expected to grow at a compound annual growth rate of 66% from 2012 to 2017. The traffic demand is generally uneven in both temporal and spatial domains.
User-Based Integrated Offloading Approach for 3GPP LTE-WLAN Network Shashi Ranjan∗ , Nadeem Akhtar† , Mahima Mehta∗ , Abhay Karandikar∗ ∗ Indian
Institute of Technology, Bombay, Mumbai - 400076, India {shashisvk, mahima, karandi}@ee.iitb.ac.in † Center of Excellence in Wireless Technology, Chennai - 600113, India {nadeem}@cewit.org.in Abstract—The increasing use of smartphones and contentrich and data-hungry applications has resulted in significant growth in demand for mobile data services in recent years. To meet this requirement, one of the widely-used approaches is to offload cellular data traffic to Wireless LAN as it operates in unlicensed bands. Offloading decisions are typically based on signal quality. In this paper, we propose an integrated offloading approach which takes both signal quality and network load into consideration. We focus on solutions that do not require significant changes to current network implementations and standards. We investigate the performance of this approach for different scenarios via extensive simulations and observe improved system and user performance in terms of throughput. The evaluation adheres to the methodology laid down by the global standards body 3GPP for such comparisons.
I. I NTRODUCTION The rapid and widespread proliferation of smartphones, smart-devices (e.g. energy meters) and growing use of dataintensive services such as video streaming, the data rate requirement has grown at an unprecedented rate in the recent years. According to Cisco forecasts [1], mobile data traffic is expected to grow at a compound annual growth rate of 66% from 2012 to 2017. The traffic demand is generally uneven in both temporal and spatial domains. This results in the creation of traffic hotspots which, in turn, makes the task of meeting data rate requirement more challenging. As the cellular networks are already operating at almost peak capacity after implementing the possible physical layer enhancements, alternative approaches are being investigated for further capacity improvement. One such option is to deploy heterogeneous network which involves the co-existence of the existing macro base station (BS) based cellular network with low-power small cells that may use the same or different Radio Access Technology (RAT). Intelligent offload of data traffic to small cells is being investigated as a capacity improvement solution [2]. In this paper, we focus on a heterogeneous network comprising of Long Term Evolution (LTE) macrocellular network with an overlay of IEEE 802.11 Wireless LAN (WLAN) access points. WLAN is the most promising candidate for this kind of solution because of it is cost-effective and uses unlicensed spectrum. Since 70% of the mobile data is generated indoors [4], WLAN is suitable for mobile data offloading leveraging c 2014 IEEE 978-1-4799-2361-8/14/$31.00
the fact that WLAN hotspots can offer high data rate in a small coverage area with limited mobility. Furthermore, most of the smartphones and tablets are WLAN-enabled and thus, capable of using offloading feature. In general, any WLAN-enabled device, connected to the cellular network, can be offloaded. However, the challenges is to select an appropriate WLAN such that the Quality of Service (QoS) requirements of the user being offloaded is not compromised, while ensuring that offloading gain is maximised. The connection manager in smartphones often incorrectly decides to connect to an undesirable WLAN (from offloading perspective). An inappropriate WLAN choice may lead to increased load in the network which may deteriorate the overall network performance. In other cases, offloading opportunities are missed altogether, thereby causing underutilisation of available resources. Furthermore, connection managers are proprietary implementations. Hence, there is no consistent behaviour across operating systems and devices, when it comes to decisions regarding WLAN selection. Therefore, the user experience becomes very much dependent on the device. In 3GPP Release 12, a study is in progress on WLAN interworking [3] for offloading. There are two approaches being considered: User Equipment (UE)-controlled offload with some network assistance and network-controlled offload, based on UE measurement reports. While both the schemes have their pros and cons, here we consider the UE-based approach since it requires minimal changes in the network. In UE-based approach, decision variables like Signalto-Interference-and-Noise-Ratio (SINR), network load, etc., assist in ensuring that all the available radio resources are optimally utilized to improve the overall network performance. The thresholds for decision variables can be set by the UE and load information can be obtained from beacons transmitted by the WLAN access points. The schemes discussed in this paper take into account the constraints posed by the current network implementations and follow the broad contours of the solutions being discussed in 3GPP. In addition, the simulation model used for performance evaluation closely follows the methodology used by 3GPP for comparative evaluation of various solutions. We have customised the ns − 3 simulator to ensure this. In this paper, we explore the extent to which the cellu-
