PAPER INVITED TO 13TH INTERNATIONAL SYMPOSIUM ON WIRELESS COMMUNICATION SYSTEMS 2016, 20–23 SEPTEMBER 2016, POZNAŃ, POLAND 1

Mobile and Residential

INEA Wi-Fi Hotspot Network
Bartosz Musznicki, Karol Kowalik, Piotr Kołodziejski, and Eugeniusz Grzybek
INEA, Poznań, Poland
{bartosz.musznicki, karol.kowalik, piotr.kolodziejski, eugeniusz.grzybek}@inea.com.pl

Abstract—Since 2012 INEA has been developing and expand- in Figure 1, which shows the changes in the number of users
ing the network of IEEE 802.11 compliant Wi-Fi hotspots (access throughout an average day. Each data point corresponds to
points) located across the Greater Poland region. This network a mean value obtained in the period of one month (further
consists of 330 mobile (vehicular) access points carried by public
buses and trams and over 20,000 fixed residential hotspots discussed in Section II-D). As in every graph in the paper,
distributed throughout the homes of INEA customers to provide confidence intervals (error bars) present the standard deviation
Internet access via the “community Wi-Fi” service. Therefore, of a data point. The graph for mobile network shows highest
this paper is aimed at sharing the insights gathered by INEA numbers of users in the morning and in the afternoon what
throughout 4 years of experience in providing hotspot-based corresponds to the busiest commuting hours. Stationary net-
Internet access. The emphasis is put on daily and hourly trends
in order to evaluate user experience, to determine key patterns, work, though, exhibits the highest usage in the evening what
and to investigate the influences such as public transportation apparently relates to Internet activities performed at home.
trends, user location and mobility, as well as, radio frequency
noise and interference. Network type:
100 Mobile Hotspot
Residential Hotspot

I. I NTRODUCTION 80
Users

There are currently about 47 million public Wi-Fi

hotspots [1] around the globe. This form of Internet access is 60

often provided by cafés, shops, airports, and railway stations

to let the customers access the web, connect to the favourite 40

social networking site, upload photos, read and send e-mails

etc. Such an urban Internet access is also often used by 0 5 10
Hour
15 20

tourists who do not use international data roaming packages,

as well as, by teenagers when they have smartphones equipped Figure 1. Number of INEA hotspot users throughout an average day
with no or limited data packages provided by mobile network
operators. The rest of the paper is organized as follows. First, we
INEA is the largest regional fixed-access telecommunica- give an overview of INEA mobile hotspot network, then
tions operator in the Greater Poland, which provides advanced we introduce the evaluation environment to move to ex-
multimedia services to over 220,000 of homes, businesses, and periment methodology and different signal quality related
institutions through different access mediums and technolo- measurements. Later on, we go through similar steps with
gies, i.e., Hybrid Fibre-Coaxial (HFC), Gigabit Passive Opti- the residential hotspot network to present a range of user
cal Network (GPON), point-to-point Carrier Ethernet optical experience related tests and observations. Finally, we sum up
fibres, IEEE 802.16e WiMAX [2], IEEE 802.11 Wi-Fi [3], with conclusions.
as well as, twisted pair based xDSL and IEEE 802.3 Ethernet.
INEA is also the major shareholder of the Greater Poland II. M OBILE HOTSPOTS
Broadband Network (Polish: Wielkopolska Sieć Szerokopas- The INEA mobile hotspot network consists of 330 mo-
mowa [WSS]) that operates over 4,500 km of optical DWDM bile access points (AP) mounted in public buses and trams,
infrastructure with IP/MPLS architecture running on top of deployed in cooperation with municipal transportation oper-
it to support Next Generation Access (NGA) services in 576 ators of Poznań and Konin, i.e. Miejskie Przedsi˛ebiorstwo
distribution nodes (points of presence), for the benefit of local Komunikacyjne (MPK) w Poznaniu [5] and Miejski Zakład
communities, businesses, and administration. Komunikacji (MZK) w Koninie [6], respectively. Each mobile
Alongside the well-established services, INEA subscribers AP device is a RouterBoard RB751U [7] enclosed within
are offered free Wi-Fi hotspot access provided by over 20,000 a protective enclosure and equipped with an external 2.5 dBi
fixed residential hotspots (with over 5,000 new hotspots omni-directional antenna. Together with a cellular network
planned in the in the upcoming months) and 330 mobile based uplink (a 4G modem), those mobile access points
hotspots carried by public buses and trams [4]. There are provide wireless Internet access to the passengers and other
hundreds of customers using the service each day, as depicted users in the close proximity of the vehicle.
PAPER INVITED TO 13TH INTERNATIONAL SYMPOSIUM ON WIRELESS COMMUNICATION SYSTEMS 2016, 20–23 SEPTEMBER 2016, POZNAŃ, POLAND 2

