«End-to-End QoS Provision over Heterogeneous IP and non IP Broadband Wired and Wireless Network Environments A dissertation submitted in satisfaction ...»
mechanisms. The paper addresses the end-to-end QoS problem of scalable video streaming traﬃc delivery over a heterogeneous IP/UMTS network. It proposes and validates through a number of NS2-based simulation scenarios a framework that explores the joint use of packet prioritization and scalable video coding, by evaluating two FGS video encoders, together with the appropriate mapping of UMTS traﬃc classes to the DiﬀServ traﬃc classes. It is observed that, the scalable extension of H.264/MPEG-4 AVC can achieve better quality gains compared to MPEG-4 FGS, due to the applied motion-compensated prediction technique.
End-to-End QoS Issues of MPEG-4 FGS video streaming traﬃc delivery in an IP/DVB/UMTS networking environment
5.1 Introduction The Fine Grain Scalability (FGS) feature of MPEG-4 is a promising scalable video solution to address the problem of guaranteed end-to-end QoS provision.
In the MPEG-4 FGS standard, a video is encoded into two bitstreams: the Base Layer (BL) and the Enhancement Layer (EL). The BL must be completely received to decode and display a basic quality video. The FGS EL can be cut anywhere at the granularity of bits and the received part can be decoded and improved upon the basic quality video. This FGS, which is achieved by a bitplane coding technique , allows the server to adapt the transmission rate ﬁnely to changing network conditions. In typical scenario for transmitting MPEG-4 FGS encoded videos over heterogeneous networks, the BL is transmitted with the high reliability (achieved through appropriate resource allocation and/or channel error protection) and the EL is transmitted with low reliability (e.g. in a best eﬀort manner and without error control). However, scalable coding only solves part of the problem, and packet loss is very common with unpredictable channel conditions. To address this problem, both an eﬃcient scalable video coding scheme with a ﬂexible delivery technique and a scalable network management framework are needed. By using rate-allocation mechanism, prioritized packetization and diﬀerential forwarding, the application-layer QoS can be provided to the end user.
Concerning the network perspective, an all-IP setting seems to be able to resolve the inter-working amongst the diverse ﬁxed core and wireless/mobile access technologies and the end-to-end QoS provision could be established through the appropriate mapping amongst the QoS traﬃc classes/services supported by the contributing underlying networking technologies. Building this concept, this work concerns a heterogeneous cluster of networks consisting of two DiﬀServaware, a DVB network acting as a trunk network and a UMTS network acting as an access network.
For the ﬁxed networks, the Diﬀerentiated Services (DiﬀServ)  model, proposed by IETF, provides a less complicated and more scalable solution because Integrated Services (IntServ)  requires maintenance of the per-ﬂow state across the whole path for resource reservation. In the DiﬀServ model, resources are allocated diﬀerently for various aggregated traﬃc ﬂows based on a set of bits. DiﬀServ model support two diﬀerent services: (1) the Expedited Forwarding (EF)  that supports low loss and delay/jitter, and (2) the Assured Forwarding (AF)  that provides QoS better than the best eﬀort, but without guarantee. For streaming video applications, in which the encoding and decoding process is more resilient to packet loss and delay variations, besides EF, the AF can be employed.
In order to provide traﬃc diﬀerentiation in a Digital Video Broadcasting (DVB)  network, Bandwidth Management (BM) techniques can be applied on queues containing 188 byte long MPEG-2 TS packets. This technique is based on the dynamic uplink bandwidth reallocation into a number of independent virtual channels according to a predeﬁned set of priority policies. The assignment of an IP ﬂow at a virtual channel is achieved through a ﬁltering mechanism, which is able to monitor traﬃc and based on some pre-deﬁned ﬁlters (IP source and destination addresses, source and destination ports, protocol type, etc) to encapsulate that traﬃc to a speciﬁc virtual channel.
