Vous êtes sur la page 1sur 8

DYNAMIC SPECTRUM MANAGEMENT FOR 5G

Dynamic Reuse of Unlicensed Spectrum:


An Inter-Working of LTE and WiFi
Youjia Chen, Ming Ding, David Lpez-Prez, Jun Li, Zihuai Lin, and Branka Vucetic

Abstract spectrum as todays WiFi, there is a strong concern


of WiFi starvation and other forms of unfairness, as
The dynamic exploitation of unlicensed spec- WiFi would always detect the channel as busy and
trum by mobile operators is becoming a trend of back off. Simulations and analytical results show that
future 5G networks, with several efficient solutions this inter-system interference will greatly limit perfor-
being standardized for enabling spectrum sharing. mance in dense deployments [1].
Solutions with various design philosophies diversify With these caveats in mind, vendors and opera-
the network architectures and protocols. Among tors are actively seeking solutions for the simultane-
them, LAA aims at the physical-layer coexistence ous usage of the licensed and unlicensed spectrum
of LTE and WiFi within the unlicensed spectrum, in cellular operation, while ensuring fair coexistence
while LWA and LWIP focus on aggregating the link with WiFi. In this light, the Third Generation Part-
capacity of LTE in the licensed spectrum and WiFi in nership Project (3GPP) has recently standardized
the unlicensed one. In this article, a comprehensive a set of new technologies as part of LTE Release
survey of these three spectrum sharing technologies 13, that is, licensed-assisted access (LAA) [2], LTE-
are provided. Moreover, a novel analytical frame- WLAN aggregation (LWA) [3], and LTE-WLAN radio
work is proposed to evaluate the network perfor- level integration with IPsec tunnel (LWIP) [3]. These
mance of these technologies by incorporating both solutions enable radio aggregation of licensed and
spatial and time domain analyses and integrating dif- unlicensed spectrum for operator-controlled access,
ferent types of cells in one network as a whole. Sim- make the access to unlicensed spectrum transparent
ulation results are provided to compare the system to the operators evolved packet core, and simplify
throughput of these spectrum reuse technologies. the overall network maintenance by avoiding multi-
ple solutions for network management, security, and
Introduction authentication.
The dynamic usage of unlicensed spectrum by Generally speaking, the above-mentioned
mobile network operators has recently attracted solutions can be classified into two categories, as
considerable attention. It dramatically alleviates shown in Fig. 1:
the long-standing problem of spectrum scarcity in Those that attempt to adapt the current LTE
Long Term Evolution (LTE) systems, and offers user radio access technology to dynamically reuse
equipments (UEs) larger bandwidth and thus better the unlicensed spectrum
performance. The unlicensed spectrum band of Those that attempt to directly reuse the unli-
main interest is around 5 GHz, as it has large band- censed spectrum using IEEE 802.11 protocols
width globally available. However, this band poses Due to this fundamental variation, network architec-
a challenge to coexistence for LTE unlicensed oper- tures, interfaces, and protocols designed for these
ations, due to its wide usage by WiFi. two kinds of solutions are completely different.
Unlike LTE, which relies on a base station Besides the aforementioned coexistence issues,
(BS)-centric scheduled access mechanism, WiFi uses another major concern is the efficient usage of
a more distributed one based on carrier sense mul- the spectrum resource to enhance system perfor-
tiple access with collision avoidance (CSMA/CA). mance. Solutions with various radio access strate-
In more detail, all nodes in a WiFi network contend gies and dynamic spectrum management schemes
to access the channel using energy detection and a diversify the system performance. To obtain a
predefined threshold. Data transmissions are only better understanding of the performance of these
performed when the channel is unoccupied, that is, emerging technologies, characteristics of LTE and
the detected energy is smaller than the predefined WiFi should be comprehensively considered during
threshold. A maximum channel occupancy time is performance evaluation.
adopted to reinforce fairness, and an extra backoff For LTE, its network performance has been ana-
period is used to reduce collisions. lyzed using stochastic geometry theory in recent
This CSMA/CA mechanism, although polite and years, which is a powerful mathematical tool to
simple, has been shown to greatly degrade WiFi per- analyze large-scale wireless networks. By modeling
formance in dense deployments with a large num- the distributions of BSs and UEs as spatially random
ber of devices, as the distributed access becomes point processes, several key performance metrics
inefficient. Moreover, if the current LTE with con- can be derived, such as coverage probability and
tinuous transmission operates directly in the same average network capacity [4, 5]. For WiFi, howev-

Digital Object Identifier: Youjia Chen, Zihuai Lin, and Branka Vucetic are with the University of Sydney; David Lpez-Prez is with Bell Laboratories, Nokia;
10.1109/MWC.2017.1700088 Jun Li is with Nanjing University of Science and Technology and Southeast University.

52 1536-1284/17/$25.00 2017 IEEE IEEE Wireless Communications October 2017


LTE eNb LTE eNb LTE eNb

LTE LTE LTE


Licensed Licensed Licensed
band EPC band EPC band EPC

Unlicensed Xw
band Unlicensed LWIP-SeGW
UE UE Unlicensed UE
LTE band band
802.11
802.11
LAA eNB LWA Wi-Fi AP WI-FI AP
IPSec tunnel
Unlicensed Unlicensed Unlicensed
band band band
UE WI-FI AP UE WI-FI AP UE 802.11 WI-FI AP
802.11 (a) 802.11 (b) (c)

FIGURE 1. Architectures with three different spectrum sharing strategies: a) LAA; b) LWA; c) LWIP.

