Evaluation of High-speed Train Communication Handover Models Based on DEA

Transcription

1 Evaluatio of High-speed rai Commuicatio Hadover Models Based o DEA Yuzhe Zhou State Key Laboratory of Rail raffic Cotrol ad Safety Beiig Jiaotog Uiversity Beiig P.R. Chia @btu.edu.c Bo Ai State Key Laboratory of Rail raffic Cotrol ad Safety Beiig Jiaotog Uiversity Beiig P.R. Chia boai@btu.edu.c Abstract Broadbad commuicatios for high speed trai is becomig a mai tred i high mobility commuicatios. he mai bottleeck of this commuicatio etwork is hadover sice the hadover occurs so frequetly ad the delays are so log that broadbad real-time commuicatio caot apply. Various hadover models have bee developed ad studied recetly. However o comprehesive evaluatio method for these models is employed. o this ed we borrow Data Evelopmet Aalysis (DEA) method to evaluate six typical hadover system models. Hadover models that to be evaluated are itroduced. A brief presetatio of DEA ad its characters is provided. A specific procedure of the evaluatio is proposed. he the results of the evaluatio are obtaied by ruig the DEA. Fially we give our commets ad coclusios to all the hadover models. We hope our work will supply a gap i the system evaluatio area. I. INRODUCION High speed railway is playig a importat role i mass trasportatio all over the world. he developmet of high speed railways makes it more coveiet for people's lives ad works. Meawhile it puts forward various higher requiremets o services of high mobility (i.e. above 300km/h) users. O the oe had iformatio of the trai operatio the system state moitorig eed real-time trasmissio to the cotrol ceter as a real-time dyamic iformatio trasmissio feature of the railway etwork. O the other had with the developmet of social iformatizatio people eed to stay coected to the etwork (especially the Iteret) through a variety of commuicatio termials. Nowadays wireless commuicatio has a wide rage applicatio i high speed rails. Wireless access etworks like satellite GSM/GPRS ad special wireless access etworks are beig used for trai operatio ad passeger applicatio. However these access methods caot meet the curret or future requiremets through the developmet of railways ad commuicatios i.e. the high mobility high data rate real-time ad more reliability. May ew wireless access techologies such as WCDMA WiMAX ad LE etc. have emerged. Each method has its ow features with higher Quality of Service (QoS). However they are ot vastly applied i the railway commuicatios. hese ew access etworks are uder research ad field test for applyig to high speed trai (HS) commuicatios. However which wireless method is the best suitable oe for the high speed rails? Evaluatios of these methods should be provided. Sice the distributio ature of the high speed railway the wireless access etwork for it also has the distributio feature i.e. it covers the whole railway through a lot of cells. herefore the costructio of distributive commuicatio system is a huge proect. he evaluatio of this kid of system is quite differet. May factors may cout for the evaluatio maily the system performace ad the ivestmet. I the view of esurig reliability ad real-time performace of the system it must esure the uderlyig techology to be reliable ad real time at first. I additio these performaces largely rest o the hadover performace. So hadover performace is the bottleeck of the wireless cellular etworks i order to provide QoS for the users ad to support users high mobility. Evaluatio of the system performace ca be obtaied by simply evaluate the hadover model which ivolves oly two adacet cells. For the ivestmet part we ca firstly divide the etwork accordig to its distributio feature ad focus o the evaluatio of oe or two sigle cells ad the we put the evaluatios together. Uder this coditio we ca cosider these mai factors oitly via evaluatig the hadover model. I the evaluatio we cosider the efficiecy factor the reliability factor ad the ecoomic factor oitly. Note that trade-offs are ofte beig made betwee efficiecy ad reliably factors. A evaluatio method which ca combie these factors is eeded. Data Evelopmet Aalysis (DEA) is a ecoomic method to evaluate the cost/reveue of a system. I [1] DEA is used to evaluate the wireless commuicatio sector's cost/reveue i differet coutries. Ad i [2] the authors employ DEA to make the choice of the best wireless access modes amog GPRS CDMA ad SCDMA which is goig to apply i a certai city. he authors i [3] apply DEA to evaluate a Artificial Neural Network model. However these evaluatios maily focus o the ecoomic part ad the results are refereces for maagemet or costructio. Ideed they have give us a revelatio that DEA ca be used to evaluate various factors of various systems. I this paper we chose DEA to oitly evaluate the combiatio of efficiecy reliability ad ecoomic factors. Ad additioally we combie obectivity ad subectivity i the evaluatio i.e. the quatitative ad qualitative aalyses are provided simultaeously. he purpose of this paper is to evaluate the performace ad cost of various hadover models for high speed trai commuicatios ad give firm remarks of these access techics.

