Highest Urgency First (HUF): A latency and modulation aware bandwidth
allocation algorithm for WiMAX base stations
, Ying-Dar Lina
, Yuan-Cheng Laib
, Che-Wen Wua a
National Chiao-Tung University, 1001 University Road, Hsinchu, Taiwan 300, ROC
bNational Taiwan University of Science and Technology, 43, Sec. 4, Keelung Rd., Taipei, 106 Taiwan, ROC
a r t i c l e
i n f o
Received 14 December 2007
Received in revised form 26 October 2008 Accepted 2 November 2008
Available online 12 November 2008 Keywords: WiMAX Bandwidth allocation Algorithm Latency Modulation
a b s t r a c t
The mobile WiMAX systems based on IEEE 802.16e-2005 provide high data rate for mobile wireless net-works. However, the link quality is frequently unstable owing to mobility and air interference and there-fore impacts the latency requirement of real-time applications. In the WiMAX standard, the modulation/ coding scheme and the boundary of uplink/downlink sub-frames could be adjusted subject to channel quality and the trafﬁc volume, respectively. This provides us a chance to design a MAC-layer uplink/ downlink bandwidth allocation algorithm that is QoS/PHY-aware.
This work takes into account the adaptive modulation and coding scheme (MCS), uplink and downlink trafﬁc volume, and QoS parameters of all ﬁve deﬁned service classes to design a bandwidth allocation algorithm that calculates the slot allocation in two phases. The ﬁrst phase decides the boundary of uplink and downlink sub-frames by satisfying requests with pending latency violation and proportionating according to trafﬁc volume, while the second phase allocates slots to mobile stations considering urgency, priority and fairness. Simulation results show our algorithm achieves zero latency violation and higher system throughput compared to existing non-QoS/PHY-aware or less-QoS/PHY-aware approaches.
Ó 2008 Elsevier B.V. All rights reserved.
IEEE 802.16, known as WiMAX, is an emerging next-gener-ation mobile wireless technology standardized based on the cable network protocol, DOCSISfrom which it inherits some features such as the point-to-multipoint system architecture and Quality of Service (QoS) service classes. Different from its predecessor, Wi-MAX transmits data over the air interface rather than over the cable, so that mobility further speciﬁed in the 802.16e-2005, can be supported. The widely used Wi-Fiis point-to-multipoint and also supports mobility, however, arbitrary contentions for bandwidth lengthen the delay and degrade efﬁciency. To tackle this, WiMAX further separates the air interface into downlink (DL) and uplink (UL) channels and adopts a control center named base station (BS) for managing the DL/UL transmissions and allo-cating bandwidth for mobile stations (MSs1).
With the ever-growing bandwidth demand of time-sensitive multimedia applications, the bandwidth in wireless environment becomes relatively scarce. Though service classes and parameters
such as minimum reserved rate, maximum sustained rate and maximum latency, have been deﬁned in the standard for service differentiation, an appropriate bandwidth allocation algorithm is required in BS to achieve satisfactory quality along with the fol-lowing considerations. First, the Grant Per Subscribe Station (GPSS) scheme which is mandatory in the standard and more ﬂexible than the Grant Per Connection (GPC) in the DOCSIS. In GPSS the BS grants bandwidth to a MS rather than to certain connec-tion, so that the MS can respond to connections of different QoS requirements. Second, the modulation types and coding schemes (MCS) which decides the data rate between BS and MS and the translation from bytes to physical slots, shall be adaptive to the varying distance and air interference. Third, among all QoS requirements, the maximum latency is most critical to the quality of time-sensitive multimedia applications and thus should be properly satisﬁed.
A number of designs have been proposed to deal with the above-mentioned considerations. The MLWDF (Modiﬁed Largest Weighted Delay First)  is throughput-optimal and using the waiting time of head-of-line packet as scheduling metric for real time trafﬁc, but the QoS service classes are not involved. The Up-link Packet Scheduling (UPS) and DFPQ (Deﬁcit Fair Priority Queue)employ service classes to meet differentiation and fair-ness, while the TPPfurther uses the dynamic adjustment of the downlink (DL) and uplink (UL) to maximize the bandwidth 0140-3664/$ - see front matter Ó 2008 Elsevier B.V. All rights reserved.
*Corresponding author. Tel.: +886 922511430. E-mail address:email@example.com(Y.-N. Lin). 1
The terminal station is named subscribe station (SS) in the standard 802.16d-2004 for ﬁxed systems, and mobile station (MS) in the standard 802.16e-2005. Below we use MS to represent the terminal station.
Contents lists available atScienceDirect
utilization. However, they do not concern the physical-layer characteristics such as MCS. In, the authors cover this and Strict Priority is applied, though latency is ignored and starvation could occur easily for the low-level service classes even an admis-sion control scheme is installed.
