QoE Enhancement of Audio Video IP Tansmission with IEEE 8.e in Mobile Ad Hoc Netwoks Toshio Nunome and Shuji Tasaka Depatment of Compute Science and Engineeing, Gaduate School of Engineeing, Nagoya Institute of Technology, Nagoya 8, Japan E mail: {nunome, tasaka}@nitech.ac.jp Abstact This pape compaes QoE (Quality of Expeience) of audio video IP tansmission with IEEE 8.e and that with in mobile ad hoc netwoks. We also assess QoE of the tansmission with busting, which is an optional function of. QoE assessment of audio video IP tansmission in mobile ad hoc netwoks involves difficult issues. The wieless nodes can move, and then channel condition changes dynamically. The outing paths may also change as the wieless nodes move. In this pape, we tackle the difficulty of the assessment by means of a continuous time QoE assessment method, which utilizes the method of successive categoies continuously in time. Fo the assessment, at fist, we pefomed compute simulation, whee a mesh topology netwok with mobile nodes is assumed fo tansmitting audio, video, and data taffic. We then caied out subjective expeiment. The assessment esults show that and enhance QoE of audio video tansmission ove mobile ad hoc netwoks. I. INTRODUCTION New geneation wieless netwoks ae expected to povide a wide vaiety of sevices with high peceived QoS (Quality of Sevice) fo uses. The peceived QoS is efeed to as Quality of Expeience (QoE) in ITU T []. It is the most impotant among QoS s at all levels since the uses ae the ultimate ecipients of the sevices. Audio video IP tansmission foms an indispensable basis of netwoked multimedia applications even in mobile ad hoc netwoks (MANETs). Enhancement of the output quality of audio video steams is one of the most impotant issues. The IEEE 8.e (Enhanced Distibuted Channel Access) is an enhanced vesion of the (Distibuted Coodination Function) and can suppot elative pioity sevices fo multimedia tansmission []. Audio, video, and data taffic have diffeent equiement fo QoS fom each othe. QoS diffeentiation plays a key ole in poviding QoE cost efficiently in a wieless mobile envionment. A vaiety of studies on QoS contol with the IEEE 8.e in ad hoc netwoks have been epoted [] []. Refeences [] and [] popose coss laye design achitectues with the netwok and MAC layes. Lee et al. employ coss laye design with the physical and MAC layes []. Dynamic pioity assignment schemes ae poposed in [], [7] and [8]. Zhu et al. popose a scheme to contol contention window (CW) size dynamically accoding to the taffic condition [9]. Calafate et al. employ a pobe based admission contol []. Most of the studies discussed above assess only end to end level and lowe level QoS fo thei pefomance evaluation. As mentioned ealie, QoE is the most impotant QoS fo the uses. Howeve, none of the studies assess QoE. QoE of audio video tansmission ove infastuctue netwoks with the IEEE 8.e is assessed in []. The stuctue and taffic chaacteistics of the infastuctue netwoks ae lagely diffeent fom those of the ad hoc netwoks. Thus, the effect of in the ad hoc netwoks on QoE can diffe fom that in the infastuctue netwoks. In [] and [], the authos have also assessed QoE of audio video tansmission ove ad hoc netwoks without mobility. In ode to achieve high QoE, the authos have employed link quality based outing, which takes account of link quality such as signal stength and link utilization level fo oute selection. Howeve, these efeences do not conside mobile nodes, which is one of the most impotant chaacteistics in mobile ad hoc netwoks. In addition, it does not assume sevice diffeentiation in the MAC laye; the audio, video, and data taffic ae tansmitted with the same pioity. In this pape, we assess the QoE of audio video IP tansmission with the IEEE 8.e ove mobile ad hoc netwoks. As in [], we employ a continuous time QoE assessment method, which is poposed in [] and utilizes the method of successive categoies continuously in time. In this study, as a fist step, we assume gid topology netwoks with mobile nodes. It is a simple example of mobile ad hoc netwoks. The est of this pape is oganized as follows. Section II summaizes the IEEE 8.e. Section III illustates a methodology of the simulation, and Section IV descibes the assessment method of QoE. The assessment esults ae pesented and discussed in Section V. II. IEEE 8.E The IEEE 8.e intoduces fou ACs (Access Categoies) to suppot diffeentiated channel access fo applications with QoS equiements []. A station which suppots the has an individual souce buffe fo each of the fou ACs, and the channel access function based on the CSMA/CA is independently caied out pe AC. In addition, the IEEE 8.e MAC defines a (Tansmission Oppotunity) as an inteval of time when a specified station has ight to initiate sequences of fame exchange onto the wieless medium. In the, a station can tansmit multiple MAC fames duing a fom the same souce buffe (AC) until the time eaches the T XOP Limit[AC] []. We efe to this scheme as the busting in this pape.
