Dynamic Fair Scheduling With QoS Constraints in Multimedia Wideband CDMA Cellular Networks

1. Dynamic Fair Scheduling With QoS Constraints in Multimedia Wideband CDMA Cellular Networks Liang Xu, Member, IEEE, Xuemin (Sherman) Shen, Senior Member, IEEE, and Jon W. Mark, Fellow, IEEE IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 1, JANUARY 2004 Speaker: Chun Hsu 許君

2. Outline • Introduction • System Model • Soft Capacity • CDGPS • C-CDGPS • Numerical Results • Conclusion • Comments

3. Introduction – Soft capacity • Wireless networks are expected to support multimedia traffic, and to make efficient use of the radio resource. • To efficiently support QoS for multimedia traffic in wireless networks is to employ a central scheduler. • The scheduler should be efficient in utilizing the radio resources and be fair in scheduling services. • An important feature of CDMA cellular networks is the so-called soft capacity. • The uplink capacity is subject to the variation of signal-to-interference ratios (SIRs) and bandwidth demands of. • For instance, if one user is allocated a large bandwidth while the other users are allocated relatively small bandwidths, the total achievable throughput could be larger than that for uniform bandwidth allocation

4. Introduction – Challenge and Goal • In order to avoid overbooking the capacity, a conventional scheduler may have to make a conservative estimate of the soft uplink capacity, • which may result in an inefficient utilization of the radio resources. • The total uplink throughput may have to be sacrificed for fairness. On the other hand, a short-term unfair bandwidth allocation may improve the total throughput. • Goal • to design a scheduler more efficient in utilizing the soft uplink capacity • to tradeoff the fairness with throughput

5. Introduction – Contribution • A formulation of the soft uplink capacity is given, where a “nominal capacity” is defined to interpret the uplink resource in each cell. • Based on the nominal capacity, adapt the CDGPS scheme to exploit the soft-capacity in a more efficient way. • A modified CDGPS scheme, credit-based CDGPS (C-CDGPS) is proposed. • The C-CDGPS scheme can improve the delay and throughput, as compared with the original CDGPS scheme. • Long-term fairness among all users can also be guaranteed due to the bounded user credits, although short-term unfairness may be introduced by the credit-based scheduling.

6. System Model • In this paper, maximum achievable throughout on all dedicated channels is referred as the uplink capacity. • Due to the multiple-access interferences (MAI) among uplink channels, the uplink capacity is considered to interference-limited, • i.e., the capacity is not limited by the number of available spreading codes. • In order to achieve a bounded delay for a user, it is required that each traffic source is shaped by a leaky-bucket regulator with parameter (σ, ρ), • where σ and ρ are token buffer size and token generation rate, respectively.

7. Soft Capacity • As the uplink capacity of a CDMA system is soft in nature, an important issue is how to formulate and deploy the capacity. • Let the spread bandwidth be W(?). Denote Ri the channel rate and Pi the received signal power of the ith user in the cell. • The SIR of the ith user can be written as • where Pn is the background noise power at the BS and Iinter is the intercell interference power from other cells. • The bit energy of user is Eb=Pi/Ri. • Ie be the equivalent spectral density of the interference plus background noise for user i. • The equivalent power of interference plus noise is WIe.

8. Soft Capacity • Then, the QoS at the BS receiver can be expressed in terms of Eb/Ie and (1) can be manipulated to yield • To achieve the target BER, a minimum Eb/Ie requirement, (Eb/Ie)0 need to be guaranteed for all users. • Let γi be the desired SIR (SIR threshold) of user i. For QoS satisfaction, γi must be greater than or equal to the specification ri0= (Eb/Ie)0Ri/W, so that the actual can be achieved as • The uplink capacity can be defined as • f((Eb/Ie)i) is an increasing function of the achieved which accounts for the effect of forward error control.

9. Soft Capacity • Consider a simplified system model, where only two users sharing the DS-CDMA uplink. • There are no other interference and background noise except for the MAI between the two users. R1A+R2A<R1B+R2B R1B<R2B R1A=R2A

10. Soft Capacity - Nominal Capacity(1/2) • From (1), if SIRi can be maintained at the desired level, for any , the minimal received powers are [24] • Since the power value should be positive and limited, the following necessary condition must be satisfied. • However, when the left-hand side of (6) is close to 1, the optimal power levels may be too high to be sustainable. Therefore, it is necessary to impose the inequality • We define (1-δ) as the nominal capacity.

11. Soft Capacity - Nominal Capacity(2/2) • Let . From(5), (7), the nominal capacity (1-δ) of a target cell can be determined using • In general, Pu can be set by the service provider. • As Pu represents a target limit on the sum of the received signal powers from all the mobiles within the cell, • the benefit of setting Pu is limiting the interference from this cell to other cells. • In practice, Pu can also be dynamically adjusted according to the location of mobiles. • For instance, if most of the mobiles move closer to the BS and, hence, their interference to other cells decrease, Pu can be increased accordingly.

12. CDGPS • The CDGPS scheme is based on the GPS fair scheduling discipline, and developed for the rate-scheduled wideband CDMA system. • To insure that the uplink system will not be overloaded during the dynamic scheduling, C has to be chosen as the lower bound of the soft capacity, • In (9), all the users N are persistently and simultaneously transmitting at the same channel rate, which gives a safe but conservative estimate of the uplink capacity.

