Congestion and Error Control for Layered Scalable Video Multicast over WiMAX

Chih-Wei Huang and Jenq-Neng Hwang Department of Engineering

University of Washington, Box 352500 Seattle, WA 98195, USA

{cwh, hwang@ee.washington.edu}

David Chih-Wei Chang SoC Technology Center

Industrial Technology Research Institute Chutung, Hsinchu, Taiwan 310, R.O.C.

{cwchang@itri.org.tw}

Abstract-Quality of service (QoS) of wireless multimedia can be seriously degraded due to the more dynamically changing end-to-end available bandwidth caused by the wireless fading/shadowing and link adaptation. Moreover, the increased occurrence of wireless radio transmission errors also results in higher bursty rate of packet loss when compared with wired IP networks. An end-system driven solution featuring embedded probing for layered multicast of scalable video is thus proposed in this paper. By taking advantage of QoS features offered by one of the four proposed WiMAX service flow arrangement, this system aims at more flexible layer constructing and subscription while reliable in diverse channel conditions and fitting users’ demand. The system optimality comes from the best tradeoff of number of video layers subscription with number of additional FEC packets insertion to simultaneously satisfy the estimated available bandwidth and the estimated wireless channel error condition.

I. INTRODUCTION

With the gradual paradigm shift from analog to digital media, from push-based media broadcasting to pull-based media streaming, and from wired interconnectivity to wireless interconnectivity, wireless broadband access with provisioned quality of service (QoS) for digital multimedia applications to mobile end users over a wide area is the new frontier of telecommunications industry. Thanks to the perfect synergy between WLAN and the mobile WiMAX, and cost-effective chipset designs based on volume, a new scalable wireless distribution system architecture without large investment is envisaged. This all IP based system architecture requires a few WiMAX base-stations for broadband access coverage as well as acting as the back-haul for the WLAN networks in its coverage footprint. Quality of Service (QoS) plays the most critical role toward success of many multimedia networking applications over IP networks. There has been a substantial amount of research regarding end-to-end QoS in both network-centric and end-system centric perspective, such as scalable multimedia coding, adaptive protection, traffic classification, and network adaptation [1]. Many existing rate-based protocols for multimedia (mainly UDP) traffic try to be TCP-friendly. Some protocols, such as TFRC [2] and SMCC [3], use the explicit TCP throughput equation for rate adjustment. For better efficiency in wireless networks, TFRC Wireless [4], MULTFRC [5], and VTP [6] have also been proposed as extended solutions. To be TCP-friendly, a lot of efforts (especially for the multicast case) are required to

acquire the round trip time (RTT) information. However, TCP is known to be RTT-biased, i.e., a TCP session with a larger RTT gets lower throughput when competing a bottleneck link with another TCP session with a smaller RTT. It is our belief that the rate should be the dominant factor to determine the bandwidth share for rate-based protocols.

Furthermore, in order to ensure effective dissemination of compressed multimedia data over IP based wireless broadband networks, the main challenges came from the integrated wired and wireless heterogeneous networking systems, where the quality of service (QoS) is further degraded due to the more dynamically changing end-to-end available bandwidth caused by the wireless fading/shadowing and link adaptation. Moreover, the increased occurrence of wireless radio transmission errors also results in higher bursty rate of packet loss when compared with wired IP networks. To overcome all these extra deficiencies caused by the wireless networks, several additional QoS mechanisms (spanning from physical, MAC, network and application layers) have to be incorporated.

An end-system driven application-layer QoS solution featuring embedded probing for layered multicast of scalable video is proposed in this paper. By taking advantage of the QoS features offered by one of the four proposed WiMAX service flow arrangement, this solution aims at more flexible layer constructing and subscription while reliable in diverse channel conditions and fitting users’ demand. Through effective integration of packet loss classification (PLC) [7], end-to-end available bandwidth probing [8,9], congestion control via layered structure and packet level FEC [10], for layered multicast applications over WiMAX for disseminating scalable extension of H.264/AVC compressed video [11] is proposed. The optimality comes from the best tradeoff of number of video layers subscription with number of additional FEC packets insertion to simultaneously satisfy the estimated available bandwidth and wireless channel error condition.

