46 JOURNAL OF MULTIMEDIA, VOL. 1, NO. 6, SEPTEMBER 2006 Seeh Qualty Measurement Methods wth Alyng PLC Algorthms on Real-tme Transmsson Control Sheme for VoIP Serve Jnsul Km Dgtal Meda Laboratory, Informaton and Communatons Unversty, Seoul, Korea Emal: jskm72@u.a.kr Hyun-Woo Lee, Won Ryu and Byung Sun Lee BN Serve Researh Grou / Broadband onvergene Network Researh Dvson, Eletrons and Teleommunatons Researh Insttute, Daejeon, Korea Emal: {hwlee, wlyu, bslee}@etr.re.kr Seung Ho Han, Mnsoo Hahn Seeh and Audo Informaton Laboratory, Informaton and Communatons Unversty, Daejeon, Korea Emal: {sae0128, mshahn}@u.a.kr Abstrat Aordng to the dsovery of avalable and sutable audovsual serves over IP, researhes on the Qualty of Serve (QoS) for varous multmeda alaton serves are requred. Ths aer resents evaluaton of seeh qualty alyng mroved Paket Loss Conealment (PLC) algorthms wth an end-to-end bdretonal transmsson ontrol sheme. We roose real-tme transmsson ontrol sheme to erform better qualty montorng of VoIP serve. In order to evaluate the newly roosed RTCP-XR sheme for VoIP serves, we measure n real-tme the qualty degree of the reeved seeh sgnal whle alyng our mroved Paket Loss Conealment (PLC) algorthms n G.711 and G.729A odes usng an mlemented Sesson Intaton Protool (SIP)- based SoftPhone and a 1-ort gateway systems. Overall, for the evaluaton wth sentf exattude of seeh qualty aled our PLC algorthms we examne wth made hone-alls through the heterogeneous network systems. The exermental results onfrm the aled PLC algorthms for mrovement of seeh qualty over IP n real envronments wth the QoSfator transmsson ontrol mehansm. Index Terms VoIP, QoS, Seeh Qualty, RTCP-XR, PLC, PESQ, E-Model, SIP, BN serves I. INTRODUCTION As the nterests n suortng IP ommunaton Based on Multmeda Convergene Serve usng Mult-funtonal Systems:3D Vrtual Montorng wth Dynam Voe and Vdeo over IP based on SIP, by Jnsul Km, Hyun-Woo Lee, Ryu Won, and Byung Sun Lee whh aeared n the Proeedngs of the IEEE Consumer Communatons and Networkng Conferene 2006, Las Vegas, U.S.A., January 2006. 2006 IEEE. tremendously nrease, a varety of IP-based real tme multmeda alaton serves have been develoed radly n resdental and enterrse ommunaton markets beause of ther attratve serve enhanements and ost savngs. Moreover, wth the nreasng demand for Internet telehony serves by usng the Voe over IP (VoIP) tehnology, tehnques for mrovng seeh qualty over IP networks are one of the hot ssues n researh area today. Also, QoS-fator montorng to manage n real-tme are requred to guarantee the qualty of VoIP serves. An mortant fator for VoIP QoS ontrol tehnque nvolves the rate ontrol, whh s based largely only on network marments suh as the aket loss rate or the delay due to the network ongestons [1], [2]. The seond mortant fator s based on the fator that dfferent arts of seeh have dfferent eretual mortane and eah art of seeh does not ontrbute equally to the overall voe qualty [3]. Hene, n our aroah, voe akets that are eretually more mortant are marked,.e. aqure rorty. If there s any ongeston, the akets are less lkely to be droed than akets that are of less eretual mortane. The rorty markng based QoS shemes are oen loo ones and do not make use of hanges n the network marments. To measure the seeh qualty we adot the well-known PESQ (Peretual Evaluaton of Seeh Qualty) algorthm, whh s one of objetve eretual voe qualty measurement tehnques, based on the ITU-T seeh qualty evaluaton standards [4]. Non-ntrusve methods do not need a referene sgnal and are arorate for montorng the real traff. The ITU-T E-model [5] s the most wdely used non-ntrusve voe qualty measurement
