Proximity-Based Data Offloading via Network Assisted Device-to-Device Communications †
Alexander Pyattaev† , Kerstin Johnsson , Sergey Andreev† , and Yevgeni Koucheryavy†
Tampere University of Technology, Tampere, Finland; Intel Corporation, Santa Clara, CA, USA E-mails:
[email protected],
[email protected],
[email protected], and
[email protected]
Abstract—Analysts predict explosive growth in traffic demand on mobile broadband systems over the coming years due to the popularity of streaming video, gaming, and other social media services. While 4G wireless technologies are making a significant effort to keep up with this demand, the expectation is that cellular deployments will fall short of the required capacity unless there is a dramatic shift towards smaller cells. There is already significant interest in femto- and pico-cell deployments for this reason. However, there is another method of creating small cells that the wireless industry has yet to capitalize on, namely direct connectivity between clients in close proximity. 3GPP is currently working to enable device-to-device (D2D) communications within Release 12 of LTE-Advanced. By comparison, IEEE has already defined a D2D communications protocol, termed WiFi Direct, which is based on the 802.11 standards. WiFi Direct not only serves to offload user data onto direct links, but does so using the unlicensed bands. To benefit users further, WiFi Direct can be enhanced by enabling the LTE network to assist during peer discovery and direct connection establishment. In this paper,1 we discuss the network/client requirements and performance benefits of network-assisted WiFi Direct. We assume that clients are continuously under management by the LTE network, which assists them with service/peer discovery and direct connection establishment. We show that networkassisted WiFi Direct can significantly improve the performance of proximal applications and reduce the power consumed by the clients involved, while also improving capacity of the LTE network. Index Terms—Proximity awareness, data offloading, network assistance, device-to-device communications, beyond 4G.
I. I NTRODUCTION While the latest 4G wireless technologies contain numerous advancements that increase network capacity, it is not clear whether cellular networks are capable of handling the impending growth in data traffic predicted in [1]. It could seem that the only viable solution is to radically reduce cell size to improve spectrum reuse. While this is already happening to some extent, as demonstrated by the increasing numbers of femto- and pico-cells in cellular deployments, the industry will not realize the full potential of this “small cell revolution” until operators take advantage of D2D communication opportunities. This is particularly true given that much of the expected mobile traffic growth comes from peer-to-peer (P2P) services that commonly involve clients in close proximity [2]. 1 Part of this work has been completed when Alexander Pyattaev was on an internship at Intel Corporation.
Depending on client mobility patterns, some services are better suited for proximity-based network offloading than others. For example, if D2D peers are non-stationary, the quality of the link may change dramatically over short periods of time [3], thus making it difficult to guarantee service. In these cases, the best candidates for network offloading are delay-tolerant services, i.e. those that can be queued until the D2D link recovers or the data session can be moved back to the infrastructure network (e.g. video-on-demand or file transfers). However, if both clients are stationary, many other P2P services, such as cooperative streaming and social gaming, can be offloaded onto D2D links with good results. In all cases, in order to justify offloading from the client’s perspective, the D2D link must provide improved throughput, delay, and/or power performance compared to the infrastructure path. While there is an ongoing effort in 3GPP to develop an LTE-based D2D solution for Release 12 [4], [5], in this work we present an alternative technique based on an existing WLAN D2D protocol called WiFi Direct, which implements the IEEE 802.11-2012 standard. Not only is this protocol already standardized, it also does not require the use of expensive licensed bands. The only expected downside is that it consumes more radio and battery resources in the peer discovery process. Fortunately, this can be mitigated with minimal network assistance. In our study, we focus on 3GPP LTE Release 10 [6], but the “beyond 4G” networks are generally targeted as potential deployment platforms. Current WLAN technologies running on the unlicensed bands can be made to cause very little interference to LTE networks. But while this makes WiFi Direct a great choice for the network, this may not always be the case for the client. WiFi Direct lacks a fast and resource efficient way of notifying clients when/if they are in D2D range. Thus, if a client is searching for a particular user who is out of range for a long period of time, it will suffer significant battery drain. This is where the LTE network can be of help. If clients are continuously connected to the LTE network, it knows which cell(s) they are associated with, which tracking area(s) they are in, and their locations within a few meters (if location services are enabled). Therefore, the network can quickly and without significant overhead determine if/when clients are potentially within D2D range and inform them when this is the case. In this work, our goal is to understand how network-assisted WiFi Direct performs relative to LTE (i.e. the direct vs. infrastructure path). From the network perspective, we are interested
