Addressing network capacity demands with LTE MIMO

Field trials of LTE with

44 MIMO
Multi-antenna reception and transmission is a key enabler of the high performance offered by LTE.

JOH A N F U RU S KO G, K A R L W E R N E R , M AT H I A S R I BAC K
A N D B O H AGE R M A N

The Long Term Evolution (LTE)
standard for mobile broadband
includes multi-antenna trans-
mission modes that improve
performance in terms of cover-
age, spectral efficiency and peak
throughput. To fully exploit the
potential of these techniques,
the antenna system at the LTE
base station, or eNodeB (eNB),
and in the user equipment (UE)
must be designed with the
intended performance profile in
mind. Here we present a com-
parison of the performance of
several multi-antenna setups in
field trials.
Multi-antenna technologies in LTE
Multiple antennas for reception and
transmission at the eNB and in UEs
are a key enabler of the high perfor-
mance offered by 3rd Generation
Partnership Project (3GPP) LTE.1 The
standard supports multi-antenna tech-
nologies that improve both link- and
system-level performance in a wide
range of scenarios.2 In LTE, data is
mapped to layers after having been
encoded and modulated. The num-
ber of transmitted layers is called the
transmission rank. 3GPP LTE Release 8
(Rel-8) permits up to four layers to be
transmitted in the downlink.2 These
layers are precoded and mapped to
antenna ports in a procedure that
is semi-statically congured to use
one of seven transmission modes (see
Table 1).3
Transmission modes 1 and 7 are iden-
tical from a UE perspective: a single
layer is transmitted in both cases. In
mode 1, the layer is transmitted from
one antenna port, whereas in mode 7,
the layer is transmitted from one anten-
na port, or transparently to the UE, a
combination of antenna ports.
In mode 2, a single layer is encod-
ed with a space-frequency block code
(SFBC) based on the Alamouti code, and
transmitted from multiple antennas.
Transmission mode 3 is identical to
mode 2 when the rank is one. For high-
er ranks, a predened codebook of
precoder matrices is cycled across the
frequency band, along with a layer
permutation designed to give each layer
similar average channel quality. Apart
from the rank indicator (number of
layers), no feedback is required to
select the precoders in modes 2 and
3. These modes are thus suitable in
scenarios where timely channel-
dependent feedback cannot be made
available for instance, in high-speed
scenarios.
Transmission mode 5 (multi-user
MIMO) enables single-layer transmis-
BOX B sion to several users who simultaneous-
Field Trials ly share the same frequency allocation.
The measure- In closed-loop spatial multiplexing
ment campaign (mode 4), one or more layers are trans-
was performed mitted using a precoder matrix that is
using the Ericsson selected from a predened codebook.
Research LTE The precoder selections are typically
test bed, which based on channel measurements made
comprised a by the UE. The precoders are applied
single mobile UE to the layers transmitted from the
and a single eNB. eNB with the goal of maximizing per-
The trials addres- formance according to the instanta-
sed downlink neous channel conditions. Depending
performance on the congured uplink feedback
only. mode, the UE may feed back
multiple preferred precoder matrix indi-
cators (PMIs), where each PMI is valid for
a particular sub-band (frequency-selec-
tive precoding); or
one PMI that is valid for all sub-bands
(wide-band precoding).
Transmission mode 6 is similar to trans-
mission mode 4, except that it is restrict-
ed to rank-one transmission.
The various transmission modes spec-
ied in the standard may be used with
any antenna setup, and the codebooks
are designed for a wide range of anten-
na setups and scenarios. Nevertheless,
the choice of antenna setup is critical,
and the best choice is dependent on the
target performance prole and propaga-
tion environment.

Field trial campaign

LTE test bed
BOX A Terms and abbreviations
The measurement campaign was per-
3GPP 3rd Generation Partnership Project LTE Long Term Evolution formed using the Ericsson Research
CDD cyclic delay diversity MIMO multiple input, multiple output LTE test bed, which comprised a single
CDF cumulative distribution function MMSE minimum mean square error mobile UE and a single eNB. The test
CQI channel-quality indicator PMI precoder matrix indicator bed supported up to 44 ([number of
eNB eNodeB, LTE base station SFBC space-frequency block code transmit antennas]x[number of receive
FDD frequency division duplex SIMO single input, multiple output antennas]) MIMO in the downlink. The
HARQ hybrid automatic repeat request SNR signal-to-noise ratio trials addressed downlink performance
HSPA high-speed packet access UE user equipment only, and the frequency bandwidth for