JOURNAL OF MULTIMEDIA, VOL. 1, NO. 6, SEPTEMBER 2006 47 method and may be used to redt the onversatonal MOS (Mean Onon Sore) dretly from the IP network and the termnal arameters [6]. The Real-tme Transort Protool (RTP) and RTP Control Protool (RTCP) ommunatons use the RTCP- Reever Reort to feedbak IP network ondtons from RTP reevers to RTP senders. However, the orgnal RTCP rovdes overall feedbak on the qualty of end-to-end networks as a whole [7]. The RTP Control Protool Extended Reorts (RTCP-XR) s a new VoIP management rotool whh defnes a set of metrs that ontans nformaton for assessng the VoIP all qualty by the IETF [8]. The RTCP-XR reorts the aket loss rate, the aket dsard rate (beause of jtter) and the dstrbuton of lost/dsarded akets. The loss/dsard dstrbuton desrbes the all n terms of bursts (erods durng whh the loss/dsard rate s hgh enough to ause noteable qualty degradaton) and gas (erods durng whh lost or dsarded akets our nfrequently and hene qualty s generally aetable). To guarantee the all qualty, the RTCP-XR an reort the all qualty dretly n terms of the estmated R fator or the mean onon sore (MOS). The R fator s a onversatonal-qualty metr n the range of 0 to 100. And the MOS-LQ s a eretual lstenng-qualty metr whle the MOS-CQ s a onversatonal-qualty metr, both n the range of 1 to 5. The RTCP-XR an be mlemented as software ntegrated nto IP hones and gateways nexensvely. Then, the messages ontanng key all-qualty-related metrs are exhanged erodally between IP hones and gateways. However, n order for montorng the QoS-fator on end-to-end VoIP networks, the RTCP-XR s nsuffent to estmate the seeh qualty beause t doesn t have any seeh qualty montorng funtonalty. To solve ths roblem we roose the udate of some omonents n RTCP-XR sheme. Paket loss ours nherently n an IP network due to the network haratersts suh as the delay, the jtter, and network ongestons. The aket loss stands mssng seeh samles for a tme nterval at reever and brngs about the degradaton of the reeved seeh qualty. Sne transmsson s a real-tme roess, t s mossble to retransmt the mssed data when the data s mssed. A onealment tehnque s requred for the mssed aket reonstruton. Seeh odes have ther own PLC mehansms aganst the aket loss. At the reever, a new seeh sgnal s generated to redue the effet of aket loss by utlzng the PLC mehansm. [9, 10] In ths aer, we roosed an end-to-end bdretonal transmsson ontrol mehansm to manage the QoS-fators suh as the jtter, the delay, and the aket loss rate. The newly resented QoS-fator transmsson ontrol mehansm for the QoS management and montorng of the seeh qualty durng mutual onversatons over IP networks s assessed wth the erformane analyss of the roosed PLC algorthms n VoIP seeh oders. Varous tyes of aket losses are generated by the Error Inserton Deve (EID) and then the PESQ s used to evaluate both the orgnal and the roosed PLC algorthm n G.711 and G.729A. Also, to measure and reort the R-fators, our exerments sefy the QoS erformane wth the objetve measurement method over the IP network,.e., E-Model and PESQ. II. VOIP NETWORK ARCHITECTURE AND PROCEDURE FOR QOS-FACTOR MANAGEMENT A. VoIP System Arhteture Hstorally, voe has been delvered, ether wreless or wred, n a dedated, one-to-one rut swthed manner. Ths s beause of the unque requrements voe traff laes on a network. It requres the low lateny and jtter erformane of a voe all, but laes hgh demands on the data aaty at the same tme. To guarantee the QoS for VoIP serves we roosed a new arhteture wth a QoS measurng server. Our roosed arhteture, Fg. 1 s desgned wth a SoftPhone and a 1-ort gateway as the man User Agent (UA) whh arres QoS measurement omonents. Our network also nludes SIP servers and a QoS-fator montorng server for the all and QoS ontrol. For the aurate QoS delverng evaluaton of the RTCP-XR Sheme on the QoS-fator montorng server, we aled alls through the PSTN to the SIP-based UA-SoftPhone, the SIP-based UA-SoftPhone to the PSTN, and the SIP-based UA-SoftPhone to the SIP-based UA-SoftPhone. B. Real-tme QoS-Fator Measurement and Transmsson Beause resent IP networks are not desgned to suort the QoS, the qualty measurement beomes more mortant and urgent for more relable hgher qualty multmeda serves over IP networks. We exlore the mat of the ndvdual aket loss, the delay, and the jtter on the eretual seeh qualty n VoIP systems. The VoIP serve