Download Raptor Code Network System and more Summaries Network Programming in PDF only on Docsity!
Raptor Code-Aware Link Process for Spectrally Efficient Unicast
Video Streaming over Mobile Broadband Network.
G.Kamalakannan^1 , G.Angeline Prasanna^2
(^1) M.Phil Research Scholar, PG & Research Dept. of Computer Science Kaamadhenu Arts & Science College, Sathyamangalam - 638 503 kamal.saka@gmail.com (^2) Associate Professor of Comp. Science, PG & Research Dept. of Computer Science Kaamadhenu Arts & Science College, Sathyamangalam - 638 503 angelineprasanna@gmail.com
Abstract
This paper proposes novel Raptor-aware link adaptation (LA) when application layer Forward Error Correction (AL-FEC) with Raptor codes is used for live, high quality, video unicast over mobile broadband networks. The use of Raptor code AL-FEC is taken into account for the adaptation of the modulation and coding scheme (MCS) used in the physical layer. A cross-layer optimization approach is used to select the Raptor code parameters and the MCS mode jointly, in order to maximize transmission efficiency. The proposed methodology takes into consideration the channel resources required to accommodate the Raptor overheads. Simulation results show that packet loss is eliminated and the amount of radio resource required is reduced significantly. Automatic repeat request (ARQ) based unicast systems require up to 115.6 percent more channel resources, by comparison to the proposed Raptor- aware LA system without retransmissions. Furthermore, the Raptor- aware LA system can enhance the link budget by up to 4 dB, increasing coverage in LoS locations, and can improve total good put by 46.7 percent compared to an ARQ-based system.
INTRODUCTION
MOBILE video traffic is rising significantly [1], mainly due to a significant rise in smart phone video applications. Thus supporting high quality video on mobile devices is very important. The efficiency of future wireless networks must be optimized to meet the convergence of video and data while strict quality of service (QoS) is guaranteed for each of the competing user streams. It is well-known that high quality video transmission over wireless networks is challenging because of the time-varying channel quality, the high data rates required and the stringent QoS demands of video. Mobile WiMAX [2] and 3GPP LTE [3] represent mobile broadband standards that offer high user data rates and QoS support for video applications. Both technologies use similar Downlink (DL) PHY layers and have strong similarities in their MAC layers, while radio resource efficiency is pivotal in supporting QoS for multimedia services [4]. Thus this work is also applicable to LTE broadband networks. The use of Raptor code FEC at the application layer, applying cross-packet FEC, has been adopted for robust video broadcasting, by the 3GPP Multimedia Broadcast and Multicast Services (MBMS) standard [5] and DVB-H [6]. The error correcting capability of Raptor codes, even in severe channel conditions, has been well established. Raptor codes have been studied for wireless video broadcasting and multimedia download, for example over 3GPP MBMS networks in [7], [8], over WiMAX in [9], over DVB-H in [10]. Unicast video transmission is generally provided as a value-added service on top of video multicasting but it permits individualized error protection based on receiver feedback and channel quality indicators (CQIs) generated by the MAC/PHY layers. In unicast transmission LA is performed and, typically, the ARQ mechanism is enabled to allow for retransmissions if packets are received in error at the MAC layer. The use of Raptor code AL-FEC for unicast video trans-mission has not been thoroughly investigated in the literature, however prior work has focused on unequal error protection for video and/or scalable video coding, e.g., [9]. Adaptive AL-FEC for unicast video transmission
has been proposed in [11], [12], identifying the need to carefully select the FEC code overhead suitable for the level of packet loss experienced. AL-FEC codes introduce additional over-heads that place high demands on valuable radio resources, in particular when high overhead codes are used. If the packet loss rate is overestimated, then wireless bandwidth will be wasted and if it is underestimated the received video quality will suffer due to errors and/or missing packets. In this work we propose a novel Raptor-aware LA method, where Raptor code AL-FEC is used as a cross-packet erasure code to protect unicast live H.264/AVC video streaming, without ARQ. The MCS mode is adapted jointly with the Raptor code rate used, according to the specific channel conditions and the channel feedback received. In unicast transmission it is common for the PHY to receive feedback on the channel quality in order to adapt the MCS on a per user basis. We focus on minimizing the bandwidth required by Raptor code AL-FEC while maintaining quasi-error free reception, by selecting the Raptor code parameters jointly with the MCS mode, using a novel cross-layer design. It has been shown , e .g. [13], that cross-layer designs are necessary since they enable intelligent resource allocation to optimize spectrum efficiency, while providing high QoS.
