We propose analytic models and conduct simulation experiments to investigate customer Satisfaction Indication (SI) for the two algorithms. Our study indicates that URM outperforms RRM in terms of customer SI. Index Terms—Channel allocation, multimedia broadcast multicast service (MBMS), performance modeling, soft handoff, universal mobile telecommunications system (UMTS).

T

I. I NTRODUCTION

HE 3GPP working group proposed the Multimedia Broadcast Multicast Service (MBMS) for multicasting multimedia content in the Universal Mobile Telecommunications System (UMTS) [8]. In UMTS, the Wideband Code Division Multiple Access (WCDMA) is the air interface between the User Equipments (UEs) and the UMTS Terrestrial Radio Access Network (UTRAN). Two wireless transmission modes for MBMS are defined in WCDMA, which are Point-ToPoint (PTP) and Point-To-Multipoint (PTM). In PTP and PTM modes, the dedicated logical channels and the common logical

Manuscript received June 16, 2007; revised November 11, 2007; accepted December 19, 2007. The associate editor coordinating the review of this paper and approving it for publication was M. Guizani. Y.-C. Lai’s work was partially supported by the National Science Council (NSC), R.O.C., under Contract NSC-97-2218-E-002-026. P. Lin’s work was partially supported by the National Science Council (NSC), R.O.C., under the contract number NSC-96-2627-E-002-001-, NSC-96-2811-E-002-010, NSC-96-2628E-002-002-MY2 and NSC-95-2221-E-002-091-MY3, Ministry of Economic Affairs (MOEA), R.O.C., under contract number 93-EC-17-A-05-S1-0017, Telcordia Applied Research Center, Taiwan Network Information Center (TWNIC), and Excellent Research Projects of National Taiwan University, 97R0062-05. Y. Fang’s work was partially supported by US National Science Foundation under grant CNS-0721744 and the National Science Council (NSC), R.O.C., under the NSC Visiting Professorship with contract number NSC-96-2811-E-002-010. Y.-C. Lai is with the Department of Computer Science & Information Engineering, National Taiwan University, Taipei 106, Taiwan, R.O.C. (e-mail: yclai@csie.ntu.edu.tw). P. Lin is with the Department of Computer Science & Information Engineering and the Graduate Institute of Networking & Multimedia, National Taiwan University, Taipei 106, Taiwan, R.O.C. (e-mail: plin@csie.ntu.edu.tw). Y. Fang is with the Department of Electrical & Computer Engineering, University of Florida, USA (e-mail: fang@ece.ufl.edu). W.-H. Chen is with the Mobile Application Software Division, High Tech Computer Corp., Taipei 231, Taiwan, R.O.C. (e-mail: Albert Chen@htc.com.tw). Digital Object Identifier 10.1109/T-WC.2008.070653

channels are used to deliver MBMS data to UEs, respectively. The MBMS users may receive MBMS data via a dedicated logical channel (by executing the radio bearer setup procedure [2]) or via a common logical channel (by simply tuning to the common logical channel without Node B’s awareness). A common logical channel can serve more than one MBMS call at the same time. The drawback for the usage of a common logical channel is that there is no associated control channel for common logical channels to report the channel quality, and the power control can not be done. To cover the whole cell, the transmission power of a common logical channel should be large enough. As mentioned in [8], [15], in WCDMA, the power control brings significant capacity gain. The usage of a common logical channel may bring high interference to the dedicated logical channels, and the capacity for dedicated logical channels decreases. It is recommended to use a common logical channel to multicast the MBMS data while the number of MBMS calls is sufficiently large. 3GPP 25.346 proposed the Channel Type Switching procedure to switch an MBMS user between PTP and PTM modes for more flexible channel allocation, whose details can be found in [1]. In UMTS, the radio resources are not only used to serve the MBMS calls but also other circuit-switched/packet-switched calls (also known as traditional calls) where dedicated logical channels are used to deliver the user data. The radio resource allocation for both the MBMS and traditional calls is a critical issue, which affects QoS of both MBMS and traditional calls significantly. This paper proposes two channel allocation algorithms: Reserved Resource for Multicasting (RRM) and Unreserved Resource for Multicasting (URM). We propose analytic models and conduct simulation experiments to investigate the performance of the two algorithms. II. C HANNEL A LLOCATION A LGORITHMS This section describes the RRM and URM algorithms. Suppose that a UE can only carry a traditional call or an MBMS call at a time. A traditional call is served by the dedicated logical channel. An MBMS call is served by either the dedicated logical channel (i.e., PTP mode) or the common logical channel (i.e., PTM mode). As mentioned in [1], [3], [14], the dedicated logical channel supports soft handoff, and the common logical channel supports soft combining, where a handoff1 call can use different logical channels to connect to more than one Node B at the same time in the overlapping
1 The handoff process is that when a communicating UE moves from one cell to another, the channel in the old Node B is released, and a channel is required in the new Node B.

1536-1276/08$25.00 c 2008 IEEE

Suppose that the system is at one of the two states.t (tc. If so. respectively. We consider a homogeneous system. one dedicated logical channel is allocated to serve the new or handoff MBMS call request. The details of soft handoff model will be elaborated in Section III. The following instructions are executed: First. This assumption will be relaxed by using general residence time distribution in our simulation experiments. The output measures derived in our models include: • λh. the system turns on the common logical channel to serve the MBMS request.e. The system determines whether there are free dedicated logical channels. For a new or handoff traditional call request.t (Pnc.g. where all cells are assumed to be identical. The RRM and URM algorithms are exercised as follows: Algorithm RRM. Otherwise. URM checks whether the number of MBMS calls in the system is less than or equal to θ.. path loss). the probability that a traditional handoff call (an MBMS handoff call) is dropped because there is no free channel in the new cell • Pnc. NO.m ) for a traditional call (an MBMS call) is exponentially distributed with the mean 1/μt (1/μm ). There are C ∗ = C − θ − 1 dedicated logical channels that can be used . the call is forced to terminate. and URM switches the state from OFF to ON. At the ON state. The θ value depends on several physical layer factors (e. it is preferred to use the dedicated logical channels (i. The state is set to ON. and readers may refer to [5]. one dedicated logical channel is allocated to serve the new or handoff MBMS call request. When a traditional call is completed. Let nt be the number of traditional call users in the system. Otherwise (i. the system decides whether the dedicated logical channel or the common logical channel can be used to serve the MBMS call. URM checks whether there are handoff calls in the handoff queue. The UE residence time ti in Cell i has an exponential distribution with the density function fm (ti ) = ωe−ωti . If so.m . i.e. A NALYTIC M ODELS This section proposes analytical models for the RRM and URM algorithms. the new traditional call request is blocked. and the state switches from ON to OFF. Next. If C − nt ≤ θ. one dedicated logical channel is allocated to serve the request. Otherwise.m ): the call incompletion probability for a traditional call (an MBMS call) A.e. The request is served by the common logical channel. The call holding time tc.m ): the forced termination probability for a traditional handoff call (an MBMS handoff call). III. URM takes the same actions as that in RRM. 11. The URM checks whether there are handoff calls in the handoff queue. The MBMS calls (that are currently served by the dedicated logical channels) are moved to the common logical channel. If so. the common logical channel is turned on to serve more than one MBMS user simultaneously. Then the system checks whether there are handoff traditional calls buffered in the handoff queue. The system is always at the ON state and the MBMS call requests can always be served. Analysis for Algorithm RRM The RRM algorithm turns on the common logical channel to serve MBMS calls. one dedicated logical channel is released.. ON and OFF. We consider the following two cases to handle a new or handoff MBMS call request. [6]. If the traditional handoff call moves out of the overlapping area before a dedicated logical channel is available. VOL. A handoff call can connect up to 2 radio links in the overlapping area between two adjacent cells. At the OFF state. If the number of MBMS calls < θ. If so.t (Po. NOVEMBER 2008 area between adjacent cells. URM moves the MBMS calls (that are currently served by the common logical channel) to the dedicated logical channels. the new MBMS call request is blocked and the handoff MBMS call request is buffered in the handoff queue. URM checks the number of MBMS calls in the system.4376 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. then URM turns on the common logical channel for the incoming MBMS call. A Node B turns on at most one common logical channel to serve the MBMS calls. The common logical channel is turned off. and there are at most C ∗ = C−θ−1 dedicated logical channels. the released channel is assigned to a handoff call in the queue.t and λo. The URM does not reserve any common logical channel to serve the MBMS calls in advance. and the handoff traditional call is buffered in the handoff queue.. The traditional and MBMS call arrivals to a cell form Poisson processes with arrival rate λo. For a new or handoff traditional call arrival. The analysis of θ is out of the scope of this paper. to save the bandwidth. For an MBMS call request. 7. Algorithm URM. Assume that there are C logical channels in a cell. When a served MBMS call leaves the system at the OFF state. A FIFO handoff queue is executed in each cell to buffer handoff calls that cannot be served immediately due to the lack of channels in the new cell.m ): traditional (MBMS) handoff call arrival rate • Po. the common logical channel is not turned on. If number of MBMS calls in the system is larger than θ. the common logical channel is turned off.. the system check whether there are free dedicated logical channels. If so. We assume that turning on one common logical channel causes damage to θ dedicated logical channels. the released channel is assigned to one buffered handoff call.t (Pf. Case URM2: The system is at the OFF state. the number of MBMS calls ≥ θ). The new or handoff MBMS call is then served by the common logical channel. Case URM1: The system is at the ON state. The RRM reserves a common logical channel to serve the new or handoff MBMS calls. if the system is at the ON state. If C − nt > θ. On the other hand. RRM releases one dedicated logical channel. the released channel is assigned to one buffered handoff call. PTP mode) to serve the MBMS calls. and there are C dedicated logical channels.t (λh.m ): new call blocking probability for a traditional call (an MBMS call) • Pf. If so.

