OpenSignal Insights Report Sample

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 109

OpenSignal Insights

CLERKENCELL HONGKONG

JUNE - AUGUST 2013

SAMPLE REPORT

copyright OpenSignal Inc. 2013

CONTENTS
EXECUTIVE SUMMARY
HOW TO USE THE REPORT
PROPORTION OF TIME ON 3G/LTE
PROPORTION OF TIME ON LTE
PROPORTION OF TIME WITH NO NETWORK CONNECTION
PROPORTION OF TIME ON WIFI
LTE DOWNLOAD SPEED
LTE UPLOAD SPEED
LTE LATENCY
LTE DATA SUCCESS RATE
3G DOWNLOAD SPEED
3G UPLOAD SPEED
3G LATENCY
3G DATA SUCCESS RATE
COVERAGE AND DATA METRICS SUMMARY
METHODOLOGY

OpenSignal

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

NSIGHTS SAMPLE REPORT


This sample report is based on data from Hong Kong.
We have redacted the names of the MNOs, replacing them with
ClerkenCell, Ucom, T-Uqbar, 3, CMHK, PCCW and SmarTone. In an
actual report this step is not taken.
The report is constructed for the network we have called ClerkenCell: the
benchmarks throughout the report show how ClerkenCell compares to
its competitor networks within the market.

ClerkenCell is a made up network


Since all networks in this market have LTE
(with the exception of CMHK)
we analyse the performance on 3G network types and LTE network
types. If it is more appropriate for your market, we can add a separate
section for HSPA+ or combine HSAP+ and LTE into a single 4G section.
In addition to this PDF we deliver an Excel spreadsheet containing the
data found in the "Source Data" tables throughout this PDF, as well as
data on the region by region performance shown in the maps.
We look forward to working with you to provide the deepest insights into
customer experience in your market.

OpenSignal

CLERKENCELL | HONG KONG | JUNE -

AUGUST 2013

EXECUTIVE SUMMARY

OVERVIEW
LTE Data Speeds

Network Coverage
.:

Overall ClerkenCell offers very strong performance on metrics


relating to network coverage. Alongside Ucom, ClerkenCell users

spend less time without a network connection and more time on


3G/LTE connections than all other networks.
'.' ClerkenCell users spend only 7.02% of the time without being
connected to the network. However, the median, which is more
representative of the 'typical user' (as it is not skewed by outliers)

.:

speeds for the period.


In terms of median LTE download speeds, ClerkenCell ranks 3rd
'.'
with speeds of 19.74 Mbps only marginally behind Ucom with
speeds of 20.38 Mbps. The median offers a better picture of typical
user experience than the mean since the latter is more strongly

suggests ClerkenCell users spend only 0.5% of the time without


network coverage.
'.' SmarTone has the poorest network coverage, with users lacking a

The data shows a very competitive market for LTE download


speeds with only small variations between all networks. SmarTone
and 3 are recording the highest average (mean) LTE download

effected by extreme cases.


'.'

ClerkenCell is also exhibiting marked increases in LTE download


speeds between the beginning of June and the end of August
(2013). All other HK networks are also showing some increases in

network connection 16.3% of the time. The median user on PCCW


is lacking network coverage 0.85% of the time, 70% more than

LTE speeds during the period of this report with the exception of

ClerkenCell users.
'.' It is clear that a significant proportion of CMHK users are on

SmarTone, whose LTE speeds appear to be more erratic. This can


also be seen from the larger than average inter-quartile distance for

2G-only data plans or live in areas that only have 2G coverage


available as they spend far more time on 2G networks than any
other networks.

SmarTone's LTE download speed distribution.


'.'

Average LTE download speeds of 20.9 Mbps for the HK market are

high by international standards.


'.'
For LTE upload speeds we are observing Ucom, PCCW and

LTE Network Coverage

SmarTone leading with very similar perfornance, ClerkenCell follows


'.' Ucom users are reporting the best LTE network coverage of all

closely behind.

HK networks, although 3 and PCCW are within levels of statistical


significance. These users are spending a majority (67.8%) of their
time connected to LTE networks indicating a strong LTE rollout.
'.' ClerkenCell users come in a strong fourth place with over 60%

LTE Latencies and Data Success Rates


'.'

(60.8%) of time spent on LTE.


'.' SmarTone and CMHK's LTE network rollout appears to be
significantly less advanced than the HK market average with users

'.'

spending only 54.5% and 54.2% of their time on LTE networks,

average.
This difference is manifested in a different LTE distribution for Ucom
and ClerkenCell than the other HK networks, with a relatively greater
proportion of 1010 and ClerkenCell latencies located in the middle
bucket.

respectively.
'.' On all HK networks we are seeing that users appear to have access
to the LTE network less during peak periods compared to off-peak

The data shows SmarTone and 3 recording the lowest LTE latencies
with ClerkenCell reporting 27.9% higher LTE latencies than the

'.'

ClerkenCell is reporting strong performance for the LTE data


success rate of 90.9% but the overall HK market is quite uniform on

periods.

this metric with 6 networks tied for 1st place.


WiFi Usage Behaviour
Overall we are seeing that, joint with Ucom, ClerkenCell customers
are spending less time connected to WiFi than all other HK

3G Data Performance

networks.

'.'

For all HK users there is a clear trend that consumers are spending

Strong performance from ClerkenCell puts it in second place


with a mean speed of 3.3 Mbps. Ucom leads the way, however

significantly more time on WiFi during off-peak periods compared


to peak periods, indicating that users are more likely to use WiFi

ClerkenCell has an average speed almost 1 Mbps faster than the

connections when at home and during leisure time.

0.5 Mbps.

OpenSignal

next fastest network (PCCW), while Ucom leads ClerkenCell by only

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

EXECUTIVE SUMMARY

OVERVIEW

(CONTINUED)

mm

Ucom

The difference between networks is clear from the bucket


distribution plot, which shows ClerkenCell and Ucom have a
the lowest bucket compared
relatively small proportion of tests inT-VQ_Xk
with all other networks. This leads to the median 3G download
speeds for Ucom and ClerkenCell being even further ahead than

Air-r-

wwww
wwwwwww
www Contacting OpenSignal
www
w

other HK networks. For example, the mean 3G download speed on


ClerkenCell is 49.5% better than the average of other HK networks,

but the median 3G download speed is 84.8% better.

ClerkenCell also exhibits strong 3G upload


LMMspeed performance

[email protected]

tftff

For information on subscribing to Insights for other markets:

wwv

leading the HK market alongside Ucom with speeds of 1.0 Mbps.


PROPORTION
ClerkenCell is in joint third place for 3G latencies with 118ms, Ucom
OF TIME
QjMob
leads with 98 ms.
MEIRICS
Proportion of Time on 3G/LTE

For technical support:


+ 44 845 834 0987

[email protected]
Winning
If you are
interestedNetworks
in publicly disseminating statistics or

other information based on this report, please consult:

WQL Ucom
LMM
mm
AiH" Ucom
WQU m Ucom
m
Q:Mb x-vcfi
[email protected]
For all other queries:

Proportion of Time on LTE

Proportion of Time with No


Network Connection

Proportion of Time on WiFi*

OpenSignal

[email protected]

CLERKENCELL | HONG KONG | JUNE -

AUGUST 2013

EXECUTIVE SUMMARY

RESULTS TABLES
LTE DATA
Winning Networks

MEIRICS
METRICS WON PER
NETWORK

Q:Mob Air+

LTE Download Speed


For each network the table shows
the count of metrics where the
LTE
Upload
network
hasSpeed
achieved the best, or
joint best, average* performance.

Q:Mob

lSm

This gives an immediate view of


LTE Latency

LTE Data Success Rate

Ucom

3G Download Speed

3G Upload Speed

LMN

lmm Ucom

AjH-

<| Air HHl


ucom
+
Air+
Winning Networks

Q:Mob

how well n
etworks ar
e performi
ng
relative to
one anothe
r across all
metrics.
To see whi
ch metrics

Ucom

3G Latency

3G Data Success Rate

LMN

mm cf

page.

*Notes:
Average here refers to the mean performance.
Proportion of time on WiFi does not contribute to the "Metrics
Won per Network" table

OpenSignal

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

EXECUTIVE SUMMARY

RESULTS TABLES

(CONTINUED)

are won by
which netw
orks refer t
o the table
below and
those on th
e following

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

OpenSignal
HOW TO USE THE REPORT

CONTENTS, MAPS
TABLE OF CONTENTS

*
ltjftaI H'H-.-[__

This report comes with an interactive table of contents to help you navigate quickly.

an

To enable this feature:


ifHowiou*Wereport
IfPrcportiixioltlineon3G/UE
ifProportionoMkneonLTE
1?Proportioncfiln-*wlthnontcrt
IfProportionofikneonWifi
IfLTEDownloadSpeed
IfLTEupload5p*W
IfLTElatency
IfLIEDataSuccessRate

rj

Windows: Press
OpenSignalInilght*
i'-l|HONGKONG|JUNE-AUGUST

Mac: Press

ED

and you will see the table of contents


and select "table of contents"

If3GDohtmadSpeed
If3GUploadSpeed
If3GLatency
If3GDaiaSuccessR*ie

Clicking on a section title will take you to it.

IfMeitwdowift

MAPS

Two sets of regions are used in this report: Where the quantity of
data allows we break our averages down over 58 sub-regions of Hong
Kong. The above image shows detail from such a map. These maps
throughout the "Proportion of Time on ..." sections (for example in
Proportion of Time on 3G/LTE).

For the sections concerning throughput, latency and data success


rates, we have divided Hong Kong into 7 sub-regions (which are
aggregates of the 58 regions used in the other maps). One relatively
large region encompasses the areas of low population density.

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

OpenSignal

HOW TO USE THE REPORT

BENCHMARKS, GRAPHS, ANNOTATIONS


BENCHMARKS

This gives the percentage difference


ClerkenCell and the average (mean) of
the other networks in Hong Kong.

89.9%

draw

AVERAGE (MEAN)
TIMEN 3G/LTE

COMPARED TO
THE AVERAGE

IPARED TO>JEX'
NETWORK

"~\
The raw value of the metric, the name
of the metric, and the unit used - in this

The ranking ClerkenCell among all


networks in Hong Kong. If the networks

When ClerkenCell is top we show how


it compares to the next best network.

case %.

are close a draw is indicated.

Otherwise we show the difference


between the benchmarked brand and
the top network for the metric.

The bars show the performance of each HK network


for a given metric. Each bar also displays the 80%
confidence interval (overlaid), which is the range in
which we are confident the true mean lies and is used
to determine if there is a draw.
Close-up of a confidence interval bar.

Higher bars indicate networks where users have access to


3G/LTE network types for a greater proportion of time.
Use the bar height to compare the proportion of time on
3G/LTE for users of each of the networks.
The vertical lines showine range in which we are confident
the true mean lies foiach network.

Tips on how to interpret the graphic.

Methodology Notes
Shows the mean values of proportion of
time on 3G/LTE found across users on
each network.

Explains how the metric is calculated.


For more details on the OpenSignal Methodology
and how this report was created, refer to the
section at the end of this document.

OpenSignal

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

PROPORTION OF TIME ON 3G/LTE

mm

AVERAGE PERFORMANCE

89.9%
AVERAGE (MEAN)
TIME ON 3G/LTE

+11.

DRAW

T-UQBAR

ClerkenCell

UCom

COMPARED TO
THE AVERAGE

UCOM

T-Uqbar

Higher bars indicate networks where users have access to 3G/LTE network
types for a greater proportion of time.
Use the bar height to compare the proportion of time on 3G/LTE for users of
each of the networks.
The vertical lines show the range in which we are confident the true mean lies
for each network. Where two bars have different heights but these lines overlap,
we cannot be certain which network offers most reliable access to 3G & LTE
network types.

OpenSignal

Air+

Cellys

PEAK VS OFF-PEAK

AIR+

LMM

Methodology Notes
Shows the mean values of proportion of
time on 3G/LTE found across users on each
network.
This is calculated by first finding the mean
proportion of time on 3G/LTE for each user
then taking the mean.
The vertical lines show the 80% confidence
interval.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON 3G/LTE

+7. 0%

7%

Q:Mob

100

90

LU

O 80
<n
c

o
E

o
c
.2 70
+
a-

o
o.
o

60

50

Peak

Off-peak

ClerkenCell

T-Uqbar

Cellys

UCom

Air+

LMM

Higher bars indicate more access to 3G/LTE.


