LTE Drive Test PDF

Signals Ahead
LTE Drive Test - Part 1

(Stockholm)
Redefining Research
Volume 6, No. 3
March 12, 2010

Michael W. Thelander
(510) 338 1284
[email protected]
www.signalsresearch.com
Subscription information
Signals Ahead is published 18 times
per year and is only be available to our
paid subscribers. For our corporate
customers, we have established the
following rate structure.
Corporate rates
Group license1. ......................... $3,995
Global license........................... $7,995
Platinum package2 ................... $9,495
Payment options
To subscribe to Signals Ahead,
please fill out the form on the last
page of this issue and return it to us
or contact us via email at [email protected] and we will respond
to your inquiry. This process is also
automated on our web site at
www.signalsresearch.com.
Once payment is received, we will
notify you of your user account
information. We accept checks and all
major credit cards and can create an
invoice upon request.
Terms and conditions
Any copying, redistributing, or republishing of this material, including
unauthorized sharing of user accounts,
is strictly prohibited without the written consent of SRG.
Up to five (5) unique users from the

same corporation.
2
The platinum package includes five
(5) hours of analyst time during the
subscription period.
uring the week of February 8th we conducted what is arguably the most scientific
and exhaustive independent performance analysis ever conducted of an LTE
network(s). During our tests, which were almost entirely conducted in vehicular
or pedestrian mode, we transferred nearly 186GB of data, or nearly twenty-five times
more data usage than what a typical fixed broadband user consumes in an entire month.
While we looked at some of the more basic performance KPIs, such as downlink and
uplink data rates, we leveraged a sophisticated drive test tool to capture and analyze
numerous underlying performance KPIs, including CINR, RSSI, resource block allocation,
modulation scheme (Antenna 1 and Antenna 2), MIMO type, and BLER. Further, during
the post-processing phase we analyzed these KPIs by several different means, including,
throughput versus CINR, CINR versus RSSI, handover success rate, throughput during a
handover, and CINR versus modulation scheme, to name a few.
As part of this research project we obtained the XCAL LTE drive test tool and XCAP LTE
post-processing tool from Accuver. To the best of our knowledge the XCAL/XCAP solution
was the only solution that was available which could interface with the Samsung and LG
LTE devices, not to mention other vendors HSPA+ or Mobile WiMAX chipsets, and this
capability proved invaluable during this entire effort. With the use of this tool and our
methodical approach to testing and analyzing the network, we have arguably reached
fairly conclusive and statistically valid opinions about how LTE performs today.
Overall, the size of the two networks was better than we anticipated, although both
networks are small in comparison to other networks that we have tested while coverage
gaps still exist in some areas. And while our in-building testing was somewhat limited, from
what we observed the in-building performance of an LTE network deployed at 2600MHz
can be quite good as long as there is great coverage to the exterior of the building.
In terms of the actual performance we recorded an adjusted average downlink PHY
Layer throughput of 16.8Mbps (peak of nearly 50Mbps) in the Stockholm network and
32.1Mbps (peak of nearly 85Mbps) in the Oslo network. Had we used a less stringent test
methodology or gone looking for peak data rates both KPIs in the two networks would
have been higher.
Despite throughput numbers that rival many fixed broadband services and an overall
performance that we believe was astounding, we also identified areas for improvement
in the two networks. In particular, once network optimization occurs we would expect
a material improvement in the cell handover success rate, not to mention the actual
throughput itself, as both KPIs are impacted by the presence of high interference from
neighboring cells.
In addition to presenting the results and comparing and contrasting the performance
of the two networks, we put things into perspective and discuss what these capabilities
mean for the typical end user. The appendix at the end of the report contains additional
figures which didnt make their way into the main report.
ffective immediately, any unauthorized use of our

research material will result in the non-refundable
cancellation of your subscription. We also reserve the
right to post your companys name, with logo, to the SRG
Wall of Shame. If you received this issue from someone
outside of your organization and it did not come directly
from SRG then the licensing terms for our research are being
violated. If you forward this research to external organizations, either in whole or in part, or if you share the contents
of the report beyond the authorized allocation within your
organization then the licensing terms for our research are
being violated.
If you value the information and insight that we provide
then I strongly urge you to respect our hard work and livelihood and subscribe to our research. If you do not have a
platinum license or a global license, you may want to upgrade
your license so that you can share this issue across your entire
organization with our blessing.
If you or your organization is interested in distributing this
report to outside organizations, please feel free to contact us
to discuss licensing terms and fees.
If you would like to leverage a figure or quote from this
report and you have at least a global license, please contact
us for permission and we will be happy to provide it.
If you would like a high-resolution version of this issue, please
contact us and we will ensure that you receive it. If you would like
to purchase the raw data for your own analysis please contact us to
discuss terms. If you havent done so already, please read the revised
licensing terms which immediately precedes this paragraph.
186GB in an LTE Network been there, done that
As most readers can appreciate, two weeks in Europe is a long

time away from home. While we were gone the New Orleans
Saints defeated the Indianapolis Colts by 31 to 17 in Super
Bowl XLIV. The United States Olympic team was well on its
way toward bringing home 37 medals. The top 181 American
Idol hopefuls got whittled down to a more reasonable 24 contestants during a hellacious Hollywood Week. 47,000 people
descended on Barcelona for four days of the Mobile World
Congress.
On a more personal level, the two-year-old heir apparent
to the SRG dynasty went from a cold start to proudly reciting
his ABCs (through Z) and counting from 1 to 20 while his
sister fortunately bided her time before entering this world at 6
pounds, 9 ounces literally days after we returned home (refer-
Figure 1 Isabella Sunah Thelander

born March 2, 2010 at 0844 hours
Source: Mom and Dad
ence Figure 1). Oh yeah, and the time that we spent in Europe
also gave us more than ample opportunity to transfer nearly
186GB (download = 163.74, upload = 22.25GB) in LTE networks in Stockholm, Sweden and Oslo, Norway before heading to MWC.
In this special two-part issue of Signals Ahead we provide
what is arguably the most scientific and exhaustive independent performance analysis of a LTE network(s). Thanks to
Accuver, we were able to use its XCAL LTE drive test tool to
capture the performance characteristics and underlying KPIs
for the Samsung USB dongle, thus providing us with detailed
information on how LTE performs in a commercial network,
not to mention the real-world experience of a typical end user,
although a typical end user would never stress test the network to the degree that we did.
To put things in perspective, our total usage during five
days of testing was nearly twenty-five times higher than the
typical fixed broadband user in most developed markets over
the course of an entire month. Further, the overwhelming
majority of the results and our analysis of the data are based
on vehicular testing with only a modest amount of pedestrian
mode testing and even less stationary testing. From a practicality standpoint, testing from a mobile environment allowed
us to cover more territory in an expeditious fashion. Further,
the continued randomness of the LTE modems position relative to the serving eNode B made our results statistically significant, even if the results were arguably lower than what they
could have been from a stationary position.
Our total usage during five days of

testing was at least 25 times higher
than the typical fixed broadband
user in most developed markets
over the course of an entire month.
March 12, 2010 | Signals Ahead, Vol. 6, Number 3
As we have demonstrated in the past, testing from a stationary position is problematic at best, since moving a few feet
or even inches, or simply turning the modem by 45 degrees,
can significantly increase/decrease the achievable throughput.
Even if the locations are randomly selected, testing from only
a few fixed locations can significantly skew the results either
positively or negatively.
To the best of our knowledge the XCAL LTE drive test
solution was the only drive test tool that had the ability at the
time of the testing to interface with the Samsung dongle and
extract the critical information that we needed on a real-time
basis and for post-processing purposes, although it may be
possible to capture very rudimentary performance data (e.g.,
throughput) through other means. We also used the companys
XCAP LTE post-processing tool to analyze more than 150MB
of log files that we created during the course of our testing.
All of the figures that we present in this report leveraged this
tool although for formatting purposes of this report we did
export some the data into an Excel format before recreating
the figures.
The report is divided into two issues and eight sections,
including this introductory section. The following section
includes the key highlights and findings from drive testing the
two networks. Section 3 contains our test methodology. The
next two sections contain detailed results for specific test scenarios Section 4 is Stockholm and Section 5 is Oslo. Section
6 puts the results into perspective. Section 7 provides some
quick concluding remarks. Section 8 is an expanded appendix
which includes additional results that didnt make their way
into the main report.
We encourage readers of this report to refer back to our
Mobile WiMAX (SA 090909, Wireless in Washington
and the surround Portland area) and HSPA+ (SA 070109,
HSPA+ - up close and personal Down Under) drive test
reports in order to compare and contrast the results. Annual
licensees to our Signals Ahead research product should already
have these reports in their position. They may also be purchased separately. Dont forget to also check out the new video
that we posted to the SRG website
Key Highlights, Findings and Caveats