The paper is organized as follows: Sections 2 addresses several critical techniques for effective wireless multimedia dissemination. We then discuss the QoS issues related to layered multicast of scalable video over WiMAX in Section 3. In Section 4, an example of application layer QoS design for wireless multimedia system is illustrated, followed by a conclusion in Section 5.

II. CRITICAL TECHNIQUES FOR MULTIMEDIA DISSEMINATION

1141-4244-0957-8/07/$25.00 ©2007 IEEE.

A. End-to-End Available Bandwidth Estimation There has been active research in available bandwidth

estimation recently, such as Delphi [12], TOPP [13], Pathload [14], IGI [15], Pathchirp [16], and Spruce [17]. However, these tools have special requirements for their probes, such as fixed packet size and exact inter-packet spacing, which may not be available for most practical multimedia applications. Moreover, existing tools generate hundreds kilobytes of probe traffic (or more) and take a long time to give the available bandwidth estimation. We have noticed that it is not required to estimate the absolute available bandwidth for congestion control purposes. It is good enough to know whether the path has enough bandwidth of a specific rate (e.g., the cumulative rate of the next layer) for congestion control purposes.

We have proposed a delay trend model [8,9], which uses tens of variable-size packets to check whether the path has the available bandwidth of a specific rate. The model relies on correctly detecting whether there are trends among the measured one-way delays (OWDs) and packet sizes. A reliable and consistent trend detection algorithm is the core function in the model. Our proposed trend detection algorithm, called Fullsearch, uses statistical tests to check the existence of trend. Define ( )I X is 1 if X holds, and 0 otherwise. If the measured OWDs are

hD , 1, ,h M= … . The test used is:

1

2 1

( )

( 1)2

M h

h uh u

Fullsearch

I D DS M M

−

= =

>= −

∑∑ (1)

where OWD is defined as the relative one-way trip time measured by the receiver as the time difference between the receiving time and the packet sending timestamp recorded in the field of multimedia UDP packet header plus a fixed bias.

As Eq. (1) shows, the Fullsearch algorithm reflects the statistical relationship among all pairs of measurements. If there is no trend among the measurements, the test result is around 0.5; if there is a strong increasing trend, the test result approaches 1. A trend threshold th can be used to check whether there is an increasing trend. If the test result is larger than th , the measurements have an increasing trend, otherwise no increasing trend.

B. Packet Loss Classification In a wireless network environment, common channel errors

due to multi-path fading, shadowing, and attenuation may cause bit errors and packet losses, which are quite different from the packet loss caused by network congestion. In congestion control, the packet loss information can serve as an index of network congestion for effective rate adjustment; therefore wireless packet loss can mistakenly lead to dramatic performance degradation. We have proposed a PLC algorithm [7] based on trend detection of OWD when it falls in the ambiguous zone where the PLC is not straightforward. The

algorithm can greatly benefits rate-based congestion control algorithms for multimedia over IP networks.

Generally, the classification algorithms of packet loss depend on the analysis of statistical behavior of some observed values: packet timestamp and packet serial number in the packet header. Spike-train [18] and ZigZag [4] investigate the OWD difference between two classes of packet loss. Unfortunately, these methods may produce unreliable classification performance when OWD is around the threshold, which depends on the network topology.

In our proposed PLC method, we also exploit the OWD of received packets to assist packet loss classification. Since network congestion is directly related to the congestion packet loss, we adopt the following based on similar reasoning as in Pathload. More specifically, when a packet loss is observed at time t, it should be considered as a congestion loss if the trend is in an ascending phase; otherwise, it is categorized as wireless loss.