Use the bar heights of networks in the left group to compare proportion of time
on 3G/LTE for users of the networks in off-peak times. Similarly, the group on
the right represents reliability found during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for
each network.

OpenSignal

Q:Mob

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean proportion of time on 3G/
LTE is calculated for each network across
each period and shown.
The vertical lines indicate the 80% confidence
interval.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON 3G/LTE

MEDIAN PERFORMANCE

97.7o/o
AVERAGE (MEDIAN)
TIME ON 3G/LTE

+2.
COMPARED TO
THE AVERAGE

0.

0
UCOM

100

75

UJ

m
c
o
a>
50

o
o

Q.
O

T-Uqbar

Highest value
Upper quartile
Median
Lower quartile
Lowest value

Higher means more access to 3G/LTE.


The median (the line through the middle of
the box) can be thought of as showing the
"typical" user experience.
Compare the proportion of time on 3G/LTE
of users on each network by judging the
relative height of the lines.

Cellys

Methodology Notes
These offer another good means of under
standing the typical user experience of a
network, since they discount the effects of a
small number of users having exceptionally
poor or good service. Like "the mean", "the
median" can also be referred to as simply "the
average".

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON 3G/LTE

9
0

PERFORMANCE TRENDS OVER TIME

8
0
7
0

60

70
60

90

LU

60

m
c 90
o
0) 80

E 70
c
o

60
90
80
llerkenCell

UCom

3G (HSPA

|2G

3G

1 No signal

T-Uqbar

70

f~

60
90
80

70
60
90
80
70
60

A line that is going up from left to right represents a network whose average
proportion of time on 3G/LTE has increased consistently across the three month
period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the
true value lies for that month.

OpenSignal

Methodology Notes
For each month and network the average
(mean) proportion of time on 3G/LTE is plot
ted as single point.
The bars indicate the 80% confidence inter
val.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON 3G/LTE

PROPORTION OF TIME BY NETWORK TYPE


100

75

#-
CD

50

CD

'
o
O)
CO
l_

Q)

25

Air+

The different colours in each bar represent the different network types
experienced by users of that network.
The tallest coloured bar in the stack is the network type that, on average, users
experience most on that network.

OpenSignal

Cellys

LMM

Methodology notes
3G devices only - not LTE
In this stacked area plot we show data from
only those users who do not have an LTE
plan. The mean time users spend on each
network type is shown. For a similar plot for
LTE users, see the following section: Propor
tion of Time on LTE.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON 3G/LTE

Q:Mob

CLERKENCELL ABSOLUTE PERFORMANCE

More than 35 %
15 to 35%
5 to 15%

-5 to 5 %
-15 to -5 %
-35 to-15%
Less than -35 %

98-100%
95 - 98 %
90 - 95 %
85 - 90 %
80 - 85 %
70 - 80 %
0 - 70 %

Areas that are a strong green colour represent regions where users experience a
high mean proportion of time on 3G/LTE.
Areas that are a strong red colour represent regions where users experience a
low mean proportion of time on 3G/LTE.

OpenSignal

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) propor
tion of time on 3G/LTE within each area is
calculated.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON 3G/LTE

CLERKENCELL VS OTHER NETWORKS

ClerkenCell

Ucom

T-Uqbar

CMHK

PCCW

SmarTone

Off-peak

90.61

90.58

88.60

87.54

62.11

85.15

81.22

Peak

90.44

87.20

89.10

87.14

62.39

86.15

84.58

ClerkenCell

Ucom

CMHK

T-Uqbar

Lower quartile

87.28

3
86.29

Median

97.65

98.45

96.97

99.96

100

99.91

Upper quartile

PCCW

SmarTone

87.07

82.22

80.68

73.83

97.46

84.26

95.73

97.25

100

98.32

100

100

CMHK

PCCW

89.10

60.93

85.06

81.47

88.41

85.69

58.71

86.48

81.74

92.25

87.99

84.95

57.16

81.28

78.76

91.40

93.26

85.69

85.09

63.44

82.63

78.41

2013-07-23

89.57

86.50

88.29

84.69

60.87

80.28

78.39

2013-08-06

92.61

88.22

88.81

84.46

62.68

81.51

80.98

2013-08-20

88.98

85.02

89.55

86.59

61.09

81.74

76.01

2013-09-03

91.03

87.61

91.94

80.36

61.19

85.59

75.84

ClerkenCell

Ucom

T-Uqbar

2013-05-28

93.03

93.70

87.19

2013-06-11

91.60

94.40

2013-06-25

91.92

2013-07-09

SmarTone

SouthCountry

Areas that are a strong green colour represent regions where users experience
a relatively high mean proportion of time on 3G/LTE compared to the mean for
that area across all other networks.
Areas that are a strong red colour represent regions where users experience a
relatively low mean proportion of time on 3G/LTE compared to the mean for that
area across users of all other networks.

OpenSignal

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) proportion
of time on 3G/LTE within each area is calcu
lated and then compared to the mean of all
other networks.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON 3G/LTE

SOURCE DATA
Average Performance (%)

ClerkenCell
Average time on 3G/LTE

89.94

SmarTone

T-Uqbar
89.02

1.34

83.94

60.03

83.08

78.79

Peak vs Off-peak (%)

Median Performance (%)


Performance Trends over Time (%)

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON LTE

AVERAGE PERFORMANCE

60.8%
AVERAGE (MEAN)
TIME ON LTE

+0.

7%

COMPARED TO
THE AVERAGE

4. 2%
AIR+

ClerkenCell

Air+

UCom

Higher bars indicate networks where users have access to LTE for a greater
proportion of time.
Use the bar height to compare the proportion of time on LTE for users of each of
the networks.
The vertical lines show the range in which we are confident the true mean lies
for each network. Where two bars have different heights but these lines overlap,
we cannot be certain which network offers most reliable access to LTE network
types.

OpenSignal

Cellys

LMM

Methodology Notes
Shows the mean values of proportion of time
on LTE found across users on each network.
This is calculated by first finding the mean
proportion of time on LTE for each user then
taking the mean.
The vertical lines show the 80% confidence
interval.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013


Off-peak

1 ClerkenCell

1 T-Uqbar

Cellys

1 UCom

1 Air+

LMM

PEAK VS OFF-PEAK

80

Q:Mob

PROPORTION OF TIME ON LTE

Peak
Q:Mob

Higher bars indicate more time with access to LTE.


Use the bar heights of networks in the left group to compare proportion of time
on LTE for users of the networks in off-peak times. Similarly, the group on the
right represents the proportion of time spent on LTE during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for
each network.

OpenSignal

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean proportion of time on LTE
is calculated for each network across each
period and shown.
The vertical lines indicate the 80% confidence
interval.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON LTE

MEDIAN PERFORMANCE

63.9%
AVERAGE (MEDIAN)
TIME ON LTE

40100

c
o
o
~

50

c
o
o

Q.
O

25

-0.8%
COMPARED TO
THE AVERAGE

9. 7%
UCOM

ClerkenCell

Highest value
Upper quartile

UCom

Air+

Higher means a greater proportion of time


spent on LTE.
The median (the line through the middle of

Median

the box) can be thought of as showing the

Lower quartile

"typical" user experience.


Compare the proportion of time on LTE

Lowest value

of users on each network by judging the


relative height of the lines.

OpenSignal

Q:Mob

Cellys

Methodology Notes
These offer another good means of under
standing the typical user experience of a
network, since they discount the effects of a
small number of users having exceptionally
poor or good service. Like "the mean", "the
median" can also be referred to as simply "the
average".

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON LTE

PERFORMANCE TRENDS OVER TIME

80
70
60
50

A line that is going up from left to right represents a network whose average
proportion of time on LTE has increased consistently across the three month

Methodology Notes
For each month and network the average
(mean) proportion of time on LTE is plotted as

period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the

single point.
The bars indicate the 80% confidence inter
val.

true value lies for that month.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON LTE

PROPORTION OF TIME BY NETWORK TYPE


100

75

#-
0J
c
O

50

CD

'
o
O)
CO
l_

CD

25

ClerkenCell
3G (HSPA4)

Air+

UCom
12G

LTE

Cellvs

LMM

Q:Mob

3G

No signal

Each coloured bar within a stack represents a particular network technology


experienced by users.
The tallest coloured bar within a stack for a particular network indicates the
network technology that, on average, users experience most.

OpenSignal
-.W

Methodology notes
In this stacked area plot we show data from
only those users who have an LTE plan. For
this set of users the mean time they spend on
each connection type is calculated.

CLERKENCELL I HONG KONG I JUNE -

.fjj5,:'Discovery

Chung-

AUGUST 2013

PROPORTION OF TIME ON LTE

USMuiWo

mxvm&m
SouthCountry

ANCE

H)J8Pui0
S!GE

CLERKENCELL ABSOLUTE PERFORM

85-100%
75 - 85 %
65 - 75 %

Areas that are a strong green colour represent regions where users spend more
time LTE.
Areas that are a strong red colour represent regions where users spend less

55 - 65 %
45 - 55 %
35 - 45 %

time on LTE.

0 - 35 %

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) proportion
of time on LTE within each area is calculated.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUS


T 2013

More than 35 %
15 to 35%
5 to 15%

-5 to 5 %

PROPORTION OF TIME ON LTE

-15 to -5 %
-35 to-15%
Less than -35 %

CLERKENCELL VS OT

HER NETWORKS

Areas that are a strong green colour represent regions where users experience
a relatively high proportion of time on LTE compared to the mean for that area

Methodology Notes
Data is geocoded and segmented by area,
based
on thisPCCW
the average (mean)
proportion
SmarTone
CMHK
of time on LTE within each area is calculated
53.60
63.13
55.80
and then compared to the mean of all other
51.72 networks.
59.78
53.04

across all other networks.


Ucomusers experience
Areas that are a strong red ClerkenCell
colour represent regions where

(!

a relatively low proportion of time


on LTE compared to69.57
the mean for that area65.38
60.86
Off-peak
across users of all other networks.
Peak
58.35
61.86
59.17
ClerkenCell

Ucom

CMHK

SmarTone

PCCW

Lower quartile

45.94

52.35

48.25

38.07

46.42

38.19

Median

63.91

70.80

70.47

56.53

66.01

57.49

80.34

85.30

83.27

74.94

83.24

75.86

Upper quartile

ClerkenCell

Ucom

CMHK

PCCW

SmarTone

2013-05-28

56.87

74.64

62.89

49.38

58.52

61.42

2013-06-11

58.93

69.99

67.58

52.24

66.46

64.80

2013-06-25

58.91

69.10

64.49

50.77

64.90

55.38

2013-07-09

61.69

61.78

67.36

59.77

59.54

56.27

2013-07-23

61.48

62.98

65.97

57.37

61.56

51.70

2013-08-06

62.89

71.34

66.90

60.94

62.92

48.17

2013-08-20

65.35

64.08

68.40

57.44

68.48

51.98

2013-09-03

69.14

63.12

61.45

58.64

61.60

51.15

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON LTE

SOURCE DATA
Average Performance (%)

ClerkenCell
Average time on LTE

60.82

SmarTone
67.81

63.79

Peak vs Off-peak (%)

Median Performance (%)


Performance Trends over Time (%)

54.17

62.32

54.52

CLERKENCELL I HONG KONG I JUNE -

OpenSignal

AUGUST 2013

PROPORTION OF TIME WITH NO NETWORK CONNECTION

AVERAGE PERFORMANCE

DRAW

6.7%

+43.0%
52.

AVERAGE (MEAN)

T-UQBAR

TIME WITH NO NETWORK CONN.

COMPARED TO
THE AVERAGE

UCOM

3%
AIR+

o
<D
C

c 15
o
o
J*

+-

<D

O 10

o
c
o
"
o

hB

Q.
O

ClerkenCell

UCom

T-Uqbar

Air+

Lower bars indicate more reliable performance.


Use the bar height to compare the proportion of time with no network
connection for users of each of the networks.
The vertical lines show the range in which we are confident the true mean lies for
each network. Where two bars have different heights but these lines overlap, we
cannot be certain which network is more reliable.

Cellys

LMM

Q:Mob

Methodology Notes
Shows the mean values of proportion of time
with no network connection found across
users on each network.
This is calculated by first finding the mean
proportion of time with no network connec
tion for each user then taking the mean.
The vertical lines show the 80% confidence.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME WITH NO NETWORK CONNECTION

PEAK VS OFF-PEAK

20

15
O
CD
C

c
o
o
o

CD
C

g 10

CD

E
c

I5
o

Peak

Off-peak

ClerkenCell

T-Uqbar

Cellys

UCom

Air+

LMM

Lower bars indicate more reliable performance.


