DHL: Dynamic History Length for Packet Order Recovery in Time-Sensitive Networks

DHL: Dynamic His o y Leng h o Packe O de Reco e y in Time-Sensi i e Ne wo ks How-Hang Liu ∗, Hosein K. Naza i ∗, Tobias Scheine ∗, S e an Senk ∗, Giang T. Nguyen †, F ank H. P. Fi zek ‡ ∗Deu sche Telekom Chai o Communica ion Ne wo ks, TU D esden, Ge many †Chai o Hap ic Communica ion Sys ems, TU D esden, Ge many ‡Cen e o Tac ile In e ne wi h Human-in- he-Loop (CeTI), TU D esden, Ge many E-mails: { i s name.las name}@ u-d esden.de∗†‡ Abs ac —The in eg a ion o 6G and Time-Sensi i e Ne wo k- ing (TSN) is desi ed o p o ide bo h mobili y and de e minis ic communica ion. In TSN, F ame Replica ion and Elimina ion o Reliabili y (FRER) imp o es s eam esilience using edundan pa hs. Howe e , he o iginal FRER wi h a ixed his o y leng h is ine ec i e when acing a iable-delay links ha cause eo de ing and bu s y a i als o packe s. We p opose he Dynamic His o y Leng h (DHL) algo i hm ha dynamically adjus s he FRER elimina ion bu e based on eal- ime link delay measu emen s. We implemen and e alua e DHL in OMNeT++/INET, which p o ides a comp ehensi e and high-accu acy model o TSN swi ches. We compa e DHL wi h he FRER a ia ions. In a 100 ms dual-link simula ion wi h la ency amps and a 10 ms ou age, DHL educes duplica e packe deli e ies by 39 % compa ed o baseline. In e ms o he 99 h pe cen ile in e -a i al in e als, DHL shows an imp o emen o 4.3 ms compa ed o he so ing and shaping algo i hm, allowing i o be e ma ch he sende ’s 1 ms ansmission in e al. Index Te ms—Time-Sensi i e Ne wo king (TSN), F ame Repli- ca ion and Elimina ion o Reliabili y (FRER), OMNeT++ I. INTRODUCTION The ad en o 6G [1] has ans o med he indus ial land- scape, d i ing Indus y 4.0 inno a ions by enabling new use cases and e enue s eams beyond adi ional consume ma - ke s. Eme ging applica ions, such as lexible manu ac u ing and obo ics [2], demand ul a-low la ency, de e minis ic commu- nica ion, and high eliabili y o ensu e seamless ope a ion. To mee hese s ingen equi emen s, he in eg a ion o 6G mobile communica ion wi h TSN [3] has eme ged as a p omising solu ion, combining he mobili y o wi eless 6G ne wo ks wi h he de e minis ic gua an ees o TSN. This syne gy o e s signi ican bene i s o no el use cases bu emains an open esea ch p oblem due o he challenges o main aining end- o- end synch oniza ion and eliabili y ac oss he e ogeneous wi ed and wi eless domains. TSN, o iginally designed o wi ed ne wo ks, p o ides de- e minis ic communica ion h ough s anda ds such as IEEE 802.1CB, which implemen s FRER [4]. FRER enhances e- liabili y by duplica ing ames ac oss edundan links a a duplica ion node and elimina ing duplica es a an elimina ion node, as illus a ed in Fig. 1. This mechanism ensu es obus deli e y in wi ed TSN en i onmen s, such as indus ial con ol sys ems. Howe e , ex ending TSN p inciples o 6G ne wo ks in oduces signi ican challenges. In 6G-TSN scena ios, whe e he 6G ne wo k ac s as a i ual link wi hin he TSN domain, wi eless channels in oduce unp edic able delays and po en ial disconnec ions. These a ia ions can signi ican ly impai he e ec i eness o FRER, as delayed o los ames dis up he elimina ion p ocess, leading o packe loss o unin ended duplica ion ha can deg ade applica ion pe o mance. A common app oach o mi iga e delayed o los ames in TSN sys ems is o employ la ge bu e s a swi ches o main ain an ex ended his o y o sequence numbe s. While his s a egy can imp o e eliabili y by accommoda ing delayed ames, i equi es cos ly ha dwa e upg ades and inc eases p ocessing imes due o la ge his o y leng hs, which may comp omise he s ingen la ency equi emen s o 6G-TSN applica ions. Fu he mo e, he li e a u e highligh