The QoS provision in Universal Mobile Telecommunications System (UMTS)  is achieved through the concept of bearers. A bearer is a service providing a particular QoS level between two deﬁned points invoking the appropriate schemes for either the creation of QoS guaranteed circuits, or the enforcement of special QoS treatments for speciﬁc packets. The selection of bearers with the appropriate characteristics constitutes the basis for the UMTS QoS provision. Each UMTS bearer is characterized by a number of quality and performance factors. The most important factor is the bearers Traﬃc Class; four traﬃc classes have been deﬁned in the scope of the UMTS framework (i.e., Conversational, Streaming, Interactive and Background). The appropriate mapping of UMTS traﬃc classes to the aforementioned DiﬀServ service classes could oﬀer a vehicle for the end-toend QoS provision over a heterogeneous DiﬀServ/UMTS network. In this chapter, I evaluate two diﬀerent mapping approaches of traﬃc classes for the end-to-end QoS provision over a heterogeneous DiﬀServ/UMTS network .
To address the end-to-end QoS guarantees across heterogeneous network, like DiﬀServ/DVB/UMTS, the paper proposes and validates through a number of experimental scenarios an architecture that explores the joint use of rate adaptation with scalable coding, packet prioritization, together with the appropriate mapping of UMTS traﬃc classes to the DiﬀServ traﬃc classes.
This chapter is organized as follows. In Section 5.2, the proposed video coding and prioritization framework for providing QoS guarantees for MPEG-4 FGS
video streaming traﬃc delivery over a heterogeneous DiﬀServ/DVB/UMTS network is presented, in which key components such as the scalable video coding and diﬀerential forwarding across diﬀerent heterogeneous network domains, including ﬁxed and wireless/mobile networks are employed. The testbed conﬁguration details for the media delivery experimental studies and the results of these studies are discussed in Sections 5.3 and 5.4, respectively. Finally, Section 5.5 draws conclusions and discusses directions for further work and improvements.
5.2 Proposed Arrhitecture
The proposed framework is focused on the integration of rate allocation within MPEG-4 FGS video streaming; prioritized packetization based on content and heterogeneous QoS-aware network systems for providing end-to-end QoS over IP/DVB/UMTS systems. The proposed framework is shown in Figure 5.2.
This work deals with the following key components: (1) scalable source encoding with constant quality rate allocation, (2) prioritized packetization, and (3) diﬀerential forwarding across heterogeneous network domains. They are brieﬂy
• Scalable Coding with rate allocation The video sequence is encoded using MPEG-4 FGS codec, where the estimated minimal bandwidth, provided by the network monitoring system, gives the rate constraint for BL. Then the rate-allocation module scales the EL stream based on the feedback of the available bandwidth, to preserve constant quality by referring to R-D samples, produced by the video analysis of the video sequence.
• Prioritized Packetization Fixed length packetization scheme is adopted, to packetized BL and EL bit-streams, as proposed by MPEG-4 . It applies priorities based on the loss impact of each packet to the end-to-end video quality.
• Diﬀerential Forwarding - The focus of network-level QoS mapping is to ensure the vertical QoS continuity across diﬀerent network domains. Basically, the application, network, and data link layers are involved in this mapping. The main motivation is to assign diﬀerent priorities to parts of a video bit stream that represents the content on the application layer. The BL in case of scalable bit-stream, is regarded as most important for the decoding process and, therefore, should be transmitted with a higher priority than less important EL, and so on. These priorities at the application layer are then mapped to diﬀerent Diﬀerentiated Service Code Points (DSCPs)  at the network layer. That is, packets containing important parts of the bitstream receive a higher packet priority than packets containing less prioritized parts. This can be realized by using diﬀerent QoS classes that diﬀer only in the drop probability (e.g., AF11 for high priority packets and AF12 for low priority packets). These diﬀerent priorities at the network layer may be mapped to QoS mechanisms available at DVB BM virtual channels and the UMTS traﬃc classes. UMTS oﬀers four diﬀerent classes, which can be used for service diﬀerentiation between real-time traﬃc (e.g., video streaming) and best-eﬀort traﬃc. Authors propose the mapping of DSCPs to the BM virtual channels and UMTS traﬃc classes in order to ensure the vertical QoS continuity across diﬀerent network domains.