2 he rest of this paper is orgaized as follows: Sectio II itroduces the DEA method. Sectio III gives a quick overview of hadover models of high speed railway commuicatio. Relative data of hadover models are gathered ad aalyzed. he a detailed DEA evaluatio of these hadover models is provided i Sectio IV. Fially Sectio V cocludes this paper. II. DAA ENVELOPMEN ANALYSIS Data Evelopmet Aalysis is a data orieted approach for evaluatig the performace of peer etities called Decisio Makig Uits (DMUs) which covert multiple iputs ito multiple outputs. We may wat to kow the overall performace of DMUs by their iputs cosumed ad outputs geerated. It is desirable that we are able to combie various iputs ad outputs ito oe measure. DEA provides this feature. Other alteratives such as aalytic hierarchy process is too complicated ad subective. However sesitivity to selectio of iputs ad outputs ad umber of variables is the mai drawback of DEA. hrough umerical results we fid that this sesitivity issue is egligible via peer compariso. As itroduced by Chares Cooper ad Rhodes i.e. the CCR model [4] there are DMUs to be evaluated. Each DMU has m differet iputs ad s differet outputs. Specifically x of iput i ad produces amout DMU cosumes amout i y of output r. Assume that x 0 r i ad yr 0 ad at least oe positive iput ad oe positive output. he ratio of outputs to iputs is used to measure the relative efficiecy of the DMU DMU to be evaluated relative to the ratios of all of o the = 1 2. We ca iterpret the CCR model as follows: uy0 max v X0 uy ⑴ s. t 1 1 v X u0 v0. where uv are s 1 m 1 weight factor vectors respectively. Y0 X are output ad iput vectors of the evaluated DMU. he 0 LP dual problems of the CCR model are referred as Farrell models ad have the iput orieted form or the output orieted form of: mi or st. X X max 1 Y Y st. X X0 1 Y Y ⑵ ⑶ he optimal values ca be less tha equal to or greater tha 1. Now we are capable to rak the DMUs accordig to their aggregated output to aggregated iput ratios by. I Sectio IV we apply iput orieted ad output orieted forms to evaluate. he result of the evaluatio is to explai the relative efficiecy. A DMU is to be rated as fully (100%) efficiet o the basis of available evidece if ad oly if the performaces of other DMUs do ot show that some of its iputs or outputs ca be improved without worseig some of its other iputs or outputs. he efficiecy of a system cosists of three compoets: techical efficiecy (E) allocative efficiecy (AE) ad cost efficiecy (CE). echical efficiecy reflects the ability of DMUs to obtai maximal output from a give set of iputs. Allocative efficiecy reflects the ability of DMUs to use the iputs i optimal proportios. Cost efficiecy reflects the ability of DMUs to produce a give amout of output with miimum cost. We should poit that the defiitio of a DMU is geeric ad flexible. o allow for applicatios to a wide variety of activities we use the term DMU to refer to ay etity that is to be evaluated i terms of its abilities to covert iputs ito outputs. I the followig part of this paper we use DMU to refer to the hadover model to be evaluated. Accordig to the practical sigificace of the iput ad output variable of each DMU of the DEA method that is a iput value smaller ad a output value bigger beig better which guaratees that the DMU has the highest beefit. Note that there is o limit to use various factors as the iputs ad outputs. he the performace metrics ad costs of hadover models ca be applied to DEA. III. HS COMMUNICAION HANDOVER MODELS Hadover is the mechaism that trasfers a ogoig call from oe cell to aother as a user moves through the coverage area of a cellular system. Hadover model ivolves i structural etwork cofiguratio ad performace metrics. Besides as oe of the most importat fuctioality of a mobile system the hadover procedure eeds to be desiged accordig to the ature of the etwork architecture. herefore differet characterized access etwork architectures determie differet hadover models. I this sectio six wireless access etworks ad related hadover models for high speed trai commuicatio are reviewed ad aalyzed ad related data iformatio is provided for evaluatio i Sectio IV. A. Satellite to rai Satellite commuicatio has large coverage ad ulike the cellular etwork much less groud equipmet ext to the tracks is eeded. Almost o hadover is eeded due to the height of the satellite. However blid spots exist i its coverage area where there are tall buildigs moutais ad tuels. here is a method of applyig the Wireless Local Area Networks (WLAN) as complemet for loopholes of the satellite commuicatio [5]. Yet its cotiuous coverage is much smaller tha we had thought. So hadovers are still required. I additio satellite sigal suffers high losses i bad weather coditios. Satellite commuicatio liks with limited badwidth (typically 4MB) ad high costs (satellite retig cost ad user equipmet cost) caot meet the eeds of a large umber of passegers. What's more satellite commuicatio has a cosiderably roud lik delay (about 4 secods for good) makig it ot suitable for real-time applicatios.

