3G LTE Simulations Using Measured MIMO Channels

Abstract

In this article we present downlink simulation results for a realistic implementation of the LTE (Long Term Evolution) 3G standard. In contrast to previous studies, actual measured channels (as opposed to computer generated artificial channels) have been used in the simulation. The used 2 times 2 MIMO channels were measured using two realistic receiver mockups, one laptop and one handset, as well as a pair of reference dipole antennas. The results suggest that LTE is able in practice to support multi stream transmission with very high data rates, even for small hand held terminals. Also, the improvements of 2 times 2 MIMO over SISO transmission are clearly shown.

Full text

Page 1

3G LTE Simulations Using Measured MIMO
Channels
Yngve Sel´ en and Henrik Asplund
Ericsson Research
Ericsson AB, Isafjordsgatan 14E, SE-164 80 Stockholm, Sweden.
Email: <first name>.<last name>@ericsson.com
Abstract—In this article we present downlink simulation re-
sults for a realistic implementation of the LTE (Long Term
Evolution) 3G standard. In contrast to previous studies, actual
measured channels (as opposed to computer generated artificial
channels) have been used in the simulation. The used 2×2 MIMO
channels were measured using two realistic receiver mockups,
one laptop and one handset, as well as a pair of reference dipole
antennas. The results suggest that LTE is able in practice to
support multi stream transmission with very high data rates,
even for small hand held terminals. Also, the improvements of
2 × 2 MIMO over SISO transmission are clearly shown.
I. INTRODUCTION
The use of mobile broadband has shown a tremendous
increase recently: the portion of packet data traffic in the
WCDMA networks surpassed that of voice in spring 2007
[1] and the increase in packet data traffic is expected to
continue as the prices of HSPA subscriptions drop and as the
coverage increases. This puts high pressure on operators to
increase the capacities of their networks, and on the industry
for enabling such an increase also in the long term future
via more efficient and flexible communication standards. LTE
(Long Term Evolution) [2] is one track of 3G evolution, which
is currently being standardized within the Third Generation
Partnership Project (3GPP). The technique is based on orthog-
onal frequency division multiplexing (OFDM) in the downlink
and single carrier frequency division multiple access (SC-
FDMA) in the uplink. LTE offers several important benefits
both for operators and end-users. Among the most important
are [1], [2]
• High throughput: LTE allows for peak throughput rates
above 200 Mbps, and LTE peak rates of up to 160 Mbps
have already been demonstrated in experimental systems
[3].
• Low latency: The latency requirements are much tighter
in LTE than in WCDMA, with radio access network
round trip times below 10 ms.
• Flexibility: LTE supports both the frequency division
duplex (FDD) and time division duplex (TDD) modes on
the same base station platform [3]. Also, LTE supports
a greater bandwidth flexibility, from below 5 MHz up to
20 MHz, than WCDMA.
The goal of the present article is to demonstrate the link
performance that can be expected from LTE, and to show that
MIMO transmission with multiple data streams will indeed be
a reality in the LTE networks, also for small handsets. To this
end, we present simulations of LTE downlink transmission
for a single user occupying 10 MHz of bandwidth in the
2.66 GHz band.1The novelty of the present simulation study
is that actual measured 2×2 MIMO channels for some realistic
antenna setups have been used. Also, the simulator resembles
a realistic implementation which includes channel estimation
errors, realistic link adaptation, error vector magnitude (EVM)
etc. (this is further described in Section III). This allows stud-
ies of how the expected link performance of an actual deployed
system can be expected to vary with different settings.
The results, presented in Section III-C, show that, under
realistic conditions, LTE will be able to support very high
data throughput as well as multi stream transmission. This
conclusion holds also for user equipments (UEs) with small
form factors, such as handsets and laptops.
II. CHANNEL MEASUREMENTS
The equipment used for channel measurements consists
of a base station (BS) located on the roof of Ericsson’s
headquarters in Kista, Stockholm, Sweden, and a UE mounted
inside a measurement van. The BS transmits pilot symbols
over a bandwidth of 20 MHz at a carrier frequency of 2.66
GHz. These are measured by the UE and are used to estimate
the MIMO channel. The UE continuously logs the channel,
at a rate high enough to sample the channel more than twice
per wavelength, as the van drives along a pre-defined route;
see, e.g., Figure 7 (the map covers an area of about 800 by
850 m). The route includes parts with line-of-sight as well
as highly shadowed areas. The GPS positions of the UE are
logged simultaneously with the channels, so that each channel
measurement can be associated with a geographical position.
This procedure is then repeated for different antenna setups
on the BS and the UE sides.
A. Antenna Configurations
We present results for three different antenna setups. On
the TX side (base station) a pair of cross polarized (±45◦)
antennas with a common phase center was used. This enables
a very compact antenna installation, see Figure 1. On the RX
side the following receiver antenna setups were used:
1This is in operating band VII evaluated for UTRA/FDD [4], and corre-
sponds to a wavelength λ of about 11 cm.