Related Work and Our Contribution
The first point of differentiation with related work is that the selection of the MCS mode is usually not considered when using AL-FEC based on rateless codes. Typically more robust MCS modes are selected at the expense of through-put. However, MCS selection should take into account the use of AL-FEC to protect the video data stream and the additional robustness that this provides. Thus more efficient MCS modes could be used. For example, in [11] adaptive unicast video streaming with rateless codes is investigated with no consideration of the MCS used. Fez et al. [12] studies the use of adaptive LDPC AL-FEC for delivery of files in a carousel system over FLUTE [14] based on user feedback. The analysis there does not account for the MCS selected and its impact on channel resources and file delivery does not impose the strict latency constraints of live video trans-mission. In [15] adaptive rateless coding is studied for unicast video streaming over mobile WiMAX, for 16QAM 1/2 only, without considering link adaptation. In [16], considering SVC video, the MCS mode is selected according to the channel SNR and the SVC video layer, while the RS-type AL-FEC is adjusted for each SVC video layer. In [9] the authors identify the need to predict the amount of redundant data according to the channel loss in order to reduce FEC data overheads for IPTV broadcasting, but the selection of MCS mode is not considered. In [8] the joint selection of the AL-FEC Raptor code rate and the code rate at the PHY layer FEC was investigated in order to minimize energy for downloads over FLUTE. However, this work also does not adjust the modulation scheme, since link adaptation for unicast transmission is beyond its scope. Failing to take into account the MCS mode and how this can be adjusted when adaptive AL-FEC codes are used, results in wasting valuable channel bandwidth. A solution to this problem has not been reported in the literature. Here we select the MCS mode and the Raptor AL-FEC code rate jointly, so that the bandwidth required is minimized and transmission effi-ciency is maximized, while observing a given QoS at the application layer. Our work shows the benefits that Raptor-aware LA offers. When rateless codes are used for unicasting they can be based on two types of feedback, either feedback from the higher layers (transport or application layer) or feedback from the MAC/PHY layers. The former can provide a statistical average of Raptor decoding failure for a number of source blocks, which could be used to adjust the FEC code overhead [11], [12]. The time interval for the Raptor decoder feedback depends on the source block size and introduces latency through the round trip time, as shown in [11]. Nevertheless, such feedback is not able to follow the rapidly changing nature of a wireless channel, because the duration of a source block (containing a large number of symbols) commonly exceeds the channel coherence time. The second type of feedback, generated from the PHY, is based on the channel state estimation used for LA at the MAC, such as the channel quality indicator. The CQI feedback frequency in WiMAX/LTE is every 1 to 3 OFDMA frames (5 ms each), according to [2], [17], depending on the rate of channel change. Our work here does not consider feedback from the Raptor decoder at the higher layers, as [11]. Unlike [11], where redundant symbols are sent until a source block is correctly acknowledged, our proposed system does not expect a higher layer feedback, thus no latency is introduced.
