interface selection and handover application for android-based mobile devices

Accepted Manuscript

Energy-aware network/interface selection and handover application for android-based mobile devices Mehmet Fatih Tuysuz , Murat Ucan PII: DOI: Reference:

S1389-1286(16)30415-7 10.1016/j.comnet.2016.11.021 COMPNW 6063

To appear in:

Computer Networks

Received date: Revised date: Accepted date:

1 June 2016 25 October 2016 29 November 2016

Please cite this article as: Mehmet Fatih Tuysuz , Murat Ucan , Energy-aware network/interface selection and handover application for android-based mobile devices, Computer Networks (2016), doi: 10.1016/j.comnet.2016.11.021

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

Energy-aware Network/Interface Selection and Handover Application for Android-based Mobile Devices Mehmet Fatih Tuysuz, Murat Ucan Department of Computer Engineering – Harran University [email protected], [email protected] Abstract—Considering recent number of energy-hungry applications, large-screen mobile devices, fast processors, multiple hardware integrated network connectivity, high amount of data consumption and audio-video communication times, it is clear

CR IP T

to say that high battery capacity is what users ask for. However, although processing power doubles roughly every two years, progress in battery technology did not even double in the last decade. In order to meet the ever growing demand, vendors can simply increase battery sizes of mobile devices or focus on novel energy-efficient hardware and software solutions. In this context, this paper1 proposes an energy-aware network/interface selection and handover application for Android based mobile devices. The proposed application computes and reports power consumption levels of each Point of Attachment in the vicinity for various web-applications (e.g. Facebook, Twitter, Skype, etc.), making use of real packet measurements and realistic

AN US

computations, and then enables stations to handover horizontally/vertically to optimize energy efficiency. Results of extensive tests clarify that the proposed scheme not only saves energy but also increases overall throughput and hence, provides a better service quality.

Keywords: Energy Efficiency, Network Selection, Android, IEEE 802.11, Wireless Networks, Cellular Networks

1.

INTRODUCTION

M

Wireless networks and mobile devices have been experiencing an outstanding progress. Users demand uninterrupted, continuous, and seamless services with Quality of Service (QoS) from any source to any device at any time while on the

ED

move or stationary. In order to satisfy the increasing traffic demands and the service requirements, the next generation of wireless infrastructures (5G networks) paradigm will include a high deployment of base stations (BS) and several different radio access technologies (RATs), such as Wireless Local Area Networks (WLAN), Long Term Evolution

PT

(LTE), Worldwide Interoperability for Microwave Access (WiMAX), etc. as illustrated in Figure 1. However, there is no single RAT that can simultaneously offer high amount of bandwidth, low-latency, wide coverage and high QoS levels for

CE

mobile users. Therefore, next generation wireless systems have to make use of various solutions and technologies that enable a cooperative heterogeneous wireless environment where users will be always best connected (ABC) anytime and anywhere [1]. In this context, the main promise of the heterogeneous network integration is to provide mobile users with

AC

high performance and wide coverage, maintaining optimal energy efficiency.

In wireless networks, handover, also known as handoff, is the procedure of shifting an ongoing call or a data session from one Point of Attachment (PoA) to another. Consequently, handover procedure allows mobile stations to dynamically associate with the most suitable PoAs among available ones. If a handover occurs within the domain of a single RAT, it is called as horizontal handover. On the contrary, vertical handover (VHO) takes place among different RATs. Figure 1 demonstrates both horizontal and vertical handover procedures. 1

This work was supported by The Scientific and Technological Research Council of Turkey (TÜBİTAK) under Grant No: 114E075.

CR IP T

ACCEPTED MANUSCRIPT

Figure 1. Next Generation Network Scenario

As stations in heterogeneous wireless networks continuously seek channels to initiate horizontal or vertical handovers,

AN US

designing an energy-aware well-performed vertical handover procedure is significant to minimize the energy consumption while still supporting essential QoS. Handover duration and its accuracy is also essential for the energy efficiency. It is because, a possible improper association to a new network may let stations consume even more power than before until a proper association, if ever, is selected [2].

M

Research studies [3-6] show that, high amount of energy, up to 10–50%, is consumed by wireless interfaces (e.g., WiFi, GSM, 3G, LTE) among all energy-hungry components of a mobile device. Other researches related to the energy management of WiFi networks show that an important amount of power is also wasted at the Access Points (APs) [7, 8]

ED

due to network contention. In a nutshell, high power consumption is one of the major focus for both wireless interface and mobile device design, as the progress of current batteries is not following the Moore‘s Law [4].

PT

In the literature, there have been many works intend to decrease power consumption of mobile devices, such as an energy-aware Medium Access Control (MAC) [9], link adaptation for power-aware transmission and reception [10],

CE

optimum RF transmission power level [11], collision resolution mechanism [12], energy-aware periodic channel scanning [13], etc. All of the aforementioned methodical solutions actually reduce power consumption after a received signal strength (RSS) based association to a PoA has been already made. However, from a station‘s point of view, which

AC

PoA to associate with is also essential to optimize energy saving. Associating with a proper PoA, which is expected to let a station consume the least amount of energy among all PoAs, may reduce power consumption dramatically from the beginning of first transmission until the end of the association.

Existing energy-efficient network selection proposals [14-28] are either mobile-initiated or network-assisted and mostly operate based on a particular metric like cost-function [14, 16], traffic-volume [17, 19], fuzzy-logic [18,26], mobility pattern-aware [20], location-assisted [21], context-aware [22], speed-sensitive [24], delay-tolerant [27], etc. However, well-known three bottlenecks of these approaches are (i) computation of power consumption with inadequate

ACCEPTED MANUSCRIPT

information, (ii) using exactly the same energy forecast for different RATs, and (iii) obtaining results under a particular channel/traffic condition.

In the literature, there have been several Android-OS based mobile applications that suggests stations which network to associate with, in case there are more than one network in vicinity that a station can connect. Some of them are; WiFi Analyzer Lite developed by Martin Hloušek, WiEye - WiFi Scanner developed by Smuwireless, WIFI Scan developed by OHT, Wifi Scanner developed by Anda Studio, Wi-Fi Analytics Tool developed by Amped Wireless, Wifi Analyser

CR IP T

developed by keuwlsoft, etc. However, all of these applications consider only the Wi-Fi interface and hence vertical handover is not supported. Besides, these applications take only the RSSI value as input and hence, provide an RSSbased AP list in descending order. Although, received signal strength is the most significant parameter for a station to measure its performance and also the expected power consumption, an estimation made without taking the channel density and the collision probability in each AP into account would be misleading. In fact, an AP that has a lower RSS and low channel utilization can be more energy-efficient than an AP that has a high RSS with high channel utilization.

