1. Page 1 박 수 현 컴퓨터 및 정보통신공학부 Scalable Video Coding and Transport over Broadband Wireless Networks Dapeng Wu, Yiwei Thomas Hou, And Ya-Qin Zhang, Proceedings of the IEEE, Vol. 89, No. 1, January 2001

2. Page 2 목 차목 차  Introduction  Scalable Video Coding  Network-Aware End Systems  Adaptive Services  Summary

3. Page 3 Problems for real-time video transmission over wireless networks  Unreliability –Compared with wired links, wireless channels are typically much more noisy and have both small-scale(multipath) and large-scale (shadowing) fades, making the BER very high.  Bandwidth fluctuations –When a mobile terminal moves between different networks [e.g., from a wireless local area network (LAN) to a wireless wide area network (WAN)]. –When a handoff happens, a base station may not have enough unused radio resource to meet the demand of a newly joined mobile host. –The throughput of a wireless channel may be reduced due to multipath fading, co-channel interference, and noise disturbances. Last but not least, the capacity of a wireless channel may fluctuate with the changing distance between the base station and the mobile host.  Heterogeneity –Unicast vs Multicast Unicast video distribution using multiple point-to-point connections. Multicast video distribution using point-to-multipoint transmission.

4. Page 4 3 basic components of Adaptive QoS support from networks  Scalable video coding –Scalable video is more suitable than nonscalable video under a time-varying wireless environment. –Scalable video representation is a good solution to the heterogeneity problem in the multicast case. –Scalable video representations naturally fit unequal error protection.  Network-aware adaptation of end systems –To solve these problems Unreliability Bandwidth fluctuations –Network awareness –Network adaptation  Adaptive services –Service contract –Call admission and resource reservation –Mobile multicast mechanism –Substream scaling –Substream scheduling –Link-layer error control

5. Page 5 Scalable Video Coding  SNR scalability –SNR-scalable coding quantizes the DCT coefficients to different levels of accuracy by using different quantization parameters –The SNR-scalable encoder operates in the same manner as both the nonscalable video encoder and decoder one. SNR-scalable encoder with 2 levels SNR-scalable decoder with 2 levels 1) The raw video is DCT transformed and quantized at the base level. 2) The base-level DCT coefficients are reconstructed by inverse quantization. 3) Subtract the base-level DCT coefficients from the original DCT coefficients. 4) The residual is quantized by a quantization parameter, which is smaller than that of the base level. 5) The quantized bits are coded by VLC. 1) Decoded by VLD and inversely quantized. 2) The base-level DCT coefficient values are added to the enhancement-level DCT coefficient refinements. 3 ) The summed DCT coefficients are inversely DCT transformed, resulting in enhancement-level decoded Video. DCT : Diserete Cosine Transform Q : Quantization VLC : Variable Length Coding VLD : Variable Length Decoding IQ : Inverse Quantiazation IDCT : Inverse DCT

6. Page 6 Scalable Video Coding  Spatial scalability –For the base layer, the raw video is Spatially down-sampled. DCT transformed Quantized VLC coded. –For the enhancement layer, the raw video is The raw video is spatially down-sampled, DCT transformed,and quantized at the base layer. The base-layer image is reconstructed by inverse quantization and inverse DCT. The base-layer image is spatially up-sampled. Subtract the up-sampled base-layer image from the original image. The residual is DCT transformed, and quantized by a quantization parameter, The quantized bits are coded by VLC. Spatially/temporally scalable encoder with 2 levelsSpatially/temporally scalable decoder with 2 levels

7. Page 7 Scalable Video Coding  Temporal scalability –Temporally scalable video is encoded by making use of temporally up-sampled pictures from a lower layer as a prediction in a higher layer. –The block diagram of temporally scalable codec is the same as that of spatially scalable codec. –Temporal down-sampling uses frame skipping. Prediction GOP border Key Picture T0T0 T0T0 T1T1 T2T2 T2T2 T3T3 T3T3 T3T3 T3T3 T x : Temporal Layer Identifier

8. Page 8 Network-Aware End Systems  Network monitoring –On-demand monitoring When applications ask the monitor to collect status information about a certain resource in an online fashion. –Continuous monitoring The monitor notifies the application when the status of a previously requested resource changes in a certain way. –Centralized case status information from the entire network is maintained at a central host and shared by all other hosts. –Distributed case Monitors collect only local network status information and obtain nonlocal status information on demand from other network monitors. CriteriaType of monitoring Method of monitoringActivePassive Monitoring frequencyOn demandContinuous Replication of informationCentralizedDistributed

