1 Stefa Paulse Istitute of Commuicatios Techology Flesburg Uiversity of Applied Scieces tadeus.uhl@fh-flesburg.de Tadeus Uhl Istitute of Commuicatios Techology Flesburg Uiversity of Applied Scieces tadeus.uhl@fh-flesburg.de The ew, parametrised VS Model for Determiig the Quality of Video Streams i the Video-telephoy Service This paper describes a ew measurig method for determiig the quality of video streams i the videotelephoy service i IP eviromets (VToIP). The method uses the so-called VS Model ad belogs to the group of parameter-based measuremet techiques (offlie operatio, i.e. without itrusio). The method yields results that come very close the curves gaied by the correspodig "Perceptual Evaluatio of Video Quality (PEVQ)" It is easy to implemet ad operates quickly ad efficietly. All i all, there a may advatages to usig this ew measurig method. 1. Itroductio Quality of Service (QoS) ad Quality of Experiece (QoE) play a very importat role i moder digital etworks. The term is becomig a household word ad ca be foud amog other thigs i the defiitio of Next Geeratio Networks accordig to the ITU-T Stadard Y.2001 [1]. I 2009 the Europea Parliamet ad Europea Coucil published directives for the stadardisatio of etworks ad services [2-3], placig great priority o quality of service. The QoS respectively QoE i moder etworks should be measured cotiuously - preferably automatically. This makes specialised measuremet systems ad methods idispesable. There are, however, hardly ay stadardised QoE ad QoS measurig methods for video-commuicatios applicatios such as video-telephoy. There are at preset oly two stadards: ITU-T Rec. J.247 [4] ad ITU-T G.1070 [5] to resort to. A third method for measurig the QoE of video services, the PEVQ (Perceptual Evaluatio of Video Quality) algorithm, has yet to be stadardised [6]. It is oe of the sigal-based QoE measurig methods. Accordig to the Germa licese holders, the compay Opticom [7], the PEVQ algorithm is i lie with Recommedatio J.247. So it seems it would make sese to work with this method to measure QoE i the VToIP service. However, the QoE measurig methods metioed so far are very complex ad the liceces expesive. There simply are o simple, parametrised QoS measurig techiques that are quick ad easy to use. The VS Model (VS = video streamig) described i this paper aims to close that gap. The VToIP service operates accordig to Recommedatio H.323 [8]. This Recommedatio defies the ecodig of audio ad video sigals. Codecs H.263, H.263+ ad H.263++ are provided to video streamig. A VToIP coectio ca be established, cotrolled ad termiated usig a rage of sigallig protocols. I practice, however, the SIP [9] protocol is by far the most widely used. Real-life measuremets of the VToIP service have revealed that a refresh rate of 25 frames per secod is widely used. Commo formats for the service are: CIF, QCIF, QVGA ad QQVGA. From a practical poit of view, this paper takes all these observatios ito cosideratio.
2 To begi with, the ew model will be formulated ad illustrated usig a actual video codec as a example. Followig that, there is a aalysis of the model's practicability, i.e. its suitability for everyday use will be put to the test i a compariso study. The results of the ivestigatio will the be preseted graphically, ad iterpreted. The paper cocludes with a summary ad a outlook o future work. 2. The ew, parametrised VS Model for determiig QoS i Video Streamig Paper [10] describes i detail a large-scale ivestigatio that, amog other thigs, demostrated that packet loss is oe of the major factors of impairmet i a VToIP eviromet. The QoS curves display a distictly expoetial character. Other parameters that affect QoS values iclude: image format ad size, ecodig rate ad burst size (average umber of cosecutively lost packets i case of radom loss). The size of the jitter buffers i the termial equipmet have a sigificat ifluece o QoS as well. Mior jitters ad some out-of-order packet sequeces ca be smoothed out i the jitter buffer. If these impairmets are too great i relatio to the size of the jitter buffer, additioal losses will occur, ad the model will register them. All of these factors must be cosidered whe creatig a ew, parametrised model for determiig QoS i VToIP. Fig. 1 shows the ewly established parametrised VS Model for determiig the quality of video streams i the VToIP service. The cocept for the VS Model is based o the A-Model [11], that was created for the VoIP service. Impairmets i the etwork Codec type ad image Jitter buffer size Jitter Reorderig Packet loss Jitter buffer Losses due to reorderig & jitter buffer Markov Model Burst size Total losses Ecodig rate Cogitive Model VS-factor [MOS] Fig. 1. Block diagram for the VS Model I practice, it is assumed that whe the QoS is determied, the packet streams from a RTP sessio are collected by a protocol aalyser ad the passed o to a suitable evaluatio tool. The ew VS Model is such a tool. It works o the followig priciple: All etwork impairmets are collected ad processed i the first block of the diagram. The effects of jitter ad out-of-order packet delivery are coverted ito losses, ot forgettig that some errors ca be compesated with the aid of the jitter buffer. The values attaied i this block ad the packet losses from the etwork are passed o to the secod block where total losses ad burst size are determied. For the VS Model we assume the Markov property "memorylessess", that is widely used i aalyses of etworks. The esuig recalculated parameters are passed o to the third ad fial block. Further iputs for the third block iclude iformatio about the image size, the codec type ad the ecodig rate that are beig used. These data are gaied from measurig the RTP streams. The last block, called "Cogitive Model" calculates ad outputs the VS-factor as a value o the MOS scale [12].
