- Research Article
- Open Access
Cross-Layer Resource Scheduling for Video Traffic in the Downlink of OFDMA-Based Wireless 4G Networks
© Feroz A. Bokhari et al. 2009
- Received: 27 June 2008
- Accepted: 30 December 2008
- Published: 18 January 2009
Designing scheduling algorithms at the medium access control (MAC) layer relies on a variety of parameters including quality of service (QoS) requirements, resource allocation mechanisms, and link qualities from the corresponding layers. In this paper, we present an efficient cross-layer scheduling scheme, namely, Adaptive Token Bank Fair Queuing (ATBFQ) algorithm, which is designed for packet scheduling and resource allocation in the downlink of OFDMA-based wireless 4G networks. This algorithm focuses on the mechanisms of efficiency and fairness in multiuser frequency-selective fading environments. We propose an adaptive method for ATBFQ parameter selection which integrates packet scheduling with resource mapping. The performance of the proposed scheme is compared to that of the round-robin (RR) and the score-based (SB) schedulers. It is observed from simulation results that the proposed scheme with adaptive parameter selection provides enhanced performance in terms of queuing delay, packet dropping rate, and cell-edge user performance, while the total sector throughput remains comparable. We further analyze and compare achieved fairness of the schemes in terms of different fairness indices available in literature.
- Token Bank
- Packet Schedule
- Fairness Index
- Packet Arrival Rate
- Multiuser Diversity
The approaching fourth-generation (4G) wireless communication systems, such as the Third-Generation Partnership Project's Long Term Evolution (3GPP LTE)  and the IEEE 802.16 standards family (e.g., ), are projected to provide a wide variety of new multimedia services, ranging from high quality voice to other high-data-rate wireless applications. Another notable 4G wireless effort is the WINNER project, which aims to develop an innovative concept in radio access in order to achieve high flexibility and scalability with respect to data rates and radio environments . Concepts developed in the WINNER project are applicable to evolving 4G standards due to common system considerations such as orthogonal frequency-division multiple access- (OFDMA-) based air interface, and support of relays and multiple-antenna configurations.
Unlike wireline networks, wireless resources are scarce. The data-rate capacity that a radio-frequency channel can support is limited by Shannon's capacity law. Moreover, due to the time-varying nature of wireless channel, radio resource management, especially packet scheduling and resource allocation, is crucial for wireless networks. Traditionally, the research on packet scheduling has emphasized QoS and fairness issues, and opportunistic scheduling algorithms have focused on exploiting the time-varying nature of the wireless channels in order to maximize throughput. This segregation between packet scheduling and radio resource allocation is inefficient. As fairness and throughput are reciprocally related, an intelligent compromise is necessary to obtain the required QoS while exploiting the time-varying characteristics of the wireless channel. Therefore, it is important to merge the packet scheduling and the resource allocation to design a cross-layer scheduling scheme .
A number of scheduling schemes in the literature analyze physical- (PHY-) and MAC-related design issues by assuming that all users are backlogged, that is, all users in the system have nonempty buffers. However, it is shown in  that this assumption is not always correct, since the number of packets in the buffers can vary significantly, and there is a relatively high probability that the buffers are empty. For example, in time-slotted networks, the packets in the queues are aggregated into time slots. Consequently, empty queues and partially filled time slots will affect the system performance. Furthermore, these non-queue-aware scheduling algorithms lack the capability to provide required fairness among user terminals (UTs). Hence, it becomes necessary to consider queue states in scheduling and resource allocation .
In recent years, some schemes have considered integrating packet scheduling and radio resource scheduling into queue and channel aware scheduling algorithms. In , a weighted fair queuing (WFQ) scheduling scheme is proposed, where the largest share of the radio resources is given to the users with the best instantaneous channel conditions in a code division multiplexing (CDM-) based network. Another example of a queue- and channel-aware scheduling algorithm is the modified-largest weighted delay first (M-LWDF) algorithm, where priorities are given to the users with maximum queuing delays weighted by their instantaneous and average rates . The associated decision metrics in these schemes are based on the combination of the delay and instantaneous channel rates. Finding an optimal metric based on these parameters is difficult due to varying requirements for different service classes.