E R I C S S O N R E V I E W 1 2010
downlink transmission was set at
10MHz. FIGURE 1 The drive routes in Sector 1 (red) and Sector 2 (yellow) together with views of
The test bed incorporated the basic each sector as seen from behind the antenna installations.
LTE frequency division duplex (FDD) 2009 Europa Technologies Image 2009 Sanborn
principles and had similar parame-
ters.4-5 The relevant multi-antenna tech-
Sector 2
niques were aligned with 3GPP Rel-8,
and therefore it was anticipated that the
measured results would be representa-
tive of the relative throughput gains
achievable for LTE with multiple trans-
mit and receive antennas in the scenar-
ios considered.2
Table 2 shows some basic parameters
of the test bed. Given that the test bed
eNB
was a prototype, the absolute perfor-
mance is not necessarily representative
of commercial products.
The results were obtained using LTE
transmission mode 4. The codebooks
dened in the standard were used for Sector 1
frequency-selective precoding.2 Nine
precoders are selected over the 10MHz
bandwidth, each valid for a subset of the
band. The total downlink transmit pow-
er was normalized to be independent of
transmission rank and number of trans-
mit antennas.
200m
Field trials
An important goal of the eld trials
was to measure the relative perfor-
mance of different antenna setups at above rooftop height. The eNB antennas enclosures mounted on horizontal bars,
the eNB. A similar campaign for MIMO were prototypes developed by Ericsson enabling recongurable inter-antenna
high-speed packet access (MIMO-HSPA) specically for the measurement cam- spacing ranging from 8cm to 3m. The
was described in MIMO-HSPA Test Bed paign. Figure 1 shows the eNB anten- eNB was equipped with a dynamic
Performance Measurements.6 na arrangement in each sector. calibration system that ensured coher-
The eld trials were conducted in a Each eNB antenna enclosure housed ent transmission from the antennas.
business district in northern Stockholm, four linearly arranged columns with The UE antennas were mounted on
Sweden. The UE was driven along two a horizontal spacing of 0.7 (~8cm). the roof of a van, arranged in a square
separate routes in two sectors (Sector Each column consisted of a dual-polar- pattern with 20cm inter-antenna
1 and Sector 2) at speeds ranging from ized (+45 and -45 degrees), co-local- spacing. Two types of antennas were
5-30km/h. Figure 1 shows an aerial ized antenna pair. In total, there were used: horizontally polarized antennas
photograph of the area with drive routes eight antenna ports per enclosure. Each from SATIMO, and vertically polarized
drawn in. Sector 1 is suburban with at, sector had multiple separate antenna antennas from Kathrein. Both types
open areas and relatively low buildings.
Sector 2 is urban and occupied by ve-
to nine-oor ofce buildings on undu- Table 1: Transmission modes in 3GPP LTE Rel-8
lating terrain. In each sector, the maxi-
TRANSMISSION MODE DESCRIPTION
mum distance between the eNB and UE
was around 720m. Also, two power set- 1 Single-antenna port, port 0
tings were used in each: 2 Transmit diversity
34dBm and 18dBm in Sector 1; and
3 Large-delay CDD (open-loop spatial multiplexing)
34dBm and 24dBm in Sector 2.
4 Closed-loop spatial multiplexing
Antenna system conguration 5 Multi-user MIMO
The eNB site installation was designed 6 Closed-loop single-layer precoding
to imitate a conventional macro-
7 Single-antenna port, port 5
cellular site with antennas slightly

E R I C S S O N R E V I E W 1 2010
Addressing network capacity demands with LTE MIMO

