GC'12 Workshop: Quality of Experience for Multimedia Communications Power-Driven VoIP Quality Adaptation Over WLAN in Mobile Devices Is-Haka Mkwawa and Lingfen Sun School of Computing and Mathematics University of Plymouth, Plymouth, PL4 8AA,UK (Is-Haka.Mkwawa,L.Sun)@plymouth.ac.uk Abstract Network parameters such as packet loss and delay have been extensively investigated as the major factors in determining when and how to carry out VoIP quality adaptation to enhance quality of experience. However, a little effort has been put into investigating the impact of non-network parameters such as limited battery resources in mobile devices on the VoIP QoE. Battery life which is not long enough to complete VoIP communication session will adversely affect users QoE. In order to mitigate battery life limitation, this paper proposes VoIP quality adaptation scheme whereby an acceptable quality is maintained by changing video send bitrate in order to conserve power and hence, prolong VoIP communication session. Through experiments, preliminary results have shown the effectiveness of the proposed scheme in terms of power saving while maintaining acceptable QoE. The power saving was between 1-3% of the total system power. Index Terms QoE, IMS, SIP, Android, WLAN, Quality Adaptation, energy, power I. INTRODUCTION Mobile phones are now packaged with a range of multimedia applications such as YouTube, Google Talk and Skype. However, these applications consume considerable amount of power especially when video communication is involved [1], [2]. While mobile devices technologies such as CPU and storage capacity have immensely evolved over the past few years, unfortunately, technologies to extend battery life has not experienced the same evolution. This limitation has prompted an extensive research into power conservation techniques such as power aware operating systems and mobile resources management [3]. The Quality of Experience (QoE) in VoIP applications is a subjective measure that takes into account what end users are experiencing when using VoIP services. The Mean Opinion Score (MOS) [4] is mainly used to measure QoE. The importance of QoE to VoIP service providers is vital. Poor QoE in VoIP services increase technical support budget, customers churn rate and lower revenue streams. International Telecommunication Union (ITU) have pioneered the investigation of VoIP quality in terms of QoE and proposed quantitative methods for measuring QoE [5]. However, the proposed methodology only considered network parameters such as packet loss and delay. It has been argued that it is not sufficient to measure QoE [6] by only considering This research is supported by the EU FP7 GERYON project (contract No. 284863). network parameters [7]. This argument triggered a broader definition of QoE which entails network parameters together with user related factors such as cognitive, psychological, behavioural and context in which VoIP services are consumed and created [6]. In this context, it is vital to consider users related factors, their devices, environment and network parameters. A scenario may arise whereby a user is on an important mobile VoIP video call session in which a battery life is not long enough to complete the session, moreover, the environment does not have facilities to charge the battery. If this situation is left without any action, then the battery life will greatly jeopardize the QoE. In this scenario, power conservation techniques have an important role to play. Since CPU, LCD, GPS, Wi-Fi/3G and AV applications are well known to be the major power consumers, any effort to reduce power consumption of these mobile components will prove vital. In this scenario only GPS is not needed and is therefore switched off, LCD must not be switched off because of the importance of video display. Popular power saving techniques for Wi-Fi/3G such as sleep scheduling cannot be applied in this scenario because of the real time communication nature. Thus, in this scenario, AV applications such video codecs are the main components to be adapted to conserve power. This approach leads to a powerdriven VoIP quality adaptation scheme that aims to conserve power while keeping VoIP quality at an acceptable level. As a proof of concept and to demonstrate the proposed scheme, an IMS based VoIP