Professional Documents
Culture Documents
Bell Labs Technical Journal 18(2), 37–56 (2013) © 2013 Alcatel-Lucent. • DOI: 10.1002/bltj.21604
Panel 1. Abbreviations, Acronyms, and Terms
3D—Three-dimensional MAC—Media access control
3G—Third generation MIMO—Multiple input multiple output
3GPP—3rd Generation Partnership Project NLOS—Non-line-of-sight
AAA—Active antenna array OFDM—Orthogonal Frequency Division
CDF—Cumulative distribution function Multiplex
CoMP—Coordinated multipoint PF—Proportional fair
CPRI—Common public radio interface PMI—Precoding matrix index
CQI—Channel quality indication RF—Radio frequency
CSI—Channel state information SINR—Signal-to-interference-plus-noise ratio
eNB—Evolved NodeB SNR—Signal-to-noise ratio
HPBW—Half-power beam width SON—Self-optimizing network
ISD—Inter site distance TTI—Transport time interval
LOS—Line of sight TS—Technical specification
LTE—Long Term Evolution UE—User equipment
multi-antenna systems have been applied to improve Some self-optimizing network (SON) systems [5]
performance. Classic beamforming concentrates the already make use of antenna downtilt adaptation to
signal energy in the direction of the receiver, and at optimize cell performance, but they do so in a static or
the same time reduces interference in other direc- quasi-static way. However, with upcoming advanced
tions. MIMO systems exploit the spatial diversity of antenna technologies, completely new and highly
the radio channel or enable spatial multiplexing dynamic schemes for horizontal and vertical beam
through adaptation of the antenna weights accord- pattern adaptation on a per resource (transport time
ing to the channel characteristics [3]. interval (TTI)) and per UE basis are now becoming
All these techniques usually operate only on the feasible, which allows the use of 3D beamforming in
horizontal antenna pattern. The vertical pattern is combination with interference coordination tech-
fixed and has a narrow half-power beam width niques. Some of these functionalities can already be
(HPBW) which must be wide enough to cover the realized with the current 3rd Generation Partnership
cell range and small enough to guarantee a high Project (3GPP) Technical Specification (TS) 36.211
antenna gain. However, the resulting inter-cell inter- Release (Rel.) 10 standard [2], but others would
ference is the same whether UEs close to the evolved require extensions, which need to be taken into
Node B (eNB) are being served or UEs close to the account and are under discussion for future releases.
cell edge are being served. This paper provides an overview of these new
3D beamforming, which enables fully dynamic possibilities. It highlights the technological back-
adaptation of the vertical beam pattern per resource ground, the basic principles, and several realization
and per UE, offers promising performance improve- options for 3D beamforming. We investigate the
ments. It can increase signal strength by pointing the potential benefits in single- and multi-cell scenarios
vertical main lobe directly at the receiver at any loca- and also consider a vertical distribution of users.
tion. In addition, it reduces inter-cell interference Concepts for combined 3D beamforming and interfer-
when serving users closer to the eNB. With beam ence coordination are also addressed. Though there
coordination or MIMO schemes, the variation in ver- has been a good deal of theoretical and simulation
tical dimension has the capability to exploit addi- work, practical measurements are quite rare. We
tional diversity or spatial separation, which then can present field trials which verify the feasibility and
be used to either improve signal quality or increase benefits of 3D beamforming in a real deployment sce-
the number of simultaneously served users. nario at the Bell Labs testbed in Stuttgart.
0.8
0.4
0.2
0
0 5 10 15 20 25 30
Downtilt (°)
CDF—Cumulative distribution function
eNB—Evolved NodeB
Figure 1.
CDF of downtilts in a multi-cell system (inter site distance = 500 meters, eNB height = 30 meters).
ϕ
System Description and Performance Evaluation
(System Simulation)
Antenna elements
We used an Orthogonal Frequency Division
Multiplexing (OFDM)-based cellular system to inves-
RF transmission
signal tigate the 3D beamforming capabilities of the down-
RF—Radio frequency
link. The resources were arranged in a radio frame
according to LTE-Advanced 3GPP Release 10 [2].