MS 7 8 move BTR BTS 9 7 8 MR MR Fig.. MS Inte-node distance Netwok topology. MS: Media Sende MR: Media Receive BTS: Backgound Taffic Sende BTR: Backgound Taffic Receive Duing a, successive sequences of fame exchange ae sepaated by SIFS (Shot InteFame Space). When a ends, the station goes into backoff. The busting can educe ovehead due to contention if a station has moe than one MAC fame in a souce buffe. III. METHODOLOGY OF SIMULATION At fist, we pefomed compute simulation with ns (netwok simulato vesion ) []. The netwok consists of 8 nodes as shown in Fig.. We employed OLSR (Optimized Link State Routing) as the outing potocol; it is a poactive outing potocol. The outing potocol is one of the most impotant elements to constuct ad hoc netwoks. It can affect the QoE and QoS lagely. In this pape, as a fist step towad the QoE assessment in mobile ad hoc netwoks, we have employed a pimitive ad hoc outing potocol. Moe sophisticated outing potocols discussed in [] and [] can be beneficial to futhe enhancement of QoE; this is a futue study. The numbe in a cicle in Fig. means the node numbe. Nodes and ae Media Sendes (MSs), and Nodes 7 and 8 ae Media Receives (MRs). Nodes and 9 ae Backgound Taffic Sende (BTS) and Backgound Taffic Receive (BTR), espectively. Except fo Nodes 7 and 8, the distance between two vetical o hoizontal adjacent nodes is constant; it is efeed to as the inte node distance. We employed thee values of : m, m, and m. Nodes 7 and 8 move along the dotted ectangle depicted in Fig.. In ode to assess the influence of the node mobility, the movement speed of each node was set to / m/s. In the simulation, a node can eceive a MAC fame when the distance between the node and the sende is smalle than 7 m. We utilized the feespace model of ns as the popagation model. In ns, the signal stength of a MAC fame can be calculated by the popagation model and distance between the tansmitte and the eceive. In calculating SNR, we assumed eceive noise stength based on Oinoco 8.b cad. We also used empiical cuves of BER vesus SNR povided by Intesil wieless LAN chipset to obtain BER []. In Fig., we assume Nodes and as the audio and video souces. Nodes and tansmitted the media steams to Nodes 7 and 8, espectively, with UDP. We efe to the tansmission TABLE I SPECIFICATIONS OF THE AUDIO AND VIDEO. audio video coding method ITU T G.7 MPEG µ law GOP I image size [pixel] aveage MU size [byte] aveage MU ate [MU/s] aveage MU inteval [ms] aveage bit ate [kb/s] duation [s] TABLE II PARAMETERS OF THE. AC CW min CW max AIFS Limit [µs] [ms] audio 7. video. data 7 unit at the application level as a Media Unit (MU); we define a video fame as a video MU and a constant numbe of audio samples as an audio MU. We used an audio steam of ITU T G.7 µ law and an MPEG video steam, which has been pepaed by encoding a pat of a Japanese music clip. Table I shows the specifications of the audio and video. Nodes and 9 ae used to handle an independent intefeence taffic flow fo the media steams. Node geneated fixed size IP datagams of bytes each at exponentially distibuted intevals and then sent to Node 9. We set the aveage amount of the intefeence taffic to kb/s in the simulation. We assessed the QoE of,, and. is the oiginal scheme in the IEEE 8. standad. means the IEEE 8.e without busting. is the scheme which combines the IEEE 8.e and busting. Table II shows paamete values of the in the simulation. These ae default paamete values in []. In this table, the value of Limit fo data is ; this means that busting is not employed fo the data tansmission. In addition, we have assigned the same pioity to the contol infomation packets fo the outing potocol as that fo the audio tansmission. IV. METHODOLOGY OF QOE ASSESSMENT In this pape, we assess QoE of the output audio video steam by a subjective expeiment. It was conducted as follows. Fo subjective assessment, we made stimuli, which ae objects to be evaluated, by actually outputting the audio and video MUs with the output timing obtained fom the simulation. Each stimulus lasts seconds. We put the stimuli in a andom ode and pesented them to assessos, using a laptop PC with headphones. The assessos ae male and female in thei twenties. A subjective scoe is measued by the ating scale method, in which assessos classify each stimulus into one of a cetain
