On Modeling and Measuring Quality of Experience Performance in IEEE 802.11n Wireless Networks Jaya Rao 1 , Abraham O. Fapojuwo 1 , Salman Naqvi 2 , Isaac Osunkunle 2 , Chad Rumpel 2 and Dave Morley 2 1 Department of Electrical and Computer Engineering, University of Calgary, Alberta, Canada 2 Shaw Communications Inc., Calgary, Alberta, Canada E1mail:{jbrao, fapojuwo}@ucalgary.ca,{Salman.Naqvi, Isaac.Osunkunle, Chad.Rumpel, Dave.Morley}@sjrb.ca Abstract – This paper addresses the problem of characterizing the quality of experience (QoE) of users accessing different types of applications in 802.11n wireless local area network (WLAN) deployment environments. Using the generalized logistic function with fractional powers, an analytically tractable QoE model that can be flexibly applied to different application types (real time and non, real time) is formulated as a function of several quality of service (QoS) parameters. An 802.11n WLAN testbed is then used to validate the accuracy of the QoE model via experimental QoE measurements. The paper also investigates the impact of different network operational parameters including the number of users supported, percentage of real time traffic streamed and level of co,channel interference on both the QoS and QoE performance. Next, an optimization problem is solved to maximize the overall QoE achievable subject to the QoS constraints of the applications in use. Overall, the developed QoE model can be used as an effective tool for enhancing the Access Point (AP) capacity and the QoE performance by optimally adapting the network operational parameters. Keywords, Quality of Experience, Quality of Service, 802.11n, WLAN I. INTRODUCTION Recent widespread deployment of wireless local area networks (WLANs) in private and public areas has generated increasing level of interest in both the industry and academia on conducting quality of experience (QoE)1centric performance evaluations [1]. From the perspective of the network operator, assessing the QoE for services offered over infrastructure1based IEEE 802.11n (current dominant standard) WLAN is essential to determine the contribution of their WLAN infrastructure to the overall level of user satisfaction. In this context, QoE bridges the gap between the performance achieved at the application layer in regards to user experience and the network layer in terms of QoS parameters (e.g. latency, packet loss rate, goodput) [2][3]. While it is imperative that the QoE achieved satisfies the users’ expectations, it becomes highly formidable to assess and optimize the QoE in practical 802.11n networks due to the following factors: 1) The applications subscribed by the users have varied intrinsic characteristics resulting in different levels of susceptibility to the QoS parameters; 2) In real life networks, multiple applications with application distribution profile that can vary both spatially and temporally are supported simultaneously. As such, it is difficult to account for the impact of changing interference and load conditions on the QoE performance. For this reason, there needs to be an analytically flexible QoS1to1QoE mapping model generic to multiple application types for enabling swift assessment of how the operational changes at the network layer affect the overall QoE achievable. For the Voice over Internet Protocol (VoIP) application, the most popular metric used to characterize the QoE of the users is the mean opinion score (MOS) [4], determined using a psycho1acoustic algorithm [5] that maps the QoS parameters to a scale ranging from 0 to 5. The media delivery index (MDI) [6], which consists of the delay factor (DF) and the media loss rate (MLR) metrics, is commonly used for measuring QoE of the real time protocol (RTP) video application. A logarithmic model is proposed in [7] to characterize the QoE for HTTP traffic (web browsing application) as a function of data rate and its accuracy is validated using measurement results. For the transmission control protocol (TCP) based video streaming (e.g. YouTube), its QoE performance is evaluated in [8] considering the QoS metrics of buffering time, bandwidth and packet losses. The authors then present a linear model between the QoE and QoS parameters [8]. While these techniques are capable of effectively capturing the users’ degree of delight or annoyance (i.e. QoE) based on the achieved QoS performance, the QoS1to1QoE mapping models currently in use are specific to a single application type and, therefore, not suitable for multiple applications that are used simultaneously [9]. Similar to previous works [4]1[6], this paper uses different metrics to evaluate the QoE achievable for each application type. However, in contrast to [4]1[9], multiple applications are considered to be in use simultaneously, which is more representative of practical WLAN deployments. Using concepts related to sigmoidal behavior from Utility Theory [11], a QoS1to1QoE mapping model applicable to both Real Time (RT) and Non1Real Time (NRT) applications is formulated. The proposed QoS1to1QoE mapping model is then used in an optimization problem to determine the optimal number of users supported per AP that maximizes the overall QoE performance. The contributions of this paper are threefold: First, a logistic function model suitable for multiple application types is used to analytically characterize the interrelationship between QoS (as determined at Layer 3) and QoE (Layer 7 concept). Second, by incorporating a weighted sum QoE model that represents the utility achieved by both the users and the network operator, an optimization problem is formulated to maximize the overall QoE subject to QoS constraints of the applications supported. From the analysis of its structure, the problem is recast and solved as a generalized geometric programming (GGP) problem. Third, using a WLAN testbed supported by core network infrastructure, the QoE performance achieved for different application types are determined through extensive experimental measurements. In addition to validating the accuracy of the QoS1to1QoE mapping models, the experimental results provide valuable insights in terms of how the QoS achieved at the network layer translates to QoE for various application types. While the experiments focus on the 802.11n standard, the analysis performed for QoS and QoE models can be directly extended to other existing and future 802.11 standards (i.e. b, g, ac, etc.) as well. The detailed descriptions of the analysis and measurement results are presented in the following sections of the paper.