were omnidirectional in the hori-

FIGURE 2 Selection of cumulative distribution functions (CDFs) of throughput achieved zontal plane.
during test drives along the measurement routes. Top left: Sector 2, low power setting.
Top right: Sector 2, high power setting. Bottom left: Sector 1, low power setting. Bottom Results
right: Sector 2, low power setting. Power gain versus multistream gain
Below we compare two categories of
CDF CDF antenna setups: those intended to cre-
1.0 1.0 ate a channel that is correlated at the
0.9 0.9 transmit side, and those intended to
0.8 0.8 create a channel that is uncorrelated at
0.7 0.7
the transmit side. The latter category is
typically well suited for multi-stream
0.6 0.6
transmission (spatial multiplexing) to
0.5 0.5
users with good channel conditions
0.4
[z] 1x2: SIMO
0.4 that is, to users near the cell center with
0.3 [b] 1x4: SIMO 0.3 [z] 1x4: SIMO a high signal-to-noise ratio (SNR). The
[k] 4x2: x x [k] 4x2: x x
0.2 [l] 4x2: //// 0.2 [l] 4x2: ////
former is best suited for beamforming
0.1
[c] 4x4: x x
0.1
[c] 4x4: x x toward users with bad channel condi-
[a] 4x4: //// [a] 4x4: ////
tions that is, to users near the cell edge
0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120 with a low SNR.
Throughput [Mbps] Throughput [Mbps] The correlation level at the transmit
side depends not only on the antenna
CDF CDF
setup but also on the propagation envi-
1.0 1.0
ronment. However, for simplicity, this
0.9 0.9
article denotes the two archetypical set-
0.8 0.8 up types: correlated and uncorrelated.
0.7 0.7 The correlated conguration was
0.6 0.6 obtained by having closely spaced trans-
0.5 0.5 mit antennas of equal polarization.
In the 44 MIMO example, the corre-
0.4 0.4
[z] 1x2: SIMO lated conguration is denoted with [a]
0.3 [b] 1x4: SIMO 0.3 [b] 1x4: SIMO
[k] 4x2: x x [j] 2x4: x in the gure; in the 42 MIMO exam-
0.2 [l] 4x2: //// 0.2 [l] 2x4: // ple, with [l]. Four co-polarized columns
[c] 4x4: x x [v] 4x4: xx
0.1 [a] 4x4: //// 0.1 were used in a single antenna enclosure,
0 0 resulting in an inter-element distance
0 20 40 60 80 100 120 0 20 40 60 80 100 120
of 0.7.
Throughput [Mbps] Throughput [Mbps]
The uncorrelated conguration was
obtained using two dual-polarized
columns with wide spacing. In the 44
MIMO example, the uncorrelated con-
guration is denoted with [c] in the
gure; in the 42 MIMO example, with
[k]. The inter-element distance between
the dual-polarized pairs was approxi-
Table 2: Parameters of the Ericsson Research LTE test bed mately 25.
In all four examples described above,
PARAMETER DESCRIPTION
the UE antenna conguration com-
Carrier frequency 2.6GHz prised antennas of both vertical and
System bandwidth 10MHz, UE always scheduled over entire bandwidth horizontal polarization two of each
kind for the four-antenna UE setup, and
Link adaptation, closed loop operation Rank, PMI and channel-quality indicator
one of each kind for the two-antenna
(3GPP mode 4) (CQI) feedback on millisecond timescale. Time
UE setup.
between channel measurement and transmission
of the resulting transport format is around 5ms
Figure 2 shows the performance of
the correlated and uncorrelated bench-
Hybrid automatic repeat request (HARQ) HARQ with chase combining mark congurations for drive tests in
Receiver Minimum mean square error (MMSE) receiver Sector 2 with the low power setting (top
LTE duplex mode FDD left), Sector 2 with the high power set-
ting (top right), and in Sector 1 with the
Number of antennas (downlink) eNB: 1, 2 or 4. UE: 2 or 4
low power setting (bottom left). For a

E R I C S S O N R E V I E W 1 2010
given number of receive antennas
in the UE, CDFs corresponding to FIGURE 3 CDFs of throughput in Sector 1 using the low power setting (left),
each benchmark eNB conguration and corresponding proportion of rank selections (right).
clearly intersect for all two- and some
four-antenna UE cases. None of the CDF Proportion
benchmark congurations consistent- 1.0 1.0

ly performed best during all parts of 0.9 0.9

any given route. The uncorrelated setup 0.8 0.8
performed best above the intersection 0.7 0.7
point, in the high throughput region,
0.6 0.6
while the correlated setup performed
0.5 0.5
best below the intersection point in
the low throughput region. The level 0.4 0.4