In this work, a bandwidth allocation algorithm, Highest Urgency First (HUF), is proposed to tackle those challenges with Orthogonal Frequency Division Multiple Access with Time Division Duplex (OFDMA-TDD). OFDMA-TDD, the most prevalent physical-layer technology for the WiMAX systems, has high capacity owing to the OFDMA technique and ﬂexibility in the mobile environment. The algorithm consists of four steps: (1) translating the data bytes of requests to slots reﬂecting the MCS of every MS, and calculating the number of frames to satisfy the maximum latency for every re-quest of the service ﬂows; (2) pre-calculating the number of slots required by DL/UL requests which must be transmitted in these scheduled frame, and then deciding the portion of DL/UL sub-frame; (3) allocating the slots for every ﬂow using the U-factor, which indicates the latency, priority and fairness of every band-width request, and (4) allocating the slots of ﬂows to the corre-sponding MSs.
The rest of this work is organized as follows. Section2briefs the 802.16 PHY and MAC features and reviews related studies to justify our problems. Section3describes the detailed procedures of the proposed algorithm. Section 4 presents the simulation environ-ments and evaluation results. Finally, Section 5 concludes this work with some future directions.
Since the WiMAX supports high data rate and long distance in the mobile environment, rather than pure contention among MSs which causes signiﬁcant re-transmissions, a BS must coordinate the decision of transmissions from/to MSs which involves opera-tions in PHY and MAC. In this section, we sketch the WiMAX PHY features which affect the transmission data rate and therefore the bandwidth allocation, and describe the QoS consideration and scheduling ﬂow in the WiMAX MAC. Some related works
investigating the allocation problems are discussed, leading to the statement of the research goals.
2.1. Overview of the WiMAX system 2.1.1. PHY layer features
Orthogonal Frequency Division Multiplexing (OFDM) is a multi-plexing technology that sub-divides the bandwidth into multiple frequency sub-carriers and exploits the frequency diversity of the multi-path channel by coding and interleaving the information across the sub-carriers prior to transmission. The OFDMA, ex-tended based on the OFDM, further supports multiple accesses. Re-sources are available in OFDMA in the time domain in terms of symbols and in the frequency domain in terms of sub-carriers which are grouped into sub-channels. The minimum frequency-time resource unit is one slot which contains 48 data sub-carriers
and a slot duration of two symbols for DL while three symbols for UL in the mandatory PUSC (Partial Usage of Sub-Channels) mode. The 802.16 PHY supports TDD, Frequency Division Duplex (FDD), and Half-Duplex FDD modes. However, the TDD is preferred in WiMAX since it only needs one channel, enabling the adjust-ment of unbalanced DL/UL loads, while the FDD needs two chan-nels. Besides, the design of a transceiver is easier in TDD than in FDD.
As shown inFig. 1, an OFDMA-TDD frame is composed of (1) preamble for synchronization, (2) DL-MAP and UL-MAP for control and element information describing bursts for all MSs, and (3) the DL/UL data bursts carrying data for MSs. The amount of data car-ried in a slot varies with different adaptive MCS which decides the transmission data rate according to the link quality between the BS and MSs.Table 1summarizes the number of bytes in a slot for all supported MCSs in WiMAX. As an example, the slot capacity when BPSK and code rate of 1/2 are used is 48(bits) 1/2 = 3(by-tes) since a sub-carrier under BPSK carries 1 bit.
2.1.2. MAC layer with QoS
Five uplink service classes, the Unsolicited Grant Service (UGS), Real-time Polling Service (rtPS), Non-real-time Polling Service
S S +1 S +3 S+5 S+7 S+9 S +11 S+13 S +15 S+17 S+20 S +23 DL Burst 1 DL Burst 3 DL Burst 2 DL Burst 6 DL Burst 7 DL Burst 5 DL Burst 4 Preamble UL-MAP FCH DL-MAP UL-MAP UL Burst 1 UL Burst 2 UL Burst 3 UL Burst 4 RNG /BW -REQ
Downlink Subframe Uplink Subframe
Slot Duration Slot Duration
(nrtPS), Best Effort (BE), and the replenished Extended Real-time Polling Service (ertPS) are supported in the 802.16e-2005. A BS re-serves bandwidth for UGS ﬂows observing the maximum sustained rate, whereas for rtPS ﬂows it polls the MSs periodically according to the pdetermined time interval and receives bandwidth re-quests for further allocation. ertPS ﬂows are treated similarly to UGS except that MSs which the ﬂows belong to can further change the reservation size either by contending for chances or using pig-gyback request ﬁeld of management packets. nrtPS and BE contend for the transmission opportunities, but nrtPS has extra opportuni-ties to be polled, while BE depends only on contention. Among all service classes except the UGS and ertPS which are provided with sufﬁcient bandwidth, the rtPS must be much concerned since it supports real-time applications having the maximum latency requirement and variable packet sizes. Table 2 summarizes the characteristics of these service classes.
The scheduling ﬂows within BS and MS are shown inFig. 2and elaborated as follows. While the DL scheduler in a BS simply dis-tributes DL data to MSs, the UL scheduler needs to reserve grants for MSs for the UGS and ertPS ﬂows as well as for the UL bandwidth requests of rtPS, nrtPS and BE ﬂows submitted through polling or contention. The scheduling results are then passed to the frame builder, in which the DL-MAP/UL-MAP is generated. The DL-MAP/ UL-MAP portrays the DL/UL sub-frame information to notify the PHY layer when to send/receive data bursts. As for the MS side, the scheduler schedules the UL data based on the number of granted slots documented in the UL-MAP. Obviously, the band-width allocation algorithm exercised by the BS’s scheduler is crit-ical and must be designed carefully in order to optimize the system performance.