C. Layered Coding and FEC Structure Our scalable video is created by the motion compensated

temporal filtering (MCTF) scalable extension of H.264/AVC [11], which is an emerging compression technology with coding efficiency comparable to original H.264/AVC standard. With MCTF using lifting framework, temporal decomposition can be achieved nicely for SNR, temporal and spatial scalabilities. Aggregation and zero-padding of network abstraction units may need to compose packets in identical size when encoding FEC based on block erasure codes.

Video data and error protection codes are formatted in layers, where each layer is assigned to a multicast group as Fig. 1 with rate ,i jr , i and j are indexes of video and FEC layers respectively. Layers with 0j = contain only video streams, otherwise j indicates level of protection for a specific i . There are V enhancement video layers and F protection.

To encode layered FEC, we decide feasible block parameters ,( , )i j in k for video layer i through network analysis. Gilbert/Elliot’s model is used to simulate the bit error rate resulting in the packet loss. This method can be approximated as a two-state Markov chain with parameters p and q representing transition probabilities from (packet) loss state to received state and from received state to loss state respectively. During the process, target decoding error rates,

FECe as well as ik are set first then apply the estimated ( , )p q value sets representing wireless channel conditions into to evaluate minimum ,i jn satisfying ( , ) ,FEC n k FEC ie e< .

In other words, by modeling the network conditions using

Fig. 1: Rates of ( , )i j in layers.

115

F number of ( ),p q sets, we can derive the corresponding ,i jn , 1 ~j F= in ascending order to generate F protection

layers. In case we want to judge how many FEC layers to subscribe after the whole scenario has been constructed, we can again choose a data layer i , plug in newly estimated ( ),p q , find minimum newn through the same process, and then pick j with ,i jn closest to but larger than newn .

Regarding inter-layer dependency and protection levels for every layer, the overall data rate is , ,

0 0

fv

v f i ji j

R r= =

=∑∑ , where v is the number of enhancement layers subscribed and f , same for every i , is the FEC layers requested. Unequal protection to different video layers is supported by different ,FEC ie , not by f . For instance, under channel condition ( ) ( ), 0.8, 0.2p q = , we preset ,0 0.001FECe = , ,5 0.01FECe = , and 0 5 8k k= = resulting in 0, 17jn = and 5, 15jn = . It also implies that a video layer requires all lower ones to be available for decoding (due to the cumulative layer structure of the adopted scalable codec), while not every sub-layer FEC is needed. Therefore, receivers acquire distinct video quality and amount of protection by subscribing to proper groups.

D. Embedded Layered Probing and Join Decision In order to increase the data rate, either for more video data

or loss recovery, available bandwidth estimation has to be performed in advance to prevent congestion. In contrast to extra probing packets, we embed probing streams in regular ones through effective scheduling of packet transmission and take advantage of the fact that streaming systems usually have decoding buffer to tolerate some amount of delays [10].

As shown in Fig. 3 about our embedded probing, the stream is periodically separated into probing and regular intervals alternatively with period T . The length of probing interval, pt , is further divided into certain uniform probing regions according to number of possible layers to go, e.g., 1pr to 4pr . In each region, previously generated packets are delayed in transmission creating temporarily higher sending rate within it.

At the receiver side, the interval and region a packet belongs to can be distinguished from its RTP timestamp; then Fullsearch delay trend detection is applied to packets in time slots at objective rates. The duration of each probing region is set to be the time to send 50 packets in regular intervals of base layer ( , ) (0, 0)i j = and remains the same for every layer.

Other important facts are that the probing rate we are looking for is an aggregation rate ,v fR not ,i jr . The

aggregation rates of all possible target subscriptions should be included in a probing interval. Given that the layered streams are ready at the server, what layers to subscribe is based on three types of information: channel estimation ( ),p q , probing results (available bandwidth), and observed packet loss rates. Relying on PLC, we are able to continuously monitor ( )ˆ ˆ,p q , (update every T seconds for enough samples) and congestion packet loss (update every second). Table 1 addresses all possible cases and reactions in our proposed architecture where five moves are allowed in the system. “--“ means do not care because the rate is getting lower or low congestion loss is a must for a positive probing. Every period T , receivers come out a ( ),p q pair, and map it to demand FEC, j , for current video layer i based on the condition specified in Section II-C . If the new j is larger than current j , we categorize it as “worse” in wireless condition. “Better” and “same” are defined analogously. For each condition, we increase FEC, decrease FEC, and change video quality accordingly taking into account of the observations. For example, if the wireless channel is worse and probing for more FEC in the same video layer is failed, we go path II with more FEC but less quality; if channel does not change and probing for higher rate at the same protection level is positive, it is time to subscribe more video contents.