A. User perspective C. Evaluation environment

To use INEA mobile hotspot, one needs to select INEA After the initial tests, in order to perform a comparative
Hotspot from the list of available networks, i.e. the list of study, not only 330 mobile access points have been used, but
Service Set Identifiers (SSID) detected by end user device. also a selected part of INEA’s stationary 5 GHz Wi-Fi network
Before gaining Internet access, the user is redirected to a local – with similar number of concurrent customers (around 150),
captive portal (a landing page) with three options to access the and with 10 stationary APs (base stations). Each stationary
service, as presented in Figure 2. Upon successful authentica- access point was a RouterBoard RB433 [9] enclosed within
tion the user is granted Internet access. a protective enclosure and connected to a 16 dBi or 19 dBi
external sector antenna.
B. Architecture and operation
Figure 2 provides an architectural overview of the mobile
hotspot system. A centralized Remote Authentication Dial D. Experiment methodology
In User Service (RADIUS) database integrated with INEA’s
customer relationship management (CRM) system is used to Each access point in the experiment has been queried
perform authentication and authorisation. The RADIUS server using Simple Network Management Protocol (SNMP) every
also allows the service provider to obtain detailed information 15 minutes from 10 June 2016 till 10 July 2016. In this way,
required by the data retention policy, such as, login and logout we have collected information about the noise floor perceived
times, physical (MAC) addresses of the devices etc. Each by each AP receiver, the number of connected clients, and the
access point is configured with a firewall and traffic control Received Signal Strength Indicator (RSSI) value for each
mechanisms responsible for policing connected devices. All connected client. The customers of these networks were not
mobile INEA hotspots share the same SSID, however they informed that the study was being performed, and hence,
do not support network-coordinated roaming (handover), and the collection of all the parameters which could be used to
hence, roaming has to be triggered by the user device. Should track customers (such as MAC addresses of customers Wi-
it take place, the user must re-authenticate but is not requested Fi cards) was disabled. The 15 minutes interval was used
to manually select the access method again since session in order to obtain the data frequently enough to get the full
cookies are used to regain proper credentials and access. picture of changes in the performance throughout a full day,
The interaction with the system after the customer is pre- without affecting AP operations and customer experience in
sented with the captive portal depends on the selected option: the process.
• free of charge access for INEA customers – the user is
redirected to the site that performs authentication against
INEA customers database, E. Results
• first 15 minutes free of charge access for visitors – a The comparison of mobile hotspots and stationary Wi-Fi
dedicated account is automatically created in RADIUS access points is influenced by the patterns of people moving
database to allow free 15 minutes during a 24 hour period, around the area and also by the number of active hotspots.
• paid access for external users who require more than Since most of the public buses are turned off for the night
15 minutes Internet access – the user is redirected to (except of night lines), while trams are still on, the number of
a payment website. After a successful payment, a ded- active mobile hotspots drops in the night, as shown in Figure 3.
icated RADIUS account is created with time limitations Around the noon, when there is a slightly lower demand
corresponding to the purchased package. The payment is for public transportation, and thus, some of the buses stay in
realized by means of integration with Przelewy24 system bus depots, the decrease in the number of active hotspots can
that provides wide range of payment methods for the be observed. As a consequence, when analysing results in the
Polish market, i.e. SMS payments, bank transfers etc [8]. following subsections, one should keep these patterns in mind.
Afterwards, when the authorisation phase has succeeded, the
customer is allowed to access the Internet.
220
Number of Active APs

200

180

Network type:
160 Mobile Hotspot

0 5 10 15 20
Hour

Figure 2. High-level architecture of INEA mobile hotspots authentication and

authorisation system Figure 3. Number of active mobile INEA hotspots throughout a day
PAPER INVITED TO 13TH INTERNATIONAL SYMPOSIUM ON WIRELESS COMMUNICATION SYSTEMS 2016, 20–23 SEPTEMBER 2016, POZNAŃ, POLAND 3

Network type:
Mobile Hotspot
−104 Stationary WiFi −66
Network type:
Mobile Hotspot
Noise Floor [dBm]

RSSI [dBm]
−68 Stationary WiFi

−105

−70

−106 −72

0 5 10 15 20 0 5 10 15 20
Hour Hour

Figure 4. Average noise floor observed throughout an average day Figure 5. Average RSSI observed throughout an average day