Based on extensive testing over a five-day period we can make
the following observations and key conclusions.
1. In general we found the network coverage to
be greater than we had expected, but both networks are still clearly in the early stages of
being deployed. Based on comments in the press and
input from the vendors we had expected rather small networks in the two cities. However, both networks covered
3
fairly meaningful portions of their respective cities. This

was more noticeable in Oslo where we half expected to be
able to walk the network given that there were only ~100
cell sites activated at the time. Instead, these cell sites covered a fairly large geographic region as evident in our drive
test route (reference Figure 31) although there was good
evidence that additional cell sites were still being deployed
to infill coverage gaps that currently exist.
That said, neither of the two networks was large by any
stretch of the imagination and they fall short when compared with the HSPA+ network that we tested in Australia or even the Mobile WiMAX network that we tested in
Portland, Oregon. We guesstimate, for example, that the
Stockholm network only covered a 10 square mile section
of the city. Further, any network that is only comprised of
a few hundred cell sites shouldnt be considered meaningful given that most mobile operators deploy thousands if
not tens of thousands of cell sites, both for coverage and for
capacity purposes. That said, both networks were more than
large enough for testing purposes and they represented a
commercially-deployed network versus a contrived vendor
deployed network with a very limited number of cell sites,
thus these networks more than met our requirements.
TeliaSonera is in the process of expanding both networks
and in the coming year it will be building out its LTE networks in other cities. Further, the operator is looking to
use its existing GSM spectrum (900MHz and 1800MHz)
while if it is successful in the forthcoming auction it will
also be able to use 800MHz to deploy LTE in less densely
populated regions within the two countries.
2. We found that in-building coverage was excellent in at least some areas, although inevitably
better in-building coverage is required elsewhere in the network. We didnt spend a lot of time
testing in-building coverage but we did use and test the
LTE network from our hotels and throughout the Galleria
mall in downtown Stockholm. In those areas where we did
conduct our in-building testing the results were quite favorable, and we would argue generally better than the results
that we obtained outdoors in our pedestrian tests.
For example, we walked throughout the entire Galleria
mall in downtown Stockholm and we consistently observed
downlink PHY Layer data rates of 25-50Mbps. As the
mall stretched for at least two full city blocks and included
a lower-level floor with an entrance to the subway station
and dead end hallways, we find this performance to be
quite impressive. We theorize, and TeliaSonera concurs,
that we may have benefited from the absence of interference

from neighboring cells (e.g., the CINR was actually better
indoors). Unfortunately, we were not successful in logging
this particular test scenario so we are working off of our
vivid memory of the experience.
Elsewhere, we witnessed 15-25Mbps from our hotel room
in Stockholm and 55Mbps from deep within the hotel lobby
at the Grand Hotel in Oslo a historic hotel with thick
walls and pillars separating us from the outside. However,
in our hotel room in Oslo, which was located in an innercourtyard of the hotel (the cheapest room in an otherwise
expensive hotel), we couldnt detect the LTE signal. We
also were relegated to EDGE versus 3G with our mobile
phone.
Although we had a very favorable in-building experience
we are not suggesting that the in-building coverage is consistently as impressive throughout the two cities. Inevitably,
there are consumers who are not happy with their in-building coverage at the moment, although we believe that if
coverage is great to the exterior wall of the building then
the coverage is at least good to 1-2 walls within the building.
Therein lies the crux of 2600MHz getting good coverage
to the exterior wall.
TeliaSonera intentionally designed its networks to provide
in-building coverage from the outside macro network, but
the operator is also a big believer in leveraging various inbuilding solutions to further improve the user experience.
Over time we believe that these solutions will play a greater
role in its network rollout strategy.
3. The adjusted average downlink PHY Layer data
rate in a 2x10MHz radio channel was 16.8Mbps
and 32.1Mbps in a 2x20MHz channel. Our test methodology, which is almost entirely dominated by vehicular and
to a much lesser extent pedestrian testing, does two things.
First, it artificially reduces the throughput from what could
have been achieved in an ideal location and from a stationary position. Second, by constantly moving throughout the
network while measuring the performance, we ensure that
our results are statistically valid. Had we limited our tests to
a few locations we may have achieved substantially higher
results or we could have even observed lower throughput.
We have demonstrated this phenomenon several times in
the past, most notably while doing our HSPA+ and HSPA
testing in Telstras network. During pedestrian mode tests
we observed an adjusted average downlink PHY Layer data
rate of 18.7Mbps and 33.1Mbps, respectively.
While less meaningful, we observed peak data rates of

approximately 50Mbps in a 2x10MHz channel and nearly
85Mbps in a 2x20MHz implementation. Had we gone
looking for great data rates we would have inevitably been
able to report even higher data rates, but such an approach
would not be consistent with our test methodology.
We also note that the Samsung modem which we used during our testing was a Category 3 device, meaning that it was
limited to 75Mbps in a 2x10MHz channel and 100MHz
in a 2x20MHz channel. Although we were not close to
exceeding those two limits, that does not suggest that we
would not have observed other performance benefits due to
the use of a higher category device.
Based on the number of resource blocks that we were observing throughout much of the testing, we feel pretty comfortable in stating that network loading was not a factor in the
results. With more users in the network the achievable data
rates would have been lower but it would not be a linear
function as it is seldom the case that all users can fully utilize all of the network resources (e.g., fading and other RF
challenges already reduce the achievable throughput of any
single user during most instances in time). Further, most
mobile data applications and services do not consistently
require constant bandwidth as active usage periods and data
transfers are followed by intervals of inactivity (e.g., reading
a web page or an email). We discuss this point in the Some
Perspective section of this report.
The achievable end user data rates are based more on an
operators business decision than they are based on the performance capabilities of the technology itself.
It is also an operators prerogative to set its own expectations regarding throughput. As networks become loaded
an operator can always deploy additional network resources
(use more spectrum, deploy new cell sites) in order to boost
the average end user data rates. The achievable end user data
rates, therefore, are based more on an operators business
decision than they are based on the performance capabilities of the technology itself. That said, the more capacity
that can be made available for a given amount of network
resources and the more economical it is to deploy additional
capacity makes it easier to deliver higher performance.
4. The adjusted average uplink PHY Layer data
rate in a 2x20MHz radio channel was 10.8Mbps.
For various reasons we did not spend a lot of time testing
the uplink performance. In Stockholm, the current vendor
release at the time limited the performance of the uplink
to approximately nera the full uplink capability is being
rolled out, even as this report goes to publishing. In Oslo,

we faced time constraints although the network supported
the full uplink capability.
In the three uplink-only tests that we did conduct we
observed an adjusted average uplink PHY Layer data rate
of 10.8Mbps. This test occurred while walking the streets
of Oslo. During this test the peak data rate was 40Mbps. In
the Stockholm network we observed data rates in excess of
5Mbps on numerous occasions due entirely to ACK messages being sent in the uplink.
As part of our planned follow-on testing we plan to spend
more time looking at the performance of the uplink.
5. Latency, as determined by pinging an out-ofnetwork server was as low as 14ms and generally no higher than 20-30ms. As was the case with
the uplink tests we didnt spend a lot of time testing the
latency since 1) we would expect very favorable latency in
a lightly-loaded network and 2) the results while mobile
would likely have been influenced by the network optimization issues that we discuss in an upcoming bullet.
While we didnt necessarily test the latency in other
instances, we knew it had to be quite low since we have
never experienced Internet surfing like we experienced in
the LTE network. The web pages would only have loaded
faster if deterministic web browsing was possible and not
even Google has figured out how to implement that feature.
Our web surfing experience could only have been possible
through the combination of a high throughput/low latency
network.
As part of our planned follow-on testing we plan to spend
more time looking at latency and jitter.
6. Handovers between LTE cell sites generally
worked reasonably well although this is an
area for significant improvement that can be
addressed with network optimization. Based on
an analysis of most of the log files for Stockholm we calculated that the Stockholm network had a handover success
rate of 77% (578 successful handovers out of 750 attempts).
A similar analysis done on the Oslo network resulted in a
handover success rate of 92% (822 successful handovers out
of 893 attempts).
The lower success rate in the Stockholm network could be
due to several factors, although we assume all largely associated with the need for network optimization. These factors include reducing the antenna tilt in order to limit the
interference from the neighboring cell sites and updating of
5
the neighboring cell list with the complete list of available

cells. As discussed elsewhere, the Stockholm network had a
much denser deployment of cell sites and without any network optimization, this inevitably created more interference
relative to the Oslo network.
In the Oslo network part of our testing was done in an area
of the network that was optimized and even though the
actual throughput in this area was lower than it should have
been due to some vendor-related testing that was taking
place at the time, the handover success rate was arguably
quite good. This factor, as well as the deployment of a less
densely built network helped to improve the overall cell
handover rate (e.g., there was generally less interference at
the cell edge versus in the Stockholm network).
We note that in reviewing the data from both networks we
observed numerous instances where multiple handovers
occurred in a relatively short period of time as the modem
sought to identify the best serving cell. While the ping
ponging between cell sites isnt necessarily a good thing
and it will probably go away with network optimization, to
us this ping ponging effect is a good indication that cell
handovers can work quite well when the network is humming. That said, much of the Oslo network had not been
optimized and at least some of the dropped handovers was
due to the absence of a cell and there is a need for further
improvement, even more so in the Stockholm network.
Our methodology for these calculations involved looking at
the bulk of the log files for each network. We then counted
the number of instances where the Serving Cell ID changed
with no interruption to the transfer of data. We also counted
those instances where the Serving Cell ID changed with
an interruption to the throughput or those instances where
no Serving Cell ID was identified in order to determine
the number of unsuccessful handovers. Further, we tried to
exclude those dropped handovers that seemed to be caused
by coverage gaps. Although this was a somewhat manual
and tedious process, our approach helped ensure that a failed
handover was only counted once in many failed handovers
the modem bounced unsuccessfully between several cells,
but we only counted this event as a single failed handover.
7. Both networks would benefit from some much
needed network optimization which should
meaningfully improve the user experience. When
we met with TeliaSonera in Sweden the operator commented that LTE was the easiest technology it had ever
deployed We just turned it on and it worked. However,
as currently implemented the LTE networks are being
used for data-only, best-effort applications and there is no