er, the existing work cannot effectively predict the coexistence of these LTE-like deployments with
performance of a large-scale WiFi network. The the existing WiFi networks in the unlicensed spec-
time-domain analysis of WiFi, based on the Markov trum. As shown in Fig. 1, LTE and WiFi interwork
chain, only considers a standalone, single-cell WiFi with each other in the physical layer, with LAA
system [6, 7]. The spatial domain analysis of WiFi, eNodeBs (eNBs) and WiFi APs contending for the
based on stochastic geometry, only treats downlink unlicensed channel.
(DL) transmissions, and investigates a snapshot of Listen-Before-Talk: For fair coexistence, a new
WiFi access points (APs) that can transmit simulta- feature for channel access is adopted in the LAA,
neously under the CSMA protocol [8, 9]. However, that is, listen-before-talk (LBT). The LBT technique
it neglects collisions, exponential backoffs, and most is a procedure whereby radio transmitters first
important uplink (UL) transmissions, with the latter sense the medium and transmit only if the medi-
being the main source for WiFi performance deg- um is detected to be idle, known as clear chan-
radation in real systems. In this circumstance, better nel assessment (CCA). An energy detection (ED)
network-level analysis is needed for WiFi networks. threshold is set beforehand to determine the exis-
In this article, we aim to achieve basic compre- tence of ongoing transmissions in the channel.
hension of the emerging technologies in dynamic At the same time, a random extended CCA pro-
spectrum sharing. First, we provide a comprehensive cedure is adopted in LBT before data transmission
survey of the three main spectrum sharing technolo- to mitigate collisions. This LBT scheme, designed
gies standardized in the 3GPP. Second, we present for LAA, resembles the medium access control
a novel evaluation framework, incorporating both (MAC) protocol used by WiFi in many ways. More
time domain and spatial domain analyses, to inves- specifically, CCA plays the role of CSMA in 802.11,
tigate the network performance of these spectrum and the eCCA procedure is similar to the exponen-
sharing technologies. With this framework: tial random backoff scheme designed for collision
We theoretically study for the first time a large- avoidance in WiFi.
scale WiFi system, which is mathematically Figure 2 shows an example of DL data transmis-
challenging due to its CSMA/CA protocol, sion for the LAA in an unlicensed band. If the detect-
requiring a nontrivial combination of a stochas- ed energy is below the ED threshold during the initial
tic geometry analysis in the spatial domain and CCA, the LAA node begins transmission immediately.
a Markov chain analysis in the time domain. Otherwise, the channel is assessed to be busy, and
We analyze the co-channel deployment of LTE then a defer duration and a backoff period consisting
and WiFi networks in the unlicensed spectrum, of a random number of additional extended CCA
where the sub-frame boundary issue in LAA has time slots have to pass before a transmission can be
been considered and the channel contention attempted again. Once the transmission opportunity
among different kinds of cells is modeled. is gained, the data transmission is limited by a maxi-
Most importantly, we systematically investigate mum channel occupancy time (MCOT) for fairness
the performance of different types of cells, syn- purposes. Different priority classes may correspond to
chronous and asynchronous, scheduling-based different MCOTs in the LAA.
and contention-based, thus making the compar- Frame Structure: Note that the LTE technology
ison of LAA and LWA/LWIP possible. works in a synchronous manner, in which data trans-
Third, we compare these different spectrum missions are aligned among cells. However, the LBT
sharing technologies in terms of system through- procedure of the LAA may be completed at any
put, thereby shedding new light on the deployment time instant. Therefore, the channel access time of
and operation of unlicensed spectrum. an LAA cell is generally not aligned with the LTE sub-
frame boundary, which appears every 1 ms. Since
Diverse Approaches to Spectrum Sharing the transmission opportunity can be taken by any
In this section, we present the fundamentals of other contender, while the LAA eNB is waiting for
the LAA, LWA, and LWIP in the context of radio the coming subframe boundary, a reservation signal
channel access. is necessary to fill the gap. However, reservation
signals contain no information bits, and thus waste
LAA time resources and incur additional signal overhead.
The LAA technology is aimed at using LTE-like As shown in Fig. 2, to efficiently utilize radio
infrastructure to access the unlicensed band. resources (i.e., lengthen data transmission time)
Thus, its main concern is the harmonious and fair and alleviate the reservation signal overhead, the