Page 2

Fig. 1.
in the figure.
The base station, with the used dual-polarized TX antenna marked
Fig. 2. Configuration A: Reference antennas on the roof of the measurement
van. The used RX antennas have been marked in the figure.
A) Reference antennas: One electric dipole (horizontal po-
larization) and one magnetic dipole (vertical polariza-
tion) placed on the roof of the measurement van with an
antenna distance of about 1 m (corresponding to about
9λ). See Figure 2.
B) Handset mockup: A custom modified Sony Ericsson
K800i handset with two internal receiver antennas. See
Figure 3(a). The antenna separation is about 9 cm, which
is slightly below the carrier wavelength λ = 11 cm.
The handset was placed inside the measurement van
during the channel measurement. This gave an additional
pathloss of about 10 dB, as compared to the reference
antenna placement on the roof.
(a) Configura-
tion B: Handset
mockup.
(b) Configuration C: Lap-
top mockup.
Fig. 3.Realistic mockups with the two used RX antennas marked.
02468 1012 141618
−50
−40
−30
−20
−10
0
10
Time delay [μs]
Relative power [dB]
Fig. 4.
dynamic range of the measurements.
Example of a measured power-delay profile showing the very high
C) Laptop mockup: A Hewlett-Packard laptop with RX an-
tennas on the top of the screen (built in) and on the sides
of the screen (added, custom built). See Figure 3(b). The
two RX antennas used (one on the top and one on the
side of the screen) are marked in the figure. The antenna
spacing is about 15 cm, which corresponds to about
1.4λ. The laptop was placed inside the measurement van
during the channel measurement.
Both mockups were positioned centrally in the measurement
van with no disturbing objects, such as persons or seats,
in the near vicinity of the antennas. Note that the above
configurations have 2 TX and 2 RX antennas, i.e., we study
2 × 2 MIMO transmission.
B. Channel Characteristics
The measured channels were evaluated to determine the
measurement SNR. As can be seen in Figure 4 the dynamic
range in the measured impulse responses is very large. A high
SNR in the measurements is very important when later uti-
lizing the measured channels in simulations, as measurement
noise can potentially change the correlation properties of a
MIMO channel.
For simulation purposes it is also important to sample the
channel adequately in time and frequency. An example of the
time-frequency behavior of the channel for one pair of transmit
and receive antennas is shown in Figure 5. As the speed
of the van varied due to traffic the resolution of the spatial
sampling was also different during different segments of the
measurement routes. At all times the speed was sufficiently
low to ensure Nyquist sampling, e.g., at least two channel
samples per traveled wavelength.
One of the key properties for characterizing MIMO per-
formance is the correlation of the fast fading experienced for
different selections of transmit and/or receive antennas. Figure
6 shows the estimated receive antenna correlation for the three
configurations used at the terminal. In addition, the estimated
transmit correlation between the dual-polarized antennas at the
BS is also plotted. The correlations were estimated using the

Page 3

Fig. 5.
bandwidth and during a 2 s long segment.
Example of the fast fading in time and frequency over a 20 MHz
0 0.20.4
Correlation, |ρc|
0.60.81
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
CDF


Configuration A (ref. antennas)
Configuration B (handset)
Configuration C (laptop)
TX correlation
Fig. 6.
receive antennas for the three different UE antenna configurations. Also, the
correlation between the two transmit antennas is shown.
Cumulative distribution functions for the correlation between
measured fast fading coefficients over blocks of 20 MHz × 1 s,
where the one second segment corresponds to between 0
and 10 m due to the varying speed of the van. No effort
was made to separate shadowing from fast fading as it was
assumed that the influence on the correlation estimates would
be minor. As is evident from the figure, the correlations
between the antennas are low for all the configurations, both
at the transmitter and at the receiver. However, the reference
configuration shows a tendency of having somewhat higher
correlation values in the upper tail of the distribution. This is
probably a result of the scatterer-free mounting on the vehicle
roof which may lead to higher correlation in the case of line-
of-sight, in contrast to the mockups that were placed inside the
van where more polarization cross-scattering can be expected.
III. SIMULATIONS
A. Stored Channels
Link simulations are typically done using computer-
generated, stochastically modeled channels. Most channel


dB
−10
0
10
20
30
40
50
Fig. 7.
settings and Configuration C (laptop). The x-mark shows the location of the
BS. Note that the effect of the EVM, which limits the SNR (see Section III-B)
is not included in this plot.
Channel SNR (in dB) over various locations for the used simulation
−1001020 304050 60
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SNR (dB)
CDF


