An SDN-based Virtual Cell Framework for ... - Semantic Scholar

An SDN-based Virtual Cell Framework for Enhancing the QoE in TD-LTE Pico Cells Salvatore Costanzo∗ , Rudraksh Shrivastava†‡ , Dionysis Xenakis∗ , Konstantinos Samdanis† , David Grace ‡ and Lazaros Merakos ∗ ∗ Dept. of Informatics and Telecommunications, University of Athens, Athens, Greece † NEC Europe Ltd, Kurfrsten-Anlage 36, 69115 Heidelberg, Germany ‡ Dept. of Electronics, University of York, York, YO10 5DD, UK {scostanzo, nio, merakos} @ di.uoa.gr, {rudraksh.shrivastava,samdanis} @ neclab.eu, {david.grace} @ york.ac.uk Abstract—In this paper, we propose a Software Defined Network (SDN)-based framework for enhanced Quality of Experience (QoE) in the presence of Time Division-Long Term Evolution (TD-LTE) pico-cell hotspots. The proposed framework enables users to utilize radio-resources from multiple base stations and enhance their QoE, by utilizing the concept of virtual cells and by customizing the TD-LTE frame at the picocell stations. In this way, the proposed framework resolves pseudo-congestion and enables elastic radio-resource management while meeting the individual traffic requirements at the end users. The formation of virtual cells is driven by a QoE-centric approach that monitors whether the Mean-Opinion Score (MOS) of specific applications drops below a prescribed user-specific threshold. Accordingly, the TD-LTE frame is centrally optimized by an SDN Controller so as to meet the individual QoE requirements at the end users. System-level simulations demonstrate that the employment of the proposed framework leads to notable improvements in terms of MOS, end-to-end delay and packet loss rate, as compared to existing approaches with static and adaptive frame reconfiguration in TD-LTE picocells. Keywords—QoE, Virtual Cell, SDN, TD-LTE-A Pico Cells.

I. I NTRODUCTION The increasing utilization of smart-phone devices and tablets have boosted the use of video-centric applications, social media, and cloud-based services, leading to a tremendous increase in mobile data traffic and diverse service patterns with higher uplink demands [1][2]. Such emerging mobile applications and services are typically delay-sensitive and bandwidth intensive, putting pressure to mobile operators towards better utilizing their network resources. Elastic resource management is a key requirement for effectively meeting the individual quality of experience (QoE) demands at the end users. In mobile networks, users perceive different channel conditions and experience different levels of QoE. At the same time, the vast majority of mobile applications are offered to the end users as Over-The-Top (OTT) services that are typically treated in a best-effort manner via the use of the default bearer at the mobile network operator. Even though such an approach was sufficient in the early phase of the LTE system (mainly due to the support of high rates), lately, it has raised certain revenue issues for the mobile operators and a QoE gap for application and service providers that cannot assure the desired service quality at the end users. Software Defined Networking (SDN) architectures and solutions are increasingly integrated in wireless networks to allow network programmability and enable mobile operators to tune the network resources according to the user needs [3]. Besides,