Figure 2. Assignment to users is in physical resource blocks
Principle of beamforming antenna. extended in time and frequency. Each resource block
consisted of 12 subcarriers and 14 time symbols. For
fact leads to an approximation of the exact main lobe a system with 10 MHz bandwidth, the number of
steering mode: the selection among a small number resource blocks was 50. These resource blocks were
of fixed downtilts (2 or 3), and also dynamically per equally distributed among the UEs in a cell. At the
resource block and UE. Depending on the location of eNB side, the antenna was either a single sector
Figure 3.
Serving users with three fixed downtilts.
antenna or one consisting of four antenna elements noise-limited systems with no or negligible inter-cell
with λ/2 spacing to realize horizontal beamforming. interference, and a multi-cell scenario reflecting inter-
Performance potential and benefits of the 3D beam- ference-limited systems, which is the typical case in
forming operation modes introduced above were many cellular radio systems aiming at spectral
evaluated with an LTE downlink system simulator efficiency.
comprised of multiple sites and cells. The main system
parameters used for the various evaluations are sum- Single-Cell Scenario
marized in Table I. As different aspects of 3D beam- This deployment scenario with only a single cell is
forming were investigated, e.g., a comparison of the well suited to investigate the impact of the vertical
operation modes in noise-limited and interference- main lobe adaptation of the antenna pattern, i.e.,
limited systems, or multi-cell systems with and with- exploiting the maximum vertical antenna gain at all
out coordination, some of the parameters differ from locations within the cell. The evaluation of one single
case to case. We present the most important results isolated cell reflects a noise-limited system where the
and conclusions from our simulation studies below. impact of interference on the performance can be
The system simulation first determines the distri- neglected. This is typically the case in cellular systems
bution of the signal-to-interference-plus-noise ratio with high frequency reuse factors, since adjacent cells
(SINR) in the considered area based on a random operate in different frequency bands. One interesting
positioning of UEs. Link level abstraction models aspect is that coverage can be extended by pointing the
then map the SINR values to a corresponding antenna pattern main lobe towards UEs close to the cell
throughput performance used to derive the two met- edge, rather than using a fixed downtilt. In particular,
rics of interest in this study: Spectral efficiency meas- the objective was to compensate the additional path
ured in bit/s/Hz/cell is the expectation of the sum loss when the 3D beamforming system is operating at a
rate per frequency unit in one cell, and cell edge user 2.6 GHz carrier frequency instead of 2.1 GHz. This is
throughput is defined as the five percent quantile of a typical scenario if the operator changes over from
all observed UE throughputs in kbit/s. third generation (3G) to LTE and maintains its site loca-
tions. The additional path loss at 2.6 GHz is approxi-
3D Beamforming Without Coordination mately 2 dB for a typical cell radius of 250 meters. It
At first, we evaluated the performance poten- was assumed that the 3D beamforming system is
tial of 3D beamforming without coordination. We dis- equipped with an antenna that can switch between a
tinguished between a single cell scenario representing near- and a far downtilt for the inner cell and cell edge,
Scenario 1 Scenario 2
Scenarios Beamforming without coordination Beamforming with coordination
Single cell Multi-cell Multi-cell with Implicit Horizontal and vertical
vertical UE coordination beam coordination
distribution
Purpose Compensation 3D beamform- Evaluation of per- Performance Performance improve-
of path loss at ing downtilt formance poten- improvement with- ment using additional
different parameter tial in dense urban out inter-eNodeB adjacent cell interfer-
frequencies optimization scenarios with communication ence information
high buildings
Inter site
— 500 meters
distance
Antenna eNB: 32 m; UE: 1.5 m eNB: 30 m; UEs dis- eNB: 32 m; UE: 1.5 m
height tributed between
0 m and 90 m
eNodeB Measured Shape accord- Shape according Shape according Shape according to
