CHAPTER 1 Introduction
3.4 Numerical results and Discussions
In the analysis below, we assume the spread spectrum bandwidth (W) of the WCDMA network is 5 MHz, the uplink full-rate transmission is 128kbps and the half-rate 64kbps.
The size of the NUC WQ (B) is 4 and the size of FRUC WQ (Q) is 10. The mean service time of NUCs (1/µn) is assumed to be 2 minutes and the mean service time of FRUCs (1/µf ) is assumed to be 10 minutes. The arrival rate of FRUCs (λf) is fixed at 0.01 connections/sec.
and that of NUC (λn) varies in the range of 0.005-0.03 connections/sec, i.e., ρn varies in the range of 0.6-3.6 connections. We compare four connection schedulers: SAll, SNRA, SNPrm and
S-NWQ. The iterative algorithm for each scheduler has been developed in C language. The program was run on a laptop PC with 1.6GHz Pentium CPU and 512MB RAM. For each traffic load, the stationary state probabilities can converge in less than one minute.
To choose a suitable α value for the cost function in (3.31), we use the operation data from ChungHwa Telecom (CHT) in Taiwan, and make assumptions if operation information is unavailable. In (3.29), D (the 128kbps transmission charge per hour) is NT$562.5, E=60 (i.e., the number of busy hours per day equals to 2), F (the number of cells) is 1000. The number of flat-rate users, X, is 2 hundred thousands, Y is assumed to be 0.001 (i.e., one out of a thousand users would quit per month due to a percentage increase of blocking probability above β, and Z (the monthly fee of a flat-rate user) is NT$850. β should be chosen to reflect the level of user dissatisfaction; it was chosen to be 0.02, which is the target blocking probability for flat-rate subscribers of CHT. Given that, we can obtain the factor α = 0.504.
Note that α is less than ρn (0.6-3.6) in our experiments, i.e., the cost weighting factor of FRUCs is less than that of NUCs.
Fig. 3.6.a plots the cost function as NUC traffic increases. The FRUC traffic is fixed at 0.01 connections/sec. The cost function represent the revenue loss of the operator; the less the better. The results indicate SNRA has the least amount of revenue loss among all schedulers.
36
When the NUC traffic less than 0.015 connections/sec, the revenue losses of SAll, SNWQ, and SNPrm are as small as that of SNRA, but the losses rise rapidly as the NUC traffic increases above 0.02 connections/sec, in particular for SNPrm. This indicates when the system traffic load is high, waiting queues and preemption are necessary, but rate-adaptation is not. This is because sub-rated connections are less "bandwidth efficient," and results in system throughput reduction and revenue loss. SNPrm suffers the biggest revenue loss when the NUC traffic load is high. This indicates that preemption is essential in reducing the revenue loss.
Fig. 3.6.a: The cost function (C) with α=0.504 and β=0.02, (B=4, Q=10) channel is needed because of the same reason that guard channels leads to system throughput
0.00
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
37
reduction and revenue loss.
Fig. 3.6.b: The numbers of guard channels (GC) with α=0.504 and β =0.02, (B=4, Q=10)
Fig. 3.7.a plots the blocking probabilities of NUCs as the NUC arrival rate increases. All schedulers provide very low blocking probabilities for NUCs, except SNPrm; the NUC blocking probability of SNPrm increases more rapidly as NUC traffic increases. This is because sub-rated FRUCs are lessefficient in using spectrum. If FRUCs can only be sub-rated, but cannot be preempted, there would be more sub-rated FRUCs when the system traffic load is high. As a result, the overall system throughput decreases, and more NUCs are blocked. Therefore, preempting FRUCs is essential in reducing the blocking probability of NUCs. When the NUC traffic is high and no NUC waiting queue is used (as in SNWQ), the blocking probability slightly rises. This indicates the NUC waiting queue is necessary when the system load is close to its capacity.
0
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
38
Fig. 3.7.a: Average NUC blocking probabilities (PBN) with B=4, Q=10
Fig. 3.7.b plots the blocking probabilities of FRUCs as the NUC arrival rate increases.
The results indicate that the FRUC blocking probabilities of SNRA and SNPrm are about the same;
SNRA outperforms SNPrm by a small margin. Even though FRUCs cannot be preempted in SNPrm, the blocking probability of SNPrm is still higher than that of SNRA. This is also because sub-rated FRUCs are less "bandwidth efficient". In addition, the FRUC blocking probabilities of SAll and SNWQ are higher and rise more rapidly as NUC traffic increases, because FRUCs are impaired by both preemption and sub-rating. The fluctuations of FRUC blocking probabilities in SNWQ, when the NUC traffic increases from 0.01 to 0.015 connections/sec, are caused by the change in the number of guard channels
0.0%
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
39
Fig. 3.7.b: Average FRUC blocking probabilities (PBF) with B=4, Q=10
Fig. 3.8.a plots the average waiting times (i.e., queueing times) of NUCs as the NUC traffic increases. The waiting times of NUCs in schedulers SAll and SNRA are very insignificant under all traffic loads, i.e., NUCs are rarely queued. This is because serving FURCs can be preempted to free radio resources. If FRUCs cannot be preempted, such as in SNPrm, the average waiting time of NUCs increases steadily as the traffic of NUCs increases.
