«End-to-End QoS Provision over Heterogeneous IP and non IP Broadband Wired and Wireless Network Environments A dissertation submitted in satisfaction ...»
type of payload that identify each packet. When a packet does not reach the destination, it is counted as a lost packet. Furthermore, not only the actual loss is important for the perceived video quality, but also the delay of packets/frames and the variation of the delay, usually referred to as packet/frame jitter. The packet/frame jitter can be addressed by the so called play-out buﬀers. These buﬀers have the purpose of absorbing the jitter introduced by the network delivery delays. It is obvious that a big enough play-out buﬀer can compensate any amount of jitter. There are many proposed techniques in order to develop eﬃcient and optimized play-out buﬀer, dealing with this particular trade-oﬀ. These techniques are not within the scope of the described testbed. For our experiments the play-out buﬀer is set to 1000msecs.
In order to measure the improvements in video quality by employing H.264/MPEGAVC, I use the Peak Signal to Noise Ratio (PSNR) and the Structural Similarity (SSIM)  metrics. P SN R is one of the most widespread objective metric for quality assessment and is derived from the Mean Square Error (MSE) metric, which is one of the most commonly used objective metrics to assess the application level QoS of video transmissions .
Let’s consider that the video sequence is represented by v(n, x, y) and vor (n, x, y), where n is the frame index and x and y are the statial coordinates. The average P SN R of the decoded video sequence among frames at indices between n1 and
n2 is given by the following equation:
where V denotes the maximum greyscale value of the luminance. The average M SE of the decoded video sequence among frames at indices beteen n1 and n2
is given by:
Note that, the P SN R and M SE are well-deﬁned only for luminance values.
As it mentioned in , the Human Visual System (HVS) is much more sensitive to the sharpness of the luminance component than that of the chrominance component, therefore, I consider only the luminance PSNR.
SSIM is a Full Reference Objective Metric  for measuring the structural similarity between two image sequences exploiting the general principle that the main function of the human visual system is the extraction of structural information from the viewing ﬁeld. If v1 and v2 are two video signals, then the SSIM
is deﬁned as:
where L is the dynamic range of pixel values and K1 = 0.01 and K2 = 0.03, respectively.  deﬁnes the values of K1 and K2.
At the ﬁrst scenario, I examine the transmission of H.264 scalable video streams consisting of two layers. The BL is encoder at 256Kbps, while the EL is encoded at 512 Kbps. As video source is used the Foreman YUV QCIF video sequence (176x144) consisting of 400 frames. The underlying network for the ﬁrst measurement is a simple Best Eﬀort network, like Internet, without implementing any QoS model for guarantee end-to-end video quality. The video frame is sent every 33 ms for 30 fps video. Figure 6.3 shows the PSNR graph for the experimental scenario described above. The Y axis represents the P SN R value in dB while the Xaxis represents the frame number of video sequence.
Figure 6.3: Scalable video transmission over best-eﬀort networks
As one may observe from Figure 6.3, during severe network congestion caused by interference by background traﬃc, the PSNR values are between 10dB and 12dB. The average value of PSNR, Pavg is 29.038dB. Note that, a frame is counted as lost also, when it arrives later than its deﬁned playback time.
The same measurementis repeated, but instead of using a best-eﬀort network, we use a network that implements the proposed model. The mapping of packets is based on Table 6.1. The DiﬀServ routers implement WRED queue management. In this scenario, according to Figure 6.4, the overall PSNR is better than ¯ without using prioritization. The Pavg value is 31.054dB. Figure 7.4 depicts the SSIM metric of both scenarios (BE and QoS-enabled networks) for foreman video sequence.
Figure 6.4: Scalable video transmission over DiﬀServ/802.11e Heterogeneous Net-work
The same measurement is repeated for four diﬀerent YUV video sequences consisting of 300 to 2000 frames. For all the scenarios, it is considered the simple but eﬃcient error concealment scheme described in the previous section. The
average P SN R and SSIM for the above scenarios are shown in Table 6.4, where:
• in Scenario 1 it is transmitted scalable H.264/MPEG-4 AVC video stream in a best eﬀort network.
• in Scenario 2 it is transmitted scalable H.264/MPEG-4 AVC video stream over a DiﬀServ/802.11e heterogeneous network.
Figure 6.5: SSIM measurements of scalable video transmission over DiﬀServ/802.
11e Heterogeneous Networks As it seems in Table 6.4, the proposed prioritization scheme improves the overall quality of the received video. By isolating the losses and the delays to packets that contain and C partitions it can achieved signiﬁcant gains to video quality. By distributing the traﬃc to all traﬃc classes, anyone can achieve equal or even better video quality, in the lowest price, by sending lowest traﬃc to the cost eﬀective EF/AC3 traﬃc. From the network provider perspective, the utilization of the network is more eﬃcient, by serving more users, at the level of quality they pay.
6.4 Conclusions Nowadays, continuous media applications over heterogeneous all-IP networks, such as video streaming and videoconferencing, become very popular. Several approaches have been proposed in order to address the end-to-end QoS both
from the network perspective, like DiﬀServ and 802.11e access categories, and from the application perspective, like scalable video coding and packetized prioritization mechanisms. In this chapter, the end-to-end QoS problem of scalable video streaming traﬃc delivery over a heterogeneous DiﬀServ/802.11enetwork is being addressed. 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 scalable extension of H.264/MPEG-4 AVC, together with the appropriate mapping of 802.11e access categories to the DiﬀServ traﬃc classes. The proposed prioritization scheme in conjuction with the proposed DiﬀSer/802.11e classes coupling have improvements in the overall quality of the received video, by isolating the losses and the delays to packets carrying less important partitions.