qualty evaluaton s arred out by frstly enodng the nut seeh re-modfed wth gven network arameter values and then deoded to generate degraded outut seeh sgnals. In order to obtan an end-to-end (E2E) MOS between the aller-ua and the allee-ua, we aly the PESQ and the E-Model method. In detal, to obtan the R fators for E2E measurement over the IP network we need to get Id, Ie, Is and Ij. Here, Ij s newly defned as n equaton (1) to reresent the E2E jtter arameter. R fator = R0 Is Id Ij Ie + A (1) The ITU-T Reommendaton rovdes most of the values and methods to get arameter values exet Ie for the G.729A ode, Id and Ij. Frst, we obtan Ie value after the PESQ algorthm aled. Seond, we aly the PESQ values to Ie value of R-fator as shown n Fg.2. We measure the E2E Id and Ij from our urrent network
48 JOURNAL OF MULTIMEDIA, VOL. 1, NO. 6, SEPTEMBER 2006 envronment. By ombnng Ie, Id and Ij, the fnal R fator ould be omuted for the E2E QoS erformane results. Fnally, obtaned R fator s reonverted to MOS by usng equaton (2), whh s redefned by the ITU-T SG12. R 6.5: MOS= 1 7 7 7 6.5 R 100: MOS= 1 R+ R R 1000 6250 1000000 R 100: MOS= 4.5 Fg. 1. VoIP network arhteture 2 3 Fg. 2. Proedure of QoS-Fator measurement and transmsson III. QOS-FACTOR TRANSMISSION CONTROL MECHANISM BASED IN RTCP-XR SCHEME A. Oeraton Proedure wth Voe Paket Transmsson The RTP rotool s used for transmttng real-tme data nformaton and the RTCP, for sendng the ontrol nformaton. The man funton of the RTCP s to rovde a detaled reresentaton of the voe akets exhanged durng an RTP sesson. Its struture nludes the sender reort (SR) and the reever reort (RR) transmtted erodally to all artants n the sesson. It ams at rovdng a feedbak on the qualty of the transmsson (e.g., delay jtter, average aket loss, et.), where transmtters (2) send sender reorts and reevers send reever reorts usng the RTCP-XR. Whle the SR nludes transmsson and reeton nformaton for atve senders n the sesson, the RR would also ontan the reeton nformaton for non-atve senders. The QoS measurement server reeves the QoS-fator and reorts them to the QoS-montorng nterfae every 2-seond. Fg. 2 also shows that the roedure of the RTP/RTCP/RTCP-XR aket transmsson wth the QoS measurement roedure. B. Conversatonal Seeh Qualty Montorng wth Modfed RTCP-XR Sheme For the onversatonal seeh qualty we have develoed an RTCP-based aket struture to rovde an end-to-end bdretonal transmsson ontrol that an reort the delay, the jtter, and the aket losses n a tmely manner. Our aket struture s smlar to the RTCP-XR, whh s rmarly defned to rovde more detaled statsts, artularly for multast alatons [8]. In our ase, the RTCP-XR sheme was sefally desgned to reort the delay, the jtter, and the aket losses for every frame of voe sgnal. Also, the loss and the dsard rates are desgned to be alulated for eah sesson at the endreever n order to get real-tme measured values. The orgnal RTCP-XR aket tye defned an be used for seeh qualty montorng. However, t s not famlar wth real-tme seeh qualty reortng for bdretonal onversaton through VoIP networks. We modfed the RTCP-XR aket sheme n order for onversatonal seeh qualty reortng and montorng beause ommunaton serves lke the VoIP must be reognzed as the mutual onversaton one. For the delay montorng as one of the sgnfant QoSfator BT-5 s formatted wth both sub-blok 1 and subblok 2. SSRC_1 and SRRC_2 are for the sender and the reever numbers whh s defned randomly. The DLRR n sub-blok 1 reorts the one way delay between the sender and the reever. The DLRR n sub-blok 2 reorts the round tr delay whh s measured usng the one way delay nformaton from DLLR n sub-blok 1. Fg. 3 shows the message format-ii of BT-5 for delay montorng. In Fg. 4, nformaton of the jtter and the akets s ontrolled n BT- 6. At frst, n order to manage the jtter, we ategorze t nto three levels of mn_jtter, max_jtter, and mean-jtter. Those are reorted as the umulatve effets of jtter values obtaned through the jtter buffer n our ases. Seond, to get the real-tme mutual ommunaton seeh qualty montorng feld of aket ount nformaton of the RTP/RTCP s nluded n the reort blok. The Tx/Rx RTP Pakets format n BT-6 s desgned for montorng sender/reever RTP. The Rx RTCP Pakets are the RTCP XR akets whh are reeved at the QoS measurement server, and the Tx RTCP Pakets are the RTCP XR akets whh are sent from the SoftPhone (or 1ort Gateway) to the QoS measurement server. The bdretonal QoS-fator s also measured n real-tme and reorted to the QoS-