2.2. Related work
A number of works concerning the bandwidth allocation over IEEE 802.16 can be found. Andrews and Kumaranpropose the MLWDF to maximize the channel capacity for multiple MSs per-forming real-time applications to support QoS. It uses the head-of-line packet’s waiting time or the total queue length as the scheduling metric for throughput optimality and satisfaction with delay requirement. Wongthavarawat and Ganzpropose the Up-link Packet Scheduling (UPS) for service differentiation. It exploits the Strict Priority to select the target class to be scheduled, in which each service class adopts a certain scheduling algorithm
for its own queues. However, this scheme only concerns the uplink and hence the overall bandwidth is suffered and low priority clas-ses tend to suffer from starvation. The Deﬁcit Fair Priority Queue (DFPQ) revises the UPS by replacing the Strict Priority with the use of maximum sustained rate as the deﬁcit counter for the transmission quantum of every service class, and therefore can dynamically adjust the DL and UL proportion according to the counters. Nevertheless, this scheme is suitable only for the GPC mode and setting an appropriate maximum sustained rate is not trivial. The Two Phase Proportionating (TPP)introduces a simple approach to dynamically proportionate the DL and UL sub-frames and considers the minimum reserved rate, maximum sustained rate and the requested bandwidth of service classes in terms of the A-Factor to grant the bandwidth for MSs proportionally. How-ever, it could lead to inappropriate grants owing to the proportion. All above schemes do not consider the MCS which affects the trans-mission data rate and the service quality. Sanyenko’s approach
involves the MCS, but does not provide the latency guarantees. 2.3. Problem statement
To integrate all features in WiMAX PHY and QoS service classes and solve the above-mentioned problems, a well-designed algo-rithm is demanded to satisfy the following metrics. First, it must be aware of the adaptive MCS in PHY and translate the requested bandwidth to appropriate number of slots to meet the bandwidth demand. Second, the QoS requirements of service classes, such as minimum reserved rate, priority and maximum latency, need to be satisﬁed. Among them the maximum latency guarantee is most important for real time applications belonging to the rtPS class. Third, for fairness, the allocation algorithm should serve the service classes fairly to avoid starvation, given the presence of an admis-sion control scheme. In this article, an operational admisadmis-sion con-trol is assumed, since without it all bandwidth allocation algorithms will inevitably subject to starvation. The problem state-ment leads to designing a modulation, latency and priority-aware downlink and uplink bandwidth allocation in a WiMAX BS. 3. Highest Urgency First
This section elaborates the concept and procedures of the pro-posed Highest Urgency First (HUF) algorithm. The HUF uses an ur-gency parameter considering latency, fairness and priority to Table 1
Slot sizes of different MCSs in WiMAX.
Modulation BPSK QPSK 16QAM 64QAM
Code rate 1/2 2/3 3/4 5/6 1/2 2/3 3/4 5/6 1/2 2/3 3/4 5/6 1/2 2/3 3/4 5/6
Bytes 3 4 4.5 5 6 8 9 10 12 16 18 20 18 24 27 30
Service classes and the corresponding QoS parameters.
Feature UGS ertPS rtPS nrtPS BE
Request size Fixed Fixed but changeable Variable Variable Variable
Unicast polling N N Y Y N Contention N Y N Y Y QoS parameters Min. rate N Y Y Y N Max. rate Y Y Y Y Y Latency Y Y Y N N Priority N Y Y Y Y
Application VoIP without silence suppression, T1/E1
Video, VoIP with silence suppression
Video, VoIP with silence suppression
FTP, web browsing
E-mail, message-based services
schedule all requests, and divides the allocation procedure into two phases. The ﬁrst phase determines the bandwidth of DL/UL sub-frame while the second phase allocates bandwidth for re-quests from MSs. Each phase manipulates different metrics to achieve high throughput, latency guarantee and fairness.
3.1. Overview of the algorithm
Our approach aims at dynamically adjusting and ﬁlling up the DL/UL sub-frames in TDD mode, while each sub-frame is further allocated to service queues of different QoS requirements such as latency, priority and fairness. Slots in a frame can carry different amount of data owing to the MCS in PHY, and the varying data rate may further affect how bandwidth allocation is performed. Based on the above characteristics, the HUF is proposed to well utilize the bandwidth. An urgency parameter which considers three met-rics, namely latency in terms of deadline, fairness in terms of num-ber of requested slots and priority of service ﬂows, is used to decide the servicing order of all data/requests. The deadline repre-sents the number of frame durations left before an uplink request or a downlink packet must be served. A request having a deadline equaling to one must be dispatched in this frame so as to satisfy the latency requirement. The other two contribute to the urgency in terms of the urgency factor, i.e. U-factor, in which a higher value indicates a more urgent request. While the priority is trivial as being a metric, the rationale behind the employment of number of requested slots is that, requests demanding large amount of bandwidth shall be allocated as early as possible. They are rela-tively hard to be scheduled compared to requests of small amount and therefore tend to miss the deadline.