s addi ional challenges wi h FRER, including issues wi h ame eo de ing and bu s ansmission unde ce ain condi ions [5]–[7], exace ba ing he complexi y o ensu ing de e minis ic communica ion in 6G- TSN en i onmen s. To ackle hese issues, we p opose a no el FRER elimina ion algo i hm, namely DHL. DHL dynamically adjus s he his o y leng h in he bu e on he ly, only inc eases he leng h when i is absolu ely equi ed. To p ecise adjus men o he his o y leng h DHL con inuously measu e link condi ions ega ding delay. Op ionally, DHL can imco po a e so ing and shaping unc ions a he elimina ion node. These mi iga e ame eo de - ing and bu s ansmission [5]–[7]. We in eg a ed he algo i hm in o OMNeT++ by ex ending a TSN module. This enables e icien 6G-TSN sys em e alua ion. We alida ed i using li e a u e-based scena ios [7]. These included a ying link delays and simula ed link ailu es. Ou esul s show educed duplica ion a es 39 %. They also indica e ewe ou -o -o de ames and less ecei ing ji e wi h he educ ion a es 40 % and 35 %, espec i ely. To acili a e eplica ing he p oposed DHL in u u e e- sea ch we ha e ex ended he open-sou ce oolki in OMNeT++ ©2025 IEEE. Pe sonal use o his ma e ial is pe mi ed. Pe mission om IEEE mus be ob ained o all o he uses, in any cu en o u u e media, including ep in ing/ epublishing his ma e ial o ad e ising o p omo ional pu poses, c ea ing new collec i e wo ks, o esale o edis ibu ion o se e s o lis s, o euse o any copy igh ed componen o his wo k in o he wo ks. Sende Recei e Swi ch 1 Swi ch 2 Sequence Reco e y Func ion (SRF) Sequence Gene a ion Func ion Fas Pa h Slow Pa h 123 1' 2' 3' TPID (2 by es) Rese ed (2 by es) SeqNum (2 by es) R-Tag 1 2 3 Fig. 1. This igu e illus a es he in eg a ion o a nex -gene a ion mobile ne wo k wi h TSN. I highligh s a basic FRER wo k low: he duplica ion node add ags and duplica es each packe using a sequence-gene a ion mechanism, while he elimina ion node applies sequence- eco e y logic o iden i y and disca d any duplica e packe s. and eleased he sou ce code o he esea ch communi y: h ps://gi hub.com/TUDComne sTSN/FRER_so ing_shaping. II. BACKGROUND AND RELATED WORK In his sec ion, we will p o ide a b ie o e iew o he common algo i hms u ilized in he Sequence Reco e y Func- ion (SRF). Nex , we highligh he scena ios ha lead o he c acks in FRER, speci ically ocusing on eo de ing and bu s ansmission. Following ha , we p esen a gene al pe spec i e on he cu en esea ch ocuses ela ed o 5G and FRER, as well as he exis ing implemen a ions o FRER in OMNeT++. As shown in Fig. 1, he sequence gene a ion unc ion a - aches a 6-by e edundancy ag (R- ag) o he 802.1Q heade . The TPID is 0xF1C1 [4], wi h a ese ed ield o u u e use. The las 2 by es hold he sequence numbe , iden ical o duplica e packe s. The SRF can use ei he he Ma ch Reco e y Algo i hm (MRA) o Vec o Reco e y Algo i hm (VRA) o emo e duplica e packe s by sequence numbe . This pape ocuses solely on he MRA; he VRA is ou o scope. The undamen al concep o MRA is illus a ed in Fig. 2. The MRA p ima ily s o es he highes sequence numbe ecei ed and compa es i wi h he nex incoming sequence numbe . I he incoming numbe ma ches he highes sequence numbe , he MRA disca ds he duplica e packe s; o he wise, i o wa ds he packe s as usual [4]. Fo example, as shown in Fig. 2, he sende ansmi s packe s a egula in e als. I is impo an o no e ha po 1 is connec ed o a as e link, while po 2 expe iences g ea e delays. A ime uni 2, he MRA ecei es bo h packe 1 and packe 1’ om he di e en po s, success ully elimina ing he duplica e. Simila ly, a ime uni 8, ano he duplica e elimina ion occu s o he packe wi h sequence numbe 4. Howe e , he MRA is limi ed when packe s expe ience la ge delays. Fo ins