978-1-4673-6337-2/13/$31.00 ©2013 IEEE
in system capacity; from the user’s perspective, we care about throughput, medium access time, and power efficiency. Since these questions are difficult to address analytically, we perform extensive system level simulations to mimic the behavior of D2D and infrastructure communications between client source/destination pairs and compare their performance. II. E VALUATION METHODOLOGY This section introduces the network entities and respective mechanisms required to enable network-assisted WiFi Direct. In effect, it describes our evaluation methodology, which can be exploited to accommodate a wide variety of prospective D2D technologies and P2P usage models. A. Network entities In our study, we consider a heterogeneous wireless network composed of multiple communicating entities with diverse capabilities comprising a variety of radio technologies (see Figure 1 for entity diagram). First, there is an underlying 3GPP LTE network represented by E-UTRAN Node B (eNB) base stations. Each eNB is connected to the core network, providing cellular connectivity to all wireless clients associated with it. Client Devices eNodeB Application MAC coordinator LTE links
LTE
Wi-Fi
interface
interface
(MAC+PHY)
(MAC+PHY)
Rogue Devices
Rogue Wi-Fi AP
D2D links
Application Wi-Fi links
Wi-Fi interface (MAC+PHY)
Fig. 1.
Network entity diagram
Each eNB is accessed by a number of multi-radio client devices capable of communicating over LTE and/or WiFi. Each client runs applications that use the device’s MAC coordination function to determine which wireless technology to use. The MAC coordinator can be regarded as a layer 2.5 entity implemented in hardware or middleware, but it can also be implemented in software as a virtual network interface. Depending on the recommendation of the MAC coordinator, a client may direct data flows onto the LTE or WiFi interface. In this study, we also account for interference on the unlicensed bands from devices engaged in regular WLAN communications with neighboring WiFi Access Points (APs). These devices compete for channel resources with multi-radio clients. Since we assume they are not associated with the cellular network, their activity on the unlicensed bands cannot be monitored or managed by the LTE network, hence we refer to them as “rogue” devices.
B. Traffic flows and network loading In our methodology, we assume that N LTE clients are placed uniformly across the deployment area following the recommendations in [7]. All clients have an LTE and a WiFi interface, and they are capable of engaging in LTE and WiFi Direct communications concurrently. The client density is high enough that each client is within D2D range of at least one other client. However, only 50% of clients are data “sources”, i.e. have data to send. Their traffic loads are modeled as full buffers with packets of 1500 bytes each. Instead of modeling content distribution and demand among clients explicitly, we assume that a certain percent, x, of source clients are within D2D range of their P2P “destination” clients. For simplicity, we assume that P2P communication is unidirectional, i.e. there is only one source and one destination client in any given P2P session. However, since destination clients are chosen randomly from within D2D range of source clients, two source clients in close proximity could be randomly given each other as destinations, effectively creating bidirectional P2P communication. Nevertheless, in the analysis, this would still be two separate P2P sessions. Rogue devices also have full buffers with packets of 1500 bytes, but their traffic always travels to the APs they are associated with. To simplify the evaluation methodology, we do not model WiFi AP downlink traffic. Instead, we adjust the number of rogue clients to mimic the desired level of competition on the unlicensed bands. III. E XAMPLE SCENARIO A. Conventional cellular deployment In order to estimate the benefits of network-assisted WiFi Direct, we construct an example scenario based on modern urban conditions. The LTE infrastructure network comprises 19 hexagonal cells of 3 sectors each (see Figure 2). Each eNB supports LTE Release 10 technology, and the distance between neighboring eNBs (inter-site distance) is 200 meters, resulting in a cell radius of approximately 110 meters. A wrap-around technique is used to improve precision of the simulation at the edges of the deployment area [8]. All cells share the same 60 MHz bandwidth, which is split into three pairs of 10 MHz bands for FDD operation. Every cell is divided into three sectors, and each sector is allocated a pair of 10 MHz bands, resulting in a 1x3x3 frequency reuse pattern. 3GPP LTE clients associate with eNBs based on the best downlink SINR, with a handover threshold of 1 dB. For more details on the configuration of the reference LTE network, the interested reader is directed to Table I and relevant standardization documents (e.g. 3GPP TR 36.814-900 and ITU-R M.2135-1). For performance verification purposes, we implement a calibration scenario from 3GPP TR 36.814-900, Table A-2.1, and run the corresponding tests. Our simulation results fall well within the required limits for both cell-center and cell-edge spectral efficiency targets (see Figure 3).