1) Noise floor: The noise floor is the measure of back- average RSSI observed throughout the day is presented in
ground noise created by all the noise sources of the Ra- Figure 5. The client side hardware, i.e., Customer Premises
dio Frequency (RF) environment. This ambient noise comes Equipment (CPE), used in the stationary Wi-Fi scenario con-
from variety of devices (microwave ovens, Bluetooth devices, sists of an outdoor wireless receiver with a directional antenna.
wireless peripherals, ZigBee-based sensors, etc.), damaged Moreover, most of the CPEs are located within the Line-
connectors, radar equipment, as well as, atmospheric and of-Sight (LoS), so there are no obstacles which could cause
thermal noise. Therefore, we investigate the noise floor values significant variance in the observed RSSI values. Hence, not
in order to verify if mobile hotspots exhibit similarities with surprisingly, the average RSSI curve for stationary Wi-Fi is
stationary Wi-Fi networks. almost flat with a typical value around -65 dBm, which is
In Figure 4 the average noise floor observed throughout an considered a recommended value for the system.
average day is presented, with the data points showing the The average RSSI for mobile hotspot network, on the
mean value for a given hour. The data for both networks ware contrary, changes significantly throughout the day. The high
collected in different geographical locations and in different values during the night are most probably related to the trams
radio frequency bands (2.4 GHz for mobile hotspots and which park at tram depots and provide Wi-Fi connectivity for
5 GHz for stationary Wi-Fi) and on various radio channels the employees and the residents of the neighbouring areas.
spread over the set of available channels. This graph clearly We assume that those users try to maximize signal strength
illustrates the correlation between the noise floor for both staying nearby (in the optimum range), and thus, during the
mobile hotspot network and for stationary Wi-Fi network. night the RSSI exhibits highest values. Similar situation occurs
Thus, it is valid to assume that the phenomenon is caused by around 1 PM when some of the vehicles and drivers go back
sun activity or the processes taking place in the atmosphere. to bus depots, and again high RSSI values can be observed.
Obviously, the artificially caused influences exist as well, Figure 6 presents RSSI percentage histogram, which shows
but due to averaging function performed over the whole month, the percentage of samples collected for both types of networks
they diminish on the presented diagram. Both curves exhibit for given RSSI value. The diagram is not cumulative thus
similar characteristics, and yet, the one for stationary Wi- the bars for mobile hotspot network cover (are presented in
Fi network is around 2 dB lower than the mobile hotspot front of) the bars for stationary network. It demonstrates that
case. The difference is caused by the fact that stationary stationary Wi-Fi is often actively used, and thus, we were able
access points are operating on rooftops and are equipped to collect much more samples than for the mobile hotspot
with antennas of much higher gain, i.e. 16 dBi or 19 dBi, network. Moreover, Figure 6 confirms that the stationary Wi-
as compared to 2.5 dBi, and thus, they are exposed to noise
approaching from wider areas then mobile hotspots. Although,
the correlation between the two curves suggests that observed
5% Network type:
noise occurs due to natural activities and is not caused by Mobile Hotspot

mobility or changes in human-related patterns. 4%

Stationary WiFi

Our findings show the risks of drawing conclusions that

Percent of users

noise floor solely depends on the human and device activities 3%

(spontaneous and intended radio transmissions). The results 2%

presented in Figure 4 imply that noise characteristic may be
highly related to sun or atmospheric activities, while human- 1%

related activities contribute to the average level of noise.

0%
2) Received Signal Strength Indicator (RSSI): This section −100 −80 −60 −40
RSSI [dBm]
studies the strength of the signal, expressed in the form RSSI
from the client to the access point, and thus, all the reported Figure 6. Distribution of mobile hotspots and stationary Wi-Fi RSSI
measurements were gathered from the access points. The
PAPER INVITED TO 13TH INTERNATIONAL SYMPOSIUM ON WIRELESS COMMUNICATION SYSTEMS 2016, 20–23 SEPTEMBER 2016, POZNAŃ, POLAND 4

Fi network is designed and deployed with care. Therefore,

most often observed RSSI values are around -65 dBm. On the
other hand the mobile hotspot network attracts various types of
users, which sometimes try to get and stay connected despite
very low signal, as when on the fringes of the radio coverage
range. Hence, the range of RSSI values for mobile network is
much wider and typically around 10 dB below the stationary
network. For this reason, the bitrates experienced by clients
are often low, what as a consequence may occasionally lead
to reduced overall capacity. Figure 7. High-level architecture of residential INEA hotspot network