testbed is developed where two Android Developer 1 mobile handsets are used as IMS clients for VoIP calls registration, session initiation and termination over Wi-Fi access network. AMR and H264 codecs are used for voice and video communication, respectively. The use of AMR and H264 codecs enables VoIP quality adaptation by changing audio and video send bitrates (SBR). Preliminary results have shown that the proposed scheme considerably extends the battery life and hence prolongs the communication while maintaining an acceptable VoIP quality. The rest of this paper is organized as follows. Section II outlines the most relevant related work. Section III presents the testbed and the experimental setup. Evaluation of power consumption of relevant mobile components is carried out in Section IV. The proposed scheme is outlined in Section V. Experimental results and evaluations are described in Section 978-1-4673-4941-3/12/$31. 212 IEEE 1276
VI. Future work and conclusions follows in Section VII. II. RELATED WORK There exists a few power-driven VoIP quality adaptation schemes in the literature, the most recent and relevant one has been proposed by Csernai and Gulyas in [8]. A framework that reduces the power consumption of wireless mobile devices during streaming video content over Wireless LAN networks was proposed. The proposed framework optimized the overall power efficiency by controlling the sleep cycles of wireless network adapters based on video QoE. The framework used various SBR to adjust sleep intervals of the wireless adapter in order to maintain video quality while maximizing power efficiency. The same approach was also investigated in [9] where the Power Save Mode (PSM) introduced in IEEE 82.11 standard [1] was used. Although the framework was claimed to save 2-3% of total system power but did not answer some key concerns of practical interests in the real world implementation of VoIP communication. A major concern in this framework is the accuracy of sleep cycles in the presence of important VoIP signalling traffic such as SIP, which is crucial for maintaining the state of the real time VoIP communication session without any delays. Global adaptation framework for quality of experience of mobile services was proposed in [11], the framework aimed at efficiently detecting the changes of a mobile system status and carrying adaptations in a global manner in order to enhance QoE. Although the framework emphasized on monitoring battery level but it was not specific on what adaptation actions to take if the battery level was low. Furthermore, the framework was not implemented for analysis and evaluation. As argued in the case of the framework proposed in [8], under the presence of VoIP signalling traffic the sleep cycles technique is not realistic. The ideal technique is to reduce power consumption of the AV application while keeping the VoIP quality at an acceptable QoE level. Therefore, this paper proposes power-driven VoIP quality adaptation scheme over wireless mobile devices in order to extend battery life and hence prolong VoIP communication session at an acceptable QoE level. The scheme periodically monitors the battery power and takes actions on what SBR to use in order to reduce power consumption of an AV application. The framework proposed in [8] showed that the higher the bitrates the more the power is consumed. This is because the power consumption of H.264/AVC depends on the quantization and entropy encoder parts which include, the computational complexity due to the number of pixels processed per unit time and coding level based computational complexity which results into several SBR [12]. III. EXPERIMENTAL TESTBED Figure 1 depicts the testbed developed to evaluate the proposed scheme. The testbed is based on an Open IMS Core for RTP session establishment and termination using Session Initiation Protocol (SIP) protocol. Fig. 1. Edge router WLAN Power-driven VoIP quality adaptation testbed Two Android Developer Phone 1 (ADP1) mobile phones ported with IMSDroid [13] were used as IMS clients connected via an open source Netgear WGR614v8 Wi-Fi router. The VoIP voice and video call with AMR and H264 codecs were used (IMSDroid supports AMR-NB and H264 base profile 1, 2 and 3). Android power monitoring tool (Powertutor) [14] was deployed into mobile