Configuration A (ref. antennas)
Configuration B (handset)
Configuration C (laptop)
EVM 4%
Fig. 8.
dB. The SNR limiting EVM at 4% is also marked in the figure.
Cumulative distribution functions (CDFs) of the channel SNRs in
models have various levels of simplification which, while
beneficial from a complexity point of view, often result in
certain properties not being captured by the model. Among
those are primarily the variability and dynamics that occur
under real-life conditions.
By the above reasoning, link simulations using stored chan-
nel data can be a good complement to traditional simulations,
especially as a tool for validating that the design choices made
during development of a system are sufficiently robust against
real-life channel conditions.
Simulations using stored channel data can be quite chal-
lenging to perform due to the large data sets required to store
the channels over km-long routes. The approach used herein
was to replace the stochastic channel generation module in the
link simulator with a module that read a segment of channel
impulse responses from file, interpolated the channel data
to symbol-rate and scaled the output, and finally shortened
the impulse responses by discarding delay samples containing
only noise. One benefit of using playback of measured chan-

Page 4

nels is that the user speed can be artificially varied to any
desired value. In the simulations presented below a pedestrian
speed (up to 1 m/s) has been used.
B. Simulation Setup
We simulated downlink transmission to a single user in
an interference free environment over numerous one second
segments along the measurement route, each represented by a
dot on the throughput maps in Figures 9-11. This was done
for each of the antenna settings described in Section II-A.
The UE utilized 10 MHz bandwidth in the 2.66 GHz band.
Link adaptation was enabled, whereby the UE could select
between 15 combinations of code rate (from 0.1 up to 0.93)
and modulation type (QPSK, 16QAM and 64QAM), and also
select the number of simultaneous data streams (one or two
for 2 × 2 MIMO). The link adaptation was based on realistic
estimation of the current signal to interference plus noise ratio
(SINR) and also the channel estimation algorithm was realistic.
An EVM of 4% (limiting the SNR to a maximum of 28.0 dB)
was assumed at the transmitter. The simulator used a linear
MMSE receiver and an LTE turbo codec with hybrid ARQ
allowing 3 retransmissions. After all control signaling and all
pilot symbols have been accounted for, the maximum possible
throughput available to the user was about 67 Mbps for the
selected settings.
White Gaussian noise, with constant variance over the mea-
surement route, was added to the received signal. This additive
noise dominated over the channel measurement noise for all
presented results. The noise level was the same for all antenna
configurations, and so was the transmit power. Their respective
values were set in the simulator such that the received SNR
became as shown in Figure 7 for Configuration C (laptop).
For Configuration A (reference antennas) the SNR was about
10 dB higher, and for Configuration B (handset) it was about
5 dB lower; see Figure 8. These differences were due to
lower pathloss for Configuration A (the RX antennas were
placed outside the measurement van) and different antenna
efficiencies.
C. Simulation Results
In Figure 9 the throughput values for Configuration A
(reference antennas) are shown. The throughput was often
high, also at locations where the UE did not have direct line
of sight to the BS (the throughput varied between 10 and
67 Mbps, with a mean value of 48 Mbps). For 85% of all
simulation points, the mean number of data streams was above
1.5. This can to a great extent be explained by the relatively
high SNR; above 20 dB in most locations (see Figure 7). For
the other antenna configurations, the SNR was significantly
lower due to the fact that the RX antennas were located inside
the measurement van and not outside on the roof.
In Figure 10, the throughput values for Configuration B
(handset) are shown. They vary between 1 and 63 Mbps with
a mean value of 26 Mbps. Furthermore, the mean number of
parallel data streams was above 1.5 for as many as 68% of


Mbps
10
20
30
40
50
60
Fig. 9. Throughput map for Configuration A (reference antennas). The mean
throughput is 48 Mbps.