JOURNAL OF MULTIMEDIA, VOL. 1, NO. 6, SEPTEMBER 2006 49 Montorng deve. The fnal result value of the umulatve aket loss s also nluded n our BT-6 sheme. The format of our BT-7 sheme s smlar to the standard RTCP-XR. However, the loss and the dsard rates are omuted as soon as the all sesson stats by usng the umulatve akets ontrolled by BT-6. The auray of eah aket ountng n real-tme s mortant. Eseally, for tme synhronzaton we set our exeuton tme of both the SoftPhone and the 1-ort gateway to the NTP (Natonal Tme Protool). That s, the host system tme s synhronzed wth the natonal standard tme. Fnally, by alyng the QoS-fator obtaned from the bdretonal seeh qualty measurement MOS-LQ and MOS-CQ s defned searately for the lstenng qualty and the onversatonal qualty. Other fators n BT-7 are just added to the standard format sheme as shown n Fg. 5. Fg. 3. BT=5 delay montorng for transmsson ontrol Fg. 5. BT=7 loss & dsard late, MOS-LQ, & -CQ for transmsson ontrol IV. PROPOSED PLC ALGORITHMS When aket losses our over IP networks, the PLC algorthms emloyed n seeh odes reonstrut lost seeh frames based on the revously reeved seeh nformaton. The PLC algorthm emloyed n G.711 Aendx I reeatedly nserts th erod deteted from the revous seeh n hstory butter [9] whh s alled the th erod relaton method. The PLC algorthm emloyed n G.729A estmates an extaton sgnal and synthess flter arameters from last good frame [10]. In ths aer, two tehnques are roosed to onstrut new PLC algorthms for G.711 and G.729A odes, resetvely. The roosed PLC algorthms adatvely utlze both-sde nformaton for lost frames based on the lass lassfaton and adatvely ontrol gans aordng to the average magntude varaton aross the frames. We aly the roosed algorthm to our system to rove akets loss omensaton n both off-lne test envronment n whh the deoder s modfed by our PLC algorthm and real-tme envronment, by the transmsson ontrol mehansm based on the RTCP-XR sheme wth gven QoS-fators. Fg. 4. BT=6 jtter, akets for transmsson ontrol A. Alyng G.711 Frst of all, the roosed PLC algorthm uses adatvely both-sde nformaton for lost frames wth the buffer to store a sueedng frame. When both-sde nformaton for lost frame s avalable, the mssng seeh s reonstruted by adatve both-sde nformaton utlzaton. The last good frame s lassfed nto ether the lass A or the lass B by the deson tree. To make the deson tree, we use the CART (Classfaton And Regresson Tree) [11] and analyze 1000 Korean dalogue seeh. The ZCR (Zero Crossng Rate) [12], the rato between the average magntude [12] of the urrent frame and that of the adjaent frame, the average magntude rato between the subframes n urrent frame, and rato between the frst value and the seond value of the autoorrelaton funton [12] are evaluated as the feature arameters. If the last good frame s lassfed nto the lass A, the
50 JOURNAL OF MULTIMEDIA, VOL. 1, NO. 6, SEPTEMBER 2006 roosed PLC algorthm uses only sueedng frame nformaton to reonstrut mssng seeh by usng the th erod relaton method whh has same roedures wth that emloyed n G.711 Aendx I. On the other hand, f the last good frame s lassfed nto the lass B, the roosed PLC algorthm uses both revous frame nformaton and sueedng frame nformaton to reonstrut mssng seeh. Reonstrutng seehes by usng revous nformaton and sueedng nformaton are averaged and the averaged sgnal beomes the fnal reonstrutng seeh for mssng seeh. Seond, the roosed PLC algorthm ontrols adatvely gans aordng to the average magntude varaton aross the frames. The varaton s evaluated by G. G I only n nreasng ase and