phones to periodically monitor power levels of relevant mobile components, the data collection was done at an interval of one second. The mobile phone was stationery, situated at about 2m from the access point. OpenSips [15] is used as an open source presence server and OpenXCAP [16] as an open source implementation of the XDM specification, for centralised storage of monitored battery life and power level consumption of relevant mobile components, the HTTP was used to upload these data at an interval of 5 seconds. IV. POWER CONSUMPTION EVALUATION This section evaluates power consumption of relevant mobile components, the evaluated components have shown to significantly contribute to overall system power consumption. In this experiment, all data was collected and evaluated at one end (the callee phone) of the two mobile phones. A. Wi-Fi Interface Power Usage Figure 2 illustrates the power consumption of the Wi-Fi interface. The evaluation was done by recording the Wi-Fi interface power usage in the following steps, Step 1: The Wi-Fi interface was off and as the result there was no power consumption recorded. Step 2: The Wi-Fi interface was enabled with no data transmission, this depicts that an average of 38 mw was used by the Wi-Fi interface just for listening, this step is known as the low-power state. 1277
8 7 Wifi Power Usage ------------------------------Step 3---------------------------- 6 5 -------------Step 4-------------- AV Power usage -----Step 5----- 6 5 4 4 3 Step 2 ------------Step 3-------------- 3 2 2 1 -----Step 2------- --Step 1-- 1 2 3 4 5 6 7 8 9 1 1 -----Step 1---- 1 2 3 4 5 6 7 8 Fig. 2. Wi-Fi power consumption Fig. 3. AV application power consumption Step 3: The VoIP communication was on, this phase is known as high-power state which according to [14], occurs when there is uplink and/or downlink data transmission. The average power consumption at the high-power state recorded an average of 725 mw. Step 3 validates the Wi- Fi power consumption model outlined in [14] (c.f., (1) and (2)), which depends on the number of packets transmitted and received per second, uplink channel rate and uplink data rate. The uplink channel rate was at 54Mbps. The Wi-fi interface manufacturer is the Texas Instruments WL 1251B chipset and the CPU is MSM721A chipset, including ARM11 application processor, ARM9 modem, and high-performance DSP. and, β wifi = 71mW + β cr (R chnl ) R data (1) β cr (R chnl ) = 48.768 R chnl (2) where, β wifi, R chnl, R data and β cr are high-power state consumption of the Wi-Fi interface, Wi-Fi uplink channel rate, Wi-Fi uplink data rate and Wi-Fi power coefficient, respectively. B. AV Application Power Usage Figure 3 illustrates the power consumption of the AV application (IMSDroid). The evaluation followed the following steps Step 1: The AV application was off and therefore no power consumption was reported. Step 2: The AV application was launched whereby the SIP and RTP stacks were readied and VoIP registration process was done, an average of 23 mw power consumption was recorded. This step involved SIP signalling traffic exchange between the mobile phone and IMS entities, presence and xcap servers. Step 3: The voice only VoIP communication was then established, this involved SIP signalling traffic for session negotiation which was followed by RTP voice media communication, this phase reported an average of 16 mw power consumption. Step 4: The receive only video transmission, at this step the power consumption jumped to an average of 37 mw. Step 5: The send video button was pressed and the power consumption increased to an average of 47 mw. C. LCD Power Usage The power consumption of the LCD (TFT-LCD at glass touch-sensitive HVGA screen) remained constant at 533 mw as it was always on for the VoIP video communication display. The brightness level was 12. D. Battery Discharge Rate Figure 4 shows the battery state of discharge during the VoIP communication. It shows that the rate of battery discharge was higher at the transition from the low-power to high-power state. During the high-power state, the rate of battery discharge was at the steady state. Voltage (mv) 4.1e+6 4.5e+6 4e+6 3.95e+6 3.9e+6 3.85e+6 3.8e+6 3.75e+6 1 2 3 4 5 6 7 8 Fig. 4. E. Total Power Consumption Battery state of discharge Battery Discharge In Figure 5, the total power of the mobile phone is shown which includes all mobile components at each step, Step 1: The Wi-fi interface is off Step 2: The Wi-fi interface is on and on listening mode Step 3: The AV application is launched Step 4: The AV application is registered to the IMS 1278