The timing diagram for the soft handoff model to serve the traditional calls. i ⎪ i!Mt ⎪ ⎭ Λi t (1) A new traditional call is blocked when the system is at state s(n) (where n ≥ C ∗ ) as it arrives.. Because the cells may overlap. and j outstanding traditional handoff calls).: CHANNEL ALLOCATION FOR UMTS MULTIMEDIA BROADCASTING AND MULTICASTING 4377 Λt Λt Λt C∗ 3M t Cell 1 Cell 2 Cell 3 Fig. where n ≥ 0 represents the number of traditional calls in a cell.t )] ηγ + (η + γ)μt ηγ(1 − Po.n leaves the cell. or leaves the cell) at time t + t1 . Let E[zi ] = 1/ω.t and Pnc.t . zi also has the same distribution as ti .t (1 − Po. the probability density function for t1 is given by (1) fn (t1 ) = (C ∗ Mt + jWt )e −(C ∗ Mt +jWt )t1 . From [14]. The MBMS call requests can be served in the system.t (1 − Pf.t C∗ +1 λh.t ηγ(1 − Po. Suppose that the overlapping period between Cell i and Cell i + 1 is τi .t . (3) Let tw be the maximum waiting time of an outstanding traditional handoff call (which is waiting for a second link).m = Pf. 1 illustrates the timing diagram for the soft handoff model. The Markov process for RRM channel occupancy time of a traditional call at a cell. Then ti can be expressed as ti = τi−1 + xi + τi .t is derived as follows. A connecting traditional call utilizes one dedicated logical channel during the non-overlapping period xi . Suppose that among the C ∗ + j traditional calls (including C ∗ traditional calls which are being served. if i > C ∗ ⎪ ∗ C∗ ∗ ⎪ C !M ⎪ (C Mt + jWt ) ⎩ t 1≤j≤i−C ∗ Since π0 + π1 + · · · + πn + · · · = 1. Assume that the second dedicated logical channel is found at time ui . The soft handoff model in [14] is applied to derive the output measures for the traditional calls.t + To compute Po. the remaining call holding time). [14]. Suppose that a traditional handoff call Ht. Let xi be the time period when the UE stays in the non-overlapping area of Cell i. Then the channel occupancy time of the connecting traditional call in Cell i + 1 is min(zi+1 . the density function fw for tw is fw (tw ) = Wt e −Wt tw . Let Mt = μt + ω. (5) . Fig.t t πi = π0 . (η + γ)μt + ηγPf. The call blocking probability for original traditional calls is Po. Then 1/Mt is the mean 3 2 1 x x τ2 2 1 0 u τ0 u τ1 u x τ3 4 2 0 t t 3 1 t 2 z t 3 1 z z Λt λh.t )λo.LAI et al. Let πi be the steady-state probability for s(n). Since cell residence time ti is exponentially distributed. By applying the memoryless property of the exponential distribution. from the memoryless property. and we have Po. Fig. the overlapping area) at time t. ω γ η γ γη From [7].t and Pf. Let Λt = λo.t C∗ + 2 0 Mt 1 2M t 2 C ∗M t C ∗ M t + Wt C C ∗ M t + 2Wt C t Fig. 2. Then from (1). Let Gn be the probability that k outstanding traditional handoff calls leave the new cell due to completion or dropping before Ht. expires. the UE will enter overlapping area i before it moves from Cell i to Cell i + 1. and j outstanding traditional handoff calls arrive at the cell earlier than Ht.t = Po.n enters a new cell (i.n . the πi probability can be computed by ⎧ i ⎪ Λt ⎪ i!M i π .t C∗ (C ∗ Mt + jWt ) 1≤j≤i−C ∗ + i=1 ⎫ −1 ⎪ ⎪ ⎬ +1 . λh. Let Wt = μt + γ. 2 illustrates the Markov process for RRM. we have π0 = ⎧ ⎪ ⎪ ⎨ C∗ ⎪ ⎪i>C ∗ C ∗ !Mt ⎩ ∗ ∗ ΛC λi−C t h. 1.t be the net traditional call arrival rate (including new and handoff traditional calls) to a cell. E[xi ] = 1/η and E[τi ] = 1/γ.e.m = Pnc. The Ht. and zi is the residual cell residence time in cell i.t for a traditional call can be computed as follows: λh. Then we have tw = min (remaining call holding time. where ti represents the time period during which a UE can receive the signal from Cell i. overlapping period).m = 0.t = i≥C ∗ πi .t )Pf. In this situation. the first outstanding traditional call leaves the handoff queue (either completes. Under the memoryless property of the exponential distribution. and the cell is at state s(n) (where n = C ∗ + j and j ≥ 0).n call request is buffered in (k) the handoff queue in the new cell. and is looking for a second dedicated logical channel from Cell i + 1 during τi . we have E[zi ] = E[ti ] = E[τi−1 ] + E[xi ] + E[τi ] or 1 1 1 γ + 2η 1 = + + = .t ) + λh.t (2) (η + γ)μt + ηγPf. we use a Markov process with states s(n). The 1/Wt is the mean waiting time of an outstanding traditional handoff call (which is waiting for a second link). (4) Pf. all dedicated logical channels in the new cell are busy.t and Pnc.t = = ηγ [λo. if i ≤ C ∗ ⎪ ⎨ t 0 ∗ i−C ∗ ΛC λh.t + λh.