3 The mathematical depedecies eeded to do this are stored i the block i the form of a table. Its cotets are calculated through the followig steps: Step 1 Usig a suitable too such as the software tool from paper [10] for istace, determie the PEVQ curves as a fuctio of packet losses, burst size ad ecodig rate for various codecs ad image formats. The curves serve as a basis for further calculatios. This approach replaces aalyses i special studios i which groups of people are eeded to voice a subjective opiio. This solutio saves time ad moey whe it comes to developig parametrised QoS/QoE models. Step 2 Approximate each of the PEVQ curves as a fuctio of packet losses at burst sizes of "1" to "5" ad selected ecodig rates usig formula (1). The approximatio method least squares was used. a packetloss burstsize b packetloss burstsize VS factor = P e + Q e (1) The costats a ad b are selected so that they display values equal to or smaller tha 0, with sigificatly smaller values beig chose for b. The result of this is that the two summads are resposible for the steepess of the curves whe packet loss is small (2 d summad) or large (1 st summad). All costats (P, Q, a, ad b) are ow calculated iteratively as best possible values for each ecodig rate. Step 3 Determie the formulas for the costats P, Q, a ad b. Oce calculated the costats are recorded as fuctios of the ecodig rate. Their correspodig formulas ca the be calculated by meas of polyomial approximatio. The degree of the polyomial is caused by the complexity of the curve. Formulas (2) to (5) show the simple relatioship: P = w Q = z a = x b = y 1 1 Bitrate + w 1 Bitrate +... + w1 Bitrate + w0 (2) 1 1 Bitrate + z 1 Bitrate +... + z1 Bitrate + z0 (3) 1 1 Bitrate + x 1 Bitrate +... + x1 Bitrate + x0 (4) 1 1 Bitrate + y 1 Bitrate +... + y1 Bitrate + y0 (5) The steps described here will ow be demostrated usig as a example the video codec H.263, the CIF image format ad the umerical tool described i paper [10]. I this ivestigatio the followig values will be assumed for ecodig rates: 305 kbps, 1702 kbps, 3809 kbps, 4978 kbps ad 7413 kbps. The burst size displays a expoetial distributio with mea values of 1 to 5. The packet losses are subject to a biomial distributio with probability P. 31 calculatios were performed per packet loss value. This meas that at a probability of error of 5 %, cofidece itervals ca be achieved that are less tha 10 % of the mea values uder ivestigatio. The results obtaied for 3 selected ecodig rates are preseted i Figs. 2 to 4. Figs. 2 to 4 show that all QoE curves develop expoetially. The curves also show that the ecodig rate has a great ifluece o the QoE values. The small ecodig rate delivers very poor QoE values which are ot acceptable i the praxis. Burst size also has a sigificat ifluece o QoE. A burst size of 1 delivers the worst quality of experiece i all three cases ad improves with icreasig burst size. Whe small groups of packet losses occur frequetly, the sychroisatio of I/P/B images breaks off more ofte tha whe ifrequet large groups of packet losses occur. The more ofte sychroizatio breaks off, the more frequetly a static image will occur, which is reflected i a deterioratio of the QoS/QoE.