AN US

There have been also a couple of works [36, 37] on context-enhanced techniques, which discuss exactly the problem of the cost of probing RATs. Authors in [36] state high amount of energy requirement of multi-interface smart devices. As a solution, they propose a technology agnostic method, which is a context-aware node discovery (CANDi) algorithm. CANDi provides a priori knowledge towards the node discovery mechanism by allowing it to search nodes in the near vicinity at the ‗right time and at the right place‘. To validate the efficiency of CANDi, authors realize extensive tests and

M

results show that CANDi can save high amount of energy (up to 50%) during the node discovery process. Additionally, authors in [37] study the effective utilization of WiFi – WiMax radio access technologies from the end users' point of view. In this context, authors propose a novel handover decision mechanism in a WiFi-WiMAX integrated HetNet

ED

environment. The proposed method, in short, designs a handover policy in a WiFi-WiMAX HetNet environment, so that cost-to-pay per bit and hence the average power consumption, are minimized.

PT

Although there have been many analytical/algorithmic/computational works on energy-efficient interface selection in the literature, there is not any application implemented in practice that can support energy-aware vertical handover. In our

CE

previous work [29], we proposed a computational method that assists mobile devices to associate with a PoA that is expected to consume the least amount of energy among all PoAs, considering essential parameters, such as RSS, channel utilization, collision probability, traffic class of the station and power consumed in each wireless states. In this work, we

AC

further our former work and present an energy-aware network/interface selection and handover application that can run in Android-based mobile devices. To the best of our knowledge, this is the first application in Google Play Store that enables stations energy-aware network/interface selection and handover. The proposed application shortly computes and reports power consumption levels of each PoA in the vicinity for various web-applications (e.g. Facebook, Twitter, Skype, etc.), making use of real packet measurements and realistic computations, and then enables stations to handover horizontally/vertically to optimize energy efficiency.

ACCEPTED MANUSCRIPT

2.

HANDOVER IN WIRELESS & CELLULAR NETWORKS

A. Handover in Wireless Local Area Networks WLAN [30] is a local area wireless networking technology that mainly uses the 2.4 and 5 GHz radio bands. The traditional procedure used for a WLAN handover starts with the channel-scanning phase. In order to detect available networks, stations initially transmit Probe Request Frames and wait for Probe Response Frames on each channel. With the end of the channel-scanning phase, stations obtain a list of APs, their signal strengths, available transmission modes,

CR IP T

etc. [13].

After the channel scanning, Re-authentication phase, the procedure of transferring associations from one AP to another, starts. Authentication is essential to associate to the next AP. As soon as the station has been authenticated with the next AP, the re-association phase starts. With the end of this phase, the station associates to the next AP. It should be noted that, channel scanning is the main factor that dramatically affects the handover latency and the power consumption.

AN US

Therefore, it has to be limited to provide seamless and energy-efficient handover operation. B. Handover in 3G Networks

3G is the third generation of mobile telecommunications technology. 3G networks mainly have three types of handover operation; (i) hard handover, (ii) soft handover and (iii) 3G-GSM inter RAT handover. In hard handover, connections are first broken and then re-established. Hence, users sometimes may notice a short communication break. In soft handover,

M

the device is connected to more than one cell throughout the handover process. As it has more than one connection active throughout the handover, soft handover leads to more consistent and seamless communication opportunity. In addition to

ED

the hard and soft handover, handover between a 3G and a 2G GSM network is called inter-RAT handover [31].

The Radio Network Controller (RNC) manages the 3G handover decision. As in WiFi environment, RNC initiates a

has a better RSSI exists.

PT

handover if the RSSI of a specific communication channel reduces below a certain threshold and a different channel that

PROPOSED ENERGY-AWARE NETWORK SELECTION APPLICATION

CE

3.

A. Protocol Description

AC

In a network that has more than one station, an important amount of energy is mostly wasted due to issues, such as idle listening2, frame errors, collisions, overhearing3, protocol overhead and traffic fluctuations. Consequently, high RSS reduces the probability of frame errors and provides higher energy efficiency and performance. Accordingly, association with an AP that has low channel utilization (no station or only a few stations connected to the AP) also reduces the probability of collisions and duration of idle listening. Simply, association with an AP that has high RSS and low channel utilization can provide stations with optimum energy saving. 2

Stations mostly waste energy staying in receiving state as there is no information when to receive a frame(s). Besides, they also waste energy staying in idle state when competing with other stations for the channel access. 3 Stations may receive packets that are sent by other stations in the same network.

ACCEPTED MANUSCRIPT

A mobile device connected to an AP can be in one of four states: transmission, receiving, idle and doze state. Most power is consumed in the transmission state since all bits are transferred one by one until the last one. Mobile devices also consume high amount of power both in receiving and idle states (see Table 1). Devices consume the least power in the doze state as Power saving mode (PSM) deactivates network interface through passive periods. Nevertheless, applications are mostly coded to maximize frame transmission, therefore, they usually remain active. In this context, our proposed application measures signal strengths and channel utilizations of APs in the vicinity in real-time. The proposed

CR IP T

application also compares power consumption levels of each AP for six well-known web-applications (Facebook, Twitter, Google, YouTube, WhatsApp, and Skype), making use of real packet measurements and realistic computations.

In fact, there are three different ways of measuring/estimating signal strength and channel utilization of an Android device; (i) channel scanning with iw utility, (ii) sniffing with Wireshark, or (iii) transmitting pings to APs.

iw is an nl80211-based CLI configuration utility, which supports all drivers that have been added to the kernel [32]. To

AN US

be able to use iw, root permissions has to be granted by the device. In this context, execRootCmd function, which enables command prompts on the terminal running at application layer, is used to enable iw utility.

Using the iw utility, stations can scan WiFi channels and get channel scanning results. Simplest iw command to perform a full channel scanning is; "iw dev wlan0 scan‖. Running this command, stations can learn the SSIDs and received signal

M

strengths of each AP. However, channel utilization ratios of APs cannot be measured through this process. A particular AP running in a specific channel frequency, such as ―MyWiFi‖ in 2437 MHz, can also be selected and channel scanning can be performed only for the AP on the selected channel with the command; ―iw dev wlan0 scan freq 2437 MyWiFi‖.

ED

What is actually done here is transmitting a probe request frame to the AP and then, receiving a probe response frame from the AP.

PT

The iw-based Round Trip Time (RTT) in between a probe request transmission and a probe response reception can be

CE

defined as,

(1)

AC

Multiple RTT intervals obtained from APs can actually let stations analyze channel utilization ratios of each AP. For instance, if RTT intervals of AP1 is higher than RTT intervals of AP 2, we can simply say that channel-busy-time (CBT) of AP1 is higher than CBT of AP2. In case both AP1 and AP2 have same/close signal strengths, a station associated with the AP1 stay more in the idle state and hence, waste more energy. Consequently, channel utilization ratios of APs can be estimated if stations transmit multiple probe requests, and wait for the replies to analyze them. In this work, we first implemented stations with the iw utility and performed SSID-aware specific channel scannings. However, obtained RTT intervals were too high (e.g. 60-70 ms.) than we were expecting (e.g. 2-3 ms.). In order to find out the reason of this differentiation, we added a laptop to our test-bed environment and sniffed all frames

ACCEPTED MANUSCRIPT

transmitted/received with the Wireshark sniffing application. Accordingly, we have seen that; in an iw-based channel scanning process, stations transmit multiple (4-5 times) probe request frames and the AP response with multiple (8-12 times) probe response frames. Due to the aforementioned procedure of the iw utility, we decided not to use the iw as it is impossible for stations to compute the average

intervals of APs without sniffing the frames from a central point.