lar network and users benefit from offloading in different scenarios and investigate the impact of thresholds for the various decision variables. Some literature is available on using different approaches to make offload decision. In [7], the offloading mechanism is investigated solely based on SINR as the decision variable, where the WLAN network is selected by a UE based on only the first received beacon frame. This strategy works well in a non-fading environment. However, in case of multi-path fading, it may lead to inappropriate offload decision, resulting in either performance degradation or frequent handovers/ping-pongs. In this paper, we implement a modified SINR-based offload scheme where we consider the average SINR of beacon frames received over a time frame. In [8], cost function based approach is suggested but it is only applicable for network-based offloading and introduces significant complexity. In our work, the offloading decision is taken by UE with minimal assistance from the network. Our contributions in this paper are: 1) WLAN offload schemes that are feasible within the constraints posed by 3GPP Release 12 study on WLAN interworking, 2) Implementation of offload schemes in ns − 3 simulator and 3) Performance evaluation as per 3GPP methodology. The proposed algorithm has a potential of being considered in on-going discussion in Release 12/13 of 3GPP. The paper is organized as follows. Section II discusses the fundamentals of LTE-WLAN interworking. Section III presents the network selection schemes and the methodology for performance evaluation. Section IV provides the details of simulation model and the performance metrics. We also provide results and discuss the inferences there and conclude the paper in Section V. II. LTE-WLAN I NTERWORKING Cellular networks and WLANs are different from system architecture perspective, with different protocol stacks, network access schemes, QoS and mobility mechanisms etc. Hence, direct interworking between the two is not simple, even though both are IP-based networks. Therefore, solutions have to be specifically designed to enable interworking between the two. I-WLAN [5] is the first architecture defined for the integration of WLAN networks with cellular architecture of 3rd Generation Partnership Project (3GPP). This architecture describes the interfaces between the networks, data and control paths, and the protocols for the access and authentication. I-WLAN is only useful for untrusted WLAN networks, a type of WLAN not owned by the cellular service provider e.g. Corporate WLAN, and very complex in terms of implementation (like dual authentication for both IPsec and WLAN link-layer) that usually results in increased handover delays, reduced battery life and degraded performance. After the introduction of Evolved Packet Core (EPC) in Release 8 [6] which defines interworking functionality between 3GPP and non-3GPP access technologies, new options for mobility through the use of inter-technology handover over several access network technologies became possible. 3GPP has also
defined trusted WLAN network that makes it possible to have seamless handover between the two networks. Interworking only addresses the protocol-level differences but does not take into account performance aspects. This is particularly important in deployments where the same operator owns both the LTE and WLAN networks. In such cases, closer coordination between the two networks can help make judicious decisions on handover to WLAN, such that user’s experience is satisfactory and network-level Key Performance Indicators (KPIs) are maintained at the desired level. The Evolved Packet System (EPS) addressed this issue to some extent by providing support for diverse non-3GPP accesses that possess different characteristics in terms of security, bandwidth etc, a new entity called Access Network Discovery and Selection Function (ANDSF) was introduced to provide information about non-3GPP access networks such as WLAN, to the UE [6]. The information, based on the operator configuration, can be [9]: 1) UE Location: UE may send its current location to the ANDSF server, based on one of these: geographical coordinates, macro cell-ID or a WLAN location (SSID or BSSID). 2) Access network discovery information: Discovery information sent by ANDSF server allows UEs to map its current location to a list of alternate access networks that may also be available. 3) Inter-System Mobility Policy (ISMP) and Inter-System Routing Policy (ISRP): These are set of operator defined rules for a UE, a number of prioritized rules that control which network should be used based on its location and/or time, that are capable of routing IP traffic over single or multiple radio access interfaces respectively. The ANDSF is accessible to UEs either through any 3GPP or non-3GPP access technologies that are interconnected through the EPC [6]. Thus, ANDSF serves as a dynamic database, controlled by the operators. This server can be queried by mobile users to discover neighbouring WLANs. As LTE connection is always on for the offloading scenario under consideration, data flows are switched only after connection with WAP is established. Therefore, connection set-up delay can be ignored. In the next section, we illustrate the existing and our proposed WLAN offloading schemes which have been implemented in this paper. III. WLAN O FFLOADING S CHEMES We consider a scenario where all the users, by default, are connected to the LTE network and exactly one application is established already. The purpose of the offloading algorithms is to determine the set of users that can be detached from the LTE network and associated with a suitable WAP. Generally, this is based on a set of decision variables such as SINR, network load etc. We describe a set of offloading schemes considering different association criteria.