G D only n dereasng ase are resetvely alulated by 2E E 2E E G =, G = 2 (3) I 1 2 2 1 D E 1 E 1 where E s the average magntude, and s the frame ndex. Evaluated G I or G D s set as G. The attenuaton fator s nreased or dereased u to G for frst loss erod of 10 mse. From after that to the erod of 60 mse, the attenuaton fator s lnearly dereased from the G to 0. In other words, G.711 Aendx I PLC attenuates the reonstrutng seeh wth fxed fator [9] whle the roosed algorthm has varyng attenuaton fator aordng to evaluated G. B. Alyng G.729A For G.729A ode when both-sde nformaton s avalable, the roosed PLC algorthm modfes the LSP (Lne Setrum Pars) nterolaton oeffents for the frst subframe based on the lass of the last good frame whh s lassfed nto ether the lass A or the lass B by the deson tree. It s same wth that used n the roosed G.711 PLC algorthm. In the lass A, the modfed LSP nterolaton s gven as where frame, q = q + q = (4) (revous) (future) 0.48 0.52, 1,...,10 q s the -th LSP oeffent for the urrent lost (revous) q, for the revous frame, and (future) q, for the future frame. And, n the lass B, the modfed LSP nterolaton s gven as q = q + q =. (5) (revous) (future) 0.52 0.48, 1,...,10 Seond, the roosed PLC ontrols adatvely gans aordng to the average magntude varaton aross the frames. The PLC algorthm emloyed n G.729A attenuates the adatve and fxed odebook gans as follows. g = g g = g (6) m (m-1) m (m-1) 0.98, 0.9 where m s the subframe ndex, g s the fxed odebook gan, and g s the adatve odebook gan. The roosed PLC algorthm modfes the attenuaton fators based on the varaton of average magntudes between revous two frames. The varaton s evaluated by G wth same way roosed n G.711 PLC algorthm. Wth ths G, the adatve and fxed odebook gans are attenuated by the new fators as follows. G> g = g g = g 0.7 1: = 0.98, = 0.9 G < g = g g = g 1: G m (m-1) m (m-1) 0.988, 0.92 m (m-1) m (m-1) g g g g 0.7 : m (m-1) m (m-1) 0.977, 0.88 V. SIMULATION SURROUNDINGS Ths aer evaluates the erformane of the QoS-fator transmsson ontrol mehansm wth the roosed PLC algorthm under dfferent aket loss smulatng ondtons. The PLC algorthm s tested n both n off-lne envronment and real-tme envronment wth G.711 and G.729A odes. In the off-lne envronment, Fg. 6 shows the exermental roedure. Frst, seeh data are enoded n the enoder. Seond, aket losses are generated by the EID. Thrd, enodng fles are deoded wth and wthout aket losses. Fnally, deodng fles wth aket losses are evaluated wth the referene seeh deoded wthout a aket loss n the seeh qualty evaluaton module. The erformane of the roosed PLC algorthms for G.711 and G.729A are omared to that of the G.711 Aendx I PLC algorthm and that of the PLC algorthm emloyed n G.729A, resetvely. In real-tme envronment, our VoIP system ales the QoS-fator transmsson ontrol mehansm based on the RTCP-XR sheme for the QoS-fator management and montorng. Fg. 7 outlnes the onfguraton of smulaton envronment alyng roosed QoS-fator sheme and mroved PLC algorthms under tests. The PLC algorthms are tested eah other nteratvely usng the SoftPhone and 1ort gateway. The erformane of the roosed PLC algorthm for G.729A s omared to that of the PLC algorthm emloyed n G.729A. And the referene system for G.711 has no PLC mehansm and smly nserts slene for a mssng erod. Thus, the roosed PLC algorthm for G.711 s emloyed n G.711. To show the effet of roosed PLC algorthm, the seeh qualty of the referene system s omared to that of the system wth the roosed PLC algorthm when aket losses our. (7)