A Pricing framework for adaptive multimedia services over QoS-enabled Heterogeneous networking environments
7.1 Introduction The network resources in the Internet are dynamically shared among a large number of users, posing a signiﬁcant challenge in the guaranteed provisioning of quality-of-service (QoS) to individual users. During the last several years, QoS issues in the Internet have attracted signiﬁcant research interest as well as commercial investments. One of the ways to achieve QoS guarantees on a per-ﬂow basis is to make a priori reservations of buﬀer and bandwidth resources in the network. This approach is used in the Integrated Services (IntServ) architecture , using a reservation setup protocol such as RSVP . The perﬂow management required at the routers in this approach, however, calls into question the scalability of this approach. The Diﬀerentiated Services architecture (DiﬀServ)  is an alternate method that achieves improved scalability by aggregating data packets into a small number of service classes and deﬁning router behaviors expected by packets belonging to each of these classes.
DiﬀServ allows up to 64 diﬀerent service classes, which serve only to deﬁne the treatment a packet will receive in relation to other packets, but without absolute guarantees on performance. In the absence of guarantees, as in IntServ, the role of capacity planning for traﬃc from various classes of service becomes critical to achieving satisfactory service. The user demands for various levels of service can change rapidly due to a variety of reasons, and therefore, capacity planning involving manual participation through service-level agreements (SLAs) between providers is not likely to be very eﬃcient in the use of network resources. Mechanisms for capacity planning and congestion control through pricing, however, can be signiﬁcantly more eﬃcient and also more responsive to changes in, and the demand for, the network resources. This paper explores a practical, ﬂexible and computationally simple user-centric pricing strategy that can achieve QoS provisioning in DiﬀServ networks with multiple priority classes at close to peak eﬃciency, while also maintaining stable transmission rates from end-users.
A network supporting multiple classes of service also requires a diﬀerentiated pricing structure rather than the ﬂat-fee pricing model adopted by virtually all current Internet services. While network tariﬀ structures are often determined by business and marketing arguments rather than costs, we believe it is worthwhile to understand and develop a cost-based pricing structure as a guide for actual pricing.
In economically viable models, the diﬀerence in the charge between diﬀerent service classes would presumably depend on the diﬀerence in performance between the classes, and should take into account the average (long-term) demand for each class. In general, the level of forwarding assurance of an IP packet in DiﬀServ depends on the amount of resources allocated to a class the packet belongs to, the current load of the class, and in case of congestion within the class, the drop precedence of the packet. Also, when multiple services are available at diﬀerent prices, users should be able to demand particular services, signal the network to provision according to the requested quality, and generate accounting and billing records.
The ﬁrst main goal of our work is to develop a pricing scheme in a diﬀerentiated heterogeneous network environment based on the cost of providing diﬀerent levels of quality of service to diﬀerent classes, and on long-term demand for multimedia services. DiﬀServ supports services, which involve a traﬃc contract or service level agreement (SLA) between the user and the network. If the agreement, including price negotiation and resource allocation, is set statically (before transmission), pricing, resource allocation and admission control policies (if any) have to be conservative to be able to meet QoS assurances in the presence of network traﬃc dynamics. Pricing of network services dynamically based on the level of service, usage, and congestion allows a more competitive price to be oﬀered, and allows the network to be used more eﬃciently. Diﬀerentiated and congestionsensitive pricing also provides a natural and equitable incentive for applications to adapt their service contract according to network conditions.
The rest of the chapter is organized as follows. In Section 7.2, the rate allocation scheme for scalable video coding and the proposed pricing strategy for providing QoS guarantees for scalable video streaming traﬃc delivery over a heterogeneous DiﬀServ/WLAN network is presented. In Section 7.3, I demonstrate how video-streaming applications can beneﬁt from the use of the proposed architecture. Finally, Section 7.5 draws the conclusions and discusses directions for further work and improvements.
7.2 Proposed Arrhitecture The proposed architecture integrates the concepts of scalable video streaming, prioritized packetization based on the H.264 data partitioning features and mapping DiﬀServ classes to MAC diﬀerentiation of 802.11e. The proposed architecture is depicted in Figure 7.2. It consists of three key components: (1) Scalable video encoding (Scalable extension of H.264/MPEG-4 AVC), (2) Pricing stategy module, and (3) DiﬀServ/802.11e class mapping mechanism in order to assure the optimal diﬀerentiation and to achieve QoS continuity of scalable video streaming traﬃc delivery over DiﬀServ and 802.11e network domains. Each one of these components is discussed in detail in the following subsections.
7.2.1 Constant Quality Rate Allocation Method To best utilize FGS encoding, a rate allocation algorithm is needed to transfer the rate constraint into the rate assigned to each frame, and at the same time, to maximize the visual quality. There are a number of schemes proposed in the literature . The simplest one is constant bit-rate allocation (CBR).
However, CBR often results in quality ﬂuctuation, hence, signiﬁcantly degrades the overall quality. To solve this problem, variable bit-rate (VBR) allocation is proposed for constant quality reconstruction by allocating rate according to the complexity of each frame . Wang et al.  proposed an optimal rate allocation using an exponential model. In , a constant quality rate allocation is proposed that minimizes the sum of absolute diﬀerences of qualities between adjacent frames under the rate constraint. The solution is computed by solving a set of linear equations. 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 allocation 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.
I propose a constant quality rate allocation algorithm for FGS using a novel composite rate-distortion (R-D) analysis. The rate allocation is formulated, as a constrained minimization of quality ﬂuctuation measured by the dynamic range of all distortions. The minimization is solved by ﬁrst computing a composite R-D curve of all frames in the processing window. Then, for any given rate budget, the constant quality that can be achieved is calculated from the composite R-D curve. Finally, this constant quality is used to allocate the rate for each video frame.