13. CDGPS – delay bound and fairness • Let traffic flow i be regulated by a Leaky-Bucket with token buffer size σi and token generation rate ρi. • Theorem 1: If guarantee rate gi>=ρi, then the packet delay of flow i is bounded by • Short term fairness • CDGPS can guarantee each backlogged flow with at least the minimum rate which is proportional to its weight. • In addition, the surplus bandwidth is redistributed among all of the backlogged flows in proportion to their weights. • Long term fairness • comparing the delay bound (12) with the ideal GPS delay bound σi/gi, the difference between the two bounds is within T.

14. CDGPS – Resource Allocation for Soft Capacity (1/2) • To overcome the inefficient use of uplink capacity, they developed the following optimal resource allocation approach. • Let Bi(k) be the amount (in bits) of flow i traffic requested for transmission in slot k, the throughput Hi(k) of user during slot can be written as • The optimization problem is to choose the vectors S={γ1(k),…,γN(k)} and R={R1(k),…,RN(k)} , such that

15. CDGPS – Resource Allocation for Soft Capacity (2/2) • subject (Eb/Ie) to constraints • Capacity constraint (7) • GPS fairness constraints

16. CDGPS – Soft Capacity Allocation Procedure (1/2) This procedure is to find a solution for (14)

17. CDGPS – Soft Capacity Allocation Procedure (2/2)

18. Credit-Based CDGPS (C-CDGPS) • In some situation the short-term fairness may not be necessary. • For instance, for delay-insensitive data users, a short-term unfair allocation of resource in a slot may not degrade their QoS in terms of throughput. • A short-term unfairness may benefit user QoS and increase resource utilization, as long as the long-term fairness can still be guaranteed. • C-CDGPS scheduling scheme is introduced to improve the throughput of the delay-insensitive data users, and, at the same time, to guarantee the long-term fairness.

19. C-CDGPS - Credit • The credit of a user is increased if it does not receive as much as its deserved fair share of bandwidth in a time slot and is decreased if it receives more than its fair share. • Note that there has to be a limit on the credit accumulation, otherwise, a user with very large credit can capture bandwidth for a long period of time. • With the C-CDGPS, it can be shown that a credit bound exists so that the long-term fairness can be guaranteed. • The total amount of extra bandwidth which can be shared by greedy flows besides their guaranteed bandwidth is • B0={j:Sreq,j(k)>Gj} for the time slot k.

20. C-CDGPS – scheme strategy • and the credits of the flows in set B0 are updated by • If a greedy flow does not receive its fair share of e(k), it will receive more credits. • The strategy is to give the greedy flow that has the largest amount of credits the first priority to obtain extra bandwidth. • In the first iteration, all the other greedy flows are only given the guaranteed bandwidth. • If the user with the highest credit does not use up e(k), the user with the second highest credit will be given a share of e(k) in the next iteration. • The iterations continue until the extra bandwidth is used up and Ki(k) will be updated by (23).

21. C-CDGPS – Fairness and delay bound • Fairness is not guaranteed within each slot by the C-CDGPS rate-allocation procedure. • However, the short-term unfairness can be limited if the credits of greedy flows are bounded. • Since Kmax and Kdiffare bounded, Theorem 4 implies that the weighted fairness constraint (16) can be approximately satisfied in the long term.

22. C-CDGPS – delay bound

23. Simulation 1 • The uplink capacity is assumed to be a constant C=2Mb/s. • Four homogeneous best-effort packet data flows are considered, and assigned the same weight. • Static scheduler assigns a fixed channel rate that equals to the guaranteed rate in the CDGPS scheme, to each packet flow.

24. Average delay comparison (minimal delay) GPS performs hypothetically bit-by-bit scheduling which can instantly respond to the traffic variation. (better than static) CDGPS is more flexible in allocating bandwidth and can make use of idle network resource to improve delay performance.

25. Maximum delay comparison

26. Throughput comparison: CDGPS and GPS traffic arrival rates fluctuate: 1.25s-1.35s

27. Simulation 2 & 3 • In the second simulation, four homogeneous Poisson data traffic flows are simulated, each of which is guaranteed 1/4 of the capacity C. • The traffic load (normalized by C) is the sum of average arrival rates of the four data flows. • Nominal capacity (1-δ) =0.67 • In the third simulation, ten voice flows, one VBR video flows and four best-effort data flows are simulated. • Voice: ON-OFF model, weight=8 • VBR: eight-state MMPP model, weight=260 • BE: Poisson process model, weight=2

28. Throughput comparison

29. Average delay comparison

30. Maximum delay comparison

31. Maximum and minimum credits in C-CDGPS

32. Delay Performance Of Heterogeneous Traffic Scheduled By CDGPS

33. Conclusion • Efficient dynamic fair scheduling schemes have been proposed to support QoS of multimedia traffic in the uplink of CDMA cellular networks. • The analysis and simulation results show that bounded delay can be provisioned for real-time application by using the CDGPS service discipline with a high resource utilization, and weighted fairness can be assured among different users. • The resource utilization and delay can be further improved by C-CDGPS which exploits the soft uplink capacity.

34. Comments • Maybe we can compare the soft capacity under each type of distribution for service request, to find out the highest soft capacity. • Then, based on C-CDGPS, allocate the resource to approximate the highest soft capacity distribution.