ance, a ime uni 5, he highes sequence numbe acknowledged is 4. I packe 2’ a i es la e om he slowe pa h, he MRA will no disca d i , as i is s ill conside ed Sende Ing ess po 1 Ing ess po 2 MRA (Leng h 2) Eg ess po 3 1Time Uni 2 3 4 5 6 7 8 9 10 11 1 2 3 4 5 7 8 9 10 11 1234567 8910 Packe s Los 1' 2' 3' 4' 5' 1 1' 123 Elimina e Duplica e 2 3 4 2' 4 2' 4 3' 3' 4 4' Elimina e Duplica e 7 5' 8 6' 9 7' 7 5' 8 6' 9 7' 10 Bu s Duplica e 6' 7' 6 Fig. 2. Illus a ion o he MRA algo i hm in FRER. Ing ess po s 1 and 2, along wi h Eg ess po 3, co espond o he po s on swi ch 2 depic ed in Fig. 1. Ing ess po s 1 and 2 a e connec ed o he as link and slow link, espec i ely. Eg ess po 3 is a e he ope a ion o MRA and is connec ed o he ecei e . We index o iginal packe s by iand deno e hei duplica es by i′. alid despi e he ac ha packe 2 was al eady accep ed a ime uni 2. A u he challenge a ises o bo h MRA and VRA [5]– [7] when he as link expe iences packe loss and subsequen eco e y, causing alid sequence numbe s o be ecei ed and o wa ded om bo h he as and slow links. As obse ed a ime uni s 9 and 10, a e he SRF, he sequence numbe s o wa ded by he eg ess po a e ou o o de (5’, 8, 6’, 9). In his scena io, he eg ess po o wa ds a bu s o packe s. O iginally, he sende ansmi ed one packe pe ime uni , bu now wo packe s a i e in a single ime uni . This occu s because some packe s (5, 6) a e los on he as link, and hei duplica e packe s (5’, 6’) a e e en ually ecei ed om he ing ess po 2, highe sequence numbe packe s (8, 9) con inue o a i e om he ing ess po 1 simul aneously. As a esul , alid packe s om bo h ing ess po s may be accep ed wi hin he same ime uni . No e ha he bu s and diso de si ua ion o VRA in he igu e did no occu ; howe e , i he his o y leng h eaches 4, hen he bu s issue will a ise. Resea ch on in eg a ing 5G/6G wi h FRER can be g ouped in o wo main di ec ions. The i s di ec ion includes building a eal ha dwa e es bed o a ious use case scena ios, such as au onomous mobile obo s [8], milling p ocess applica ions [9], and oil ield ope a ions in pe ochemical con ex s [10]. The sec- ond di ec ion includes designing scheduling algo i hms o se- lec ing and ou ing a ic ac oss mul iple edundan pa hs. Such algo i hms, o example, can be applied in space– e es ial ne wo ks [11] o in- ehicle ne wo ks [12]. In addi ion o de el- oping es beds and designing scheduling algo i hms, Aijaz [13] has implemen ed a simula ion app oach o e alua e FRER wi h 5G. This esea ch ocuses speci ically on he 5G domain, discussing duplica ion and elimina ion poin s ac oss a ious 5G componen s and p o ocol laye s. Unlike he p e iously men ioned s udies, ou ocus is speci - ically on eco e y algo i hms o FRER. To he bes o ou knowledge, he only s udy add essing issues in FRER eco e y algo i hms is [7], which applies ne wo k calculus on abs ac models o e alua e he wo s -case impac o adding so ing and shaping unc ions a e he elimina ion unc ion. Howe e , he p esen ed app oach is based on a ixed his o y leng h and lacks de ailed implemen a ion guidance. Fu he mo e, i s shaping unc ion elies on ATS, which is a ely implemen ed in comme cial ha dwa e due o i s complexi y. Rega ding he OMNeT++ wi h he FRER implemen a ion, only a ew s udies ha e been conduc ed in his a ea [5], [14], and bo h we e eleased be o e he o icial in oduc ion o FRER unc ionali y in he OMNeT++ INET lib a y. One s udy implemen ed FRER in OMNeT++ wi h con ol-plane capabili- ies [14], while ano he iden i ied issues in FRER p o ocols bu did no add ess bu s y o eo de ed ansmissions [5]. In summa y, his wo k ocuses on he algo i hm in he SRF o educe he accep ance o duplica e packe s. Mo eo e , we implemen ed he so ing and shaping algo i hm a e he SRF o