600
we employ reliable results from publications on ad-hoc WLAN deployments. Calibrating against WiFi performance results in [11], we achieve near perfect alignment (see Figure 4), and reasonable coherence with FlashLinq technology.
400
2
10
0
200
400
600 600
Clients eNB 400
200
Fig. 2.
0 Position, m
200
400
600
Spectral efficiency, bps/Hz
Position, m
200
1
10
Our 802.11-2012 Our 802.11g-2003, calibration QualComm FlashLinq [11] QualComm 802.11g-2003 [11]
0
10
Network deployment
0
Fig. 4.
100 90
250
300
WiFi calibration, spectral efficiency in 1x1 km area
70
Parameter
60
Client Tx power limit Mobility model Observation period
50 40 30 3GPP calibration[7] SLS implementation
20 10 0 0
100 150 200 Number of node pairs
TABLE I BASELINE SIMULATION PARAMETERS
80
CDF, %
50
0.1
0.2
0.3
0.4
0.5
bps/Hz
Fig. 3.
Propagation model Shadowing model Medium access Power and rate control Frequency resources Signaling mode RF equipment Antenna configuration
LTE calibration, spectral efficiency
B. Conventional WLAN deployment We assume that all APs and their respective clients (i.e. rogue devices) run the same version of the technology, namely IEEE 802.11-2012 [9]. To mimic realistic deployments, rogue devices are positioned around their respective APs. APs may be located anywhere inside the deployment area, recreating hot-spots similar to those in cafes, transportation hubs, etc. A rogue’s distance to its AP is constrained by the maximum tolerable path loss. APs and rogues do not move during the simulation, thus handover is not considered. Our study assumes that all WiFi connections (AP and D2D) use the same frequency bands and have to yield to any active transmission for which the received power exceeds the designated threshold. For more details on the configuration of WiFi networks the reader is referred to Table I and Atheros driver documentation available online [10]. For calibration purposes,
Propagation model Shadowing model Medium access Power and rate control Frequency resources Signaling mode RF equipment Antenna configuration
Value/Source Core parameters 23 dBm IRP per interface Random direction, 3 km/h speed 10 seconds LTE ITU-R M.2135-1 [12], Tbl. A.2.2-1, A1-3 ITU-R M.2135-1 [12], Sect. 1.3.1.1 Round-Robin scheduling Closed-loop SINR target at 15 dB 10 + 10 MHz FDD in each sector, short CP 2 out of 20 special subframes, 10 ms frame ITU-R M.2135-1 [12], Tbl. 8-4 1x2 (diversity reception at eNB) WiFi Empirical, based on [13] Correlation only, based on [14] CSMA/CA, -76 dBm yielding threshold Open-loop SINR target at 25 dB 20 MHz TDMA Green-field, control rate 18 Mbps, RTS/CTS Noise figure 7 dB, noise floor -95 dBm 1x1 (single antenna)
C. Suggested additional functionality If the network is to improve the speed and efficiency of WiFi Direct discovery, the client must be able to register its WLAN ID and available content with the network and inform the network of which client(s) and/or content it is interested in. Similarly, the network must be able to inform clients when they are in proximity, supply them with each other’s WLAN ID, and potentially indicate what content is available. In order to enable this information exchange, the following additional functionality is needed. • On the LTE network: – Content updates that allow the network to maintain a database of what content is available at each client.