3 B. LCX Leaky coaxial cable (LCX) ca provide a uiform coverage of radio sigals without mutual iterferece. It ca reuse several frequecy poits i the log segmet for HS with high badwidth efficiecy. he typical m-legth LCX [6] is deployed close to the trai so the sigal quality is good ad trasmissio power is low. However striget requiremets o the slot size together with heavy loss of radio sigal which should istall a great umber of repeaters to compesate for trasmissio losses makig the relay high cost. Its etworkig iitial ivestmet cost could be very high. C. RoF Radio over Fiber (RoF) [7] is a recetly ew framework which is very suitable for HS broadbad wireless access ad has a promisig future. Multiple RAUs are located alog the railway ad coected to oe cotrol ceter via a fiber rig. he electrical sigals at the cotrol ceter are coverted ito optical sigals ad trasmitted to RAUs via the fiber. he they are coverted back to electrical sigals ad radiated by the ateas from RAUs. he 60 GHz techology is applied to gai a very broadbad (1GB) compared with curret access techologies. he RoF model is cost effective: oe cotrol ceter ca associate with may RAUs ad these about-100mradius microcells are equivalet ad the ateas are liearradiated so much trasmissio power is saved. Moreover there is o eed for hadovers whe the trai moves from oe RAU to aother uder the same cotrol ceter. hus RoF model ca reduce the umber of hadovers sigificatly. Besides hadover time ca be reduced sice optical switchig time is typically 10-6 s. D. RS-assisted Relay statio (RS) is widely used i commuicatio systems to icrease capacity expad etwork coverage ehace weak-field zoe ad eve have a iformatio gatherig fuctio. he purpose of the relay assisted hadover model is to esure sigal coverage ad stregth to meet the miimum reliable commuicatio requiremets i the overlappig area. I this model [8] the RS is located i the middle of overlappig regio ad associated with the source Base Statio (BS). he RS first act as a repeater ad a diversity gai is achieved. Whe the hadover triggerig coditio is satisfied the RS starts its power cotrol to maitai the coectio of HS with source BS whe the target BS is preparig hadover. After the hadover is completed the RS stops its trasmissio. E. SFN I the Sigle Frequecy Network (SFN) all cells of the cellular system operate at the same frequecy. his is reasoable sice the whole trai ca be treated as a user termial. Assumig that all the BSs i the SFN are sychroized o ew low-layer coectio eeds to establish. A good realizatio is to apply Coordiated Multipoit rasmissio/receptio (CoMP) [9]. I this model two adacet cells form a Cooperative rasmissio Set (CS) where the two enbs are cooperatively workig. Same frequecy is used to sed the data to oe trai withi the CS so soft hadover is available at the cell boudary. CS cotiuously recostructs alog with the trai's movig directio. Oly oe frequecy is used for oe trai so the badwidth ca be much broader. Each trai ca achieve diversity gai ad seamless hadover. F. Dual-soft I the Dual-soft hadover model [10] two ateas are istalled i the frot ad the rear of the HS respectively. Whe the HS is at the cell boudary the frot atea performs hadover to the target BS ad the rear atea is still commuicatig with the source BS so that the coectio ca be maitaied durig the etire hadover procedure. Sice bicastig is the most basic data forwardig scheme to support fast hadover i stadard LE system. he goal of this techique is to elimiate the data forwardig delay betwee the servig BS ad the target BS. Dual-soft hadover model takes advatage of bi-castig ad dual ateas the a soft hadover ca be effectively seamless reducig the hadover delay ad executio overhead. IV. HANDOVER MODES EVALUAION BASED ON DEA he effectiveess of the various wireless access etworks for high speed railway commuicatio a large part of that depeds o the coverage of etworks. I additio the mai problem of the etwork coverage is to move across differet cells without disruptio of commuicatio i.e. the hadover issue. I this Sectio we will focus o applyig DEA to evaluate the effectiveess of the various hadover models preseted i Sectio III. Ad the we use the evaluatio results to ifer the effectiveess of relevat access etworks. I our evaluatio each hadover model is oly beig cosidered of