978-1-4799-8958-4/15/$31.00 ©2015 IEEE

OTT applications and service providers can communicate with mobile operators to acquire network information and specify the service requirements [4]. In this direction, novel business models can be supported to allow operators to get into the revenue loop between OTT services and the subscriber [5]. In this paper, we propose an SDN framework that aims at enhancing the QoE perceived by the users running OTT applications. We further focus on the scenario where the LTE network is composed by TD-LTE picocells that can better handle symmetric traffic and effectively match instantaneous traffic demands by adapting the UL/DL frame ratio [6]. In this direction, we consider the presence of an SDN Controller with a global view of the TD-LTE picocell status and the QoE requirements of the end users. Based on the available knowledge, the SDN Controller centrally orchestrates the instantiation of virtual cells and adapts the UL/DL frame ratio in neighboring picocells when needed. The proposed framework aims at resolving the so-called pseudo congestion problem, where a picocell has a surplus of resources in the opposite direction as compared to the desired one. Besides, the proposed framework aims at creating virtual cells in regions with overlapping network coverage in order to enable cell-edge users utilize resources from multiple pico-cells. We envisage that the SDN-enhanced employment of UL/DL frame reconfiguration in conjunction with the instantiation of virtual cells can enhance service quality and improve resource utilization [7]. The rest of the paper is organized as follows. Section II discusses related works and paper contributions. Section III introduces the proposed SDN framework for QoE-aware flexible network management. Section IV includes system-level simulation results on the performance of the proposed framework. Finally, Section V concludes the paper. II. R ELATED W ORK QoE can be monitored by using subjective or objective techniques. As discussed in [8], quality of service (QoS) is the connection between network performance and application requirements, while QoE is the connection between application performance and the user. One of the most common methods for subjective determination of the QoE consists of computing the Mean Opinion Score (MOS) in line with the International Telecommunication Union (ITUT) Rec. P.800. Nonetheless, measuring the MOS of speech quality is rather complicated, given that MOS is averaged over a large number of user’s

QoMEX 2015

opinion. MOS can be measured using objective techniques as well, such as the E-Model and the Perception Evaluation of Speech Quality (PESQ) model [8]. The PESQ model estimates MOS by comparing a reference signal with the received (degraded) real signal. On the other hand, the E-Model can be readily used to estimate MOS in real-time. In this paper, we adopt the E-Model to assess the QoE of individual users in real-time and adapt the UL/DL frame ratio depending on the current user requirements. Aiming to meet the individual QoS requirements at the end users (i.e. in terms of throughput and packet delay), the authors in [9] have proposed a dynamic cell specific UL/DL frame reconfiguration mechanism. Different from this work, the framework proposed in this paper enables customized TD-LTE frame ratio in line with the QoE demands at the end users. Besides, in this work we utilize the SDN paradigm [4] to achieve efficient synchronization of the TDLTE picocells in order to enable an SDN-enhanced formation of virtual cells. In this manner, we allow interaction between the OTT application providers and network operators. In [10] the authors propose a downlink scheduling method for improving the QoE for VoIP traffic in LTE networks. To achieve this, they consider that the scheduling decision is made based on feedback provided by the users. The authors in [11] present a QoE-based framework for network planning and propose an analytical method based on queueing theory to dimension access networks such as LTE. Different from [10] and [11], we propose an SDN-based network management architecture and solution based on the formation of virtual cells and the flexible adaptation of the UL/DL ratio at the TD-LTE picocell. Besides, as shown in the numerical results, the proposed framework improves the QoE of users suffering from service quality degradation without affecting the performances of others. In [12], the authors propose an architecture that uses a central QoE server that collects performance indicators from different network elements and takes actions to improve the QoE of particular users. Decentralized data offloading mechanisms are proposed in [13], to improve end-to-end delay and enhance the MOS required for service continuity. Different from [12] and [13], we adopt a hybrid approach where, even though the users’ QoE is estimated locally at the pico-eNBs, we employ centralized SDN-based resource management to enhance the QoE of specific users by forming virtual cells. In [14] the authors propose a radio resource management (RRM) strategy, which dynamically reallocates resources assigned to users with adverse channel conditions towards other users aiming to improve their MOS. To accomplish this they consider two expulsion criteria: (i) the direct expulsion of unsatisfied users that underutilize radio resources and (ii) the expulsion of besteffort users that can tolerate service delays. Different from this work, in this paper we provide additional resources to users with MOS below a certain level depending on the type of application. To achieve this, we allow cell-edge users to utilize resources from multiple pico eNBs according to the virtual cell concept. Such a strategy provides sufficient resources to users with bad channel conditions while enhancing their individual MOS. In summary, the proposed framework extends the state