Second way of measuring/estimating the channel utilization ratios of APs is sniffing all transmitted frames between stations and APs. However, as briefly mentioned, sniffing requires additional server that monitors the traffic of APs. It

intervals of APs. Therefore, it is also not efficient to use.

CR IP T

also requires additional message exchanges between stations and the sniffer to let stations know about specific RTT

The last way of measuring/estimating the channel utilization ratios of APs is to connect those APs one by one and then transmit/receive control frames between the station and its AP for a while to analyze the traffic density of each AP. Simplest way to do it is the ping messages. A station can transmit multiple pings (e.g. 100 pings) to its connected AP and

AN US

then compute the average RTT interval. The station can repeat this procedure for all other APs in the vicinity and then compare the RTT results of different APs. In this context, Figure 2 shows the flowchart of the proposed energy-aware AP selection algorithm.

As illustrated in Figure 2, the proposed algorithm lets stations perform full channel scanning at first. In case the number of scanned APs is more than five, the proposed algorithm forms an RSS-based AP list and select only the first five AP

M

that have higher RSS than the others. Selection is made to limit the total running time of the algorithm. Besides, APs that have low RSS most probably consume more power than the first five. In case there are no more than five APs, the proposed algorithm forms an RSS-based AP list in descending order. Later on, the proposed algorithm sets a dynamic test

ED

counter according to the number of APs in the list. In order to limit/reduce the total running time, test counter is set to 20, 25, 33, 50 and 100 pings, when the number of APs that will be analyzed is 5, 4, 3, 2 and 1, respectively. This way, either the number of APs that will be analyzed is 5 or 1, it will take close times for our algorithm to terminate and show results

PT

in all scenarios. After the counter set, the station implemented with the proposed algorithm associates with the AP that has the highest RSS in the list. The station transmits ping messages (the number of transmitted pings can be 20, 25, 33,

CE

50 or 100 according to the counter set) to the AP, and save the average RTT interval, average signal strength and also the packet loss rate for the AP. The station then determines the IP addresses of the closest servers of popular webapplications (e.g. Facebook, Twitter, Google, YouTube, WhatsApp and Skype) by; “nslookup host_url” command.

AC

nslookup is a Linux terminal command that is used to find IP address(es) of the queried link on DNS server. Whenever the station learns these IP addresses, it transmits ping messages to the IP addresses of Facebook, Twitter, Google, YouTube, WhatsApp and Skype in an order, and then saves the average RTT intervals, average signal strengths and also the packet loss rates for all of these web-applications. With the end of this process, the algorithm removes the currentlyanalyzed AP from the list and checks the list whether there is an AP to be analyzed. If so, the algorithm repeats the same procedure until there is no AP to be analyzed.

M

AN US

CR IP T

ACCEPTED MANUSCRIPT

ED

Figure 2. Flowchart of the proposed energy-aware AP selection algorithm.

In order to minimize processing time in the analysis phase of the application, multi-threaded programming is

PT

implemented. In this regard, AsyncTask is used to manage waiting processes of threads that work simultaneously, and require results of an earlier thread(s) to terminate. AsyncTask class enables processes that run on different threads in the

CE

background to integrate into the main thread that run on foreground with ease. Thus, drawback of multi-threaded structure, in terms of processes waiting for one another, is also handled, making use of the AsyncTask. As an example, multi-threaded flowchart of the proposed energy-aware network selection application in case there are only 2 available

AC

APs is illustrated in Figure 3.

AC

CE

PT

ED

M

AN US

CR IP T

ACCEPTED MANUSCRIPT

Figure 3. Multi-threaded flowchart of the proposed ping transmissions when there are two available APs.

As is shown in Figure 3, ping transmissions for the Facebook are handled by Thread-3 in the analysis of the first AP. Whenever these ping transmissions are done, Thread-4 is created and ping transmissions for the Twitter are handled. While Thread-4 works on Twitter pings, at the same time, Thread-3 analyses ping results of Facebook coming from the Linux terminal commands, computes average RTTs and makes the system ready for the analysis of power consumption.

ACCEPTED MANUSCRIPT

While Thread-4 works on Twitter pings, Thread-3 and Thread-2 can simultaneously analyze the ping results of the AP 1. However, Thread-4 is never created before Thread-3 finishes the ping transmissions. This way, simultaneous ping transmissions for two different web-applications (i.e. Facebook and Twitter) are prevented. In fact, simultaneous ping transmission for more than one web-application results in higher average RTT results than expected, which means inaccurate computations. In a similar way, in order to create Thread-9 and work on the analysis of the AP 2, the proposed algorithm waits for the analysis process of the AP1 to be completed. This way, switching or in other words; associating

CR IP T

and disassociating between APs are lowered to the minimum level.

It should be noted that, the proposed algorithm utilizes ping messages transmitted consecutively with 1 millisecond (ms.) intervals. This means a station can transmit 100 ping messages in 100 ms. Besides, to be able to transmit pings with 1 ms. intervals, root permissions have to be granted by the device. The command that lets stations transmit ping messages with 1 ms. intervals is; ―ping –c ping_count –i 0.001 host‖, where -c is used to set how many pings (ping_count) will be sent, -i is used to set the transmission interval in between two consecutive pings. The value 0.001 means the interval is set as 1

AN US

ms. Finally, the host is the IP address or the URL that the pings will be transmitted to. B. Computation of Expected Amount of Energy Consumption

With the end of the analyzation phase, the proposed algorithm allows stations to compute expected amount of power consumption for each AP separately in case of a communication with one of the abovementioned web-applications. The proposed algorithm also compares unit power consumption of each AP and suggests stations application-based energy-

M

aware AP selection. In this regard, total amount of expected power consumption of a station‘s WiFi interface can be formulated as below,

)

(

ED

(

)

(

)

(2)

where Pt(i,j), Pidle(i,j) and Pr(i,j) are the amount of power consumption of the network i and the traffic type j in the

PT

transmitting, idle and receiving states, respectively. tt, tidle and tr are the time intervals that the station will stay in the transmitting, idle and receiving states.

CE

depends on the number of frames that will be transmitted, data rate, frame size, frame error rate (FER) and collision can be computed as follow, ( )

AC

probability Pc. Accordingly,

(

It should be noted that ( ) is the size of a ping message, of transmitted pings. Finally, (

)

(

)

(

)

(3)

is the transmission rate of ping messages,

is the number

) is equal to the packet loss rate measured by the proposed algorithm.