antenna beam pattern ing to 3GPP to 3GPP TR 36.814; to 3GPP TR 3GPP TR 36.814;
pattern of AAA with TR 36.814; 36.814;
Horizontal HPBW: Horizontal HPBW: 70°
horizontal
Horizontal 70° (4 elements); Horizontal HPBW: (4 elements);
HPBW: 75°
HPBW: 70° (1 Vertical HPBW: 10° 70° (4 elements); Vertical HPBW: 6.5°
and vertical
element); Vertical HPBW: 10°
HPBW 6.2°
Vertical HPBW:
10°
Number of 1/1 19/57 7/21
sites/cells
Cell: 120° Cell: 120° sector Cell: 120° sector
sector
Propagation 3GPP SCME case 1
Line-of-sight
model (non-line-of-sight)
UEs per cell 10 randomly distributed 15 randomly distributed
Frequency
10 MHz bandwidth; reuse 1
resources
Scheduling
Round robin Proportional fair
algorithms
Major study Compensation “Two fixed Rural flat scenario: 42% gain in cell 50% gain in cell edge
outcome of increased downtilt opera- 5% gain in spec- edge throughput throughput
path loss if tion” mode tral efficiency
12% gain in 16% gain in spectral
carrier fre- primarily
Skyscrapers sce- spectral efficiency
quency steps improves cell
nario: 80% gain in efficiency
from 2.1 GHz edge user
spectral efficiency
to 2.6 GHz is throughput,
possible whereas
“Direct steering
with limitation“
boosts spectral
efficiency
3D—Three-dimensional
3GPP—3rd Generation Partnership Program
AAA—Active antenna array
eNB—Evolved NodeB
HPBW—Half-power beam width
SCME—Spatial channel model extended
TR—Technical recommendation
UE—User equipment
1,9
Spectral efficiency (bit/s/Hz)
1,8
1,7
15°/8°
15°/9°
1,6 15°/10°
17°/8°
17°/9°
1,5 17°/10°
19°/9°
19°/10°
1,4
80 100 120 140 160 180
Border between near and far area (meters)
Figure 4.
Spectral efficiency for combinations of two fixed downtilts versus the distance of the border between near- and
far area from the base station.
Table II. Performance comparison of two different radio systems operating at different carrier
frequencies with individually optimized parameters.
Metric One fixed downtilt 10° 2.1 GHz Two fixed downtilts 17° and 9° 2.6 GHz
Spectral efficiency 1.86 bit/s/Hz 1.91 bit/s/Hz
Cell edge user
437 kbit/s 399 kbit/s
throughput
respectively. This parameter setting is summarized in of near downtilt = 17° (serving 23 percent of the
the “Single Cell” column in Table I, scenario 1. cell) and far downtilt = 9° (serving 77 percent of
As a first step, and as shown in Figure 4, differ- the cell) to be the most appropriate setting.
ent combinations of near and far downtilt were com- Table II provides a comparison of the baseline
pared, and an appropriate border maximizing the system (2.1 GHz with one fixed downtilt at 10°)
spectral efficiency between the regions was identi- and the 3D beamforming system (2.6 GHz with two
fied for each combination. Surprisingly, the main fixed downtilts at 17° and 9°). In spite of the
outcome was that the achievable maximal spectral increased path loss, spectral efficiency can be slightly
efficiency does not depend on the radius of the bor- improved with 3D beamforming since the received
der. For each border a different combination of near signal power in the near area is significantly higher
and far downtilt provides this optimal value for the as a result of the steep near downtilt of 17°.
spectral efficiency. Overall, we found the combination The much more important fact is that the cell edge
160 12°
Increasing
DT limitation
Increasing
120
far DT
7° 16°
5°
80
16°
Increasing
fixed DT
40
10°
0
0.8 1 1.2 1.4 1.6 1.8
Spectral efficiency (bit/s/Hz/cell)
DT—Downtilt
Figure 5.
Cell edge user throughput over spectral efficiency for different operation modes and configurations.
user throughput can be conserved. Thus 3D beam- UE, as well as the vertical HPBW. The “Multi-Cell”
forming is able to compensate a 2 dB increase in column in Table I, scenario 1 represents the param-
path loss. eters for a classic macrocellular scenario. The objec-
tive was to study the pure impact of the vertical
Multi-Cell Scenario antenna pattern on the performance. We therefore
In the remainder of this paper we focus on a utilized a fixed sector pattern of 70° on the horizon-
multi-cell scenario with a frequency reuse factor of tal plane. The baseline again was a system with one
1 since it is widely deployed in densely populated fixed downtilt which varied between 10° and 18°.
suburban and urban areas where mutual interfer- As the downtilt increased, spectral efficiency also
ence between the cells is the major performance- increased until the optimum was reached at 16°.