The HUF consists of two phases, ﬁrst of which decides the size of DL/UL sub-frames based on the minimum reserved rate, the data/requests whose deadline equals to one and other non-urgent demand, while in the second phase DL and UL independently dis-patches its own bandwidth to the individual queues of DL and UL according to the minimum reserved rate of every service queue,
data/requests in queue whose deadline equals to one, and the U-factor of the data/request. Finally, the HUF follows GPSS by grant-ing the accumulated bandwidth of ﬂows to the correspondgrant-ing MSs. The components and operations of the HUF algorithm are illus-trated inFig. 3and explained in Section3.2.
3.2. Detailed procedures of the HUF algorithm
3.2.1. Data/request translation and deadline determination
In the uplink, a service ﬂow in MSs expedites a bandwidth re-quest to BS whenever necessary, while in the downlink data are en-queued, scheduled and ﬁnally sent down to MSs. The transmis-sion unit in WiMAX is a slot whose capacity depends on the cur-rent MCS. Therefore, when a new frame starts, according to the MCS the required size of data/requests is ﬁrstly translated into number of slots as
#of slots ¼ BQS
bytes per slot; ð1Þ
where the BQS denotes the requested size, and bytes_per_slot repre-sents the capacity of a slot. Since a slot contains 48 data sub-carriers in Mobile WiMAX PHYand the MCS decides the number of bits carried in a sub-carrier, we can thus have
bytes per slot ¼48 Mod bits Coding rate
8 : ð2Þ
Regarding the service classes such as UGS, ertPS and rtPS, the maximum latency parameter is expected to be guaranteed for real-time applications. Thus, in this algorithm the deadline is de-ﬁned as
deadline ¼ ML
where ML means the maximum latency for the service ﬂow and FD represents the frame duration. If the maximum latency is not set in the service ﬂow, the deadline of the requests belonging to that ﬂow Downlink Scheduler Uplink Scheduler MS Scheduler Frame Builder Uplink sub-frame Downlink sub-frame frame QoS Parameters QoS Parameters Downlink queues Bandwidth request queues (Virtual) MS queues
Uplink grants Uplink data
Downlink Data Uplink Data
BS MAC Layer
MS MAC Layer
is set to 1. Otherwise, the corresponding deadline is calculated upon the arrival of a data/request, and then decreased by one after a frame duration. A deadline equaling to zero indicates the violation of the maximum latency requirement.
3.2.2. First phase: DL/UL sub-frame allocation
In order to ﬁll up the frame to achieve high throughput while considering the latency requirement for the service ﬂows, the HUF uses the urgent data/requests with deadline equaling to one and non-urgent data/requests as the metrics to decide the DL/UL sub-frame size. Besides, the minimum reserved rate is also neces-sarily considered. Detailed procedure to decide the DL/UL ratio is as follows:
(i) For DL and UL, respectively, sum up the number of data/ request slots whose deadline equals to one in all queues so as to reserve bandwidth for those that must be served in this frame.
(ii) For DL and UL, respectively, sum up the amount of slots translated from the minimum reserved rate of every service ﬂow. Exclude those that have been considered in i. (iii) Sum up the number of reserved slots calculated from i and ii.
Divide them by the number of DL/UL sub-channel in a slot duration to obtain the amount of symbols to be reserved. Notably in PUSC mode a slot duration spans two symbols in DL yet three in UL.
(iv) The amount of remaining symbols is thus calculated by sub-tracting the number of reserved symbols from the total number of symbols in a frame. Proportionate the remaining symbols for the DL and UL according to their amount of bandwidth requested by data/requests having deadlines lar-ger than one. Letting DR and UR represent the above requested bandwidth for DL and UL, respectively, the pro-portion can be derived as
UR DR¼ ðSrem ðSDDL xÞÞ=SDUL x ¼ Srem ðSDDL xÞ SDUL x ; ð4Þ
where Sremindicates the number of remaining symbols and SDDL and SDUL represents the number of symbols in a DL and UL slot duration, respectively. x which is the number of slot durations DL obtains can be found after solving the equa-tion, in whichSremðSDDLxÞ
SDUL speciﬁes the amount of slot
dura-tions distributed to the UL.
In short, the HUF reserves symbols for data/requests which must be served in this frame, and then proportionates the remain-ing symbols for the non-urgent data/requests to decide the DL/UL sub-frame size.
3.2.3. Second phase: urgency-based bandwidth allocation
After the DL and UL sub-frame sizes are determined in ﬁrst phase, the HUF scheduler starts to allocate independently the bandwidth of DL/UL sub-frame to MSs. The essence of HUF is to ensure the requirements of maximum latency and priority among all service ﬂows, and allocate the bandwidth to MSs fairly. Hence, HUF allo-cates the bandwidth in the precedence based on that requested slots whose deadline is one and satisfying the minimum reserved rate of every ﬂow. Then, when there is bandwidth left in a sub-frame, HUF deﬁnes the U-factor to select the other data/requests to be served. The allocation procedure in the uplink is portrayed as follows:
(i) For each service ﬂow, allocate bandwidth ﬁrstly to requests whose deadline equal to one and then to others until the minimum reserved rate is complemented.