18 16 14 12 Total power usage -----------Step 7---------- ---Step 8--- Step 3 --Step 6--- Step 4 Step 5 Step 5: Video communication where an average of 75 Bytes were recorded for both downlink and uplink. The downlink video communication was started before the uplink. 1 8 1 Uplink Bytes 6 8 4 ---Step 1-- --------Step 2------------ 2 1 2 3 4 5 6 7 8 9 1 Packets (Bytes) 6 4 --------------Step 5--------- Fig. 5. Total power consumption of the mobile phone 2 -Step 3- -Step 2- ---Step 1--- -------------------------Step 4---------------------- Step 5: VoIP session establishment is made Step 6: RTP audio traffic communication Step 7: Incoming video communication and Step 8: Outgoing video communication F. Uplink and Downlink Bytes The downlink and uplink Bytes are demonstrated in Figures 6 and 7, respectively, which clearly illustrate the traffic behaviour at each step during the VoIP communication session. Packets (Bytes) 1 8 6 4 2 ---Step 1--- -Step 2- -Step 3- --------------Step 4-------------- 1 2 3 4 5 6 7 8 9 Fig. 6. Downlink Bytes Downlink Bytes --------------Step 5---------------- Figures 6 and 7, involved the following steps, Step 1: The Wi-fi interface was off and therefore, no downlink/uplink Bytes were recorded. Step 2: The AV application was launched and the IMS registration performed. This steps recorded an average of 11 and 9 Bytes for downlink and uplink, respectively. This is due to the SIP registration and presence signalling. The downlink had more activities of downloading XML files for power monitoring at the time of registration. Step 3: VoIP session establishment is established. This step reported an average of 14 Bytes due to SIP INVITE message flows. Step 4: RTP audio traffic communication where an average of 45 Bytes were observed for both downlink and uplink. 1 2 3 4 5 6 7 8 9 1 Fig. 7. Uplink Bytes V. PROPOSED POWER-DRIVEN VOIP QUALITY ADAPTATION SCHEME By knowing power consumption of each component during VoIP communication and the state of battery discharge, it is important to identify which component to adapt in order to conserve power while maintaining acceptable VoIP QoE. It is clearly observed in Figure 2 that Wi-Fi interface is the highest consumer of the overall power. One would argue that since its power consumption model is dependent on the transmitted and received data rate, adapting video and audio send bitrates could significantly reduce Wi-Fi interface power consumption, and as the result, battery life will be extended. But this is not the case for the Wi-Fi interface. As described in Section IV and proved in [14], there are two states of Wi- Fi interface power consumption, low-power state when there is no data being transmitted and/or received, and high-power state when there is data being transmitted and/or received. The two state of Wi-Fi power consumption are dependent on the number of packets transmitted/received per second. Authors in [14] proved that the high-power state is entered when number or transmitted/received packets per second is at least 8 packets/second. The high-power state is noticed at the start of voice communication where the transmitted/received packets are at an average of 1 packets per second. This is because interdeparture and inter-arrival is at 2 ms per packet of the AMR Codec. The number of packets transmitted/received per second increased when the video communication was commenced, but this did not change the Wi-Fi power consumption. Figure 8 depicts the dependency of Wi-Fi power consumption states to the number of packets transmitted/received per second. In this context, it is suggested to reduce power consumption of the AV application because codecs power consumption is dependent on the transmitted and/or received data rates as seen in Figure 3. Adapting SBR should be carried out while maintaining acceptable QoE. 1279