IEEE Wireless Communications October 2017 53


them idle, yielding a 50 percent On-Off cycle. During
CCA the CAST off periods, WiFi nodes have full access to
Backoff the band. LTE-U uses the MAC channel element acti-
DL UL Busy Defer Data
vation and deactivation feature of carrier aggregation
Tx to activate and deactivate the unlicensed carrier and
MCOT B C D realize the CAST duty cycle [10].
A
Since a WiFi device may start transmission at any
A: Channel clear - start time during CAST Off periods, if WiFi transmissions
transmission do not end before the next CAST On period, collision
B: Ready to transmit may occur. CAST has been enhanced with a short
but channel is busy Reservation Ending
C: Channel clear - start signal Subframe partial CCA to avoid colliding with WiFi transmissions at the
random wait period Initial subframe duty cycle flip, thus enhancing coexistence [11].
Subframe partial
D: Done waiting - boundary subframe
start transmission LWA and LWIP
FIGURE 2. An illustration of LAA data transmission. Different from LAA, as shown in Fig. 1, the LWA
and LWIP leverage WiFi infrastructure to access
the unlicensed spectrum. Thus, coexistence and
concept of a partial subframe is introduced into regulatory requirements in unlicensed spectrum
the LAA, including an initial partial subframe and an are not a concern.
ending partial subframe. The initial partial subframe Generally speaking, LWA and LWIP integrate
with length of 1/2 the normal subframe can start LTE and WiFi technology to enhance system perfor-
transmission in the middle of an LTE subframe. As mance. By exploiting both LTE and WiFi interfaces
illustrated in Fig. 2, if the first subframe boundary in parallel, users can benefit from the best of both
comes later than 0.5 ms, an initial partial subframe technologies, that is, LTE provides reliable connec-
can be transmitted after the reservation signal. The tivity and mobility management, while WiFi boosts
ending partial subframe can be transmitted to make data capacity through its large bandwidth. The main
the most use of the MCOT, whose length can be difference between LWA and LWIP is the different
chosen among {3, 6, 9, 10, 11, 12} orthogonal fre- layers where the integration of LTE and WiFi occur.
quency-division multiple access (OFDMA) symbols. LWA: LWA supports link aggregation at the
It is used when there is not enough time to transmit Packet Data Control Protocol (PDCP) layer, as
a full subframe comprising 14 OFDMA symbols. shown in Fig. 3. That is, in DL transmissions, PDCP
Enhanced LAA: In 3GPP Release 13, only DL packet data units (PDUs) of the same IP flow can
transmission in unlicensed spectrum is consid- be independently routed by the LTE eNB through
ered, while in Release 14 both DL and UL trans- the LTE and WiFi links. The reordering mecha-
missions are supported, which is also referred to nism at the PDCP layer ensures in-sequence data
as enhanced LAA (eLAA). In the eLAA, channel delivery to the upper layers at the UE. Since LWA
access is always driven by the LAA eNB, which works at an LTE radio protocol layer and benefits
contends for the channel. Once the LAA eNB has from PDCP radio link statistics, each PDCP PDU
gained access to the the channel through the afore- split can quickly adapt to radio and traffic fluctua-
mentioned LBT procedure, the subframes within its tions of LTE and WiFi links.
MCOT can be assigned to DL or UL transmissions LWA defines a new interface, Xw, between LTE
dynamically according to the LAA eNB schedul- and WLAN, which has similar functions to the X2
ing decisions. It is important to note that the UEs interface in LTE and is used to route PDCP PDUs to
that are provided with an UL grant and thus may WiFi APs. Specifically, the Xw interface is terminat-
transmit data in unlicensed band should perform ed at the WLAN termination (WT), a newly defined
an extra CCA prior to transmission to comply 3GPP logical node, which may be in control of one
with regulations and ensure the idle status of the or more WiFi APs and provides seamless mobility
channel. If the channel is sensed busy, the UE will among these WiFi APs. LWA also defines a new
ignore the UL grant and will not transmit. Ether-type to identify the PDCP PDUs routed over
Comparison with LTE-U: LTE in unlicensed WLAN, which allows the WiFi APs to differentiate
spectrum (LTE-U) was proposed by the LTE-U LWA traffic from other WLAN traffic. Note that
Forum. Both LTE-U and LAA use carrier aggre- software or firmware upgrades of legacy WiFi APs
gation to augment an existing LTE licensed car- to recognize this new Ether-type are needed to
rier with an unlicensed one. LTE-U, however, enable LWA operation.
is a simplified scheme without LBT targeted at LWIP: In contrast to LWA, LWIP provides a
early deployments. LTE-U operates in accordance more universal solution to access the unlicensed
with the existing Release 10/11/12 LTE physical band, realized via legacy WiFi APs (no new Ether-
(PHY)/MAC standards. The absence of LBT, how- type needed). LWIP supports link switching of
ever, restricts its use to regions such as the United IP flows at the IP layer, that is, the IP data is
States, where this is not required by government tunnelled from the LTE eNB to UE transparently
regulations. Instead, LAA has worldwide scope, over legacy WiFi APs relaying on IP connectivity.
and thus includes LBT and other features (e.g., Such tunnelling is realized via an LWIP-security
minimum bandwidth occupancy, transmit power gateway (SeGW) introduced between the LTE
spectral density) required to conform with, for eNB and UE, where IPsec is adopted to guar-
instance, European and Japanese regulations. antee data security when traversing the WLAN
To allow fair coexistence with WiFi and other network. As shown in Fig. 3, the IP packets trans-
technologies operating in the unlicensed band, LTE-U mitted between the LWIP-SeGW and the UE are
adopted CAST, which is based on an On-Off duty encapsulated with IPsec. Thus, the path between
cycle. For example, an LTE-U eNB would transmit on LWIP-SeGW and UE is also called an IPSec tun-
every other frame and leave the channel between nel, illustrated in Fig. 1.