(ii) Calculate the average-U-factor for every service ﬂow. (iii) Identify the ﬂow with the highest average-U-factor and serve
its head-of-line request. Recalculate the average-U-factor and repeat step iii until.
The average-U-factor of a service ﬂow can be derived as average-U-factor ¼ Pn i¼1U-factori n ; where ð5Þ U-factori¼ Ni ðP þ 1Þ Di ; ð6Þ
indicates the urgency of the ith request in the ﬂow and n represents number of requests. As shown in Eq.(6), the U-factoricomprises three metrics, namely Di, P and Ni. Dimeans the deadline of the ith bandwidth request. For ﬂows not having a deadline, the HUF automatically associates them with a value which is the maximum deadline among all UL requests. P stands for the ﬂow priority, which is deﬁned in the 802.16 standard and ranges from zero (lowest) to seven (highest). Niis the number of slots translated from the re-quested size. Once the head-of-line requests of all queues are dis-patched, the HUF performs step ii, namely recalculating the average-U-factors and so forth, repeatedly until the UL sub-frame is fulﬁlled. The downlink is treated similarly to the uplink. Fig. 3. Procedure of the Highest Urgency First (HUF).
3.2.4. Grant bandwidth to MSs
After allocating bandwidth to data/requests of each queue, the HUF scheduler further distributes the bandwidth to every MS by
totaling up the allocated bandwidth of the service queues of the same MS. Based on the grants, the scheduler generates the corre-sponding DL and UL MAPs which are sent every frame to notify the MSs of the transmission and receiving opportunities. Finally the HUF updates the deadline of every data/request by Deadline = Deadline 1.
This section elaborates an example of the HUF, in which the parameters and system proﬁle frequently adoptedare shown inTable 3. It is assumed that both DL and UL have four queues and the minimum reserved rates are 34, 30, and 20 slots for Queue 1, 2, and 3, respectively. In the ﬁrst phase, HUF decides the DL/UL sub-frame sizes. According to the requests whose deadline equals to one and the aggregated number of slots for the minimum reserved rate of all DL/UL ﬂows, ((90 + 20)/10) 2 = 22 and ((64 + 20)/ 12) 3 = 21 symbols are reserved for DL and UL, respectively, with 72 21 22 = 29 symbols remained. The HUF then proportionates the remaining symbols for DL and UL by solving Eq.(4)where DR is 50 + 50 + 20 20 = 100 and UR is 40 + 30 + 25 20 = 75. So, DL ob-tains additional x ¼ 10029
753þ1002ﬃ 7 slot durations equaling to
2 7 = 14 symbols and UL obtains additional2927
3 ¼ 5 slot dura-tions equaling to 3 5 = 15 symbols. Finally, the sizes of DL and UL sub-frames are 22 + 14 = 36 and 21 + 15 = 36 symbols, respectively.
(a) DL/UL requested slots and (b) system proﬁle in the example.
Direction Type Number of requested slots
UL (sum of all UL queues) deadline = 1 64 deadline = 2 40 deadline = 5 30 deadline = 7 25
Min. Rev. Rate 34 + 30 + 20 64 = 20 DL (sum of all DL queues) deadline = 1 90
deadline = 3 50 deadline = 4 50 deadline = 6 20 Min. Rev. Rate 20
PHY parameter Value
Number of symbols in a frame 72
Number of UL sub-channels 12
Number of DL sub-channels 10
UL slot duration (symbols) 3
DL slot duration (symbols) 2
In the second phase, the HUF allocates slots to DL and UL re-quests, respectively. Take the UL as an example, while 36
3 12 ¼ 144 slots have been allocated in the ﬁrst phase, in the second phase the bandwidth is reserved for requests whose dead-line equals to one and also for the minimum reserved rate of the queues. This is accomplished by 144 (34 + 20 + 10) (0 + 10 + 10) = 60 slots, since the HUF only needs to allocate addi-tional 34 34 = 0 slot for Queue 1, 30 20 = 10 slots for Queue 2, and 20 10 = 0 slot for Queue 3. Therefore, 60 slots are left to be allocated to queues according to their average-U-factors, in which the queue of the largest average-U-factor is served ﬁrst. As shown inFig. 4in which priority of each queue is conﬁgured to 0, the average-U-factor of all queues are calculated as
ð20ð0þ1Þ2 þ25ð0þ1Þ7 Þ=2 ﬃ 6:79, 5ð0þ1Þ5 ¼ 1, and25ð0þ1Þ5 ¼ 5 for Queue
1, 2, and 3, respectively. Thus, the HUF selects the head-of-line re-quest in Queue 1 to serve ﬁrst, and recalculates the average-U-fac-tor of Queue 1 as25ð0þ1Þ
7 ﬃ 3:57. Similar procedures are executed until the sub-frame is fulﬁlled. Finally the HUF calculates the total number of slots each queue has just gained which are 34 + 20 + 15 = 69, 20 + 10 = 30, and 10 + 10 + 25 = 45 for Queue 1, 2, and 3, respectively, and then grant them to every MS, namely 69 + 45 = 114 for MS1, and 30 for MS2. Finally, the deadline of all requests is decreased by one before entering the next frame.