8 7 6 5 4 3 2 1-1 1 2 3 4 5 6 7 8 9 1 Fig. 8. Number of packets transmitted/received Wi-Fi power usage (mw) Transmitted/recevied packets per second As it can be seen in Figure 3, there is a significant difference in power consumption between audio and video codecs and hence the proposed scheme will only concentrate on adapting video SBR which also have wider range of variations than the former. From the empirical results, adapting audio SBR of AMR from 12.2 Kbps to 4.75 Kbps did not significantly save the power. The proposed scheme makes use of the model proposed in [17] (c.f., Equation (3)) where the MOS value of at least 3.5 was proved to be an acceptable video QoE for slight movement video content type such as in VoIP video call. The proposed range of SBR between 1-5 Kbps provided maximum QoE of 4.2 at 15 frames per second when using QCIF video format. The SBR of 5 Kbps provided a good QoE of at least 3.5 MOS value. QCIF was specifically used in [17] because it was the recommended size for small handheld terminals which were the target applications of the research. However, the proposed scheme is not limited to QCIF and can be extended to higher resolutions. MOS = a 1 + a 2 F R + a 3 ln(sbr) 1 + a 4 P ER + a 5 (P ER) 2 (3) where, FR denotes the video frame rate, SBR presents the video send bitrates and PER is the packet error rates. The values of coefficients a i, i = 1... 5 can be found in [17]. In this experiment P ER =. Table I defines the battery charge levels (BCL) with the corresponding SBR values for VoIP quality adaptation. TABLE I MAPPING OF BCL TO SBR Level BCL (%) SBR (Kbps) MOS 1-75 5 4.2 1 75-5 3 4.2 2 5-4 1 3.5 2 4-15 5 3.5 3 15- only voice capacity and the current battery charge level at the time of registration and then at an interval of 5 seconds provided mobile phones are still registered. 5 seconds is used as a standard from the well known RTCP communication. Step 2: Mobile phones will use the presence server and XDM server to retrieve power capabilities at the time of the VoIP session establishment. Step 3: At the beginning of the VoIP session establishment mobile phones will choose the right video SBR which corresponds to the acceptable QoE and battery charge level. Step 4: Mobile phones will monitor power capabilities through the published data in the XDM server at an interval of 5 seconds. Step 5: From the battery charge level, mobile phones will compute the state of the battery discharge and calculate the remaining time to reach 15% of the maximum battery capacity. The 15% is the threshold set for poor battery charge level. Step 6: The remaining time to reach 15% of the maximum battery charge capacity will determine the appropriate video SBR adaptation actions to be carried out while maintaining QoE at an acceptable level. Table I will be used to determine which SBR to use in the adaptation. Step 7: If the poor battery charge level is reached then the video transmission and the LCD will be switched off. Only the voice communication will be left to continue. Algorithm 1 Power-driven VoIP Quality Adaptation Scheme MBCC = Max Batt Charge Capacity() BCL = Batt Charge Level() while Phone is Still Registered to the IMS do for Time Interval of 5 Seconds do PUT Max Batt Charge Cap(MBCC,XDMS) PUT Batt Charge Level(BCL,XDMS) while VoIP Session is Ongoing do BCL = GET Batt Charge Level(Callee,XDMS) MBCC = GET Max Batt Charge Cap(Callee,XDMS) Comput Batt DisCharge Rate(Callee,BCL) Comput Remaining Time(BCL,MBCC,.15) Adapt SBR If Needed() if BCL Threshold is reached then Switch off video transmission Switch off LCD end if end while end for end while The proposed scheme involves the following steps for which the algorithm is presented in Algorithm 1, Step 1: Mobile phones upload their power capabilities to the XDM server, i.e., maximum battery charge VI. EXPERIMENTAL RESULTS It can be seen that there is significant saving of power consumption when changing video SBR from 5Kbps to 5Kbps 128