A. ANDSF based Approach The ANDSF server may provide a list of available WLANs to UE based on its subscription, location, time of the day etc. The ANDSF Management Object (MO) sent to the UE includes validity areas, applicable time duration, and availability of access networks belonging to different RATs, including WLAN and RAT-specific information such as WLAN SSID. We consider two ANDSF model approaches, based on LTE Cell-ID and UE position, respectively. In both approaches, the ANDSF server sends a list of WAPs to UE, on request. In the Cell-ID based method, the list includes WAPs in the same geographical area corresponding to the coverage of the LTE macro cell to which the UE is connected. In locationbased method, ANDSF provides a list of WAPs within a range characterized by a distance threshold Dth with respect to the UE position. The threshold is predefined by the network operator. B. SINR based Approach The information provided by ANDSF can, at best, be considered semi-static. In addition, it only provides information about the presence of one of more WAPs in a given geographical area. However, there is no provision for the UE to decide which of the available WAPs to connect, except for the priority assigned by ANDSF. Therefore, UE needs additional criteria to make this decision, e.g. WLAN link quality. A well-known measure of radio link quality is the ratio of received signal power to the sum of noise and interference power,i.e., SINR. It can be used to determine which of the available WAPs are suitable for offloading. In other words, the LTE-connected UE measures the SINR on the WLAN link and compares it with the pre-defined threshold. If the SINR exceeds this threshold, then UE connects to the WAP, else stays connected to the LTE network. The algorithm may formally be described as: ( W APi , if SIN Ri ≥ SIN Rth Access = LT E, otherwise In case, multiple WAPs meet the SINR criteria, then the priority defined by ANDSF can be used to select a WAP. If no such rule is provided, then the UE selects that W APi from the set of available WAPs which is the first to satisfy the above condition for offload. We use the Received Signal Strength (RSS) of beacon transmissions to determine the SINR. In order to avoid fading related temporal variations in RSS, a number of successive beacons, Nbeacon , are measured to calculate average RSS. Any UE whose SINR with respect to a certain WAP meets the criteria is handed over to that WAP. C. Load based Approach SINR provides a good indication of the link quality but does not guarantee that the network has enough resources to admit a new user. In cellular networks, admission control can be performed in the Radio Access Network (RAN) but
WLAN is opportunistic in this respect. Hence, even if a user has good SINR, it may not get good throughput if the network is overloaded. Therefore, network load also needs to be considered when deciding a suitable WAP for offloading a user. In this approach, we evaluate the target WAP’s Medium Access Control (MAC)-layer throughput over a given time interval ∆tHO . This is compared with a reference load, which in itself is a function of the peak WAP throughput. For example, if maximum throughput is 20 Mbps and throughput evaluated during ∆tHO period is 10 Mbps, then system is considered to be 50% loaded. If the observed throughput is less than the pre-defined load threshold, Loadth , the user is offloaded to WLAN. Otherwise, it continues to be attached with the LTE network. D. Proposed Integrated Approach Each of the decision variables discussed above is important, but a decision taken considering only one of them may not yield the desirable user and network performance. Hence, we define an integrated algorithm, which takes into account both these variables. The flow-chart shown in Figure 1 illustrates our integrated algorithm. start
Yes
Initialize model
WLAN loaded ? No
t=0 Check for WLANs
t = t + ∆tHO
Check again after ∆tHO
No
SINR≥ SIN Rth ?
Yes WLAN available ?
No
User offloaded to WLAN
Yes
stop
Fig. 1.