Use the bar heights of networks in the left group to compare proportion of time
with no network connection for users of the networks in off-peak times. Similarly,
the group on the right represents reliability found during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for
each network.

Q:Mob

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean proportion of time with
no network connection is calculated for each
network across each period and shown.
The vertical lines inidicate the 80% confidence
interval.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME WITH NO NETWORK CONNECTION

MEDIAN PERFORMANCE

5%

0.4%

+41.5%
37.

AVERAGE (MEDIAN)

COMPARED TO
THE AVERAGE

TIME ON LTE

UCOM

50

o
o
c
c
o
o

c
o
c

20

o
c
o

'
oa.
<_ 10

ClerkenCell

T-Uqbar

Air+

Highest value

Lower is more reliable.

Upper quartile

The median (the line through the middle of


the box) can be thought of as showing the

Median

"typical" user experience.

Lower quartile

Compare the proportion of time with no


network connection of users on each
network by judging the relative height of the

Lowest value

lines.

Cellys

Q:Mob

Methodology Notes
These offer another good means of under
standing the typical user experience of a
network, since they discount the effects of a
small number of users having exceptionally
poor or good service. Like "the mean", "the
median" can also be referred to as simply "the
average".

0%

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME WITH NO NETWORK CONNECTION

PERFORMANCE TRENDS OVER TIME

a-

CO

CM
CM

to

to

1
O

m
o

CM

A line that is going down from left to right represents a network whose average
proportion of time with no network connection has decreased consistently
across the three month period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the
true value lies for that month.

m
o
CO

Methodology Notes
For each month and network the average
(mean) proportion of time with no network
connection is plotted as single point.
The bars indicate the 80% confidence inter
val.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME WITH NO NETWORK CONNECTION

CLERKENCELL ABSOLUTE PERFORMANCE

&

Areas that are a strong green colour represent regions where users experience a
low mean proportion of time with no network connection.
Areas that are a strong red colour represent regions where users experience a
high mean proportion of time with no network connection.

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) proportion
of time with no network connection within
each area is calculated.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME WITH NO NETWORK CONNECTION

CLERKENCELL VS OTHER NETWORKS

Areas that are a strong green colour represent regions where users experience a
relatively low mean proportion of time with no network connection compared to
the mean for that area across all other networks.
Areas that are a strong red colour represent regions where users experience a
relatively high mean proportion of time with no network connection compared to
the mean for that area across users of all other networks.

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) proportion
of time with no network connection within
each area is calculated and then compared to
the mean of all other networks.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME WITH NO NETWORK CONNECTION

SOURCE DATA
ClerkenCell

Ucom

T-Uqbar

CMHK

PCCW

SmarTone

Off-peak

6.06

5.98

7.88

10.83

11.15

10.66

14.23

Peak

6.49

9.14

10.73

9.51

11.95

ClerkenCell
Lower quartile
time on no signal
Average
Median
Upper quartile
ClerkenCell

6.13
11.31
Average Performance (%)

Ucom

T-Uqbar

ClerkenCell
0
0.42 6.56

0.63

8.53

11.24

Ucom

7.04

CMHK

T-Uqbar

0.45

7.47

0.77

7.73

T-Uqbar

PCCW
0

14.13

14.81
I

0.50

SmarTone

0
12.94

12.16

0.76
11.54

CMHK

PCCW

12.44

SmarTone

0.85

16.30

16.46
SmarTone

2013-05-28

5.45

4.46

7.77

8.47

12.28

10.15

14.08

2013-06-11

4.49

5.12

6.47

11.46

11.68

9.73

13.46

2013-06-25

5.41

6.53

8.56

12.58

16.08

13.81

16.53

2013-07-09

5.80

5.48

9.09

12.42

13.16

11.06

16.23

2013-07-23

7.04

8.33

6.99

12.92

13.67

14.89

16.17

2013-08-06

4.60

6.49

7.55

12.66

9.26

12.22

14.85

2013-08-20

8.06

9.05

6.58

11.84

14.22

12.43

16.47

2013-09-03

5.45

7.10

6.70
16.90
Peak vs Off-peak (%)

13.28

10.36

18.86

Median Performance (%)


Performance Trends over Time (%)

CLERKENCELL I HONG KONG I JUNE -

OpenSignal

PROPORTION OF TIME ON WIFI

AVERAGE PERFORMANCE
\

32.9%

DRAW

AUGUST 2013

19.7%

23. 6%

AVERAGE (MEAN)
TIME WITH NO NETWORK CONN. /
'

UCOM

COMPARED TO
THE AVERAGE

T-UQBAR

Please note: we do not colour these benchmarks since time spent on WiFi is not necessarily an indicator of network performance.

ClerkenCell

UCom

T-Uqbar

Air+

Higher bars indicate networks whose users spend more time on WiFi.
Use the bar height to compare the proportion of time on WiFi for users of each
of the networks.
The vertical lines show the range in which we are confident the true mean lies for
each network. Where two bars have different heights but these lines overlap, we
cannot be certain on which network users spend more time on WiFi.

OpenSignal

60

LMM

Q:Mob

Methodology Notes
Shows the mean values of proportion of time
on WiFi found across users on each network.
This is calculated by first finding the mean
proportion of time on WiFi for each user then
taking the mean.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON WIFI

PEAK VS OFF-PEAK

Cellys

50

40

c
o
0)

1 30
c
o
o
Q.

O
A- 20

10

Peak

Off-peak

ClerkenCell

T-Uqbar

Cellys

UCom

Air+

LMM

Higher bars indicate more time spent on WiFi.


Use the bar heights of networks in the left group to compare proportion of time
on WiFi for users of the networks in off-peak times. Similarly, the group on the
right represents the proportion of time spent on WiFi during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for
each network.

OpenSignal

Q:Mob

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean proportion of time on WiFi
is calculated for each network across each
period and shown.
The vertical lines indicate the 80% confidence
interval.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON WIFI

MEDIAN PERFORMANCE

-23.o%
38.

21 .2o/o
AVERAGE (MEDIAN)

COMPARED TO
THE AVERAGE

TIME ON WIFI

- 6%
Q:MOB

Please note: we do not colour these benchmarks since time spent on WiFi is not necessarily an indicator of network performance.
100

II
o

a-

CO

CM
CM

to

to

1
O

m
o

ClerkenCell

Highest value
Upper quartile
Median
Lower quartile

CM

UCom

T-Uqbar

Air+

Higher means a greater proportion of time


spent on WiFi.
The median (the line through the middle of
the box) can be thought of as showing the
"typical" user experience.

Lowest value

m
o
CO

Cellys

LMM

Methodology Notes
These offer another good means of under
standing the typical user experience of a
network, since they discount the effects of a
small number of users having exceptionally
poor or good service. Like "the mean", "the
median" can also be referred to as simply "the
average".

OpenSignal

Q:Mob

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON WIFI

PERFORMANCE TRENDS OVER TIME

A line that is going up from left to right represents a network ofr which user's
average proportion of time on WiFi has increased consistently across the three
month period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the
true value lies for that month.

OpenSignal

Methodology Notes
For each month and network the average
(mean) proportion of time on WiFi is plotted
as single point.
The bars indicate the 80% confidence inter
val.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON WIFI

CLERKENCELL ABSOLUTE PERFORMANCE

More than 35 %
15 to 35%
5 to 15%
-5 to 5 %
-15 to -5 %
-35 to-15%
Less than -35 %

Areas that are a dark blue colour represent regions where users spend more
time on WiFi.
Areas that are a light blue represent regions where users spend less time on
WiFi.

OpenSignal

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) proportion
of time on WiFi within each area is calculated.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON WIFI

CLERKENCELL VS OTHER NETWORKS

ClerkenCell

Ucom

T-Uqbar

CMHK

PCCW

SmarTone

Off-peak

34.17

33.23

39.49

40.66

43.78

39.89

47.26

Peak

21.28

19.57

30.47

29.56

34.71

29.82

36.74

Ucom

ClerkenCell

T-Uqbar

CMHK

PCCW

SmarTone

Lower quartile

5.25

7.69

7.46

7.31

8.24

6.47

13.13

Median

21.12

19.57

27.33

25.33

33.31

25.27

33.98

44.14

40.4

55.78

52.06

63.04

57.80

67.55

Upper quartile

CMHK

PCCW

33.65

40.08

33.87

40.35

39.29

32.92

42.92

34.05

43.53

27.58

41.86

33.56

40.86

35.22

40.86

29.14

27.74

35.56

36.41

37.44

31.64

43.07

2013-07-23

31.21

30.04

37.39

39.99

39.91

34.85

45.57

2013-08-06

30.97

41.18

36.87

38.89

40.90

39.03

42.96

2013-08-20

31.09

34.01

42.64

39.75

43.40

39.41

46.55

2013-09-03

33.07

38.24

45.14

35.65

41.07

35.57

37.18

ClerkenCell

Ucom

T-Uqbar

2013-05-28

29.31

26.22

38.35

2013-06-11

34.64

26.61

2013-06-25

25.65

2013-07-09

Areas that are a dark blue colour represent regions where users experience a
relatively high proportion of time on WiFi compared to the mean for that area
across all other networks.
Areas that are a light blue colour represent regions where users experience a
relatively low proportion of time on WiFi compared to the mean for that area
across users of all other networks.

OpenSignal

SmarTone

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) proportion
of time on WiFi within each area is calculated
and then compared to the mean of all other
networks.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

PROPORTION OF TIME ON WIFI

SOURCE DATA
Average Performance (%)

ClerkenCell

T-Uqbar

SmarTone

Average time on wifi

33.09

32.47

42.92

40.62

42.75

39.

47.51

Peak vs Off-peak (%)

Median Performance (%)


Performance Trends over Time (%)

CLERKENCELL I HONG KONG I JUNE -

OpenSignal

AUGUST 2013

LTE DOWNLOAD SPEED

AVERAGE PERFORMANCE
Methodology Notes
Shows the mean values of download speed
found across speed tests taken over LTE on
each network.
The vertical lines show the 80% confidence
interval.

20.7 Mbps
AVERAGE (MEAN)
LTE DOWNLOAD SPEED

-1.0%
COMPARED TO
THE AVERAGE

DRAW

CELLYS

7. 7%
AIR+

UCOM

20

(fl

Q.

XX

w15
o
0)
0)
a.
w

T3
10

o
o

ClerkenCell

Ucom

Air+

Cellys

LMM

Q:Mob

Higher bars indicate faster performance.


Use the bar height to compare the LTE download speed of the networks.
The vertical lines show the range in which we are confident the true mean lies for
each network. Where two bars have different heights but these lines overlap, we
cannot be certain which network is faster.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE DOWNLOAD SPEED

PEAK VS OFF-PEAK

Off-peak

Peak

ClerkenCell

Ucom

T-Uqbar

Higher bars indicate faster performance.


Use the bar heights of networks in the left group to compare LTE download
speed of networks in off-peak times. Similarly, the group on the right represents
speeds found during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for

Air+

LMM

Q:Mob

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean download speed of tests
taken over LTE for each network across each
period is shown.
The vertical lines show the 80% confidence
interval.

each network.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE DOWNLOAD SPEED

CUMULATIVE FREQUENCY
100

75

- ClerkenCel

o
c

0)
3
0)

50

0)

- Air+
Cellys

>

4-

re

Q:Mob

3
3

25

10

20

40

30

Download speed (Mbps)

Generally, shallower lines show a greater likelihood that a user will get a fast
download speed than steep lines.
To find the percentage of tests run on LTE on a particular network that achieve
less than a certain download speed (for example 5 Mbps), look at where the

50

Methodology Notes
Here we represent the full download speed
distribution from speedtests run over LTE on
each network.

line for the network is above the point (5 Mbps). For example, if it is at 10% then
10% of tests willl achieve this download speed or less and 90% will achieve
higher speeds.

OpenSignal

CLERKENCELL I HONG KONG I JUNE -

AUGUST 2013

LTE DOWNLOAD SPEED

BUCKET DISTRIBUTION OF PERFORMANCE


Use the bar heights to see what percentage of LTE tests fall within a particular
bracket of download speed.
A good distribution will have a high right bucket (>20 Mbps). For example, if this
bucket is greater than 50% then this indicates that more than half the tests run
on LTE for the network are over 20 Mbps.
A high left bucket indicates a large proportion of tests on LTE achieving speeds
of less than 5 Mbps.