and Pnc.t is ∞ Pf.k j=C ∗ .k + Hi.j.k Gi−1.k = δ1 πi+1.t .j. we have where Gn the probability that a traditional handoff call (entering the new cell at state s(n)) is forced to terminate as (n−C Pr {force termination|s(n)} = 1 − Gn ∗ (n−C ∗ +1) state transitions is omitted.k+1 + δ11 πi. and we have Po.t . λh. j = C − i. k) where 0 ≤ i ≤ θ.j+1.t − λh. 11.t . Let πi.j−1.j+1. k) ∈ S. the description for the (C − j)Mm + jMt + (i − C + j)Wm + kWt (C − j)Mm + jMt + (i − C + j)Wm + (k + 1)Wt . j. k ≥ 0. In URM.k (k+1) (k+i−C+j+1) πi. × Li. + (j + 1)Wt δ14 With the memoryless property. (10) From (9) and (10). j-1.t .j.j−1. Pf. j-1.k Gi−1. If λh.k (l−i+C−j) (l−1) (l−1) (l−1) (l−1) (l−1) (l−1) + Ni.j. The balance equation for the Markov process is expressed as (δ1 + δ4 + δ5 + δ8 + δ9 + δ12 + δ14 )πi.t . NO.j. (8) With (2)-(4) and (8).j+1. j.k .k−1 .old > δλh. Note that δ is a predefined threshold set to 10−7 . k i. k ≥ 0} . Compute Pnc.t. k ≥ 0.k−1 . k ≥ 0 or i ≥ θ + 1. the details are described in pp.j. From (7). The derivation of Pf. j.t for RRM.k + δ7 πi. 0 ≤ j ≤ C. × Li. k)|i ≥ 0. C − i ≤ j ≤ C.t.t by using (4) and (8).k Gi−1. s(i.j.j−1. j+1. The state s(i. Step 5.4378 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS.k−1 + Mi. j = C + 1.t . πi. k ≥ 0.t . VOL.k = ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ (1) ⎪ ⎪ G ⎪ i. Gi.k ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ where Gi. (11) (9) In (11). for n ≥ C ∗ (7) is obtained by using (6).j.k = (1) (k+i−C+j+1) can be obtained by iterating the if 0 ≤ i ≤ θ. 7.k + δ6 πi.j.old ← λh.t = n≥C ∗ (n−C 1 − Gn ∗ +1) πn .j.k Gi−1. k-1 = = Pr {tw > t1 } = C ∗ Mt ∞ tw =0 tw t1 =0 (1) fw (tw )fn (t1 )dt1 dtw δ12 i-1. 1-3 [10].j. 3.t is given by θ C Pf.k Gi. j.t : Step 1.j.k Di. j traditional calls (which are being served). respectively.k ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ (l) Gi.k + i≥θ+1 C j=C ∗ k≥0 πi.k + i≥θ+1 j=C ∗ +1 k≥0 1 − Gi. j.k + δ3 πi−1. j.t = θ i=0 C j=C−i k≥0 +1) .t for URM is similar to that of Pf. Otherwise. For all reachable / sates s(i.k + Ni.t . the following iterative approach is executed to compute λh. which can be found in pp. Then the state space S for this Markov process of URM is S = {s(i. C − i < j ≤ C.j. Compute λh.j. and Pf.j. j = C − i. ∗ (l−1) if i ≥ θ + 1. and the force termination probability Pf. We model the algorithm by using a three-dimensional Markov process.k πi.j.j.n keeps searching the second link within the time period tw ) as (1) Gn (1) i+1.k Gi. Fig. Select an initial value for λh. and k outstanding traditional handoff calls (which are waiting for the second link). Due to page limitation.k (k+i−C+j+1) πi.j. C ∗ + 1 < j ≤ C.k+1 . Step 4. j. Po. go to Step 6.k−1 + Mi.j. a new traditional call is blocked when the system is at the state s(i. Step 2.k following equations: ⎧ (1) ⎪ Gi. The values for λh. for n ≥ C ∗ .j. NOVEMBER 2008 Then we have the probability Gn (that Ht. Gn computed recursively by using (k) for n ≥ C ∗ can be i.j−1.j. k) denotes the state where a cell is occupied by i MBMS calls.k−1 + δ13 πi+1. k) ∈ S.j.t converge.j. k+1 G(k) = G(1) × Gn−1 . Step 3. Po.t .j.k . and Pf.t and Pf. j. Compute Po. we have πi. The probability that a traditional handoff call (entering the new cell at state s(n)) can get a free channel is given by (n−C Pr {get a channel|s(n)} = Gn ∗ The Markov process for URM +1) . k δ5 i. B. the steady state probability πi.t by using (2). k-1 δ11 δ1 δ 2 i. respectively.m for Algorithm URM can be derived as follows.k = 1. n n (k−1) (6) Fig.m . k).k be the steady-state probability for state s(i. k δ9 δ10 i. The Po. k ≥ 0 or i ≥ θ + 1.j.j.j. C ∗ ≤ j ≤ C.j.k + δ10 πi.k)∈S πi. By convention.t .j. Step 6. and Pf.k can be solved by using the successive over-relaxation method [4].j.k Gi.t by using (3). Analysis for Algorithm URM This section proposes an analytic model for the URM algorithm. j. 3 illustrates the state transition diagram for URM.t = i=0 j=C−i k≥0 C 1 − Gi.j.k + i≥θ+1 k≥0 1 − Di. j. j+1. 4-8 [10]. if 0 ≤ i ≤ θ. k δ7 δ8 δ3 δ4 i-1.k = 0 if state s(i. j. where the soft handoff model is applied to both traditional and MBMS calls.j. ⎪ ⎪ ⎪ ⎪ ⎪ (1) ⎪ ⎪ G ⎪ i. Po. k = 0.j.k + Hi. k+1 C ∗ Mt + jWt .j. Due to page limitation.j.j. then go to Step 2. k δ6 δ13 i+1. k > 0 or 1 ≤ i ≤ θ. Pf.

k ⎪ ⎪ ⎪ ⎪ [(i − C + j)/(k + i − C + j)]jM ⎪ t ⎪ ⎪ ⎪ jM + (i − C + j)W + kW .LAI et al. jMt . j = C.k Wi−1.j. k > 0. ⎪ ⎪ ⎪ [(i − C + j)/(k + i − C + j)](C − j)Mm ⎪ if i ≥ θ + 2. j = C.k in (11) can be computed by the following equations: (C − j)Mm + (i − C + j)Wm ⎧ (1) . k = 0.j.k + Hi. ⎪ Pf. k) state where 0 ≤ i ≤ θ. C ∗ + 1 ≤ j ≤ C.k (k+i−C+j+1) πi.k × Pi. C − i < j ≤ C − 1. j = C . ⎪ ⎪ ⎪ ⎪ if 1 ≤ i ≤ θ.j. k > 0 ⎪ ⎪ ⎪ ⎪ or i ≥ θ + 1.k−1 + Qi.j.t for URM.k = (l) ⎧ (1) ⎪ Wi.j. if i ≥ θ + 1.k + Ni.k Wi−1. k ≥ 0 and i ≥ θ + 1. C ∗ + 1 ≤ j ≤ C − 1. j = C ∗ . C ∗ + 1 ≤ j ≤ C − 1. C − i ≤ j ≤ C.j. respectively.k = (i − C + j)Wm ⎪ (1) (l) (l−1) + . otherwise. and the system is at the s(i. j = C.k−1 + Mi. iMm + jMt + kWt if i ≥ θ + 1. k > 0.k × Pi. C ∗ + 1 < j ≤ C.k .k Di−1.j.j.j. k ≥ 0.j.j+1. ⎪ if 1 ≤ i ≤ θ.t .j.k = iMm + jMt + kWt iMm + jMt + (k + 1)Wt and Pi. j = C − i. k > 0 ⎪ ⎪ ⎪ ⎪ or i ≥ θ + 1. λh. k = 0 or i ≥ θ + 1. k > 0 ⎪ ⎪ ⎪ ⎪ or i ≥ θ + 1. ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ 0. (C − j)Mm + jMt + (i − C + j)Wm + kWt if 1 ≤ i ≤ θ.k (k+i−C+j+1) πi.j−1. k > 0.j.k G(l−1) ⎪ i−1.j.k Wi−1.k Wi. k ≥ 0 or i ≥ θ + 1.j. k > 0.k−1 .k = iMm . Qi. C − i < j ≤ C. k > 0 or i ≥ θ + 1.j.t and Pnc. k > 0 ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ W (1) ⎪ i. (l−1) (l−1) (l−1) Ni. k ≥ 0.k = 1. otherwise. ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ = ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ + i≥θ+1 j=C ∗ +1 k≥0 1 − Wi.j+1. j = C − i.j. k > 0.j.j. C ∗ + 1 ≤ j ≤ C − 1. k > 0. ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ [k/(k + i − C + j)]jMt + kWt . Similar to the derivation for RRM. k > 0.j+1.j. j = C.j.j.k + i≥θ+1 j=C ∗ +1 k≥0 πi. k = 0. C − i < j ≤ C − 1. ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ [k/(k + i − C + j)]jMt + kWt ⎪ ⎪ ⎪ (C − j)M + jM + (i − C + j)W + kW .k × Pi. ⎪ if 1 ≤ i ≤ θ.k Wi.k−1 .k ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ jMt + (i − C + j)Wm + kWt . jMt + (i − C + j)Wm + kWt if 1 ≤ i ≤ θ.k Di. (C − j)Mm + jMt + (i − C + j)Wm if 1 ≤ i ≤ θ.j. C − i < j ≤ C. iMm + jMt + kWt jMt + kWt . k > 0 ⎪ ⎪ ⎪ ⎪ or i ≥ θ + 1.j. (C − j)Mm + jMt + (i − C + j)Wm + kWt if 1 ≤ i ≤ θ.j. j = C − i. ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ (l−1) (l−1) ⎪ W (1) × Li. k > 0 or i ≥ θ + 1. k > 0 ∗ ⎪ or i ≥ θ + 1.j.j. k = 0.t . j.k Wi−1.j.m = i=0 j=C−i k≥0 πi. k = 0 ⎪ ⎪ ⎪ ∗ ⎪ or i ≥ θ + 1. 0. k > 0.j. Di. and Pnc. ⎪ ⎪ ⎪ (1) (l−1) ⎪ D ⎪ i. .k−1 . j = C ∗ . C − i < j ≤ C.k ⎪ ⎪ ⎪ (l−1) (l−1) ⎪ ⎪ Mi. which is θ C Mi. A new MBMS call is blocked when all dedicated logical channels are occupied. ⎪ ⎪ ⎪ ⎪ if 1 ≤ i ≤ θ.: CHANNEL ALLOCATION FOR UMTS MULTIMEDIA BROADCASTING AND MULTICASTING 4379 and ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ = Li. × Li.j. otherwise.j. ⎪ ⎪ (1) (l) ⎪ D ⎪ i.k−1 . j = C ∗ .k ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ or 1 ≤ i ≤ θ. Hi.k .j. ⎪ m t t ⎪ ⎪ ⎪ if 1 ≤ i ≤ θ. j = C ∗ . ⎪ Di. k = 0.j.k−1 + ⎪ i.k = [(i − C + j)/(k + i − C + j)]jMt .j.k + Qi. ⎪ D ⎪ i. 0. ⎪ m t m t ⎪ ⎪ ⎪ if 1 ≤ i ≤ θ. C ∗ + 1 ≤ j ≤ C.k . if i = 0.k + Qi.j.t for URM are expressed as that in (2) and (3). k > 0. otherwise.k where Wi. Po. C − i < j ≤ C − 1. j = C. ⎧ (C − j)Mm ⎪ ⎪ ⎪ (C − j)Mm + jMt + kWt . C + 1 ≤ j ≤ C − 1.j. if i ≥ θ + 1.k Di. j = C.j. (C − j)Mm + jMt + (i − C + j)Wm + kWt ⎪ ⎪ ⎪ if i ≥ θ + 2.k Di−1.j. ⎪ ⎨ (C − j)Mm + jMt + (i − C + j)Wm + kWt (l) Di. (1) (k+1) where Di. 8-9 [10].j.k [k/(k + i − C + j)](C − j)Mm .j.j. k ≥ 0. j = C ∗ + 1. and the common logical channel is not turned on. ⎧ jMt + kWt ⎪ ⎪ (C − j)M + jM + kW . ⎪ t m t ⎪ ⎪ ⎪ if 1 ≤ i ≤ θ.k + Ni. otherwise. ⎪ ⎪ ⎪ ⎪ if 0 ≤ i ≤ θ.t . The details of the derivation for the Pf. j = C ∗ .m probability are given in pp. j = C. j = C − i. k > 0. Pf.j.j. ⎪ (i − C + j)Mm ⎩ ∗ if i = θ + 1.k .j. k > 0. k > 0.k Wi.k + Hi. The MBMS call blocking probability is θ C C Po. otherwise.j. 0.k = 0.j. if 0 ≤ i ≤ θ. ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ 0. k > 0.m = i=0 j=C−i k≥0 C 1 − Wi. (C − j)Mm + jMt + (i − C + j)Wm + kWt ⎪ ⎪ ⎪ if i = θ + 1. k = 0. C + 1 ≤ j ≤ C. C − i < j ≤ C − 1. k > 0. and the iterative approach applied to the analysis for RRM is used to obtain λh.

which includes two fields: callType indicates whether the event is for a traditional call or an MBMS call.t (Tc.m ] .t Ti. The QDELETE event represents the deletion of an event from the handoff queue. 11.t .m ≤ μ1 . Two counters.m ) be the probability that a traditional (an MBMS) call is connected but is eventually force-terminated. We repeat the simulation runs until Nt and Nm exceed 5. Pf. the complexity for the successive over-relaxation method is depend on the predefined threshold δ (used to obtain converge values of the steady state probabilities).m ] = 1/μm ) be the expected non-interrupted traditional (MBMS) call holding time.m .m . and depend on the input parameter setups. We adopt the successive over-relaxation method to compute the steady state probabilities πi for RRM and πi.m for URM are the same as that for RRM. callType.m Tc.t (N Tc. a call in conversation moves into the overlapping area of the two cells).m Ti. and E[Tc. The COMPLETE event represents the completion of a call. the total effective call holding time N Tc. Ti. (C − j)Mm + jMt + (i − C + j + 1)Wm + kWt The derivations for λh.m + ηγ(1 − Po.t (Ti. The event are inserted into an event list and are deleted/processed from the list in a non-decreasing timestamp order. we have 0 ≤ SI ≤ 1. The ARRIVAL event represents a new call arrival at a cell.m .t ] = 1/μt (E[Tc.t (No.t + Pc. which is defined as follows: Let Pc. During the execution of simulation.t + λo. Pf. The state. j) to count the number of traditional calls and MBMS calls. which are the same as those in the H ARRIVAL event. The callType and newCell fields in this event are the same as those in the H ARRIVAL event. Denote Cell(i.m λo.m Ti. we use the C++ programming language. tr stores the residual call holding time of the call.j) provides a FIFO handoff queue to buffer the handoff calls due to the lack of free channels.t ≤ μt and Pi.t λo. the total number of blocked traditional (MBMS) calls No.m + Pc.m )λo. j) and CRm (i. we consider a 8 × 8 wrapped mesh cell structure. Pnc. The H COMPLETE event represents the completion of a soft-handoff call.t Tc.m ηγ(1 − Po.m Tc. the total number of traditional (MBMS) calls Nf.m ) for incomplete traditional (MBMS) calls. j). and ta indicates the timestamp when a soft-handoff call is buffered in the handoff queue. This event includes the callType and oldCell fields. Po. which are λh.000 to ensure the stability of the simulation results. NOVEMBER 2008 and (1) Wi.m )Pf. and compute the output measures. we use the Satisfaction Indication (denoted as SI) to reflect the customers’ satisfaction about the network. j) turns on the common logical channel to serve the MBMS users.m By using the iterative approach for the analysis of RRM. oldCell and newCell fields are the same as those in the H ARRIVAL event.t + Pc. cur ts. and Pnc.m = and Pnc. which are the same as those in the H ARRIVAL event. As mentioned in [4]. Pf. j) are maintained for the Cell(i.k (C − j)Mm + jMt + (i − C + j)Wm + kWt = . to indicate the time when an event occurs. the total effective call holding time N Ti.m ) be the expected effective call holding time for a traditional (an MBMS) complete call. which indicates the progress of the simulation.m and Pnc. NO.k for URM.t (Pc. callType.t (Nf. This event consists of six fields: callType is the same as that in the ARRIVAL event.t Ti. j) for a period tx . ts .m) for complete traditional (MBMS) calls. the total number of handoff calls Nh. We define six types of events as follows.j. This event consists of five fields.m ) that are forced to terminate.e..t ] Pi. Due to page limitation. We calculate six output measures Po.m . then moves to the overlapping area of one of the four neighboring cells (selected with equal routing probabilities) for a period tτ . oldCell and newCell stores the ID of the cell which the user handoffs from and to.m (η + γ)μm + ηγPf.t (N Ti.t (Nh.t . D.m Pi. the total number of complete traditional (MBMS) calls Nc. S(i. C. In the implementation of the proposed analytical models. (η + γ)μm + ηγPf. Pi. 7. Then SI is expressed as: SI = + λo. Following the previous study [9]. tc .000. Then we obtain the output .t (Nc. The RELEASE event represents that the occupied channels are released at the old cell due to the forced termination of a handoff call or the completion of a handoff call. and finally moves