54 IEEE Wireless Communications October 2017


Although LWA and
Split LWIP
Split LWA bearer LWIP require less
bearer change in existing net-
IP LWIPEP works and will become
Xw
PDCP LWAAP commercially available
PDCP IPsec
Delivery Delivery in a short term, more
RLC MAC RLC protocol protocol
IP MAC work is still needed in
MAC 802.11 MAC QoS provisioning, flow
LWIP-SeGW 802.11
PHY WLAN control and security to
terminal PHY LTE- WLAN AP
and AP eNodeB improve performance.
LTE eNodeB
(b) Moreover, WiFi spec-
(a)
ifications continue to
FIGURE 3. Comparison of the protocol architectures of LWA and LWIP: a) protocol architecture of LWA evolve, providing
[12]; b) protocol architecture of LWIP [3]. higher rates and
better mobility.
Different from the link aggregation capacity in and LTE-U will increase the traffic in unlicensed
LWA, with which PDCP packets of the same data spectrum, how to maintain a balance in unlicensed
flow can be simultaneously transmitted in both LTE spectrum, guaranteeing fair access to resources
and WiFi links, the IP packets of a data flow can for all other systems, such as WiFi, Bluetooth, and
be transmitted through either an LTE or a WiFi link ZigBee, has also raised deep concerns.
but never over both (no aggregation) in LWIP. This Although LWA and LWIP require less change
is because TCP cannot cope with the out-of-order in existing networks and will become commercially
delivery of packets transmitted over two radio inter- available soon, more work is still needed in quality
faces, which would lead to poor TCP performance. of service (QoS) provisioning, flow control, and
security to improve performance. Moreover, WiFi
Summary specifications continue to evolve, providing higher
The 3GPP dynamic spectrum sharing technologies rates and better mobility. Such enhancement may
(LAA, LWA, and LWIP) allow mobile operators to require additional optimizations; thus, the collabo-
benefit from additional capacity in the unlicensed ration of 3GPP and IEEE may allow LWA to further
band. The main difference between LAA and increase the benefits of these radio access technol-
LWA/LWIP is that LAA uses LTE-like radio access ogies (RATs).
technology in the unlicensed spectrum, while LWA
and LWIP directly access the unlicensed spectrum
by following IEEE 802.11 specifications.
A Framework of Performance
This fundamental difference leads to different Analysis for LTE and WiFi Interworking
protocol designs and network structures. For LAA, A key issue when discussing advantages and dis-
the LTE and WiFi technologies meet and coex- advantages of these different interworking tech-
ist with each other in the unlicensed spectrum nologies is their performance [13]. Thus, a new
through LBT. LWA and LWIP are aimed at aggre- framework is required to analyze the system per-
gating the capacity of LTE and WiFi by reusing WiFi formance of a large-scale network that includes
infrastructures. The former aggregates the LTE and different types of cells in licensed and unlicensed
WiFi links in the PDCP layer, while the latter does spectrums, such as LTE cells, LAA cells, and WiFi
it in the IP layer. cells. Here, WiFi cells include legacy WiFi cells
It is important to note that an anchor cell or and those new WiFi cells in which LWA or LWIP
component carrier (CC) in the licensed spectrum is adopted.
is needed for both LAA and LWA/LWIP, since con- To involve the analysis of different RATs in one
tinuous and periodic presence of reference sig- framework, their different characteristics should
nals and control channels cannot be guaranteed be considered. Generally, stochastic geometry is
in unlicensed spectrum. That is, the LTE cell in the a powerful math tool for the performance analy-
licensed band is the primary cell, where a UE com- sis of large-scale networks in the spatial domain.
municates and maintains its connection with the However, unlike the well-scheduled LTE systems,
network. LAA cells and WiFi cells working in LWA/ the CSMA or CSMA-like protocols adopted in unli-
LWIP can be viewed as secondary cells, which are censed spectrum make time domain analysis indis-
activated for supplementary bandwidth extension. pensable. Besides, the interactions between the
spatial domain analysis and the time domain analy-
Challenges sis cannot be neglected. Moreover, the interactions
A number of WiFi operators in different countries between different types of cells when they coexist
have expressed their concerns and approached in the same spectrum should be covered too. We
government regulatory bodies indicating that LAA present such a framework in the following.
and LTE-U operations may have a detrimental
impact on the existing and future use of unlicensed Performance Metrics
or shared spectrum. For fair coexistence, it is nec- As one important performance parameter, the
essary that deploying an LAA or LTE-U eNB should signal-to-interference-plus-noise ratio (SINR) of a
not impact WiFi services more than adding a WiFi typical UE can be formulated as
AP. Maximizing the capacity of LTE in unlicensed P (r) h
spectrum in this coexistence spirit still needs more SINR = , (1)
I agg + N 0
effort. Moreover, since the implementation of LAA