Consequently, expected amount of power consumption of a station in the transmitting state is approximately equal to, (

)

( )

(

)

(

)

(4)

ACCEPTED MANUSCRIPT

where

is the power consumption of the network interface card in the transmitting state. In a similar way,

is

approximately equal to, ( )

(

)

(

)

(5)

Accordingly, expected amount of power consumption of a station in the receiving state is approximately equal to,

where

( )

)

is the downlink transmission rate and

(

(

)

(6)

is the power consumption of the network interface card in the

receiving state. Finally,

)

CR IP T

(

depends on the experiment time texp (i.e. time interval between the first and the last ping) and the time spent

(7)

AN US

in the transmitting and receiving states, respectively.

Accordingly, expected amount of power consumption of a station in the idle state is approximately equal to, (

)

(

)

(8)

In order to analyze signal quality and compute the average RTT interval of a station connected to a 3G network, the

M

proposed application transmits pings from the station to the IP addresses of the closest servers of abovementioned webapplications. Afterwards, the proposed application allows stations to compute the expected power consumption values for

ED

these web-applications. In this regard, expected amount of power consumption of a station connected to a 3G network is computed as follow,

frames (

PT

In the Cell_DCH state, a station consumes energy both to maintain the Cell_DCH state (

), and to transmit or receive

). Consequently, total amount of expected power consumption of a station‘s 3G interface in the Cell_DCH

AC

CE

state can be formulated approximately as below, (9)

Since the goal is to compute the energy cost of transferring n Mbps data, the number of transfer blocks (Ntb) required for transferring n Mbps data can be formulated as the ratio of n to the Maximum Transport Block Size (MTBS). (

)

It should be noted that the energy cost of transferring n Mbps data also depends on the packetization interval (Ip) of frames and the successful transmission rate (S(Ci)) between the device and the associated PoAi. Therefore, the total amount of expected power consumption of a station‘s 3G interface can be re-formulated as below,

ACCEPTED MANUSCRIPT

(

where

( )

)

(

( )

(

)

is the energy consumption of maintaining Cell_DCH state in unit of time, and

)

is the energy consumption

of transmitting/receiving one transfer block.

Finally, the parameters used by the proposed application to compute expected amount of energy consumption of a station

Name

Value

Transmitting state Idle state Receiving state Data rate DIFS SIFS Slot time

1.3 740 900 11 50 10 20

CR IP T

that is connected to a WLAN or a 3G network are shown in Table 1 and Table 2 4, respectively.

Unit

W mW mW Mbps µs µs µs

AN US

Table 1. Parameter values of IEEE 802.11b WLAN.

Name Transmitting state Idle state Receiving state EDCH Data rate

Value

Unit

2.8 82 495 850 7.2

W mW mW mW Mbps

ED

C. Overview of the Proposed Aplication

M

Table 2. Parameter values of 3G network.

The proposed energy-aware network selection and handover application shortly enables mobile devices to discover available RATs and their channel frequencies in the vicinity, order them according to the expected amount of energy

PT

usage, and then handover to the most energy-efficient PoA, with the user confirmation. In this context, Figure 7-a shows the home screen of the proposed energy-aware network selection and handover application that can be downloaded from

CE

Google Play Store. Through the home screen, it is possible to perform channel scanning and ordering networks based on expected amount of power consumption (with the Start the Process button), add known Wi-Fi networks (Figure 7-b), and

AC

activate movement detection (Figure 7-c).

4

Parameters in Table 1 and Table 2 are derived from the works in [30], [33], [34] and [35].

(a) Home screen

(b) Addition of Known Wi-Fi

CR IP T

ACCEPTED MANUSCRIPT

(c) Movement Detection

AN US

Figure 7. Energy-aware network selection and handover application for Android devices.

Additionally, Figure 8 shows the test results of a sample set-up that has 2 Wi-Fi and one cellular networks. While results of specific networks can be seen in single pages (Figure 8-a, and Figure 8-b), results of a specific web-application, such

AC

CE

PT

ED

M

as YouTube, can be seen in detail in a different page once clicked (Figure 8-c).

(a) Results of a Wi-Fi network

(b) Results of Cellular Network

(c) YouTube results

Figure 8. Test results of a sample set-up.

Google has started to provide ActivityRecognition and StepCounter services to the application developers with Android 4.3 Nexus5. ActivityRecognition service has the ability of interpretation activities, such as walking, running, stair climbing, by processing the information received via accelerometer, gyroscope, GPS and other sensors. StepCounter, on the other hand, provides both hardware and software support in order to perform step counter process accurately.

ACCEPTED MANUSCRIPT

The proposed application also utilizes the StepCounter service and hence, it can investigate movements of mobile device user. Since, StepCounter service also consumes energy5, initiatition of movement detection is left to the user as shown in Figure 7-c. Shortly, users can switch movement detection on and off as a user preference, by clicking ―Detect Movement‖ from the main menu of the application.

In Android applications, there is a class named service for processes that are working in the background and do not affect user-application relationship. Routine background processes, such as database updates and reporting daily error-logs, can

CR IP T

be implemented without disturbing the user, by creating a sub-service class. In the proposed application, if movement detection is turned on, it keeps working in the background even if the application is switched off, as the step counter process of the application is implemented with the aforementioned service structure.

The proposed application counts steps of the mobile device user in a background service, after movement detection is activated. If pre-defined step-counter threshold6 is exceeded, the application checks the signal quality of the associated PoA. In case step-counter threshold is exceeded and the signal quality of the current PoA is low, application informs the

AN US

user about low signal quality with a pop-up notification and ask the user whether to start the process of a new network/interface selection.

The proposed application calculates the signal quality of the PoA with the ―calculateSignalLevel (int rssi, int numLevels)‖ function that is found in Android SDK. This function takes the RSSI and the number of signal levels as

M

input parameters and then, computes the signal strength (level) based on given number of signal levels. In the proposed application, number of signal levels is set to 5, by utilizing the calculateSignalLevel function. If step-counter threshold is

ED

exceeded and the measured signal is less than 3 level, user notification process starts. 4.

EVALUATION OF THE PROPOSED APPLICATION

PT

Throughout this section, 6 different scenarios are tested, comparisons are made and the measured results and the results generated by the application are discussed in detail at the end of each evaluation scenario. These scenarios are shortly based on; (i) investigation of additional energy cost that is consumed by the application, (ii) investigation of the average

CE

RTT intervals, (iii) investigation of the impact of number of ping frames on the accuracy of channel density, (iv) investigation of the impact of AP channel utilization on the throughput and power consumption, (v) investigation of the impact of different signal strength levels and RATs on the throughput and power consumption, and (vi) investigation of

AC

the impact of handover operation on energy saving.

To the best of our knowledge, the proposed application is the first application in Google Play Store that provides stations with energy-aware network/interface selection and handover. Since there is no similar application in the literature, we

5

With Android 4.4, energy consumption of StepCounter is reduced dramatically with hardware sensor batching, as device's application processor remain in a low-power idle state until batches are delivered. 6 Step-counter threshold is set to 25 in the proposed application.