Flowchart of Integrated Algorithm
We first select the set of users which satisfy the SINR criteria and then offload them, one-by-one, to the WAP, till the load criteria is satisfied. Users are offloaded over a period of time, in staggered manner to avoid a sudden traffic surge in the WAP. IV. P ERFORMANCE E VALUATION We compare the performance of the various schemes described in Section III by simulations. The objective is to analyse the offloading gain achieved by handing over LTE users to the WLAN under different traffic load conditions. A. Simulation Model We model an LTE cellular network comprising a regular hexagonal grid of seven cell sites, with three sectors per site. The inter-site distance is 500 meters, which corresponds to
the well-known Urban Macro model. Macro base stations are deployed at each cell-site, with three directional antennas, each covering one sector. Users are dropped randomly, with uniform distribution, in each sector. Furthermore, there is a hotspot region in each sector, with higher user density than in rest of the sector. The hotspot location is chosen randomly, with a minimum separation of 75m from the macro base station. The hotspots are assumed to be circular areas with radius of 40 meters. At the centre of each hotspot, a WAP is deployed. WLAN is also assumed to be deployed by the cellular operator, thereby making it a trusted network from the interworking point of view. Figure 2 illustrates the network model, with hotspots located in the centre cell.
users in the simulation model, the algorithms described in this paper are applicable to mobile users also. Parameters for LTE network are based on 3GPP model and adopted from [11] and [12]. These are listed in Table I. The WLAN network consists of IEEE 802.11g APs, placed at the centre of each hotspot region. As WAPs are considered to be operator-deployed, their locations are known a priori. Table II lists the relevant model parameters for WLAN. TABLE II WLAN N ETWORK M ODEL Parameter Tx power for WAP Noise figure for WAP Antenna-height for WAP Antenna parameters Range of WAP Path loss Multi-path fading
Statistical Characterization 23 dBm 4 dB 2.5 m Isotropic Antenna 60 m 140.3 + 36.7log(R), R in Kms Rayleigh fading
The procedure for dropping users is described as Algorithm 1. Each WAP is operating at different non-overlapping channel Procedure
Fig. 2.
User and Hotspot distribution within the cellular area
TABLE I LTE N ETWORK M ODEL Parameter Tx power for BS and MS Noise figure for BS and MS Antenna-height for BS and MS Antenna parameters for BS
Antenna parameters for MS HARQ Transmissions Min. distance b/w eNB and UE Path loss Shadowing Multi-path fading
Statistical Characterization 46 dBm and 23 dBm 5 dB and 9 dB 32 m and 1.5 m Parabolic Antenna, BW = 70◦ , Max. Attenuation = 25 dB Isotropic Antenna 4 for both UL and DL 35 m 128.1 + 37.6log(R), R in Kms Log-normal variable with mean = 8 dB 3GPP TS 36. 104 Annex B. 2 [13]
We assume that, irrespective of its location, a UE can always connect to the LTE network and all UEs are capable of accessing WLAN, if available. However, association with a WAP depends on UE’s distance from the WAP, link quality, network load etc. All users are stationary and therefore, the total number of users in each sector remains same throughout the simulation period. We focus on users in the centre cell while the remaining cells contribute only to interference to the LTE users in the centre cell. We drop only a single user per sector in the neighbouring cells, which is always connected to LTE network and uses all available resources, thus creating a fully loaded cell. Note that even though we consider stationary
1) Fix the total number of users, Nusers , dropped within the central macro geographical area 2) Randomly and uniformly drop the configured number of hotspots, N , within each sector 3) Fix the probability of positioning users inside hotspots as Ph 4) Calculate Nusers lpn = bPh .Nusers /(3 ∗ N )c 5) Randomly and uniformly drop Nusers lpn users within each hotspot 6) Randomly and uniformly drop the remaining users to macro geographical area, with (Nusers − Nusers lpn ∗ N )/3 within each sector 7) Drop one user in each sector of the interferer cells End Procedure
Algorithm 1: Procedure of UE placement for Hotspots (channel 1, 6 and 11), so as to ensure that there is no interference among the WLAN networks. It also means that the SINR-based schemes actually use SNR as the criteria, since interference is absent. We consider Constant Bit Rate (CBR) traffic, with one downlink flow per user, at a fixed rate of 2Mbps and packet size of 576 bytes. All flows are initiated at the same time. The generation and parameterization of CBR traffic is implemented by ON/OFF application. We consider proportional fair scheduling at the MAC layer of macro BSs. The simulation model described above is implemented in ns − 3, a discrete event simulator particularly useful for network based simulations. The simulator required significant modification and addition of new modules in order to validate the 3GPP model as described in [11]. B. Evaluation Procedure We consider two scenarios, with 10 and 15 users per sector, corresponding to medium and high load, respectively. At