requirement for real-time applications, such as voice/VoIP,
or handovers to the legacy HSPA+/HSPA networks.
In looking at the results it is also evident that both networks
could benefit from some network optimization and this is
an activity that the operator plans to pursue later this year.
Presently, many of the antennas point into neighboring cells
or even extend into faraway cells as was the case in Stockholm where cell sites from across the river played havoc
with our testing. Our analysis of the data suggests that this
problem was more prevalent in Stockholm, but largely due
to the higher densification of cell sites in the network. We
note that TeliaSonera owns its own backhaul network in
Stockholm so it is much easier to provision the cell sites
than it is in Oslo.
The lack of meaningful network optimization had several
implications. It reduced the typical end-user data rates since
the CINR values were lower than they otherwise would
have been. It also impacted the handover success rate since
at the edge of the cell the Samsung modem had too many
options to chose from while in some cases the list of neighboring cells was incomplete or the modem was using a cell
site that it was not supposed to be using given its location in
the network.
Although the handover issue was more evident in Stockholm in Oslo we encountered cells behaving badly or cells
allocating a widely inconsistent number of resource blocks.
At this point it isnt clear what was causing this issue in the
Oslo network, but we theorize that the backhaul or the provisioning of the cell site could have played a role. In other
words, the problem wasnt systemic although it did tend to
impact the results with many of the test scenarios.
Our point in raising the network optimization issue and in
backing it up with our analysis later in this report is to help
illustrate that the performance of the two LTE networks
will improve over time and that many of the performance
issues that we faced had more to do with the deployment of
the network than the technology itself. We also note that
most users would not even notice the issues that we identify in this report since most applications are very forgiving
while dropped data sessions are frequently masked by the
modem reconnecting on its own.
8. The availability of higher modulation schemes,
including 64QAM, was quite good in the two
networks, although there were some large variances in the results. The Accuver XCAL drive test tool
captures the modulation schemes being reported by the two
6
antennas recall LTE uses MIMO. Therefore, there are

actually two distributions of modulation schemes that need
to be analyzed.
Based on an analysis of all of the test results from the Stockholm network, the primary antenna reported 64QAM for
19% of the time, 16QAM for 35% of the time, and QPSK
for 45% of the time. The secondary antenna reported 15%,
29% and 55%, respectively. In the Oslo network the primary
antenna reported 64QAM for 55% of the time, 16QAM for
31% of the time, and QPSK for 9% of the time. Conversely,
the secondary antenna reported 26%, 12% and 20%, respectively. Note that the sum of the values does not always equal
100% and in some cases it falls well short of the mark. This
result is because in some instances the secondary antenna
is not reporting/using the available network resources (see
MIMO discussion below).
9. MIMO (Open Loop Spatial Multiplexing) was
widely available in both networks, although
there were some large variances in the results.
Both networks took advantage of the MIMO feature,
namely open loop spatial multiplexing in which two unique
data streams are sent between the serving eNode B and the
modem. In the case of the Stockholm network Open Loop
SM was used 99% of the time while in the remaining 1%
the network defaulted to SIMO (Single Input, Multiple
Output). In the Oslo network Open Loop SM was used
58% while in the other instances the network defaulted to
2TX SFBC (Space-Frequency Block Coding).
10. We were impressed with the high data rates that
we were able to obtain even with relatively low

CINR values. As part of our analysis we created scatter
plots of the downlink PHY Layer throughput versus the
Serving Cell CINR, or the strength of the desired signal
versus the strength of all received signals, including both
the desired signal as well as the surrounding interference.
In theory high CINR values with low interference (RSSI)
are the most desirable but this is not always possible since
challenging RF conditions exist while one cannot completely remove the interference from neighboring cells, in
particular in an N=1 network.
As discussed in an earlier finding, the two networks had
not undergone any optimization so the CINR values were
artificially deflated relative to what they could and should
have been in a fully optimized network. That said, for the
relatively low CINR values that we observed we were able
to achieve meaningfully high data rates. We even observed
megabit-per-second data rates with a negative CINR.
For example, in the Stockholm network the CINR values

were generally below 20dB and always below 25dB, yet
we were able to measure data rates approaching 50Mbps.
To be more specific the adjusted average CINR across all
test scenarios in Stockholm was 12.4dB and the adjusted
median was 14dB. In Oslo the adjusted average CINR was
14.5dB and the adjusted median was 16.5dB. We theorize
and TeliaSonera concurs that the CINR values in Oslo
could have been higher since the network is not as densely
deployed at the moment. With network optimization the
CINR values in both networks should improve. The operators target is a minimum CINR of -1.5dB.
11. While there is a natural inclination to compare
these results both against themselves and with
other technologies, such a comparison can be
problematic (but fun and insightful). Our sole
purpose in doing these tests and in publishing this report
was to get a first-hand look at how LTE performed and
to provide that insight to the industry at large. We ended
up doing our testing in Stockholm and Oslo since we grew
tired of waiting for the hotline with a certain operator based
in New Jersey to ring.
However, it wasnt lost on some of our readers or on Ericsson and Huawei that we would be in a unique position to
provide detailed competitive intelligence on the performance of the two vendors solutions. Further, given operator interest in the prospects of continuing down the HSPA
path with HSPA+ or jumping to LTE, not to mention the
merits of other operators deploying Mobile WiMAX versus
LTE (or HSPA+), there is a natural and important opportunity to compare the three technologies.
Since we already provide the information that readers need
to do these comparisons as part of our analysis the HSPA+
and Mobile WiMAX studies were published earlier in Signals Ahead we might as well weigh in on the matter and
shed some perspective on how to make the comparisons as
well as some of the factors which make these comparisons
problematic.
a. Stockholm versus Oslo. There are several noteworthy differences between the two networks which we
think are worth mentioning. However, we firmly believe
that most of these differences can be explained by the
deployment of the network itself versus the maturity of
each vendors solution. That said, well go where no analyst has gone before and provide some quantified opinions on the two vendors LTE solutions.
At the time of our testing the Stockholm (Ericsson network) only supported 10MHz channels and the uplink
throughput was limited to no more than 6Mbps. Conversely, the Oslo (Huawei network) supported the full
20MHz channels and what we believe were the full
uplink capabilities. As this report is going to press the
Stockholm network is being upgraded to support the
two limitations that existed when we did our testing.
Based on the use of 20MHz channels and the high
CINR values that we observed, we would have expected
even higher throughput in Oslo than what we witnessed.
The doubling of the channel bandwidth by itself should
have met a doubling of the data rate while the measurably higher CINR would have further increased the lead.
Although we did see some very high data rates it was not
consistent across the network. The cluster in the Oslo
network where the vendor testing was taking place was
also an optimized network and because of the testing the
maximum number of available resource block was only
52, or on par with what is possible in a 10MHz channel. Assuming nothing else was going on, comparing
the results from this cluster with the Stockholm results
would be a fair comparison and in that regard Oslo fell
short of expectations. However, in other regions we
observed the super-high data rates that came close to the
limits of the Samsung device.
We also note that some of the cell sites in the Oslo network assigned resource blocks in a very erratic fashion
and although this insight isnt readily available in some
of the results we do highlight this underlying issue in
the text. We believe that the performance of these cells
was not consistent with the capabilities of the Huawei
solution but at this time we do not know the root cause
of the problem other than that we see no reason why it
cannot be resolved through network optimization.
Many of the performance

differences between the two
networks can be attributed to
network optimization issues.
With respect to handovers, the Oslo (Huawei) network
did meaningfully better than the Stockholm network.
In addition to a much higher handover completion rate,
the throughput at the cell edge was more consistent and
the data rates during the handover process were sub-
Come join us!
stantially better than they were in Stockholm, even after

adjusting for the differences in the channel bandwidth.
Finally, it appears that the two vendors are taking different approaches toward optimizing their respective solutions. Ericsson tends to focus on achieving Open Loop
Spatial Multiplexing at the expense of higher order
modulation schemes for the two antennas. Conversely,
Huawei seems to focus on getting the highest possible
modulation scheme from the primary antenna even if it
means sacrificing the use of Open Loop SM. We will
leave it to others to determine which approach represents
a better and more efficient use of network resources.
b. LTE versus HSPA+ versus WiMAX. Readers should
keep several considerations in mind when comparing
the performance of these three technologies using the
information that we provide in this report and in earlier
Signals Ahead issues. The LTE network in Stockholm
used twice as much spectrum/channel bandwidth as
the HSPA+ network in Melbourne while the LTE network in Oslo used four times as much spectrum/channel bandwidth. Further, both LTE networks leveraged
MIMO, which has yet to be implemented in Telstras
HSPA+ network. All things being equal one would
expect a doubling in the average/peak throughput for a
commensurate doubling of the channel bandwidth, plus
an additional boost due due the introduction of MIMO.
The best comparison between HSPA+ and LTE would
be to compare LTE with DC-HSPA (with MIMO) an
exercise that we will be taking on later this year, albeit
perhaps without MIMO being implemented.
When comparing the performance

of various technologies, once
must take into consideration the
channel bandwidth and the total
amount of spectrum being used.
With respect to Mobile WiMAX, the network in Portland
used 1x10MHz channels so one would expect the average
data rates in Stockholm to be twice as high and the average
data rates in Oslo to be four times as high. Readers of this
report who are subscribers to our Signals Ahead newsletter
are encouraged to look back in our Mobile WiMAX report to
see how LTE stacks up against Mobile WiMAX (if you dont
have an annual license to our research, we encourage you to
purchase the report as well as the HSPA+ report).
CTIA Wireless, Las Vegas, NV