Fig. 3.8.a: Average NUC waiting times (WTN) with B=4, Q=10
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
40
Fig. 3.8.b plots the average waiting times of FRUCs as the NUC traffic increases. The waiting times of all schedulers show the same trend of rising as the NUC traffic increases.
Even when the system traffic is low, the average waiting time of FRUCs in SNRAis as large as 60 seconds, which is unacceptable for real-time applications. SNPrm provides the shortest waiting time, while SNRA the longest. The difference can be as high as 100 seconds when the NUC traffic is 0.03 connections/sec. Note that the fluctuations of FRUC waiting times in SNWQ, when the NUC traffic increases from 0.01 to 0.015 connections/sec, are also caused by the change in the number of guard channels. This change of guard channels also results in fluctuations of SNWQ results in later figures.
Fig. 3.8.b: Average FRUC waiting times (WTF) with B=4, Q=10
Fig. 3.9.a depicts the probability that a NUC is queued. The results indicate that NUCs in SAll and SNRA are very rarely put into the waiting queue because FRUCs can be preempted to free radio resources. On the other hand, NUCs in SNPrm are more likely to be queued. The probability that a new NUC is queued increases steadily and rapidly as the traffic load increases. The NUC queueing probability in SNPrm can be as high as 50%. This indicates that
0
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
41
preempting FRUCs is critically in reducing the queueing probability of NUCs. Fig. 3.9.b depicts the queueing probabilities of FRUCs as the NUC traffic increases. In general, the probability that a FRUC is queued increases as the traffic load increases. SNRA has the largest FRUC queueing probability, because FRUCs cannot be sub-rated. Other schedulers provide about the same queueing probabilities under all traffic loads.
Fig. 3.9.a: Average NUC queueing probabilities (PQN) with B=4, Q=10
Fig. 3.9.b: Average FRUC queueing probabilities (PQF) with B=4, Q=10
0%
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
NUC arrival rate (FRUC is fixed at 0.01 connections/sec.) 0%
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
NUC arrival rate (FRUC is fixed at 0.01 connections/sec.) SNPrm
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
NUC arrival rate (FRUC is fixed at 0.01 connections/sec.) 0%
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
NUC arrival rate (FRUC is fixed at 0.01 connections/sec.) SNPrm
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
42
Figs. 3.10.a and 3.10.b present the probabilities that a serving full-rate and half-rate FRUC would be preempted before completion. All schedulers display the same trend of rising preemption probabilities as the traffic load increases. The preemption probability of SNRA is lower than other schemes by a small margin. This is because sub-rated FRUCs are less bandwidth efficient.
Fig. 3.10.a: Average preempted probabilities of a serving full-rate FRUCs ( PFPrm) with B=4, Q=10
Fig. 3.10.b: Average preempted probabilities of a serving half-rate FRUCs ( PSPrm) with B=4, Q=10
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
43
Figs. 3.11 presents the probabilities that a serving full-rate FRUC would be sub-rated.
As the NUC traffic increases, the sub-rating probabilities of SAll and SNWQ first rise and then decline. The decline is because when the system traffic is high, FRUCs are more likely to be preempted. On the other hand, the sub-rating probability of SNPrm increases more rapidly and saturates later as the traffic load increases. This is because as the NUC traffic increases, SNPrm cannot preempt FRUCs; it can only sub-rate more FRUCs.
Fig. 3.11 Average sub-rated probabilities of a serving full-rate FRUCs (PFS) (B=4, Q=10)
Fig. 3.12 plots the average transmission rate of FRUCs. Since FRUCs may be sub-rated and/or preempted, the average transmission rate of FRUCs is reduced. In SNRA, no FRUCs are sub-rated. In SAll and SNWQ, a FRUC can be sub-rated and preempted; the average transmission rate is reduced to as much as 70% of the full rate transmission when the system traffic is
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
44
Fig. 3.12 Average transmission rate of serving FRUCs (TF) (B=4, Q=10)
3.5 Conclusions
This chapter, we investigate four combinations of scheduling techniques, queueing, guard channels, preemption and rate-adaptation, on their effectiveness in scheduling UMTS R99 uplink connections to reduce the revenue loss of the operators serving both normal and flat-rate users. We proposed a cost function representing the revenue loss due to both blocked normal user connections and lost flat-rate users. The optimum numbers of guard channels was determined by an iterative algorithm. The analytic results indicate when α , the cost weighting factor of flat-rate users, is less than ρn, queueing and preemption are essential for connection scheduling to maximize the revenue. Rate-adaptation is ineffective, because half-rate connections are less bandwidth-efficient. Sub-rating FRUCs reduced the system throughput and the operator revenue. In addition, no guard channel is needed, if queuing and preemption are used, because guard channels increase the blocking probability of FRUCs and reduces system throughput.