commuicatio betwee groud statios (or satellite) ad HSs typically two adacet cells classic hadover sceario. Cosiderig the coverage of a base statio we ca calculate the umber of base statios which covers a specific legth of railways. Furthermore the effectiveess of the etire coverage ca be calculated. A. Data Acquisitio able I shows the performace metrics ad costs of the six hadover models preseted i Sectio III. All the data is acquired from the cited papers or related materials ad has bee further processed (e.g. trai speed is coverted ito hadover rate). We take six factors for evaluatio. Note that there is o umber or type costrait to the choice of factors. For the readers' coveiece some items eed to be explaied. he hadover rate item here refers to how may secods betwee two hadover triggers. It's calculated by the ratio of two middle poits of adacet overlappig areas to the trai speed. he hadover delay is the time from hadover trigger to hadover completio. he item cost is more flexible sice some of these models have ot bee realized yet we ust give our estimated values. he SNF ad Dual-soft model oly eed slight chage of addig special equipmet to a HS. he RSassisted model eeds oe relay statio betwee two base statios. he RoF model eeds to establish ew RAU to form a microcell. hese values ca be modified if there are more accurate data ad the evaluatio ca be cotiued. It is iterestig to otice that the cost items i able I are a little tricky because they are ot comparable. For example the SFN ad Dual-soft models oly eed ew equipmet o trais; the Satellite model has o base statios; other models have to establish ew statios. Ad each coverage area of BS is ot the same. So we further average the cost values i able I with

4 respect to the coverage areas of each model ad the the averaged cost values i able II is obtaied. ABLE II AVERAGE COSS Modes\Items Coverage per Cell Cost/km Satellite 250km 4 LCX 0.3km 100 RoF 0.1km 50 RS-assisted 4.8km 2 SFN 4.8km 0.2 Dual-soft 1.4km 0.1 B. DEA hrough aalyzig the distributio commuicatio system the evaluatig model for wireless access system icludes three kids of factors: efficiecy reliability ad ecoomic (gree). he efficiecy decides the capacity data rate ad usermobility adaptability of a commuicatio system. his meas the higher data rate the more users it ca hold ad the higher mobility the users are the more efficiet a system is. he reliability determies the success probability seamless access ad robustess of the commuicatio liks. he ecoomic decides the costructio ivestmet ad maiteace cost icludig the power cosumptio of the wireless access system. It is the mai referece to evaluate the implemetatio of the model. I the DEA the six hadover models are regarded as DMUs. Items such as cost trasmissio power ad hadover delay are iputs; Badwidth hadover rate ad success probability are outputs. Because a iput value smaller ad a output value bigger are better. By settig up a liear programmig model a hypothetical sythetic model S is costructed based o the iput ad the output of the six hadover models. A weighted average iput (or output) of the six hadover models is applied as the iput (or output) of the hypothetical sythetic model S. I the costraits of this liear programmig model the output of the sythetic model S must be greater tha or equal to the output of the model A (ay oe of the six hadover models) ad the iput of S must be less tha or equal to the iput of A. If the iput of S is less tha the iput of A the the sythesis model has more output with less iput. herefore model A is viewed to be relatively iefficiet compared with the sythetic model S which meas A ca be cosidered relatively iefficiet tha other hadover models. able III shows the results after ruig this DEA. We calculate both output orieted ad iput orieted form of CCR model. hree assumptios are made: First oly techical metrics are cosidered o cost ivolved this is a very ideal assumptio. Secod cost values i able I are cosidered with techical metrics. hird averaged cost values i able II are cosidered with techical metrics. C. Model Evaluatio We first aalyze the output orieted DEA results. he results are obtaied via solvig the relevat DEA models usig (3) preseted i Sectio II. he lower the factor is the better the output relative efficiecy of the model is. Note that