IEEE Wireless Communications October 2017 55


throughput and inter-cell interference in the net-
System performance work. Moreover, channel contention only occurs
among the active cells in unlicensed bands. The
calculation of the density of an active cell, ~l s, is
Intra-cell analysis Inter-cell analysis presented in the following subsection.
LTE: intra = 1 LTE:
Intra-cell Inter-cell
time
intra = 1 Intra-Cell Analysis
Time time LAA: Markov chain
efficiency The intra-cell analysis mainly includes two parts:
domain efficiency and probability the spatial domain analysis and the time domain
intra intra LAA, Wi-Fi:
Wi-Fi: Markov chain Pr[x0 M] analysis.
Spatial Domain: First, the distance r in Eq. 1,
Licensed: between the typical UE and its serving eNB/AP,
UE number determines the received signal power necessary
per cell Impact HPPP
Inter-cell to compute the UE SINR. No matter the type of
interference Unlicensed:
Spatial Serving distance cell, LTE, LAA, or WiFi, the distribution of such serv-
for signal: r MHCP M Impact
domain ing distance, r, can be derived from the density of
eNBs/APs and the association criteria adopted.
Stochastic Impact Second, the distribution of the number of UEs
geometry Density of active cells in a cell, n, is known to follow a negative binomial
distribution, that is,
u
FIGURE 4. The proposed analysis framework. n NB ,
u + K s
where K is a distribution parameter. Thus, the dis-
where P, h, and N 0 denote the transmission tribution of the number of UEs in an active cell, n,
power, channel gain, and additive white Gaussian follows a truncated negative binomial distribution.
noise power, respectively. Moreover, z() denotes Pr[n 0] can be used to calculate n and thus ~ls.
the path loss function, and I agg denotes the Time Domain: The intra-cell time efficiency,
cumulative interference, which comes from the h intra, is the focus here, which reflects the time
transmitters in other co-channel cells. Note that, fraction used for data transmission when consider-
different link directions or cell types may have dif- ing the cell as a standalone system:
ferent transmission powers, path loss functions, Inside an active LTE cell, due to centralized
and cumulative interference. scheduling and exclusive use of licensed spec-
The coverage probability is defined as the prob- trum, the time resource is fully used in full buf-
LTE
ability that SINR is above a determined threshold fer scenarios (i.e., hintra = 1).
g0, that is, Pr[SINR > g0]. Moreover, the SINR-de- Inside an active WiFi cell, due to the CSMA/
pendent area system throughput (AST) in bits per CA protocol, statuses can be differentiated:
second per square kilometer can be calculated as idle, transmission, and collision. Since data
can only be transmitted in the transmission
AST = B ~ls h log2(1 + SINR), (2) status, the time resource is not fully used, and
WiFi
hintra should be carefully studied. In this case,
where B denotes the bandwidth allocated, ~l s we adopt the existing Markov chain framework
WiFi
denotes the density of active cells with transmis- in [6, 7]. Note that hintra depends not only on
sions, and h denotes the time efficiency (i.e., the the parameters adopted in CSMA/CA, but also
time fraction spent on data transmissions). In the on the UE number inside the cell (i.e., ).
LAA
proposed analysis framework, h is divided into The calculation of h intra should consider two
two parts: h = h intra h inter, that is, intra-cell and aspects. First, a similar Markov chain model can
inter-cell time efficiency. be used to analyze the time efficiency of the
Before the detailed interpretation of the pro- CCA/eCCA procedure adopted in LAA. How-
posed analysis framework, a diagram is plotted in ever, different from WiFi cells, there are only
Fig. 4 to illustrate its major components and the idle and transmission statuses inside LAA cells.
targeted parameters that impact the system per- Intra-cell collisions do not take place as traffic
formance. is scheduled. Second, the time in MCOT occu-
pied by reservation signals should be excluded
Density of an Active Cell from the effective transmission time.
In practice, an LTE eNB/WiFi AP will mute its
transmission if there is no UE connected to it, Inter-Cell Analysis
aiming to reduce interference and energy con- In large-scale networks, especially with dense
sumption. Considering the increasing densities of deployments, inter-cell interference, Iagg, great-
eNBs and APs deployed in current and future net- ly degrades the system performance. Since in
works, the common assumption that all the cells licensed spectrum all active LTE cells can transmit
are active is no longer appropriate. simultaneously, Iagg comes from all other active
Modeling the distribution of eNBs/APs as a cells. However, in unlicensed spectrum where
homogeneous Poisson point process (HPPP) with contention-based channel access is adopted,
density ls and the distribution of its UEs as another the inter-cell interference only comes from the
HPPP with density l u, the density of active cells, cells that grab the chance to transmit at that time
denoted by ~l s, can be derived by the probability instance. Moreover, due to channel contention,
that there is at least one UE inside each cell. where one cell cannot always access the channe;, thus,
0 < ~ls < ls, and a larger lu leads to a larger ~ls. the inter-cell time efficiency, hinter, should also be