March 22-25
LTE World Summit, Amsterdam, Netherlands
May 18-20
Invited speaker and moderator
However, one must also consider that the Mobile WiMAX

network used 30MHz of spectrum (N=3) outside of downtown Portland and 60MHz (N=6) of spectrum in downtown
Portland. In other words, the Stockholm network, with only
20MHz of allocated spectrum, used less spectrum than the
Portland network. Likewise, the Oslo network (40MHz)
used less spectrum than was used in downtown Portland, but
10MHz more spectrum versus than what was used the Portland suburbs.
Given the Mobile WiMAX focus on TDD, the only fair
comparison would be to compare a 1x20MHz implementation of Mobile WiMAX with a 2x10MHz implementation of
LTE both using an N=1 frequency reuse. Alternatively, one
could compare TD-LTE with Mobile WiMAX. Given ongoing activities in China this latter option may be possible to
pursue later this year.
Test Methodology
For the LTE drive test we used the Accuver XCAL LTE drive
test tool to collect the underlying performance characteristics
of the LTE modem, and in turn the performance of the LTE
network as experienced by an individual user. As was the case
with our HSPA+ and Mobile WiMAX drive tests, no attempt
was made to look at overall network efficiency although this
is an area of personal interest that we would like to pursue in
the future. The XCAL LTE drive test tool is a software-based
solution which easily installs onto a Windows-based notebook
computer. This feature proved to be invaluable since it allowed
us to do some pedestrian and in-building testing without having to lug around a bunch of additional test equipment.
We used the Accuver XCAL LTE drive test tool to collect the underlying performance characteristics of the LTE
modem, and in turn the performance of the LTE network
Once the software is installed it is only a matter of establishing a network connection with the Samsung application
manager, launching the XCAL LTE software, and starting a
log file. In order to plot the results using Google Earth, we also
used a Garmin GPS receiver which interfaced with the data
collection tool.
Figure 2. Accuver XCAL LTE Drive Test Tool in Action
After we started generating the log file with the click of the
mouse, it was simply a matter of running a test scenario, which
in our case meant driving the streets of Stockholm or Oslo
while listening to the continuous replaying of Lady GaGa in
the local radio station. While seemingly a simple task, hearing
the same Lady GaGa song a half dozen times before breakfast
and then at least 2-3 times an hour throughout the day is no
laughing matter.
The frustration of listening to her music was also compounded by the CD player in our Volvo SUV which literally ate
our hotel parking ticket and then attempted to play the ticket
every 30 seconds or so. Fortunately, we were able to MacGyver the ticket out of the player but we had no such luck
removing Lady GaGa from the radio stations rather limited
playlist. Due to sensitivities surrounding some of the language
and references in her songs, we will be using another very
popular song in Sweden and Norway at the time of our tests
from the Black Eyed Peas in a forthcoming video montage that
we will be posting to the SRG website.
Both infrastructure vendors Ericsson in Stockholm and
Huawei in Oslo provided us with access to a server located
immediately outside of the operators network which we used
for our throughput testing. In the case of the Stockholm tests
we predominantly used the UDP protocol with some FTP
testing also taking place. Due to logistical issues, we used a
Source: Accuver and SRG
combination of FTP (FileZilla) and UDP in Oslo with the

majority of the results based on using FTP. When using the
FTP protocol, especially in Oslo, we established multiple FTP
sessions in order to ensure that the broadband pipe was full of
data and to more closely approximate the use of UDP.
We set the TCP window size on our notebook computers to 1048576 and the MTU size to 1360. If we not done
so we would not have seen the higher end of the results. In
theory, installing the Samsung connection manager should
have resulted in the TCP window sizes changing but this was
seemingly not the case.
We also used the Net Meter shareware application to look at
application layer throughput. In addition to providing a GUI
that shows the throughput on a real-time basis much like
NetPerSec this particular application automatically creates a
log file of the results on a per second basis. By combining this
file with the log files from the Garmin GPS receiver we could
look at the application layer throughput and it proved to be a
nice complement to the XCAL LTE results.
The tests that we conducted were done independently
and financed, if you will, by our large base of Signals Ahead
clientele. Ericsson and Huawei gave us a drive tour of their
respective networks, provided us with a USB dongle, access to
their server, on-call support for questions, and a free meal, but
otherwise had no involvement in the data collection process.
Out of necessity, Huawei did participate in the data collection process that involved its test modem, but we collected the
majority of the Oslo test results on our own. During our stay
in Stockholm we also visited with senior management from
TeliaSonera with much of their comments already appearing
in the Signals Flash! report that we published while the tests
were ongoing.
As we did with our HSPA+ and Mobile WiMAX drive
tests, just prior to the publication of this report we provided a
pre-brief to the two vendors and the operator. These pre-briefs
were done as a courtesy and to help us get resolution to certain questions that we had identified during the course of our
analysis.
Although we had a seemingly unlimited number of KPIs that
we could have analyzed and presented in this report, we elected
to focus on a few select KPIs. Organizations who are interested
in obtaining the complete set of log files are encouraged to contact us for additional details. These KPIs include the following:
Figure 3. Stockholm LTE Network
DL Physical (PHY) Layer Throughput (maximum, average,
median, CDF, distribution and adjusted average)

UL PHY Layer Throughput (average, adjusted average and
maximum)
CINR (average and median)
RSSI (serving cell, Top 1, Top 2 and Top 3 cells)
Resource Block (RB) Allocation (average and median)
Modulation Scheme (none, QPSK, 16QAM and 64QAM)
MIMO Type (SIMO, 2TX SFBC and 2Tx Open Loop
SM)
BLER
We also leveraged the capabilities of the XCAP tool to

analyze these KPIs by several different means, including:
DL
PHY Layer Throughput versus Cell ID (e.g., handover

performance)
DL PHY Layer Throughput versus CINR (scatter plot and
real-time)
CINR versus RSSI (scatter plot and real-time)
DL PHY Layer Throughput versus CINR/RSSI (realtime)
CINR versus Modulation Scheme
Finally, we provide geo-plots of several of the KPIs,
including:
Vehicular Speed
DL PHY Layer Throughput
CINR
Modulation Scheme (antenna 1 and antenna 2)
MIMO Type
10
Source: TeliaSonera Website
Adjusted average throughput removes all instances of

0Mbps throughput in the calculations and it is perhaps a better indication of the typical end user throughput.
Of note, we created a new KPI which we dubbed adjusted
average throughput. By our definition, adjusted average removes
all instances of 0Mbps throughput in the calculations and it is
perhaps a better indication of the typical end user throughput.
We made these adjustments since most of these instances were
due to extraneous factors which erroneously lowered the real
throughput from what we observed. These factors include:
Starting the log file before starting the transfer of data
Stopping the log file after the transfer of data had finished
Coverage gaps where cell sites had yet to be deployed
Lengthy (multi kilometer) tunnels, in particular in the Oslo
area
The time spent re-establishing a connection to the server
following a dropped call
We used a similar methodology when doing our Mobile

WiMAX tests. For completeness sake, we present both sets of
results. The geo-plots also show gaps of missing information,
which in some cases was also caused by the GPS receiver resetting itself or losing its position.
11
Ostermalm Drive Test 1225
Sodermalm Drive Test 1 0500
Old Stockholm Pedestrian Walk
28
Stockholm West Drive Test 0820
23
Stockholm Pedestrian Walk
22
Ericsson Lobby Stationary Test
21
27
26
Feb 10: 0700 hrs
19
20
Feb 10: 0642 hrs
Feb 9: 0745 hrs
17
18
25
Feb 9: 0710 hrs

Feb 9: 0725 hrs
15
16
24
Feb 9: 0655 hrs
14
Feb 9: 0629 hrs
Feb 9: 1413 hrs
Feb 10: 1300 hrs
Feb 8: 1440 hrs
Feb 10: 0845 hrs
Feb 10: 0840 hrs
Feb 10: 0820 hrs
Feb 10: 0815 hrs
Feb 10: 0805 hrs
Feb 10: 0745 hrs
Feb 9: 0645 hrs
13
Feb 9: 0605 hrs
Feb 9: 0548 hrs
Feb 9: 0535 hrs
Feb 9: 0500 hrs
Feb 9: 1225 hrs
12
Feb 9: 1140 hrs
11
Feb 9: 1050 hrs
10
Feb 9: 1025 hrs
4
Feb 9: 1210 hrs
Feb 9: 1013 hrs
Feb 8: 1321 hrs
Drive from Kista
Date/Time
Scenario
No.
Table 1 Stockholm Results Summary
3,734.2
1,814.8
521.8
6,761.3
30.9
1,008.2
370.2
24.2
1,423.4
4,398.4
1,354.7
919.6
1,635.2
1,668.2
583.0
319.2
1,478.0
821.0
1,032.4
657.0
1,748.0
1,634.0
1,251.5
1,944.7
559.1
699.8
22.2
33.6
Transfer Size
(MB)
25:08
14:13
4:37
56:27
00:24
08:13
07:13
02:58
14:06
37:56
13:21
09:49
13:29
13:22
10:09
4:26
12:37
13:22
10:15
10:07
25:25
19:43
10:23
22:12
12:50
7:11
1:44
1:09
Transfer
Time
(min:sec)
2.0
1.4
N.A.
10.6
3.6
5.6
6.6
13.6
6.4
8.7
10.0
11.0
10.2
10.1
16.7
15.7
11.2
13.1
17.8
10.2
9.5
10.9
9.9
10.2
7.2
10.2
11.9
40.7
Avg Speed
(mph)
19.8
17.0
15.1
16.0
10.3
16.4
6.8
15.3
13.5
15.5
13.5
12.5
16.2
16.6
7.4
9.6
15.6
8.2
13.4
12.9
9.2
11.1
16.1
11.7
14.2
13.0
1.7
3.9
21.4
18.7
25.9
19.5
15.5
16.7
12.1
15.3
15.5
16.8
15.6
13.7
17.7
18.0
7.4
13.0
20.7
13.5
17.9
17.9
17.9
15.5
17.5
13.3
17.4
15.3
10.2
3.9
43.3
49.3
29.5
49.2
43.2
40.6
36.9
41.0
46.0
49.3
45.3
47.7
49.2
38.8
40.6
33.8
47.1
49.1
46.2
49.2
41.6
47.7
46.1
48.6
48.4
43.9
22.0
23.6
Avg PHY Adjusted Avg Max PHY