of the intersection point along the ver- 0.3 0.3

4x4: eNB: xx, UE:|-|-
tical axis gives the relative proportion 0.2 4x4: eNB: xx, UE:|||| 0.2
of how often one setup outperformed 0.1
2x2: eNB: x, UE:|-
0.1
2x2: eNB: x, UE:||
another along a drive route. The inter- 0 0
section point is dependent on the 0 10 20 30 40 50 60 70 80 eNB: xx eNB: xx eNB: x eNB: x
UE:|-|- UE:|||| UE: |- UE: ||
scenario. Throughput [Mbps]
Rank 1 Rank 3
In general, CDFs that correspond to Rank 2 Rank 4
test drives with the low power setting
intersect at higher points than those
for test drives with the high power set-
ting. This may be attributed to the fact
that, given single-layer transmission, antennas, multi-antenna transmis- but includes setups with two transmit
increased received power gives UEs sion using the correlated bench- antennas. Clearly, the closely spaced
operating with low receive power and mark eNB setup outperforms single- co-polarized setup [i] performs better
a low SNR a more signicant increase antenna transmission (single input, than the dual-polarized setup [j] for
in throughput than UEs operating with multiple output or SIMO) in every case. users with the least favorable chan-
high receive power and a high SNR. The In most cases, the uncorrelated setup nel conditions. On the other hand, the
correlated setup gives a larger power shows a gain over SIMO, and in some dual-polarized setup performs bet-
gain than the uncorrelated setup. In the cases it dramatically improves through- ter for users with better channel con-
higher power region, it is usually much put. Due to the larger reference signal ditions. The results also suggest that
more benecial to transmit multiple lay- overhead in multi-antenna transmis- a four-antenna eNB setup with two
ers, and the uncorrelated setup creates sion, SIMO occasionally outperforms closely spaced dual-polarized pairs [v]
channels that better support this kind the uncorrelated setup. captures the benets of both the two-
of transmission. For similar reasons, the Figure 2 (bottom right) shows results antenna eNB setups shown. One oth-
drive routes in the more challenging in Sector 2 for the low power setting er visible benet of that setup for
Sector 2 produced curves that intersect
at higher points than those of the more
open Sector 1. FIGURE 4 CDFs of throughput at cell center (left) and at cell edge (right).
The intersection point for CDFs
that correspond to the UE congura- CDF CDF
1.0 1.0
tion with four antennas was lower
than those that correspond to the UE 0.9 0.9

conguration with two antennas. In a 0.8 0.8

relative sense, due to its lesser ability to 0.7 0.7
improve SNR by coherently combining 0.6 0.6
the signals received by multiple anten- 0.5 0.5
50% increase 113% increase
nas, the two-antenna UE setup bene- at median at median
0.4 0.4
ts more from the increase in received
0.3 0.3
power provided by the correlated setup.
By contrast, the four-antenna UE bene- 0.2 0.2
4x4: MIMO 4x4: MIMO
ts more from the uncorrelated setup, 0.1
2x2: MIMO
0.1
2x2: MIMO
which improves the ability of the chan- 0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
nel to support multiple layers.
Throughput [Mbps] Throughput [Mbps]
Increasing the number of receive
antennas always improves perfor-
mance. For a given number of receive