numbe of categoies. We adopt the following five categoies of impaiment: impeceptible assigned intege, peceptible, but not annoying,, annoying, and vey annoying. The intege value is egaded as the subjective scoe. In audio video steaming in mobile ad hoc netwoks, its quality can fluctuate quite lagely. In the ating scale method, each assesso is supposed to give a subjective scoe fo a stimulus. Howeve, it is difficult fo the assessos to give the aveage of the peceived quality at the end of each stimulus because of the tempoal fluctuation. Thus, we asked the assessos to give a scoe fo each fagment of a stimulus as stated below. While a stimulus is pesented to each assesso, he/she classifies evey instantaneous quality into one of the five categoies of impaiment accoding to his/he subjective assessment. The assesso inputs a scoe by the laptop PC s keyboad wheneve his/he classification changes fom a scoe that had been input immediately befoe. The input scoe is kept until the assesso changes it to anothe; it is sampled evey second. The sampled value is assumed as the subjective scoe fo the fagment fo the second. In this pape, we utilize the method of successive categoies [7]. In ode to obtain an inteval scale as the QoE paamete, we fist measue the fequency of each categoy with which the fagment of the stimulus was placed in the categoy by the ating scale method. With the law of categoical judgment, we can tanslate the fequency obtained by the ating scale method into an inteval scale. We then pefom Mostelle s test, which tests the goodness of fit between the obtained inteval scale and the measuement esult. The inteval scale of which we have confimed the goodness of fit is efeed to as the psychological scale [7]. The assessos assessed stimuli fo the thee schemes:,, and. Fo each scheme, thee wee thee stimuli, which coespond to the inte node distances of m, m, and m. It took about minutes fo an assesso to finish all assessment which includes the pesentation of the oiginal audio video steam, a stimulus fo pactice, and =9stimuli. V. ASSESSMENT RESULTS This pape discusses the quality of the audio video steam eceived at Node 8. At fist, Figs. though depict the aveage numbe of hops fom Node to Node 8 as a function of elapsed time. We find in these figues that the numbe of hops with has almost the same tendency as that with. On the othe hand, has a little diffeence fom the two schemes. This is because does not change the oute so quickly as o owing to the queuing delay fo contol packets of the outing potocol; in and, high pioity is given to the contol packets fo channel access. A. Application Level QoS In this study, we also assess the application level QoS, which is closely elated to the QoE. Although we have employed the thee values of in the assessment, we only show the assessment esults with = m because of space limitations. Aveage numbe of hops Aveage numbe of hops Aveage numbe of hops Inte-node distance = m Teminal 8 Fig.. Aveage numbe of hops (Inte node distance = m). Inte-node distance = m Teminal 8 Fig.. Aveage numbe of hops (Inte node distance = m). Inte-node distance = m Teminal 8 Fig.. Aveage numbe of hops (Inte node distance = m). The coefficient of vaiation of MU output inteval fo audio is shown in Fig.. It is defined as the atio of the standad deviation of the MU output inteval (i.e., the pesentation time inteval of two MUs at the destination) of a steam to its aveage and epesents the smoothness of the output steam. In this figue, blanks of lines (e.g., aound time ) mean pause