In this chapter, we consider uplink connection scheduling only. We did not study
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
0.005 0.01 0.015 0.02 0.025 0.03 NUS arrival rate (FRUS is fixed at 0.01 sessions/second)
SAll
45
needed for UMTS networks with α larger than ρn, which is possible when the number of flat-rate users increases or the normal user traffic decreases. In this situation, a more sophisticated scheduler is needed. The scheduler should give priority to NUCs when the FRUC blocking probability is below the departure threshold, β. When FRUC blocking probability is above the threshold, FRUCs should have priority.
46
CHAPTER 4
Flate-Rate Packet Scheduling for the WCDMA Systems with HSDPA
4.1 Introduction
A Universal Mobile Telecommunications System（UMTS）network consists of three interacting domains: the Core Network (CN), the UMTS Terrestrial Radio Access Network (UTRAN) and Mobile Stations (MSs). Fig. 4.1 depicts the network architecture of the UMTS network. The CN includes Circuited-Switch (CS) domain (i.e., MSC/VLR and GMSC) and Packet-Switched (PS) domain (i.e., SGSN and GGSN). The UTRAN includes multiple Radio Network Controllers (RNCs), each of which connects to multiple Node-Bs. The air interface of UTRAN is based on Wideband CDMA (WCDMA) technology and the details can be found in the 3GPP Release 99 specifications [27]. The peak transmission rate between a Node-B and a stationary mobile station (MS) is 2 Mbps.
To provide a higher data transmission rate for packet data services, WCDMA has evolved into High Speed Downlink Packet Access (HSDPA) described in 3GPP Release 5 specifications [28]. The HSDPA is expected to achieve a peak data rate over 10 Mbps, which is a significant improvement over the peak data rate (2 Mbps) of the 3G WCDMA Release 99.
The idea behind HSDPA is that the network transmits the downlink packets to the MS with maximum carrier-to-interference ratio (max. C/I) first at a high data rate. To enable HSDPA, the radio packet scheduler is moved from the radio network controller (RNC) to Node-Bs.
47
Fig. 4.1: The network architecture of the UMTS network
The packet scheduler in Node-B tracks the channel quality of each MS by measuring the SIR (Signal to Interference Ratio) on the CPICH (Common Pilot Indicator Channel) and allocates the High Speed Downlink Shared Channel (HS-DSCH) to the MS with the best SIR value [29-32]. As a result, the network can achieve the maximum throughput for downlink packets. To prevent the MSs with poor radio channel quality from starvation, traditional Round Robin (RR) packet schedulers can be used to ensure service fairness [33-34], but RR schedulers do not fully utilized the advantages of HSDPA. Proportionally Fair (PF) packet schedulers realize a reasonable trade-off between radio efficiency and fairness [35]. The network transmits downlink packets to the MS whose normalized instantaneous SIR value, the instantaneous SIR value divided by the average SIR value of the on-line transmission period, is the largest among all MSs. The numerical results show that both its system throughput and its worst case user throughput are larger than those of RR schemes.
However, the packet schedulers described above did not consider the revenues of mobile operators. When the mobile operators begins to provide flat-rate packet services to users, revenue, instead of fairness or capacity, is the most important consideration in the HSDPA network for the operators. Flat-rate users pay fixed monthly charge to access the
Node B RNCRNC HLR
CS Domain
SGSN
PS Domain
Internet PSTN
MSC/VLR
GGSN GMSC
MS
Node B RNCRNCRNCRNC HLR CS Domain
SGSN
PS Domain
Internet PSTN
MSC/VLR
GGSN GMSC
MS
48
HSDPA network without limiting the packet transmission. Since usage incurs no extra cost, flat-rate users could occupy most of the network radio resources. Without special treatments for different classes of user’s packets, users charged by usage may be blocked out from accessing the HSDPA network and compromise the mobile operator’s revenues. Therefore, it is important for a packet scheduler in a Node-B to fairly utilize the network resources for both the users charged by volume and the flat-rate users.
