You are on page 1of 5

LTE-A Field Measurements:

8x8 MIMO and Carrier Aggregation

Karl Werner, Henrik Asplund, Björn Halvarsson, Anton K. Kathrein


Niklas Jaldén, and Daniel V.P. Figueiredo
Ericsson AB, Stockholm, Sweden Kathrein-Werke KG, Rosenheim, Germany
{karl.werner, henrik.asplund, bjorn.halvarsson, anton.k@kathrein.de
niklas.jalden, daniel.vp.figueiredo}@ericsson.com

Abstract—The introduction of 8x8 MIMO and carrier components of the system are common with the Ericsson radio
aggregation in the 3GPP LTE Rel. 10 opens up for increased user base station hardware product line for LTE. The system is
throughput. The potential gains using these techniques have been aligned with the LTE FDD mode of 3GPP LTE Rel. 10 (LTE
evaluated in a field measurement campaign with a testbed Advanced).
implementation. A downlink throughput exceeding 1 Gbps has
been achieved combining 8x8 MIMO in an outdoor macro
The testbed consists of one eNB and one user equipment
scenario with carrier aggregation using three component carriers (UE). The UE is scheduled for downlink transmission over the
(3x20 MHz). The relation between the achievable throughput and entire bandwidth with full buffers at a carrier frequency of 2.7
the channel richness arising from the physical environment and GHz. The system is configurable to operate with one or more
antenna spacing was demonstrated. The performance of MIMO component carriers of 20 MHz. The results in this paper where
setups ranging from 1x2 up to 8x8 was evaluated in indoor-to- carrier aggregation is applied use three 20 MHz carriers for
indoor, outdoor-to-indoor, and outdoor-to-outdoor deployments. transmission. The impact of interference is not covered in this
It was observed that each added transmit or receive antenna study. Regarding higher layers and control features, the testbed
increased the throughput. These gains were achieved with a is not aligned with the commercial requirements and therefore
compact UE antenna that is reasonable in size for
implementation in a consumer device.
absolute performance can not be extrapolated to be
representative of the products. In addition, the UE radio is
Index Terms— indoor deployments, mimo, carrier aggregation derived from an eNB radio, and, therefore, expected to have
higher performance than commercial UEs.
INTRODUCTION The eNB system supports up to eight antennas for downlink
transmission and the UE system supports up to eight receive
Multiple input multiple output (MIMO) systems have antennas with a linear MMSE receiver. The choice of the
attracted much attention as a key enabler for an increased codebook is transparent to the UE in TM9, because the
spectral efficiency. Several multi-antenna transmission modes reference symbols intended for demodulation (DM-RS) are
are present in the 3GPP Long Term Evolution (LTE) standard precoded together with the data. An open-loop spatial diversity
for mobile broadband in its release 10 [1], also known as LTE- scheme for MIMO has been implemented, in which the
Advanced. These modes permit the design of algorithms aimed precoders are cycled over frequency, with a fixed pattern. The
at improving the performance in a wide range of scenarios. codebook is not aligned with the standard, instead each layer is
Support for non-codebook based precoding and MIMO mapped to a single antenna; the mapping from layer to antenna
transmissions up to eight layers was introduced in LTE Rel. 10. is different for each precoder, and thus cycled over frequency.
Other downlink additional features in LTE Rel. 10 include In order to extract performance figures per rank, a fixed
Carrier Aggregation (CA) [1]. transmission rank choice pattern is used where each subframe
Previous work, covering field measurements of antenna is assigned a specific rank. The rank pattern is repeated with a
configurations up to 4×4 MIMO (number of transmit antennas cycle of 10 ms and link adaptation is performed per rank. This
× number of receive antennas) for HSPA and LTE Rel. 8, has setup makes the results less dependent on rank selection
been presented in [2,3]. In [4], an initial study presented gains algorithms and also allows comparison of different ranks.
of 8×8 MIMO for an indoor base station (eNB) with omni
directional antennas. In this paper, the understanding of 8×8 MEASUREMENT SETUP
MIMO setups is taken further to show practical gains with Herein we present measurements and evaluations from
3GPP LTE Rel. 10 with support for 8×8 MIMO in a variety of outdoor-to-outdoor (O2O) scenarios with a macro eNB as well
scenarios, both outdoor and indoor. as outdoor-to-indoor (O2I) and indoor-to-indoor (I2I) scenarios
TESTBED DESCRIPTION with a pico eNB. Further, to evaluate the influence of antenna
spacing on the system performance, the O2I and I2I
The results and measurements presented in this paper have measurements were repeated with both large and small antenna
been obtained with a testbed developed by Ericsson. The main separation at both the eNB and UE side. This paper focuses on