4 Fig. 2. PEVQ values as a fuctio of packet losses ad burst sizes at a ecodig rate of 305 kbps for the Codec H.263 ad the image format CIF Fig. 3. PEVQ values as a fuctio of packet losses ad burst sizes at a ecodig rate of 3809 kbps for the Codec H.263 ad the image format CIF
5 Fig. 4. PEVQ values as a fuctio of packet losses ad burst sizes at a ecodig rate of 7413 kbps for the Codec H.263 ad the image format CIF Fig. 5. Approximatio of parameters P ad Q as fuctios of the ecodig rate for the Codec H.263 ad the image format CIF
6 Fig. 6. Approximatio of parameter a as fuctios of the ecodig rate for the Codec H.263 ad the image format CIF Fig. 7. Approximatio of parameter b as fuctios of the ecodig rate for the Codec H.263 ad the image format CIF
7 Figs. 5 to 7 show the parameters superimposed o the ecodig rate ad the results obtaied from the approximatios that were coducted usig Equatios (2) to (5). The actual formulas for the Codec H.263 ad the image format CIF are therefore: P = 3,54 10 Q = 7,02 10 8 15 Bitrate Bitrate 4 2 3,45 10 + 1,36 10 10 4 Bitrate + 2,39 Bitrate 3 9,66 10 7 Bitrate 2 + 3,02 10 3 Bitrate 0,51 (6) (7) 10 2 6 a = 7,00 10 Bitrate + 8,00 10 Bitrate 2,39 10 (8) 11 3 7 2 b = 3,68 10 Bitrate 5,23 10 Bitrate + 1,94 10 Bitrate 2,80 (9) 3. Compariso Study 3.1. Numerical Aalyses The software tool preseted i paper [10] was used here, too. The followig parameters were assumed for the umerical compariso study: # Nodetermiistic distributed packet losses of 0 to 20 % ad costat burst size of 1 at a ecodig rate of 1702 kbps # Nodetermiistic distributed packet losses of 0 to 20 % ad odetermiistic burst size of 2 at a ecodig rate of 1702 kbps # Nodetermiistic distributed packet losses of 0 to 20 % ad costat burst size of 1 at a ecodig rate of 4978 kbps # Nodetermiistic distributed packet losses of 0 to 20 % ad odetermiistic burst size of 2 at a ecodig rate of 4978 kbps # Video codec H.263 # Image format CIF # Image refresh rate of 25 images/s # 31 measuremets per value of each of the variables (here: packet losses). This esures that cofidece itervals are achieved that are less tha 10 % of the mea values uder aalysis (with a probability of error of 5 %). # PEVQ ad VS Models as the QoS measurig techiques # As the referece video a AVI file from the compay Opticom [7] was chose. The file has duratio of 8 secods ad a resolutio of 352 x 288. Figure 8 shows a screeshot of the referece video. 2 3 Fig. 8. Screeshot of the referece video The results of the compariso study are preseted graphically i Figs. 9 to 12.
8 Fig. 9. QoS values as a fuctio of packet losses gaied from differet measurig methods for the Codec H.263, the image format CIF, burst size 1 ad a ecodig rate of 1702 kbps (umerical aalysis) Fig. 10. QoS values as a fuctio of packet losses gaied from differet measurig methods for the Codec H.263, the image format CIF, burst size 2 ad a ecodig rate of 1702 kbps (umerical aalysis)
9 Fig. 11. QoS values as a fuctio of packet losses gaied from differet measurig methods for the Codec H.263, the image format CIF, burst size 1 ad a ecodig rate of 4978 kbps (umerical aalysis) Fig. 12. QoS values as a fuctio of packet losses gaied from differet measurig methods for the Codec H.263, the image format CIF, burst size 2 ad a ecodig rate of 4978 kbps (umerical aalysis)
10 Figs. 9 to 12 show that the quality of service deteriorates expoetially as packet losses icrease. This is the case for both QoS measurig methods used here. Furthermore, the curves fall less steeply as burst size icreases. The upshot of this is: it is far better for the service if fewer, larger budles of packets are lost tha lots ad lots of smaller budles. The reaso for this lies i the properties of huma visio. The PEVQ ad the VS Model curves proceed very close to each other, meaig that the umerical compariso study has proved that the VS Model is quite suitable for practical use. 3.2. Aalyses i the real Eviromet Fig. 13 shows the real eviromet used i the aalysis. The Video Streamig Quality Measuremet System (VSQMS), that was described i detail i paper [13], is the maistay of the eviromet. The measuremet system is capable of establishig a RTP coectio betwee two VoIP statios ad usig it to trasmit video streams ecoded accordig to H.263. The VSQMS cliet fuctios here as a "mirror" that seds the icomig RTP packets back to the VSQMS aget. The aget comprises both the PEVQ Algorithm ad the ew VS Model. Both are methods to determie values for QoE respectively QoS. Fig. 13. The real eviromet Figs. 14 ad 15 show examples of the results obtaied from the real aalysis eviromet. The curves from Fig. 14 show a remarkable agreemet with the correspodig curves from the umerical aalyses (cf. Fig. 9). Comparig Figs. 10 ad 15 leads to similar coclusios, i.e. icreases i burst size lead to higher QoS values. Here it becomes evidet that the two measurig methods yield comparable QoS values i a real eviromet. Cosequetly, the usability of the ew VS Model i real-life situatios has bee proved here as well.