III. R ESIDENTIAL HOTSPOTS

INEA operates the network of over 20,000 residential compliant radio module incorporated in the CPE. When boot-
hotspots that follow the idea commonly known as “commu- ing, the wireless channel the AP will operate on is automati-
nity Wi-Fi” and provide Internet access to the members of cally selected from those that are the least used by other Wi-Fi
the community. Hence, according to INEA’s policy for the devices in the radio range. Service differentiation and separa-
Data Over Cable Service Interface Specification (DOCSIS) tion from the home Wi-Fi router is achieved by broadcasting
access network, every Cisco EPC3925 cable modem (a CPE), an additional Basic Service Set Identifier (BSSID) with net-
equipped with an integrated IEEE 802.11n 2x2 MIMO 2.4 work name, i.e. SSID, set to INEA_HotSpot_WiFi. Then
GHz Wi-Fi module [10] by default operates not only as the logical interface associated with that BSSID is internally
a personal (home) Wi-Fi router, but also as a virtualized linked (within the CPE) to a dedicated DOCSIS 3.0 service
community access point. Each customer is allowed to opt out flow [13], i.e. a MAC layer transport tunnel that logically
of this complimentary service, but as a result will no longer be separates hotspot traffic from home user’s traffic in the last
allowed to access the distributed residential hotspot system [4]. mile section between CPE and Cable Modem Termination Sys-
This decision has no effect on other services the customer is tem (CMTS). The data traffic related to the hotspot is limited
subscribed to in a monthly fee model and still can enjoy home to 2 Mbit/s (downlink) and 1 Mbit/s (uplink) according to
Internet access available in packages up to 250 Mbit/s. the applied Quality of Service (QoS) policy. Therefore, the
common Hybrid Fibre-Coaxial access medium in the last-
mile of the DOCSIS-based network can be efficiently used
A. User perspective to provide both high speed Internet access to the local (home)
To access the Internet via community Wi-Fi, subscriber, as well as, to provide the community Wi-Fi for
INEA customers have to connect to the SSID called visitors.
INEA_HotSpot_WiFi and login with the credentials they Following the guidelines of RFC 6598 [14], a DHCP server
use to access INEA customer care portal. When the first assigns an IPv4 address which belongs to the 100.64.0.0/10
authentication is successful, in most cases, depending on prefix for every end user device connected to the residential
the configuration of end user device, the credentials input hotspot network and the data is further transmitted via CMTS
by the user are stored in the (mobile) device for further in a Layer 2 Over Generic Routing Encapsulation (L2oGRE)
use. Therefore, the next time the customer turns on the tunnel to an aggregation router, as presented in Figure 7. Then
Wi-Fi while being in the radio range of INEA residential the traffic undergoes the process of Network Address Transla-
hotspot system, the customer will be automatically logged tion (NAT) performed by INEA’s Carrier-Grade NAT (CGN)
in to the network without the need to repeat the process by system. To ensure uninterrupted and transparent services, the
hand. Moreover, Protected Extensible Authentication Protocol tunnel aggregation and CGN systems are built in a strict
(PEAP) [11] based approach allows INEA to provide seamless high availability architecture. Security measures applied to
authentication for customers that are moving between access INEA_HotSpot_WiFi are separated from the mechanisms
points. For example, let us consider a customer who uses his used by the home router Wi-Fi service. The authentication
or hers home Wi-Fi provided by INEA on a daily basis. When is based on IEEE 802.1X standard [15] and uses PEAP
visiting a friend or a relative who is also a member of INEA and RADIUS together with a centralized user database.
Wi-Fi community (or lives nearby a community member),
the user will be able to access the Internet through the local C. Experiment environment
CPE. Therefore, regardless if the device is a smartphone, a
tablet, or a notebook, it can connect to the network without To tackle the assessment of user experience in the commu-
the user being aware of the automated processes taking place nity Wi-Fi network, an approach opposite to the one applied
in the background. in Section II has been used. Therefore, an empirical study
involving end user devices has been performed to evaluate the
actual usefulness and quality of the service. The methodology
B. Architecture and operation was twofold: first to investigate the experience of a smartphone
Each INEA community Wi-Fi access point is in fact an addi- user who moves through the test area, secondly to evaluate the
tional service provided by a single 2.4 GHz IEEE 802.11n [12] service when a user keeps a fixed position inside an apartment.
PAPER INVITED TO 13TH INTERNATIONAL SYMPOSIUM ON WIRELESS COMMUNICATION SYSTEMS 2016, 20–23 SEPTEMBER 2016, POZNAŃ, POLAND 5

Figure 8. Heroes of the Second World War Subdivision (foreground) and Figure 9. Path of the mobile user in residential hotspot access test
National Army Subdivision (background) in Poznań, Poland