to the non-overlapping area of one of the neighboring cells. [14].t + λo.m ) be the probability that a traditional (an MBMS) call is completed.m E[tc. oldCell and newCell. Simulation Model This section describes the simulation models for the two channel allocation algorithms. indicates whether the Cell(i.4380 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. Each Cell(i. and Pnc.m . respectively. VOL. CRt (i. [11].m ). respectively.t .m ) for traditional (MBMS) calls.m = Po. j) as one of wrapped mesh cells. Our models adopt the discrete event approach which has been widely used in many Personal Communications Service (PCS) studies [7]. In the models. tr . The periods tx and tτ are generated by the exponential or gamma random number generator. we do not elaborate the models further. oldCell indicates the cell where the call is originated.m ) be the expected effective call holding time for a traditional (an MBMS) incomplete call. Simulation Validation The analytical models are validated by simulation experiments. The following variables are used to calculate output measures: the total number of ARRIVAL events Nt (Nm ) for traditional (MBMS) calls. we obtain λh. The tc . and a larger SI m implies a higher customer satisfaction.m + Pc.t Tc. tτ stores the residence time in the overlapping area. is maintained. A traditional user or an MBMS user resides in the non-overlapping area of a Cell(i.m . We maintain a timestamp.m ). The area filed indicates that the call is completed in the non-overlapping area or overlapping area. The H ARRIVAL event represents a softhandoff call arrival (i. Po.m for URM. Tc. tc stores the elapsed connection time of a call before the event is processed.t E[tc.m . The event consists of five fields.t (Pi. 1 Note that since Pi.m λo. a simulation clock.j.

..0 = mμ ..... . . .... η. ..0 71.. .... . . .tμ51 = γ ... . .1 9520.... . .. .cnP t. .. 96 ◦... 94 . ...1 84... .... . ....0 724259. ...1 15.. . .. . ...◦ 88 ◦ : E[τi ] = 0..◦. .. .. .....25μt . . . .... ..... . .. . 100 ×. ....... .2 895068.7 81414..4 = θ . .. ... ....0 noitalumiS MRU 558874. . .m = μt ..... .1 )%( rorrE 44.... . ..0 52. .. .. .. ....0 45. .......... ... ..2 noitalumiS MRR 0 0 0 lacit yla n A 92343......61 = C( t. . .. . .... . ◦...... . μm = 0. ..e. ◦.2 noitalumiS MRR 0 0 0 lacit yla n A 92343.. ..... .. .. .0 noitalumiS MRU 8132.◦....9 96.. .. .... ......15/μt . 6 (a) studies the effects of μm .... Most of operators used in the implementation of the analytical model is “+” and “×”. .. .. 98. ...t . ... ..m values of analytic models and simulation experiments for different λo. .. Dashed: RRM ... Pf. . Effects of τi and xi : Fig..........t with different E[τi ] setups. .... ..... ◦. × 93 .. .0 10............ Po...... .◦........... . ... 6 (b) investigates the effects of θ on the performance for RRM and URM.. and η are normalized by μt ......... . .... . ... ..... . ...1 82.. .. ..... ..0 583113..× .. ..... . . ... .... ... .m ..cnP 0 0 0 noital u miS M R R 33794..6 96.....0 noitalumiS MRU 682314. .. ........ ...2 )%( rorrE 50.0 876359.. ... ... . ..×.... .....2 51363.. .... ...0 54. ◦ .... .. γ = 20μt . ... ... .. .1 32...0 lacitylanA 81854...0 5099. .......cnP fo noitadilaV :)V( esaC 27... ... This figure indicates that when λo... .. .... .cnP t........ IV......1 0 0 0 7 t. Effects of λo. .... .1 8360..... .. 87 : E[τi ] = 0. . .◦..0 lacitylanA 556759.. . .... ..m (Unit: μt ) (b) Fig....... ........ ..... . ..oλ )tμ30....... ... .0 )tμ :tinU( η 1.............. .. .0 lacitylanA 6 5 )tμ :tinU( t.0 726793.t for both URM and RRM.2 37600..1 7863...45 μm (Unit: μt ) (a) 80 1 2 3 4 θ (b) 5 6 7 Fig....0 = mμ . . . ◦.1 11953.. . . ... γ. . .t and Pnc..cnP dna t. .......... ... .. if the expected traditional call holding time is 1/μt = 3 mins. SI ......... ..0 )%( rorrE 70652..5μt : η = 1. . . .0 noitalumiS MRU 17491...2 5...... . . .. .oλ . ◦ .....×.◦ ...21 55726...25μt .tμ = m.. . . ........25/μt .m (θ = 4.. ..: CHANNEL ALLOCATION FOR UMTS MULTIMEDIA BROADCASTING AND MULTICASTING .. . .... . . ◦ . . ..... ....cnP dna t.... ... ....4 92343.cnP )%( )%( )%( )%( )%( )%( 96. ..... ×. .... ....5μt ) each figure.. . The parameter setups are given in the caption of each figure... .0 noitalumiS MRU 26794.... ..e. . .... .. .. ..... ... .... This figure indicates that for RRM.....× .0 56983. .. ... .. .. .... ...0 60....... 97.... . . ... ....2 48533. .... ..4 = θ .... .2 92343... .... ... 99 ..4 51363..0 = η . ...........1 37110. ...cnP fo noitadilaV :)II( esaC 63..2 67313..1 noitalumiS MRU 2996670. .. . ◦ .. .. ...... . . .0 noitalumiS MRU 799804..◦ ..t = 5μt ◦ : λo.. ... .oλ .. . ...0 02.. ... . ...m ..... .. ..0 187323. ..t . ..0 lacitylanA )%( 0 0 0 )%( rorrE 69.t . .. . ... ... ...0 )%( rorrE 819623. . Fig..tμ = m.... . ◦ .. .. 5 (a) shows that with small λo. ..... .. 89 ×: E[τi ] = 0.... ....... ....30 0... .. .... As shown in Fig...m increases (i... ........ .0 50. .. ... ... . . .1 04... . .. .. ... 5.0 48. .....61 = C( γ tsniaga m.... 6.2 61043.2 67023... . ...t . .... .oλ .... .... . .. μm ..... . ×.cnP 0 0 0 noital u miS M R R 16363. ...25 0....5μt : η = 1.......0 )%( rorrE 74. The reasons causing these phenomena has been treated as a technical report...... .1 noitalumiS MRU 782923.. .◦ . .2 278548. ... .. . ........ . .35/μt 86 4 5 6 7 8 4381 SI (%) 98........oλ .. When λo... .... . ..cnP t. . ...... . . . . .. ×.. . . .......... ... . .0 noitalumiS MRU 60140.. Dashed: RRM Solid: URM ×: η = 0... . ... . but with high λo. .t . . . ..... .... .. ◦ . . ×.LAI et al... ......... . . . .. ... .... .......2 45730. ...........2 37....... .. . . . .... .0 48.... ....61 = C( η tsniaga m.... . ×. .. . ..... . . .... .1 07.1 )%( rorrE 3937570... 4... . ...... . .. . ..... . . ..... .. . ..0 lacitylanA )%( 0 0 0 )%( rorrE 67......t . . ....... .tμ51 = γ .. ... .. ......t and E[τi ] (θ = 4.....t ...cnP 0 0 0 noital u miS M R R 51363.. . .0 48.tμ = m.cnP fo noitadilaV :)IV( esaC 18. .....0 )%( rorrE m... .cnP fo noitadilaV :)I( esaC 08....1 72..... ..... .. 92 .... .×. ... ...... ..0 lacitylanA 80381....1 . .... . .... ..1 508889.0 lacitylanA 53210....0 2632.. . ........... 95 .... . ◦. ... . ..0 noitalumiS MRU 9769..