limiting effect. The carrier frequency is 2.6 GHz This is because interference is reduced as the down-
throughout all studies. The basic objective was to tilt becomes steeper. However, at 17° the cell edge
optimize the setting of the different 3D beamform- user throughput drops off because the far region of
ing operation modes and to compare them with the cell can no longer be covered.
respect to achievable spectral efficiency and cell edge Figure 5 also shows the results for a case with
user throughput. The most significant parameters two fixed downtilts. The near downtilt is fixed at 17°,
impacting the results shown in Figure 5 are the size and the far downtilt varies from 5° onwards in 1°
of the cell, the antenna heights of the eNB and the steps. We observe that both spectral efficiency and
1,8
1,6
1,4
1,2 Rural eNodeB Skyscrapers
1
0 10 20 30 40 50 60 70 80 90 100
Building height (meters) and indoor ratio (%)
3D—Three dimensional
UE—User equipment
Figure 6.
Increasing spectral efficiency gains of 3D-beamforming over one fixed downtilt in vertical scenarios.
appropriately in the city or skyscraper scenario with one Thus 3D beamforming also proves its superiority
Δh
fixed downtilt. The elevation angle Φe is Φe = arctan
x
, ( ) over conventional antenna systems in multi-cell deploy-
ments, especially in vertical scenarios with tall build-
where Δh is the vertical distance between the eNB
ings. The fact that the beam can also be steered in an
antenna and the UE, and x is the respective horizon-
uptilt direction allows the eNB antenna installation to
tal distance. In these two scenarios it is even better to
be below rooftop. As a further advantage, tall buildings
operate with one fixed uptilt because of the large
can then be exploited to provide isolation between cells.
number of indoor UEs in our model. This effect is
also visible in Figure 6 (dark blue lines with trian-
gles). However, outdoor UEs would no longer be 3D Beamforming in Combination With
covered. Direct steering also suffers from the increas- Interference Coordination
ing variety of UE heights, but it can adapt much bet- The advantage of 3D beamforming is based on
ter to these vertical scenarios, as indicated by a improving the SINR at the receiver. On the one hand,
smaller decrease in spectral efficiency when stepping this can be achieved by maximizing the desired user
through the scenarios (black lines). Consequently, signal, in particular, with a direct steering mode of
the main benefit of 3D beamforming becomes clearer operation. On the other hand, a reduction of the
and clearer. Though in a flat scenario, the expected expected interference provides the same effect. Possible
gain in spectral efficiency ranges around five percent, approaches discussed in the previous section include
this value increases significantly—up to 80 percent— limiting the minimal possible downtilt and applying a
in the skyscraper scenario. Figure 6 also shows the rather steep downtilt for UEs close to the eNB in the
expected performance for indoor UEs only (dotted near area. However, up to now, only statistical effects
lines). It can be clearly seen that 3D beamforming have been exploited for interference reduction. The
can keep the spectral efficiency constant although question is how established interference coordination
building height and mean wall penetration loss techniques can be advantageously combined with 3D
increase. beamforming, and which additional performance
Figure 7.
Implicit coordination schemes.
Coordination gain:
42% cell edge user throughput
12% spectral efficiency
1000
DT 15° (Baseline)
DT 15° (hor. sorting)
900 DT 17°, 13°, 9°
DT 17°, 13°, 9° (vert. sorting)
DT 17°, 13°, 9° (hor. sorting)
800
31% 3D-BF gain
Direct steering
Direct steering (vert. sorting)
700 Direct steering (hor. steering)
4% 3D-BF gain Direct steering, limit 15°
Figure 8.
Gains of implicit coordination schemes.
spacing and it selects one of eight possible horizontal Distributed Horizontal and Vertical Beam Coordination
beams for each UE. The baseline is one fixed downtilt Applying a UE-specific downtilt already shows
of 15°. Vertical and horizontal sorting for “three fixed gains in cell edge user throughput and spectral effi-
downtilts” and direct steering with and without a ciency when applied in each cell independently or
downtilt limitation are compared. The highest gain is with implicit coordination. Since these gains result
achieved by a combination of 3D beamforming and from statistical reduction of interference, we expected
horizontal sorting. The gain with vertical sorting is even higher gains with an explicit and more dynamic
smaller, mainly due to the limited angular separation coordination of resource allocation which takes into
in the vertical direction. A smaller HPBW could account the actual UE distribution and scheduling
improve this. Maximum gain is achieved with direct requirements. Such a resource coordination scheme
steering and downtilt limitation (a 42 percent requires exchange of information on scheduling
increase in cell edge user throughput, and a 12 per- decisions and resulting interference between eNBs. It
cent increase in spectral efficiency with the system is therefore not implementable with current 3GPP
setup used). It also shows the highest absolute spec- Rel. 10 systems.