As a representative region of operation, two subdivisions been recorded. The location discrepancies were caused by the
of the Rataje residential area in Poznań, Poland have been limited GPS/GLONASS visibility between the buildings what
selected – Osiedle Armii Krajowej (English: National Army has affected the accuracy (a circle on a building or in the
Subdivision) and Osiedle Bohaterów II Wojny Światowej middle of a street). The total number of SSIDs and BSSIDs,
(English: Heroes of the Second World War Subdivision). as well as, the frequencies they were operating on have been
Both subdivisions have been built in the 1970s from pre- presented in Table I, along with the capabilities advertised by
fabricated concrete blocks to meet the increasing demand every access point.
caused by rapid population growth. Most of the constructions In the group of 1375 unique SSIDs, 313 distinct
are four-story residential blocks (90 to 135 flats each) with INEA_HotSpot_WiFi access points have been spotted.
a few buildings reaching 10 or 16 storeys, please see Figure 8. In the area of interest, only 8.83% of BSSIDs have been
According to the 2012 census, the population density in Rataje broadcasted at 5 GHz, i.e. 152 out of 1874, with the remaining
was 7755,38 person per km2 , and hence, the selected subdivi- operating at 2.4 GHz. Most of the BSSIDs (96.48%) advertised
sions give a valid overview of the community Wi-Fi concept that the use of Wi-Fi Protected Access (WPA) and/or Wi-Fi
in one of the largest Polish cities. The results are presented Protected Access II (WPA2) protocol is required to connect.
and analysed in the following subsections. Some of the BSSIDs, i.e. 306 (16.33%), supported Wi-Fi
Protected Setup (WPS) security measures. The three most
D. Mobile access evaluation popular 2.4 GHz channels were 1 (34,67%), 11 (28,05%),
and 6 (23,40%), as presented in Fig. 10.
1) Experiment methodology: Mobile access experience has
The distribution of the maximum RSSI observed for 313
been evaluated with the use of Nexus 5X smartphone (run-
residential INEA_HotSpot_WiFi access points has been
ning Android 6.0.1 Marshmallow) that features an IEEE
presented in Figure 11. The maximum signal from 103 APs
802.11a/b/g/n/ac (2x2 MIMO) module, as well as, a GPS and
have been received at the level no weaker than -70 dBm.
GLONASS receiver [16]. During the test the user has been
This value of reference has a practical significance because
moving on the pavement between the buildings holding the
it is officially used by Apple iOS 8 and later as a threshold
phone in the hand in front of the user. The objective was to
for initiating a scan to roam to a different BSSID for the same
record the maximum RSSI (dBm) of every BSSID spotted
while traversing the area, together with other parameters of
interest, i.e. SSID, AP capabilities, GPS/GLONASS based
Table I
geographical coordinates, radio frequency, and Wi-Fi channel. T HE GENERAL STATISTICS OF RESIDENTIAL HOTSPOTS MOBILE ACCESS
Therefore, a free Android application called WiFi Tracker EVALUATION
(version 1.2.20) developed by Ian Hawkins [17] has been used
to record the data gathered by scanning the radio environment INEA_HotSpot_WiFi BSSIDs 313
in one second intervals during the whole test. Then the mea- all SSIDs 1375
surements have been exported to a comma-separated values all BSSIDs 1874
(CSV) file to perform desired analyses. 2.4 GHz BSSIDs 1722
2) Results: In the mobile access test performed on 2 July 5 GHz BSSIDs 152
2016 around 3 PM, the user has been moving on the pavement WPA/WPA2 protected BSSIDs 1808
between the buildings following the path depicted in Figure 9.
WPS capable BSSIDs 306
Green circles correspond to the geographical coordinates of
WEP protected BSSIDs 31
the location where the maximum RSSI of a BSSID has
PAPER INVITED TO 13TH INTERNATIONAL SYMPOSIUM ON WIRELESS COMMUNICATION SYSTEMS 2016, 20–23 SEPTEMBER 2016, POZNAŃ, POLAND 6

600
80

60
Number of BSSIDs

400

Number of APs
40

200

20

0 0

1 2 3 4 5 6 7 8 9 10 11 −90 −80 −70 −60 −50

2.4 GHz channel number RSSI [dBm]

Figure 10. Distribution of 2.4 GHz Wi-Fi channels usage Figure 11. Maximum RSSI distribution of residential
INEA_HotSpot_WiFi access points