JOURNAL OF MULTIMEDIA, VOL. 1, NO. 6, SEPTEMBER 2006 51 Seeh Coder Enoder Deoder Paket loss Deodng seeh wth aket loss EID Deodng seeh wthout aket loss Fg. 6. Exermental roedure Seeh Qualty Evaluaton Module PESQ sore We use 200 Korean dalogue seeh utteranes from 2 male and 2 female seakers as test data. The duraton of eah utterane s about 10 seonds. VI. SIMULATION RESULTS The smulaton shows that the roosed PLC algorthms wth G.711 and G.729A odes over IP n real envronments alyng our QoS-fator transmsson ontrol mehansm based on the modfed RTCP-XR sheme aheve muh better result for seeh qualty than other VoIP systems whh use standard sheme and no PLC algorthm durng mutual onversaton n IP network envronment. In the followng result tables, f refers to frame er aket. Fg. 7. Confguraton of smulaton envronment alyng roosed QoSfator sheme and mroved PLC algorthm A. Paket Loss Desgn A ommunaton hannel an be modeled by the Glbert- Ellot model [13] or the Bellore model [14]. In the smulaton, we use the STL (Software Tool Lbrary) nluded n ITU-T Reommendaton G.191 [15] to model aket losses. The STL rovdes the EID whh generates losses based on the Glbert-Ellot model and the Bellore model. Varous atterns of losses an be generated by the modfaton of arameters n the EID. To model varous aket loss envronments, we desgn burst and random aket losses wth 1%, 3%, 5%, 10%, and 15% loss rates and t s onsdered that a aket ontans 1 seeh frame, 2 seeh frames, or 3 seeh frames. B. Performane Evaluaton For PLC algorthm erformane evaluaton, we use the system evaluaton method whh s PESQ defned by ITU- T Reommendaton P.862 for objetve assessment of qualty. After omarng an orgnal sgnal wth a degraded one, the outut of PESQ rovdes a sore from -0.5 to 4.5 as a MOS-lke sore. The referene seeh for the real-tme envronment smulaton and the off-lne envronment smulaton s the orgnal seeh and the deoded seeh wthout any aket loss, resetvely. In the real-tme envronment, after gettng the PESQ, t s used for E-Model evaluaton method as we menton above. Fnally, the MOS s aqured. A. PESQ Measurement n Off-lne Test Envronment Table I shows the results of the PLC erformane evaluaton for G.711. At 1f, the degree of mrovement for burst losses s hgher than that for random losses. PESQ sores are nreased by 0.016 for random losses and 0.103 for burst losses on the average, resetvely. On the other hands, at 2f and 3f, the degree of mrovement for random losses s hgher than that for burst losses beause of the aketzaton effet. At 2f, the degree of the mrovement for random losses s 0.112 whle that for burst losses s 0.047 on the average. At 3f, the degree of the mrovement s 0.119 for random losses and 0.022 for burst losses, resetvely. Table II shows the results of the PLC erformane evaluaton for G.729A. At 1f, the dfferene of the degree of mrovement sore between random losses and burst losses s not large. PESQ sores are nreased by 0.074 on the average. On the other hands, at 2f and 3f, the degree of mrovement for random losses s hgher than that for burst losses beause of the aketzaton effet. At 2f, the degree of the mrovement for random losses s 0.169 whle that for burst losses s 0.042 on the average. At 3f, the degree of the mrovement s 0.124 for random losses and 0.0389 for burst losses, resetvely. Thus, the suerorty of the roosed PLC algorthm s roved n the off-lne test envronment. TABLE I PESQ RESULT FOR OFF-LINE ENVIRONTMENT WITH G.711 PLC Orgnal PLC Proosed PLC Loss tye 1f 2f 3f 1f 2f 3f 1% 3.983 3.987 4.018 3.984 4.025 4.066 3% 3.644 3.630 3.621 3.641 3.717 3.714 Random 5% 3.440 3.412 3.390 3.451 3.524 3.504 10% 3.124 3.038 2.965 3.149 3.186 3.123 15% 2.893 2.784 2.667 2.940 2.959 2.845 Burst 1% 3.975 3.991 3.983 4.021 4.029 4.009
52 JOURNAL OF MULTIMEDIA, VOL. 1, NO. 6, SEPTEMBER 2006 3% 3.558 3.589 3.600 3.637 3.621 3.600 5% 3.317 3.262 3.305 3.412 3.297 3.323 10% 2.839 2.758 2.755 2.972 2.817 2.783 15% 2.504 2.381 2.407 2.664 2.455 2.446 Burst 0% 4.2 93 4.4 4.2 93 4.4 3% 3.2 64 3.3 3.2 65 3.4 5% 3.0 57 3 3.3 64 3.3 10% 2.7 50 2.6 2.8 53 2.7 TABLE II PESQ RESULT FOR OFF-LINE ENVIRONTMENT WITH G.729A PLC Orgnal PLC Proosed PLC Loss tye 1f 2f 3f 1f 2f 3f Random Burst 1% 3.952 3.912 3.939 3.976 4.003 4.005 3% 3.569 3.488 3.500 3.615 3.637 3.605 5% 3.319 3.239 3.234 3.386 3.417 3.359 10% 2.936 2.818 2.776 3.038 3.030 2.932 15% 2.658 2.537 2.475 2.787 2.751 2.642 1% 3.953 4.031 4.019 3.991 4.049 4.039 3% 3.531 3.634 3.698 3.587 3.663 3.717 5% 3.275 3.347 3.406 3.352 3.393 3.434 