tral efficiency. Maximum cell edge user throughput The scheme described in the following assumes
is achieved with three fixed downtilts. This is due to the availability of specific UE feedback on adjacent
reduced interference at the cell edge, whereas direct cell interference. The eNBs exchange this informa-
steering boosts spectral efficiency by maximizing the tion so the schedulers can use the information for
signal strength at the UE. their own scheduling decisions. These schedulers
•29.9%
900.0
850.0
α=3 α=2
800.0
750.0 •19%
700.0 •9.6%
α=1 •6.5%
650.0
600.0
550.0 Fixed DT 12°, no beam coordination
500.0 Fixed DT 12°, horizontal beam coordination
450.0 3D beamforming with downtilt limitation to 11°, no beam coordination
400.0 3D beamforming with downtilt limitation to 11°, horizontal and vertical
350.0 beam coordination
α = 0.5
300.0
1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6
Spectral efficiency (bits/s/Hz)
3D—Three dimensional
DT—Downtilting
UE—User equipment
Figure 9.
Gains of beam coordination schemes.
apply constraints on specific resources for a specific fairness parameter α for weighting the estimated
UE to avoid interference. The basic coordination data rate based on channel quality indication (CQI)
algorithm for a proportional fair (PF) scheduler, reporting. So each eNB schedules the UE with the
described in [10], already provides some gains with- highest score, for which none of the PMI constraints is
out 3D beamforming. The combination with 3D violated. The PF score is calculated from the average
beamforming adds an additional degree of freedom rate r_avt, the expected instantaneous rate r_instt and
for coordination. The specific UE feedback consists of the weighting parameter α according to
the “worst case interfering precoding matrix indices”
scoret = r_instt/(r_avt)α
(WCIs) of each adjacent cell, which are communi-
cated to each eNB. The WCIs are the precoding Figure 9 shows the simulation results of cell edge
matrix indices (PMIs), which will cause maximum throughput versus the spectral efficiency, with the
interference when applied in the adjacent cell. In parameters from the “Horizontal and Vertical Beam
combination with 3D beamforming, these WCIs are Coordination,” column in Table I, scenario 2 applied.
estimated for the specific downtilt used to serve the All curves represent cases with direct steering, and
UE. The PF schedulers use this information to derive the baseline is a system with a fixed downtilt of 12°
resource allocations which avoid assigning PMIs that without beam coordination (black curve). If horizon-
would cause significant interference in adjacent cells tal coordination is applied without 3D beamforming,
or disturb the cell itself. The scheduling parameters we see a gain of 19 percent in cell edge user through-
include the number of interfering WCIs that must be put and 6.5 percent in spectral efficiency (gray curve).
considered to determine the PMI constraints, and a The gain of the dark blue curve shows the statistical
−2
−4
Relative Rx power (dB)
−6
0
−5 −8
−10
−15 −10
−20
10
−12
5
0 10
−14
8
−5 6
Do −10 4
wn −16
til 2
−15 z)
t(
de 0 (MH
gr −2 ncy −18
ee −20 que
) −4 Fre
−25 −6
−8 −20
−30 −10
Rx—Receiver
Figure 10.
Relative Rx power over frequency and downtilt (positive tilt values indicate uptilts).
Figure 11.
Relative Rx for optimum tilt and for fixed reference tilt.
regions near the base station, it was evident that the [4]. In areas where there was a strong deviation of
optimal tilt was a perfect match with the geometric the geometric downtilt from the fixed downtilt, we
tilt. In regions shadowed by the rooftop, we observed observed receive power improvements of up to more
a receive power improvement of up to 15 dB here at than 10 dB.