4. Evaluation results
The HUF algorithm is evaluated using the OPNET simulator with the WiMAX module developed by the INTEL Corp. The evaluation scenarios cover the MCS awareness, latency-aware dynamic adjustment, latency guarantee, and fairness in service classes. Each scenario considers a set of algorithms supporting certain function-ality. For example, the DFPQ and MLWDF are involved when dis-cussing the latency guarantee, while the DFPQ, TPP and UPS are considered for the fairness. Furthermore, only the rtPS and BE are involved in the evaluation because the UGS as well as ertPS is granted with ﬁxed bandwidth, and the nrtPS differs from the BE merely in the priority.
4.1. Simulation environment
The simulation topology is depicted in theFig. 5. A number of MSs and a BS are connected via a gateway to a video conference endpoint and an FTP server.
The video conference application used in the simulation has variable packet size and is constrained by the latency requirement.
The WiMAX system proﬁle and application parameters are summarized inTable 4(a) and (b), respectively.
4.2. Modulation-aware allocation
Whenever the MCS is changed due to interferences, for consis-tent video conferencing quality the data rate of MSs is sustained by granting each of them adapted number of slots.Table 5depicts the setup of the modulation awareness test of HUF, in which two MSs whose MCSs change along with time, are involved. FromFig. 6we observe that though the modulation alters, the throughput is still kept the same. This is because more slots are granted as the capac-ity of a slot shrinks due to an un-scalable MCS. Notably the system is not stable during the ﬁrst 10 s because of performing Network Entry.
4.2.1. Latency-aware dynamic DL/UL adjustment
Dynamic DL/UL adjustment considering the latency require-ment not only maximizes the link utilization but retains the qual-ity of real-time applications. In this section, we evaluate the latency-aware adjustment supported by the HUF and compare its performance with the static approach and TPP. Six MSs are dedi-cated to downloading ﬁles using FTP with BE while an increasing number of MSs performing video conferencing with rtPS are adopted to enlarge the link load. Proﬁles of the applications are conﬁgured according to Table 4. Throughput and violation rate are investigated and shown in Fig. 7. Violation rate, deﬁned as the ratio of the number of packets whose delay exceeds the maxi-mum latency requirement to the number of all packets, is used to judge whether the adjustment is latency-aware.
As depicted inFig. 7(a), the throughput of dynamic adjustment, whether using TPP or HUF, is about 7% higher than the static adjustment when overloaded with 41 MSs. This is due to the fact that the former dynamically exploits the bandwidth according to the DL and UL trafﬁc loads, while the latter tends to abuse link re-sources because of not concerning the actual requirement.Fig. 7(b) shows that the degraded throughput of static adjustment contrib-utes to the increased violation rate. Although the TPP has similar throughput to HUF, its violation rate is considerably higher than that of HUF, whose rate is close to zero. This is because the TPP de-cides the DL/UL allocation simply by considering their loads, while the HUF further reserves bandwidth for requests that must be served in the current frame.
Fig. 5. Simulation topology.
(a) System proﬁle and (b) application parameters in the simulation.
System parameter DL UL
System bandwidth 1.5 MHz
FFT size 1024
Frame duration 5 ms
Useful symbol Time (Tb= 1/f) 60ls
Guard time (Tg= Tb/8) 7ls
OFDMA symbol duration (Ts= Tb+ Tg) 67ls
Sub-channels 10 12
Number of slot per sub-channel 1 1
Number of symbols per slot 2 3
Application Parameter (b)
Video conference Frame size:
– Lognormal distribution – Average: 4.9 bytes
– Standard deviation: 0.75 bytes
Frame inter-arrival time:1 30s
4.3. Latency guarantee with different requirements
We compare the latency-awareness of the proposed algorithm with the MLWDF, which is throughput-optimal and considers the waiting time of head-of-line packet to meet the latency quarantee, and with the DFPQ which uses EDF  for rtPS to satisfy the requirement. The evaluation scenario involves two ﬂows of the vi-deo conference application referencing toTable 4(b). Among the QoS parameters of the two ﬂows presented in Table 6, only the maximum latency is conﬁgured differently to 50 and 150 ms, respectively. The load of the link is increased by simultaneously increasing the input of the ﬂows.
While throughput and average latency are the general criteria to evaluate the performance of a bandwidth algorithm, the viola-tion rate is taken into account to estimate the satisfacviola-tion with dif-ferent latency requirements, as discussed inFig. 8involving three algorithms. The criteria of the evaluation are throughput, average latency of packets and violation rate. The throughput and average Table 5
Setup of the modulation awareness test. The MCS changes along with time.