This work addresses the QoS requirements of live, high bitrate, at 1.03 Mbps, H.264/AVC video transmission. The work is focused on a pedestrian scenario with a mobile station (MS) speed of 1 km/h, in a dense urban environment. At low MS speeds the channel coherence time is long and the channel may remain in deep fades for longer periods. This results in long packet error bursts and it has been shown, e.g., [21], [23], that in such channel conditions the ARQ retransmission mechanism is not very effective. Here only direct transmissions between the base station and a mobile terminal are considered. In summary, this paper studies the performance of a novel Raptor-aware LA methodology when live, high quality H.264/AVC video is sent over a unicast mobile WiMAX link. Our novel cross-layer optimization approach maximizes the data good put while maintaining a near zero packet loss at the application layer. The remainder of this paper is organized as follows. Section 2 outlines the principles of Raptor code AL- FEC and summaries the IEEE 802.16e protocol. Section 3 describes the simulator developed and the proposed cross-layer design. In Section 4 the cross-layer optimization methodology is explained, while in Section 5 the Raptor-aware LA performance is analyzed. Conclusions are presented in Section 6. BACKGROUND Raptor Codes: The use of cross-packet FEC at the application layer, based on erasure codes to protect multimedia data from packet loss, is well established [28]. Raptor codes are a class of rate-less or fountain codes [29] first introduced in [30] that have been widely selected because of their unique properties. Their flexibility, in terms of the number of source symbols, K, and encoded symbols they generate, N, overcomes the limitations of other well-known erasure codes, such as Reed-Solomon (RS) codes, that can only support a limited number of input and encoded symbols (typically K <_ N _ 255) [28]. A rateless code can generate as many repair symbols as desired from the source symbols. Also Raptor codes have low decoding complexity, linear in K, enabling software-based implementation, whereas RS decoding is prohibitively complex, non-linear in K [8]. Finally, Raptor codes operate close to ideal erasure codes, offering overhead efficiency [28]. The Raptor decoder will recover the source data with high probability, if any of K(1+ð) symbols (source or repair) are successfully received, where d is a small real and ð > 0 [29]. The Raptor encoder partitions incoming data packets into several source blocks. Each source block consists of a number of source symbols, K, each of length T bytes. For each source block a number of repair symbols, R, also of length T bytes are generated. Raptor codes, as specified in 3GPP MBMS [5], are systematic codes. The encoding and decoding algorithms are described in [5]. For the systematic Raptor codes, K encoded symbols are identical to the original K source symbols while R repair symbols are generated. In total N = K + R Raptor encoded symbols are transmitted for each source block. The Raptor code rate c is defined here as c = K/N. IEEE 802.16e Medium access control (MAC) layer—The 802.16e MAC layer [2] includes a number of adjustable features, such as adaptive MCS, ARQ, packet fragmentation and aggregation, variable size MAC Protocol Data Units (PDU), application specific service flows and PDU scheduling based on QoS. Packets from the higher layers arrive at the convergence sub layer (CS) of the MAC as MAC service data units (SDUs). Based on their QoS requirements, MAC SDUs are classified into service flows. There is the option for SDU fragmentation into MAC Protocol Data Units and this feature is assumed here, because small PDU sizes (less than 200 Bytes) were shown to improve the error rate in [31]. SDUs are also partitioned into ARQ blocks of fixed size. The MAC PDU is the data unit exchanged between the BS and MS MAC layers. Once a PDU has been constructed, it is placed in the appropriate service flow queue and managed by the scheduler, which determines the PHY resource allocation (i.e., bandwidth and OFDMA symbol allocation) on a frame-by-frame basis. Each transmitted PDU is either received correctly or in error. The standard specifies a number of ARQ feedback
mechanisms, such as cumulative ACK and selective ACK (S-ACK) [2]. Physical layer (PHY)—The mobile WiMAX standard has adopted Scalable-OFDMA (S-OFDMA)[2].
Table 1 shows 404 IEEE TRANSACTIONS ON MOBILE COMPUTING
the relevant parameters for the S-OFDMA PHY. Simulations were performed assuming a 10MHz channel profile (highlighted in italics in Table 1). Our mobile WiMAX PHY layer simulator is described in [32]. The payload data is modulated using the full range of MCS modes as defined in [2] and shown in Table 2. Assuming a Partial Usage of Sub-Channels (PUSC) DL, as the only mandatory sub-channelization method in [2], the modulation symbols allocated to a sequence of slots in each DL OFDMA frame are assigned to a number of logical sub channels. According to [2] a slot is the minimum PHY resource allocation unit and for PUSC DL it is defined as one sub channel by two OFDMA symbols. For the 10 MHz profile, an OFDMA symbol consists of 30 sub channels for PUSC DL, each containing 24 data subcarriers [33]. Hence a slot contains 48 data subcarriers. Based on this, the slot payload capacity Psl for each MCS mode is calculated as shown in Table 2, where m represents the MCS modulation order and r the coding rate. The channel resources (in terms of slots) required for data transmission over a mobile WiMAX network are evaluated based on the slot payload capacity for each MCS mode. In our simulator a typical DL/UL ratio of 22:15 data symbols is assumed [33]. Our analysis methodology, however, can be used for any allowed channel bandwidth or DL/UL ratio. 2.3 GHz and the FFT size is 1,024. The MS speed is set at 1 km/h to simulate a typical pedestrian user. Each radio channel is made up of a large number of channel realizations, corresponding to the duration of the simulated trans- mission of 2,000 UDP packets at 1.03 Mbps. ESM PHY abstraction—To simplify the interface between the PHY link level and the MAC simulator, while modeling dynamic system behavior, a technique known as Effective SINR Mapping is used [25], [34].