× .... .. . .m iteratively until the computation results converge to fixed values.1 )%( rorrE m... . . .. . ..... . ... .8 .. ◦.... ..cnP t.0 26. . ... .. . λo. ........... i. .... . ......2 92343.. .... .0 = η ... .. .... . . .. ...... . λo. ....... .0 159304.. ..cnP dna t. We observe that SI is a decreasing function of λo...0 788604....0 )%( rorrE m.......... .. Fig. ....tμ6 = t. .. .. . .... . . .oλ ..0 80..2 51363.. .... ..0 64.4 = θ .. ... λo....1 lacitylanA )%( 0 0 0 )%( rorrE 48.tμ6 = t. .... .. .. ..cnP dna t..... .. .. ..× . μm = 0.. ..... . .. .... .....0 = η ... . ... ............1 4491.. ... ..0 noitalumiS MRR 0 0 lacit yla n A 97158. .... .. .. ... . ... ... .. .... . ...1 432179.....×... ........... ..... . ..... .. γ = 20μt ) (b) Effects of θ (λo. ... ..... ..1 )%( rorrE m.. ..tμ51 = γ . × ....0 48. . μm . ... ◦..... ..tμ = m... . γ.....cnP 0 0 noital u miS M R R 26198......oλ 7 6 5 )tμ :tinU( t. .m . .. ... .0 53. λo..0 noitalumiS MRU 15934..2 88373. . × ..m .. . . × ..2 51363. ...... .. The URM outperforms RRM in terms of SI.... ....... ....... ..... ....... ......... .. ... . ..m .... .. we only describe the phenomena observed in )tμ30.2 )%( rorrE 92. . For URM.. .. .. . .. .... ...2 26769.. .2 0... . .... .. .... .. .. ..... . . .... . ... 100 ×. . ... .0 = mμ . ...... . ..... . .2 05559.......... ..... ...0 26257............7 64983.. .. . ..... .1 lacitylanA )%( 0 0 0 )%( rorrE 24.2 63593. and θ: Fig..0 50.. .... . . ◦.m = μt . . ... η = 0.. Fig....... ... . .... λo.. . 5 (a) plots SI as a function of λo...... . ..0 89142... .0 65249....1 37. .....0 425783.... γ = 20μt ) 100 99 98 SI (%) ..... . ...0 17. ... γ..1 38...t = 6μt . ... ... .tμ3. .......0 397093.. . .. ..... .. ... .cnP fo noitadilaV :)III( esaC 66.. .....2 92343...0 )tμ :tinU( mμ 54. ..... .2 92343... μm . .1 48.84%............ .... . Fig. ... ... ... ..... ..... .. .... . Po...0 48.. URM outperforms RRM in terms of SI. .cnP t.....e. .2 51363.. ..... . .. ..... .. .. .1 18970....cnP 0 0 0 noital u miS M R R 5983.... ..1 516217.0 )%( rorrE 687764.. .... .. 5 (a) also investigates the effects of the expected length of the overlapping period (i.... . Dashed: RRM Solid: URM × ◦ × ◦ × ◦ × ◦ ◦ × 90 Dashed: RRM Solid: URM ×: λo. .2 5........ .0 = η ..3 96. .. .... ..1 9102. .....2 lacitylanA 02 01 5 )tμ :tinU( γ 02 01 5 )tμ :tinU( γ )tμ51 = γ . ..tμ6 = t.cnP dna t.... .... .. . the input parameters λo. .... .....tμ3. . .×. ... and θ on the performance of RRM and URM.. .. ...0 91... and the analysis and the simulation results match well for different conditions.. . .... . .... ....1 03....2 lacitylanA 1. the performance for URM approaches that for RRM.... .1 26. . ... ..oλ ... .... ....tμ6 = t...... .. . ..... .. 97.........◦.0 ...oλ ... θ setups. .... ... Fig.t ..cnP fo noitadilaV :)VI( esaC 62...0 )%( rorrE 25.t and Pnc. .1 )tμ :tinU( m.... SI is a decreasing function of θ for both URM and RRM. .. ........... . ........ ..2 noitalumiS MRR 0 0 0 lacit yla n A 73844.. ◦...... (%) ....... .4 noitalumiS MRR 0 0 0 lacit yla n A 733.....tμ = m...... . . . ..... . .0 948804.. . . . .............0 464111. .. λo....... . . . . . . .. . .... . ... . .1 noitalumiS MRU 88872... For example.0μt 85 95 0... . . .◦.tμ6 = t... We investigate the effects of λo....... .4 = θ . . ... then λo.. .. .... . .. . ..2 noitalumiS MRR 0 0 0 lacit yla n A 15543... .. . .. . ... .t = 6μt .......... .... . .0 )%( rorrE m.m between simulation and analysis results for RRM and URM are within 3. .....25μt ◦ : η = 0.oλ tsniaga m.. ........ 5 (b) studies the effects of λo........ .0 = mμ . .. . λo.... .5μt ) (b) Effects of λo..2 lacitylanA 54... .. . ..21 73266.. .. .... ... . the effects of μm can be ignored.0 lacitylanA 18891...... .. .. . 95 .0 48.2 77.. .... . ... .. and readers may refer [10].......0 = η .... 90 Solid: URM . ..0 52. .... ... . . .. ... .. 97 ...◦.2 40373..m < 3μt .... . ..... .. ×.. ..........0 615071..... . ....7 × .. . ... ......tμ3......... ... .. ... ....0 52.... λo...0 10.0 980157. ◦.........0 523049..... μm . . ...0 432893. .......◦.. C = 16.. ........... . . . ..oλ . . . ....2 ◦.0 179424. . . .t = 6μt means that the expected time for two successive traditional call arrivals at a cell is 30 seconds. The complexity of this iterative approach is dependent on the value of the predefined threshold δ and the input parameter setups. . . .25μt . ...... .... . . ..0 = mμ .....t = 6μt : λo... ..... × ◦ 97 96 × × × × × SI (%) ... ...... .....0 )%( rorrE 57.......t = 7μt ◦ ×: η = 0. ..... ..4 lacitylanA 7 6 5 θ 7 6 5 θ )tμ30.. ..1 )tμ :tinU( m.... 98 . ..tμ3.. .0 463111. .. ... .... ..1 )%( rorrE 137504...... × ...... Pnc... . ... ...... . Due to page limitation..... .............cnP 0 0 0 noital u miS M R R 51363. .. .61 = C( θ tsniaga m. Pnc. . . .. . 6 (b). . in the following.....m = μt .m . ... .... .. . μm = 0....61 = C( m. . . .... .. .... ....3 )%( rorrE 640614.... . ... .... .. ....40 0..0 625179.... η..... .....×. ..1 lacitylanA )%( 0 0 0 )%( rorrE 48. ....m ... .×... . .. ... .0 lacitylanA )%( 0 0 )%( rorrE 18. ... ..0 lacitylanA 393594.... . . ... ... ◦ ..... . ..... . . ... ....25μt ◦ : η = 0... .. ........◦ . .. E[τi ] has insignificant effects on SI for URM and RRM. ... ◦.... .. . ........... . ......35 0.. ....0 371571... .....4 = θ . 91 .. . ........ × ....... .. .. .... ... . .. ...... ..0 304123.. .oλ tsniaga m..2 58. .. ..... .0 )tμ :tinU( mμ )tμ30. . .. . . . .t (Unit: μt ) (a) λo... We assume that there are sixteen logical channels per cell.... ... (a) Effects of λo. ..........t .61 = C( mμ tsniaga m....t .. . ....2 lacitylanA 5.....2 0... The figure shows the errors for Pnc.1 333396. ......2 )%( rorrE m.. Fig. 4 shows the Pnc. measures λh...... . . . . . In our study. .0 53.. η = 0.0 34513.. .tμ3. .... Pf... ... ..0 24. . . . ... ..... ..... . E[τi ]) on the performance for URM and RRM. 97........... ........2 41206.. P ERFORMANCE E VALUATION This section investigates the performance of RRM and URM... SI increases when μm increases......0 )tμ :tinU( η )tμ30.m ...1 87. .. . ..... ...1 10. .5 .. . λh.... ...... ...oλ 5..... ..2 22793. . .oλ ........oλ ..... 98.. . . .... ....... . more MBMS call requests arrive in a period)... . .... Then we investigate the effects of the variance of the UE residence time in the non-overlapping and overlapping areas... . .cnP dna t.... ... ......× . .... ... ..0 881293. ..0μt 1 2 3 4 5 λo.....0 985824.. .tμ51 = γ . .. ... (a) Effects of μm (θ = 4.. . . .. ..2 6826..4 .. . .. .. ....oλ Validation of simulation and analytic results 05.. which are considered as the operators with low computation complexity.0 504983...