Layer
PHY Layer
Layer
Throughput Throughput Throughput
(Mbps)
(Mbps)
(Mbps)
14.4
14.1
16.8
12.9
8.1
10.7
15.1
10.0
12.5
11.9
12.5
9.0
12.7
12.5
7.7
9.9
22.0
11.5
12.4
8.7
12.7
11.2
13.0
10.5
13.3
12.0
10.0
7.5
Avg CINR
(dB)
17.0
15.0
17.0
15.0
9.5
10.0
17.0
12.0
14.0
13.0
14.5
10.0
13.0
14.0
7.5
11.5
14.0
12.0
14.0
11.0
15.0
12.5
14.0
10.8
14.5
12.0
13.0
7.0
Median CINR
(dB)
48.6
49.4
50.0
47.3
49.5
47.6
49.5
49.0
49.2
45.9
47.9
49.2
49.3
34.7
50.0
48.5
50.0
50.0
50.0
50.0
50.0
50.0
50.0
48.9
50.0
16.8
Avg RB
Allocation
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
50.0
3.0
Median RB
Allocation
Source: Signals Research Group, LLC
-66.7
-57.8
-79.9
-73.0
-81.3
-69.5
-64.7
-70.5
-68.5
-71.0
-74.6
-76.1
-70.9
-69.8
-77.7
-70.7
-47.8
-72.3
-73.1
-72.1
-65.6
-74.9
-70.0
-75.8
-71.6
-77.1
-78.9
-75.1
Avg RSSI
(dBm)
Figure 4. Oh the places we did go! Geo plot of all Test Routes with Speed (mph) - Stockholm
Vehicular and Pedestrian Speeds

(Mph)
30 <= x < 50
25 <= x < 30
20 <= x < 25
15 <= x < 20
12.5 <= x < 15
10 <= x < 12.5

7.5 <= x < 10
5 <= x < 7.5
2.5<=x < 5
0 <= x < 2.5
Details Results Stockholm

The LTE network in Stockholm currently covers the Stockholm area with limited, non-contiguous coverage extending to
the north in Kista the technology corridor and home to Ericsson and to the south in Farsta home of TeliaSonera. Figure 3 gives a birds eye view of where LTE coverage currently
exists. Based on conversations with the operator we expect that
this coverage will expand appreciably in the coming year and
ultimately becoming fairly ubiquitous if the operator is able to
leverage yet-to-be-auctioned 800MHz spectrum for its LTE
network rollout.
Figure 4 provides a geo-plot of the vehicular and drive test
routes that we used in Stockholm. The figure tends to underrepresent the amount of time spent collecting the data since
some of these routes were covered multiple times. In total we
covered 73 miles while doing the tests.
Table 1 provides a summary of the test results from the
Stockholm network. Consistent with our test methodology, we
are presenting all of the results that we collected. In a couple
of instances we did not successfully activate the logging of the
data so those results are not available.
12
Table 1 also provides information for a couple of other KPIs

that we would like to highlight. First, it is evident that in
many of the test scenarios we achieved a maximum data rate
of approximately 50Mbps. With the Samsung modem, which
is a Category 3 device, the maximum achievable data rate is
75Mbps in a 10MHz channel. We would have likely obtained
higher peak data rates if we had gone looking for them, but
that would not have been consistent with our more stringent
test methodology.
Second, the average and median number of resource blocks
allocated to the Samsung modem gives a pretty good indication that network loading was not a concern. In a 10MHz LTE
radio channel the maximum number of downlink resource
blocks is 50 although it is theoretically possible for multiple
(2-3) users to share a single resource block.
Finally, the results in the table do not include any of the
data that we collected with the Net Meter tool on Sunday, February 7th, or one day prior to receiving the XCAL tool from
the local Accuver distributor.
Figure 5. Stockholm DL Throughput Results CDF and Pie Chart Distribution

Avg PHY Data Rate = 13.9Mbps
Median PHY Data Rate =11.4Mbps
Adjusted Avg PHY Data Rate = 16.8Mbps
Adjusted Median PHY Data Rate =14.8Mbps

Adjusted Avg BLER =3.2
Adjusted Median BLER =2.0
40-45Mbp
35-40Mbps 1.8% 45-50Mbp
1.1%
3.8%
CDF
100
DL PHY Layer
Throughput
80
30-35Mbps
9.1%
0-5Mbps
18.5%
60
25-30Mbps
11.5%
40
5-10Mbps
17.5%
20-25Mbps
9.7%
20
0
5
10
15
20
25
30
Mbps
35
40
45
50
15-200Mbps
12.6%
10-15Mbps
14.5%
As indicated in Figure 5, the average throughput during

our tests in the Stockholm network was 13.9Mbps and the
adjusted average throughput was 16.8Mbps. As discussed in
our Test Methodology section the adjusted average throughput ignores instances where no data was being transferred and
it is perhaps a better reflection of the real throughput. For
completeness sake we report both metrics. Further, it is somewhat evident that if we had limited our testing to stationary or
pedestrian mode tests that we would have obtained a higher
average throughput across the entire network. We prefer our
approach since it is more statistically accurate even if it understates the true performance of the network.
The average throughput during our

tests in the Stockholm LTE network
was 13.9Mbps and the adjusted
average throughput was 16.8Mbps.
Figure 6 provides a geo plot of the downlink PHY Layer
data rates over the course of the tests done in the Stockholm
LTE network. We intentionally set the minimum threshold to
250kbps although in reality anything below a couple of megabits-per-second indicated terrible conditions in our view. A
13
studious reader will observe some gaps in the drive routes

where data is seemingly missing or the data rate was greater
than 0Mbps but below 250kbps a rarity unless there wasnt
any data transfer whatsoever. These gaps are due to several factors as discussed in the Test Methodology section. Also note
that for our adjusted average throughput calculations the minimum threshold was set for >0kps versus the 250kbps threshold
that we used in the figures.
The single biggest factor for many of these gaps was the time
spent trying to re-establish the Iperf session with the server.
Had we been able to stop the vehicle and/or re-drive the route
several times we could have filled in many of these gaps with
real, and occasionally quite good, throughput. In other words,
in at least some instances the measured CINR was quite good
this will be evident in some of the upcoming figures even
though no data transfer was taking place. In other instances,
we encountered areas within the network where establishing
a data connection was problematic at best due to interferencerelated issues.
As alluded to in the key highlights section, we encountered
more dropped calls than we would have preferred. Based on
analyzing the data it is evident that LTE handovers can work
quite well, but in some areas there is an abundance of interference from neighboring cells, including neighboring cells that
are actually quite some distance away. In these areas the SamMarch 12, 2010 | Signals Ahead, Vol. 6, Number 3
Figure 6. Stockholm Vehicular and Pedestrian Modes Geo plot of DL PHY Layer Data Rates
DL PHY Layer Data Rates (Mbps)

40 <= x < 55
35 <= x < 40
30 <= x < 35
25 <= x < 30
20 <= x < 25
sung modem naturally had difficulty completing the handoff.