The performance of a mobile WiMAX system is studied when the proposed Raptor-aware LA is used. It is then com-pared with an ARQ-enabled mobile WiMAX system [2] using S-ACK, following the recommendations in [36]. A detailed analysis of the ARQ-enabled mobile WiMAX sys-tem was given in [21]. In order to study the Raptor code AL-FEC performance over a mobile WiMAX network, the transmission of a flow of UDP packets is simulated through the MAC and PHY layers of 802.16e. Simulation is per-formed for a single MS user, however the 3GPP SCM channel model used generates a time- correlated fading link of several thousand realizations for a generic user, which means that the statistics of the simulation model apply to any user. Thus results can extend trivially to multiple links. Simulation is run for different mean channel SNR values and results are averaged over the small scale fading statistics for each mean SNR, as depicted in Fig. 2 so our results are representative and extendable to multiple links. Multi-user interference is not studied in this work. The simulation would include multiple links/users in order to study net-work congestion, but this is beyond the scope of this work. The unicast transmission of high bitrate live video certainly poses the problem of the bandwidth required and how many unicast users can be served, given the available net-work bandwidth. Thus, this study aims to minimize the bandwidth required per user (while maintaining a required level of QoS) so that more users may be served. Application Layer According to the H.264/AVC standard [37], video data are incorporated in network abstraction layer units (NALUs) with a maximum length of SP Bytes. It is assumed that each H.264 NALU is encapsulated in one RTP/UDP packet, as depicted in Fig. 3. UDP packets are of fixed size, SP Bytes, as also assumed in [7], [8], [38]. According to the standard [5], the Raptor encoder operates on a stream of incoming RTP/UDP packets at the application layer, to encode the UDP packets. In order to minimize jitter in the received video sequence due to the encoding process, as in [7], [28], [39] it is assumed that the video sequence is sent as a constant bit rate (CBR) flow. The Raptor encoder collects UDP packets to form blocks of source data. Each source block of size K _ T Bytes consists of K source symbols of length T Bytes. In order to form a source block of size K _ T Bytes, it is assumed that n UDP packets are required. For each source block of K source symbols a number of repair symbols R, of length T Bytes are generated by the Raptor encoder. The systematic Raptor encoder specified in [5] is used here, with Raptor code rate c (c ¼ KKþR). The
Raptor encoder generates l repair UDP packets per source block (containing the 406 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 14, NO. 2, FEBRUARY 2015 Fig. 3. Cross-layer Raptor FEC simulation. Raptor repair symbols) [5], also of length SP Bytes. According to the 3GPP standard [5] the source and repair data form two separate UDP flows on the network. The process is depicted in Fig.
- The receiver recovers the source symbols from the FEC encoded packets. 2. 802.16e MAC and PHY layers At the MAC layer each UDP packet is mapped to one MAC SDU and then transmit according to the 802.16e MAC layer protocol. Thus the MAC SDU error rate is equal to the UDP PER. SDUs are fragmented into PDUs. PDU packing is not enabled here. It is assumed that each MAC SDU is partitioned into ARQ Blocks, of length T also, as will be discussed next. At the MAC a Raptor-aware scheduler is assumed for the DL video transmission, which takes both flows (source and repair packets) into account, according to the selected Rap-tor code rate c. Resource allocation is dependent on the selected data modulation MCS mode, according to Table 2. The slot payload capacity increases with higher MCS modes, decreasing the number of slots required for the same bit rate. The full range of MCS modes supported by 802.16e (Table 2) has been simulated, using the PUSC DL sub-channelization method for a 10 MHz channel profile (Table 1). 802.16e MAC and PHY layers According to [2], at the receiver the MAC waits to acquire all the ARQ blocks forming a MAC SDU before reassembling the SDU. SDUs are delivered as UDP packets to the Raptor decoder at the application layer. An SDU will not be delivered to the higher layers if any of its ARQ blocks are lost or discarded [2]. This policy however, aggravates the SDU packet error rate seen at the receiver in comparison to the ARQ block error rate (BLER). From our work in [21] it is clear that for the same mean channel SNR and MCS mode, the UDP PER is significantly higher than the BLER. If the whole SDU is discarded then much larger overheads are required for the AL- FEC. Since the AL-FEC is in place, sys-tem performance and FEC overhead efficiency would greatly benefit if SDUs with missing ARQ blocks were delivered to the Raptor decoder. The traditional IP receiver policy ignores a significant amount of correctly received data. The detrimental effect of this policy on wireless transmission for 3GPP MBMS has been also studied in [24], [27] in conjunction with Raptor AL-FEC, where a permeable layer receiver (PLR) was pro-posed to allow the
The heuristic cross-layer optimization algorithm proposed here identifies the pair of MCS, m and Raptor code rate, c, ðmi; ciÞ that maximizes good put- per-frame at each mean channel SNR value si, for a source block length K and a given target EUDP. The symbol length T remains constant. This methodology jointly controls the AL-FEC redundancy and the MCS mode according to the specific channel conditions and the system attains the highest good put for the least amount of PHY layer resource, offering the required level of QoS. The cross-layer optimization process is summarized by Algorithm 1.