Coefficient of vaiation of output inteval fo audio...... Inte-node distance = m Node 8 9 % confidence inteval Psychological scale.. impeceptible Inte-node distance = m peceptible, but not annoying annoying vey annoying 7 8 9 Fig.. Coefficient of vaiation of MU output inteval fo audio (Inte node distance = m). Fig. 7. Psychological scale (Inte node distance = m). MU loss atio of video.8... Inte-node distance = m Node 8 9 % confidence inteval Fig.. MU loss atio of video (Inte node distance = m). Psychological scale. impeceptible Inte-node distance = m peceptible, but not annoying. annoying vey annoying 7 8 9 Fig. 8. Psychological scale (Inte node distance = m). peiods due to oute disconnections. We find in Fig. that has the smallest coefficient of vaiation of MU output inteval fo audio among the thee schemes. This is because the audio steam has the highest pioity than the othe steams in. Figue epesents the MU loss atio of video vesus elapsed time. It is the atio of the numbe of MUs lost to the total numbe of MUs tansmitted fom the sende teminal. In this figue, we can obseve that has smalle MU loss atio of video than the othe two schemes especially in the peiod of aound time 9 to time 9. This is because and assign the smalle CW values fo the video AC than. When the netwok condition is not good, the schemes attempt much etansmission in a shot peiod and then can each the limit of etansmission easily. B. QoE We calculated the QoE paamete fo all the inte node distances employed in the simulation. We pocessed the esult in the peiod of time though. We then pefomed Mostelle s test, which tests the goodness of fit between the obtained inteval scale and the measuement esult. As a esult of the test, we found that the null hypothesis that the obtained inteval scale fits the obseved data can be ejected at significance level. By emoving about % of the fagments, we saw that the hypothesis cannot be ejected; we conside the inteval scale as the psychological scale, i.e., QoE paamete. Note that we can select any oigin in the inteval scale. In this pape, fo convenience, we egad the minimum value of the psychological scale as the oigin. Unde this condition, we also calculated the lowe boundaies of the categoies and got.9 fo Categoy, 8 fo Categoy,. fo Categoy, and. fo Categoy. Figues 7, 8, and 9 plot the psychological scale vesus the elapsed time fo the inte node distance m, m, and m, espectively. In these figues, hoizontal dotted lines show boundaies between the categoies. In Figs. 7 though 9, we notice that the values of the psychological scale ae stongly affected by the oute changes; the degadation of QoE occus when the numbe of hops changes as we can see in Figs. though. The outing potocol is one of the dominant factos fo audio video quality in ad hoc netwoks. We find in Figs. 7 though 9 that and have highe values of the psychological scale than in almost all the duation. Thus, and ae also effective
Psychological scale... impeceptible Inte-node distance = m peceptible, but not annoying annoying vey annoying 7 8 9 Fig. 9. Psychological scale (Inte node distance = m). in audio video tansmission ove mobile ad hoc netwoks as well as infastuctue wieless LANs []. We can obseve in Figs. 7 though 9 that tends to have lowe QoE than. This is because we employ an audio dominant content in the expeiment, and busting does not wok effectively fo the audio steam. In Figs. 7 and 9, we see that the ecovey of the psychological scale value afte the oute change with delays against that with o. This is due to the queuing delay fo contol packets. We notice in Figs. 7 though 9 that the values of the psychological scale fo the inte node distance m fo the duation fom time to ae the best among the thee inte node distances. This is because the oute fo the inte node distance m fo this duation does not include long links, which ae included in the oute fo the inte node distance m o