Methodology Notes
Each bucket shows the proportion of speedtests run that fall within that range of down
load speed, giving a simple way of comparing
the speed distributions.
The percentage value of the bucket is the
likelihood that a speedtest run will fall within
that download speed bracket.

60

40

20
0
60
40
20
0
60

0)

5 o
c 60
O

>
+

<5 Mbps

>20 Mbps

5-20 Mbps

Download speed (Mbps)

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE DOWNLOAD SPEED

MEDIAN PERFORMANCE
\

19.7 Mbps
AVERAGE (MEDIAN)
LTE DOWNLOAD SPEED

100

S> 75
Q.

T3
0)
CD
Q.

(fl
re ou
o
c
o
Q

25

+2.!

5%

COMPARED TO
THE AVERAGE

10. 5%
Q:MOB

ClerkenCell

Highest value
Upper quartile
Median
Lower quartile
Lowest value

Air+

Ucom

Higher is faster.
The median (the line through the middle of
the box) can be thought of as showing the
"typical" user experience.
Compare the download speed of a typical
test over LTE on a network by judging the
relative height of the lines.

OpenSignal

LMM

Cellys

Q:Mob

Methodology Notes
These offer another good means of under
standing the typical user experience of a
network, since they discount the effects of a
small number of users having exceptionally
poor or good service. Like "the mean", "the
median" can also be referred to as simply "the
average".

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE DOWNLOAD SPEED

PERFORMANCE TRENDS OVER TIME


30
O

25

ID

25
20
15
30

+-

c
o
o

June 2013

Julv2013

A line that is going up from left to right represents a network whose average
download speed has increased consistently across the three month period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the
true value lies for that month.

OpenSignal

Auqust 2013

Methodology Notes
For each month and network the average
(mean) LTE download speed is plotted as
single point.
The bars indicate the 80% confidence inter
val.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE DOWNLOAD SPEED

CLERKENCELL ABSOLUTE PERFORMANCE

Areas that are a strong green colour represent regions of hig


h mean download
speed for speedtests run over LTE.

More than 30 mbps


25 - 30 mbps
20 - 25 mbps

Areas that are a strong red colour represent regions of low mean download
speed for speedtests run over LTE.

16-20 mbps
13-16 mbps
10-13 mbps
Less than 10 mbps

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) LTE down
load speed within each area is calculated.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUS


T 2013

More than 35 %
15 to 35%

5 to 15%
-5 to 5 %

LTE DOWNLOAD SPEED

-15 to-5%
-35 to-15%
Less than -35 %

CLERKENCELL VS OT

HER NETWORKS

Areas that are a strong green colour represent regions of relatively high mean

Methodology Notes
Data is geocoded and segmented by area,
the average (mean)Q:Mob
LTE down
based
on thisLMM
Cellys
load speed within each area is calculated
17.94
16.51
23.96
and then compared to the mean of all other
21.06 networks. 21.42
22.90

download speed for speedtests run over LTE compared to the mean in that area
found for the other networks.
Ucom
Air+
ClerkenCell
Areas that are a strong red colour represent regions of relatively low mean
20.90
19.30
Off-peak
download speed for speedtests
run over LTE compared
to the mean found21.38
in
that
Peak
area across all other networks.
20.45

22.52

24.18

Ucom

ClerkenCell

CMHK

PCCW

SmarTone

<5 Mbps

7.11%

7.18%

2.99%

16.48%

6.79%

9.42%

5 - 20 Mbps

41.78%

43.65%

38.92%

35.16%

47.22%

53.5%

>20Mbps

51.11%

49.17%

58.08%

48.35%

45.99%

37.08%

ClerkenCell
OpenSignal

Ucom

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013


Air+

LMM

Cellys

Q:Mob

Lower quartile

12.54

13.42

11.00

8.47

9.82

12.68

Median

19.74

20.38

18.53

19.02

16.34

22.05

28.40

25.29
LMM

33.43

Upper quartile

27.68
ClerkenCell

2013-06-01

27.96

31.75
Air+

Ucom

17.65

2013-07-01

22.09

2013-08-01

21.27

LTE DOWNLOAD SPEE

Cellys

Q:Mob

16.01

18.40

Notes 23.15
Methodology
20.41
16.78
Shows the mean values of download speed
22.84
24.25
21.61
found across speed tests taken over LTE on
each network.
The vertical lines show the 80% confidence
interval.

16.13

18.9

19.56

24.10

18.68

16.86

27.30

SOURCE DATA
Average Performance (Mbps)
ClerkenCell
Download speed (Mbps)

20.69

20.81

22.41

18.97

Peak vs Off-peak (Mbps)

OpenSignal

Bucket Distribution of Performance


Median Performance (Mbps)
Performance Trends over Time (Mbps)
CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE UPLOAD SPEED

AVERAGE PERFORMANCE

18.73

23.62

6.4

DRAW

Mbps

AVERAGE (MEAN)
LTE UPLOAD SPEED

AIR+

13. 2%

-4.7%
CELLYS

COMPARED TO
THE AVERAGE

LMM

(A

Q.

XX

D
<U
0)

Q.

(fl
T3
CC
O

Q.

z>

ClerkenCell

Ucom

Air+

Cellys

LMM

Higher bars indicate faster performance.


Use the bar height to compare the LTE upload speed of the networks.
The vertical lines show the range in which we are confident the true mean lies for
each network. Where two bars have different heights but these lines overlap, we
cannot be certain which network is faster.

OpenSignal

LTE UPLOAD SPEED

PEAK VS OFF-PEAK

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

Q:Mob

ClerkenCell

Higher bars indicate faster performance.


Use the bar heights of networks in the left group to compare LTE upload speed
of networks in off-peak times. Similarly, the group on the right represents speeds
found during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for
each network.

OpenSignal

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean download speed of tests
taken over LTE for each network across each
period is shown.
The vertical lines show the 80% confidence
interval.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE UPLOAD SPEED

CUMULATIVE FREQUENCY
100

Q:Mob

T-Uqbar

75

n
- ClerkenCel

o
c

0)
3
0)

- Air+

50

0)

Cellys

>
re

Q:Mob

25

10

Upload speed (Mbps)

Generally, shallower lines show a greater likelihood that a user will get a fast
upload speed than steep lines.
To find the percentage of tests run on LTE on a particular network that achieve
less than a certain upload speed (for example 5 Mbps), look at where the line for
the network is above the point (5 Mbps). For example, if it is at 10% then 10% of
tests willl achieve this upload speed or less and 90% will achieve higher speeds.

OpenSignal

20

30

Methodology Notes
Here we represent the full download speed
distribution from speedtests run over LTE on
each network.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE UPLOAD SPEED

BUCKET DISTRIBUTION OF PERFORMANCE


Use the bar heights to see what percentage of LTE tests fall within a particular

bracket of upload speed.


A good distribution will have a high right bucket (>10 Mbps). For example, if this
bucket is greater than 50% then this indicates that more than half the tests run
on LTE for the network are over 10 Mbps.
A high left bucket indicates a large proportion of tests on LTE achieving speeds
of less than 5 Mbps.

<5 Mbps

OpenSignal

5-10 Mbps

Upload speed (Mbps)

Methodology Notes
Each bucket shows the proportion of speedtests run that fall within that range of down
load speed, giving a simple way of comparing
the speed distributions.
The percentage value of the bucket is the
likelihood that a speedtest run will fall within
that download speed bracket.

>10 Mbps

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE UPLOAD SPEED

MEDIAN PERFORMANCE

6.2

+2.

Mbps

AVERAGE (MEDIAN)
LTE UPLOAD SPEED

8. 9%

4%

COMPARED TO
THE AVERAGE

Q:MOB

30

m
a.
xx
o
<D 20
0)
Q.
V)
T3
re
o
Q.

10

ClerkenCell

Highest value
Upper quartile
Median

Ucom

Air+

Higher is faster.
The median (the line through the middle of
the box) can be thought of as showing the
"typical" user experience.

Lower quartile

Compare the upload speed of a typical


test over LTE on a network by judging the

Lowest value

relative height of the lines.

LMM

Cellys

Q:Mob

Methodology Notes
These offer another good means of under
standing the typical user experience of a
network, since they discount the effects of a
small number of users having exceptionally
poor or good service. Like "the mean", "the
median" can also be referred to as simply "the
average".

OpenSignal

CLERKENCELL I HONG KONG I JUNE -

AUGUST 2013

LTE UPLOAD SPEED

1
0

PERFORMANCE TRENDS OVER TIME

8
6
4

10

4
10
O
ID

c
o
o

More than 10 mbps


8 -10 mbps

5-8 mbps
3-5 mbps
2 - 3 mbps
1-2 mbps

Less than 1 mbps

June 2013

July 2013

A line that is going up from left to right represents a network whose average
upload speed has increased consistently across the three month period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the
true value lies for that month.

OpenSignal

LTE UPLOAD SPEED

August 2013

Methodology Notes
For each month and network the average
(mean) LTE download speed is plotted as
single point.
The bars indicate the 80% confidence inter
val.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

CLERKENCELL ABSOLUTE PERFORMANCE

More than 35 %
15 to 35%

5 to 15%
-5 to 5 %
-15 to-5%
-35 to-15%
Less than -35 %

Areas that are a strong green colour represent regions of high mean upload
speed for speedtests run over LTE.
Areas that are a strong red colour represent regions of low mean upload speed
for speedtests run over LTE.

OpenSignal

LTE UPLOAD SPEED

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) LTE down
load speed within each area is calculated.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

CLERKENCELL VS OTHER NETWORKS

ClerkenCell

Ucom

Air+

Cellys

LMM

Q:Mob

Off-peak

6.86

7.19

5.63

6.73

6.57

7.11

Peak

5.83

6.42

6.79

5.30

8.33

7.69

Ucom

ClerkenCell

CMHK

PCCW

SmarTone

<5 Mbps

39.81%

40.46%

36.73%

45.95%

49.42%

32.06%

5 -10 Mbps

34.95%

38.82%

35.37%

35.81%

28.4%

40.77%

>10 Mbps

25.24%

20.72%

27.89%

18.24%

22.18%

27.18%

ClerkenCell

Air+

Ucom

LMM

Cellys

Q:Mob

Lower quartile

3.33

3.77

2.52

2.25

4.07

3.39

Median

6.15

6.27

5.06

5.45

6.51

6.75

8.13

10.61
LMM

10.26

Upper quartile

9.13
ClerkenCell

9.98

9.16
Air+

Ucom

Cellys

Q:Mob

2013-06-01

6.92

5.82

4.74

5.18

6.80

9.27

2013-07-01

7.61

8.06

5.56

7.59

7.64

5.71

2013-08-01

5.79

7.00

6.62

7.45

7.61

7.27

Areas that are a strong green colour represent regions of relatively high mean
upload speed for speedtests run over LTE compared to the mean in that area
found for the other networks.
Areas that are a strong red colour represent regions of relatively low mean
upload speed for speedtests run over LTE compared to the mean found in that
area across all other networks.

OpenSignal

LTE UPLOAD SPEED

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) LTE down
load speed within each area is calculated
and then compared to the mean of all other
networks.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

SOURCE DATA
Average Performance (Mbps)
ClerkenCell
6.43

Upload speed (Mbps)

6.81

6.10

6.16

7.41

7.23

Peak vs Off-peak (Mbps)

Bucket Distribution of Performance


Median Performance (Mbps)
Performance Trends over Time (Mbps)
CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

OpenSignal

LTE LATENCY

AVERAGE PE

RFORMANCE
Methodology Notes
Shows the mean values of latency found
across speed tests taken over LTE on each
network.
The vertical lines show the 80% confidence
interval.

62.6

ms

AVERAGE (MEAN)
LTE LATENCY

60

,-.40
CO

E,
>.
o
c

0)
IB

4-

20

27. 9%
COMPARED TO
THE AVERAGE

43. 1%
AIR+

ClerkenCell

Air+

Ucom

Cellys

LMM

Lower bars indicate faster performance.


Use the bar height to compare the LTE latency of the networks.
The vertical lines show the range in which we are confident the true mean lies for
each network. Where two bars have different heights but these lines overlap, we
cannot be certain which network is faster.

OpenSignal

LTE LATENCY

PEAK VS OFF-PEAK

60

40

(fl
E,
>.

o
c
0)
4-

20

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

Q:Mob

Peak

Off-peak
ClerkenCell

Ucom

T-Uqbar

Lower bars indicate faster performance.


Use the bar heights of networks in the left group to compare LTE latency of
networks in off-peak times. Similarly, the group on the right represents speeds
found during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for
each network.