SSID if its signal is at least 8 dB higher (active communica-

user can count on decent quality, what has been additionally
tion) or 12 dB higher (connection in idle state) [18].
proven with a test performed on 21 July 2016 at 10 AM with
Signal level not lower than -70 dBm can be considered as a Speedtest.net application installed on Samsung Galaxy S3 Neo
value allowing users to enjoy the most popular service types, (running Android 4.4.2 KitKat) with the results of 1.94 Mbit/s
including voice and video calls. Therefore, because another download, 0.72 Mbit/s upload and 8 ms RTT [21]. Although
81 APs fell into a range between -71 dBm and -75 dBm, the test succeeded, in some outdoor conditions, a sufficient
empirical tests of a Voice over IP (VoIP) call quality have link quality will not be possible due to long distance to the
been performed for those conditions. The calls have been nearest AP or difficult radio propagation conditions.
conducted between a user of Nexus 5X and a user of a
fixed-line VoIP phone via INEA’s commercial VoIP exchange
(proxy), with CSipSimple (version 1.02.03 r2457), a popu- E. Stationary access evaluation
lar free VoIP client for Android [19]. The ITU-T G.711A 1) Experiment methodology: The evaluation of stationary
recommendation [20] compliant codec has been used and access has been performed by measuring the RSSI and the
yielded satisfactory quality with only minor glitches that did Round Trip Time (RTT) to the IP default gateway, i.e. the ag-
not sacrifice the overall experience. As long as the user stayed gregation router, as observed by the user wirelessly connected
connected to the BSSID, the call could be continued without to INEA_HotSpot_WiFi access point located in different
interruptions. During the pavement walking test, most of the apartment in the same building. A laptop running Windows
time the user has been in the radio range of at least one hotspot 7 equipped with an IEEE 802.11n card has been chosen as a
which could be reached with a required RSSI level. Without test platform. The sampling has been conducted in the period
the use of dedicated roaming mechanisms, when the user of one week during the average busy-hour of the Internet
moved out of the radio range and the Wi-Fi connection was access services in INEA in the summer, i.e. between 9:30 PM
lost, the phone started to scan for another BSSID to connect and 10:30 PM. This part of the day has been selected based
to. Together with the process of 802.1X-based authentication, on the assumption that it will provide the most challenging
it can take up to several seconds to re-establish the voice RF conditions (noise, interference, medium congestion etc.)
connection and continue the conversation. Therefore, 59% for the residential Wi-Fi usage. RSSI has been recorded
of INEA residential hotspots (184 out of 313) can be used every 2 seconds with Homedale (version 1.61) [22], a free
outdoors in the urban environment for VoIP calls, provided Wi-Fi scanner. The ICMP-based RTT has been measured
that the user stays within the optimum radio range of a single every 1 second with hrPing (version 5.06) [23], a feature-rich
access point (usually a few meters from the nearest building). freeware ping utility.
For the levels lower than -75 dBm, call quality gets worse as 2) Results: The mean results of busy-hour measurements
signal quality drops, and yet, a user can still benefit from basic are presented in Figure 12, where the solid line corresponds
web browsing, e-mail and instant messaging (243 community to mean RSSI and the dashed line corresponds to mean
hotspots, i.e. 78% of the total count). The phone starts to RTT. The mean RSSI for every busy-hour oscillated between
keep losing connection to the AP when the RSSI drops to -63 and -68 dBm with the confidence intervals (error bars)
-80 dBm (70 APs have been discovered with RSSI at -80 dBm being not visible since the widest was only 0.146 dBm.
or worse). On the other hand, the confidence intervals for mean RTT are
Another real-world test scenario involved a parent engaged distinguishable because sample series are significantly varied.
in a web browsing session while sitting on a bench by As validated with the tests, the key to drawing valid con-
children’s playground. In those conditions the quality of expe- clusions from the analysis of the RTT samples in suboptimal
rience is rather of best effort type because it heavily depends radio conditions is focusing not (only) on the mean values,
on the RSSI, and hence, the proximity to the nearest access even when thousands of samples are available, but (also)
point. Therefore, according to the experiments, the user will on the wider scope of statistical parameters, as analysed in
often have to find an optimal spot to seat oneself. Then the Table II. As it turns out, a pretty steady mean RTT that
PAPER INVITED TO 13TH INTERNATIONAL SYMPOSIUM ON WIRELESS COMMUNICATION SYSTEMS 2016, 20–23 SEPTEMBER 2016, POZNAŃ, POLAND 7