E R I C S S O N R E V I E W 1 2010
Vinjett
Addressing network capacity demands with LTE MIMO
FIGURE 5 The selected parts of the drive route in Sector 2 with a
picture of the eNB antenna installation overlaid with schematic
drawings of the antenna setups. 2009 Europa Technologies Image
2009 Sanborn 2009 Google Gray buildings 2008 Sanborn
Cell-edge
Use-case: Low power
a four-antenna UE is the ability to
transmit more than two layers when
channel conditions allow it.
Polarization-matching aspects
The UE antenna congurations de-
scribed thus far were composed of
purely vertically and horizontally polar-
ized antennas. Such congurations are
well matched to eNB antenna congura-
tions and transmission modes that rely
on polarization diversity for example,
multi-stream transmission with dual-
polarized antennas. Note, however, that
perfectly orthogonal polarizations are
hard to obtain in practical UE imple-
mentations. The ability to support multi-
stream transmission using polarization
diversity diminishes the more similar
the UE antennas are to each other in
polarization. To illustrate this principle,
tests were performed for both 44 and
22 MIMO using exclusively vertically
polarized antennas in the UE.
Figure 3 (left) shows the resulting
CDFs. In terms of throughput, perfor-
mance decreases in the high-through-
put region with an unmatched UE
antenna conguration, due to the
reduced ability to sustain multi-
layer transmission. This is also evident
E R I C S S O N R E V I E W 1 2010
increasing capacity demands:
Cell-centre
Use-case: High power
eNB
from the bar charts in Figure 3 (right),
which show the transmission-rank
proportion. The CDFs also suggest
that the opposite holds true in the
low-throughput region, where single-
layer transmission dominates. This
effect may be ascribed to the ability
of an eNB with dual-polarized anten-
nas to match the effective polariza-
tion of the transmission through prop-
er choice of precoder. The polarization
can be matched when the UE has anten-
nas with parallel polarization, but not
when the polarization is mixed as in the
orthogonal case. The intersection point
for the 44 setups is different from that
of the 22 setups. The gain from match-
ing polarization is mainly one of SNR;
the CDFs for the 22 setups intersect
at a higher point than those for the 44
setups, which is in line with the results
reported above.
LTE mobile broadband now
and tomorrow
The rst commercial LTE networks
are currently being launched. Multi-
antenna technology is thus already
a vital component in the evolution
of mobile broadband. And its impor-
tance will continue to increase as user
trafc grows and puts greater demands
on network capacity. Two use cases
exemplify the potential of LTE multi-
antenna technology to meet these
a user close to the cell center, limited by
cell capacity and peak throughput; and
a user close to the cell edge, limited by
network coverage.
Figure 4 (left) shows the rst use case.
Those parts of the Sector 2 measure-
ment route that primarily support the
transmission of multiple streams were
selected with the transmit power set to
high. A typical 22 MIMO conguration
(which represents the rst generation
of multi-antenna-capable LTE products)
is presented together with a 44 MIMO
system that uses the uncorrelated trans-
mit antenna setup.
The second use case is illustrated in
Figure 4 (right), comprising the parts of
Sector 2 that are closer to the cell edge,
with the transmit power set to low. A
44 MIMO setup using the correlated
transmit antenna conguration is pre-
sented together with the reference used
in the rst case.
Figure 5 shows the selected parts of
the drive route in Sector 2. Note that
3GPP LTE Rel-8 supports up to twice
the frequency bandwidth used in the
measurements. For the cell center case, a
50 percent increase in median through-
put was measured; for the cell edge
case, the median throughput was more
than doubled. These examples clear-
ly illustrate the potential of the multi-
antenna technologies in LTE Rel-8 to
appease growing demands on network
capacity.
Johan Furuskog
joined Ericsson in 2007
after graduating from
Uppsala University in
Sweden with an MSc in
engineering physics. His work as
a research engineer at Ericsson
Research in Kista, Sweden, has mainly
concerned test-bed development and
field trials with focus on MIMO channel
characteristics and LTE multi-antenna
performance. Johan is currently
involved in different test-bed projects,
at both Ericsson Research and
Business Unit Networks, which target
the design and deployment of future
systems for radio access.
Mathias Riback
who joined Ericsson in
2004, is currently project
manager for a future
access test-bed at
Business Unit Networks, Kista,
Sweden. Before this he worked as a
senior research engineer at Ericsson
Research with the main focus on radio
propagation research and test-bed
development for both HSPA and LTE
with emphasis on multi-antenna
aspects. Mathias holds an MSc in
electrical engineering from the
Royal Institute of Technology,
Stockholm, Sweden.
Bo Hagerman
received an MSc EE, Lic.
Tech. EE and PhD in Radio
Communication Systems
from the Royal Institute of
Technology, Stockholm, Sweden in 1987,
1993 and 1995, respectively. From 1987
to 1990, he was a member of the
technical staff in the Ericsson Radio
Systems Research and Development
department, where he worked in the
area of signal processing with applica-
tions to GSM receivers. He joined the
Radio Access Technologies Research
department at Ericsson Research,
Stockholm in 1995, where he is currently
a Senior Specialist in the area of
Advanced Antenna Systems, working
with research on multiple antenna
systems and heterogeneous deploy-
ments in cellular networks.
Karl Werner
joined Ericsson in 2007.
As a research engineer at
radio access technologies
in Stockholm, Sweden, he
has mainly worked with test-bed
implementation and with field
measurement campaigns targeting
performance of antenna systems for
LTE. He holds a PhD in Signal Processing
and an MSc in computer engineering
from the Royal Institute of Technology
in Stockholm, Sweden, and Lund
University, Lund, Sweden, respectively.

References

1. Dahlman, E. Parkvall, S. Skld, J. & Beming, P., 2007, 3G

Evolution: HSPA and LTE for Mobile Broadband. 2nd ed, Oxford, Academic Press
2. 3GPP TS 36.211 3rd Generation Partnership Project, Technical Specification
Group Radio Access Network: Evolved Universal Terrestrial Radio Access
(E-UTRA), Physical Channels and Modulation. Release 8
3. 3GPP TS 36.213 3rd Generation Partnership Project, Technical Specification
Group Radio Access Network: Evolved Universal Terrestrial Radio Access
(E-UTRA); Physical Layer Procedures. Release 8
4. Riback, M. & Karlsson, J., 2008. Initial Performance Measurements of LTE.
Ericsson Review, Vol. 85 (1), pp.22-28
5. Johansson, B. & Sundin, T., 2007. LTE test bed. Ericsson Review,
Vol. 84 (1), pp.913
6. Riback, M., Grant, S., Jongren, G., Tynderfeldt, T., Cairns, D. & Fulghum, T., 2007:
MIMO-HSPA Testbed Performance Measurements in proceedings of PIMRC,
Athens

E R I C S S O N R E V I E W 1 2010