Modulation BPSK QPSK 16QAM 64QAM
Coding scheme 1/2 1/2 3/4 1/2 3/4 1/2 2/3 3/4
Bytes per slot 3 6 9 12 18 18 24 27
Time period for MS1 0 10 10 15 15 20 20 25 25 30 30 35 35 40 40 50
Time period for MS2 40 50 35 40 30 35 25 30 20 25 15 20 10 15 0 10
0 5 10 15 20 25 30 35 40 45 50 55 60 65 5 10 15 20 25 30 35 40 45 50
s0 5 10 15 20 25 30 35 40 Th ro u g h p ut (kb p s)
# of granted slots for M S2 # of granted slots for M S1 Thrput of M S1 Thrput of M S2
Fig. 6. Modulation-aware allocation: throughput is sustained as the MCS changes.
0 10 20 30 40 50 60 70 80 90 10 0 36 37 38 39 40 41 # of MSs
STA460 480 500 520 540 560 580 600 36 37 38 39 40 41 # of MSs
s)HUF TPP STA
latency are the general criteria to evaluate the performance of a bandwidth allocation algorithm. Besides, the evaluation scenario focuses on the satisfaction with different latency requirements, and thus takes the violation rate into account. Fig. 8 discusses the throughput as well as the latency of three algorithms. From
Fig. 8(a) we can observe that generally the throughput increases
as more MSs participate in. However, performance of the DFPQ starts to degrade when the number of MSs reaches 32. This is be-cause the EDF, which is an optimal scheduling algorithm in re-source sufﬁcient environment, malfunctions when overloaded
. The corresponding average latency inFig. 8(b) is thus found to exceed 1000 ms suddenly from 32 MSs in DFPQ. The throughput is similar between the MLWDF and HUF, though the average la-tency differs noticeably when number of MSs approaches 30 since the MLWDF only considers the waiting time of the head-of-line packet, resulting in high average latency when heavily loaded. The HUF achieves high throughput while retaining low average latency.
Fig. 8(c) further examines the violation of the three algorithms in latency. Even when the number of MSs comes to 34, the HUF has no violation of the maximum latency being 50 and 150 ms. Never-theless, the violation rate of MLWDF grows drastically when 28 Table 6
QoS parameters of two trafﬁc ﬂows.
QoS parameter Flow #1 Flow #2
Service class rtPS rtPS
Minimum reserved rate (bps) 2400 2400
Maximum sustain rate (bps) 1,000,000 1,000,000
Maximum latency (ms) 50 150
Polling time (ms; speciﬁc to the rtPS class) 20 20
0 5 10 15 20 25 30 35 40 4 8 12 16 20 24 28 30 32 34 # of MSs Thr oughput ( kbps ) DFPQ MLWDF HUF 0 10 20 30 40 50 60 70 80 90 4 8 12 16 20 24 28 30 32 34 # of MSs Violation Rate (%) DFPQ_50ms DFPQ_150ms M LWDF_50ms M LWDF_150ms HUF_50ms HUF_150ms 0 200 400 600 800 1000 1200 1400 4 8 12 16 20 24 28 30 32 34 # of MSs Latenc y (ms) DFPQ M L W D F HUF
MSs are involved and is close to 70% and 80%, respectively, when maximum latency is conﬁgured to 150 and 50 ms and 34 MSs are present. This indicates that considering the head-of-line pack-et’s waiting time may not be sufﬁcient to guarantee the latency requirement. The DFPQ has a violation rate of 58% for 50 ms and 78% for 150 ms for 34 MSs resulted from the degraded throughput. 4.4. Fairness
A bandwidth allocation algorithm is said to be fair if the differ-ence in normalized services received by different ﬂows in the scheduler is bounded. In the evaluation which compares the fairness of the HUF with DFPQ, TPP and UPS, two sets of MSs are involved with one performing rtPS-based video conferencing and the other uploading ﬁles via BE-based FTP. The application proﬁles are shown inTable 4(b) while the parameters of service classes are presented inTable 7.
The fairness between rtPS and BE can be formulated as
Fairnessr b¼ ThrtPS SrtPS ThBE SBE ½8; ð7Þ
where SrtPSand ThrtPSare the requested bandwidth and the corre-sponding throughput of rtPS, yet SBEand ThBEare those of BE. The results are depicted in Fig. 9, in which small values suggest fair allocation.
Fig. 9(a) shows that TPP and HUF are fairer than DFPQ and UPS. That is because the UPS uses Strict Priority to allocate bandwidth to all service classes in which BE tends to get starved as the rtPS be-comes demanding. In DFPQ, the maximum sustained rate is em-ployed as the Deﬁcit counter; however determining an appropriate maximum sustained rate for all service classes is not trivial. Thus, if the maximum sustained rate is not conﬁgured prop-erly, the fairness suffers.Fig. 9(b) further explains the results. As shown in the ﬁgure, all approaches allocate fairly, namely 17% for rtPS and 83% for BE, when four MSs are employed. However, UPS and DFPQ start to distribute excessive number of slots to rtPS for eight MSs owing to its high priority, resulting in the starvation of BE. Contrastively, the HUF is quite fair even when 16 MSs are in-volved. TPP behaves similarly to the HUF, but becomes much unfair when heavily loaded because it tends to grant excessive slots to service classes.