Algorithm 1.
Cross-Layer Optimization for Selection of Optimum Raptor Code c and MCS Mode For each mean channel SNR and for each source block length K:
- Select the suitable MCS modes mi that with Raptor AL-FEC delivers UDP PER _ EUDP. The candidate modes constitute a setMs.
- 8 mode mi _ Ms locate the Raptor code rates c that attain PER _ EUDP and form a set of appropriate code rates, Ci ¼ fc1; c2;... ; clg. If no code rates achieve the target EUDP then mode i is excluded from the setMs and Ci ¼? f g.
- 8 mode mi _ Ms form the candidate pairs of code rates for the candidate mode i as ðmi; c1Þ, ðmi; c2Þ;... ; ðmi; clÞ.
- Form the set P of all suitable pairs for all modes i in Ms, and their appropriate code rates Ci. Within the set P of all candidate pairs, find the pair ps ¼ ðmk; cyÞ for which the goodput-per-frame is maximized. This is the recommended pair for the specific mean channel SNR and aptor parameter K.
RAPTOR-AWARE LA PERFORMANCE ANALYSIS
The performance of the Raptor-aware LA is studied in terms of PER, channel resources required and total system good-put. The optimized Raptor unicast system performs Raptor-aware LA using the optimum pairs ðm; cÞ shown in Table 4, selected by the cross-layer optimization algorithm. Table 5 reports the performance results for K ¼ 1;820 , showing the PER attained and the average percentage of slots required per DL-subframe, for mean channel SNR values in the range of 8 to 22 dB. It is observed that zero PER is attained at all SNR values of interest. It should be noted that the minimum observable PER in these simulation results is limited to 5 _ 10_4; given the number of simulated UDP packets (2,000). The results in Table 5 also show that the DL channel resources required decrease for higher throughput MCS modes and higher code rates, as expected. The optimized Raptor unicast system is compared with an ARQ-enabled mobile WiMAX system [2], with block lifetimes of 65 and 90ms. Based on the WiMAX Forum recommendations [17], [36], a maximum of four ARQ retransmissions are allowed for a block lifetime of 90 ms, and up to three retransmissions for a block lifetime of 65 ms. The ARQ-enabled mobile WiMAX system was analyzed in [21] and the ARQ performance results here are taken from that work. For a given ARQ block lifetime at each mean channel SNR value, the MCS mode delivering the highest good-put with UDP packet error rate PER _ EUDP is selected, based on the good put-per-frame metric [21]. The constraint PERUDP _ EUDP is applied for a target value EUDP ¼ 10_2 and if that is not achieved it is then assumed that the ← The optimized Raptor system with Raptor-aware LA, uses higher throughput MCS modes at low channel SNR values, whereas for both ARQ block lifetimes only MCS mode 0 can be used, for SNR values up to 14 dB, in order to achieve PER _ 10_2. ← The optimized Raptor system offers a quasi-error free video service for mean channel SNR values as low as 8 dB. By comparison, the ARQ-enabled sys-tem cannot deliver PER _ 10_ below 12 dB, when the ARQ block lifetime is 65 ms, and it is assumed that the video service would not be available. For a block lifetime of 90 ms the service would not be available below 10 dB. Thus the optimized Raptor system offers an extension of service range. ← The optimized Raptor system requires less DL channel resources than the ARQ-enabled system, for most of the channel SNR range studied. The highest reduction occurs at 14dB, where the ARQ-enabled mobile WiMAX system (for both block lifetimes) uses MCS mode 0 and requires 33.4 per- cent of slots per DL sub frame. The optimized Rap-tor system at the same SNR value requires only 9. Percent of the available DL resources, with the optimum pair [mode ¼ 2, c ¼ 0.7]. In this case the ARQ-enabled mobile WiMAX system requires 8. Percent more channel resources than the optimized Raptor system. ← the optimized Raptor
system offers zero PER, whereas the ARQ-enabled system has a residual PER _ 0:01 at all SNR values (which may affect the video quality for some decoders). The ARQ block lifetime of 65 ms has a slightly higher residual PER than when the block lifetime is 90 ms.