Further, we understand that the list of available neighboring
cells that the modem used to pick the next serving cell was not
always complete at the time of our testing, thus compounding
the problem.
Finally, at the time of our tests the operator was using a
dynamic IP allocation and when we dropped a session we frequently got reassigned a new IP address instead of receiving
the same IP address. When using the Iperf application this
twist increased the amount of time required to establish a new
session while moving at vehicular speeds. The network has
since been set up to preserve a released IP address for 180 seconds. We experienced a similar issue in Oslo but since we were
using FTP the impact was not as great. We also note that the
typical user would never encounter this problem while using
most applications since the modem automatically reconnects
to the network if it is disconnected in other words, the typical user wouldnt even know that the session was dropped with
the only real indication being a short chime from the modem.
Sodermalm Drive Test Results

Figure 7 provides a close-up view of the downlink PHY Layer
data rates in the Sodermolm area in central Stockholm.
14
15 <= x < 20
10 <= x < 15
7.5 <= x < 10
2.5 <=x < 7.5
0.25 <= x < 2.5
As indicated in Table 1 we had ten separate drive test loops

in the Sodermalm area although each loop was only distinguished by the time and place some drive test loops repeated
regions covered by other drive test loops. In the main section
of this report we focus on Sodermalm Drive Test #5. In the
appendix we provide additional insight for Sodermalm Drive
Test #8.
Figure 8 provides a geo plot of the DL PHY Layer data
rates for this particular test.
Figure 9 provides a corresponding geo plot of the CINR
values for this test scenario. Not surprisingly, there is a strong
correlation between higher CINR values and higher throughput values. It will also become evident once we present some
additional figures that the beginning of test occurred to the
right in the figure where the inverted L appears note the
high CINR values in Figure 9 and no measurable throughput
in Figure 8.
Figure 10 provides another means of analyzing the distribution of the downlink PHY Layer throughput.
Figure 11 provides a wealth of information and with a little tutorial most readers should be able to quickly grasp the
information that is being presented. We use this framework
throughout this report. There are 3 figures embedded within
the main figure. The top figure plots downlink PHY Layer
Figure 7. Sodermalm Vehicular Mode Geo plot of DL PHY Layer Data Rates
>10Mbps for 63% of the time; >25Mbps for 25% of the time
40 <= x < 55
35 <= x < 40
30 <= x < 35
25 <= x < 30
20 <= x < 25
15 <= x < 20
10 <= x < 15
7.5 <= x < 10
2.5 <=x < 7.5
0.25 <= x < 2.5
Figure 8. Sodermalm Drive Test #5 Geo plot of DL PHY Layer Data Rates

40 <= x < 55
35 <= x < 40
30 <= x < 35
25 <= x < 30
20 <= x < 25
15
15 <= x < 20
10 <= x < 15
7.5 <= x < 10
2.5 <=x < 7.5
0.25 <= x < 2.5
Figure 9 .Sodermalm Drive Test #5

Geo plot of CINR Values
CINR Values (dB)

20 <= x < 25
17.5 <= x < 20
15 <= x < 17.5
12.5 <= x < 15
10 <= x < 12.5
7.5 <= x < 10

0 <= x < 7.5
-5 <= x < 0
-15 <= x < -5
Figure 10. Sodermalm Drive Test #5 DL Throughput Results CDF and Pie Chart Distribution
Avg PHY Data Rate =15.6Mbps
Median PHY Data Rate = 14.4Mbps
Normalized Avg PHY Data Rate = 20.7Mbps
Normalized Median PHY Data Rate = 19.1Mbps

Avg RB Allocation =48.5
Median RB Allocation = 50.0
CDF
100
30-35Mbps
4.8%
35-40Mbps
7.6%
0-5Mbps
12.0%
80
DL PHY
Layer
Throughput
60
25-30Mbps
12.4%
5-10Mbps
14.0%
40
20-25Mbps
15.8%
20
0
5
10
15
20
25
30
35
10-15Mbps
13.8%
15-20Mbps
19.6%
40
Mbps

16
Figure 11. Sodermalm Drive Test #5
DL PHY Layer Throughput versus Cell ID, DL PHY Layer Throughput versus CINR, and CINR versus RSSI Time Plots
DL Throughput (Mbps)
50
Cell ID
450
Serving Cell ID
45
400
40
350
35
Comcast DL
rate to home
office
(15.1Mbps)
30
25
300
20
Average rate during
Mobile WiMAX tests
(3.8Mbps)
15
10
5
250
Upper limit of
AT&T
advertised
3G rates
(1.7Mbps)
200
150
100
50
PHY Layer Throughput
0
1
50
100
150
200
250
300
350
Seconds
400
450
500
550
600
650
700
CINR (dB)
30
50
45
25
Serving Cell CINR
40
20
35
30
15
25
10
20
15
10
-5
5
0
-10
1
50
100
150
200
250
300
350
Seconds
400
450
500
550
600
650
700
RSSI (dBm)
-40
Serving Cell RSSI
Top 1 Cell RSSI
Top 2 Cell RSSI
Top 3 Cell RSSI
-50
-60
-70
-80
-90
-100
1
50
100
150
200
250
300
350
Seconds

17
400
450
500
550
600
650
700
throughput (primary axis) versus Serving Cell ID (secondary

axis). The middle figure plots downlink PHY Layer throughput (primary axis) versus Serving Cell CINR (secondary axis).
The bottom figure plots the Serving Cell CINR versus RSSI,
including the received signal strength from the top 3 strongest
cells as measured by the modem during the handover process.
The X axis for all three figures is a function of time and the
same scale applies to all three figures in this case the results
for the entire test scenario are presented although in other
instances in this report we use the template for this figure to
highlight a particular section of the overall test scenario.
In the top figure it is possible to analyze the downlink PHY
Layer throughput during a cell handover, which is indicated by
a change to the Serving Cell ID line. By our count there were
20 successful handovers and 3 instances where the handover
was not successfully completed as indicated by the absence of
the PHY Layer throughput line. In those instances, the bottom figure can be used to shed some light into what was taking
place, namely relatively high signal levels received from multiple adjacent cells (e.g., interference).
The seemingly dramatic swings in the

downlink PHY Layer throughput are
actually fairly typical of what takes
place in a mobile wireless network.

Ideally we would have used a scanner in addition to the
Samsung dongle when collecting the data and this would have
shed far more light into the issue of interference. However, we
can at least capture the interference as measured by the modem
when it is looking to complete a handover. Our belief is that
these handovers would have stood a better chance of having

been successfully completed in an optimized network and this
is an important metric that we hope to revisit later this year.
One final comment is that the seemingly dramatic swings
in the downlink PHY Layer throughput as a function of time
may appear to be abnormal (e.g., not good). As we have not
presented results in this fashion in the past we went back to
look at some of the Mobile WiMAX throughput data to see
how it would have looked when plotted as a function of time.
The plots were markedly similar on a percentage basis. Further,
the scale on the Y axis, combined with some of the peak values
that we obtained, makes some of the lower throughput values
seem extremely low. We have, therefore, highlighted a few of
these points and equated them to other reference points which
we and our readers should be familiar with. That said, some of
the troughs should get much better once network optimization
takes place.
We are impressed by the high

data rates that we achieved
with relatively low CINR.

Figure 12 provides scatter plots of downlink PHY Layer
throughput versus CINR. Those instances where we were not
obtaining any throughput but where we had good CINR are
indicated by the vertical stacking on the Y axis. Setting aside
those data points, which all occurred at the beginning of the test
scenario and perhaps due to us forgetting to start a file transfer
before starting the log file, we are actually impressed by the
high data rates that we achieved with relatively low CINR.
Figure 12. Sodermalm Drive Test #5 DL PHY Layer Throughput versus CINR Scatter Plot
CINR (dB)
25
20
15
10
5
0
10
15
20
25
30
35
40
45
50
-5
-10

-15
18

RSSI (dBm)
-100
-15
Figure 13. Sodermalm Drive Test #5 CINR versus RSSI Scatter Plot
RSSI (dBm)
-100
OK
Great
-90
-80
-15
-10
-5
10
15
20
25
-70
CINR (dB)
-60
-50
Poor
Good
-40

Figure 13 provides a scatter plot for the CINR versus RSSI.

The thresholds that we used to make the determination of Great,
Good, OK and Bad data points were somewhat arbitrary but
ideally instances of high CINR and low RSSI are the most
desirable while instances of low CINR and high RSSI are least
desirable since it suggests an abundance of interference from
neighboring cells. With further network optimization the dots
in the lower left-hand quadrant should move up or to the right,
while the dots in the OK and Good quadrants should shift
to the Great quadrant. In an N=1 frequency reuse network,
or even in an N=3 frequency reuse network, it will never be
possible to completely eliminate interference. However, we do

believe that some additional improvements can be achieved
and that this would boost the overall throughput, not to mention address some of the handoff issues that we encountered.
Figure 14 provides a plot of the Serving Cell CINR versus
the modulation scheme for the two antennas recall that LTE
uses MIMO and hence each antenna can, and frequently does,
use different modulation schemes based on the channel conditions that it experiences.
Figure 14. Sodermalm Drive Test #5 Serving Cell CINR versus Modulation Schemes for Antenna 1 and
Antenna 2
CINR (dB)
Modulation Antenna 1
Modulation
Scheme
Serving Cell CINR
64QAM
25
20
15
16QAM
10
5
0
QPSK
-5
-10
-15
400
450
500
550
600
seconds