D. Simulation tool summary In the course of this study, we have developed an advanced system-level simulator (SLS) based on the LTE evaluation methodology described in TR 36.814-900 and current 802.11 specifications. This simulator is a flexible tool designed to support diverse deployment strategies, traffic models, channel characteristics, and wireless protocols. It models all of the conventional LTE infrastructure and client deployment choices (hexagonal vs. square cells, environment with or without wraparound, uniform vs. clustered client distribution, etc.). Every client has its own dedicated traffic generator, enabling a variety of data patterns across the deployment. Channels are modeled to incorporate all relevant source, destination, and environment characteristics, and each client is capable of supporting multiple radio interfaces, which actively interact up and down the stack. IV. E XPERIMENTS AND RESULTS For a complete picture of the benefits to network and client from offloading onto WiFi Direct, we analyze the performance of network-assisted WiFi Direct under a variety of interference conditions (i.e. with and without WiFi APs and associated rogue devices). We do not model any particular type of client traffic, but instead consider different client densities in order to observe how network offloading onto WiFi Direct performs under different load conditions. We also vary the percentage of approved WiFi Direct connections (i.e. those that outperform their alternative infrastructure path) from 0% to 30%. Based on current P2P traffic statistics and client behaviors, we consider it unlikely that more than 30% of clients will be within D2D range of their peers, but this could change in the future.
The results for total cell throughput are presented in Figure 5. In these curves, the throughputs from LTE and WiFi Direct data sessions are totaled per cell, based on the source client’s cell association. One can easily see that offloading LTE traffic onto WiFi Direct links results in a significant boost in cell throughput, actually doubling it at the 30% offload level. However, if interfering rogue devices are present, throughput gains are more modest, but they are still nearly 50% at the 30% offload level. 700 D2D, no rogues
650
D2D, with rogue nodes LTE baseline
600 Throughput, Mbps
In the WiFi middleware: – Interfaces and capabilities for supervised operation. • In the device OS kernel: – MAC coordination function. • In the core network: – Content database maintaining information about the content available at each client. – WLAN ID database maintaining the WLAN IDs of all clients registered for network assisted D2D. With this additional functionality, a client can send a request to the network for a specific client, service or content, and the network can check whether that client or content is in close proximity. When D2D peers are in proximity, the network can provide them with each other’s WLAN ID for the purpose of quick, resource efficient WiFi Direct discovery and connection establishment. Since this procedure occurs over LTE control channels, we assume it is instantaneous and error free. We also assume that the client’s MAC coordination function recommends a WiFi Direct connection whenever it is preferable to the infrastructure path. Alternative offloading implementations are of course possible, but they are out of scope of this paper. •
550 500 450 400 350 300 0
5
Fig. 5.
10
15 Ooad, %
20
25
30
Total cell throughput (sum rate WiFi + LTE)
Energy efficiency is typically measured in bits per Joule and is therefore agnostic to the particular technology involved. Since device energy consumption figures are generally vendor specific, we use the power coefficients from Table II, which are not based on any particular implementation. TABLE II N ORMALIZED ENERGY EXPENDITURE Operation mode Offline Idle/circuit power Energy sensing Data reception Data transmission
LTE 0 0.1 N/A 0.5 1.0
WiFi 0 0.1 0.25 0.5 1.0
The energy efficiency curves in Figure 6 clearly indicate that communication over WiFi is significantly more power efficient than that over LTE. This is in large part due to WiFi’s higher data rates. In addition, LTE clients are allocated small frequency chunks across multiple time slots, thus their transceiver circuitry has to stay active for extended periods of time, while the actual data rate is relatively low. By comparison, the WiFi MAC activates the transceiver only when it is actually accessing the channel. Even when WiFi users are forced to delay their channel access due to RTS or CTS messages, they can sleep during those periods of time. Then, when they finally do get access to the channel, they utilize the entire bandwidth. As a result,
5
Mbit/relative unit (see Table II)
V. C ONCLUSIONS
D2D, no rogues
4.5
D2D, with rogue nodes LTE baseline
4 3.5 3 2.5 2 1.5 1 0.5 0 0
5
10
Fig. 6.