10% 2.808 2.896 2.936 2.901 2.953 2.974 15% 2.478 2.543 2.585 2.586 2.604 2.674 TABLE IV RESULT FOR REAL-TIME ENVIRONMENT WITH G.729A PLC Loss tye Random Burst Orgnal PLC Proosed PLC PESQ R MOS PESQ R MOS 0% 3.5 72 3.7 3.5 72 3.7 3% 3.2 63 3.4 3.3 68 3.5 5% 3.1 62 3.2 3.2 65 3.4 10% 2.9 57 2.9 3 58 3.0 0% 3.5 72 3.7 3.5 72 3.7 3% 3.2 64 3.3 3.3 67 3.5 5% 3.1 61 3.2 3.2 64 3.2 10% 2.9 56 2.9 3.0 57 2.9 B. PESQ, R-fator and MOS Measurement n the Real-tme Envronment In Table III, the PLC erformane evaluaton results for G.711 for the real-tme envronment are summarzed as the PESQ, the R-fator, and the MOS. The roosed PLC aheves the PESQ gans of from 0.1 to 0.3 as to loss tyes. The orresondng gans for the R-fators and the MOSs are also aheved by these PESQ gans. Hgh gans are aheved at random losses wth hgh loss rates. In Table IV, the PLC erformane evaluaton results wth G.729A for the real-tme envronment are summarzed as the PESQ, the R-fator, and the MOS. The roosed PLC aheves the PESQ gan of 0.1 for all tyes of loss. The orresondng gans for the R-fators and the MOSs are aheved by these PESQ gans. Fnally, even though the MOS s not mroved for burst losses at hgh loss rates, the MOS gan of 0.1 s generally aheved. TABLE III RESULT FOR REAL-TIME ENVIRONMENT WITH G.711 PLC No PLC Proosed PLC Loss tye PESQ R MOS PESQ R MOS Random 0% 4.2 93 4.4 4.2 93 4.4 3% 3.5 71 3.6 3.8 78 3.9 5% 3.0 59 3 3.4 68 3.5 10% 2.5 48 2.5 3.0 58 3 VII. CONCLUSION In ths aer, the erformane evaluaton of seeh qualty onfrms that our roosed PLC algorthm outerforms more effently than the orgnal algorthm n the G.711 and G.729A seeh odes. Also, we have roosed an end-to-end bdretonal QoS-fator transmsson ontrol mehansm to manage the seeh qualty n real-tme on modfed RTCP-XR sheme for QoS montorng of mutual onversatons over IP networks. The newly resented QoS-fator transmsson ontrol mehansm s assessed wth the erformane analyss of the roosed PLC algorthms n off-lne and real-tme envronment. Fnally, we have shown that, based on the smulaton results the seeh qualty for VoIP serve an be sgnfantly mroved by usng the roosed PLC algorthm wth newly resented transmsson ontrol mehansm. REFERENCES [1] R. Eejae, M. Handley and D. Estrn, RAP: An End-to-end Rate-based Congeston Control Mehansm for Realtme Streams n the Internet, Pro. IEEE INFOCOM, USA, 21-25 Marh 1999. [2] F. Bertell, G. Rugger and G. Shembra, TCP-Frendly Transmsson of Voe over IP, Proeedngs of IEEE Internatonal Conferene on Communatons, New York, USA, Arl 2002.
JOURNAL OF MULTIMEDIA, VOL. 1, NO. 6, SEPTEMBER 2006 53 [3] J. C. De Martn, Soure-drven Paket Markng for Seeh Transmsson over Dfferentated-Serves Networks, Proeedngs of IEEE ICASSP 2001, Salt Lake Cty, USA, May 2001. [4] ITU-T Reommendaton P.862, Peretual evaluaton of seeh qualty (PESQ), an objetve method for end-to-end seeh qualty assessment of narrowband telehone networks and seeh odes, Feb. 2001. [5] ITU-T Reommendaton G.107, The E-Model, a omutatonal model for use n transmsson lannng, 2000. [6] R.G. Cole and J. H. Rosenbluth, VoIP over IP Performane Montorng, Journal on Comuter Communatons Revew, vol 31, no.2, Arl 2001. [7] H. Shulzrnne, S. Casner, R. Frederk and V. Jaobson, RTP: A Transort Protool for Real-Tme Alaton, IETF RFC 3550, July 2002. [8] T. Fredman, R. Caeres, A. Clark, RTP Control Protool Extended Reorts, IETF RFC 3611, Nov. 2003. [9] ITU-T Reommendaton G.711 Aendx I, A hgh qualty low-omlexty algorthm for aket loss onealment wth G.711, Set. 1999 [10] ITU-T Reommendaton G.729 Annex A, Codng of seeh at 8 kbt/s usng onjugate-struture algebra-ode-extedlnear-redton (CS-ACELP), Nov. 1996. [11] R. O. Duda, P. E. Hart, D. G. Stork, Pattern Classfaton, JOHN WILEY&SON, 2001,. 