UE position 11 in Figure 11 (10 dB typically). Note,
however, that in a real system this region would usu- Trials With Adaptive Beamforming Algorithms
ally be served by a neighbor cell. The trials described in the previous section used
For a typical drive route with a distance of 115 to vertical beamsteering with a narrow vertical HPBW.
280 meters from the base station with non-line-of- The beam pattern was shaped using an open-loop
sight conditions, the direct beam was blocked by two method.
tall buildings. Here the tilt for optimum receive Here, we consider trials that apply adaptive algo-
power varied between 2° and 0°. That meant the rithms to shape the transmit beams in a closed loop
optimum tilt was 4° to 6° higher than the fixed tilt of system. We used the same beamforming measure-
6°. The receive power could be improved by 0.5 dB ment setup as we had previously and used adaptive
to 1.5 dB. The main result of these vertical downtilt beam steering to two test mobiles. Based on the CSI
trials was that the receive power at the mobile could measured from the pilot signals at the UEs and
be further increased by using vertical beam steering reported to the eNB, a zero-forcing algorithm calcu-
under most conditions. The optimum tilt range lated the optimum complex beamforming weights
strongly depends on the selected environment. In for each subcarrier. This meant we could adapt both
mid-cell areas with a strong line of sight component, the phase and amplitude of the weights. Both lab
there is little improvement in receive power with trials and field trials were conducted using this
respect to the fixed downtilt because the fixed down- setup [9].
tilt was already optimal. For regions with strong For principal performance tests in the lab trial,
NLOS conditions, we obtained significant improve- the antenna array was mounted on a rack 1.8 meters
ment in areas where the direct beam was blocked by high. Two test UEs were mounted 4 meters apart on
tall buildings. In this case, downtilts deviate from two stages with a lateral distance of 20 centimeters.
geometric downtilts. The reflected paths were domi- Each UE could be moved 160 cm in the vertical direc-
nated by rooftop diffraction, therefore the optimum tion. This setup allowed us to examine the behavior
downtilts were lower than the geometric downtilts of the algorithm for different UE locations and
Rx power (dB)
−40 −40
−45 −45
−50 −50
−55 Rx UE0 −55
−60 Rx UE1 −60
Rx UE0
−65 −65
Rx UE1
−70 −70
−10 −8 −6 −4 −2 0 2 4 6 8 10 −10 −8 −6 −4 −2 0 2 4 6 8 10
Frequency (MHz) Frequency (MHz)
(a) Lab conditions (b) Field conditions
Rx—Receiver
UE—User equipment
Figure 12.
Separation of receive signals at the UE location.
vertical separations. When examining the behavior towards a street with almost LOS conditions. The
of the algorithm for different UE locations and verti- two test UEs were attached to a test van with one
cal separations, a vertical separation of only 30 cm receive antenna per UE mounted on the roof of the
already provided very good separation for the van. We tested the separation of wanted and
received signals. Both test UEs operated at 20 MHz. unwanted signals for the two test mobiles in different
Figure 12 provides a good visualization of the excel- scenarios and in test van positions both with and
lent separation of wanted and unwanted signals. The without LOS conditions.
representation for each UE was limited to 10 MHz, as To test a typical realistic scenario, the test van
shown in Figure 12a. The lower half of the spectrum was parked at a distance of about 30 meters from the
represents UE0 and the upper half of the spectrum rep- antenna array in an area where the direct beams
resents UE1. The signal of the unwanted UE was sup- were blocked by metal fences, metal garbage con-
pressed more than 35 dB. Similar results were found tainers, and trees. One antenna (UE0) was on the
in other situations (with larger vertical separations) roof of the van with NLOS conditions and the second
even with NLOS propagation conditions. Since this antenna (UE1) was mounted at street level about
setup was tested in the lab, reflections from the walls five meters away in a position with a strong LOS
and surfaces contributed to the nicely separable paths component.
to the individual UEs served by individually adapted To modify the reflection scenario, we modified
beam weights. the azimuth direction of the transmit beam pattern
Next, we took the measurement setup to the so the horizontal pattern pointed towards different
outdoor testbed used for the measurements reported parts of nearby buildings.
in the previous section. The antenna was mounted at As expected, strong reflection in the direction of
a height of 7 meters above street level, on a wall of the two test UEs led to a good suppression of the
the Bell Labs building. The beams were steered unwanted signal by about 15 dB for both mobiles, as