of the art by dynamically programming the UL/DL ratio of TD-LTE pico eNBs, while introducing virtual cells considering the requirements of the OTT applications to satisfy QoE of end users. In the remainder of this paper, we concentrate on the network management architecture and mechanisms required to employ on-demand TD-LTE frame re-configuration and virtual cell formation. To achieve this, we start by introducing a QoE estimation algorithm that is considered to be deployed across multiple pico-cells in a distributed manner. The outcome of the estimation is used as feedback to the 3GPP Operations, Administration and Maintenance (OAM) system that assists the SDN Controller to identify the users suffering from QoE degradation. When such degradation is identified, the SDN Controller enforces frame re-configurations in a certain area covered with TD-LTE picocells in order to enhance the QoE of users with degraded performance. Using system-level simulations we compare the performance of the conventional static configuration with the adaptive cell specific UL/DL re-configuration and virtual cells considering VoIP as the main OTT application. Simulation results indicate that a QoE-aware TD-LTE re-configuration enhances user performance corresponding to the degree of re-configuration freedom, i.e. within a certain cell (adaptive cell specific UL/DL re-configuration) or additionally between neighboring cells (virtual cells). Besides, the proposed QoE framework can enable operators to optimize their network revenues in line with the real time QoE requirements of the end users. III. SDN F RAMEWORK F OR Q O E E NHANCMENT This section describes the main features of the proposed SDN-based framework for enhanced QoE in the presence of TD-LTE picocells. In Section III.A, we briefly introduce the concept of virtual cells and discuss how it can be used to enhance the QoE in the presence of TD-LTE picocells. In Section III.B, we present an SDN-based reference architecture for flexible network management that is accompanied with the logic signaling flow for enhanced QoE management on a per user and per application type basis. Accordingly, in the same section, we discuss an algorithm for assessing the QoE of the users locally at the pico eNBs and triggering the instantiation of the virtual cell function which is a logical entity residing in the SDN Controller. The details of the QoE-Assessment algorithm are described in Section III.C. A. Virtual Cell Concept for Enhancing QoE The concept of virtual cell has been introduced in [7]. Virtual cells provide efficient and flexible resource management for TD-LTE networks. In particular, virtual cells enhance the state of the art cell-specific adaptive reconfiguration [6] at the cells, by adopting a particular UL/DL ratio reflecting their local traffic conditions and enabling a certain set of users to utilize subframes from multiple eNBs. In this way, virtual cells allow customized adaptation of virtual frames as shown in Fig. 1, by integrating specific subframes from multiple eNBs to

match the UL and DL transmission needs of the user demands. As eNBs involved in a virtual cell may adopt a different TD-LTE frame configuration, the resulting virtual frame may consist of a customized selection of subframes that appears as a different logical eNB to the residing users.

issues related to QoE degradation and pseudo congestion. Such reconfiguration instructions are communicated from the SDN-controller to the pico-eNBs via the Data-Controller Plane Interface (D-CPI). The D-CPI also monitors QoE related KPIs e.g MOS scores of different applications, in order to feed the SDN controller with performance information related with virtual cells directly.

Fig. 1: Virtual Frame Configuration In certain traffic conditions, cell-specific adaptive reconfiguration is unable to meet UL-DL traffic requirements. In such situations, the proposed framework applies the virtual cell concept to further enhance the QoE of certain user flows that suffer from QoE degradation due to varying traffic conditions. This is done by providing additional resources via virtual cell formation with the neighboring eNBs.

Fig. 2: SDN Network Management Architecture Fig. 3 shows the logical signaling flow required to support the proposed two-phase QoE-enhancement approach. As a first step, each user sends a measurement report to the serving pico eNB periodically.

B. QoE Aware Flexible Network Management Fig.2 illustrates the proposed SDN-based flexible network management architecture for enhanced QoE. In this scenario, the virtual cells are created on demand for specific user applications between adjacent pico eNBs. Each pico eNB has a logical function for QoE assessment that works in cooperation with TD-LTE cell specific adaptive re-configuration function. Virtual cells are created by the SDN-based virtual cell function as shown in Fig.2. According to this architecture, OTT providers specify their service requirements and configuration needs to the SDN controller via the Application Controller Plane Interfaces (A-CPIs). This provides a degree of freedom and flexibility to network operators to program their network in real-time in order to satisfy the users’ QoE requirements in an efficient fashion. The SDN controller attains a global network view by interacting with the 3GPP OAM network management plane [15]. The 3GPP OAM subsystem assists the SDN Controller to acquire on-demand (or periodically) information on the current state of the RAN. The RAN state includes information regarding the UL/DL traffic load measurements, the interference conditions and performance measurement KPIs e.g. throughput, delay, handover failures etc. This information can be then be forwarded to the SDN controller and used by the virtual cell function to suggest reconfiguration actions to the pico-eNBs in order to resolve