396-412. [12] L. R. Rabner, R. W. Shafer, Dgtal Proessng of Seeh Sgnals, Prente Hall, 1978. [13] Mushkn, M, Bar-Davd, I, Caaty and odng for the Glbert-Ellot hannels, IEEE Transatons on Informaton Theory, Vol. 35, Nov. 1989, 1277-1290. [14] Vjay K. Varma, Testng Seeh Coders for Usage n Wreless Communatons Systems, Proeedngs of IEEE worksho on Seeh Codng for Teleommunaton, Ot. 1993, 94-94. [15] ITU-T Reommendaton G.191, Software tools for seeh and audo odng standardzaton, Nov. 2000. Insttute (ETRI), Daejeon, Korea. Hs researh nterests nlude Moblty, Wreless LAN, Interworkng Tehnology and VoIP. Won Ryu was born n Busan, South Korea, n 1958. He reeved the B.S. degree n the deartment of omuter sene & statsts from Pusan Natonal Unversty, Busan, South Korea, n 1983, and the M.S. degree n the deartment of omuter sene & statsts from Seoul Natonal Unversty, Seoul, South Korea, n 1988, resetvely, and hs Ph.D. degree n nformaton engneerng from SungKyunKwan Unversty, Kyongg, South Korea, n 2002. Sne 1989, he has been jonng as a researher and urrently he s a rojet leader n the Broadband onvergene Network Interworkng Researh Laboratory, Eletrons and Teleommunatons Researh Insttute (ETRI), Daejeon, Korea. Hs researh nterests nlude All-IP based Convergene Serve & Control Plane, Moblty, NGN, Wre/Wreless Interworkng Tehnology, and VoIP. Byung Sun Lee was born n Seoul, South Korea, n 1956. He reeved the B.S. degree n the deartment of mathemats from SungKyunKwan Unversty, Kyongg, South Korea, n 1980, and the M.S. degree n the deartment of omuter sene from Dongkuk Unversty, Seoul, South Korea, n 1982, resetvely, and hs Ph.D. degree n the deartment of omuter sene from Korea Advaned Insttute of Sene and Tehnology Unversty, Daejeon, South Korea, n 2003. Sne 1982, he has been jonng as a researher and urrently he s a rojet leader n the Broadband onvergene Network Serve Researh Grou, Eletrons and Teleommunatons Researh Insttute (ETRI), Daejeon, Korea. Hs researh nterests nlude All-IP based Convergene Serve & Control Plane, Moblty, NGN, IMS Tehnology, OenAPI Tehnology and VoIP. Seung Ho Han was born n Seoul, South Korea, n 1981. He reeved the B.S. degree n avons engneerng from Hankuk Avaton Unversty, Kyongg, South Korea, n 2003, and the M.S. degree n eletrons engneerng from Informaton and Communatons Unversty (ICU), Daejeon, South Korea, n 2005. He s urrently workng toward Ph.D. degree n shool of engneerng at ICU. Hs researh nterests nlude seeh odng, and VoIP. Jnsul Km was born n Namwon, Chonbuk, South Korea, n 1972. He reeved the B.S. degree n omuter sene from Unversty of Utah, Salt Lake Cty, Utah, USA, n 2002, and the M.S. degree n dgtal meda engneerng from Informaton and Communatons Unversty (ICU), Daejeon, South Korea, n 2005. Sne 2005, he has been workng as a researher n Broadband onvergene Network Interworkng Tehnology Laboratory, Eletrons and Teleommunatons Researh Insttute (ETRI), Daejeon, Korea. Hs researh nterests nlude Moblty, NGN, Multmeda Communaton and Voe/Vdeo over IP. Hyun-Woo Lee was born n Seoul, South Korea, n 1970. He reeved the B.S. degree n eletrons engneerng from Hankuk Avaton Unversty, Kyongg, South Korea, n 1993, and hs M.S. and Ph.D. degree n ommunaton and nformaton engneerng from Hankuk Avaton Unversty, Kyongg, South Korea, n 1995 and 2005. Sne 1995, he has been workng as senor researher, n Broadband onvergene Network Interworkng Laboratory, Eletrons and Teleommunatons Researh Mnsoo Hahn was born n Seoul, South Korea, n 1956. He reeved the B.S. and the M.S. degrees n eletral engneerng from Seoul Natonal Unversty, Seoul, South Korea, n 1979 and 1981, resetvely, and the Ph.D. degree n eletral and eletrons engneerng from Unversty of Florda, Florda, USA, n 1989. From 1982 to 1985, he was wth Korea Researh Insttute of Standards and Sene (KRISS), Deajeon, South Korea. From 1990 to 1997, and he was wth Eletrons and Teleommunatons Researh Insttute (ETRI), Deajeon, South Korea. In 1998, he has been a faulty member of the Shool of Engneerng, Informaton and Communatons Unversty (ICU). Currently, he s a Full Professor n Shool of Engneerng, ICU and a Dretor n Dgtal Meda Laboratory, ICU. Hs researh nterests nlude seeh and audo odng, seeh synthess, nose reduton and VoIP. Reently, he s loated general har for Seeh Engneerng Communty, South Korea.