Note that only active cells contribute to system investigated in unlicensed spectra.

56 IEEE Wireless Communications October 2017


According to these differences, in the following, 80
we discuss the inter-cell analyses for licensed spec- LTE eNBs only
trum and unlicensed spectrum, respectively. LTE+LAA eNBs
LTE+LWA eNBs
Licensed Spectrum Spatial Domain Analy- 70
sis: As mentioned earlier, the inter-cell interference
for an active LTE cell comes from other active LTE 60
cells. Moreover, because LTE works in a synchro-
nous manner, no matter if in frequency-division

System throughput (Gbps)


50
duplex (FDD) or time-division duplex (TDD) mode,
the interference for an LTE DL transmission only
comes from the DL transmissions in other active 40
LTE cells, while that for an LTE UL transmission only
comes from the UL transmissions. In our modeling,
30
the spatial distribution of DL interference sources
follows an HPPP with density ~lLTE s , and that of UL
interference sources can also be approximately 20
modeled as an HPPP [5].
Unlicensed Spectrum Joint Spatial and Time 10
Domain Analysis: In traditional networks, only WiFi
cells are deployed in the unlicensed spectrum.
When the LAA strategy is engaged, LAA cells and 0
50 100 150 200 250
legacy WiFi cells coexist in the unlicensed band. Density of eNBs (1/km2)
When the LWA or LWIP strategy is adopted, new
WiFi cells join in, which essentially work in the FIGURE 5. Throughput gain with different strategies.
same way as legacy WiFi cells. In unlicensed spec-
trum, channel contention not only occurs inside a
cell, but also occurs among the nodes in different Through these three steps, the spatial distribu-
cells. Therefore, we conclude that: tion of interfering cells for typical cell x0 can be for-
Not all of the active cells in the unlicensed spec- mulated as fI = {xi:xi FM|x0 FM, xi x0}. Also,
trum can transmit simultaneously. In other words, the inter-cell time efficiency can be calculated by
each cell can use part of the time resource, the probability that the typical cell is retained in the
defined as inter-cell time efficiency hinter. MHCP (i.e., hinter = Pr[x0 FM]).
The inter-cell interference in one time instance
is generated only from the active cells that suc- Performance Studies
cessfully grab the transmission opportunity at In this section, we analyze the system perfor-
that time. mance with the various spectrum sharing strate-
To investigate hinter and the spatial distribution gies discussed.
of the inter-cell interference, three major steps are First, we establish the baseline performance
taken to model the inter-cell channel contention in from the traditional network with a clear cut of LTE
a tractable way: and WiFi, where LTE eNBs are deployed in licensed
Each cell in the contention is abstracted to a spectrum and WiFi APs are deployed in unlicensed
spatial point with a specific location and power. spectrum. Second, two spectrum sharing technolo-
Specifically, the center of the cell (i.e., the loca- gies are considered: LAA and LWA/LWIP. That is,
tion of its eNB/AP) can be used to represent the operator with LTE infrastructures deploys LAA
its location. Moreover, the expected transmis- eNBs or LWA APs to dynamically reuse unlicensed
sion power inside this cell represents its power. spectrum. These newly deployed LAA eNBs and
This abstraction is sufficiently accurate because LWA APs will contend for the spectrum resource
the distance between two co-channel cells is with legacy WiFi networks, which are assumed to
much larger than the average cell coverage, belong to another operator.
especially when the number of channels in unli- In Fig. 5, we present the performance gain that
censed spectrum is large. Besides, the usage of an operator can achieve by deploying LTE eNBs
the expected transmission power is reasonable and employing spectrum sharing strategies. When
due to the comparable transmission powers of only LTE eNBs are deployed in licensed spectrum,
an AP(LAA eNB) and a UE [14]. the system capacity almost linearly increases when
The contention domain or the contention range the density of LTE eNBs increases from 50/km2 to
is defined based on the CCA threshold. In more 250/km2 under the single-slope path loss model.
detail, cell A will be inside the contention domain The higher deployment density results in shorter
of cell B if the received power from A to B is high- serving distance and stronger received signals,
er than the pre-determined CCA threshold. which counterbalance the stronger interference.
We use the Matrn hard-core point process The idle mode also plays a role in the higher sys-
(MHCP), FM, to model the channel contention tem throughput.
among cells: To boost system performance, two different
An independent random mark is tagged onto strategies are considered: LAA and LWA (LWIP). In
each contending cell in an HPPP, representing more detail, one LAA eNB or LWA AP is co-located
the minimum backoff time in this cell. with every LTE eNB to use the unlicensed spectrum.
All cells that have a neighbor cell with a small- In this way, the operator obtains a performance
er mark and within its contention domain are improvement due to the extra spectrum resource in
removed, which means these cells do not the unlicensed band. Generally speaking, the perfor-
acquire the transmission opportunity in the mance gains achieved by LAA and LWA are similar.
channel contention. The difference is shown in more detail in Fig. 6.