19
Stockholm West Drive Test Results

Figure 15 provides a geo plot of the downlink PHY Layer data
rates for all of the vehicular mode tests that were conducted in
West Stockholm.
In the main section of this report we would like to focus
on the results of the Stockholm West 0642 Drive Test. In the
appendix we provide similar results for the Stockholm West
0700 Drive Test. Figure 16 provides a geo plot of the downlink
PHY layer data rates that we observed over this particular test
scenario.
Figure 17 provides the distribution of the downlink PHY
Layer data rates and the downlink data rates plotted in a
CDF plot.
It is possible to make a few observations, based on the wealth
of information contained in Figure 18. Most notably, the wide
variances in the downlink PHY Layer data rates between
roughly 350-450 seconds closely correspond to the variances
in the CINR, not to mention seemingly high RSSI from adjacent cells. Similarly, the same can be said for the results around
the 575 second mark. Finally, and somewhat lost in the noise,
there are instances of respectable data rates despite low or even
negative CINR (circa 175 and 550 as examples). This information is more evident in Figure 19.
Figure 20 provides a scatter plot of the CINR versus RSSI
for this particular drive test. The relatively high count of low
CINR and low RSSI values could be an indication of a forthcoming cell site that was not yet activated when we conducted
our tests.
Finally, Figure 21 provides a plot of the Serving Cell CINR
versus the reported modulation scheme for the two antennas.
While not specifically indicated in the figure, anytime both
antennas are reporting/using a modulation scheme, the device is
taking advantage of Open Loop Spatial Multiplexing (SM). In
those rare instances in the Stockholm network where one of the
antennas was not reporting/using a modulation scheme, SIMO
(single input, multiple output) was used the use of SIMO
occurred once during this test or at the 432 second mark.
Figure 15. Stockholm West Vehicular Mode Geo plot of DL PHY Layer Data Rates
40 <= x < 55
35 <= x < 40
30 <= x < 35
25 <= x < 30
20 <= x < 25
20
15 <= x < 20
10 <= x < 15
7.5 <= x < 10
2.5 <=x < 7.5
0.25 <= x < 2.5
Figure 16. Stockholm West 0642 Drive Test Geo plot of DL PHY Layer Data Rates

40 <= x < 55
35 <= x < 40
30 <= x < 35
25 <= x < 30
20 <= x < 25
15 <= x < 20
10 <= x < 15
7.5 <= x < 10
2.5 <=x < 7.5
0.25 <= x < 2.5

Figure 17. Stockholm West 0642 Drive Test DL Throughput Results CDF and Pie Chart Distribution
Adjusted Avg PHY Data Rate =15.6Mbps

Avg RB Allocation =45.9
Median RB Allocation=50.0
35-40Mbps
2.9% 45-50Mbps
0.1%
CDF
35-40Mbps
4.3%
100
DL PHY Layer
Throughput
80
30-35Mbps
10.1%
0-5Mbps
27.4%
60
25-30Mbps
8.8%
40
20-25Mbps
7.2%
20
15-20Mbps
13.2%
0
5
10
15
20
25
30
35
40
45
50
5-10Mbps
11.9%
10-15Mbps
13.9%
Mbps

21
Figure 18. Stockholm West 0642 Drive Test DL PHY Layer Throughput versus Cell ID, DL PHY Layer
Throughput versus CINR, and CINR versus RSSI Time Plots
High interference area
opportunity for optimization
50
Cell ID
400
Serving Cell ID
45
350
40
300
35
30
250
25
200
20
150
15
10
100
50
5
0
0
1
50
100
150
200
250
300
350
400
450
500
550
600
650
700
750
Seconds
CINR (dB)
50
30
45
Serving Cell CINR
25
40
20
35
30
15
25
10
20
15
10
-5
5
0
-10
1
50
100
150
200
250
300
350
400
Seconds
450
500
550
600
650
700
750
50
100
150
200
250
300
350
400
Seconds
450
500
550
600
650
700
750
RSSI (dBm)
-40
-50
-60
-70
-80
-90
-100
0
Serving Cell RSSI

22
Top 1 Cell RSSI
Top 2 Cell RSSI
Top 3 Cell RSSI

Figure 19. Stockholm West 0642 Drive Test DL PHY Layer Throughput versus CINR Scatter Plot
CINR (dB)
35
30
CINR
(dB)
25
35
20
30
15
25
10
20
5
15
0
10
-5
5
-10
0
-15
-5
10
15
20
25
30
35
40
45
50
30
35
40
45
50
5
10
15
20
25
-10
-15
RSSIversus
(dBm) RSSI Scatter Plot
Figure 20. Stockholm West 0642 Drive Test CINR
-110
OK
-110
-90
OK
-15
-10
-5
Great
-100
RSSI (dBm)
Great
-100
-80
10
15
20
25
30
35
10
15
20
25
30
35
-90
-70
CINR (dB)
-15
-10
-5
-80
-60
-70
-50
CINR (dB)
Poor
Good
-60
-40
-50
Poor
Good
-40
Figure 21. Stockholm West 0642 Drive Test Serving Cell CINR versus Modulation Schemes for Antenna 1
and Antenna 2
Modulation
Scheme
CINR (dB)
Serving Cell CINR
35
30
25
20
15
10
5
0
-5
-10
64QAM
16QAM
QPSK
300
23
350
400
seconds
450
500
Stockholm Pedestrian Mode Test Results

Given the balmy weather that we were witnessing in Stockholm
we were able to embark on not one but two pedestrian mode
tests during our stay. In addition to dealing with icy sidewalks
and a very cold notebook computer that would have gladly welcomed and held onto a wet tongue, we also had to deal with
closed sidewalks as building owners worked to remove ice and
snow from the tops of their buildings.
Figure 22 provides a view of the pedestrian mode test that
occurred in Old Stockholm. Note the narrow sidewalks and
tall buildings which created an ideal urban canyon. Given the
results, we believe that the urban canyon actually helped the
results in this case since it minimized the impact of neighboring cell sites.
Figure 23 provides a geo plot of the downlink PHY Layer
data rates.
Figure 24 provides a geo plot of the reported modulation
scheme for each of the antennas. A close examination of Figure 23 and Figure 24 indicates a strong correlation between
reported data rates and reported modulation schemes (e.g., the
Figure 22. A Stroll Through Old Stockholm
highest data rates were observed when both receivers reported/

used 64QAM)
Figure 25 provides additional insight into the achievable
downlink PHY Layer data rates during this test. Of note, the
average achievable data rate was 19.8Mbps while the adjusted
average data rate was 21.4Mbps.
Figure 23. Old Stockholm Pedestrian Mode Geo plot of DL PHY Layer Data Rates

40 <= x < 55
35 <= x < 40
30 <= x < 35
25 <= x < 30
20 <= x < 25
15 <= x < 20
10 <= x < 15
7.5 <= x < 10
2.5 <=x < 7.5
0.25 <= x < 2.5

24
Figure 24. Old Stockholm Pedestrian Mode
Geo plot of Modulation Schemes for Antenna 1 and Antenna 2
Modulation Type
64QAM
16QAM
QPSK
Modulation Type
64QAM
16QAM
QPSK
25
Figure 25. Old Stockholm Pedestrian Mode DL Throughput Results CDF and Pie Chart Distribution
Adjusted Avg PHY Data Rate =21.4Mbps

Avg Allocation =49.4
Median RB Allocation =50.0
CDF
100
35-40Mbps 40-45Mbp
3.0%
0.4%
DL PHY Layer
Throughput
80
0-5Mbps
6.3%
30-35Mbps
15.5%
5-10Mbps
12.6%
60
10-15Mbps
11.3%
40
20
25-30Mbps
30.0%
0
5
10
15
20
25
30
35
40
45
50
15-20Mbps
10.0%
20-25Mbps
10.8%
Mbps

For analysis purposes we are honing in on the latter portion

of the test or the period occurring from roughly 750-1550 seconds. In the geo plots the beginning of the test occurs in the
upper left portion of the figures and the test ends in the middle
right portion of the figure. Unfortunately, the King and Queen
were out for the day so we werent able to stop in for a cup of
tea or the chance to warm our toes.
Figure 26 sheds additional insight into the results. Of note
the CINR is generally quite good and the RSSI is generally low,
hence the achievable throughput during this portion of the test
is quite favorable. However, between roughly 900-1000 seconds the received power from several nearby cell sites is quite
high and this has a detrimental impact on the throughput even
though the CINR is generally, but not always, respectable.
Also worth noting, there were not any dropped handovers during this test, which we attribute to reduced interference from
adjacent cells and not to the slower speed of the modem during
the handover.
Figure 27 provides a scatter plot of the downlink PHY
Layer throughput versus the CINR. Once again, with relatively low CINR values the achievable throughput is generally
quite good. Obviously, any improvement to the CINR would
help improve the data dates.
Figure 28 provides a scatter plot of the CINR versus RSSI
during the entire Old Stockholm Pedestrian test.
26
Finally, Figure 29 provides a plot of the Serving Cell CINR

versus the reported modulation schemes for the two receivers.
This particular figure limits the data set to the range of 10001200 seconds, which includes both low data rates (1000-1050
seconds) and high data rates (1100-1200 seconds).
We also had the opportunity to do some in-building testing. Arguably we could have spent a week just looking at the
impact of throughput as a function of building type and our
location within the building. Instead, we focused on capturing
the throughput from locations within our hotel (15-25Mbps
from our room) and throughout a large indoor shopping mall
in central Stockholm. This particular shopping mall, called the
Galleria, extended for at least a couple of blocks in one direction and it included two levels throughout the complex plus
an additional lower level toward the back of the mall which
eventually made its way into the Stockholm subway system.
Unfortunately, we werent able to successfully log the results
and we didnt realize our error until after we returned home.
However, since we were able to observe the results in real-time
with the XCAL application we have a pretty good idea of what
the overall results would have looked like. In summary, the
results were much better than expected with average data rates
consistently in the 25-50Mbps range, even when riding in a
[glass] elevator or entering down dead end hallways in the
lower levels of the mall. We theorize that the walls of the mall
Figure 26. Old Stockholm Pedestrian Mode DL PHY Layer Throughput versus Cell ID, DL PHY Layer
Throughput versus CINR, and CINR versus RSSI Time Plots
45
Serving Cell ID
40
Cell ID
200
175
35
30
150
25
125
20
15
100
10
5
75
50
750
800
850
900
950
1000
1050
1100
1150
1200
1250
1300
1350
1400
1450
1500
Seconds
CINR (dB)
50
30
45
25
Serving Cell CINR
40
20
35
30
15
25
10
20
15
10
5
-5
-10
750
800
850
900
950
1000
1050
1100
1150
Seconds
1200
1250
1300
1350
1400
1450
1500
1350
1400
1450
1500
RSSI (dBm)
-40
-50
-60
-70
-80
-90
Serving Cell RSSI
Top 1 Cell RSSI
Top 2 Cell RSSI
Top 3 Cell RSSI
-100
750