9. Page 9 Network-Aware End Systems  Adaptation 1 Architecture for transporting scalable video from a mobile terminal to a wired terminal At the sender side, the compressed video bit stream is first filtered by the scaler, the operation of which is to select certain video layers to transmit. 1) Scale down the received video representation, that is, drop the enhancement layer(s) 2) Transmit what is received, i.e., do not scale the received video representation. Scaler 1) To notify the sender about the available bandwidth of the wireless channel through a signaling channel The network monitor 1) The rate control module at the sender conveys the bandwidth parameter to the scaler. 2) The scaler regulates the output rate of the video stream so that the transmission rate is less than or equal to the available bandwidth. The rate control

10. Page 10 Network-Aware End Systems  Adaptation 2 Architecture for transporting scalable video from a mobile terminal to a wired terminal At the sender side, the compressed video bit stream is first filtered by the scaler, the operation of which is to select certain video layers to transmit. 1) The network monitor notifies the sender about the channel quality (i.e., BER) The network monitor The rate control module at the sender commands the scaler to perform the following operations. 1)if the BER is above a threshold, discard the enhancement layer so that the bandwidth allocated for the enhancement layer can be utilized by FEC to protect the base layer. 2)otherwise, transmit both layers. The rate control Scaler’s operations

11. Page 11 Adaptive Services  Functions of adaptive services –Reserve a minimum bandwidth to meet the demand of the base layer. As a result, the perceptual quality can always be achieved at an acceptable level. –Adapt the enhancement layers based on the available bandwidth and the fairness policy.  Using scaling inside the network has the following advantages –Improved video quality –Low latency and low complexity –Lower call blocking and handoff dropping probability  The required components of the end-to-end adaptive services –Service contract –Call admission control and resource reservation –Mobile multicast mechanism –Substream scaling –Substream scheduling –Link-layer error control

12. Page 12 Adaptive Services  Service contract –The service contract between the application and the network could consist of multiple subcontracts. –A substream is assigned a priority according to its significance. The base layer is assigned the highest priority. The priority can be used by routing, scheduling, scaling, and error control components of the adaptive network.  Call admission control(CAC) and resource reservation CAC To provide a QoS guarantee for individual connections while efficiently utilizing network resources CAC algorithm has to check whether admitting the connection would reduce the service quality of existing connections, and whether the incoming connection’s QoS requirements can be met Resource reservation In order to maintain the specified QoS in the long time scale, the network must reserve some resource along the current path of a mobile connection. In order to seamlessly achieve the QoS on the short time scale, bandwidth must be reserved on the paths from the current base stations to the neighboring base stations so that in the event of a handoff, a termination of the connection can be avoided

13. Page 13 Adaptive Services  Mobile multicast mechanism –The mobile routing protocol needs to be proactive and anticipatory in order to match the delay, loss, and jitter constraints of a substream. –As a mobile station hands off from a base station to another, new paths are set up and old paths are torn down  Substream scaling –Scaling is employed during bandwidth fluctuations and/or under poor channel conditions. –The scaling decision is made by a bandwidth manager, which obtains the available bandwidth from a network monitor. –Fariness problem’s solutions A max–min fairness A utility-based fairness –Which represents the relationship between observed quality (i.e., utility) and bandwidth

14. Page 14 Adaptive Services  Substream scheduling Architecture for substream scheduling at a base station 1) Delay and throughput guarantees for error-free sessions 2) Long-term fairness for error sessions 3) Short-term fairness for error-free sessions 4) Graceful degradation for sessions that have received excess service time CIF (Channel-condition Independent Fair)’s properties Hierarchical packet-scheduling architecture where a priority link scheduler is shared among a CIF-Q scheduler for base-layer substreams. An FIFO scheduler for enhancement-layer substreams. Service priority is first given to the CIF-Q scheduler and then to the FIFO scheduler.

15. Page 15 Adaptive Services  Link-layer error control FEC (Forward Error Correction) 1)The throughput can be kept constant. 2)Delay can be bounded. 1)The redundancy ratio should be made large enough to guarantee recovery of corrupted bits under the worst channel conditions. 2) FEC is not adaptive to varying wireless channel conditions and it works best only when the BER is stable. 3) FEC is useless when the short-term BER exceeds the recovery capability of the FEC code. ARQ (Automatic Repeat reQuest) 1)ARQ is adaptive to varying wireless channel conditions. 1)Adaptiveness and efficiency of ARQ come with the cost of unbounded delay. VS Advantages Disadvantages Combination of bounded delay and adaptiveness H-ARQ (Hybrid-ARQ) T c : Current time D s : Slack term T d (N) : Packet N’s scheduled time

16. Page 16 Summary  We examined the challenges in QoS provisioning for wireless video transport.  To address the challenges, 3 techniques have been studied in great depth individually. Adaptive QoS support from network (Adaptive services) Network-aware adaptation of end systems Scalable video coding Combinations of 3 techniques An adaptive framework for scalable video transport over wireless networks