ACCEPTED MANUSCRIPT

were not able to compare our application with a similar application available. Instead, we have performed extensive tests to validate our approach, observing the efficiency of the proposed application in different scenarios. A. Investigation of Additional Energy Cost that is Consumed by the Application In addition to the energy consumed by standard procedures, in case the proposed application is running on a smartphone, the smartphone will consume additional energy based on the implementation of two new processes; (i) a process that performs required operations, such as channel scanning, transmitting ping frames and analyzing/finalizing the results, and

CR IP T

(ii) a process that focus on movement detection, utilizing the StepCounter service running on the background. While the first process (process_1) consumes instantaneous energy only when it is initiated, the second process (process_2) may consume energy as long as the application is in use.

It should also be noted that almost all kinds of smartphones/applications nowadays use StepCounter service. Besides, With Android 4.4, energy consumption of StepCounter is reduced dramatically with hardware sensor batching, as

AN US

device's application processor remain in a low-power idle state until batches are delivered. If this service is already activated by the smartphone, our application will only consume additional, yet negligible energy while requesting/receiving related data from the StepCounter service.

Nevertheless, to demonstrate the energy requirements of these two additional processes, we have performed specific measurements for four different states. Within the scope of this scenario, each of these four states was implemented one

M

by one, by utilizing a smartphone7 that has fully-charged battery, and is connected to an AP. Accordingly, energy consumptions were measured by a Multimeter per state after a 100-minute test period, and the results were transferred to

ED

Table 3.

The abovementioned four states are; (i) total energy consumption measured after a 100-minute test period when the

PT

proposed application is off, (ii) total energy consumption measured after a 100-minute test period when the proposed application is on, but the Movement Detection is off (process_1), (iii) total energy consumption measured after a 100minute test period when the proposed application is on, the Movement Detection is on (process_2), and the smartphone is

CE

stationary, and (iv) total energy consumption measured after a 100-minute test period when the proposed application is on, the Movement Detection is on, and the smartphone is frequently on move8.

AC

As is shown in Table 3, results show that additional energy consumed by the proposed application is in between 0.6% 2.2%. Consequently, even in worst case, additional energy consumption will not be more than 2.2% of the total energy consumption. Besides, if StepCounter service is already initiated by another application, then additional energy consumed by the application will be around 1%. When considering the probability of saving high amount of energy (e.g. 46%, see subsection E. of this section), we believe it is an affordable price to pay. 7

Samsung Galaxy S4 was used during the tests. The smartphone was fully-charged, screen was off and no additional application was in use in the beginning of all tests. 8 Throughout 100-minute test period, we have walked (frequently, not continuously) around our computer networks lab while having the smartphone in our pockets, and staying inside the coverage of the connected AP.

ACCEPTED MANUSCRIPT

State 1

State 2

State 3

State 4

100

100

100

100

Test period [min.]

91

91

90

88

Initial volt [v]

4.264

4.266

4.261

4.265

Remaining volt [v]

3.794

3.768

3.745

3.712

1

1.006

1.013

1.022

Remaining battery [%]

Consumption in unit time [x]

Table 3. Additional energy cost that is consumed by the application.

CR IP T

B. Investigation of RTT Intervals and Impact of Number of Ping Frames on the Accuracy of Channel Density In order to see the impact of channel utilization on the average RTT interval, we first run the proposed application for a case that there is only one AP to connect, yet the number of stations deployed by the AP varies in each scenario. In scenario 1, the AP has no station connected to itself, our smartphone connects to the AP and start pinging to analyze the AP. In scenario 2, our station connects to the AP and transmits pings when the AP has another station that is downloading a large file. Similarly, in scenario 3, 4, 5, 6 and 7, our station connects to the AP and transmit pings when

AN US

the AP has already 2, 3, 4, 5 and 6 stations that are downloading a large file, respectively. In each scenario, since there is only one AP to associate with, our smartphone transmits 100 ping messages to the AP, analyze the channel utilization of the AP itself, and then the smartphone transmits 100 pings to each of the IP addresses of the Facebook, Twitter, Google, YouTube, WhatsApp and Skype in an order. Results are derived from the application and can be seen in Table 4.

Sc. 1

Sc. 2

Sc. 3

Sc. 4

Sc. 5

Sc. 6

Sc. 7

1.6

2.7

3.9

4.4

5.6

9.1

10.2

Facebook

124.0

615.3

647.5

829.4

817.3

958.2

991.3

Twitter

180.2

726.2

827.3

949.6

991.9

1013

1058

101.1

444.2

656.2

801.5

903.9

947.2

987.0

ED

Google

M

AP itself

99.2

345.8

574.4

755.3

802.3

888.9

900.6

WhatsApp

157.2

816.1

907.3

911.6

972.2

1001

1086

Skype

81.3

376.3

606.5

855.6

911.5

935.7

978.6

PT

YouTube

CE

Table 4. RTT intervals (ms.) of the AP and servers in 7 different scenarios, when the number of transmitted ping is 100.

As it is clear from the results, RTT intervals dramatically increases (especially in Sc. 2) in parallel with the increase in the number of deployed station by the AP. Therefore, channel utilization of the AP has a direct impact on the RTT

AC

intervals. This clearly means that connecting to an AP that has high channel utilization will cause station to stay more in idle state and hence, waste more power. Consequently, in order to achieve energy efficiency, it is best for a station to associate with an AP that has none, or the lowest channel utilization as long as the signal strength is in an acceptable level.

Additionally, as mentioned through Section 3, the number of ping messages are decreased in case there is more than one AP that will be analyzed. This procedure may reduce the estimation accuracy of the channel utilization. However, as also

ACCEPTED MANUSCRIPT

mentioned earlier, there is a trade-off between application running-time and the total number of ping frames that will be sent. There is also a trade-off between additional energy consumption and the number of ping frames that will be sent. As an example, let‘s consider an environment that has at least five APs in its vicinity. In this case, the smartphone that runs our application has to connect each of these APs and then transmit ping frames to the AP itself, Facebook, Twitter, Google, YouTube, WhatsApp, and Skype, respectively. Therefore, 700 ping frames transmission is required for one AP. Since, the application has 5 APs on total, it requires transmission of 3500 ping frames. Since channel scanning,

CR IP T

connection and disconnection to APs already take a considerable time, with the addition of 3500 ping frames, total interval required for the application to select the most energy-efficient network/interface may take roughly half a minute. This interval is not practical as users demand quick access to the Internet. Therefore, in the proposed application, we set a dynamic test counter according to the number of APs in the list. In order to limit/reduce the total running time, test counter is set to 20, 25, 33, 50 and 100 pings, when the number of APs that will be analyzed is 5, 4, 3, 2 and 1, respectively. This way, either the number of APs that will be analyzed is 5 or 1, it will take close times (about 5-6

AN US

seconds) for our algorithm to terminate and show the results in all scenarios.

In order to minimize processing time in the analysis phase of the application even further, multi-threaded programming is also implemented. Nevertheless, in order to see the impact of the number of ping frames on the accuracy of the channel utilization, we have re-run the abovementioned set-up, where this time we let the smartphone transmit only 20 ping

M

frames per scenario regardless of the number of deployed APs in the vicinity. Results are shown in Table 5.