In this paper, we present a scheduler which comprises packet scheduling and resource mapping taking both queue and channel states into account. In the first level of scheduling (packet scheduling), users to be served are selected based on the token bank fair queuing (TBFQ) algorithm, considering fairness and delay constraints among users. Although TBFQ was originally proposed for single-carrier time-division multiple access (TDMA) systems , it has been modified in this study by introducing additional parameters that adaptively interact with the second level of scheduling (resource mapping). These parameters take into account the network loading and the user channel conditions. Based on these parameters, the second-level scheduler assigns resources to the selected users in an adaptive manner that exploits the frequency selectivity of the OFDMA air interface. The modified combined scheduling scheme is called ATBFQ.
The performance of ATBFQ is studied in the context of the WINNER wide-area downlink scenario and is compared to that of the SB scheduling algorithm (which was the baseline scheduling scheme in WINNER)  and the RR scheme by extensive simulations. The rest of this paper is organized as follows. In Section 2, the ATBFQ algorithm is described in detail, along with its parameter selection. Methods of fairness assessment are addressed in Section 3. The system model and the simulation parameters are presented in Section 4. Simulation results are provided in Section 5, followed by conclusions in Section 6.
2.1. Original TBFQ Algorithm
The TBFQ algorithm was initially developed for wireless packet scheduling in the downlink of TDMA systems [9, 11], and was later modified for wireless multimedia services using uplink as well. Its concept was based on the leaky-bucket mechanism which polices flows and conforms them to a certain traffic profile.
A traffic flow belonging to user i is characterized by the following parameters:
: packet arrival rate,
: token generation rate,
: token pool size,
: counter that keeps track of the number of tokens borrowed from or given to the token bank by flow i.
By prioritizing in this manner, we ensure that flows belonging to UTs that are suffering from severe interference, and shadowing conditions in particular, will have a higher priority index, since they will contribute to the bank more often.
2.2. ATBFQ Algorithm
In this study, the TBFQ algorithm, originally proposed for single carrier TDMA systems, is improved by introducing adaptive parameter selection and extended to suit the WINNER multicarrier OFDMA systems . The motivation behind this modification was to incorporate the design and performance requirements of the scheduler in 4G networks into the original scheme. In such networks, the utilization of the resources and hence the performance of the network can be enhanced by making use of the multiuser diversity provided by the multiple access scheme being used. Also, such networks support users with high mobility. Therefore, in order to make use of the channel feedback, faster scheduling (at a much smaller time scale) is required. Another requirement is the ability to maintain fairness and provide a minimum acceptable QoS performance to all users.
Like TBFQ, the ATBFQ scheduling principle is based on the leaky-bucket mechanism. Each traffic flow i is characterized by a packet arrival rate , token generation rate , token pool size , and a counter to keep track of the number of tokens borrowed from or given to the token bank. Each L-byte packet consumes L tokens. As tokens are generated at rate , the tokens overflowing from the token pool are added to the token bank, and is incremented by the same amount. When the token pool is depleted and there are still packets to be served, tokens are withdrawn from the bank by flow i, and is decremented by the same amount. A debt limit is set as a threshold to limit the amount a UT can borrow from the bank. It also acts as a measure to prevent malicious UTs (transmitting at unusually high transmission rates) from borrowing extensively. The packets are then queued in subqueues in a per-flow queuing (PFQ) manner such that each subqueue belongs to a particular flow, as shown in Figure 1.
At the scheduler, information is retrieved from the higher layer about all active users using the getActiveUsers() function. An active user is defined as a backlogged queue which has packets waiting to be served.
Using the borrowbudget() function, a certain budget is calculated for the priority user which depends on the token counter , and the debt limit , and is given by . keeps track of how much the user has borrowed or given to the bank. The debt limit keeps track of how much a user can further borrow from the bank in order to accommodate the burstiness of the traffic over the long term.
where is the SINR of the selected user i in chunk j. This is the most opportunistic of all scheduling algorithms for time-slotted networks. This means that the adaptive modulation and coding (AMC) policy maximally exploits the frequency diversity of the time-frequency resource, where a chunk is allocated to only one user and a user can have multiple chunks in a scheduling instant.
The resourceMap() function determines the amount of bits that can be mapped to the chunk depending on the AMC mode used.
Each time a chunk resource is allocated, the updateCounter() function is called. This function updates the bank, the counter , and the allocated budget.