15 Offload, %
20
25
30
Energy efficiency
only a handful of WiFi interfaces across the deployment are powered on at any given time, and those are all either transmitting or receiving data. One of the known issues with the IEEE 802.11 MAC is its excessive medium access time in the presence of heavy traffic. However, this understanding is based on legacy IEEE 802.11g2003 [15] behavior. Our study models the latest version of the standard, IEEE 802.11-2012. With this latest version, the MAC transfer times (i.e. the time a packet spends in the MAC layer and below) of WiFi Direct clients in the absence of rogue devices are sometimes shorter than those of LTE (see Figure 7). This is primarily because in LTE data rates are significantly lower. When rogue clients are present, the situation benefits LTE more, yet no considerable degradation can be observed. 100 90
MAC transfer time, ms
80 70 60 50 D2D, no rogues
40
D2D, with rogues LTE baseline
30 20 10 0 0
5
10
Fig. 7.
15 Offload, %
20
Average MAC transfer time
25
30
As this study shows, there is significant potential for both network and client performance improvement from network offloading onto WiFi Direct in urban environments. Since much of the predicted growth in social media traffic will be generated between clients in close proximity, ignoring this network offloading mechanism represents a significant loss in network capacity and user satisfaction. For example, in case of 30% offloading, cell throughput can be nearly doubled, while energy efficiency can be improved by as much as 6 times. Currently, WiFi Direct is known to waste a lot of radio and power resources on client discovery. However, the network assistance proposed in this paper can remedy this by providing clients with information regarding which clients and/or what content is available within D2D range at any given time. WiFi Direct provides the additional bandwidth necessary to accommodate the expected growth of data traffic, and with network assistance it performs as well as or better than LTE in terms of average user data rate, energy efficiency, and MAC transfer times. Although we have implemented just a very basic version of network offloading onto WiFi Direct, it can be clearly seen that there is significant potential behind such technology. ACKNOWLEDGMENTS This work is supported by Intel Corporation, Tampere Graduate School in Information Science and Engineering (TISE), and Internet of Things program of Tivit, funded by Tekes. R EFERENCES [1] CISCO, “Cisco visual networking index: Global mobile data traffic forecast update, 2011–2016,” tech. rep., 2012. [2] 3GPP, TR 22.803 Feasibility Study on Proximity-based Services. 2012. [3] A. Berl and H. D. Meer, “Integration of mobile devices into popular peer-to-peer networks,,” in Proc. of 5th Euro-NGI conference on Next Generation Internet networks, 2009. [4] 3GPP, SP-110638 - LTE Proximity-Based Services Study Item. 2012. [5] A. Osseiran, K. Doppler, C. Ribeiro, M. Xiao, M. Skoglund, and J. Manssour, “Advances in device-to-device communications and network coding for IMT-Advanced,” in Proc. of ICT-MobileSummit 2009 Conference Proceedings, 2009. [6] LTE Release 10 & beyond (LTE-Advanced). [7] 3GPP, TR 36.814-900 Evolved Universal Terrestrial Radio Access (EUTRA); Further advancements for E-UTRA Physical Layer aspects. 2010. [8] R. Srinivasan, J. Zhuang, L. Jalloul, R. Novak, and J. Park, “IEEE 802.16m Evaluation Methodology Document (EMD),” tech. rep., IEEE, 2008. [9] IEEE, Std 802.11-2012: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications. 2012. [10] Atheros 9k drivers http://linuxwireless.org/en/users/Drivers/ath9k. [11] X. Wu, S. Tavildar, S. Shakkottai, T. Richardson, J. Li, R. Laroia, and A. Jovicic, “Flashlinq: A synchronous distributed scheduler for peer-topeer ad hoc networks,” in 48th Annual Allerton Conference, 2010. [12] ITU, “Recommendation ITU-R M-2135 Guidelines for evaluation of radio interface technologies for IMT-Advanced,” tech. rep., 2009. [13] K. Konstantinou, S. Kang, and C. Tzaras, “A measurement-based model for mobile-to-mobile UMTS links,” in Proc. of Vehicular Technology Conference, 2007. [14] M. Gudmundson, “Correlation model for shadow fading in mobile radio systems,” Electronic Letters, vol. 27, p. 2145–2146. [15] IEEE, Std 802.11g-2003: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications. 2003.