The aggregation rates of probing regions will match target rates. Depending on channel quality, end users select region 1 and 2 or 3 and 4 for delay trend detection and make decisions to stay or take one of five moves accordingly. The only information need to feedback is the resulting change of layer subscription. III. WIMAX QOS PROPERTIES FOR ADVANCED MULTIMEDIA

TRANSMISSION SYSTEMS

WiMAX is based on 802.16d [8] and 802.16e [9] standards published in 2004 and 2006 respectively. 802.16d supports fixed terminals only using OFDM while 802.16e is able to provide services to mobile terminals by OFDMA.

A. WiMAX QoS Architecture With QoS in mind, the design of 802.16 MAC supports

Fig. 2: Layered FEC block erasure coding structure.

Fig. 3: Embedded probing.

TABLE 1 Adaptation Rules

Events Channel Estimation

Probing Result

Congestion Loss

Action Path

positive -- I Increase FEC worse

negative (Yes) II Decrease

FEC better -- -- III

positive -- IV Change Quality same

negative Yes V

116

traffics with a wide range of demand based on scheduling services. More specifically, the data handling mechanisms, which is supported by the MAC scheduler, is determined by a set of QoS parameters that quantify aspects of its behavior.

Outbound transmission scheduling selects the data for transmission in a particular frame/bandwidth allocation and is performed by the base station (BS) for downlink, and subscriber station (SS) for uplink. The following details are taken into account for each active service flow:

• The scheduling service specified for the service flow. • The values assigned to the service flow’s QoS

parameters. • The availability of data for transmission. • The capacity of the granted bandwidth. In 802.16e, five types of data delivery services associated

with certain predefined set of QoS-related service flow parameters are specified for downlink coordination.

Unsolicited grant service (UGS) is to support real-time applications generating fixed-rate data provided by fixed or variable length protocol data units (PDUs). The transmission opportunities are granted by periodic basis and parameters of tolerated jitter, service data unit (SDU) size (in case of fixed length SDU), minimum reserved traffic rate, maximum latency, request/transmission policy, and unsolicited grant interval.

Real-time variable rate (RT-VR) service is to support real-time applications with variable bit rates which require guaranteed data rate and delay. The BS is supposed to allocate the connection sufficient resources for at least min{ , }S R T∗ , where S denote the amount data arriving at the transmitter’s queue during timeinterval T with

_ _ _R minimum reserved traffic rate= . Any SDU should be delivered within the latency requirement. In the case when S R T> ∗ , delivery of each specific SDU is not guaranteed. QoS parameters are maximum latency, minimum reserved traffic rate, maximum sustained traffic rate, traffic priority, request/transmission policy, and unsolicited polling interval.

Non-real-time variable rate (NRT-VR) service supports applications that require a guaranteed data rate but insensitive to delays. Delivery of each specific SDU is not guaranteed if

_ _ _S maximum sustained traffic rate T> ∗ . Parameters are minimum reserved traffic rate, maximum sustained traffic rate, traffic priority, and request/transmission policy.

Best effort (BE) service is for applications with no rate or delay requirements. Parameters are maximum sustained traffic rate, traffic priority, and request/transmission policy.

Extended real-time variable rate (ERT-VR) service is to support applications with variable data-rates, which require guaranteed data rate and delay while sensitive to jitter, for example VoIP with silence suppression and interactive conferencing. Parameters for this type are maximum latency, tolerated jitter, minimum reserved traffic rate, maximum sustained traffic rate, traffic priority, request/transmission policy, and unsolicited grant interval.