OpenSignal

Air+

LMM

Q:Mob

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean latency of tests taken over
LTE for each network across each period is
shown.
The vertical lines show the 80% confidence
interval.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE LATENCY

CUMULATIVE FREQUENCY
100

75

- ClerkenCel

o
c

0)
3
0)

50

0)

- Air+
Cellys

>

4-

re

Q:Mob

3
3

25

100

Latency (ms)

Generally, steeper lines show a greater likelihood that a user will get a fast
latency than shallow lines.
To find the percentage of tests run on LTE on a particular network that achieve
less than a certain latency (for example 5 ms), look at where the line for the
network is above the point (5 ms). For example, if it is at 10% then 10% of tests

200

300

Methodology Notes
Here we represent the full latency distribu
tion from speedtests run over LTE on each
network.

willl achieve this latency or less and 90% will achieve higher latencies.

OpenSignal

CLERKENCELL I HONG KONG I JUNE -

AUGUST 2013

LTE LATENCY

BUCKET DISTRIBUTION OF PERFORMANCE


Use the bar heights to see what percentage of LTE tests fall within a particular
bracket of latency.
A good distribution will have a high left bucket (<40 ms). For example, if this
bucket is greater than 50% then this indicates that more than half the tests run
on LTE for the network are less than 40 ms.
A high right bucket indicates a large proportion of tests on LTE achieving speeds
of greater than 80 ms.

Methodology Notes
Each bucket shows the proportion of speedtests run that fall within that range of latency,
giving a simple way of comparing the speed
distributions.
The percentage value of the bucket is the
likelihood that a speedtest run will fall within
that latency bracket.

60
40
20
0
60
40
20
0
60

- 20
0

o
o
3

60
40
20
0
60
40
20
0
40 - 80 ms

<40 ms

>80ms

Latency (ms)

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE LATENCY
ClerkenCell

Ucom

Highest value

Upper quartile
Median

57.0

The median (the line through the middle


'.'
of the box) can be thought of as showing the
'.'
Compare the latency of a typical test
er LTE on a network
by judging the relative
height of the lines.

ms

AVERAGE (MEDIAN)
LTE LATENCY

200

150

tn

E
>.
o
100
re

50

Lower is faster.

MEDIAN PERFORMANC
E

"typical" user experience.

Lower quartile
Lowest value

Air+

48.1%
COMPARED TO
THE AVERAGE

70. 1%
LMM

LMM

Q:Mob

Methodology Notes
These offer another good means of under
standing the typical user experience of a
network, since they discount the effects of a
small number of users having exceptionally
poor or good service. Like "the mean", "the
median" can also be referred to as simply "the
average".

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE LATENCY

PERFORMANCE TRENDS OVER TIME


60

50

40

Cellys

June 2013

July 2013

A line that is going down from left to right represents a network whose average
latency has decreased consistently across the three month period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the
true value lies for that month.

OpenSignal

August 2013

Methodology Notes
For each month and network the average
(mean) LTE latency is plotted as single point.
The bars indicate the 80% confidence inter
val.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE LATENCY

CLERKENCELL ABSOLUTE PERFORMANCE

Areas that are a strong green colour represent regions of low mean latency for

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) LTE latency
within each area is calculated.

speedtests run over LTE.


Areas that are a strong red colour represent regions of high mean latency for
speedtests run over LTE.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

More than 35 %

LTE LATENCY

15 to 35%

5 to 15%
-5 to 5 %

CLERKENCELL VS OT

-15 to -5 %
-35 to-15%
Less than -35 %

HER NETWORKS

Areas that are a strong green colour represent regions of relatively low mean

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) LTE latency
within each area is calculated and then com
pared to the mean of all other networks.

latency for speedtests run over LTE compared to the mean in that area found for
the other networks.
Areas that are a strong red colour represent regions of relatively high mean
latency for speedtests run over LTE compared to the mean found in that area
across all other networks.

DRAW

ClerkenCell

Ucom

Air+

Off-peak

64.96

60.37

43.41

Peak

59.70

53.68

44.65

<40 ms
OpenSignal

1/6

LMM

Cellys

-0.1%
Q:Mob

46.36
5 52.58
OTHER NET45.56 COMPARED
TO
53.42
45.11 THE AVERAGE
45.98
WORKS

Ucom

ClerkenCell

CMHK

PCCW

SmarTone

56%

54%

57%

70%

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

33%

20%

40 - 80 ms

55%

59%

37%

40%

35%

24%

>80ms

12%

21%

6%

5%

8%

6%

ClerkenCell
Lower quartile
LTE
LATENCY
Median
Upper quartile

LMM

Cellys

Q:Mob

42

37

25

34.5

30

31

57

48

35

39

33.5

37

74.75
ClerkenCell

2013-06-01

Air+

Ucom

70

54

56.74

2013-07-01

64.54

2013-08-01

58.51

SOURCE DATA

53

Air+

Ucom
57.47

48
LMM

Cellys
39.66

52
Q:Mob

55.25

38.94

41.12

Methodology
51.66 Notes
45.03
42.82
Shows the mean values of data success rate
51.52
40.53
43.28
found across speed tests taken over LTE on
each network.
The vertical lines show the 80% confidence
interval.

35.06

49.91

41.42

41.42

Average Performance (ms)

ClerkenCell
Latency (ms)

62.57

56.89

43.72

52.69

Peak vs Off-peak (ms)

OpenSignal

Bucket Distribution of Performance


Median Performance (Mbps)
Performance Trends over Time (Mbps)

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE DATA SUCCESS RATE

AVERAGE PERFORMANCE

45.19

46.19

90.9

0. 3%

J/o

AVERAGE (MEAN)
LTE DATA SUCCESS RATE

90

0)

re 85

k.
(A
(A
0)
O

o
3

(fl
re
f-i
re

Q 80

75
ClerkenCell

Ucom

Air+

Cellys

LMM

Higher bars indicate more reliable performance.


Use the bar height to compare the LTE data success rate of the networks.
The vertical lines show the range in which we are confident the true mean lies for
each network. Where two bars have different heights but these lines overlap, we
cannot be certain which network is more reliable.

OpenSignal

LTE DATA SUCCESS RATE

PEAK VS OFF-PEAK

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

Q:Mob

90

2 85
in
in
0)

o
o
3

(fl
re
f-i
re
o

80

75
Peak

Off-peak
ClerkenCell

Ucom

T-Uqbar

Higher bars indicate more reliable performance.


Use the bar heights of networks in the left group to compare LTE data success
rate of networks in off-peak times. Similarly, the group on the right represents
speeds found during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for
each network.

OpenSignal

Air+

LMM

Q:Mob

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean data success rate of tests
taken over LTE for each network across each
period is shown.
The vertical lines show the 80% confidence
interval.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

LTE DATA SUCCESS RATE

95
90

PERFORMANCE TRENDS OVER TIME

85
80
75
95
90

85
80
75
95

goo
0J85

re

OC 80
(A
$75

O 95
OT90

re

re 85
Q
80

O
o

75
95
90

c
o
o
3

85
80
75
95
90

>

85
80
75

;:.
*

H
5
o
June 2013

July 2013

A line that is going up from left to right represents a network whose average
data success rate has increased consistently across the three month period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the
true value lies for that month.

August 2013

Methodology Notes
For each month and network the average
(mean) LTE data success rate is plotted as
single point.
The bars indicate the 80% confidence inter
val.

OpenSignal

LTE DATA SUCCESS RATE

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

CLERKENCELL ABSOLUTE PERFORMANCE

98-100%
96 - 98 %
94 - 96%
92 - 94 %
89 - 92 %
86 - 89 %
0 - 86 %

Areas that are a strong green colour represent regions of high mean data
success rate for speedtests run over LTE.
Areas that are a strong red colour represent regions of low mean data success
rate for speedtests run over LTE.

OpenSignal

LTE DATA SUCCESS RATE

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) LTE data
success rate within each area is calculated.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

CLERKENCELL VS OTHER NETWORKS

ClerkenCell

Ucom

Air+

Cellys

LMM

Q:Mob

Off-peak

89.95

91.38

90.52

90.16

90.97

91.55

Peak

92.26

90.61

92.59

90.46

91.51

90.69

ClerkenCell

Air+

Ucom

LMM

Cellys

Q:Mob

2013-06-01

91.62

91.94

92.69

90.80

89.87

90.00

2013-07-01

88.67

91.07

88.92

88.98

88.80

89.09

2013-08-01

91.76

90.47

91.71

90.64

92.77

92.57

Areas that are a strong green colour represent regions of relatively high mean
data success rate for speedtests run over LTE compared to the mean in that
area found for the other networks.
Areas that are a strong red colour represent regions of relatively low mean data
success rate for speedtests run over LTE compared to the mean found in that
area across all other networks.

OpenSignal

LTE DATA SUCCESS RATE

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) LTE data
success rate within each area is calculated
and then compared to the mean of all other
networks.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

SOURCE DATA
Average Performance (%)

ClerkenCell
Data success rate (%)

90.91

91.01

91.U

90.25

Peak vs Off-peak (%)

Performance Trends over Time (%)

Methodology Notes
Shows the mean values of download speed
found across speed tests taken over 3G on
each network.
The vertical lines show the 80% confidence
interval.

OpenSignal

CLERKENCELL I HONG KONG I JUNE -

3G DOWNLOAD SPEED

AVERAGE PERFORMANCE

AUGUST 2013

91.18

91.23

U.3 Mbps
AVERAGE (MEAN)
3G DOWNLOAD SPEED

ClerkenCell

Ucom

T-Uqbar

14. 7%

+20. 7%
COMPARED TO
THE AVERAGE

Air+

UCOM

LMM

Higher bars indicate faster performance.


Use the bar height to compare the 3G download speed of the networks.
The vertical lines show the range in which we are confident the true mean lies for
each network. Where two bars have different heights but these lines overlap, we
cannot be certain which network is faster.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G DOWNLOAD SPEED

PEAK VS OFF-PEAK

Q:Mob

a.
xx
D
0)
0)

a.
in

c
o

Peak

Off-peak
ClerkenCell

Ucom

T-Uqbar

Higher bars indicate faster performance.


Use the bar heights of networks in the left group to compare 3G download
speed of networks in off-peak times. Similarly, the group on the right represents
speeds found during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for
each network.

OpenSignal

LMM

CUMULATIVE FREQUENCY

Q:Mob

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean download speed of tests
taken over 3G for each network across each
period is shown.
The vertical lines show the 80% confidence
interval.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G DOWNLOAD SPEED

100

Air+

75

- ClerkenCel

o
c

0)
3
0)

50

- Air+

0)

>

4-

re

Q:Mob

3
3

25

0.0

2.5

5.0

Download speed (Mbps)

Generally, shallower lines show a greater likelihood that a user will get a fast
download speed than steep lines.
To find the percentage of tests run on LTE on a particular network that achieve
less than a certain download speed (for example 5 Mbps), look at where the

7.5

10.0

Methodology Notes
Here we represent the full download speed
distribution from speedtests run over 3G on
each network.

line for the network is above the point (5 Mbps). For example, if it is at 10% then
10% of tests willl achieve this download speed or less and 90% will achieve
higher speeds.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G DOWNLOAD SPEED

BUCKET DISTRIBUTION OF PERFORMANCE


Use the bar heights to see what percentage of 3G tests fall within a particular
bracket of download speed.

A good distribution will have a high right bucket (>3 Mbps). For example, if this
bucket is greater than 50% then this indicates that more than half the tests run
on 3G for the network are over 3 Mbps.
A high left bucket indicates a large proportion of tests on 3G achieving speeds
of less than 1 Mbps.

Methodology Notes
Each bucket shows the proportion of speedtests run that fall within that range of down
load speed, giving a simple way of comparing
the speed distributions.
The percentage value of the bucket is the
likelihood that a speedtest run will fall within
that download speed bracket.

75

<D

50
3

25
0
100

C 100
r

o
a.
2

Q.

75
50
25
0
100
75

75
50

25
0
<1 Mbps

OpenSignal

1 - 3 Mbps

Download speed (Mbps)

>3 Mbps

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G DOWNLOAD SPEED

MEDIAN PERFORMANCE

3.0

Mbps

AVERAGE (MEDIAN)
3G DOWNLOAD SPEED

+34. 6%

-5. 4%

COMPARED TO
THE AVERAGE

UCOM

15
in

xx
T3

CD
CD

"10
T3

re
o
c

o
Q

IT
ClerkenCell

Highest value
Upper quartile
Median

Ucom

T-Uqbar

Higher is faster.
The median (the line through the middle of
the box) can be thought of as showing the
"typical" user experience.