90
15 users/sector
10 users/sector
2
80
15 users/sector
10 users/sector
Throughput [Mbps]
Throughput [Mbps]
70 60 50 40 30
1.5
1
0.5
20 10 0
No Offload
Location−Based SNR−Based
Load−Based
0
Integrated
No Offload
(a) Average System Throughput
Load−Based
Integrated
(b) Average User Throughput Throughput Performance
the beginning of a simulation, all users are attached to the LTE network and application flows are initiated. After a few seconds, the offloading process is started. After the desired set of users have been offloaded to WLAN, we start observing the throughput performance and data is collected until the end of the simulation. Users once offloaded stay connected with the WAP until the end. For comparison, we also simulate a reference case where all users stay connected with the LTE network for the entire duration. For each scenario, the offloading approaches described in Section III are applied and the throughput observed for each case. The results are obtained by extensive simulations with 10 trials for each data point. The MAC level throughput is determined with a 95% confidence interval. For our simulation model, values of the variables used in Offloading schemes in section III and Algorithm 1 are listed in Table III. TABLE III S IMULATION M ODEL PARAMETERS Value [30, 45] 1 3 2/3 120 m 40 dB 4 102400 µs [10] 80% 0.2 sec
C. Simulation Results Figure 3a illustrates the system throughput for both scenarios, with different offloading schemes. The graphs also show the system throughput for the reference case. When WLAN is deployed, system throughput is aggregated over both networks. The results clearly indicate the beneficial impact of offloading in both scenarios but gains are much higher for the high load case (15 users/sector). When there are only 10 users/sector, the LTE network load is moderate to begin with and hence,
after offloading, both LTE and WLAN become underloaded, leading to very little throughput gain. In contrast, when offered load is high, the increase in system throughput is significant because, after offloading, both LTE and WLAN have sufficient number of users. Hence, all the available resources are fully utilised. The system throughput results also demonstrate that taking load into account is beneficial for offloading, compared to considering only SNR. However, when the two criteria are jointly used, no additional gain is seen. The SNR threshold is low and hence, does not provide any additional selectivity in the integrated scheme. As a result, the same number of users get offloaded as in the load-based scheme (see Figure 4). 100 Number of users offloaded [%]
Fig. 3.
Parameter Total number of users, Nusers Number of hotspots per sector, N Number of WAPs Probability of users’ positioning in hotspots, Ph Distance Threshold, Dth SNR Threshold, SN Rth Number of beacon intervals, Nbeacon Beacon interval Load Threshold, Loadth Handover interval, tOH
Location−Based SNR−Based
15 users/sector
10 users/sector
80
60
40
20
0
Location−Based
Fig. 4.
SNR−Based
Load−Based
Integrated
Number of Users Offloaded
Figure 3b sheds further light on the performance of different schemes under consideration. In particular, we can observe how, in the high load scenario, the average user throughput increases when offloading is applied. The interesting point to note here is that user throughput for the high load scenario becomes nearly the same as the throughput in medium load case for the threshold-based schemes, even though the number of users served is higher. This is due to the load-balancing effect where the available resources in both networks are effectively utilised to serve all the users. There is little change in user throughput in the 10 users/sector scenario because both the networks are under-loaded after offloading. Figure 4 illustrates the relative effect of all the schemes in terms of the number of users offloaded. It is clear that in
1
1
No Offload
0.8
Location−Based
0.9
SNR−Based
0.7
SNR−Based
0.7
Load−Based Integrated
0.6
CDF
Integrated 0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0 0
Location−Based
0.8
Load−Based
0.6
CDF
No Offload
0.9
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0 0
2
User Throughput [Mbps]
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
User Throughput [Mbps]
(a) CDF of User Throughput for Nuser = 45
(b) CDF of User Throughput for Nuser = 30 Fig. 5.