Mbps
10
20
30
40
50
60
Fig. 10.
put is 26 Mbps.
Throughput map for Configuration B (handset). The mean through-
all simulation points, i.e., two data streams were selected in a
majority of the cases.
In Figure 11 we show the throughput values for Configu-
ration C (laptop). They vary between 3 and 66 Mbps, with a
mean value of 35 Mbps. For 85% of all simulation points, the
mean number of streams was above 1.5.
We also show, in Figure 12, cumulative distribution func-
tions (CDFs) of the obtained throughput values. This presen-
tation makes it easier to directly compare the performances
for the various receiver antenna configurations, and it is easy
to appreciate the higher performance of the reference antennas
from this figure. Also, the laptop mockup was able to perform
better than the handset mockup. However, these differences
stem, to a large extent, from differences in the SNR due to
different pathloss and RX antenna gains. A SISO case with
reference antennas, corresponding to Configuration A, has also
been included in the plot. Here, the same transmit power as
for the 2 × 2 MIMO case was used on a single transmitter

Page 5

Mbps
10
20
30
40
50
60
Fig. 11. Throughput map for Configuration C (laptop). The mean throughput
is 35 Mbps.
0102030 40 5060 70
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
throughput (Mbps)
CDF


Configuration A (ref. antennas)
Configuration B (handset)
Configuration C (laptop)
SISO case (ref. antennas)
Fig. 12.
to Configuration A has also been included.
CDF for the throughput measurements. A SISO case corresponding
0102030405060 70
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
throughput (Mbps)
CDF


Configuration A (ref. antennas)
Configuration B (handset)
Configuration C (laptop)
SISO case (ref. antennas)
Fig. 13. CDF of the throughput measurements when the TX power has been
varied such that all configurations experience the same SNR as in Figure 7.
antenna, and the vertically polarized antenna on the roof of the
measurement van was used at the receiver side (the leftmost
antenna in Figure 2). For the SISO case, only one data stream
can be transmitted, and it is clear that 2 × 2 MIMO, with
its multi stream capacity, gives a significant improvement of
the performance. Note that the SISO curve has been obtained
using the reference antennas. Therefore, its SNR is higher than
that of the configurations B and C, which explains why it has
fewer throughput measurements at the lower end.
In Figure 13 the TX power has been scaled differently
for the different antenna setups, so that they all experience
approximately the same SNR (that which is shown in Figure
7).2The performance differences are now much smaller. This
result indicates that the antenna correlations are sufficiently
low for all the studied configurations, and that the antenna
efficiency is what distinguishes the different antenna setups
when it comes to their throughput and multi-stream potential.
IV. CONCLUSIONS
We have performed simulations of a realistic 2 × 2 MIMO
downlink transmission for a single LTE UE in an interference
free environment. The simulations used actual measured chan-
nel data for different receiver antenna configurations, including
realistic mockups (a handset and a laptop). The results show
that all configurations are able to support high data throughput
as well as multi-stream transmission. The main performance
differences were also shown to depend on the experienced
SNR (which is different due to different path loss and different
RX antenna efficiencies) while the antenna correlations were
sufficiently low for all configurations.
The simulation study strengthens the belief that multi stream
transmission will be a realistic method for increasing the data
throughput in LTE systems, also for small hand held devices.
ACKNOWLEDGMENTS
The authors would like to thank Thomas Bolin at Sony
Ericsson Mobile Communications AB in Lund, Sweden, for
supplying the modified handset (antenna configuration B), and
Anders Derneryd at Ericsson Research in Gothenburg, Swe-
den, for supplying the laptop mockup (antenna configuration
C). Also, thanks goes to Johan Furuskog at Ericsson Research,
Kista, Stockholm, Sweden, for performing the channel mea-
surements.
REFERENCES
[1] Ericsson AB, “Long term evolution (LTE): an introduction.” online:
http://www.ericsson.com/technology/whitepapers/-
lte_overview.pdf, October 2007. White Paper.
[2] E. Dahlman, S. Parkvall, J. Sk¨ old, and P. Beming, 3G Evolution: HSPA
and LTE for Mobile Broadband. Oxford, UK: Academic Press, 2007.
[3] Ericsson AB, “Ericsson first to demonstrate LTE in both FDD
and TDD modes on the same base station platform.” online:
http://www.ericsson.com/ericsson/press/releases/-
20080130-1186619.shtml, January 2008. Press release.
[4] 3GPP, “Universal mobile telecommunications system (UMTS) base sta-
tion (BS) radio transmission and reception (FDD),” January 2008. 3GPP
TS 25.104 version 8.1.0 Release 8.
2In the previously shown results, the TX power was the same for all
configurations, whereas the SNR varied due to different path loss and antenna
gains for the different configurations.