978-1-4673-6337-2/13/$31.00 ©2013 IEEE


Mbps @ 60MHz

Macro

Pico
Indoor

Fig. 1. Measurement environment. The white arrows indicate the location and pointing direction of the eNB
antennas. The yellow area indicates the corridor where the O2I and I2I measurements were collected. The colored
route is the outdoor measurement route. The color coding, legend on top left, indicates throughput on a
measurement drive using CA three carrier system (60 MHz DL).

the results using a compact UE with λ/2-spaced dual polarized more than 20λ across the antenna array. The three carriers used
patch antennas. Previously reported results [1] have shown that in the outdoor system are distributed over the antennas (note
this antenna configuration supports similar performance as a that each of the six radomes has four antenna ports) to
configuration with omni directional elements with larger maximize the inter element spacing within each carrier. In the
antenna spacing. two tx antenna case, two cross-polarized antennas in one
radome were used, in the four tx-antenna case cross polarized
Common setup for outdoor and indoor measurements
antennas in two adjacent radomes were used.
The measurements were performed in Kista, Stockholm,
Sweden. This area is characterized by 2-8 storey office 0.56
buildings with some open areas. Fig. 1 shows the locations and
pointing directions of the macro and pico sites, as well as
photos of the setup. In the O2O measurements the UE A B C D
equipment was placed inside a van with the antennas located
on the roof. In the indoor scenario the measurement equipment
was placed in a movable rack, pushed along the corridors,
while the antennas were carried by hand. For all outdoor 0.65 E
measurements, positioning was obtained through GPS. The
indoor locations were manually recorded in a map. For a
measure such as the median downlink throughput, over the
entire route, the difference was within 2% for repeated
measurements with the same parameters. This indicates good F G
repeatability of the results. The UE was configured to operate
with two, four, or eight antennas. The 2- or 4-element
subarrays were selected to get the minimal spatial separation,
i.e. one or two patches were used.
Fig. 2. Antenna arrangements for outdoor pico eNB. Note that different
Outdoor measurement setup combinations of the lettered antennas are used. Antenna footprints in meters.