Lin.... . [13] P. ..... Taipei County....... . . .. . [6] P... ... . .. .... .. λo..25μt .. . increases... . ... ... [3] 3GPP.. . ... ... . .. .. Fang... Y.... μm = 0.... .. ... . respectively.... “Actual call connection time characterization for wireless mobile networks under a general channel allocation scheme..... 3. . Y.... .... ... .. .. .... .. PHOTO Taiwan. ..... . .. .... Yen-Cheng Lai (S’08) received the B. M. ..... HERE He is currently a Postdoctoral Researcher with the Institute of Applied Mechanics... RRM and URM for the UMTS MBMS service.. . ..... .. . ... 6. . .. J.. ..... [5] M.. Technical Specification Group Radio Access Network..... . Technical Specification Group Radio Access Network.... wireless Internet services.... .. . . .... ... . .. . .. . ... ... Wacker. ..... . 5....... IEEE International Conference on Communications... ..C.. Henry. ... . Oct.. .. .. . ..... 1 Fig. . .... . [4] G...... R... .. . . .” in Proc. . where we assume that the distributions for the non-overlapping period and overlapping periods are Gamma distributions with the variances V ar[xi ] and V ar[τi ]..◦. . ....... .. the performance for URM approaches that for RRM... . .. 4.. ....... .... .× ... Technical Report 3G TS 25. Our performance evaluation results showed that URM outperforms RRM in terms of SI (i. Due to page limitation..346... ...... 9. Technol.... pp. pp... . pp. ... ... . 11. .. . 2001.. . to more precisely simulate the radio interface of UMTS. ... ........... and Ph.. .. NTU..... ACKNOWLEDGMENT The authors would like to thank the two anonymous reviewers. ◦.... ....... 95 ..... . . . Fig. .. ... ... .. Chapman.. . . no.. . . January 1999. . ...... “Channel allocation for GPRS with buffering mechanisms. . ........×.5μt ) as E[τi ] increases. .... . .... .. 2002.. . 3231–3235. ......... ... . pp.. 3GPP... 47.. no. . . .... National Taiwan University.. Lin.... .. . Lin.... .. SI for URM and t RRM is insensitive to the change of the variance of the 1 overlapping period.” IEEE Trans. ... Fig.. and A. [15] K... .. . pp. .... Z.... I. ............ ..” ACM Wireless Networks.. . WCDMA for UMTS. 3GPP.. Introduction of the Multimedia Broadcast/Multicast Service (MBMS) in the Radio Access Network (RAN). . .. where we set λo. Li..” in Proc.. Computers... ...... . . 2006.◦.. .. Dec.. 1003–1005..931......... .. ......... .....0μt . degrees in computer science and information enPLACE gineering from National Taiwan University... 7 (a) shows that when V ar[xi ] ≤ μ2 ..... ... . no... . .. 2004.... we observe SI t slightly increases.. a common logical channel is reserved to serve the MBMS calls......... .. pp... i... .. . vol.. Lai.. . ....35/μt -3 -2 -1 0 1 2 -3 -2 -1 0 1 2 3 V ar[xi ] (Unit: 10x /μ2 ) t (a) V ar[τi ] (Unit: 10x /μ2 ) t (b) R EFERENCES [1] 3GPP.. .. . Technol.-H. V. . we observe the similar phenomena as those for the overlapping area... .◦..... Bolch..... .. . “Coverage and planning aspects of MBMS in UTRAN.. .” ACM Wireless Networks... . . . .. . . ..-B..... ... . ... .. .. . . .. when 1 V ar[xi ] > μ2 .. [9] J. ... .. . . ... . .... .” IEEE Trans.e..... [13] to well simulate the UE cell residence time.. Y. .” IEEE Trans. ... ..... ........ Protocol Specification (Release 6).... then URM turns on the common logical channel to serve the new incoming MBMS calls..........m = μt . . . .. ... H. ...15/μt ◦ : E[τi ] = 0.. γ = 20μt ) (b) Effects of V ar[τi ] (θ = 4. ... ... [10] Y. .... “Estimation of capacity and required transmission power of WCDMA downlink based on a downlink pole equation... .... . .... .m = μt (low MBMS call arrival rate).-C.. . Mar.. .. ... . Mar.... ... May 2000. ......... .... degree in computer science in 2002 from Tamkang University.. ... ... . ...... ...-B. ...5μt : η = 1..... Laiho-Steffens.. Fang. pp. .. if the number of MBMS calls in the system is larger than an interference threshold θ.. and W... “Channel allocation for GPRS. . ◦ .... . .. and wireless PAN/LAN/WAN...... Honkasalo. ... .. vol.. URM decides whether the dedicated logical channel or the common logical channel can be used to serve MBMS calls. . ... ... Radio Resource Control (RRC). “Comparing soft and hard handoffs.◦ .. WCDMA.. 1..... .. . ... .. Fifth IEE International Conference on 3G Mobile Communication Technologies... . we apply the soft-handoff model for both traditional and MBMS calls while constructing analytic and simulation . SI for URM and RRM decreases. 7 1 (b) shows that when V ar[τi ] ≤ μ2 . SI for URM t and RRM is insensitive to the change of the variance of the non-overlapping period. J..O.. .. ... 375–387. .... IEEE VTC2000-Spring.. .. [12] P.m = μt .. . Analytic models and performance evaluation for channel allocation for UMTS MBMS service.. 7... .. .. . SI degrades as V ar[xi ] increases. C ONCLUSION This paper proposed analytic and simulation models to study the impact of MBMS service on the radio resource in the UMTS network.... . .. . . . .. .. ..... λo. Their comments have significantly improved the quality of this paper..... .. .. .. ...... .. . . .. When V ar[τi ] > μ2 ...... .×... ... .. ... .. ... . ... . . ....... ..... . On the other hand.. .. Jeng and Y.... 1998. Veh. . . Czerepinski... . ... ..... . .. Krause.. .... 50.....331. ........ .. . Taiwan... .. ......4382 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS... Fang.. .. . . . . .. ... .... ... Lin and Y. 7 t (b) studies effects of the variance of overlapping period. ... . and R.. .. mobile computing. ...... .. .. Stage 2 (Release 6)..” IEEE Trans.... Greiner...... ... .e... ...... ... ....... . and Y... .... 2. η = 0. [8] H..... . ..... ..... and when the MBMS call arrival rate.. ... .. .. John Wiley and Sons Ltd. ×..t = 6μt (high traditional call arrival rate) and λo. .. 95 94 × ◦ 3 Dashed: RRM Solid: URM ×: E[τi ] = 0.. ... .. .. . . . .× ..... .. [14] Y. Chuah. “UMTS release 99/4 airlink enhancement for supporting MBMS services.... ...... ... .. Luo. . .. . ... Meer. ...◦..... ...... In RRM.... ...... In other words. . . 2002. .. . . Technical Report 3G TS 25... . . .......... . . UTRAN functions.. no.. ...... .. . vol... . ×. .. .” in Proc... Pang... . ... .. [7] Y.. . .. His current research interests include design and analysis of PCS networks..... .... . and W. . We considered two channel allocation algorithms.. On the other hand....... ◦. In our study... ...... 