thigs tur out to be differet uder differet assumptios. For the techical oly assumptio Satellite model has a score of 3 which meas that the hypothetical sythetic model S ca use the same iput as the satellite model but has a output of 2 times bigger tha the satellite model. So the satellite model is relatively iefficiet. Same commets to RS-assisted model SFN model ad Dual-soft hadover model. I additio their techical performace is bad. he LCX model ad the RoF model are output relatively efficiet which meas they perform well i covertig iputs to outputs i.e. overall techical metrics are properly achieved. Whe takig the cost ito accout the SFN ad Dual-soft hadover models become relatively efficiet. he RS-assisted ad satellite models are still relatively iefficiet. Ayway the RS-assisted model improves its score cosiderig the cost. Sice the ivestmet of satellite model is so large that it is ot capable of improvig its score. Whe cosiderig the average costs all the six models are efficiet. his meas that i order to apply ew models to substitute the existig oe the average cost i the log ru must be cosidered. Oce the scores are all the same subective evaluatio ca be employed. Oe or several metrics that we are especially iterested i ca be chose to make the fial decisio. For example because we are cosiderig the hadover performace o high mobility sceario the hadover delay overrus other performace metrics. Comparig the hadover delay colum i able I we see that RoF model performs the best so RoF is chose to be the most appropriate hadover model for the high speed trai commuicatios. Now let's see the iput orieted DEA results. he results are obtaied via solvig the relevat DEA models usig (1) preseted i Sectio II. he higher the score (o more tha 1) is the better the iput relative efficiecy of the model is. here are three efficiecy measures which are preseted i Sectio II. he RoF model is still the best model. For the allocative efficiecy i techical oly assumptio SFN model has a better score tha the other four models which meas it has its iputs i better proportios. Ad the satellite model has its iputs i the worst proportios. Note that CE=E AE. Whe it comes to the cost the ivestmets of satellite LCX ad RSassisted models are ot appropriate for their performace i.e. ivestmets are higher ad performace improvemets are smaller. SFN ad Dual-soft models take better advatage of the ivestmets. For the average cost assumptio coclusios may be differet from the output orieted results. LCX ad RSassisted models still have bad AEs idicatig they are ot allocative efficiet while satellite SFN ad Dual-soft models are doig well. After the score aalysis we give our commets o the six models. Satellite has a bad techical efficiecy but a good averaged cost efficiecy which meas that it does ot meet the requiremets of the broadbad high speed trai commuicatios but i the log ru its cost does ot seem uaffordable. he existece of halys i Europe may have proved that satellite model is still alive. LCX model has a higher techical efficiecy but a lower averaged cost efficiecy which idicates that the LCX etwork coverage solutio is ot suitable for the whole lie of HS. Istead it is a alterative to other wireless access methods i tuels or other specific coditios. Japa has applied this model to the Shikase. Besides the existig models the other four models are still uder research or test. RoF performs very well i our evaluatios. Before we argue that it is the most appropriate hadover model for high speed trai commuicatio more accurate cost assessmet must be further made. RS-assisted model ca improve some techical performace metrics however the ivestmet of establishig may ew relay odes will cost too much. It is ot appropriate for the coverage of the whole railway lie. Istead it is a good solutio i weak fields alog the railway. SFN ad Dual-soft do ot improve much of the model performace. Ayway their ivestmets are relatively low compared with the RS-assisted model. So before

5 they are chose to be applied i the high speed rail commuicatios improvemets ad further evaluatios should be made. Note that we ca also apply a dyamic DEA model to evaluate the model efficiecy. Sice the techologies are developig hadover model performace metrics are chagig. Dyamic DEA will give a time varyig evaluatio ad the improvemet will be made oly if a metric is beyod a certai rage of the score i the static DEA evaluatio. his kid of model eeds more specific data ad is more complicated. We hope our work will give some valuable iformatio to our colleagues. V. CONCLUSION I this paper we apply data evelopmet aalysis to evaluate the existig ad ovel hadover models of high speed trai commuicatios. he differece of our work ad the existig system evaluatio methods is that we combie all the system metrics icludig efficiecy reliability ad ecoomic metrics ito oe evaluatio model. he former works maily cosidered the ecoomic metrics such as ivestmet ad maiteace cost to decide whether a system should be established or should be substituted by aother ew system. However our proposed work expads this method to evaluate various techical metrics idepedetly or together with the ecoomic metrics. We maily use this method to evaluate the commuicatio system especially the hadover models uder the high speed railway commuicatio tred. he evaluatio results show that the RoF model is the best appropriate system to support high mobility commuicatios. SFN ad Dual-soft models also perform well. LCX ad RS-assisted models ca be employed i special situatios. Satellite ca still remai existece sice direct liks ca be made to the trai so it would be a stadby system i case of some groud systems were destroyed. Note that the DEA method ca also apply to evaluate other kids of systems or subsystems. It ca also uify various metrics what we are iterested i ito oe evaluatio model. However sice a good evaluatio eeds much more specific ad accurate data our models do have limitatios i view of the data acquisitio. Ayway we hope the methods ad ideas use i our work ca be elighteig ad supply a gap i system evaluatios. ACKNOWLEDGMEN he authors would like to express their great thaks to the support from the 863 Pla of Chia uder Grat 2011AA Beiig Muicipal Natural Sciece Foudatio uder Grat the proect of State Key Lab uder Grat RCS2012Z013 the Key Proect of Chiese Miistry of Educatio uder Grat the Fudametal Research Fuds for the Cetral Uiversities uder Grat 2010JBZ008 ad the proect of State Key Lab uder Grat No. RCS2011ZZ002. REFERENCES [1] A. Azadeh A. Bukhari ad H. Izadbakhsh "A Maagemet based DEA model for Evaluatio of Wireless Commuicatio Sectors" i Idustrial Electroics 2006 IEEE Iteratioal Symposium o 2006 pp [2] W. Zhiqiag Z. Lixi Q. Huaku ad L. Xiaoru "Evaluatio of wireless access modes i distributio etwork commuicatio based o weighted DEA" i Sustaiable Power Geeratio ad Supply SUPERGEN'09. Iteratioal Coferece o 2009 pp [3] C. Wu X. Che ad Y. Yag "Decisio-makig modelig method based o artificial eural etwork ad data evelopmet aalysis" i Geosciece ad Remote Sesig Symposium IGARSS'04. Proceedigs IEEE Iteratioal 2004 pp [4] W. W. Cooper L. M. Seiford ad J. Zhu Data evelopmet aalysis: History models ad iterpretatios: Spriger [5] M. Ha N. Lee K. Ha ad D. Lee Fast IP hadover betwee satellite etworks ad wireless LAN etworks for high-speed trais [6] S. Wezhe W. Hogwei ad B. Big "Feasibility research o traigroud iformatio trasmissio based o leaky coaxial cable i CBC" i Service Operatios Logistics ad Iformatics (SOLI) 2011 IEEE Iteratioal Coferece o 2011 pp [7] B. Laoo D. Colle M. Pickavet ad P. Demeester "Radio-over-Fiber- Based Solutio to Provide Broadbad Iteret Access to rai" IEEE Commuicatios Magazie pp [8] L. Lu X. Fag M. Cheg C. Yag W. Luo ad C. Di "Positioig ad Relay Assisted Robust Hadover Scheme for High Speed Railway" preseted at the 2011 IEEE 73rd Vehicular echology Coferece (VC Sprig) [9] W. Luo R. Zhag ad X. Fag "A CoMP soft hadover scheme for LE systems i high speed railway" EURASIP Joural o Wireless Commuicatios ad Networkig p [10] L. ia J. Li Y. Huag J. Shi ad J. Zhou "Seamless Dual-Lik Hadover Scheme i Broadbad Wireless Commuicatio Systems for High-Speed Rail" Selected Areas i Commuicatios vol. 3 pp ABLE I MODEL DAA O BE EVALUAED Items Cost Chael Badwidth rasmissio Power Hadover Rate Hadover Delay Success Probability Modes (10000 RMB) (MB) (W) (s) (s) Satellite LCX RoF RS-assisted SFN Dual-soft ABLE III DEA RESULS Efficiet Score Output orieted Iput orieted echical Oly Cost Average Cost echical Oly Cost Average Cost Models E AE CE E AE CE E AE CE Satellite LCX RoF RS-assisted SFN Dual-soft