27
800
850
900
950
1000
1050
1100
1150
Seconds
1200
1250
1300
Figure 27. Old Stockholm Pedestrian Mode DL PHY Layer Throughput versus CINR Scatter Plot
CINR (dB)
25
CINR (dB)
25
20
20
15
15
10
10
5
50
0-5
-5
-10
10
15
10
15
20
25
20
25
30
35
40
45
30
35
40
45
-10
-15

-15
Figure 28. Old Stockholm Pedestrian Mode CINR versus RSSI Scatter Plot
RSSI (dBm)
-100
RSSI (dBm)
-100
-90
OK
OK
Great
Great
-90
-80
-15
-10
-5
-15
CINR (dB)
-10
-5
-80 10
-70 10
15
20
25
15
20
25
-70
-60
CINR (dB)
-60
-50
Poor
Poor
-50
-40
Good
Good
-40

Figure 29. Old Stockholm Pedestrian Mode Serving Cell CINR versus Modulation Schemes for Antenna
1 and Antenna 2
Modulation
Scheme
CINR (dB)
25
64QAM
20
15
16QAM
Serving Cell CINR
10
5
QPSK
0
-5
-10
1000

28
1050
1100
seconds
1150
1200
helped shield the received power of neighboring cells (e.g., better CINR and lower interference).

Despite consistently great

coverage throughout the mall
we are not going to suggest that
in-building coverage throughout
Stockholm is as good.

Despite consistently great coverage throughout the mall we
are not going to suggest that in-building coverage throughout
Stockholm is as good and it is inevitable that the bloggers will
take to the Internet to complain about indoor LTE usage in
the Stockholm area. Then again, the satisfied customers are
generally the quiet majority. That said, as was the case with
our Mobile WiMAX testing in Portland, we believe that the
use of 2500-2600MHz isnt as bad as most people think when
it comes to penetrating buildings.
Said another way, if the coverage is great to the outside wall
then the coverage will generally be good within any reasonable building structure to at least 1-2 walls. Obviously, much
depends on the outdoor coverage being great to the outside
wall of the building, and herein lies the challenges of using
higher frequency bands to deploy a cellular network. We also
note that TeliaSonera is a strong advocate of virtually any and
all solutions which can help improve in-building coverage.

29
Michael Thelander, CEO, SRG

Michael Thelander is the CEO and Founder of Signals
Research Group. In his current endeavor he leads a team
of industry experts providing technical and operator economics analysis for clients on a global basis. Mr. Thelander
is also responsible for the consultancys Signals Ahead
research product, including its widely acclaimed Chips
and Salsa series of reports that focus on the wireless IC
industry.
Previously, Mr. Thelander was an analyst with Deutsche
Bank Equity Research. Prior to joining Deutsche Bank,
Mr. Thelander was a consultant with KPMG (now known
as BearingPoint) and a communications officer with the
United States Army. Mr. Thelander has also published
numerous articles for leading trade publications and engineering journals throughout his career.
He has been an invited speaker at industry conferences
around the world and he is frequently quoted by major news
sources and industry newsletters, including The Economist, The Wall Street Journal, Investors Business Daily, Reuters, Bloomberg News, and The China Daily. Mr. Thelander
earned a Masters of Science in Solid State Physics from
North Carolina State University and a Masters of Business
Administration from the University of Chicago, Graduate
School of Business.
Signals Ahead Subscription

The Signals Ahead newsletter is available on a subscription basis. We offer four distinct packages that have been tailored to address the needs of our corporate users. The Group License includes up to five users from the same company. The
Global License is the most attractive package for companies that have several readers since it is offered to an unlimited
number of employees from the same organization. Finally, the Platinum package includes the Global License, plus up to five
hours of analyst time. Other packages are available.
Corporate Rates (18 issues)
Group License ($3,995)
Global License ($7,995)
Payment Terms
American Express
Visa MasterCard
Check
Check Number
Purchase Order
PO Number
Name:
Title:
Affiliation:
Phone: ( )
Mailing Address:
Mailing Address
Signals Research Group, LLC ATTN: Sales
5245 College Avenue, Suite 824
Oakland, CA 94618
Platinum ($9,495)
Credit Card #
Exp Date
/ /
Our fax number is (510) 338-1284.

Alternatively, you may contact us at (510) 273-2439 or at [email protected] and we will contact you for your billing
information. We will not process your payment until after the trial subscription period is completed.
Terms and Conditions: Any copying, redistributing, or republishing of this material, including unauthorized
sharing of user accounts, is strictly prohibited without the written consent of SRG.
please note disclaimer: The views expressed in this newsletter reflect those of Signals Research Group, LLC and are based on our understanding of past and current events shaping the wireless industry. This report is provided for informational purposes only and on the condition that it will not form a basis for any investment decision. The information has been obtained from sources
believed to be reliable, but Signals Research Group, LLC makes no representation as to the accuracy or completeness of such information. Opinions, estimates, projections or forecasts in this report
constitute the current judgment of the author(s) as of the date of this report. Signals Research Group, LLC has no obligation to update, modify or amend this report or to otherwise notify a reader
thereof in the event that any matter stated herein, or any opinion, projection, forecast or estimate set forth herein, changes or subsequently becomes inaccurate.
If you feel our opinions, analysis or interpretations of events are inaccurate, please fell free to contact Signals Research Group, LLC. We are always seeking a more accurate understanding of
the topics that influence the wireless industry. Reference in the newsletter to a company that is publicly traded is not a recommendation to buy or sell the shares of such company. Signals Research
Group, LLC and/or its affiliates/investors may hold securities positions in the companies discussed in this report and may frequently trade in such positions. Such investment activity may be inconsistent with the analysis provided in this report. Signals Research Group, LLC seeks to do business and may currently be doing business with companies discussed in this report. Readers should
be aware that Signals Research Group, LLC might have a conflict of interest that could affect the objectivity of this report. Additional information and disclosures can be found at our website
at www.signalsresearch.com. This report may not be reproduced, copied, distributed or published without the prior written authorization of Signals Research Group, LLC (copyright 2004, all
rights reserved by Signals Research Group, LLC).
30

LTE Drive Test PDF

Uploaded by

Copyright:

Available Formats

LTE Drive Test PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

LTE Drive Test PDF

Uploaded by

Copyright:

Available Formats

Signals Ahead

LTE Drive Test - Part 1

Up to five (5) unique users from the

ffective immediately, any unauthorized use of our

186GB in an LTE Network been there, done that

As most readers can appreciate, two weeks in Europe is a long

Figure 1 Isabella Sunah Thelander

Source: Mom and Dad

Our total usage during five days of

March 12, 2010 | Signals Ahead, Vol. 6, Number 3

Key Highlights, Findings and Caveats

fairly meaningful portions of their respective cities. This

March 12, 2010 | Signals Ahead, Vol. 6, Number 3

that we may have benefited from the absence of interference

While less meaningful, we observed peak data rates of

rolled out, even as this report goes to publishing. In Oslo,

the neighboring cell list with the complete list of available

March 12, 2010 | Signals Ahead, Vol. 6, Number 3

used for data-only, best-effort applications and there is no

antennas recall LTE uses MIMO. Therefore, there are

we were able to obtain even with relatively low

March 12, 2010 | Signals Ahead, Vol. 6, Number 3

For example, in the Stockholm network the CINR values

Many of the performance

March 12, 2010 | Signals Ahead, Vol. 6, Number 3

Come join us!

stantially better than they were in Stockholm, even after

When comparing the performance

CTIA Wireless, Las Vegas, NV

However, one must also consider that the Mobile WiMAX

March 12, 2010 | Signals Ahead, Vol. 6, Number 3

Figure 2. Accuver XCAL LTE Drive Test Tool in Action

Source: Accuver and SRG

combination of FTP (FileZilla) and UDP in Oslo with the

March 12, 2010 | Signals Ahead, Vol. 6, Number 3

Figure 3. Stockholm LTE Network

DL Physical (PHY) Layer Throughput (maximum, average,

median, CDF, distribution and adjusted average)

We also leveraged the capabilities of the XCAP tool to

PHY Layer Throughput versus Cell ID (e.g., handover

Source: TeliaSonera Website

Adjusted average throughput removes all instances of

We used a similar methodology when doing our Mobile

March 12, 2010 | Signals Ahead, Vol. 6, Number 3

Ostermalm Drive Test 1225

Sodermalm Drive Test 1 0500

Sodermalm Drive Test 2 0535

Old Stockholm Pedestrian Walk

Stockholm West Drive Test 0820

Stockholm Pedestrian Walk

Stockholm West Drive Test 0815

Ericsson Lobby Stationary Test

Stockholm West Drive Test 0805

Stockholm West Drive Test 0745

Feb 10: 0700 hrs

Stockholm West Drive Test 0700

Stockholm West Drive Test 0840

Feb 10: 0642 hrs

Stockholm West Drive Test 0642