CONCLUSIONS:
Our results have shown the benefits of a joint selection of MCS mode with the Raptor code rate, through Raptor-aware Link Adaptation and a cross- layer optimization approach. Results show that this approach increases the transmission efficiency for unicast streaming of high quality live video, while the video is delivered error-free. The ARQ-enabled WiMAX system (with a block lifetime of 65 ms) requires up to 115.6 percent more channel resources at an SNR value of 18 dB, than the optimized Raptor-aware LA system, when the maximum residual UDP PER is set 10_3. The proposed Raptor- aware LA also offers an increase in the total good put by 46.7 percent, in the SNR range 14-16 dB, and an extension of the service range by 4 dB, compared to an ARQ-enabled WiMAX system, with a block lifetime of 90 ms. The trade-off between MCS mode throughput and Raptor code overhead was investigated for a range of mean channel SNR values, Raptor code rates and source block lengths K, for different UDP PER targets. Results showed that the larger source block length, K ¼ 1;820, is more efficient than K ¼ 1;040. These encouraging results open a discussion for practical issues regarding possible implementation in mobile broad-band systems. This work does not attempt to offer a full practical solution to the problem, as the parameter set can be extended (e.g., packet sizes, PER target values) and practical, system-specific issues should be considered. Further work is required to design a practical implementation of a Raptor-aware LA algorithm. The methodology discussed in this paper is applicable to other mobile broadband networks apart from mobile WiMAX.
REFERENCES :
1] CISCO, “White paper, global mobile data traffic forecast update, 2010- 2016,” Cisco Visual Networking Index, (Feb.2012).[Online].Available: http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/ whitepaperc11520862.html 2] IEEE Standard for Local and Metropolitan Area Networks Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems, IEEE Std., 2006. 3] LTE; Evolved Universal Terrestrial Radio Access (E-UTRA) and Evolved Universal Terrestrial Radio Access Network (E-UTRAN); Overall Description; Stage 2 (3GPP TS 36.300 version 10.5.0 Release 10), 3GPP Std, 2011. 4] A. Esmailpour and N. Nasser, “Dynamic QoS-based band-width allocation framework for broadband wireless networks,” IEEE Trans. Veh. Technol., vol. 60, no. 6, pp. 2690–2700, Jul. 2011. 5] 3GPP TS 26.346 V8.0.0 (2008-10) Universal Mobile Telecommunications Syst. (UMTS); Multimedia Broadcast/Multicast Serv. (MBMS); Protocols and Codec’s, ETSI Std., 2008. 6] ETSI TS 102 005 , Digital Video Broadcasting (DVB): Specification for the Use of Video and Audio Coding in DVB-H Services Delivered Directly Over IP-Based Protocols, ETSI Std., 2006. 7] J. Afzal, T.Stockhammer, T. Gasiba, and W. Xu, “System design options for video broadcasting over wireless networks,”in Proc. 3rd IEEE Consum. Commun. Netw. Conf., Jan. 2006, vol. 2, pp. 938– 943. 8] M. Luby, T. Gasiba, T.Stockhammer, and M. Watson, “Reliable multimedia download delivery in cellular broadcast networks,” IEEE Trans. Broadcasting, vol. 53, no. 1, pp. 235–246, Mar. 2007. 9] L. Al-Jobouri, M. Fleury, and M. Ghanbari, “Robust IPTV delivery with adaptive rateless coding over a mobile wimax channel,” ISRN Commun. Netw., vol.2011, p. 11, 2011.
