Cross-layer quality-based resource reservation for scalable multimedia

Abstract

Resource reservation for Variable Bit Rate (VBR) multimedia streams is a complex task if both the requirements of service providers and of end-users need to be considered. In this paper we present a novel static reservation scheme for VBR multimedia traffic that takes into account the end-user Quality of Experience (QoE). We apply a two-step process for the reservation. In the first step, we map the target user-perceived quality to an upper limit on packet loss. The second step applies a numerical method, that uses a simple video model and a Markov queue model to determine the required bandwidth for the given level of packet loss. For the validation of the video and the system model we use simulation studies. Although our method is static, it is still able to capture the characteristics of a VBR multimedia, which is the main benefit.

Introduction

With the introduction of streaming applications content providers face new challenges. End-users demand constant and high quality in case of pay-services, that may not be satisfied due to bottleneck links on the transport path. This is typical especially in access networks, mostly over wireless channels.

The layered coding approach of multimedia divides the media stream into several layers each providing additional quality refinement to the ones below. Such a coding technique enables the network to adapt the media stream to changed transmission conditions during the session. The properties of the scalable media can be communicated to network elements with cross-layer information forwarding methods [1]. Although local adaptation with such cross-layer information could scale down the stream to meet the altered network conditions, it cannot guarantee any Quality of Experience (QoE). Bandwidth reservation can solve this problem, however, it raises the main question: how much bandwidth to reserve.

In this paper we provide a method to optimize resource reservation if the quality level required by the end-user is known. The proposed solution utilizes network bandwidth in an efficient way and takes into account the user-perceived quality as well, not only network characteristics.

We assume, that the quality curve of the given media is available at the content provider. For layered media, the quality can be measured as the function of the number of received layers. This means that the less the number of layers are received, the less the data are available for the decoder. That is, the more the data loss can be measured. The quality function can be derived from the so-called rate distortion curve [4], [6]. The higher amount of data is available for the decoder, the lower the distortion is.

We use Peak Signal to Noise Ratio (PSNR) to measure the user-perceived quality. Although PSNR is not the most sophisticated metric to measure video quality and there exist better metrics like Structural Similarity Index (SSIM) and Video Quality Model (VQM), PSNR is still widely used because of its simplicity, low complexity and robustness [5].

The system model we use is a network scenario, in which a scalable stream traverses a bottleneck link. Before the bottleneck link, a queue is installed in the Adaptation Point (AP) that performs the local adaptation based on the forwarded cross-layer information. That is, in our model we assume that application-layer information is available at the queue. The bottleneck link(s) is assumed to be anywhere inside the network of a service provider. The aim of the service provider is to guarantee a given quality level for the media, with the reservation of the lowest possible amount of resources. If the multimedia stream is buffered with a feasible delay constraint of real-time services, and the appropriate amount of bandwidth is reserved for the session, the required user-perceived quality can be guaranteed.

Application level information about the video stream (such as priorities of frames) as well, as the quality level desired by the end-user are available at the Adaptation Point. The system architecture must support the signaling of this information to the AP. The AP solves the bandwidth reservation. However, our paper focuses only on the method to determine the amount of bandwidth that is needed to fulfill the quality requirements.

The main contribution of this paper is an algorithm to compute bandwidth requirements for Variable Bit Rate (VBR) multimedia services. The calculation is based on the mapping of a QoE descriptor, the perceived quality, to a network level QoS descriptor, the expected amount of loss. With the quality curve of the video and with a model of the system we estimate the required bandwidth of the stream for a given required quality level. The reservation we propose is a two-step process. In the first step the method determines the amount of loss at which the perceived quality still conform to the end-user’s requirements. Second, the required bandwidth is computed based on the maximum allowable loss. For this latter we developed an analytical model of the queue.

The queuing scheme is modeled with a Discrete Time Markov Chain (DTMC). The arrival process for the queue is a video model based on the state-of-the-art scalable codec H.264/SVC. The video model presented in this paper is not developed to provide exact description of the original video but rather to enable efficient bandwidth reservation.

The rest of the paper is organized as follows. First the applied buffer management is introduced. In Section 4 the proposed video model is described and its validation for the investigated scalable video codec H.264/SVC [7]. Section 5 presents the model of the buffer management, while at last we show how the system model can be used for resource reservation.