The selected user i gets to transmit as long as (1) its queue remains backlogged and (2) the allocated budget is less than the total bank size and more than the number of bits that can be supported with the lowest AMC mode (binary phase-shift keying (BPSK) rate-1/2, considered in this study). If either of these conditions is not satisfied, the user is classified as nonactive. A new priority is calculated on the updated active users, and Steps 1–6 are repeated. This procedure is repeated until there are no chunk resources available or there are no active users.
2.3. ATBFQ Parameter Selection
The performance of the ATBFQ scheduler depends on its parameters that define the debt limit, the burst credit ( ), and the token generation rate. The token generation rate is critical to the extent to which the burstiness of the UT traffic can be accommodated. A UT in its burst mode transmits more data in a short interval of time than its actual statistics, and hence, requires more resources in order to maintain a certain QoS level. The debt limit is set to −5 MB. The token generation rate should be large enough to handle instantaneous bursty traffic. In simulations, this generation rate has been considered three times larger than the average packet arrival rate.
Burst credit for ATBFQ for low loading (8 users).
(Byte per frame)
BC = 1000
BC = 5000
BC = 10000
Burst credit for ATBFQ for high loading (20 users).
(Byte per frame)
BC = 1000
BC = 5000
BC = 10000
where is the transmitted throughput in bits for UT i during the time interval and is the total number of packets arriving in the queue for UT i during . In simulations, is chosen to be equal to 16 frame time durations.
In (6), the throughput is normalized to avoid ambiguity since the throughput alone as a metric does not provide an insight into the overall fairness.
where is the instantaneous throughput of UT i in a particular frame, and n is the total number of UTs. As stated in , the CDF of this normalized throughput should lie to the right of the coordinates (0.1, 0.1), (0.2, 0.2), and (0.5, 0.5).
The results using both of these fairness assessment methods are discussed in detail in Section 5.
ATBFQ is studied in the wide-area downlink scenario. To reduce the simulation complexity, the bandwidth is reduced to 15 MHz from the original 45 MHz. The chunk dimension is given as 8 subcarriers by 12 OFDM symbols or 312.5 kHz 345.6 microseconds. The frame duration is defined as 691.2 microseconds, that is, there are a total of 96 chunks per frame.
where PL is the path loss in dB, and d is the transmitter-receiver separation in meters.
The average thermal noise power is calculated with a noise figure of 7 dB. We have considered independent lognormal random variables with a standard deviation of 8 dB for shadowing. Sector transmit power is assumed to be 46 dBm, and chunks are assigned fixed equal powers.
The interference is modeled by considering the effect of intercell interference and intracell interference on the sector of interest in the central cell (denoted as sector 1 in BS 1). For this purpose, the interference from the first tier is taken into account. In this case, for a link of interest in sector 1 in BS 1, the interference will comprise 18 (6 BS × 3 sectors) intercell and 2 intracell links.
where x is a uniform random variable defined over , and AF (activity factor) is defined as a probability for a particular interfering link to be active. For example, AF of 1 denotes a high level of interference where all the links are active interferers (100% interference).
Lookup table for AMC modes and corresponding chunk throughput.
Chunk throughput (bits)
0.2311 ≥ SINR > −1.7
1.231 ≥ SINR > 0.231
3.245 ≥ SINR > 1.231
4.242 ≥ SINR > 3.245
6.686 ≥ SINR > 4.242
9.079 ≥ SINR > 6.686
10.33 ≥ SINR > 9.079
14.08 ≥ SINR > 10.33
15.6 ≥ SINR > 14.08
SINR > 15.6
where are i.i.d. random variables on with .
Summary of simulation parameters.