User Throughput CDFs
the 10 users/sector case, the load threshold is not reached at all and hence, the percentage of users that connect to WLAN is nearly the same for all threshold based schemes. In contrast, for the 15 users/sector case, load-based schemes limit the number of users offloaded and hence, result in improved overall performance, as observed from the throughput results in Figure 3. Figures 5a and 5b illustrate the Cumulative Distribution Function (CDF) of user throughput for 15 and 10 users/sector, respectively. A common feature in both graphs is the relatively larger fraction of users receiving throughput above 1Mbps, when offloading is applied. In particular, the load-based and integrated schemes result in significant increase in the number of medium-to-high throughput users. In addition, compared to the baseline case, the peak throughput is higher for all offloading schemes in the high load scenario whereas not much difference is seen in the medium load scenario. In summary, the results provided here indicate that offloading is beneficial in improving user and system throughput, although gains vary depending on the offload criteria and the network load. The integrated algorithm shows marginal improvement compared to SNR and load-based approaches. However, performance can be further improved when more users are admitted in the LTE network after offloading. V. C ONCLUSION In this paper, we have discussed a set of offloading schemes for LTE-WLAN interworking scenarios, to achieve better system and user performance. The schemes use WLAN SINR and load as the decision variables to select an appropriate WAP for offloading users, from a list of WAPs provided by the network operator. We have evaluated the throughput performance achieved when these variables are used independently as well as jointly, taking the LTE-only system as reference. In all cases, we observed varying degrees of improvement when threshold-based offloading is used. However, gains are signicant when offered load is high and WLAN load is used
for admission control. The effect is visible from the distribution of user throughout which shows that most users enjoy relatively higher throughput when threshold-based offloading is used. In this paper, we have considered stationary users and CBR trafc only. Further investigation focuses on studying the impact of user mobility and dynamic load on the user and system throughput performance. We are currently exploring the possibility of contributing to 3GPP Release 12/13 based on the algorithm proposed in the paper. VI. ACKNOWLEDGEMENT This work is supported by the India-UK Advanced Technology of Centre of Excellence in Next Generation Networks (IU-ATC) project and funded by the Department of Science and Technology (DST), Government of India. R EFERENCES [1] Cisco, Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, Feb, 2013. [2] Accuris Networks, The Business Value of Mobile Data Offload, 2010. [3] 3GPP TR 37. 834 V0. 3. 0, Study on WLAN/3GPP Radio Interworking, May, 2013. [4] Ericsson Research, Heterogeneous Networks: Meeting Mobile Broadband Expectations With Maximum Efficiency, Feb, 2012. [5] 3GPP TS 23. 234 V11. 0. 0, 3GPP system to Wireless Local Area Network (WLAN) interworking, Dec, 2012. [6] 3GPP TS 23. 402 V10. 7. 0, Architecture enhancements for non-3GPP accesses, Mar, 2012. [7] Desta Haileselassie Hagos, R¨udiger Kapitza, Study on PerformanceCentric Offload Strategies For LTE Networks, IFIP, WMNC 2013. [8] Sven Wieth¨olter, Marc Emmelmann, Robert Andersson, Adam Wolisz, Performance evaluation of selection schemes for offloading traffic to IEEE 802.11 hotspots, Wireless Networks Symposium, IEEE ICC 2012. [9] 3GPP TS 24. 312 V12. 0. 0, Access Network Discovery and Selection Function (ANDSF) Management Object (MO), Mar, 2013. [10] IEEE Std 802. 11, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, Mar, 2012. [11] 3GPP TR 36. 814 V9. 0. 0, Further advancements for E-UTRA physical layer aspects, Mar, 2010. [12] 3GPP TR 36. 839 V11. 1. 0 , Mobility enhancements in heterogeneous networks, Dec, 2012. [13] 3GPP TS 36. 104 V10. 2. 0 , Base Station (BS) radio transmission and reception , May, 2011.