IEEE Wireless Communications October 2017 57


satisfy two thresholds: virtual carrier sensing CCA,
Strategy1: Wi-Fi(operator 1) + LAA(operator2)
Strategy2: Wi-Fi(operator 1) + LWA(operator2) 82 dBm, and CCA ED, 62 dBm. The first thresh-
70 old limits the received interference from any other
throughput (G/ps/km2) 60 WiFi device to be less than 82 dBm, and the sec-
Unlicensed band

50 ond threshold limits the overall interference power


40
level. In contrast, the nodes in LAAs adopt a CCA
30
ED threshold of 72 dBm. With such configura-
20
tions, the 82 dBm threshold makes WiFi nodes
10
0 more sensitive than LAA nodes. Thus, when LAA
is used, the spatial reuse is larger, which leads to
P
P

P+ AA

P+ AA

P+ AA

P+ AA

P+ LAA

A
better performance.
0A
0A

LW

W
0 A 50L

0 A 0L

0 A 0L

0 A 0L
0L

0L

0L

0L
0 A 250
50
20
20

20 P+10

20 +20
20 +15
10

20
15

25
P+

In Fig. 6b, it is also important to note that when

+
P

P
0A

P
0A

0A

0A

0A
the UE density is very low, it can be observed that
20
20

20

20

20

20
20
Strategy 1: Strategy 1: the throughputs of both LAA and LWA cells are
AP throughput LAA throughput AP throughput LWA throughput small. The main reason is that some LAA/LWA
(a) cells have no UE in their coverage, which makes
35 no contribution to system throughput. However,
30 when the UE density keeps increasing to a relative-
Throughput (Gb/s/km2)

25 ly large value, the performance of LWA cells starts


20 to decrease while that of LAA cells does not. This
15 is because when the UE density increases, the UE
10
LAA with density 200 number in each LWA cell also increases. In LWA
LWA with density 200
LAA (DL only) with density 200 cells, which are basically WiFi cells, the growth of
5
LWA (DL only) with density 200 the contending nodes gives rise to the probability
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 of collision and thus more overhead. Thus, as in
Density of UE (1/km2) legacy WiFi cells, the system performance degrades
(b) when a large number of devices are involved.
We further plot the performance of LAA and
FIGURE 6. Throughput comparison of LAA and LWA: a) throughput in unli- LWA cells when only DL transmissions are support-
censed spectrum with different numbers of LAA and LWA cells; b) through- ed as standardized in LTE Release 13, shown by
put of LAA and LWA cells with different UE densities. the dashed lines in Fig. 6b. In such scenarios, first,
there is little performance degradation in LWA cells
caused by collisions, since DL transmissions are
Moreover, when comparing the performance scheduled in APs. Second, the system performance
of LTE only with that of LTE plus LAA or LWA in of both LAA and LWA cells is better than that when
Fig. 6a, it can be seen that the capacity of an LAA/ UL transmissions are involved. The main reasons for
LWA cell is less than that of an LTE cell when the this phenomenon is the lower contention and the
density is low (from 50/km2 to 150/km2), while the larger transmission power in the DL, which results
capacity in the unlicensed band is comparable to in higher transmission rate. Third, the performance
the capacity in the licensed band when the density of LWA is slightly better than that of the LAA when
increases. The main reason is that the transmission only DL transmissions are supported, due to the
powers of LTE eNB and LTE UE are both larger than extra synchronous cost in LAA cells. It is worth not-
those of LAA and LWA cells. Thus, when the cell ing that the performance difference depends on the
coverage is large, the signal quality in LTE cells is parameters adopted in LWA and LAA cells, such as
better than in LAA and LWA cells. However, when MCOT in the LAA, maximum backoff stages, and
the density is high, LTE suffers from severe inter-cell basic contention window size in 802.11.
interference, while the interference for LAA or LWA Compare these two different scenarios, one
cells is limited by the channel contention. joint UL and DL transmissions and the other only
Moreover, Fig. 6a draws the corresponding DL transmissions: LAA performs better than LWA
throughput in the unlicensed spectrum when in the first scenario, because the collisions existing
different densities of LAA eNBs or LWA APs are in LWA degrade its performance; however, LWA
deployed. As expected, the overall throughput in performs better in the second scenario, because in
unlicensed spectrum increases with the deployed this scenario both LWA and LAA have no collisions
number of serving nodes (LAA eNBs, LWA APs, while LAA has extra cost on frame synchronicity.
and WiFi APs). However, the throughput of a leg-
acy WiFi network whose density keeps 200/km2 Conclusions
decreases in this process. From the perspective Unlicensed spectrum sharing by mobile operators
of the legacy WiFi network, the newly deployed is a hot research topic for future 5G networks.
LAA or LWA cells also contend for the spectrum Currently, three major approaches have been pro-
resources, which were occupied by them before. posed to enable this spectrum reuse. LAA aims
Under the CSMA/CA access scheme, the more at the coexistence of LTE and WiFi in the physi-
nodes contend for the channel, the less chance to cal layer of unlicensed spectrum, while LWA and
access, and thus lower throughput. LWIP focus on aggregating licensed spectrum in
In Fig. 6b, we compare the capacity of LAA LTE and unlicensed spectrum in WiFi. Considering
with that of LWA as the UE density varies. Gen- these different strategies, LTE cells, LAA cells, and
erally speaking, the performance of LAA is slightly WiFi cells are incorporated into one network as a
better than that of LWA. This is because LWA cells, whole. From our simulation results, we conclude
which are also WiFi cells, use two energy detec- that the performance gains achieved by LAA and
tion (ED) thresholds, while LAA cells use only one LWA are similar, which present more than one
ED threshold. That is, the nodes in WiFi have to option for mobile operators to unlock the benefits