For the layered multicast application addressed in this paper, we will take RT-VR and ERT-VR for more discussion.

B. Multicast on WiMAX The BS may establish a downlink multicast service by

creating a connection with all SSs associated with the service. The SSs need not to be aware that the connection is a multicast connection and each multicast SDU is transmitted only once per BS channel, i.e., no ARQ. Since a multicast connection is associated with a service flow, it is associated with the QoS and traffic parameters of that service flow.

C. Advantages to Layered Multicasting The best news for QoS demanding service introduced by

WiMAX is to explicitly specify QoS needs, such as minimum reserved traffic rate, maximum sustained traffic rate, maximum latency, and sometimes tolerated jitter. The scheduler will guarantee minimum rate, try to achieve maximum rate and hold other constraints by vendor-specific tweaks. For uplink transmission, requesting processes are also involved. In contrary to conventional distributed coordination standards such as Ethernet and 802.11, the centralized coordination of WiMAX can result in better efficiency, guaranteed QoS support, and few collisions while necessary mechanisms for robust wireless communication like ARQ and link adaptation are still playing their roles.

In terms of layered streaming (downloading), the minimum rate constraint will be the key to have base layers video or relatively more important packets delivered so as to guarantee the acceptable quality. When it comes to multicast design, there are several possibilities involving multiple scheduling services.

• Guarantee the base video/protection layers only, RT-VR only (see Fig. 4(a)). By setting minimum rate at the base-layer rate and maximum rate at the best quality/protection rate, the application layer probing will force BS to temporarily get extra capacity if available. If the result of end-to-end available bandwidth estimation is positive, application clients (at SS) can then subscribe to more video/FEC layers.

• Adaptive base rates (minimum reserved rates), RT-VR only (see Fig. 4(b)). Similar to the first method, except through channel estimation, it adaptively sets the minimum rate at certain level of protection depending on current channel quality. Competitions with other traffic flows are needed for extra quality only. The adaptive algorithm may need careful design for not setting minimum reserved rate too high so as not grabbing too much bandwidth and thus effect inter-connection fairness.

• Use both RT-VR and NRT-VR (see Fig. 4(c)). Transmit base-layers in RT-VR but enhancement layers through the NRT-VR providing more flexibility to drop late packets and tolerate more varying receiving variable quality of clients in a WiMAX network.

117

• Use ERT-VR only (see Fig. 4(d)) if the content is jitter sensitive and not allowing much buffering at receivers.

Implementation of proposed schemes can be done by setting proper header columns and multiple MAC layer queues. Base-layer and enhancement-layer have different values of priority in the type of service column of IP header or values compatible with DiffServ framework. Besides, service flow ID (SFID) and connection ID (CID) tags are also specified for streams to be assigned to desired service type in WiMAX MAC. There are also corresponding priority queues for each service types. For example, base and enhancement layers in Fig.4(a) method have different IP priorities, identical SFID and CID. Therefore they end up with different queues in the same service type (RT-VR). Due to higher IP priority, base-layer packets will be sent out first with guaranteed minimum rate and the amount of transmitted enhancement packets depends more on contention between connections.

When break it down to components, ene-to-end available bandwidth estimation and PLC could both benefit from QoS features offered by WiMAX, e.g., knowing the minimum and maximum rate, extreme bandwidth estimation errors can be avoided.

In PLC, less collision loss could make it more reliable in differentiating wired congestion loss from wireless transmission loss since the wireless loss is more reflected in the signal quality.

In brief, we believe the advanced QoS design in WiMAX can bring in much reliable wireless multimedia transmission performance and more flexibility for service providers to customize both software and hardware in the path.