Fig. 3: Signaling Flow Mechanism Logic This report contains information regarding the end-to-end delay and the packet loss rate (PLR) of each application flow running at the users’ device. The PLR and the endto-end delay measurements are subsequently used as inputs for the QoE-Assessment algorithm that runs at the pico eNB. If the performance of a particular application drops below a predefined limit, the QoE-Assessment algorithm triggers the SDN controller to create a virtual cell and indicate appropriate actions to resolve the problem of QoE degradation for specific applications or users. The triggering conditions are described in more details in section III.C. Once the SDN Controller

receives such a request from the pico eNB, it initiates the virtual cell configuration function. In more detail, it collects information about the status of neighboring cells (e.g. traffic load, SINR, users’ location) and identifies suitable neighbor cells for creating virtual cells in proximity with the tagged pico eNB. Accordingly, the SDN Controller reports back to the serving pico eNB the respective set of neighbor cell identities (IDs) to create virtual cell. In addition it helps in identifying the list of users to be served via the virtual cell configuration, allocating the desired resources without causing any negative impact on the performance of other users in the neighbor cell. C. QoE Assessment Algorithm The parameters and variables used in the proposed QoEAssessment algorithm are listed in Table I. A pseudo-code version of the proposed QoE-Assessment algorithm is presented in Algorithm 1. The QoE-Assessment algorithm runs in all the pico eNBs in a distributed fashion. In the following, we consider a scenario where the users may host different applications on a single device. Each pico eNB calculates the MOS of each application flow running at the served users (step 4). The MOS is calculated according to the E-Model by taking into account the average delay and PLR of each flow over a specific time interval. TABLE I: List of Parameters and Variables Notation C c, (c ∈ C) u, (u ∈ U ) f , (f ∈ F ) M OST hr (f ) M OSdist (f ) G1mos , G1mos ∈ F G2mos , G2mos ∈ F γu,c γT hresh UlowSIN R , UlowSIN R ∈ U PG0 PG1 PG2 PG0out PG1out PG2out

Description Set of cells Index for a cell Total number of users in a cell Set of application flows of a user u Application specific MOS Threshold for the flow f Distance of M OS(f ) from the M OST hresh (f ) Set of flows with M OSdist (f ) ∈ [0, 1.5] Set of flows with M OSdist (f ) ∈ [1.5, 2.5] SINR of user u with cell c SINR Threshold Set of users with γ< γT hresh Probability of a flow f to be in group G0mos Probability of a flow f to be in group G1mos Probability of a flow f to be in group G2mos Outage Probability for group G0mos Outage Probability for group G1mos Outage Probability for group G2mos

In the next step, the proposed algorithm calculates the MOS distance (M OSdist ) between the MOS of a specific application flow and the application specific MOS threshold (M OST hr ) that we consider to be fixed and known. For a flow f, the M OSdist is defined as the difference between the MOS of flow f and the application specific M OST hr . Note that a different MOS threshold can be used for each application (VoIP, Video, Best effort etc.) running at the end devices. For example, the MOS threshold for VoIP flows can

be 3.5, whereas the respective one for video applications can be up to 4.0. Algorithm 1 :QoE Assessment Function 1: for ∀ c ∈ C do 2: for u=1:U do 3: for f =1:F do 4: Calculate M OS(f ) according to the E-Model 5: M OSdist (f )=CalcDistFunc(M OS(f ),M OST hr (f )) 6: if M OSdist (f ) ≤ 0 then 7: Add f to G0mos 8: end if 9: if 0