58 IEEE Wireless Communications October 2017


of unlicensed spectrum. However, in scenarios Ming Ding [M12] (Ming.Ding@data61.csiro.au) received his Unlicensed spectrum
B.S. and M.S. degrees (with first class Hons.) in electronics
with a large number of UEs, LAA provides bet- engineering from Shanghai Jiao Tong University (SJTU), China, sharing by mobile
ter performance than LWA because contention and his Ph.D. degree in signal and information processing from operators is a hot
and collision are removed inside the LAA cells. SJTU in 2004, 2007, and 2011, respectively. From Septem-
research topic for
Moreover, we confirm that these newly deployed ber 2007 to September 2011, he pursued a Ph.D. degree at
SJTU while at the same time working as a researcher/senior future 5G networks.
cells in unlicensed spectrum will degrade the per- researcher at Sharp Laboratories of China (SLC). After achiev-
formance of legacy WiFi networks. Nevertheless, ing his Ph.D. degree, he continued working with SLC as a Currently, three major
the performance impact brought by these newly senior researcher/principal researcher until September 2014, approaches have been
deployed cells is no more than that from addition- when he joined National Information and Communications
Technology Australia (NICTA). In September 2015, the Com- proposed to enable
al legacy WiFi nodes. monwealth Scientific and Industrial Research Organization and this spectrum reuse.
NICTA joined forces to create Data61, where he continued
Acknowledgment as a senior research scientist in this new R&D center locat-
ed in Sydney, Australia. He has authored about 40 papers in
The LAA aims at the
This work is supported in part by the National coexistence of LTE and
IEEE journals and conferences, all in recognized venues, and
Natural Science Foundation of China under Grant about 20 3GPP standardization contributions, as well as a WiFi in the physical
61501238, in part by the Jiangsu Provincial Sci- book, Multi-point Cooperative Communication Systems: Theory layer of unlicensed
ence Foundation under Project BK20150786, in and Applications (Springer). Also, as the first inventor, he holds
spectrum, while the
part by the Specially Appointed Professor Pro- 15 Chinese, seven Japanese, three U.S., and two Korean pat-
ents, and has co-authored another 100+ patent applications on LWA and LWIP focus
gram in Jiangsu Province, 2015, in part by the 4G/5G technologies.
Fundamental Research Funds for the Central Uni- on aggregating licensed
versities under Grant 30916011205, and in part David Lpez-Prez [M12] (dr.david.lopez@ieee.org) received spectrum in LTE and
by the Open Research Fund of National Mobile his B.Sc. and M.Sc. degrees in telecommunication from Miguel
Hernandez University, Spain, in 2003 and 2006, respectively, unlicensed spectrum
Communications Research Laboratory, Southeast and his Ph.D. degree in wireless networking from the Univer- in WiFi.
University, under grant no. 2017D04. sity of Bedfordshire, United Kingdom, in 2011. He was with
Vodafone, Spain, from 2005 to 2006, where he was involved in
References network planning and optimization. He was a research associate
[1] J. Jeon et al., LTE in the Unlicensed Spectrum: A Novel with Kings College London from 2010 to 2011. He is currently
Coexistence Analysis with WLAN Systems, Proc. 2014 IEEE a member of technical staff with Nokia Bell Laboratories. He
GLOBECOM, Dec. 2014, pp. 345964. authored the book Heterogeneous Cellular Networks: Theory,
[2] 3GPP, Feasibility Study on Licensed-Assisted Access to Unli- Simulation and Deployment (Cambridge University Press, 2012).
censed Spectrum, TR 36.889 v. 13.0.0, July 2015. He has authored over 90 book chapters, journal, and confer-
[3] 3GPP, Evolved Universal Terrestial Radio Access (E-UTRA) ence papers, all in recognized venues. He holds over 30 patents
and Evolved Universal Terrestial Radio Access Network applications. He also received his Ph.D. Marie-Curie Fellow in
(E-UTRAN); Overall Description; Stage 2, TS 36.300 v. 2007. He was also a finalist for the Scientist of the Year prize in
13.3.0, Apr. 2016. the Irish Laboratory Awards in 2013 and 2015. He was an Exem-
[4] J. Andrews, F. Baccelli, and R. Ganti, A Tractable Approach plary Reviewer of IEEE Communications Letters in 2011. He is or
to Coverage and Rate in Cellular Networks, IEEE Trans. has been a Guest Editor of a number of journals, for example,
Commun., vol. 59, no. 11, Nov. 2011, pp. 312234. the IEEE Journal on Selected Areas in Communications and IEEE
[5] T. Ding et al., Uplink Performance Analysis of Dense Cellu- Communications Magazine.
lar Networks with LoS and NLoS Transmissions, IEEE Trans.
Wireless Commun., vol. 16, no. 4, Apr. 2017, pp. 260113. Jun Li [M09, SM16] (jun.li@njust.edu.cn) received his Ph.D.
[6] G. Bianchi, Performance Analysis of the IEEE 802.11 Dis- degree in electronic engineering from Shanghai Jiao Tong Uni-
tributed Coordination Function, IEEE JSAC, vol. 18, no. 3, versity in 2009. From January 2009 to June 2009, he worked
March 2000, pp. 53547. in the Department of Research and Innovation, Alcatel Lucent
[7] D. Malone, K. Duffy, and D. Leith, Modeling the 802.11 Shanghai Bell as a research scientist. From June 2009 to April
Distributed Coordination Function in Nonsaturated Hetero- 2012, he was a postdoctoral fellow at the School of Electri-
geneous Conditions, IEEE/ACM Trans. Networking, vol. 15, cal Engineering and Telecommunications, University of New
no. 1, Feb. 2007, pp. 15972. South Wales, Australia. From April 2012 to June 2015, he was
[8] Y. Li et al., Modeling and Analyzing the Coexistence of a research fellow at the School of Electrical Engineering, Univer-
Wi-Fi and LTE in Unlicensed Spectrum, IEEE Trans. Wireless sity of Sydney. Since June 2015 he has been a professor at the
Commun., vol. 15, no. 9, Sept. 2016, pp. 631026. School of Electronic and Optical Engineering, Nanjing Univer-
[9] J. Li et al., Share in the Commons: Coexistence between sity of Science and Technology. His research interests include
LTE Unlicensed and Wi-Fi, IEEE Wireless Commun., vol. 23, network information theory, channel coding theory, wireless
no. 6, Dec. 2016, pp. 1623. network coding, and cooperative communications.
[10] C. Cano et al., Using lte in Unlicensed Bands: Potential
Benefits and Coexistence Issues, IEEE Commun. Mag., vol. Z ihuai L in [S98, M06, SM10] (zihuai.lin@sydney.edu.au)
54, no. 12, Dec. 2016, pp. 11623. received his Ph.D. degree in electrical engineering from Chalm-
[11] B. Chen et al., Coexistence of LTE-LAA and Wi-Fi on 5 ers University of Technology, Sweden, in 2006. Prior to this he
GHz with Corresponding Deployment Scenarios: A Survey, held positions at Ericsson Research, Stockholm, Sweden. Follow-
IEEE Commun. Surveys & Tutorials, vol. 19, no. 1, 1st qtr. ing his Ph.D. work, he worked as a research associate professor
2017, pp. 732. at Aalborg University, Denmark, and is currently at the School
[12] D. Lpez-Prez et al., Long Term Evolution-Wireless Local of Electrical and Information Engineering, University of Sydney.
Area Network Aggregation Flow Control, IEEE Access, vol. His research interests include source/channel/network coding,
4, 2016, pp. 986069. coded modulation, MIMO, OFDMA, SC-FDMA, radio resource
[13] A. M. Voicu, L. Simic, and M. Petrova, Inter-Technolo- management, cooperative communications, small-cell networks,
gy Coexistence in a Spectrum Commons: A Case Study of 5G cellular systems, and so on.
Wi-Fi and LTE in the 5-GHz Unlicensed Band, IEEE JSAC,
vol. 34, no. 11, Nov. 2016, pp. 306277. Branka Vucetic [M83, SM00, F03] (branka.vucetic@sydney.
[14] Y. Chen et al., A Space-Time Analysis of LTE and Wi-Fi edu.au) received her B.S.E.E., M.S.E.E., and Ph.D. degrees in
Inter-Working, IEEE JSAC, vol. 34, no. 11, Nov 2016, pp. 1972, 1978, and 1982, all in telecommunications, from the
298198. University of Belgrade. During her career she has held various
research and academic positions in Yugoslavia, Australia, the
Biographies United Kingdom, and China. She currently holds the Peter Nicol
Youjia Chen (youjia.chen@sydney.edu.au) received her B.S. and Russel Chair of Telecommunications Engineering at the Uni-
M.S degrees in communication engineering from Nanjing Universi- versity of Sydney and serves as the director of the Centre of
ty, China, in 2005 and 2008, respectively. From 2008 to 2009, she Excellence in Telecommunications. She has published more than
worked at Alcatel-Lucent Shanghai Bell, and in August 2009 she 300 research papers and coauthored four books in telecom-
joined the College of Photonic and Electrical Engineering, Fujian munications and coding theory. Her most significant research
Normal University, China. She is currently pursuing a Ph.D. degree contributions have been in the field of channel coding and
in wireless engineering at the University of Sydney, Australia. Her its applications in wireless communications. Her research has
current research interests include resource management, load bal- involved collaborations with industry and government organiza-
ancing, and caching strategy in heterogeneous cellular networks. tions in Australia and several other countries.

IEEE Wireless Communications October 2017 59

Vous aimerez peut-être aussi