The outdoor route as shown in Fig. 1 was planned such that


Indoor measurement setup
it included both line-of-sight (LoS) and non-line-of-sight
(NLoS) areas. The macro eNB had an output power of 20 W The O2I and I2I measurements were conducted in a
per carrier, used cross-polarized antennas with gain of 17 dBi. 15×70 m section of an office building. This building consists of
The array was down tilted by 2°. The antennas are well- a long corridor, marked in yellow in Fig. 1, with offices on
separated: minimum separation was 3λ, with a total distance of both sides. The inner walls are made of plaster and glass, while
Fig. 3. Throughput for two outdoor pico installations and two indoor pico Fig. 5. Median throughput for outdoor and indoor pico installations (same
installations on the indoor. as Fig. 3) on the indoor route and for the macro installation on the
outdoor route (see Fig. 1). The colored portions of the bars indicate
the outer walls are reinforced concrete and brick. One indoor performance with restrictions on maximum rank 1 (dark blue) to 7 (red).
measurement route consists of a walk through the full corridor The indoor and outdoor pico eNB alternatives cover the
and several offices, thus covering varying channel conditions measurement area in different ways. The outdoor-to-indoor
such as LoS and NLoS. The outdoor pico eNB was located solution has a better 10th percentile performance while the
65 m from the measured corridor, across the inner yard, see indoor-to-indoor pico solution has better peak performance.
Fig. 1, while the indoor pico was located in the end of the This effect is to be expected given the geometry of the
corridor. Both the outdoor and indoor pico eNB have been measurement area. Note that though the outdoor pico covers
equipped with dual polarized antennas with 7 dBi gain. Two several floors it also has higher output power.
antenna footprints were considered; the indoor pico antenna Based on the statistics in Fig. 3 it is also clear that the large
arrangements are shown in Fig. 4, and the outdoor pico and small square indoor antenna arrangements have very
arrangements are shown in Fig. 2. The inter element distance of different performance. The same applies to the square and
all considered setups are summarized in Tab. I. horizontal outdoor pico antenna arrangements.
TAB. I - ENB TRANSMIT POWER AND ANTENNA SEPARATION.
Fig. 5 shows throughput performance for the same test
cases as Fig. 3 and, additionally, for the macro installation on
Antenna type Vertical Horizontal Power the outdoor route. The colored subareas of the bars indicate the
separation (m) separation (m)
Small sq. indoor 0.21 0.13 10 mW
resulting throughput had the system been allowed to use only
Large sq. indoor 0.65 1.2 10 mW rank 1 (dark blue) only rank {1,2} (lighter blue), only rank
Large sq. outdoor 0.65 0.6 0.4 W {1,2,3} (cyan) etc.
Horiz. outdoor - 0.13 0.4 W It is clear for all scenarios that increasing the admissible
Outdoor macro - 0.3 20 W rank beyond six gives only small performance gains (at the
median). It is also clear that all setups have very large gains up
to rank three. The gains for rank four and beyond are large for
0.75 the indoor pico installation with large square antenna setup but
0.45 diminishing for the small square setup. Although not as
1.3 dramatic, this difference is also visible for the two outdoor pico
antenna setups: the horizontal arrangement has significant
0.27 gains only up to rank six.
In addition to sufficiently high SNR, a prerequisite for high
performance multi-stream transmission is a low singular value
(SV) spread of the (MIMO) channel. The singular values are a
Fig. 4. Large and small square indoor pico arrangements with antenna
footprint in meters. In the two tx antenna case only one radome was used. In measure of the relative power of the spatial subchannels that
the four tx antenna case the two topmost radomes were used. In the eight tx compose the MIMO channel.
antenna case, all four radomes were used. Fig. 6 shows SV spread for the MIMO channels measured
using four-antenna subarrays in the outdoor pico eNB and all
MEASUREMENT ANALYSIS the 8 rx antennas in the UE. The numbers are averages over the
entire measurement route and frequency band. The spread
Fig. 3 shows the throughput statistics of 8x8 MIMO for the
between SV one and two is similar across all included
indoor measurement route using the outdoor and indoor pico
subarrays; this is due to the use of cross-polarized antennas at
eNB setups in Tab. I. For each alternative, two eNB antenna
both ends. In contrast, the ratios of SV one to three and one to
arrangements are included differing in the physical positioning
four, respectively, differ much more between the subarrays. It
of the antennas, see Fig. 2 and Fig. 4.
is clear that horizontal separation between antennas gives lower
antennas) there is a consistent performance gain in every step.
Adding receive antennas improves performance at the lower
percentiles and also enables gains from addition of tx-antennas.
Adding tx-antennas improves peak performance. Note that in a
system operating in closed-loop mode, adding transmit
antennas may also enable beam-forming gains that improve
performance at the lower percentiles. The implementation used
in this measurement campaign operates in open-loop mode and
thus cannot utilize these possibilities.
Fig. 8 shows the same kind of performance measures for
the indoor pico with the small square antenna arrangement; in
this case there is a loss associated with going from a 4×8 to a
8×8 system! This is somewhat counterintuitive and warrant
further investigation.
Fig. 10 shows SV spread of 4×8 MIMO channels created
from four tx-antenna subarrays in the indoor pico eNB setup
similar to the analysis above for the outdoor pico. Below, in the
Fig. 6. Averaged singular value spread for 4x8 MIMO subchannels same figure, throughput performance for small and large
created from different subarrays of the outdoor pico eNB installation.
square antenna arrangement, respectively, is plotted on maps.
Close to the antennas, the SV spread is low for all
combinations; this is also reflected in very high performance in
terms of throughput for both the large and the small square
arrangements as visible on the maps. However, as the UE
moves further away from eNB antennas, SV spread of the four
antenna subarray with small horizontal spacing in particular,
but also the small vertical spacing increases.
The SV spread of the two subarrays with larger spacing is
much more stable. This is clearly reflected in the throughput
plotted. The high performance region in the measurements
using the large square setup covers a much larger portion of the
route. This observation may also explain the lower
performance of the small square arrangement compared to the
large square arrangement in Fig. 3.
In the indoor pico case with small separation of the
antennas, vertical spacing appears to give smaller SV spread
Fig. 7 Throughput statistics for MIMO transmission with increasing (left
than horizontal spacing when the UE is farther than
to right) number of antennas at transmitting and receiving end. Outdoor approximately 20 m from the eNB antennas. Returning to Fig.
pico eNB antenna installation is used with horizontal separation of the 8; the performance loss at median and 10th percentile between
antennas. In the two tx antenna case the two antennas in radome A were 4×8 and 8×8 is a manifestation of this effect.
used (see Fig. 2). In the four tx antenna case radomes A and B were used.
In the eight tx antenna case radomes A,B,C, and D were used.