(c.f., Figure 9). A simple non-linear regression analysis is derived which gives a power consumption prediction model for the range of 5-5 Kbps SBR at 15 FR with the goodness of fit of.99 (c.f, Equation (4) ), P ower = 95.211ln(SBR) + 311.84 (4) where SBR(5Kbps SBR 5Kbps), is the video send bitrates of H264 codec. 48 46 44 42 4 38 36 34 32 3 5 1 15 2 25 3 35 4 45 5 Fig. 9. SBR (Kbps) Power usage at different SBR Power consumption at different SBR The IMSDroid AV application shows noticeable power conservation (c.f., Figure 1) when changing from 5 Kbps to 5 Kbps. It can be seen from Equation (4) that the saving of power is at the range of 1-3% of the total system power. 6 5 4 3 2 1 1 2 3 4 5 6 7 8 Fig. 1. 5Kbps power usage 5Kbps power usage Power saving consumption at different SBR VII. CONCLUSIONS AND FUTURE WORK This paper has proposed and evaluated the power-aware VoIP quality adaptation scheme over WLAN in mobile devices. The scheme uses video SBR for adaptation in order to conserve power while maintaining acceptable video QoE. Through preliminary results, the evaluation has shown that the proposed scheme can save power consumption between 1-3% of the total power consumption in Android Developer Phone 1. The scheme can be extended to include several other video codecs, mobile devices and VoIP applications over various access networks such as UMTS and LTE. Future work will also involve intelligent SBR adaptation scheme by using well known techniques such as Fuzzy logic. Subjective quality assessment with network parameters such as packet loss rate and delay will also be in the future work to validate the proposed scheme. The impact of mobility on the QoE will also be considered. REFERENCES [1] W. Yuan, K. Nahrstedt, S. V. Adve, D. L. Jones, and R. H. Kravets, Grace-1: Cross-layer adaptation for multimedia quality and battery energy, IEEE Transactions on Mobile Computing, vol. 5, pp. 799 815, 26. [2] J. Flinn and M. Satyanarayanan, Energy-aware adaptation for mobile applications, in Proceedings of the seventeenth ACM symposium on Operating systems principles, ser. SOSP 99. New York, NY, USA: ACM, 1999, pp. 48 63. [Online]. Available: http://doi.acm.org/1.1145/319151.319155 [3] N. Vallina-Rodriguez and J. Crowcroft, Energy management techniques in modern mobile handsets, Communications Surveys Tutorials, IEEE, vol. PP, no. 99, pp. 1 2, 212. [4] ITU-T SG12, Methods for Subjective Determination of Transmission Quality, ITU-T SG12, Geneva, Switzerland, Recommendation P.8, Aug. 1996. [Online]. Available: http://www.itu.int/rec/t-rec-p.8-19968-i/en [5] ITU-T E-Model, A computational model for use in transmission planning, ITU-T Rec. G.17, Int. Telecommun. Union, 2. [6] K. Mitra, C. Ahlund, and A. Zaslavsky, A decision-theoretic approach for quality-of-experience measurement and prediction, Multimedia and Expo, IEEE International Conference on, vol., pp. 1 4, 211. [7] Kilkki, K., Quality of experience in communications ecosystem, Journal of Universal Computer Science, pp. 615 624, 28. [8] M. Csernai and A. Gulyas, Wireless adapter sleep scheduling based on video qoe: How to improve battery life when watching streaming video? in Computer Communications and Networks (ICCCN), 211 Proceedings of 2th International Conference on, 31 211-aug. 4 211, pp. 1 6. [9] V. Namboodiri and L. Gao, Energy-efficient voip over wireless lans, Mobile Computing, IEEE Transactions on, vol. 9, no. 4, pp. 566 581, april 21. [1] IEEE 82.11, IEEE Std 82.ll-1999, IEEE Computer Soc. LAN MAN Standards Committee, Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, 1999. [11] H. Sun, V. De, and G. Florio, Global adaptation framework for quality of experience of mobile services, In Proc. of the 27 IEEE Three- Rivers Workshop on Soft Computing in Industrial Applications, Passau, Germany, 27. [12] A. Ukhanova, E. Belyaev, and S. Forchhammer, Encoder power consumption comparison of distributed video codec and h.264/avc in lowcomplexity mode, in Software, Telecommunications and Computer Networks (SoftCOM), 21 International Conference on, sept. 21, pp. 66 7. [13] D. Telecom, Sip/ims client for android, Website, 211, hhttp://code.google.com/p/imsdroid/. [14] L. Zhang, B. Tiwana, R. Dick, Z. Qian, Z. Mao, Z. Wang, and L. Yang, Accurate online power estimation and automatic battery behavior based power model generation for smartphones, in Hardware/Software Codesign and System Synthesis (CODES+ISSS), 21 IEEE/ACM/IFIP International Conference on, oct. 21, pp. 15 114. [15] V. System, Open sip server, Website, 29, http://www.opensips.org/. [16] M. Amarascu, R. Klaver, L. Stanescu, D. Bilenko, and S. Ibarra, Open xcap server, Website, 26, http://www.openxcap.org/. [17] A. Khan, L. Sun, E. Jammeh, and E. Ifeachor, Quality of experiencedriven adaptation scheme for video applications over wireless networks, Communications, IET, vol. 4, no. 11, pp. 1337 1347, 21. 1281