5.2.1 Rate allocation with scalable video coding
The video sequence is encoded based on the estimated minimal bandwidth, provided by the network monitoring system, which gives the bandwidth requirements for BL. The encoding is being performed based on the collected statistics, generated by the video sequence analysis. For the EL, the generated R-D samples are stored either in the user data of each Video Object Plane (VOP) or as metadata in a separate ﬁle. Then, the rate allocation module truncates the EL stream, according to the feedback of the available bandwidth in order to increase quality by referring to information, provided by the generated R-D samples.
To make practical and eﬀective use of MPEG-4 FGS encoding, a rate control algorithm is needed to transfer the rate constraint into the rate assigned to each frame, and also to minimize the variation quality. A simple method is constant quality rate allocation (CBR) but the usage of this method does not achieve high results in overall video quality due to quality ﬂuctuations. In order to tackle this problem, variable bit-rate (VBR) allocation is proposed for constant quality reconstruction by allocating rate according to the complexity of each frame .
Authors in  propose an optimal rate allocation using an exponential model.
In , constant quality rate allocation is proposed that minimizes the sum of absolute diﬀerences of qualities between adjacent frames under the rate constraint.
However the optimality of this approach depends on the initial condition, which is computed based on the assumption that the average distortion of CBR rate allocations is close to the distortion of the constant quality rate allocation. In fact, the two distortions must be within the same R-D sample interval for all frames in order to have a valid solution to the set of linear equations. According to piecewise linear interpolation, described in , the rate allocation can be calculated by
where Ctotal is the available bandwidth, N is the total number of frames, Ri denotes the source frame rate, and Ri is the optimal rate that should allocated to i frames in order to achieve the constant distortion D. Consider Rmi, Dmi and Rni, Dni to be two ajdacent R − D points, such that Dmj D Dni and Rni. In the above equations, ∆Ri = Rni −Rmi and ∆Di = Dmi −Dnmi Rmj R represent the diﬀerence in rate and distortion at adjacent R-D points, respectively.
5.2.2 Prioritized Packetization Scheme
In order to packetize MPEG-4 video streams, ﬁxed-length packetization scheme is adopted, where video packets of similar length are formed. The packet size of video stream is also related to eﬃciency and error resiliency because a smaller packet size for example requires a higher overhead but has a better performance in error prone networks. Then, each packet is identiﬁed by a particular priority in accordance with its impact on end-to-end visual quality. For diﬀerent service preferences in terms of loss and delay, the priority can be further divided into the RLI and RDI, as authors proposed in  .
To determine packet priority with low computational complexity is an active research area today. Several features, such as initial error strength, propagation via motion vectors, and the spatial ﬁltering eﬀect were used to develop a corruption model in  to determine packet priority in terms of loss impact. For BL packets, we use a ﬁxed Equal Error Protection (EEP) scheme, where all packets are high priority and they are transmitted using the EF class.
The packet loss within the EL only aﬀects a single frame, and it does not propagates, the incurred distortion from each EL packet can be accurately calculated
within each frame, and the packet priority can be calculated as:
where ∆Di represents the incurred distortion due to the speciﬁed loss and ∆Ri is the rate of the packet concerned. Furthermore, packet dependency must be considered to that if packets containing a more signiﬁcant bitplane get lost, packets containing a less signiﬁcant bitplane in the same region get discarded anyway. By using the piecewise-linear R-D model for each bitplane, the priority of EL packets can be easily calculated online during the packetization procedure.
However, for simplicity reasons, a simpler QoS mapping policy in this framework is adopting, by using direct mapping of packets to DiﬀServ classes. All packets are formed into three groups, according the type of context that they contain, and each group of packet is mapped to one DiﬀServ class. Table 5.1 depicts the relation between the type of the EL content and the corresponding DiﬀServ classes. The ﬁrst digit of the AF class indicates forwarding priority and
the second indicates the packet drop precedence.
5.2.3 DVB Domain - BM Implementation The bandwidth reallocation among the IP virtual channels of a DVB MPEG-2 TS uplink is based on a set of predeﬁned priority policies . The paper implements
three priority policies, which are: