Optimization of Mobile Video Transmission in the High-Speed Rail Environment

Transcription

1 03 8th Iteratioal Coferece o Commuicatios ad Networig i Chia (CHINACOM) Optimizatio of Mobile Video Trasmissio i the High-Speed Rail Eviromet Rui Jiag, Youpig Zhao*, Jixig Li Istitute of Broadbad Wireless Mobile Commuicatios School of Electroics ad Iformatio Egieerig Beijig Jiaotog Uiversity, Beijig, 00044, Chia *Correspodece yozhao@bjtu.edu.c Abstract With the rapid developmet of high-speed rail (HSR) i the world, the demad for providig multimedia services (especially video-related applicatios) to passegers is ever icreasig. The characteristics of radio chael i the HSR eviromet are adverse to high-quality mobile video trasmissio. I this wor, to improve the quality of experiece (QoE) for passegers o the high-speed trai, a cogitive broadbad wireless commuicatio system is proposed. By employig a radio eviromet map (REM)-based cogitive wireless gateway, both high peetratio loss ad frequet group hadover are avoided, which ca sigificatly improve the QoE for mobile video trasmissio i the HSR eviromet. A hadover algorithm aimig to eable reliable video trasmissio ad QoE-drive hadover is proposed. The results show that the proposed hadover algorithm ca gai more utility tha the traditioal algorithm ad also reduce the umber of hadovers. Scearios with equal weights ad uequal weights for passegers are cosidered. Compared to the equal-weight sceario, higher utility ca be obtaied uder the uequal-weight sceario. Keywords Broadbad wireless commuicatio; cogitive wireless gateway; high-speed rail; hadover; quality of experiece (QoE); radio eviromet map (REM) I. INTRODUCTION Compared to the ormal speed of rail traffic, high-speed rail (HSR) is a type of passeger rail trasport that operates sigificatly faster: usually, 00 m/h for upgraded trac ad 50 m/h or faster for ew trac. With the icreasig demad for multimedia services (e.g., video coferece, voice ad video chat, olie games) to passegers ad trai cotrol safety, broadbad wireless commuicatio for HSR attracts more ad more attetio. However, due to the high mobility of highspeed trai, the radio chael is characterized by sigificat Doppler spread ad doubly selective fadig (both i time ad frequet domais) [][]. The uique chael characteristics impose great challeges for commuicatio. Also, because of metal carriages, peetratio loss atteuates the received sigal ad degrades the quality of experiece (QoE) sigificatly [3]. Aother problem is the frequet hadover. This is maily because the trai ca move through a umber of cells i a short time with high speed. As for mobile video trasmissio, multiple parameters have sigificat impact o the video trasmissio quality. I [4] ad [5], the type of video cotet is cosidered. Temporal ad spatial iformatio of the video frames are used for cotet classificatio. I [6], bit error rate is the mai cocer for radio resource allocatio to adjust to QoE requiremet. Badwidth, pacet delay, ad pacet loss rate are discussed i [7]. The variatio of these factors may lead to video quality deterioratio such as mosaic effect, image jitter, frame loss ad so o. Therefore, i the HSR eviromet, the broadbad wireless system has to be carefully desiged so as to support reliable video trasmissio cosiderig all these factors. Various techologies have bee employed for railway wireless commuicatios. Satellite commuicatio is the earliest system that was used for trai commuicatio [8]. However, this system caot support the upli coectio from the trai to a satellite. Ad the dowli coectio will be iterrupted whe the trai moves ito a tuel where sightof-lie path is o loger available. The badwidth is also limited (to oly a few MHz). Leaage coaxial cable is widely used i subway ad tuels for its relatively reliable commuicatio [9]. Its disadvatages are high atteuatio ad high cost of deploymet ad maiteace. GSM-R is the commo commuicatio system for trai operator voice ad low bit rate commuicatio [0]. Most of its features are the same as GSM, except for the added cotrol ad dispatchig capabilities. The limited badwidth of GSM-R system is t suitable for emergig multimedia commuicatios. For WLAN etwors, the limited coverage of access poit lead to frequet hadover ad degraded performace. Therefore, OFDM- MIMO based log term evolutio (LTE) techology has bee selected by the Iteratioal Uio of Railways (UIC) as the ey techology for ext geeratio wireless commuicatio stadard for railway ( LTE-R ). However, differet from applyig LTE to cellular commuicatio etwors, there are some uique features for HSR broadbad wireless commuicatio which should be tae ito cosideratio as metioed above. I additio, the HSR lies usually mae extesive use of viaducts, bridges, ad tuels. As a result, the radio chael characteristics are quite differet from those for cellular systems. A lie-of-sight (LOS) path is usually available due to smaller cell size ad much higher atea istalled o top of the high-speed trai. This wor is supported i part by the NSFC project uder grat No , the Fudametal Research Fuds for the Cetral Uiversities (0JBM03, 03JBZ00), ad the Ph.D. Program Foudatio of Miistry of Educatio of Chia uder grat No IEEE

2 Figure :Architecture of Cogitive Broadbad Wireless Commuicatio System The hadover algorithm is critical to improve the perceived quality of video i the HSR eviromet. Hard hadover is easily coducted. The ey problem is that there is a short brea for commuicatio whe hadover is coducted. This is itolerable for delay-sesitive video applicatios (e.g., video surveillace). Soft hadover ca overcome this shortcomig. How to reduce pacet loss should be carefully cosidered i implemetatio. I additio, there is a uique pheomeo i hadover process for HSR mobile commuicatios group hadover []. Group hadover is occurred whe all users i the trai trigger a hadover simultaeously because of the same movig directio. As a result, there is a hadover sigalig storm i the etwor ad the rate of hadover failure may icrease dramatically. To address the above-metioed problems, a cogitive broadbad wireless commuicatio system is proposed i this paper. I additio, a hadover scheme ad a QoE-drive cogitive hadover algorithm are proposed to improve the QoE for mobile video trasmissio i the HSR eviromet. The remaider of this paper is orgaized as follows: Sectio II describes the architecture of our proposed cogitive broadbad wireless commuicatio system. I sectio III, the hadover scheme ad cogitive QoE-drive hadover algorithm for reliable video trasmissio are discussed. Sectio IV presets some simulatio results to evaluate the proposed hadover algorithm. I Sectio V, we sums up this paper while idetifyig future research directios. II. ARCHITECTURE FOR HSR BROADBAND WIRELESS COMMUNICATIONS SYSTEM Whe the high-speed trai moves alog a fixed HSR lie, its locatio, movig directio, ad chael characteristics alog the trac are predictable to some extet. Thus, to fully exploit the prior owledge ad past experiece, a cogitive broadbad wireless commuicatio system architecture is proposed with the dedicated radio eviromet map (REM) for railway [, ]. REM is a database that ca provide comprehesive multi-domai iformatio such as the chael parameters, available wireless etwors, hadover algorithm ad resource dispatchig policy. As illustrated i Fig., a REM-Based cogitive wireless gateway (CWG) is istalled o the trai to support various types of radio devices (e.g., GSM/WCDMA/LTE smart phoes, WLAN user equipmet, etc.). With the distributed access poits (APs) located iside the carriages of the trai, all commuicatio traffic from/to these radio devices is multiplexed/demultiplexed through the cogitive wireless gateway, just lie a broadbad wireless data pipe. The cogitive wireless gateway ows two trasceivers with their ateas set up o the top of the trai, which are used to coect to the two eighborig enbs while i a hadover process. Oe is used to coect to the servig enb ad the other is for the target enb. Aother importat feature of CWG is REM-eabled casead owledge-based reasoig ad learig (REM-CKL) []. Evirometal iformatio is collected at first. With similarity fuctio (evaluatig the similarity betwee the curret scearios ad the past scearios), the best matched case i the REM is foud. Through performace evaluatio, feedbac ad optimizatio, wireless chael parameters ca be obtaied. As it is ot the focus of this paper, how to obtai the accurate chael parameters alog the HSR is ot detailed here, for iterested readers, please refer to []. I traditioal wireless commuicatio system employed for trai commuicatio, radio devices coect to the trac-side base statio directly. While i the proposed system, CWG, as a relay ode, coects to the base statio istead of radio devices. Thas to it, high peetratio loss is avoided. The much higher eergy efficiecy ad much better radio li quality ca be obtaied. Moreover, the group hadover problem ca also be solved with this architecture. As oly oe (multiplexed data pipe) hadover is eeded whe the trai moves ito the coverage area of two eighborig trac-side base statio, hadover sigalig cogestio i the wireless etwor is avoided. Therefore, less hadover drop rate is expected. Also, Hadover 3

3 ca be completed over a much shorter time as the hadover procedure ca be simplified by avoidig repetitive measuremet ad hadover sigalig. III. HANDOVER SCHEME AND COGNITIVE QOE-DRIVEN HANDOVER ALGORITHM FOR LTE-R A. Hadover Scheme I this subsectio, as i Fig., a hadover scheme is itroduced to realize seamless hadover. The iterfaces related to hadover are X iterface ad S iterface. X iterface is the iterface betwee enbs ad maily used to support mobility. S is the iterface betwee enbs ad mobile maagemet etity (MME, for mobility maagemet) ad system architecture evolutio gateway (S-GW, for pacet trasmissio). Normally, the hadover betwee two enbs is coducted through X iterface istead of S iterface to save sigalig resources ad reduce the effect o the odes i the evolved pacet core etwor. However, X iterface caot support hadover sigalig uder some circumstaces, such as where there is o X iterface betwee the source enb ad the target enb. Thus, to coduct hadover successfully, S iterface ad MME are used. I this wor, hadover through S iterface is discussed. Figure : Procedures of the Proposed Hadover Scheme i HSR Eviromet To esure video trasmissio ad reduce the pacet loss rate, the hadover procedure based o S iterface is detailed as follows: ) Oce CWG receives the sigal from enb (the target enb), it starts to eter ito the REM-CKL process. Eviromet iformatio is iputted ito CWG to gai the best matched chael parameters. Those iformatio icludes the distace betwee CWG ad enb, the distace betwee CWG ad enb, the received sigal stregth from both enbs whe CWG is movig alog the HSR lie, available badwidth ad so o. ) Thas to REM, CWG ca obtai the followig chael parameters at ay positio alog the HSR lie through reasoig ad learig, such as the umber of paths, path loss, ad agle of arrival for each path. Based o our proposed cogitive QoE-Drive hadover algorithm, CWG will ow which locatio is the best poit to trigger a hadover. 3) The CWG coducts a hadover ad seds a hadover request, enb iforms enb for hadover preparatio through S iterfaces ad MME. At the same time, the dowli data to CWG is duplicated ad bi-casted to enb ad enb. It is oted that enb will discard the data because the li betwee CWG ad enb is uavailable right ow. CWG still seds/receives data to/from enb. 4) Whe enb receives HO request commad from MME ad CWG receives HO commad from enb, it starts to coect to enb with oe of its trasceiver while eepig the coectio with enb through aother trasceiver util the li betwee CWG ad enb is set up successfully. 5) At last, CWG seds a message to iform enb of HO completio. enb otifies MME ad enb to release resources. I additio, enb eeds to iform S-GW of shuttig dow bi-castig so that the data will oly be trasmitted to enb. B. Cogitive QoE-Drive Hadover Algorithm I the followig, a cogitive QoE-drive hadover algorithm based o the REM is detailed. Usually, i commo commuicatio systems, oly curret chael coditio is cosidered to trigger a hadover. The pedig variatios of chael ad etwor coditios are igored. Therefore, whether the hadover is good for upgradig the performace is ucertai. Pig-pog hadover is tae as a adverse case for wireless commuicatio. Frequet pig-pog hadover ot oly waste etwor resource, but degrade the perceived quality of video because of sigalig cogestio ad commuicatio failure. Moreover, i most curret scearios, oly oe sigle factor is used to trigger a hadover, such as SNR, RSRQ ad the stregth of received sigal, without cosiderig QoE from all agles. To upgrade the performace of system, the dedicated REM for HSR is used i the proposed algorithm. REM ca provide us prior owledge ad iformatio about the variatios of the system ad wireless eviromet. I our algorithm, which locatio is the best choice where a hadover is triggered so that the system ca obtai the maximum reward is also cosidered. Durig the coherece time, the chael characteristic ca be assumed to be quasi-static. Therefore, the rail i the overlappig coverage zoe ca be divided ito N sectios, mared by d, d,,d N (Note: d ad d N represet the start poit 4

4 ad the last poit of the rail i the overlappig zoe, respectively). Iside each sectio, the system will experiece the same chael characteristics. To evaluate the hadover algorithm s effect o mobile video trasmissio ad the QoE of passegers, MOS (Mea Opiio Score), as a objective stadard, is used. For ay locatio d, the utility fuctio based o MOS for each user m is defied as follows: u m mos m if CWG is ot udergoig hadover at locatio ad mosm mosthreshold if CWG is ot udergoig hadover at locatio ad mosm mosthreshold if CWG is udergoig hadover at locatio where both δ ad δ are pealty factors. δ is used for this case. That is, whe CWG moves ito a deep fadig area, the commuicatio will be iterrupted ad the video experiece of passegers will be extremely bad. Whe cogitive wireless gateway is udergoig hadover, the passegers QoE will be degraded because most etwor resources are used for sigalig ad overheads. δ is used to describe this situatio. MOS is defied by the equatio as follows [4] [5]: mos a a FRa l( SBR) () 3 () a4per a5( PER) where FR is the video frame rate, SBR is the sed bit rate, the coefficiets, a, a, a 3, a 4, a 5, vary with the video cotet type of passegers. The aim of the proposed algorithm is to achieve the maximum utility. Thus, the optimality problem ca be formulated by max M N wm u (3) m m where w m is the ormalized weight of passeger m depedig o his priority level. Telecommuicatio service providers usually assig some passegers with higher priority to esure their QoE i the etwor with limited resources. I this case, passegers with higher priority ca be allocated with better radio resources. M is the umber of passegers o the trai. If all passegers have equal weights, the the optimality fuctio ca be simplified by N max u (4) Thas to the iformatio i REM, the utility fuctio is m replaced by the ew defied reward fuctio r. For i j passeger m, the reward is obtaied whe differet enb is chose as the servig enb from this locatio d i to ext locatio d j. The the fuctio (3) ca be expressed by M m m m m wm r r r r 3 i N (5) m max ( ) Subject to mosm if mos i m mos i threshold a d j i+ ad s s j i if m mosm mos i threshold r i j ad j i+ ad s s j i if mosm mos i threshold ad + N ad s s j N j i h i where N h is the umber of sectios that CWG moves across durig a successful hadover process. Which enb to be coected is specified by s, the state of CWG at locatio d, s { enb, enb}. I fuctio (5) ad coditios (6), the idices i are mootoically icreasig. If d i- is a locatio where hadover is triggered, i = i- + N h. If ot, i = i- +. enb enb Next, we defie two hadover sets H ad H for locatio d whe curret servig enb is enb or enb, to record the hadover locatios from it, but icluded. Also, the accumulative reward R ad R are used. The reward at enb locatio d for all passegers is r enb ' M w r. m m m The procedure of the proposed cogitive hadover algorithm is described as follows: ) Set the first locatio s state s = enb ad the last locatio s state s N = enb. Specify the last locatio s s accumulative reward R N N 0. For other locatios si di,,, N, the accumulative reward is R. Set i = N. ) For = - or = - N h, s { enb, enb}, compute * s s ' R R r. * s 3) If s R R, go to step 4). Otherwise, go to step 5). s * s s 4) Set R R ad empty the hadover set H. s If s s s s s, H { H, d}. Otherwise, H H. 5) If =, go to step 6). Otherwise, decrease by ad retur to step ). enb 6) Fid all the hadover locatios i the hadover set H. enb The accumulative reward R is the optimal solutio. IV. SIMULATION RESULTS OF COGNITIVE QOE-DRIVEN HANDOVER ALGORITHM I this sectio, the simulatio results of the cogitive QoEdrive hadover algorithm are aalyzed. The system parameters are summarized i Table I. With the trai speed ad carrier frequecy, the coherece time ca be obtaied. The the umbers of locatios (L) ca be calculated through overlappig distace divided by coherece time. MOS threshold is set as, below which the experiece of video quality is itolerable. The two factors pealty factors, δ ad δ, ca be adjusted to differet simulatio scearios. I this simulatio, they are set as - ad -0.0, respectively. (6) 5

5 TABLE I: SIMULATION PARAMETERS Parameters Values Carrier frequecy.6 GHz Trai Speed 40 m/h Cell Radius 500 m Overlappig Zoe 500 m Hadover Duratio 00 ms Path Loss Model WINNER Da [3] Pacet Legth 563 bytes Bitrate 00 Mbps Video Frame Rate 30 fps Numbers of users 00 Numbers of Locatios 5000 MOS Threshold Pealty Factor δ - Pealty Factor δ -0.0 As show i Fig. 3, SNR ad PER vary i lie with our expectatio whe CWG moves through the overlappig zoe coectig to two differet adjacet enbs. PER is ot oly i relatio to SNR, but to agle of arrival, umber of paths, delay of each path ad so o. Thus, the variatio of PER is slightly differet from SNR i some areas. The tred observed i this picture is that the margi of two curves of PER is larger tha SNR. Withi the distace from 00 m to 300 m, it is show that the two PERs uder differet enbs rise ad decrease alteratively from 0.0 to 0.4. Two hadover algorithms are compared i Fig.4. The first oe is a traditioal hadover algorithm. Whe the received sigal stregth from the target enb is better tha the servig enb (by a hysteresis margi) ad the situatio lasts for a timeto-trigger (TTT), the hadover ca be triggered. I this simulatio, the hysteresis margi is set as 3 db ad the TTT is set as 0 ms. The secod algorithm is our proposed cogitive QoE-drive hadover algorithm. The algorithms further cosider two differet scearios equal weights ad uequal weights for all passegers, respectively. I the simulatio for uequal-weight sceario, there are three types of passegers. The weights are 0.5, 0.3 ad 0.. Correspodigly, the umbers of the three types are 0, 50 ad 30. From Fig.4, either for equal-weight or uequal-weight sceario, it ca be see that the utility of the latter algorithm is higher compared to the first algorithm withi the distace from 80m to 30m. The reaso is that, oly received SNR is cosidered i the traditioal algorithm. However, i the cogitive QoE-drive hadover algorithm, QoE of each passeger is cosidered ad the best hadover locatios are chose to reach the maximum utility. Apart from this distace, the hadover has t started or the hadover has bee completed. I these areas, for the two hadover algorithms, if the scearios (with equal weights or uequal weights) are the same, the utilities are the same. I additio, for either the traditioal hadover algorithm or the cogitive hadover algorithm, the uequal-weight sceario has better utility compared to the equal-weight sceario. Uder uequal-weight sceario, the higher weight the passeger has, the better radio resource ca be allocated for him, resultig i a higher utility. Also because of its higher weight, higher total utility ca be obtaied. SNR(dB) PER enb enb Distace(m) (a) Distace (m) (b) Figure 3: SNR vs. Distace (a) ad PER vs. Distace (b) i the coverage overlappig zoe of two eighborig enbs Utility Traditioal HO Alg. with Equal Weights Traditioal HO Alg. with Uequal Weights Cogitive HO Alg. with Equal Weights Cogitive HO Alg. with Uequal Weights Distace(m) Figure 4: Utility compariso of two hadover algorithms with equal weights ad uequal weights for passegers enb enb 6

6 HO Requests Traditioal HO Alg. (Equal Weights/ Uequal Weights) Cogitive HO Alg. (Equal Weights) Cogitive HO Alg. (Uequal Weights) 50 5 hadover positios to meet the requiremet of QoE for passegers to the maximum extet with the QoE-drive cogitive hadover algorithm. The performace of the cogitive hadover algorithm is evaluated with the equal weights or uequal weights for passegers. Compared to the traditioal hadover algorithm, the cogitive hadover algorithm ca gai more utility ad reduce the umber of hadover times. Moreover, the simulatio results show that the total utility uder the uequal-weight sceario is higher tha that uder the equal-weight sceario Distace(m) Figure 5: Compariso of hadover locatios of two differet hadover algorithms uder equal-/uequal-weight scearios I Fig. 5, the hadover locatios of the two algorithms are compared. The umber of hadover times of the cogitive hadover algorithm is 3 (uder equal-weight ad uequalweight scearios) whereas the umber for the traditioal hadover algorithm is 5. The umber of hadover times is reduced. This is maily because cogitive hadover algorithm based o utility avoids the cost itroduced by more hadover times. Compared to the traditioal algorithm, the hadover locatios of the cogitive hadover algorithm are more itesive. This is maily because hadover will cause overhead ad decrease the utility. It is preferable to trigger hadover withi the distace where PERs are relatively high ad vary dramatically. Because of the differece of total utility, the hadover locatios differ uder equal-weight ad uequalweight scearios. V. CONCLUSION There is a huge demad for broadbad wireless commuicatios with multimedia services sprigig up. The most commo multimedia services are the video-related applicatios. However, there are a lot of challeges for video trasmissio i the HSR eviromet. Because of the high speed of the trai (50 m/h or eve faster), the wireless chael is characterized by sigificat Doppler spread ad doubly selective fadig. Metal carriages of the high-speed trai also atteuate the received sigal sigificatly. Aother adverse issue it brigs is the frequet group hadover, which deteriorates the performace of commuicatio systems. I this paper, we proposed a cogitive broadbad wireless commuicatio system. A REM-based cogitive wireless gateway (CWG) is set up o the trai to support various types of radio devices. Two trasceivers of the cogitive wireless gateway are employed to esure the reliable trasmissio of pacets durig hadover. Based o this system, a ovel hadover scheme ad a cogitive QoE-drive hadover algorithm are proposed. CWG acquires iformatio about the variatios of wireless etwor ad eviromet from REM. Through reasoig ad learig, the chael parameters alog the HSR trac ca be obtaied. The CWG ca choose the best Based o these results, our future wor will focus o validatig the performace of the proposed commuicatio system via field tests. Moreover, the QoE model of mobile video trasmissio for passegers i the HSR eviromet ad its parameters should be further studied, taig the variatios of the video cotet ad the HSR radio chael coditios ito accout. REFERENCES [] X. Wag, G. Zhu, X. Che, Z. Ta, Doppler Diversity for OFDM High-Speed Mobile Commuicatios, Proc. of IEEE Iteratioal Coferece o Commuicatios, 006, pp , Istabul, Turey. [] Jixig Li, Youpig Zhao, Radio eviromet map-based cogitive Doppler spread compesatio algorithms for high-speed rail broadbad mobile commuicatios, EURASIP Joural o Wireless Commuicatios ad Networig, Volume 0, Issue, pp.-8, DOI: 0.86/ [3] Rui Jiag, Youpig Zhao, Haichao Zhu, Performace Emulatio of Mobile Video Trasmissio i High-Speed Rail Eviromet, i Proceedigs of the th Iteratioal Symposium o Commuicatios ad Iformatio Techologies (ISCIT 0), pp. 7-, Gold Coast, Australia, Oct -5, 0. [4] A. Kha, L. Su, E. C. Ifeachor, Cotet Clusterig Based Video Quality Predictio Model for MPEG4 Video Streamig over Wireless Networ, ICC 09,pp. -5, Jue 009. [5] A. Kha, L. Su, E. A. Jammeh, E. C. Ifeachor, Quality of experiecedrive adaptatio scheme forvideo applicatios over wireless etwors, IET Commuicatios, vol.4, o., pp , July 00. [6] Huilig Zhu, Radio Resource Allocatio for OFDMA Systems i High Speed Eviromets, IEEE Joural o Selected Areas i Commuicatios, vol.30, o. 4, pp , May 0. [7] R.N. Vaz, B.W.M. Kuipers, M.S. Nues, Video quality optimizatio algorithm for video-telephoy over IP etwors, IEEE st Iteratioal Symposium o Persoal Idoor ad Mobile Radio Commuicatios (PIMRC), pp , Sept. 00. [8] V. C. M Leug, Peter Taats, Peter J Garlad, Iteractive multimedia satellite access commuicatios, IEEE Commuicatios Magazie, vol. 4, o. 7, pp , July 003. [9] D. A. Hill ad J. R. Wait, "Electromagetic characteristics of a coaxial cable with periodic slots", IEEE Tras. Electromag. Compat., vol. EMC-, pp , Nov [0] K. Kastell, S. Bug, A. Nazarov, R. Jaoby, Improvmets i Railway Commuicatio via GSM-R, VTC 006-Sprig, pp , May 006. [] Woogsup Lee, Dog-Ho Cho, Group Hadover Scheme Usig Adjusted Delay for Multi-Access Networs, ICC 0, pp.-5, May 00. [] Y. Zhao, B. Le, J. H. Reed, Networ Support The Radio Eviromet Map, (Boo Chapter) Cogitive Radio Techology, Bruce Fette (editor), pp , Elsevier, 006. [3] P. Kyösti, et al, WINNER II D.., WINNER II chael models, [Olie Available] September