Lower quartile

Compare the download speed of a typical


test over 3G on a network by judging the

Lowest value

relative height of the lines.

Air+

LMM

Methodology Notes
These offer another good means of under
standing the typical user experience of a
network, since they discount the effects of a
small number of users having exceptionally
poor or good service. Like "the mean", "the
median" can also be referred to as simply "the
average".

OpenSignal

Q:Mob

CLERKENCELL I HONG KONG I JUNE -

AUGUST 2013

3G DOWNLOAD SPEED

PERFORMANCE TRENDS OVER TIME

More than 8 mb

ps
5 8 mbps

3 5 mbps

2 3 mbps
1 2 mbps

0.5 1 mbps

Less than 0.5


bps

A line that is going up from left to right represents a network whose average
download speed has increased consistently across the three month period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the
true value lies for that month.

OpenSignal

Methodology Notes
For each month and network the average
(mean) 3G download speed is plotted as
single point.
The bars indicate the 80% confidence inter
val.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G DOWNLOAD SPEED

CLERKENCELL ABSOLUTE PERFORMANCE

More than 35 %

15 to 35%
5 to 15%
-5 to 5 %
-15 to -5 %
-35 to-15%
Less than -35 %

Areas that are a strong green colour represent regions of high mean download
speed for speedtests run over 3G.
Areas that are a strong red colour represent regions of low mean download
speed for speedtests run over 3G.

OpenSignal

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) 3G down
load speed within each area is calculated.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G DOWNLOAD SPEED

CLERKENCELL VS OTHER NETWORKS


5-X

If

M J-

ClerkenCell

Ucom

T-Uqbar

Air+

LMM

Q:Mob

Off-peak

3.27

3.80

0.32

2.80

3.29

3.65

Peak

3.32

4.05

0.32

2.35

3.37

3.14

Ucom
<1 Mbps

ClerkenCell
15.82%

1 - 3 Mbps

31.65%

>3 Mbps

52.53%
ClerkenCell

T-Uqbar

PCCW

SmarTone

17.4%

100%

40.81%

37.85%

42.28%

32.37%

23.73%

32.49%

24.06%

50.23%

35.46%

29.66%

33.66%

Ucom

Air+

T-Uqbar

LMM

Q:Mob

Lower quartile

1.45

1.50

0.28

0.88

1.18

1.45

Median

2.99

3.16

0.33

1.91

2.68

3.03

0.37

4.10
Air+

4.69
LMM

Upper quartile

4.90
ClerkenCell

6.22
Ucom

T-Uqbar

5.11
Q:Mob

2013-06-01

3.85

4.57

0.31

2.31

2.76

2013-07-01

3.23

4.03

0.31

2.67

2.07

1.67

2013-08-01

3.03

3.44

0.32

2.09

2.70

2.87

Areas that are a strong green colour represent regions of relatively high mean
download speed for speedtests run over 3G compared to the mean in that area
found for the other networks.
Areas that are a strong red colour represent regions of relatively low mean
download speed for speedtests run over 3G compared to the mean found in
that area across all other networks.

OpenSignal

3G DOWNLOAD SPEED

SOURCE DATA

2.62

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) 3G down
load speed within each area is calculated
and then compared to the mean of all other
networks.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

ClerkenCell

Average Performance (Mbps)

T-Uqbar
Download speed (Mbps)

3.28

0.32

2.65

3.30

3.48

Peak vs Off-peak (Mbps)

Bucket Distribution of Performance


Median Performance (Mbps)
Performance Trends over Time (Mbps)
CLERKENCELL I HONG KONG I JUNE -

OpenSignal

AUGUST 2013

3G UPLOAD SPEED

AVERAGE PERFORMANCE
\

0.9 Mbps

AVERAGE (MEAN)
3G UPLOAD SPEED

1.00
(fl

Q.

XX

e.0-75
D
0)
0)

a.

in

TO 0.50
O

a.

0.25

+2.
COMPARED TO
THE AVERAGE

21. 2%
AIR+

0.00
ClerkenCell

Ucom

T-Uqbar

Higher bars indicate faster performance.


Use the bar height to compare the 3G upload speed of the networks.
The vertical lines show the range in which we are confident the true mean lies for
each network. Where two bars have different heights but these lines overlap, we
cannot be certain which network is faster.

OpenSignal

3G UPLOAD SPEED

PEAK VS OFF-PEAK

1.0

Q.

xx
D

0)
0)
Q.
in

a
re
o

Q.
0.5

Air+

LMM

Q:Mob

Methodology Notes
Shows the mean values of upload speed
found across speed tests taken over 3G on
each network.
The vertical lines show the 80% confidence
interval.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

0.0
Peak

Off-peak
ClerkenCell

Ucom

T-Uqbar

Higher bars indicate faster performance.


Use the bar heights of networks in the left group to compare 3G upload speed
of networks in off-peak times. Similarly, the group on the right represents speeds
found during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for
each network.

OpenSignal

Air+

LMM

Q:Mob

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean download speed of tests
taken over 3G for each network across each
period is shown.
The vertical lines show the 80% confidence
interval.

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G UPLOAD SPEED

CUMULATIVE FREQUENCY
100

75

- ClerkenCel

o
c

0)
3
0)

50

- Air+

0)

>

4-

re

Q:Mob

3
3

25

Upload speed (Mbps)


Generally, shallower lines show a greater likelihood that a user will get a fast
upload speed than steep lines.
To find the percentage of tests run on LTE on a particular network that achieve
less than a certain upload speed (for example 5 Mbps), look at where the line for
the network is above the point (5 Mbps). For example, if it is at 10% then 10% of

Methodology Notes
Here we represent the full upload speed
distribution from speedtests run over 3G on
each network.

tests willl achieve this upload speed or less and 90% will achieve higher speeds.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G UPLOAD SPEED

BUCKET DISTRIBUTION OF PERFORMANCE


Use the bar heights to see what percentage of 3G tests fall within a particular
bracket of upload speed.
A good distribution will have a high right bucket (>1 Mbps). For example, if this
bucket is greater than 50% then this indicates that more than half the tests run
on 3G for the network are over 1 Mbps.
A high left bucket indicates a large proportion of tests on 3G achieving speeds
of less than 256 kbps.

60

Methodology Notes
Each bucket shows the proportion of speedtests run that fall within that range of down
load speed, giving a simple way of comparing
the speed distributions.
The percentage value of the bucket is the
likelihood that a speedtest run will fall within
that download speed bracket.

o
<D

40
3

20

0
80

0
80

60

J4020
g

"5 o

H
C
a>

60
40
20
0

80
60

256 kbps - 1 Mbps

<256 kbps

>1 Mbps

Upload speed (Mbps)

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G UPLOAD SPEED

MEDIAN PERFORMANCE
\

0.9 Mbps

AVERAGE (MEDIAN)
3G UPLOAD SPEED

in

o.
xx

5.3
T3
CD
CD
Q.
in

c
re

+4.<.0%
COMPARED TO
THE AVERAGE

14. 9%
UCOM

ClerkenCell

Ucom

Highest value
Upper quartile
Median

T-Uqbar

Higher is faster.
The median (the line through the middle of
the box) can be thought of as showing the

Lower quartile

"typical" user experience.


Compare the upload speed of a typical test

Lowest value

height of the lines.

over 3G on a network by judging the relative

Air+

LMM

Q:Mob

Methodology Notes
These offer another good means of under
standing the typical user experience of a
network, since they discount the effects of a
small number of users having exceptionally
poor or good service. Like "the mean", "the
median" can also be referred to as simply "the
average".

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G UPLOAD SPEED
.9

PERFORMANCE TRENDS OVER TIME


1.5
1.2
0.9

0.6
0.3
1.5
1.2
0.9

0.6
0.3
1.5
1.2

"
Q.0.9

n
5-0.6
o
CD

V 0.3
-1 5

o
iS 1-2

_o

3 0.9
0.6
0.3
1.5
1.2
0.9
0.6
0.3
1.5
1.2

.6
.3

0
0
0

O
CD

O
ID

r-

c
o
o

o
June 2013

July 2013

A line that is going up from left to right represents a network whose average
upload speed has increased consistently across the three month period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the
true value lies for that month.

August 2013

Methodology Notes
For each month and network the average
(mean) 3G upload speed is plotted as single
point.
The bars indicate the 80% confidence inter
val.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G UPLOAD SPEED

CLERKENCELL ABSOLUTE PERFORMANCE

Areas that are a strong green colour represent regions of high mean
upload
speed for speedtests run over 3G.
Areas that are a strong red colour represent regions of low mean uplo
ad speed
for speedtests run over 3G.

More than 35 %

15 to 35%
5 to 15%

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) 3G upload
speed within each area is calculated.

-5 to 5 %
-15 to -5 %
-35 to-15%
Less than -35 %

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G UPLOAD SPEED

CLERKENCELL VS OTHER NETWORKS

ClerkenCell

Ucom

T-Uqbar

Air+

LMM

Q:Mob

Off-peak

0.97

1.12

0.41

1.29

0.94

1.03

Peak

0.88

0.93

0.38

1.04

1.04

0.87

Ucom

ClerkenCell

T-Uqbar

PCCW

SmarTone

<256 kbps

4.29%

6.93%

16.53%

23.88%

23.4%

12.75%

256 kbps - 1 Mbps

39.29%

47.87%

79.73%

37.5%

36.17%

45.99%

>1 Mbps

56.43%

45.2%

3.73%

38.62%

40.43%

41.26%

Air+

ClerkenCell
Ucom
T-Uqbar
Areas that are a strong green colour represent regions of relatively high mean
0.51
0.62
0.28
Lower quartile
upload speed for speedtests run over 3G compared to the mean in that area
Median
0.89
1.05
0.38
found
for the other networks.
Upper quartile

1.36
ClerkenCell

1.41

Q:Mob

0.43Methodology Notes
0.43
0.44
Data is geocoded and segmented by area,
0.97
0.87
1.01
based on this the average (mean) 3G upload
1.86
1.33
1.56
Air+
LMM
Q:Mob

0.51

Ucom

LMM

T-Uqbar

2013-06-01

1.09

1.33

0.37

1.09

0.85

0.85

2013-07-01

0.98

1.00

2013-08-01

0.10

1.08

0.40

1.06

0.88

0.72

0.38

0.90

0.89

0.89

speed wi
thin eac
h area i
s calculat
ed and t
hen

compared to the mean of all other networks.

Areas that are a strong red colour represent regions of relatively low mean
upload speed for speedtests run over 3G compared to the mean found in that
area across all other networks.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G UPLOAD SPEED

SOURCE DATA
Average Performance (Mbps)
ClerkenCell
Upload speed (Mbps)

0.94

T-Uqbar
1.03

0.40

Peak vs Off-peak (Mbps)

1.19

0.98

0.98

Bucket Distribution of Performance


Median Performance (Mbps)
Performance Trends over Time (Mbps)
CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

OpenSignal

3G LATENCY

AVERAGE PERFORMANCE
AVERAGE (MEAN)

118.0ms '/

DRAW

20.4%

+4.:

2%

QMOB

AVERAGE (MEAN)
3G LATENCY

COMPARED TO
THE AVERAGE

UCOM

Methodology Notes
Shows the mean values of latency found
across speed tests taken over 3G on each
network.
The vertical lines show the 80% confidence
interval.

160

120

(fl
E,
>.

o
c
0)
4-

80

re

40

ClerkenCell

Ucom

T-Uqbar

Lower bars indicate faster performance.


Use the bar height to compare the 3G latency of the networks.
The vertical lines show the range in which we are confident the true mean lies for
each network. Where two bars have different heights but these lines overlap, we
cannot be certain which network is faster.

Air+

LMM

Q:Mob

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

OpenSignal

3G LATENCY

PEAK VS OFF-PEAK

150

100
o
c
0)

re

50

Peak

Off-peak
ClerkenCell

Ucom

T-Uqbar

Lower bars indicate faster performance.


Use the bar heights of networks in the left group to compare 3G latency of
networks in off-peak times. Similarly, the group on the right represents speeds
found during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for
each network.

Air+

LMM

Q:Mob

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean latency of tests taken over
3G for each network across each period is
shown.
The vertical lines show the 80% confidence
interval.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G LATENCY

CUMULATIVE FREQUENCY
100

75

- ClerkenCel

o
c

0)
3
0)

50

- Air+

0)

>

4-

re

Q:Mob

3
3

25

500

1000

1500

2000

Latency (ms)
Generally, steeper lines show a greater likelihood that a user will get a fast
latency than shallow lines.
To find the percentage of tests run on LTE on a particular network that achieve
less than a certain latency (for example 500 ms), look at where the line for the
network is above the point (500 ms). For example, if it is at 10% then 10% of
tests willl achieve this latency or less and 90% will achieve higher latencies.