5. Conclusions and future work
This work aims at designing an integrated bandwidth allocation algorithm for a WiMAX BS in order to provide (1) dynamic down-Table 7
Parameters of the rtPS and BE.
QoS parameter Type I Type II
Service class rtPS BE
Minimum reserved rate (bps) 2400 2400
Maximum sustain rate (bps) 1,000,000 1,000,000
Maximum latency (ms) 50 N/A
Polling time (ms) 20 N/A
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% UP S TP P DF P Q HU F UP S TP P DF P Q HU F UP S TP P DF P Q HU F UP S TP P DFP Q HU F UP S TP P DF P Q HU F 4 8 12 16 20 # of MSs Pe rc e n ta g e BE rtPS
b0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 4 8 12 16 20 # of MSs F airness HUF DFPQ TPP UPS
link/uplink adjustment, (2) latency guarantee for real-time appli-cations, and (3) service differentiation and fairness among all ser-vice classes. Moreover, the MCS is incorporated to be adaptive to the link status between MSs and BS to lessen the impact from long distance and interference. The GPSS is preferred not only to comply with the standard but also to enable MSs to ﬂexibly utilize the bandwidth.
The HUF is proposed to achieve the goals. It translates the requested size to number of slots according to the current MCS when a frame starts, and then allocates the bandwidth according to the Urgency of the data/request which considers latency, priority and fairness. A data/request with a deadline equaling to one needs to be served immediately, while others’ urgency is calculated and indicated as the U-factor. In the dy-namic DL/UL sub-frame determination, the HUF ﬁrstly reserves bandwidth for (1) data/requests whose deadline equals to one and (2) the minimum reserved rate of each service ﬂow, and then proportionates the remaining bandwidth for DL/UL according to the remnant non-urgent data/requests. After that the head-of-line data/request of a queue with the largest aver-age-U-factoris allocated, repeatedly, until the sub-frame is ful-ﬁlled. Finally, each MSs obtains grant from its own service queues.
Simulation result indicates that the quality is retained as the MCS adapts owing to the link quality. For dynamic adjustment, we show the throughput is as satisfactory as TPP and is 7% better than the static adjustment, and the violation rate is signiﬁcantly alleviated by 42% and 80% compared to the TPP and static adjust-ment, respectively. The HUF outperforms the DFPQ by 25% in throughput when overloaded, and incurs no latency violation when the load is within system capacity. Finally, the HUF is ob-served to be fairer than the UPS, DFPQ and TPP and, unlike the TPP, it avoids inappropriate grant for rtPS.
Though HUF is relatively tolerant to overloaded situations, as a future direction we plan to develop admission control schemes to ease the degradation in throughput and fairness. Besides, while la-tency guarantee and fairness are concerned in BSs, a bandwidth allocation algorithm for MSs is also demanded to schedule appro-priately the granted bandwidth.
 IEEE 802.16 Working Group, Air Interface for Fixed Broadband Wireless Access Systems, June 2004.
 Cable Television Laboratories Inc., Data-Over-Cable Service Interface Speciﬁcations – Radio Frequency Interface Speciﬁcation v1.1, July 1999.  IEEE 802.16 Working Group, Air Interface for Fixed and Mobile Broadband
Wireless Access Systems – Amendment for Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands, February 2006.
 IEEE 802.11 Working Group, Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Speciﬁcations, September 1999.
 W.M. Yin, C.J. Wu, Y.D. Lin, Two-phase Minislot Scheduling Algorithm for HFC QoS Services Provisioning, GLOBECOM, November 2001.
 M. Andrews et al., Providing quality of services over a shared wireless link, IEEE Communication Magazine (February) (2001) 150–154.
 K. Wongthavarawat, A. Ganz, IEEE 802.16 Based Last Mile Broadband Wireless Military Networks with Quality of Service Support, MILCOM, October 2003.
 J. Chen, W. Jiao, H. Wang, A Service Flow Management Strategy for IEEE 802.16 Broadband Wireless Access Systems in TDD Mode, ICC, May 2005.
 Y.N. Lin, S.H. Chien, Y.D. Lin, Y.C. Lai, M. Liu, in: Maode Ma (Ed.), Dynamic bandwidth allocation for 802.16e-2005 MAC, Current Technology Developments of WiMax Systems, Springer, 2007.
 A. Sayenko, O. Alanen, J. Karhula, T. Hamalainen, Ensuring the QoS requirements in 802.16 scheduling, in: ACM MSWiM’06, October 2006.  Mobile WiMAX Part I, A technical overview and performance evaluation, in:
WiMAX Forum, April 2006.
 D.P. Heyman, A. Tabatabai, T.V. Lakshman, Statical analysis and simulation study of video teleconference trafﬁc in ATM networks, IEEE Transactions on Circuits and Systems for Video Technology 2 (1) (1992) 49–59.
 C. Lu, J.A. Stankovic, G. Tao, S.H. Son, Design and evaluation of a feedback control EDF scheduling algorithm, in: Real-Time Systems Symposium, 1999.