3rd Generation Partnership Project. × ◦ × ◦ models. Sipila.. and J... .. the performance figure for E[xi ] = 1/η is ignored.... . performance modeling.. NO..... ... ..... ..... . 2005. vol.. .. ... ... ... NOVEMBER 2008 100 ..◦. . .. .. . .. ... examples on signalling procedures (Release 6). λo. .. ....... Hu.. . ... Taipei.. . ... ..... .. 41–55. .. .. in 2004 and 2008........ Chlamtac.... . . . ..... .. .. . . . . Fig.... . .. vol... ... ... Chen.. 431–441... .... ..m . .. Queueing Networks and Markov Chains: Modeling and Performance Evaluation with Computer Science Applications. ... . June 2005. vol..... . “Channel occupancy times and handoff rate for mobile computing and PCS networks... . [11] W.... . .. ... . . “Equal resource sharing scheduling for PCS data services... John Wiley and Sons Ltd.. ..... ..... Lin..... S. [12]. VOL. The Gamma distribution has been adopted in several PCS studies [9]. pp..... . . . 679–692. technical report... customer satisfaction indication) in all cases. . ... 2003.. λo.t = 6μt .. Veh. .. ............ . ........... .... . λo.... ... .. .. (a) Effects of V ar[xi ] (θ = 4. . .. and K... .×.... 3GPP... .. no.. Lin and A.. . 792–798. . . 1. . . respectively. ..... . [2] 3GPP.. vol... .... S. .. we study the effects of variance of the nonoverlapping period and overlapping period.. 5......... ... . 7. 6.. . ....... . For effects of the non-overlapping period E[xi ] = 1/η. μm = 0. . ... .. .. P.. ...t = 6μt ... . ...... . and the M... ... ...... 99 ◦ ◦ × × ◦ 98 × × × × × ◦ ◦ ◦ ◦ × × ◦ ◦ 97 × × × × × ◦ SI ◦ ◦ ◦ ◦ (%) 96 . . ... .. pages 529–533.... . 3rd Generation Partnership Project...-Y. . Technical Report 3G TS 25.....S. ...-B.×...... ... Dec.-B..D. .. . 2000. . 682–691.. ... R. . .. Antti..... ....... Sept.... 90 SI (%) 85 80 75 70 Dashed: RRM Solid: URM ×: η = 0.. .... . ... Wireless Commun. . Harri and T.. ..... ..... .. ... ... . Trivedi.. . ...25/μt : E[τi ] = 0. . .... . ...... . ....◦. .... . 1998. ...... 2004....... .... T. .. . .. .. ..... . . . .... .. ..... 3rd Generation Partnership Project... .....25μt .. 49.×... .. . . .. ...... ...... .S.25μt ◦ : η = 0.. ... June 2004. .... .. . Technical Specification Group Radio Access Network... .. .-C... ... ×... . ..... Effects of Variance of τi and xi : Based on the simulation experiments. ..... . .... .... Lin.. no. ..

in 1996 and 2001. he was an Assistant Professor in Department of Computer Science and Information Engineering (CSIE). 2003-2008).O. and a Guest Editor for ACM/Springer MONET special issue on Wireless Broad Access. From August 2001 to July 2004. respectively.: CHANNEL ALLOCATION FOR UMTS MULTIMEDIA BROADCASTING AND MULTICASTING 4383 Phone Lin (M’02-SM’06) received his BSCSIE degree and Ph.D.D degree in Electrical Engineering from Boston University in May 1997. in 2005. Dr.C. Lin is listed in Who’s Who in Science and Engineering(R) in 2006. IEEE T RANSACTIONS ON M OBILE C OMPUTING and ACM W IRELESS N ETWORKS . He holds a University of Florida Research Foundation (UFRF) Professorship from 2006 to 2009. He then joined the Department of Electrical and Computer Engineering at University of Florida in May 2000 as an assistant professor. He is an Fellow of IEEE and a member of ACM. degree from National Taiwan University. R. Fang is also active in professional activities. and a recipient of 2006 Young Electrical Engineering Award. He has served on several editorial boards of technical journals including IEEE T RANSACTIONS ON C OMMUNICATIONS. got an early promotion to an associate professor with tenure in August 2003 and to a full professor in August 2005. Dr. Lin has published more than twenty international SCI journal papers (most of which are IEEE transactions and ACM papers). Since August 2004. Dr. T. IEEE Globecom in 2002. Technical Program Symposium Co-Chair for IEEE Globecom’2004. He currently is a senior engineer of HTC Corp. wireless Internet. degree in Systems Engineering from Case Western Reserve University in January 1994 and a Ph. a recipient of Fu Suu-Nien Award of NTU in 2005 for his research achievements. the 3rd IEEE ComSoc Asia-Pacific Young Researcher Award. a recipient of K. and develops several projects based on Windows Mobile PPC/SP.C. Dr. a Guest Editor for IEEE W IRELESS C OMMUNICATIONS special issue on Mobility and Resource Management. mobile IP. National Taiwan University.C. Dr. Wei-Hao Chen received his M. He is also an Associate Editorial Member for the WCMC Journal.O. Lin has received many research awards. R.. 2000. Li Young Researcher Award honored by ACM Taipei Chapter in 2004. He has been actively participating in professional conference organizations such as serving as the Steering Committee Co-Chair for QShine. He is the recipient of the Best Paper Award in IEEE International Conference on Network Protocols (ICNP) in 2006 and the recipient of the IEEE TCGN Best Paper Award in the IEEE High-Speed Networks Symposium. National Taiwan University. Lin is a Senior Member. IEEE. and performance modeling. and a member of Technical Program Committee for IEEE INFOCOM (1998. He was an assistant professor in the Department of Electrical and Computer Engineering at New Jersey Institute of Technology from July 1998 to May 2000. the Chinese Institute of Electrical Engineering. the Technical Program ViceChair for IEEE INFOCOM’2005.O. He has published over 200 papers in refereed professional journals and conferences.O. R. IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS. a recipient of Wu Ta You Memorial Award of National Science Council (NSC) in Taiwan in 2005. 2007. . Taiwan. His current research interests include personal communications services. Taipei. Lin is an Associate Editor for IEEE T RANSACTIONS ON V EHICULAR T ECHNOLOGY.LAI et al. and wireless security. R. he has been an Associate Professor in Department of CSIE and in Graduate Institute of Networking and Multimedia. His research interests include personal communications services.C. He was a recipient of Research Award for Young Researchers from Pan Wen-Yuan Foundation in Taiwan in 2004. Taiwan.D. Yuguang “Michael” Fang (S’92-M’94-S’96-M’97SM’99-F’08) received a Ph. He was elected as the Best Young Researcher. He received the National Science Foundation Faculty Early Career Award in 2001 and the Office of Naval Research Young Investigator Award in 2002. degree from National Chiao Tung University. Dr.S.