SV spread than vertical. Note that the distance between the


antennas with the large vertical separation (radomes A and F)
is larger than the distance between the antennas with the small
horizontal separation. Even though increasing aperture size
does affect the SV spread, the horizontal/vertical orientation is
more important. This observation can explain the results
presented in Fig. 3 and Fig. 5: When covering an office
corridor from the outside, horizontal orientation of the antennas
gives better channel properties for higher order MIMO
compared to vertical orientation.
Fig. 7 shows performance statistics for UE/eNB antenna
aperture combinations going from a baseline 1×2 system up to
a full 8×8 system in the outdoor pico system. Similar to the Fig. 8. Throughput statistics for indoor pico antenna installation with
results presented in [1] (for indoor pico with omni directional small square arrangement of the antennas (see Fig. 4).
Fig. 9. Throughput statistics for outdoor macro antenna antenna setup on
outdoor route, single carrier system

significant throughput gains in each step. These gains were


largest for the indoor pico deployment but clearly visible also
in the outdoor pico and macro deployments. In fact, when
combining 8×8 MIMO in the macro scenario with Carrier
Aggregation—another feature introduced in LTE Rel. 10—an
impressive 1 Gbps downlink throughput was achieved using
3×20 MHz bandwidth. However, since these measurements
were performed under interference-free conditions these results
indicate the potential of the technology in well-isolated cells
rather than expected typical performance.
Fig. 10. Topmost plot: SV spread (here ratio of SV 1 to 3) as a function of
distance (projected to the horizontal axis) from eNB antennas for 4×8
The variations of the throughput as a function of used
MIMO subchannels using different antennas at the transmitter. Middle antenna configurations and the geographical area was studied.
plot: throughput on map (with matching horizontal axis scale) for 8×8 It was observed that the reachable throughput was strongly
system using small square antenna installation. Bottommost plot: connected to the channel richness in terms of the singular value
throughput on map for 8×8 system using large square antenna
installation. Numbers on the legend are in Mbps.
spread. This channel richness was demonstrated to be
connected to the physical environment as well as to the antenna
spacings and configurations at both transmitter and receiver.
In the open loop mode implemented in the system: All Nevertheless, good gains were achieved in all three
eight antennas are used also for one, two, three and four stream deployment scenarios even with compact antenna arrays,
transmission in the 8×8 setup while only four antennas will be particularly at the UE. This is very encouraging for the outlook
used for the same number of streams in the 4×8 setup. The 4×8 of going beyond 2×2 MIMO support in consumer devices
antenna subarray has horizontal separation, while the 8×8 array using the features of 3GPP LTE Rel. 10.
has both horizontal and vertical separation.
Fig. 9 shows performance for the outdoor measurement REFERENCES
campaign using the macro site. Similar to the indoor [1] E. Dahlman, S. Parkvall, J. Sköld, 4G LTE/LTE-Advanced for
measurements, there are significant MIMO gains from adding Mobile Broadband, Elsevier Ltd., 2011.
antennas. However the gains are somewhat lower, indicative of [2] M. Riback, S. Grant, G. Jöngren, T. Tynderfeldt, D. Cairns, T.
the less well-conditioned MIMO channel compared to the Fulghum, “MIMO-HSPA testbed performance measurements,”
indoor environment. Still, the aggregated throughput from the in proc. of IEEE 18th International Symposium on Personal,
three carriers had a peak that exceeded 1 Gbps and it was Indoor and Mobile Radio Communications (PIMRC 2007),
seldom below 400 Mbps as can be seen in Fig. 1. Athens, Greece, pp.1-5, Sep. 2007.
[3] K. Werner, J. Furuskog, M. Riback, B. Hagerman, “Antenna
CONCLUSIONS configurations for 4×4 MIMO in LTE - field measurements,” in
proc. of IEEE 71st Vehicular Technology Conference (VTC
The potential link performance gains of higher order
2010-Spring), Taipei, Taiwan, pp.1-5, May 2010.
MIMO up to 8×8 according to LTE Rel. 10, also known as
[4] K. Werner, H. Asplund, D.V.P. Figueiredo, N. Jaldén and B.
LTE-Advanced, have been evaluated in a field measurement
Halvarsson, “LTE-Advanced 8×8 MIMO Measurements in an
campaign using a testbed implementation. The studied Indoor Scenario” in proc. of International Symposium on
scenarios include macro, outdoor pico, and indoor pico Antennas and Propagation (ISAP-2012), Nagoya, Japan, Oct
installations. It was found that adding transmit and receive 2012.
antennas from 1×2 (Tx×Rx) MIMO up to 8×8 MIMO gave

You might also like