Sc. 1

Sc. 2

Sc. 3

Sc. 4

Sc. 5

Sc. 6

Sc. 7

AP itself

1.9

3.0

3.6

4.7

5.9

8.5

9.4

Facebook

167.6

513.0

Google

717.1

843.0

908.8

962.5

703.1

789.3

837.4

916.4

978.2

1012

142.4

489.0

695.3

884.2

967.9

1008

1093

ED

Twitter

602.9

156.6

90.6

351.4

495.0

625.7

782.9

846.1

918.5

WhatsApp

155.8

435.3

642.1

748.4

839.2

893.4

927.3

Skype

103.4

384.6

689.0

803.8

931.0

1011

1073

PT

YouTube

CE

Table 5. RTT intervals (ms.) of the AP and servers in 7 different scenarios, when the number of transmitted ping is 20.

As is shown in Table 4 and Table 5, RTT intervals of these two approaches approximate each other and show that as long

AC

as there is no network-based peak in congestion, or in number of collisions, less number of ping transmission also provides a good knowledge about the channel density. These results consequently validate our approach. C. Investigation of the Impact of AP Channel Utilization on Throughput and Power Consumption In this sub-section, we analyze the impact of AP channel utilization on the throughput and power consumption. Figure 9 illustrates our test-bed environment, where there are 4 APs, each of which has different amount of traffic (AP 1 has no station connected to itself, AP2, AP3 and AP4 have 1, 3 and 6 stations connected to themselves, respectively and all of

ACCEPTED MANUSCRIPT

these stations are downloading a large file) and there is one station (Samsung Galaxy S4 smartphone) in the center that is

CR IP T

implemented with the proposed algorithm and is looking for an AP to associate with.

Figure 9. IEEE 802.11 network communication scenario.

The proposed application lets the station perform a full channel scanning first and discover the above-mentioned four APs. The station associates with the AP1 at first and transmits ping messages (25 pings are sent for each AP due to

AN US

dynamic test counter) to the AP itself. Afterwards, the station transmits ping messages to the IP addresses of Facebook, Twitter, Google, YouTube, WhatsApp and Skype in an order, and then saves the average RTT intervals, average signal strengths and also the packet loss rates for all of these applications. With the end of the analysis of the AP 1, the algorithm removes the AP1 from the list and repeats the same procedure for the AP 2, AP3 and AP4.

Although the proposed application reports the results of six web-applications (Google, Skype, WhatsApp, Facebook,

M

Twitter and YouTube) separately per PoA association, throughout this section we will only focus one of them, which is YouTube, to simplify the comparison of throughput and power consumption in terms of channel utilization, signal

ED

strength and different RATs. Accordingly, expected amounts of power consumption of the station, which is computed by the proposed application, when connected to YouTube on the aforementioned four APs separately are shown in Figure 10-a. Measured energy consumptions9 in unit time10 and in unit throughput11 are shown in Figure 10-b. Finally, actual

AC

CE

PT

YouTube downlink throughput values of the station after an association with those APs are shown in Figure 10-c.

(a) Expected Power consumption per AP

(b) Measured energy consumptions per AP

(c) Throughput of the sta. per AP connection

Figure 10. Impact of AP channel utilization on the throughput and power consumption. 9

Energy consumptions were measured by a Multimeter after running the simulation set-up during a 100-minute test period. Energy consumed in unit time, measured by Multimeter. 11 Energy consumed to transmit/receive same amount of throughput. It is computed utilizing the consumption in unit time and the throughput rates. 10

ACCEPTED MANUSCRIPT

In this scenario, the proposed application suggests user to connect to the AP1 that has no station, as it is expected to save 1,53, 2,29 and 4,37 times more energy to transmit the same amount of throughput compared to the association with the AP2, AP3, and AP4, respectively. As seen in Figure 10-c, if the station connects to the AP 1, the station not only saves energy but also increases its throughput and hence provides a better QoS. In a similar way, measured results also validates that the AP1 is the most energy-efficient AP, and it is expected to save 1,73, 2,94 and 5,26 times more energy to transmit the same amount of throughput compared to the association with the AP 2, AP3, and AP4, respectively. As depicted in Figure 10-b, measured results of consumptions in unit throughput are a bit higher than the results of the

CR IP T

proposed application. It is because, in measurements, the smartphone itself also consumes power to stay awake and execute its background processes. Therefore, results of energy consumptions in unit time get closer to each other compared to the measurements of pure energy consumed only by the network/interface. Since results of energy consumptions in unit time get closer to each other, results of energy consumptions in unit throughput will be further compared to the results of the proposed application.

AN US

D. Investigation of the Impact of Signal Strength and RATs on Throughput and Power Consumption

In this sub-section, we analyze the impact of different signal strength levels and RATs on the throughput and power consumption. Figure 11 illustrates our test-bed environment, where there are 2 APs, each of which has one station (downloading a large file) connected to itself. While AP1 is positioned (in line of sight) very close (less than 5 meters) to the user, AP2 is positioned (not in line of sight) very far (more than 40 meters) to the user. Apart from the APs, there is one cellular network available, and there is also a station (Samsung Galaxy S4 smartphone) implemented with the

CE

PT

ED

M

proposed algorithm and is looking for a PoA to associate with.

Figure 11. Heterogeneous network communication scenario.

AC

As in the previous scenario, the proposed application lets the station perform a full channel scanning first and discover the above-mentioned two APs and one cellular BS. The station associates with the AP1 at first and transmits ping messages to the AP1 itself. Afterwards, the station transmits ping messages to the IP addresses of Facebook, Twitter, Google, YouTube, WhatsApp and Skype in an order, and then saves the average RTT intervals, average signal strengths and also the packet loss rates for all of these web-applications. With the end of the analysis of the AP1, the algorithm removes the AP1 from the list and repeats the same procedure for the AP 2, and the BS of the cellular network. Accordingly, expected amounts of power consumption of the station, which is computed by the proposed application, when connected to YouTube on the aforementioned two APs and one BS separately are shown in Figure 12-a. Measured

ACCEPTED MANUSCRIPT

energy consumptions in unit time and in unit throughput are shown in Figure 12-b. Finally, actual YouTube downlink

(a) Expected Power consumption per AP

(b) Measured energy consumptions per AP

CR IP T

throughput values of the station after an association with those PoAs are shown in Figure 12-c.

(c) Throughput of the sta. per AP connection

Figure 12. impact of different signal strength levels and RATs on the throughput and power consumption.

AN US

In this scenario, the proposed application again suggests user to connect to the AP1, as it is expected to save 1,46 and 1,92 times more energy to transmit the same amount of throughput compared to the association with the BS and AP2, respectively. In a similar way, measured results also validate that the AP1 is the most energy-efficient AP, and it is expected to save 1,59 and 1,89 times more energy to transmit the same amount of throughput compared to the association with the BS and AP2, respectively. Additionally, as seen in Figure 12-c, signal degradation results in a huge decrease on throughput (due to frame errors) and an important increase in power consumption (as it takes more time to transmit the

M

predefined same amount of data in an error-prone channel) for the AP2. Due to the low signal quality, AP2 is actually expected to consume even more power than the cellular network. Therefore, as suggested by the proposed application, in case the station connects to the AP1, it saves not only energy but also increases its throughput and hence, provides a

ED

better QoS.