Table II
ROUND T RIP T IME (RTT) RELATED STATISTICS OF RESIDENTIAL INEA_H O T S P O T _W I F I ACCESS POINT STATIONARY TEST ( IN MILISECONDS )

day 1 2 3 4 5 6 7
mean RTT 22.381 22.606 26.408 28.573 20.508 24.469 22.722
confidence interval 1.121 1.123 1.448 1.425 1.116 1.512 1.082
standard deviation 26.092 26.132 33.643 33.159 25.960 35.154 25.190
minimum 5.422 5.509 5.535 5.549 5.618 5.741 5.711
maximum 310.987 318.919 525.134 731.002 614.059 992.670 197.314
mean one way jitter 10.492 10.450 13.133 13.702 9.211 11.830 10.339
second quartile (median) 3.043 2.983 5.340 7.081 1.851 3.232 3.007
third quartile 19.127 18.215 22.916 23.075 14.875 20.445 19.233
maximum 144.459 154.386 253.140 346.468 303.031 467.485 94.988

varies from 20.508 to 28.573 ms can be accompanied by a still meet the expectations, especially when enhanced by voice
significant range of standard deviation, i.e. between 25.190 buffering algorithms. Although, seeing that the averages may
and 33.643 ms. Because RTT takes only positive values, that not give the full picture, the second and third quartiles should
relation suggests that there is not only a wide variation, but be of interest. The second quartile (the median) shows that
also an asymmetry in the distribution. As it turns out, in the 50% of the differences between two samples are lower than
studied case there are fewer samples of high values, with single milliseconds. Nevertheless, the third quantile indicates
some as high as 992.670 ms, while the most frequent ones that the highest 25% of differences have exceeded 23.075 ms
were as low as only 5.422 ms. As cumulatively depicted with the highest one reaching 467.485 ms, what might have
in Figure 13, the distribution of RTT throughout 7 days led to a glitch if a voice call had been taking place.
examination is visibly long-tailed with 69.55% of samples not Although the stationary access evaluation is based on a
higher than 20 ms, 74.92% not higher than 30 ms and 94.55% particular case of synthetic tests, it gives a solid overview of
not exceeding 80 ms. what can be expected in similar conditions. Aside from the less
Since a variation in the RTT is visible, it is of key impor- delay and jitter demanding 2 Mbit/s web browsing, a seamless
tance to investigate how exactly, from the user’s perspective, voice call might not be possible in every usage scenario related
the delay changes in time, so in other words, to scrutinize to INEA_HotSpot_WiFi service, but certainly there is a
the jitter. Hence, the focus should be put on one way jitter, degree of freedom a community member can enjoy and use
as one of the key parameters used by network designers and the service also outside the apartment of a fellow member.
service providers to assess the quality of real time services. Moreover, the community Wi-Fi can be considered a useful
One way jitter presented in Table II has been calculated as Wi-Fi offload solution, enabling users to use the local radio
the half of the mean absolute difference between every two communication, instead of transferring the data through 3G or
consecutive RTT samples, yielding results between 9.211 ms LTE networks. Still, the further from the nearest community
and 13.702 ms. In most cases, these values satisfy the rule Wi-Fi access point, the higher the likelihood of the packet
of a thumb that recommends the end-to-end jitter for VoIP drops and delay variations.
communication not to exceed 30 ms [24]. Even if we double
the results for the sake of the other party operating in similar IV. C ONCLUSIONS
last-mile conditions and leave a few ms safety margin for Little is known about performance of commercially avail-
the jitter in the backbone networks, the call quality should able hotspot networks. Therefore, this article shares practical

50%
60

40%
RSSI [−dBm]
50 RTT [ms]
Distribution

30%

40
20%

30
10%

20

1 2 3 4 5 6 7 10 20 30 40 50 60 70 80 90 100 200 1000

Day RTT [ms]

Figure 12. Mean values of RSSI and RTT to the gateway throughout 7 days Figure 13. Distribution of RTT measured to the gateway in stationary test of
of stationary test of a residential INEA_HotSpot_WiFi access point a residential INEA_HotSpot_WiFi access point
PAPER INVITED TO 13TH INTERNATIONAL SYMPOSIUM ON WIRELESS COMMUNICATION SYSTEMS 2016, 20–23 SEPTEMBER 2016, POZNAŃ, POLAND 8