Wide area DL (frequency adaptive)
WINNER C2 channel
Independent lognormal random variables (standard deviation 8 dB)
Sector Tx antenna
directional with WINNER baseline antenna pattern
UT receive antenna
15 MHz (i.e., 48 chunks which is 1/3rd of the baseline assumptions)
Sector Tx power
Adaptive Token Bank Fair Queuing, score based, and round-robin
brute force method (central cell is considered with interference from the 1st-tier)
BPSK (rate 1/2 and 2/3), QPSK (rate 1/2, 2/3, and 3/4), 16QAM (rate 1/2, 2/3, and 3/4), and 64QAM (rate 2/3 and 3/4)
With FEC block of 1728 bits and 10% BLER
0.6912 ms (scheduling interval)
1.9 Mbps 2IRP model for MPEG video
Packet drop criterion
Delay ≥ 0.19 sec
MATLAB and OPNET
The performance results are classified into four categories: (1) average user statistics, (2) performance of the cell-edge users, (3) effect of varying user loading and interference conditions, and (4) fairness analysis. Furthermore, the results are compared to the SB and RR algorithms. The window size plays an important role in the performance of the SB algorithm . The performance of ATBFQ has been studied with different window sizes in the WINNER context [22, 23].
5.1. User Performance
5.2. Cell-Edge User Performance
5.3. Varying User Loading and Interference Conditions
Performance indicators such as average dropped packets, average UT throughput, and average UT queuing delay have been considered in evaluating ATBFQ by comparison with the reference SB and RR schemes.
For low-to-medium loading with an AF of 0.7, it is observed again that ATBFQ outperforms the reference schemes in terms of all observed parameters. This trend changes as network loading increases to 20 UTs per sector. In this case, SB outperforms ATBFQ and RR in terms of average UT queuing delay, average packets dropped per frame, and the total sector throughput, respectively. This is due to the fact that SB is opportunistic in nature, whereas ATBFQ is fairness aware. As the number of UTs increases, SB takes advantage of the multiuser diversity to achieve higher throughput.
5.4. Fairness Analysis
In this paper, the performance of the ATBFQ scheduling algorithm with adaptive parameter selection is investigated in the context of the 4G WINNER wide-area downlink scenario. It is a queue- and channel-aware scheduling algorithm which attempts to maintain fairness among all users. Performance of ATBFQ is presented with reference to the SB and RR schedulers. Being an opportunistic scheduler belonging to the proportional fair class, SB aims to maximize throughput by making use of multiuser diversity while trying to maintain fairness. However, this comes at a certain cost, since the cell edge users in this scheme, suffering from poor channel conditions, are more severely affected. Also, due to the bursty nature of the traffic, such users experience higher queueing delays, resulting in a higher number of packet dropping.
Contrary to SB, ATBFQ is a credit-based scheme which aims to accommodate the burstiness of the users by assigning them more resources in the short term, provided that long-term fairness is maintained. For lower to medium loading, ATBFQ provides higher throughput, lower queuing delay, and a lower number of packets dropped as compared to SB and RR. At high loading, ATBFQ still outperforms SB and RR with regard to the queuing delay and packet dropping, however, with a slight degradation in the sector throughput. This is because ATBFQ attempts to satisfy users with poor channel conditions by assigning more resources, even with a lower chunk spectral efficiency. An overall improvement of the performance of cell-edge users is observed in terms of spectral efficiency and packet-dropping ratio for ATBFQ as compared to SB and RR.
The observed throughput, queuing delay, and packet dropping rate clearly indicate the superiority of the ATBFQ algorithm. This apparent improvement in the fairness performance of the ATBFQ algorithm based on these performance parameters is further validated by evaluating the fairness indices available in the literature.
The authors would like to express their gratitude to Mr. Jiangxin Hu for his technical support and Dr. Abdulkareem Adinoyi for providing his valuable comments on the manuscript. They also thank OPNET Technologies, Inc. for providing software license to carry out the simulations of this research. This work was a part of the Wireless World Initiative New Radio (WINNER) project, http://www.ist-winner.org/, with the support of the Natural Sciences and Engineering Research Council (NSERC) of Canada. Preliminary results of this work have been presented in IEEE VTC2008-Spring and IEEE VTC2008-Fall conferences.