m; the long links suffe fom the channel eo lagely. Fom the above obsevation, as fo the application level QoS, has good audio quality owing to the pioitized tansmission, and has good video quality because of its lage CW size. Howeve, fom the QoE assessment esults, peceived quality of audio video with is not bette than that with eithe of the othe two schemes. One of the easons is that we have employed the music clip as the audio and video souces, and then the audio affects subjective quality moe lagely than the video. We should assess the QoE with vaious audio video contents; this is fo futue study. VI. CONCLUSION This pape assessed QoE and application level QoS of audio video tansmission ove mobile ad hoc netwoks with IEEE 8.e. As a esult, we found that and busting impove the QoE of audio video steams even in mobile ad hoc netwoks. These schemes can achieve bette application level QoS fo audio than and can ecove oute disconnections faste. As futue wok, we have to investigate the effect of coss layeing between the outing potocols and on QoE because the outing potocol affects QoE lagely. We also need to assess the effect of paamete values fo on QoE. ACKNOWLEDGMENT The authos thank Tsuyoshi Saito fo his assistance in the simulation. This wok was suppoted by the Gant In Aid fo Scientific Reseach of Japan Society fo the Pomotion of Science unde Gant 8. REFERENCES [] Rec. ITU T G./P. Amendment, New Appendix I Definition of Quality of Expeience (QoE), Jan. 7. [] Amendment to Standad 8.: Medium access contol (MAC) enhancements fo quality of sevice (QoS), IEEE 8.e Std., Sept,. [] V. C. Fias, G. D. Delgado and M. A. Igatua, Multipath outing fo video steaming sevices ove IEEE 8.e ad hoc netwoks, Poc. SoftCOM, Sept./Oct.. [] L. Romdhani and C. Bonnet, A coss laye on demand outing potocol fo delay sensitive applications, Poc. IEEE PIMRC, Sept.. [] T. H. Lee, A. Mashall and B. Zhou, A QoS based ate adaptation stategy fo IEEE a/b/g PHY schemes using IEEE 8.e in ad hoc netwoks, Poc. ICNS, July. [] A. Iea, A. Molinao, G. Ruggei and D. Tipodi, Impoving QoS and thoughput in single and multihop WLANs though dynamic taffic pioitization, IEEE Netwok, vol. 9, no., pp., July/Aug.. [7] A. Iea, A. Molinao, S. Polito and G. Ruggi, Coodinated multihop scheduling in IEEE 8.e wieless ad hoc netwoks, Poc. IEEE PIMRC, Sept.. [8] T. B. Reddy, J. P. John and C. S. R. Muthy, Poviding MAC QoS fo multimedia taffic in 8.e based multi hop ad hoc wieless netwoks, Compute Netwoks, (7) 7. [9] H. Zhu, G. Cao, A. Yene and A. D. Mathias, E DM: A novel enhanced distibuted coodination function fo wieless ad hoc netwoks, Conf. Rec. IEEE ICC, June. [] C. T. Calafate, J. C. Cano, P. Manzoni and M. P. Malumbes, A QoS achitectue fo MANETs suppoting eal time pee to pee multimedia applications, Poc. IEEE ISM, Dec.. [] T. Suzuki, A. Noguchi and S.Tasaka, Effect of busting and tansmission eo on application level and use level QoS in audio video tansmission with IEEE 8.e, Poc. IEEE PIMRC, Sept.. [] T. Nunome, S. Tasaka and K. Nakaoka, Application level and use level QoS assessment of audio video IP tansmission ove coss laye designed wieless ad hoc netwoks, IEICE Tans. Commun., vol. E9 B, no., pp., Oct. 8. [] T. Nunome and S. Tasaka, Use level QoS assessment of audio video IP tansmission ove link quality based ad hoc netwoks, Poc. Fist Ambient Netwoks Wokshop on Mobility, Multiaccess, and Netwok Management (MNM 7), pp. 9, Oct. 7. [] Y. Ito, S. Tasaka and R. Ito, Continuous time assessment and mapping of use level QoS in audio video tansmission ove IP netwoks, Poc. IASTED Intl. Conf. COMMUNICATIONS AND COMPUTER NETWORKS, pp. 7, Oct.. [] The Netwok Simulato ns, http://www.isi.edu/nsnam/ns/. [] W. Xiuchao, Simulate 8.b channel within ns, http://www.comp.nus.edu.sg/wuxiucha/sim8chns.pdf,. [7] S. Tasaka and Y. Ito, Psychometic analysis of the mutually compensatoy popety of multimedia QoS, Conf. Rec. IEEE ICC, May..