11 Fig. 14. QoS values as a fuctio of packet losses obtaied from differet measurig methods for the Codec H.263, image format CIF, burst size "1" ad a ecodig rate vo 1702 kbps (real eviromet) Fig. 15. QoS values as a fuctio of packet losses obtaied from differet measurig methods for the Codec H.263, image format CIF, burst size "2" ad a ecodig rate vo 1702 kbps (real eviromet)
12 4. Summary ad Outlook I the course of the work described i this paper a ew parametrised QoS measurig model was developed to evaluate the quality of video streams i the video-telephoy service ad its fuctioality was proved i a real-eviromet aalysis. Whe the ew model was defied, all due cosideratio was paid to the typical parameters for the VToIP service. The ew VS Model is based o the PEVQ curves. This is of especial practical sigificace sice the PEVQ is cosidered to be the most objective QoS measurig method for video. The compariso study has proved the practicability of the ew QoS model beyod ay shadow of doubt. The ew, quick ad easy-to-use ad iexpesive VS Model ca defiitely be see as a superior alterative to the laboriously slow PEVQ method with its expesive liceces. I order to be able to assess the QoS of the VToIP service i its etirety, it is ecessary i practice to cosider the quality of the audio streams as well. The ITU-T Recommedatio G.1070 [5] cotais suggestios as to how this might be achieved. Paper [14] itroduced the ew E(IP) Model. It is very a very effective tool for determiig the quality of audio streams i a VoIP eviromet. It is a obvious ext step to combie the two models to create a joit VT Model (VT = Video-telephoy) to aalyse the complete VToIP service. Research work towards that ed has already begu at the Flesburg Uiversity of Applied Scieces. Refereces 1. Defiitio of NGN, http://www.itu.it/rec/t-rec-y.2001; page last viewed July 2012. 2. Directives 2009/140/WE for the creatio of egieerig stadards i etworks ad electroic services, http://eurlex.europa.eu/lexuriserv/lexuriserv.do?uri=oj:l:2009:337:0037:0069: PL:PDF; page last viewed July 2012. 3. Directives 2009/136/WE for the creatio of egieerig stadards i etworks ad electroic services, http://eurlex.europa.eu/lexuriserv/lexuriserv.do?uri=oj:l:2009:337:0011:0036: PL:PDF; page last viewed July 2012. 4. ITU-T Recommedatio J.247, http://www.itu.it/rec/t-rec-j.247-200808-i; page last viewed July 2012. 5. ITU-T Recommedatio G.1070, http://www.itu.it/rec/t-rec-g.1070/e; page last viewed July 2012. 6. PEVQ algorithm ad tool, http://www.opticom.de/dowload/pevq-wp-v07-a4.pdf; page last viewed July 2012. 7. The compay Opticom, http://www.opticom.de; page last viewed July 2012. 8. ITU-T H.323 protocol suite, http://www.opeh323.org/stadards.html; page last viewed July 2012. 9. IETF SIP protocol suite, http://www.ietf.org/rfc/rfc3261.txt; page last viewed July 2012. 10. Ch. Hoppe, T. Uhl, K. Nowicki, Determiig QoS i the Video Telephoe Service i a IP Eviromet, Proceedigs of the 3 rd Iteratioal Workshop o Future Multimedia Networkig, Krakow/Polad, Jue 2010, pp. 123-133. 11. S. Paulse, T. Uhl, Das eue, parametrisierte A-Modell für QoS beim Diest VoIP. Fachzeitschrift PIK (Praxis Iformatiosverarbeitug ud Kommuikatio), 34. Jahrgag, Heft 1/2011, S. 23-30. 12. Mea Opiio Score (MOS), http://www.itu.it/rec/t-rec-p.800/e; page last viewed July 2012. 13. T. Uhl, Ch. Hoppe, Neues Messsystem zur Beurteilug der Qualität des Videostroms beim Diest Video Telephoy i IP-Umgebug. Proceedigs of the WowKiVS2011, Kiel, March 2011, pp. 197-207 & Electroic Commuicatios of the EASST (ISSN 1863-2122), vol. 37 (2011). 14. S. Paulse, T. Uhl, Adjustmets for QoS of VoIP i the E-Model. Proceedigs of the WTC (World Telecommuicatios Cogress) 2010 (CD form), Viea/Austria, September 2010, pp. 69-74.