Methodology Notes
Here we represent the full latency distribu
tion from speedtests run over 3G on each
network.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G LATENCY

BUCKET DISTRIBUTION OF PERFORMANCE


o
o
%
O

Use the bar heights to s

in a particular
bracket of latency.
A good distribution will have a high left bucket (<100 ms). For example, if this
bucket is greater than 50% then this indicates that more than half the tests run
on 3G for the network are less than 100 ms.
A high right bucket indicates a large proportion of tests on 3G achieving speeds
of greater than 500 ms.

ee what percentage of 3G tests fall with


Methodology Notes
Each bucket shows the proportion

of speedtests run that fall within that range of latency,


giving a simple way of comparing the speed
distributions.
The percentage value of the bucket is the
likelihood that a speedtest run will fall within
that latency bracket.

o
o
3
J2

>
-i
+
r

O
5
o

40
20
0
60
40
20
0
<100 ms

100-500 ms

Latency (ms)

>500 ms

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

OpenSignal

3G LATENCY

MEDIAN PERFORMANCE
\
AVERAGE (MEDIAN)

AVFRARF(M

3%

COMPARED TO
THE AVERAGE

3G LATENCY

ClerkenCell

Highest value
Upper quartile
Median
Lower quartile
Lowest value

o
c

Ucom

14. 7%

+0.:

99.8 ms

LMM

T-Uqbar

Lower is faster.

The median (the line through the middle


'.'
of the box) can be thought of as showing the
"typical" user experience.
'.'
Compare the latency of a typical test
er 3G on a network
by judging the relative
height of the lines.

400
in

E
>.

CD

+-

re

200

Air+

LMM

OpenSignal

Q:Mob

Methodology Not
es
These offer another
good means of unde
r
standing the typical
user experience of a
network, since they
discount the effects
of a

small number of users having exceptionally


poor or good service. Like "the mean", "the
median" can also be referred to as simply "the
average".

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G LATENCY

PERFORMANCE TRENDS OVER TIME

June 2013

July 2013

A line that is going up from left to right represents a network whose average
latency has increased consistently across the three month period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the

August 2013

true value lies for that month.

(mean) 3G latency is plotted as single point.


The bars indicate the 80% confidence inter
val.

Methodology Notes
For each month and network the average

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G LATENCY

CLERKENCELL ABSOLUTE PERFORMANCE

More than 300 ms


200 - 300 ms
170-200 ms
140- 170 ms
110-140 ms
80- 110 ms
Less than 80 ms

Areas that are a strong green colour represent regions of low mean latency for
speedtests run over 3G.
Areas that are a strong red colour represent regions of high mean latency for
speedtests run over 3G.

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) 3G latency
within each area is calculated.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G LATENCY

MAPS: T-UQBAR ABSOLUTE PERFORMANCE

More than 300 ms


200 - 300 ms
170-200 ms
140- 170 ms
110-140 ms
80- 110 ms
Less than 80 ms

Areas that are a strong green colour represent regions of low mean latency for
speedtests run over 3G.
Areas that are a strong red colour represent regions of high mean latency for
speedtests run over 3G.

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) 3G latency
within each area is calculated.

OpenSignal

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

3G LATENCY

CLERKENCELL VS OTHER NETWORKS

More than 35 %
15 to 35%
5 to 15%

-5 to 5 %
-15 to -5 %
-35 to-15%
Less than -35 %

Areas that are a strong green colour represent regions of relatively low mean
latency for speedtests run over 3G compared to the mean in that area found for
the other networks.
Areas that are a strong red colour represent regions of relatively high mean
latency for speedtests run over 3G compared to the mean found in that area
across all other networks.

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) 3G latency
within each area is calculated and then com
pared to the mean of all other networks.

OpenSignal

3G LATENCY

CLERKENCELL I HONG KONG I JUNE - AUGUST 2013

ClerkenCell

Ucom

T-Uqbar

Air+

LMM

Q:Mob

Off-peak

119.36

109.14

135.69

166.83

109.42

120.85

Peak

115.92

89.43

142.67

137.35

97.16

116.68

SOURCE DATA
Ucom

ClerkenCell

<100ms

51%

50%

100-500 ms

47%

48%

>500 ms

3%

2%

Median

71

99.75

Upper quartile

166.75
ClerkenCell

SmarTone

49%

47%

58%

Performance (ms)
48%
Average 51%

46%

39%

7%

3%

6%

75

97.95

98.5
158
Ucom

3%

T-Uqbar Air+

T-Uqbar

65 117.97

PCCW

43%

ClerkenCell ClerkenCell
Ucom
Lower
quartile
Latency
(ms)

T-Uqbar

137.82

LMM
75

109

105.67

161

187.5
Air+

T-Uqbar
168.57

155.77

2013-06-01

101.10

137.63

145.35

2013-07-01

125.85

2013-08-01

128.66

Methodology Notes
123.39
144.18
147.09
Shows the mean values of data success rate
116.94
125.21
155.90
found across speed tests taken over 3G on
each network.
The vertical lines show the 80% confidence
interval.

61

Q:Mob
104.39

66

119.46

87

100

143.5
LMM

183.75
Q:Mob

117.96

145.24

111.62

140.61

106.48

114.5

Peak vs Off-peak (ms)

OpenSignal

Bucket Distribution of Performance


Median Performance (ms)
Performance Trends over Time (ms)
CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

100/112

3G DATA SUCCESS RATE

AVERAGE PERFORMANCE

93.9%

fA

AVERAGE (MEAN)
3G DATA SUCCESS RATE

DRAW

LMM

+2.:

2%

T-UQBAR

COMPARED TO
THE AVERAGE

3.

5%
AIR+

90

0)

4-
CO

(0
V) 85
<D
O
O
3

(fl
re

re
o

80

75
ClerkenCell

Ucom

T-Uqbar

Air+

LMM

Higher bars indicate more reliable performance.


Use the bar height to compare the 3G data success rate of the networks.
The vertical lines show the range in which we are confident the true mean lies for
each network. Where two bars have different heights but these lines overlap, we
cannot be certain which network is more reliable.

OpenSignal

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

3G DATA SUCCESS RATE

PEAK VS OFF-PEAK

Q:Mob

90

0)
re
k.

(fl
$85
o
o

3
in

re
re
o

80

75
Peak
ClerkenCell

T-Uqbar

Higher bars indicate more reliable performance.


Use the bar heights of networks in the left group to compare 3G data success
rate of networks in off-peak times. Similarly, the group on the right represents
speeds found during peak times.
To see how average experience of users varies between off-peak and peak
times compare bar heights across the two groups.
The vertical lines show the range in which we are confident the true mean lies for
each network.

DATA

SUCCESS

RATE

PERFORMANCE TRENDS OVER TIME


95
90
85
80
75
95

Q:Mob

Methodology Notes
Data is partitioned into peak times (9:0018:00 Monday to Friday) and off-peak times
(all else). The mean data success rate of tests
taken over 3G for each network across each
period is shown.
The vertical lines show the 80% confidence
interval.

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

OpenSignal

3G

LMM

90
85
80
75
95

goo

......
-

IL

____I

-H

--

---

--'

0J85

re

OC 80
(A

$75

O 95
OT90
re
re 85
Q
80

t'

75
95
90
85
80
75
95
90
85

80
75
June 2013

July 2013

A line that is going up from left to right represents a network whose average
data success rate has increased consistently across the three month period.
The steeper the line the faster the rate of improvement (or deterioration).
The vertical line over each point show the range in which we are confident the
true value lies for that month.

OpenSignal

August 2013

Methodology Notes
For each month and network the average
(mean) 3G data success rate is plotted as
single point.
The bars indicate the 80% confidence inter
val.

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

3G DATA SUCCESS RATE

CLERKENCELL ABSOLUTE PERFORMANCE

98-100%
96 - 98 %
94 - 96%
92 - 94 %
89 - 92 %
86 - 89 %
0 - 86 %

Areas that are a strong green colour represent regions of high mean data
success rate for speedtests run over 3G.
Areas that are a strong red colour represent regions of low mean data success
rate for speedtests run over 3G.

OpenSignal

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) 3G data
success rate within each area is calculated.

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

3G DATA SUCCESS RATE

CLERKENCELL VS OTHER NETWORKS

WN i
'Jr"-ifl:(SlorthWrig.Aye'

ClerkenCell

Ucom

T-Uqbar

Air+

LMM

Q:Mob

Off-peak

94.31

91.18

93.24

90.99

92.94

91.56

Peak

93.28

90.67

93.30

90.28

94.13

90.87

ClerkenCell

Ucom

Air+

T-Uqbar

LMM

Q:Mob

2013-06-01

95.07

90.96

90.79

92.20

94.73

89.86

2013-07-01

93.66

90.11

92.65

89.81

93.83

91.12

2013-08-01

93.53

91.24

95.20

91.03

92.83

92.16

Areas that are a strong green colour represent regions of relatively high mean
data success rate for speedtests run over 3G compared to the mean in that

Methodology Notes
Data is geocoded and segmented by area,
based on this the average (mean) 3G data
success rate within each area is calculated
and then compared to the mean of all other
networks.

area found for the other networks.


Areas that are a strong red colour represent regions of relatively low mean data
success rate for speedtests run over 3G compared to the mean found in that
area across all other networks.

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

OpenSignal

3G DATA SUCCESS RATE

SOURCE DATA
Average Performance (%)

ClerkenCell
Data success rate (%)

93.93

T-Uqbar
90.90

93.26

90.76

93.43

91.33

Peak vs Off-peak (%)

Performance Trends over Time (%)

CLERKENCELL | HONG KONG | JUNE -

OpenSignal

AUGUST 2013

COVERAGE AND DATA METRICS SUMMARY

FULL BENCHMARKS SUMMARY


The tables on this page and the following page add an extra level
of detail over the Results Tables of the preceding section.
The tables show:
The score achieved by ClerkenCell for each metric
The ranking of ClerkenCell relative to the other networks
'.' How the value achieved by ClerkenCell compares to the
average of other networks
'.' How this value compares to the market leader for that metric
when a ClerkenCell brand is not the market leader
'.' The market leader for the metric
>

Note: when networks are in joint equal position, just one logo is
shown.

PROPORTION OF
TIME METRICS

Network

Value

Ranking

Compared To
Average

Compared to
Best Network

Best Network
In This Market

draw 1/7 with


T-Uqbar &
UCom

Mean

89.9%

Median

97.7%

2/7

+2.8%

-0.8%

Ucom

Mean

60.8%

4/6

+0.7%

+0.7%

Ucom

Median

63.9%

4/6

-0.8%

-0.8%

Ucom

Proportion

Mean

of Time
with No
Network
Connection

6.7%

draw 1/7 with


T-Uqbar &
UCom

+43.0%

Median

0.4%

1/7

+41.5%

Mean

32.9%

draw 6/7 with


UCom

-19.7%

-19.7%

Q:Mob

Median

21.2%

6/7

-23.0%

-23.0%

Q:Mob

Proportion
of Time on
3G/LTE

Proportion
of Time on
LTE

Proportion
of Time on
WiFi

+11.7%

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

OpenSignal

COVERAGE AND DATA METRICS SUMMARY

FULL BENCHMARKS SUMMARY


DATA
METRICS

(CONTINUED)

Value

Ranking

Compared To
Average

Mean

20.7 Mbps

draw 3/6 with


Cellys & Ucom

-1.0%

-7.7%

Q:Mob

Median

19.7 Mbps

3/6

+2.5%

-10.5%

Q:Mob

Mean

6.4 Mbps

draw 4/6 with


Air+ & Cellys

-4.7%

-13.2%

LMM

Median

6.2 Mbps

4/6

+2.4%

-8.9%

Q:Mob

Mean

62.6 ms

6/6

-27.9%

-43.1%

Air

LTE Data

Success

Network

LTE Upload

Compared to
Best Network

Speed

LTE Upload
Speed

LTE
Latency
Rate

Best Network
In This Market

3G

LMM

Median

57.0 ms

6/6

-48.1 %

Mean

90.9%

draw 1/6 with


5 other net
works

-0.1%

Mean

3.3 Mbps

4/6

+20.7%

-14.7%

Ucom

Median

3.0 Mbps

3/6

+34.6%

-5.4%

Ucom

Mean

0.9 Mbps

5/6

+2.8%

-21.2%

Air

Median

0.9 Mbps

4/6

+4.0%

-14.9%

Ucom

Mean

118.0 ms

draw 3/6 with


QMob

+4.2%

-20.4%

Ucom

Median

99.8 ms

3/6

+0.3%

-14.7%

93.9%

draw 1/6
with LMM 8
T-Uqbar

+2.2%

-70.1%

Q:Mob

Download
Speed

3G
Upload
Speed

3G
Latency

3G Data
Success
Rate

Mean

OpenSignal

LMM

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

METHODOLOGY

OVERVIEW
Measuring the True User Experience
The OpenSignal data is collected from real world consumer

For KPIs relating to the data network performance, such as throughput,


latency and data success rates we run active tests of the data network.
See below for more details of how we run these tests.

smartphones, and is recorded under conditions of normal usage. Rather


than approximate the user experience, we directly measure it from
the users of our smartphone application. Our application can be freely
downloaded on either iOS or Android devices and constantly monitors
the true network experience that users are getting on those devices. We
do not augment this real world data with any simulated lab or drive test

It

Rather than approximate the user


experience, we directly measure it from
the users of our smartphone application.

data of the network performance.