IV. AN EXAMPLE SYSTEM

An end-system driven solution featuring embedded probing for layered multicast of scalable video [10] can thus be

adopted for WiMAX systems. Fig. 5 illustrates our proposed WiMAX scalable video layered multicast system, where video quality degradation resulting from wireless packet loss is protected by adequate FEC erasure codes with embedded probing performed first to assure enough available bandwidth for redundancy and fair share with other sessions. On the other hand, video quality degradation resulting from wired congestion loss can be adequately mitigated through less video/FEC layers subscriptions. The fundamental spirit of this proposed system is to decouple several important modules (scalable video layer creation, packet loss classification, bandwidth probing, and adaptive FEC insertion) and conduct an effective integrated tradeoff analysis to reach optimal number of video layers and FEC protection levels under all the resource constraints.

In simulation, we use NS2 to generate a topology like Fig. 6 with cross traffic from node 6 to node 7. We also assume RT-VR w/o ARQ is the only type of traffic so the performance of the scheduler is not an issue if the total wireless traffic is much lower than WiMAX capacity. We set 3 scalable video layers in H.264/SVC at cumulative rates 300K (QCIF 15fps), 500K (CIF 15fps), and 750K (CIF 30fps). Regarding to FEC, Table 2 illustrates all details from ( , )p q sets to resulting ,i jn , where

2V = and 6F = . All layers are not necessary to be the same rate. In addition, two more thresholds for observed congestion, Pc , and wireless, Pw , loss rates are both 1%. The system will immediately go either lower layer or get more protection considering the last probing outcome, if observing loss rates are higher than thresholds. For probing, T was 10 second and 0.8 second for each region.

The highlight of receiving video quality in terms of PSNR is

Figure 5: System Diagram.

(a)

(b)

(c)

(d)

Figure 4: Layered multicast packets arrangement in service flows.

Figure 6: Simulation network topology.

118

shown in Fig. 7. The PSNR is computed by comparing the reconstructed stream and the original stream with the best quality. We repeat decoded frames for computing PSNR if frame rates do not match, so the value shown may be lower than comparison with matched frame rates in common codec performance charts. The available bandwidth controlled by cross traffic begins from 1Mbps, down to 600Kbps then go back up to 750Kbps. Wireless condition in frame error rate also change from 3% to 10%.

During the experiments, our proposed system subscribes layers adaptively from (2,1), (1,1), (0,1), (0,2), to (0,4) following rules listed in Table 1 while maintaining satisfactory PSNR. On the other hand, the system without adaptation, always stays at either high rate of (2,1) or low rate of (0,2), break down due to low available bandwidth or not enough FEC protection.

V. CONCLUSION

By taking advantage of QoS features offered by one of the four proposed WiMAX service flow arrangement, our proposed layered scalable video multicast over WiMAX can achieve more flexible layer constructing and subscription while reliable in diverse channel conditions and fitting users’ demand. The system optimality comes from the best tradeoff of number of video layers subscription with number of additional FEC packets insertion to simultaneously satisfy the estimated available bandwidth and the estimated wireless channel error condition.

REFERENCES [1] Q. Zhang, W. Zhu, Y. Zhang, “End-to-end QoS for video delivery over

wireless Internet,” Proceedings of the IEEE, Jan. 2005. [2] S. Floyd, M. Handley, J. Padhye, J. Widrner, "Equation-based

Congestion Control for Unicast Applications," in Proceedings of SIGCOMM'00, Auguest 2000.

[3] G.I. Kwon, J.W. Byers, “Smooth Multirate Multicast Congestion Control,” IEEE INFOCOM '03, April 2003.

[4] S. Cen, P. Cosman, and G. Voelker, “End-to-end differentiation of congestion and wireless losses,” IEEE/ACM Transactions on Networking, Oct. 2003.

[5] M. Chen and A. Zakhor, “Rate control for streaming video over wireless,” IEEE Infocom, 2004.

[6] G. Yang, L.-J. Chen, T. Sun, M. Gerla and M. Y. Sanadidi, “Smooth and Efficient Real-time Video Transport in Presence of Wireless

Errors,” ACM Transactions on Multimedia Computing, Communications and Applications, 2006.