observations of a versatile service provider and presents the [6] “Miejski Zakład Komunikacji w Koninie.” http://www.mzk-konin.com.
analyses performed in mobile and residential INEA Wi-Fi pl
[7] “RouterBOARD 751 series.” http://i.mt.lv/routerboard/files/
hotspot network in comparison with INEA stationary 5 GHz rb751U-2HnD-ug.pdf, 2012. User’s Manual.
Wi-Fi access network. The paper reports results of experiments [8] “Przelewy24 – Payments, transactions, transfers, credit cards.” http://
performed over the period of one month. Some trends are www.przelewy24.pl/eng/
[9] “RouterBOARD RB433.” http://i.mt.lv/routerboard/files/
clearly visible, such as, the peaks in user number in the rb433-151007115417.pdf, 2015. Brochure.
rush hours (mobile hotspots) and in the afternoon (residen- [10] “Cisco Model EPC3925 8x4 EuroDOCSIS 3.0 Wireless
tial hotspots), or the general characteristics of noise floor Residential Gateway with Embedded Digital Voice Adapter.”
http://www.cisco.com/c/dam/en/us/products/collateral/video/
distribution, regardless of the type or location of an access cable-modems-digital-voice-emta-edva/7018345_C.pdf, 2011. Data
point. Moreover, a survey of 2.4 GHz channels usage has Sheet.
been presented with the statistics of Wi-Fi security measures [11] Protected EAP Protocol (PEAP) Version 2, 2004. IETF EAP Working
Group Internet-Draft.
deployed in urban environment. Community Wi-Fi user expe- [12] IEEE Standard for Information technology, Part 11: Wireless LAN
rience evaluation has been conducted and discussed, proving Medium Access Control (MAC) and Physical Layer (PHY) Specifica-
that the service fulfils the requirements of VoIP transmission tions, Amendment 5: Enhancements for Higher Throughput, 2010. IEEE
Std 802.11nTM -2009.
and web browsing. The synergy achieved by the combination [13] Data-Over-Cable Service Interface Specifications DOCSIS R 3.0 –
of two types of hotspot network provides coverage not only MAC and Upper Layer Protocols Interface Specification, 2015. Cable
in the city centre but also in more remote urban areas reached Television Laboratories, Inc. CM-SP-MULPIv3.0-I29-151210.
[14] IANA-Reserved IPv4 Prefix for Shared Address Space, 2012. IETF
by INEA cable network. Request for Comments: 6598.
[15] IEEE Standard for Local and metropolitan area networks — Port-Based
Network Access Control, 2010. IEEE Std 802.1XTM -2010.
R EFERENCES [16] “Nexus tech specs.” https://support.google.com/nexus/answer/6102470
[1] Wi-Fi Alliance, “Wi-Fi Alliance R
publishes 6 for ’16 Wi- [17] “WiFi Tracker.” https://play.google.com/store/apps/details?id=org.prowl.
Fi R
predictions.” http://www.wi-fi.org/news-events/newsroom/ wifiscanner
wi-fi-alliance-publishes-6-for-16-wi-fi-predictions, 7 January 2016. [18] “iPhone 6 Roaming Behavior and Optimization.” http:
[2] K. Kowalik, D. Dudek, M. Kołodziejski, B. Musznicki, E. Grzybek, //www.cisco.com/c/en/us/td/docs/wireless/controller/technotes/8-0/
and J. Jarzina, “Lessons Learned from WiMAX Deployment at INEA,” iPhone_roam/b_iPhone-roaming.html
Journal of Telecommunications and Information Technology, pp. 34–41, [19] “CSipSimple.” https://play.google.com/store/apps/details?id=com.
September 2014. http://www.itl.waw.pl/czasopisma/JTIT/2014/3/34.pdf csipsimple
[3] P. Walkowiak, R. Szalski, B. Musznicki, D. Dudek, K. Kowalik, [20] Pulse Code Modulation (PCM) of Voice Frequencies, 1990. ITU-T
and P. Zwierzykowski, “Evaluation of CARMNET System in INEA Recommendation G.711.
HOTSPOT Network,” in Proceedings of 2014 IEICE ICTF, Information [21] “Speedtest.net by Ookla.” http://www.speedtest.net/my-result/a/
and Communication Technology Forum, (Poznań, Poland), 28–30 May 2078856611, 21 July 2016.
2014. http://i-scover.ieice.org/proceedings/ICTF/2014/pdf/ALIEN-4.pdf [22] “Homedale – Wi-Fi / WLAN Monitor.” http://www.the-sz.com/products/
[4] “HotSpot – darmowy internet poza domem.” https://www.inea.pl/ homedale/
superdodatki/dla-ciebie/hotspot [23] “hrPing.” https://www.cfos.de/en/ping/ping.htm
[5] “MPK Poznań – About us.” http://www.mpk.poznan.pl/about-mpk/ [24] V. Joseph and B. Chapman, Deploying QoS for Cisco IP and Next Gen-
about-us eration Networks: The Definitive Guide. Morgan Kaufmann Publishers
Inc., 2009.
* All websites accessed on 20 July 2016, if not stated otherwise.