QoE-aware Congestion Control Algorithm for Conversational Services in Wireless Environments

QoE-aware Congestion Control Algorithm for Conversational Services in Wireless Environments

When the bottleneck link is overloaded or channel conditions are bad, the TCP through- put decreases and cannot satisfy the source rate of the multimedia application. This increases, generally, the jitter and the packet loss rate that could impact the user- perceived quality, which is also known as the QoE. Although the QoE is affected by some factors, such as the audio quality, devices, echo, etc., we focus, in this chapter, on improving the QoE through a novel congestion control algorithm. The impact of non-networking factors could be cataloged into a protocol stack to form a conceptual relationship between QoS and QoE (see [124] and [125]). The QoE is measured by MOS values. In a subjective test, the QoE is rated on a scale of 1 (bad) to 5 (excellent) by a significant number of people, and the average of the scores is called a MOS. Note that, the ITU-T Recommendation P.911 [126] provides the reference for carrying out subjective measurement of audiovisual materials. In this chapter, we propose a QoE-aware POMDP-based congestion control algorithm, referred to as MOS-TCP, which exhibits an improved performance when transporting multimedia applications, specifically over a wireless path. Our algorithm is suited for networks containing wireless branches, like the model depicted in Figure 7.1. The goal of the MOS-TCP algorithm is to control the end-to-end congestion in order to maximize the QoE, where packets can be lost due to congestion or randomly due to errors encoun- tered across the wireless path. Unlike the current TCP congestion control protocol that only adapts the congestion window to the network congestion (e.g. based on network congestion signals, such as the packet loss rate in TCP Reno, or the round-trip time in TCP Vegas), the proposed congestion control algorithm adopts a two-level congestion control mechanism. Indeed, it adapts over time the congestion window size according to the source rate and the QoE feedbacks. Moreover, we consider a set of updating policies composed of generic congestion control algorithms with general increase and decrease functions, such as AIMD, IIAD, SQRT, and EIMD. In order to capture dynamics of the network congestion and optimize the QoE, we formulate the congestion control us- ing a POMDP framework. The proposed POMDP framework allows users to evaluate the network congestion variations over time, and determines an optimal threshold-based congestion window updating policy in order to maximize the long-term discounted re- ward. In this chapter, the QoE measured through the multimedia quality (MOS) defines the reward.

POMDP-based adaptation: We formulate the QoE-aware congestion control problem using a POMDP framework. The framework allows senders to optimize the congestion window updating policy that maximizes the long-term expected QoE. Furthermore, users have a partial knowledge about the bottleneck link status. In fact, the number of packets in the bottleneck link queue depends on the congestion windows of all users, which cannot be observed. Therefore, the long term prediction and adaptation of the POMDP framework is essential for optimizing the performances of multimedia applications. This chapter is organized as follows. We introduce the QoE and explain the MOS calculation in Section 7.2. In Section 7.3, we model the QoE-aware congestion control problem that maximizes the performance of multimedia applications. Thereafter, in Section 7.4, we formulate the problem using a POMDP-based framework. We present a low-complexity algorithm to solve the POMDP in Section 7.5. Section 7.6 provides experimental results that validate the proposed congestion control method, and Section 7.7 concludes the chapter.This chapter is organized as follows. We introduce the QoE and explain the MOS calculation in Section 7.2. In Section 7.3, we model the QoE-aware congestion control problem that maximizes the performance of multimedia applications. Thereafter, in Section 7.4, we formulate the problem using a POMDP-based framework. We present a low-complexity algorithm to solve the POMDP in Section 7.5. Section 7.6 provides experimental results that validate the proposed congestion control method, and Section 7.7 concludes the chapter.To overcome the limitation of QoS-based optimization, QoE-based approaches are intro- duced as a more effective way to optimize transmission algorithms and protocols with respect to user satisfaction. QoE metrics are defined as a set of quantitative measures to assess the perceived QoS of end users [127]. Moreover, a new approach, namely QoE- aware networking, is proposed to re-formalize the service optimization problem and to improve the user experience. Because the QoE metrics reflect the end user’s experience, QoE-based approaches may improve the subjective service quality, optimize the use of network resources, and provide services to more users without noticeable degradation of users’ experience. Recently, QoE metrics are used to optimize various types of network services. In cellular systems, authors of [128] used a QoE-based approach to allocate downlink wireless resources among different applications. They defined several QoE models for different types of applications such as file downloading, voice call and videostreaming, and adopt QoE-based utility maximization to improve the user perceived quality. In [129], authors applied QoE metrics to optimize IEEE 802.11 wireless LAN. They used a machine learning approach to generate real-time QoE measurements and used the QoE feedbacks to manage wireless network resources. In [130], authors used QoE metrics for packet scheduling in multi-hop wireless networks. The packet scheduler determines the packet drop pattern that minimizes the degradation of MOS values. In P2P networks [131], scalable video coding and QoE metrics are used to optimize the performance of P2P video streaming systems. In this chapter, we seek to enable QoE- awareness in a more general setting. We integrate the QoE metrics within the TCP protocol. Since TCP is a widely adopted building block in many network services, our approach is applicable to a much wider spectrum of applications.