- Overall Description: Stage 2 (Release 8) : 3GPP Std. 3GPP E-UTRA and E-UTRAN Technical Specification TS 36.300 V8.4.0. March 2008, http://www.3gpp.org/ftp/Specs/html-info/36300.htm
- IEEE 802.16 Std. 802.16j/D5 : Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems—Multihop Relay Specification. June 2008, http://www.ieee802.org/16
- Project Presentation WINNER Deliverable D8.1, March 2004, http://www.ist-winner.org/deliverables_older.html
- Liu Q, Wang X, Giannakis GB: A cross-layer scheduling algorithm with QoS support in wireless networks. IEEE Transactions on Vehicular Technology 2006, 55(3):839-847. 10.1109/TVT.2006.873832View ArticleGoogle Scholar
- Borst S: User-level performance of channel-aware scheduling algorithms in wireless data networks. Proceedings of the 22nd Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM '03), March-April 2003, San Francisco, Calif, USA 1: 321-331.Google Scholar
- Wu D, Negi R: Effective capacity: a wireless link model for support of quality of service. IEEE Transactions on Wireless Communications 2003, 2(4):630-643.Google Scholar
- Stamoulis A, Sidiropoulos ND, Giannakis GB: Time-varying fair queueing scheduling for multicode CDMA based on dynamic programming. IEEE Transactions on Wireless Communications 2004, 3(2):512-523. 10.1109/TWC.2003.821151View ArticleGoogle Scholar
- Andrews M, Kumaran K, Ramanan K, Stolyar A, Whiting P, Vijayakumar R: Providing quality of service over a shared wireless link. IEEE Communications Magazine 2001, 39(2):150-154. 10.1109/35.900644View ArticleGoogle Scholar
- Wong WK, Leung VCM: Scheduling for integrated services in next generation packet broadcast networks. Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC '99), September 1999, New Orleans, La, USA 3: 1278-1282.Google Scholar
- Bonald T: A score-based opportunistic scheduler for fading radio channels. Proceedings of the 5th European Wireless Conference (EW '04), February 2004, Barcelona, SpainGoogle Scholar
- Wong WK, Tang HY, Leung VCM: Token bank fair queuing: a new scheduling algorithm for wireless multimedia services. International Journal of Communication Systems 2004, 17(6):591-614. 10.1002/dac.670View ArticleGoogle Scholar
- Final Report on Identified RI Key Technologies, System Concept, and their Assessment WINNER I Deliverable D2.10, November 2005, http://www.ist-winner.org/deliverables_older.html
- Test Scenarios and Calibration Cases Issue 2 WINNER II Deliverable D6.13.7, December 2006, http://www.ist-winner.org/deliverables.html
- Knopp R, Humblet PA: Information capacity and power control in single-cell multiuser communications. Proceedings of IEEE International Conference on Communications (ICC '95), June 1995, Seattle, Wash, USA 1: 331-335.View ArticleGoogle Scholar
- Jain R, Chiu D, Hawe W: A quantitative measure of fairness and discrimination for resource allocation in shared computer systems. Digital Equipment Corporation, Maynard, Mass, USA; September 1984.Google Scholar
- Sirisena H, Haider A, Hassan M, Fawlikowski K: Transient fairness of optimized end-to-end window control. Proceedings of IEEE Global Telecommunications Conference (GLOBECOM '03), December 2003, San Francisco, Calif, USA 7: 3979-3983.View ArticleGoogle Scholar
- Berger-Sabbate G, Duda A, Gaudoin O, Heusse M, Rousseau F: Fairness and its impact on delay in 802.11 networks. Proceedings of IEEE Global Telecommunications Conference (GLOBECOM '04), November-December 2004, Dallas, Tex, USA 5: 2967-2973.View ArticleGoogle Scholar
- IEEE 802.16 Std : IEEE 802.16m Evaluation Methodology Document. September 2007, http://www.ieee802.org/16
- Final report on link level and system level channel models WINNER I Deliverable D5.4, November 2005, http://www.ist-winner.org/deliverables_older.html
- Traffic model for 802.16 TG3 MAC/PHY simulations IEEE 802.16 Work-in-progress document 802.16.3c-01/30r1, March 2001, http://www.ieee802.org/16
- Chaponniere EF, Black PJ, Holtzman JM, Tse DNC: Transmitter directed code division multiple access system using path diversity to equitably maximize throughput. US Patent no. 6449490, September 2002Google Scholar
- Inteference Avoidance Concepts WINNER II Deliverable D4.7.2, June 2007, http://www.ist-winner.org/deliverables.html
- Bokhari FA, Wong WK, Yanikomeroglu H: Adaptive token bank fair queuing scheduling in the downlink of 4G wireless multicarrier networks. Proceedings of the 67th IEEE Vehicular Technology Conference (VTC '08), May 2008, Marina Bay, Singapore 1995-2000.Google Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.