How Are Data Network Tests Structured?
Through being located on consumer smartphones, we are able to
observe the network exactly as the end user experiences it. This
customer-centric approach allows us to measure the true end-to-end
experience of the mobile network. We're not interested in models,
simulations or assumptions - our goal is to directly measure user
experience through the eyes of the users themselves.
The Consumer Philosophy

Everything we do at OpenSignal begins and ends with the consumer.


We aim to collect the data that matters most to consumers and to
precisely measure what they experience. Every step of the way we ask
ourselves if our choices are most representative of the experience had
by the average consumer. We believe that if you provide consumers with
accurate and unbiased data they will be willing to provide useful data in

JJ

return, and that this process can foster greater understanding and trust
between consumer and mobile operator.
Although operators have been monitoring how their networks perform
since the very beginning, there remains a disconnect between
the standard network KPIs and what surveys say about customer
experience. We believe that the only way to bridge this gap is to measure
the network using the customer experience as our starting point.
How Do We Test?

Our data network tests are designed to measure a user's average


experience as closely as possible. Unlike other testing methodologies,
we do not optimize anything related to the testing scenario; we instead
keep conditions as close to normal usage as possible.
Firstly, we run a series of 3 ICMP ping requests and 3 HTTP HEAD
requests to assess the availability of the data network. These tests are
run on popular servers such as google.com in order to emulate real world
usage and to remove any geographical bias (whilst not over-engineering
tests to use an artificially close server). Then we attempt to download
a large file of known size from a remote server for a fixed period of time

Our application runs on user devices and takes measurements of


customer experience KPIs at regular intervals. We run different types of
tests depending on the specific KPI that we are examining. For network
availability metrics, such as the proportion of time users spend on 3G, we
record the type of connection (or lack thereof) a user has during the test
and use this to calculate an average of that user's connectivity over time.

over a standard TCP/IP connection. The speed is calculated by looking


at the amount of data successfully downloaded in the test interval. The
same is done in reverse for the upload throughput tests. To deliver the
file we use a series of popular CDNs to emulate as closely as possible
what users would generally experience when downloading files from the
internet.

Multiple HTTP connections are opened as this is what modern


browsers and applications do when downloading files. Finally, as
described below, the relative success of each of these types of tests are
used to calculate the data success rate of the network.
When and Where Do We Test?
Our tests run 24 hours a day, 7 days a week in all locations visited by
our users.

This means that we are able to build up a complete picture


of network experience; measured at all the times and locations where
people actually use their devices, rather than just those that suit a drive
test schedule. We run tests when people are indoors, when they are
outdoors and when they are in vehicles.

This allows us to model the full

range of consumer behaviour in way that is not possible through other


methodologies. We build up a true picture of how modern consumers

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

OpenSignal

METHODOLOGY

OVERVIEW

(CONTINUED)

use and experience their devices and this is something that naturally
evolves and remains up to date as consumer behaviour changes. In
order to make sure that our data is truly representative of the times
people use their devices, we ensure that the screen is on when we take
a measurement of network connectivity, indicating that the device is
currently in use.
Why Do Users Download the OpenSignal App?
The OpenSignal app is a toolkit designed to help users better understand
the network connectivity of their device. The app features a dashboard
that provides the user with more details than usually available about
the status of their current connection. The dashboard shows them the

direction of their currently connected cell tower, uses separate indicators

being transparent with our users, as well as not capturing unnecessary


additional information, ensures that the vast majority of users are willing
to share their diagnostic information with us.
Big Data
Our emphasis on providing users with real benefits from the OpenSignal
for voice and data connectivity, and displays a map of nearby Wi-Fi
connections. Users can also submit data to us by running manual data
tests or submitting manual reports about specific problems or overall
provider satisfaction levels. Finally, users can view coverage maps and
network statistics that we provide based on data aggregated across all
other app users.

app has led to over 5 million downloads worldwide. From these users we
collect millions of customer experience data points daily and have built
up a global dataset. We use the latest Big Data technologies to process
and analyse this data dataset in a scalable and repeatable way, but our
goal is to extract the insights and provide it in an easily digestible and
actionable format.

it

Ultimately, a good network connection is


fundamental to the satisfaction of mobile
users.

Our coverage maps and statistics help users identify

which operator is offering the best coverage in the areas that are most
important to them.

Why Consumer on-Device Data Instead of Drive/Lab Testing?


Drive/lab testing methodologies do not emulate real world mobile device

Ultimately, a good network connection is fundamental to the satisfaction


of mobile users. The suite of features in the OpenSignal app provide

usage. Traditional drive testing does not even attempt to measure the
indoor user experience which is where more than 50% of mobile usage

greater clarity to users on the state of their current network connection


and provides them with a set of practical solutions to use in order to go

occurs, according to numerous studies. Drive tests that do measure


indoor conditions still lack the behavioural aspect of smartphone use,

about improving both their immediate connection and their overall mobile

a very important dimension. Whether it's the locations users tend to

experience.

spend their time or the way in which a user holds their device, simulated
approaches are completely unable to model the impact of user

Privacy

behaviour. The on-device testing we employ ensures that user behaviour


is built into the KPIs we measure from the very beginning.

Privacy is an issue that we take extremely seriously at OpenSignal. On


first opening the application, users are presented with clear information

In addition, tapping into the potential of consumers allows us to achieve

on what data is being shared with OpenSignal. No demographic

a scale not possible by drive testing. We have millions of consumers

information is ever recorded, which includes names, ages or genders.


The only data we collect is diagnostic information pertaining to network

providing a continual stream of network performance data with granular


location detail. The drive testing process is expensive and there is a limit

performance. Furthermore, we identify users by assigning our own


anonymous identifier and do not ever make this available to anyone

to the number of distinct locations in which it is feasible to test, as well


as the frequency with which they can be updated. While drive testing

outside of OpenSignal so there is no way that this data could be

remains applicable for some use cases, we believe on-device data

'de-anonymized' in any way. Finally, users can simply choose not to


share data with us if they prefer and these users are still able to make

represents the future of customer experience monitoring.

use of the OpenSignal app benefits. We have, however, found that

OpenSignal

CLERKENCELL | HONG KONG | JUNE -

AUGUST 2013

METHODOLOGY

HOW THIS REPORT IS MADE


Preparing the data
First we isolate the networks in Hong Kong by extracting all those points
sent with network IDs corresponding to the brands operational there.
Additionally, we then perform a geographical query to isolate only those
data points that have been captured within Hong Kong - as defined by
the bounds sent by ClerkenCell.

[SAMPLE NOTES: In the case that one network has several brands under
a single network ID, these will be identified by investigating the network
names associated with the SIM in the reporting device and the network
which is serving the device.]
[SAMPLE NOTES: Any data filters specific to the market are described
here.]

JJ

For the data-points isolated in this way we run a pre-averaging process


designed to eliminate biases caused by individual devices reporting much
more than others - this can be the case for very keen users of the app
who often open it or change the settings in order to contribute more
data. This process works by aggregating readings from each user to a
precision of 3 decimal places in latitude and longitude to hourly level.
Thus, multiple readings from one user at one location in a particular
hour will be aggregated to a single reading; however if the user moves
location by more than about 100m within that hour then the readings
will contribute multiple times - one for each location. By contrast, if the
user does not move significantly over several hours (for example whilst
at work) they will contribute one reading per hour to the pre-averaged
dataset, no matter how many raw readings they contribute.

each point in time.


For the purposes of this report the following classification is used:

>

where such changes have occurred with one exception: speedtests


during which the network connection has changed between two 3G
connections. These do contribute to the 3G section of the report.
Identifying 'No Network'
We are able to detect directly when a user is not currently connected to
the network. We do not use an RSSI threshold to indicate no network
access only considering when a user is completely disconnected from

Identifying WiFi

Android allows us to see the network technology the device is using at

2G:GPRS, EDGE, 1xRTT, iDen

technologies, or even between a WiFi and mobile connection. The app


listens for these changes, we exclude from the averaging speedtests

the network.

Identifying 3G and LTE technologies

>

Whilst a speedtest is running it is possible that the network connection


of the user can change - for example from 3G to LTE, between 3G

Android allows us to identify if a device is currently connected to a cellular


network or a WiFi network. Occasionally (1 %) the device is unable to
determine whether the phone is connected to either a cellular network or
WiFi network. For the purposes of this metric these results are discarded.
At the Point of Averaging - Time on Metrics

Network

Percentage of data
For the "time on" metrics we average over all devices that mee

3G: UMTS, HSPA, HSPA+, CDMA, EVDO-

ClerkenCell

0, EVDO-A, EVDO-B,
eHRPD

Ucom

21%

T-Uqbar

10%

inate the

13%

effects of an unusual behaviour profile from a device contributin

CMHK

12%

PCCW

18%

g a large
amount of data. From these device averages we calculate our

SmarTone

11%

>

LTE: LTE

15%

t our
minimum data volume criteria. We use device averages to elim

overall
averages.

The Data Success Rate


The data success rate is based on proportion of latency and download
tests that succeed. A single speedtest will perform several pings and
a single download, each of these can be thought of as "sub-test". The
proportion of these sub-tests that succeed is then the data success rate
for the individual test. Unlike latency, download and upload values, the
data success rate for an individual test does not have a continuous range
of possible values and for this reason we do not examine the distribution
of these tests.

OpenSignal

CLERKENCELL | HONG KONG | JUNE - AUGUST 2013

METHODOLOGY

HOW THIS REPORT IS MADE

At the Point of Averaging - Speedtest Metrics

(CONTINUED)

When calculating the mean values of the data (whether it is an overall


mean per network, or mean over some time period or off-peak vs peak)
we first discard the top and bottom 5% of the data, this is to remove
the impact of extremes of high and low to arrive at a figure better
representing the average experience. In the calculation of distributions
- such as the cumulative frequency distribution, the bucket distributions
and the median and quartiles - we do not discard any of the data
because we are interested in the full range of values.

Contacting OpenSignal
For technical support:
+ 44 845 834 0987
[email protected]
For information on subscribing to Insights for other markets:
[email protected]
If you are interested in publicly disseminating statistics or
other information based on this report, please consult:

Filtering Throttled Networks


Since all users on T-Uqbar are subject to throttling, we do not filter out
throttling on that network. For 3, PCCW and SmarTone we filter out
throttling to create a fairer like-for-like comparison with other networks.

[email protected]
For all other queries:
[email protected]

The filtering is done by looking at the distribution of tests in the range


between 0.4Mbps and 1 Mbps, where the distribution of tests is
relatively flat. For the unthrottled 1010 and one2free networks we note
that the number of tests in the 100 kbps brackets between 0 and 0.4

Sample Size

Mbps are roughly equal with the number of tests run in the 100 kbps
brackets between 0.4 Mbps and 1 Mbps. For throttled networks, then,

Total number of users contributing data: 4,032


Total number of data points: 7,691,988

our algorithm looks at the number of tests taken in 100kbps brackets


between 0.4Mbps and 1 Mbps. This number of tests is then used as
the sample for number of tests run between 0 and 0.4Mbps, with the

Network breakdown:

remaining tests discarded.


Before throttling filter:

After throttling filter:

Download speed (Mbps)

OpenSignal

CLERKENCELL | HONG KONG | JUNE -

AUGUST 2013

You might also like