PT

E. Investigation of the Impact of Handover Operation on Energy Saving In order to see the impact of handover operation on energy saving, we have performed another set-up, where there are two APs, and each of which is located in different lab rooms. AP1 is located in computer networks lab and has one station

CE

connected to itself, downloading a large file. On the other hand, AP2 is located in computer security lab, across the computer networks lab, and has no station connected to itself. These two APs and any station in between these APs can see the signals of each other. Accordingly, our smartphone, which is programmed to download small amount of packets

AC

continuously, is located in the middle of these rooms in the beginning of the test. The smartphone first visits the room where AP1 is located for 10 minutes. Then, it visits the other room where AP2 is located for the next 10 minutes. The same steps are repeated for the next 20 minutes, so that the smartphone visits both AP1 and AP2 for 20 minutes, which makes 40 minutes of test period on total. Test-bed environment can be seen in Figure 13.

ACCEPTED MANUSCRIPT

CR IP T

Figure 13. Handover-based network communication scenario.

When the proposed application is off, regardless of the movement, the smartphone always associates with the AP1 during the 40 minutes of test period. The smartphone achieves 1.8 Mbps average throughput rate when it is in the room where AP1 is located. On the other hand, it achieves 0.7 Mbps average throughput rate when it is in the room where AP2 is located. Consequently, average throughput rate for the smartphone is computed as 1.34 Mbps during the 40 minutes of

AN US

test period.

When the proposed application is on, the smartphone associates with the AP1 for the first 10 minutes, then performs handovers to the AP2, AP1 and AP2 again for the next 10 minutes of time periods, respectively. Accordingly, the proposed application performs handover operation three times during the 40 minutes of test period. In this case, the smartphone achieves 1.8 Mbps average throughput rate when it is in the room where AP1 is located, and 3.84 Mbps

M

average throughput rate (AP2 has no additional station connected to itself) when it is in the room where AP2 is located. Consequently, average throughput rate for the smartphone is computed as 2.71 Mbps during the 40 minutes of test

ED

period.

With the end of each test, we measured the energy consumed in unit time and also in unit throughput, utilizing the

PT

average throughput rates. In this context, Figure 14 shows the measured results of energy consumption in unit time, in

AC

CE

unit throughput and average throughput rates, when algorithm is on and off.

Figure 14. Measured results of energy consumption in unit time, in unit throughput and average throughput rates, when algorithm is on and off.

As is shown in Figure 14, when the proposed application is off, the smartphone requires 1.87 times more energy to receive the same amount of throughput, compared to the scenario where the proposed algorithm is on. In other words, in

ACCEPTED MANUSCRIPT

order to receive same amount of data, our proposed application consumes 46% less energy. Therefore, it is clear to say that in case there are more than one AP in a vicinity, and stations are also on move, performing a handover can dramatically increase both energy efficiency and throughput performance.

It should also be noted that, the proposed handover operation takes longer than a standard handover operation as our application additionally utilizes ping frames to have a good knowledge of the channel density. As also mentioned earlier, the proposed handover operation roughly takes 5-6 seconds to terminate. However, the impact of this interval depends on

CR IP T

the communication time. For instance, 5-6 seconds of handover delay is huge (and non-preferable) if the user is going to connect to a PoA only for seconds. Yet, the same user can save high amount of energy, as shown in Figure 14, in case he/she is going to stay connected for minutes. 5.

CONCLUSION

This work presents an energy-aware network/interface selection and handover application that can run in Android-based

AN US

mobile devices. The proposed application shortly computes and reports power consumption levels of each PoA in the vicinity for various web-applications (e.g. Facebook, Twitter, Skype, etc.), making use of real packet measurements and realistic computations, and then enables stations to handover horizontally/vertically to optimize energy efficiency. To enable mobile devices to associate with a PoA that is expected to consume the least amount of energy among all PoAs, the proposed application takes essential parameters into account, such as RSS, channel utilization, collision probability,

M

frame error rate, traffic class of the station and amount of power consumed in each wireless states.

Proposed application allows Android-based mobile devices to achieve fast and energy-efficient communication without degrading the user quality experience from the beginning of the first transmission until the end of the association. To the

ED

best of our knowledge, this is the first application in Google Play Store that provides stations with energy-aware network/interface selection and handover. Results of extensive tests clarify that the proposed scheme not only saves

PT

energy but also increases overall throughput and hence, provides a better service quality. REFERENCES

E. Gustafsson and A. Jonsson, "Always best connected," IEEE Wireless Communications, vol. 10, pp. 49-55, 2003.

[2]

M. F. Tuysuz and R. Trestian, "A Roadmap for a Green Interface Selection Standardization over Wireless HetNets," IEEE Globecom Workshops, San Diego, USA, 2015.

[3]

R. Kravets and P. Krishnan, "Power management techniques for mobile communication," in Proceedings of the 4th annual ACM/IEEE international conference on Mobile computing and networking, 1998, pp. 157-168.

AC

CE

[1]

[4]

G. Anastasi, M. Conti, E. Gregori, and A. Passarella, "Saving energy in wi-fi hotspots through 802.11 psm: an analytical model," in Proceedings of the Workshop on Linguistic Theory and Grammar Implementation, ESSLLI-2000, 2004, pp. 24-26.

[5]

802.11ac: The Fifth Generation of Wi-Fi, August 2012. Available: http://www.cisco.com/c/en/us/products/collateral/wireless/aironet-3600series/white_paper_c11-713103.html

[6]

802.11ac Technology Introduction, March 2012. Available: https://www.rohde-schwarz.com/us/solutions/wireless-communications/wlan-wifi/infocus/technology-introduction_106713.html

[7]

E. Rozner, V. Navda, R. Ramjee, and S. Rayanchu, "NAPman: network-assisted power management for wifi devices," in Proceedings of the 8th international conference on Mobile systems, applications, and services, 2010, pp. 91-106.

[8]

J. Manweiler and R. Roy Choudhury, "Avoiding the rush hours: WiFi energy management via traffic isolation," in Proceedings of the 9th international conference on Mobile systems, applications, and services, 2011, pp. 253-266.

[9]

Shiao-Li Tsao, Chung-Huei Huang, ―A survey of energy efficient MAC protocols for IEEE 802.11 WLAN‖, Computer Communications, v.34 n.1, p.54-67, 2011.