In this chapter, we study how to use packet scheduling techniques to control data packet transmission and guarantee the revenues of mobile operators without impairing flat-rate users too much. We consider two types of downlink packets, Charged Packets (CPs) and Flat-Rate Packets (FRPs). From the viewpoints of mobile operators, revenue and customer satisfaction need to be well balanced. To garner more revenue, the packet scheduler needs to give CPs a higher priority over FRPs. On the other hand, to ensure customer satisfaction, the dropped probability of FRPs needs to be kept below a certain threshold. In this paper, we present two enhanced packet schedulers that constantly monitor the dropped probabilities of both CPs and FRPs, and schedule down-link packet transmission so that the dropped probability of CPs could be below P1 and that of FRPs could be below P2. The scheduling techniques we used include a Priority Queue (PQ) with dynamic guard slots for CPs, and a PQ with Discard Timer (DT) for FRPs. Analytic models have been used to evaluate their performance in terms of packet dropped probability and downlink radio utilization.
4.2 HSDPA Basic Principles
Instead of the Downlink Shared Channel (DSCH) used in the WCDMA, HSDPA provides a new transport channel called High Speed DSCH (HS-DSCH) to transmit the downlink packets to MSs [28, 31]. In HSDPA, a large amount of radio resources can be assigned to a single MS on a Transmission Time Interval (TTI) basis. For each TTI (also referred to as a frame, a 2 ms interval), the Node-B selects an adequate Modulation Coding
49
Scheme (MCS), such as Quadrature Phase-Shift Keying (QPSK) or 16-Quadrature Amplitude Modulation (QAM), for each served MS according to the quality of downlink radio signal and the current system load. The better quality of downlink radio signal between the Node B and the MS is, the higher data rate of MCS can be selected. Each MCS value chosen for a served MS determines the data transmission rate for the served MS in the next TTI.
The High Speed Shared Control Channel (HS-SCCH) is a downlink control channel at a fixed rate (e.g., 60 kbps) and carries downlink signaling directing the HS-DSCH
transmission. It provides packet transmission timing and coding information, so that each served MS listens to the HS-DSCH at the correct time using the correct codes for its downlink packets.
Fig. 4.2 depicts the downlink Spreading Factor (SF) codes allocation tree for the HS-DSCH and HS-SCCH in an HSDPA network. The SF codes for the HS-DSCH and HS-SCCH with orthogonal character must be fixed at 16 and 128, respectively. There are at most 15 downlink SF codes for the HS-DSCH that can be assigned to one MS in a TTI to achieve an ideal peak rate of 14 Mbps when 16- QAM full rate MCS is used in a frequency band of 5MHz. A downlink SF code for the HS-SCCH can instruct only one MS to receive the downlink packets belonged to it, and there are at most four HS-SCCH codes can be used to control downlink packet transmission for all MSs. Other SF codes, except those for the HS-DSCH and HS-SCCH, can be assigned to transport voice calls in parallel with HS-DSCH data transmission or for non-HSDPA data transmission. As a result, in a TTI at most four MSs can be instructed by four HS-SCCH codes, and the selected MSs share 15 HS-DSCH codes to receive their downlink packets.
50
Fig. 4.2: Downlink SF codes allocation tree for HS-DSCH and HS-SCCH
Fig. 4.3 depicts an example downlink packet scheduling for four MSs, MS1-MS4, in a cell. Each MS can at most monitor four HS-SCCH codes and can only be assigned one HS-SCCH code belong to it in a TTI by a Node-B, and then in the assigned HS-SCCH code, the corresponding MS can be instructed to receive its downlink packets using the downlink HS-DSCH codes assigned to it. The time interval between the HS-SCCH instruction for a MS and its correspondent HS-DSCH transmission for this MS is 4/3ms. In the example depicted in Fig. 3, MS1 and MS2 are instructed to receive downlink packets in the first TTI, MS2 and MS3 to receive downlink packets in the second TTI, and all MSs to receive downlink packets in the third TTI.
Fig. 4.3: An example downlink packet scheduling for MS1-MS4 in a cell
SF=1
-Codes for the voice calls -or non-HSDPA data transmission SF=1
-Codes for the voice calls -or non-HSDPA data transmission
First TTI HSDPA HS-SCCH Control
Downlink HSDPA HS-SCCH Control
Downlink
51
4.3 System Models and Assumptions
According to the 3GPP specifications, when a MS creates a data session in the PS Domain, a SGSN can send Radio Access Bearer (RAB) parameters in the RAB assignment request message to the RNC to indicate the downlink packet priority [36-37]. In addition, a RNC in a HSDPA network can send the downlink packet scheduling policy to a Node-B, such as packet discard timer and scheduling priority, during the radio link setup procedure [38]. Fig.
According to the 3GPP specifications, when a MS creates a data session in the PS Domain, a SGSN can send Radio Access Bearer (RAB) parameters in the RAB assignment request message to the RNC to indicate the downlink packet priority [36-37]. In addition, a RNC in a HSDPA network can send the downlink packet scheduling policy to a Node-B, such as packet discard timer and scheduling priority, during the radio link setup procedure [38]. Fig.