[7] H.-F. Hsiao, A. Chindapol, J. Ritcey, Y.-C. Chen, J.-N. Hwang, “A New Multimedia Packet Loss Classification Algorithm for Congestion Control over Wired/Wireless Channels,” ICASSP, IEEE, March 2005.

[8] Qiang Liu, JaeJun Yoo, Byung-Tae Jang, KyoungHo Choi, and Jenq-Neng Hwang, “A scalable VideoGIS system for GPS-guided vehicles,” Signal Processing: Image Communication, Vol. 20, No. 3, pp. 205-208, March 2005.

[9] Q. Liu and J. N. Hwang, “A Scalable Video Transmission System using Bandwidth Inference in Congestion Control", IEEE Int’l Conf. On Circuits and Systems, Vancouver, May 2004.

[10] Chih-Wei Huang and Jenq-Neng Hwang, “An Embedded Packet Train and Adaptive FEC Scheme for Effective Video Adaptation over Wireless Broadband Networks,” 15th International Packet Video Workshop, Hangzhou China, April 2006.

[11] Heiko Schwarz, Detlev Marpe, and Thomas Wiegand, “Overview of the Scalable H.264/MPEG4-AVC Extension,” IEEE International Conference on Image Processing (ICIP'06), Atlanta, GA, USA, October 2006.

[12] V. J. Ribeiro, M. Coates, R. H. Riedi, S. Sarvotham, and R. G. Baraniuk. “Multifractal cross traffic estimation,” In Proc. of ITC specialist seminar on IP traffic Measurement, September 2000.

[13] B. Melander, M. Bjorkman, and P. Gunningberg, "A new end-to-end probing and analysis method for estimating bandwidth bottlenecks," In IEEE Globecom - Global Internet Symposium, San Francisco, November 2000.

[14] M. Jain and C. Dovrolis. “Pathload: A Measurement Tool for End-to-End Available Bandwidth,” In Passive and Active Measurements, Fort Collins, CO, March 2002.

[15] N. Hu and P. Steenkiste. “Evaluation and Characterization of Available Bandwidth Techniques,” IEEE JSAC Special Issue in Internet and WWW Measurement, Mapping, and Modeling, 2003.

[16] V. J. Ribeiro, R. H. Riedi, R. G. Baraniuk, J. Navratil, and L. Cottrell. “pathChirp: Efficient Available Bandwidth Estimation for Network Paths,” In Passive and Active Measurement Workshop, 2003.

[17] J. Strauss, D. Katabi, F. Kaashoek, “A Measurement Study of Available Bandwidth Estimation Tools,” IMC2003, Miami Beach, Florida, USA, 2003.

[18] Y. Tobe, Y. Tamura, A. Molano, S. Ghost, H. Tokuda, “Achieving moderate fairness for UDP flows by path-status classification,” IEEE LCN2000, Tampa, FL, Nov. 2000, pp.252-261.

[19] IEEE 802.16-2004, “IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface for Fixed Broadband Wireless Access Systems,” Oct. 2004.

[20] IEEE 802.16e-2005, “IEEE Standard for Local and metropolitan area networks Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands and Corrigendum 1,” 2006.

TABLE 2

( , )p q Sets and Resulting ,i j

n

j 1 2 3 4 5 6

Lp 3% 5% 10% 15% 20% 30%

p 0.97 0.95 0.9 0.85 0.8 0.7 q 0.03 0.05 0.1 0.15 0.2 0.3

,FEC ie 0.005 0.005 0.005 0.005 0.005 0.005

ik 8 8 8 8 8 8

,i jn 10 11 12 14 15 19

, /i j in k 1.25 1.375 1.5 1.75 1.875 2.375

1M 600K 750K

3% 10%

Available BandwidthWireless

Error Rate

0 10 20 30 40 50 6015

20

25

30

35

40

Second

PS

NR

Adaptive

Non-Adaptive (2,1)

Non-Adaptive (0,2)

Figure 7: PSNR plots of adaptive and non-adaptive streams.

119