ACCEPTED MANUSCRIPT

[10] J. Zhao, Z. Guo, W. Zhu, "Power Efficiency in IEEE 802.11a WLAN with Cross-layer Adaptation", in Proceeding of ICC 2003, May 2003, pp. 2030--2034. [11] J.P. Ebert, B. Stremmel, E. Wiederhold, and A. Wolisz. ―An Energy-efficient Power Control Approach for WLANs‖, Journal of Communications and Networks (JCN), 2(3):197-206, September 2000. [12] C. Wang , B. Li and L. Li "A new collision resolution mechanism to enhance the performance of IEEE 802.11 DCF", IEEE Trans. Veh. Technol., vol. 53, no. 4, pp.1235 -1246, 2004. [13] Tuysuz, M. F., Mantar, H. A., ―Minimizing Communication Interruptions Using Smart Proactive Channel Scanning Over IEEE 802.11 WLANs‖, Wireless Personal Communications, vol. 82, no. 4, pp. 2249-2274, 2015. [14] Huaiyu L., Maciocco C., Kesavan V., Low A. L. Y., ―Energy efficient network selection and seamless handovers in Mixed Networks‖, World of Wireless, Mobile and Multimedia Networks & Workshops, 2009. WoWMoM 2009. pp.1-9, 15-19 June 2009. [15] Bastos J., Albano M., Marques H., Ribeiro J., Rodriguez J., Verikoukis C., ―Smart interface switching for energy efficient vertical handovers in ns-2‖, Communications, IET , vol.6, no.14, pp.2228-2238, September 25 2012.

CR IP T

[16] Inwhee J., Kim W. T., Hong S., ―A Network Selection Algorithm considering Power Consumption in Hybrid Wireless Networks‖, Computer Communications and Networks, 2007. ICCCN 2007, pp. 1240-1243, 13-16 Aug. 2007. [17] Minji Nam, Nakjung Choi, Yongho Seok, Yanghee Choi, ―WISE: energy-efficient interface selection on vertical handoff between 3G networks and WLANs‖, Personal, Indoor and Mobile Radio Communications, PIMRC 2004, pp. 692-698 Vol.1, 5-8 Sept. 2004. [18] Tran T., Kuhnert M., Wietfeld C., ―Energy-Efficient Handoff Decision Algorithms for CSH-MU Mobility Solution‖, Computer Communications and Networks (ICCCN), July 30 - Aug. 2, 2012. [19] Petander H., ―Energy-aware network selection using traffic estimation‖, In Proceedings of the 1st ACM workshop on Mobile internet through cellular networks (MICNET '09), New York, NY, USA, pp. 55-60. [20] Yang W. H., Wang Y. C., Tseng Y. C., Lin B. S. P., ―Energy-efficient network selection with mobility pattern awareness in an integrated WiMAX and WiFi network‖, Int. J. Commun. Syst., 23: 213–230, 2010.

AN US

[21] Bastos J., Albano M., Rodriguez J., Verikoukis C., ―Location assisted energy efficiency for multi-interfaced mobile terminals‖, ICC 2012, pp. 5660-5664, 10-15 June 2012. [22] Xenakis D., Passas, N., Di Gregorio, L., Verikoukis, C. ―A Context-Aware Vertical Handover Framework Towards Energy-Efficiency,‖ IEEE Vehicular Technology Conference, 2011. [23] Chowdhury M. Z., Yeong Min Jang, Choong Sub Ji, Sunwoong Choi, Hongseok Jeon, Junghoon Jee, Changmin Park, ―Interface selection for power management in UMTS/WLAN overlaying network‖, Advanced Communication Technology, 2009. ICACT 2009, pp. 795-799, 15-18 Feb. 2009. [24] Calhan A., Ceken C., ―Speed sensitive-energy aware adaptive fuzzy logic based vertical handoff decision algorithm‖, Systems, Signals and Image Processing (IWSSIP), 16-18 June 2011.

M

[25] SungHoon Seo, JooSeok Song, ―An energy-efficient interface selection for multi-mode terminals by utilizing out-of-band paging channels‖, Telecommunication Systems, Volume 42, Issue 1-2, pp. 151-161, October 2009. [26] Juan Fan, Sihai Zhang, Wuyang Zhou, ―Energy-Friendly Network Selection in Heterogeneous Wireless Networks‖, Vehicular Technology Conference (VTC Spring), 6-9 May 2012.

ED

[27] Kolios P., Friderikos V., Papadaki K., ―A practical approach to energy efficient communications in mobile wireless networks‖, Mobile networks and applications, Volume 17, Issue 2, pp. 267-289, April 2012. [28] SuKyoung Lee, SungHoon Seo, Golmie, N., ―An efficient power-saving mechanism for integration of WLAN and cellular networks‖, Communications Letters, IEEE , vol.9, no.12, pp. 1052-1054, Dec. 2005.

PT

[29] M. F. Tuysuz, "An energy-efficient QoS-based network selection scheme over heterogeneous WLAN – 3G networks," Computer Networks, vol. 75, Part A, pp. 113-133, 12/24/ 2014. doi:10.1016/j.comnet.2014.09.012. [30] IEEE Std 802.11, ―Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications‖, 1999 Edition. [31] I. Poole. UMTS WCDMA Handover: hard soft, softer, inter-RAT. Available: http://www.radio-electronics.com/info/cellulartelecomms/umts/umtswcdma-handover-handoff.php

CE

[32] https://wireless.wiki.kernel.org/en/users/documentation/iw [33] Kalic G., Bojic I., Kusek M., ―Energy consumption in android phones when using wireless communication technologies‖, MIPRO, 2012 Proceedings of the 35th International Convention, pp. 754-759, 21-25 May 2012.

AC

[34] Wang L., Ukhanova A., Belyaev E., ―Power consumption analysis of constant bit rate data transmission over 3G mobile wireless networks‖, ITS Telecommunications (ITST), pp. 217-223, 23-25 Aug. 2011. [35] Harjula E., Kassinen O., Ylianttila M., ―Energy consumption model for mobile devices in 3G and WLAN networks‖, Consumer Communications and Networking Conference (CCNC), pp. 532-537, 14-17 Jan. 2012. [36] M. Alam, et al. "CANDi: context‐aware node discovery for short‐range cooperation." Transactions on Emerging Telecommunications Technologies 26.5 (2015): 861-875. [37] A. Sarma, S. Chakraborty, and S. Nandi. "Deciding Handover Points Based on Context-Aware Load Balancing in a WiFi-WiMAX Heterogeneous Network Environment." IEEE Transactions on Vehicular Technology 65.1 (2016): 348-357.

ACCEPTED MANUSCRIPT

Author Biography

AC

CE

PT

ED

M

AN US

CR IP T

Mehmet Fatih Tuysuz holds the B.Sc. degrees from İnönü University, department of Electric and Electronic Engineering and Anatolian University, department of Business Administration. He holds the M.Sc. degree from Dokuz Eylül University, department of Electric and Electronic Engineering. During his M.Sc. thesis, he worked on the ―Quality of Service Enhancement of VoIP applications over wireless networks‖ and published several papers in this area. He joined Gebze Institute of Technology, department of Computer Engineering in 2008 as a Ph.D. student and graduated his Ph.D in 2013. Currently, he has been working at Harran University, Computer Engineering Department as an assistant professor. His interest includes VoIP, Wireless QoS, energy-aware communications and energy optimization in wireless networks.