add ess he diso de and bu s ansmission p oblems, based on he mainline elease o he OMNeT++ INET lib a y o FRER. III. DHL – DYNAMIC HISTORY LENGTH The bu e needs o s o e ba ches o packe s o eo de hem and elimina e duplica ed packe s. The bu e ’s his o y leng h has o mee wo design cons ain s. To minimize la ency, he his o y leng h o he bu e has o be minimized. Howe e , sho his o y leng hs a e insu icien o accommoda e packe s om di e en links wi h la ge la ency disc epancies, inc easing he likelihood o duplica e packe s. The DHL in oduces he idea o a iable his o y leng h o he bu e . DHL algo i hm con inuously measu es he la ency ha each packe expe iences om a ious links and adjus s he his o y leng h acco ding o he disc epancies be ween hei la encies. A. Dynamic His o y Leng h As shown in Algo i hm 1, DHL conside s he la ency o measu emen links. When he e a e signi ican la ency di e - ences be ween he links, he likelihood o accep ing a duplica e sequence numbe inc eases. As a esul , we need o ex end he his o y leng h. Con e sely, i he di e ences a e smalle , we can educe he his o y leng h. I his ea u e is enabled, he algo i hm is execu ed e e y τ seconds o upda e he his o y leng h. In line 2, we compu e he obse ed delay di e ence be ween he links, and in line 3, ansla e his span in o he equi ed window size acco ding o he o mula in [5], whe e ˆ Bis he p oposed new leng h o he his o y window W. Line 4 en o ces a minimum capaci y o wo, and lines 5–8 hen adjus he his o y leng h B. I he upda ed ˆ Bis smalle hen B, hen we educe he alue smoo hly by conside ing he a e age o he upda ed ˆ Band p e ious alue Bin line 7. O he wise, line 7 g ows immedia ely by assigning B←ˆ B. Finally, lines 10–11 p une he exis ing sliding window Wdown o he upda ed capaci y by epea edly calling emo eF on (W)un il |W| ≤ B. Algo i hm 1: Dynamic His o y Leng h Inpu : B,W: cu en his o y window o s o ed sequence numbe s, minDelay,maxDelay: obse ed link-delay ex emes, J: Ji e , ∆ : sende ’s in e -packe in e al (ms), τ: ime in e al Ou pu : upda ed window capaci y B, p uned his o y window W 1E e y τseconds ( ime e en ): 2δ←(maxDelay −minDelay) + J// delay span 3ˆ B← ⌈(δ×10−3)/∆ ⌉// needed slo s 4i ˆ B < 2 hen 5ˆ B←2// en o ce min window o 2 6i ˆ B < B hen 7B← ⌊(B+ˆ B)/2⌋// smoo h sh ink 8B←ˆ B// g ow immedia ely 9i |W|> B hen 10 while |W|> B do 11 emo eF on (W) B. In eg a ion o DHL in o OMNeT++ The cu en OMNeT++ INET (4.5.4) FRER implemen s an ex ended MRA, which is a duplica e il e ing mechanism using a sliding window Wo ixed his o y leng h B. Fo each incoming packe , i checks whe he he packe ’s sequence numbe is al eady in he window. I i is, he packe is d opped; o he wise, he sequence numbe is added o he window. I he window exceeds i s size limi , he oldes en y is emo ed. Accep ed packe s a e hen o wa ded. The co e de-duplica ion logic is implemen ed in he S eamMe ge module. We ex end his unc ionali y in he S eamMe ge So e module o suppo adap i e bu e - ing and in-o de packe shaping. Th ee boolean pa ame- e s con ol hese ea u es: dynamicBu e size enables he bu e upda e logic in Algo i hm 1, enableReo de ing ac- i a es he so ing algo i hm, and pe iodicEmission en- ables bo h so ing and shaping. Two iming pa ame e s go - e n he iming beha io : ime In e al (τ) igge s he me ge Time sel -message o bu e upda es du ing ini ializa- ion, while sende T ansmissionIn e al se s he in e al o he eleaseTime , which con ols ansmission iming in he shaping p ocess. Link delays a e measu ed by imes amping each packe a Swi ch1, using he ese ed ield in he R-Tag inse ed by IEEE8021 TagEpdHeade Inse e . These imes amps a e hen ead by S eamMe ge So e a Swi ch2. Du ing each me ge Time e en , handled in he handleMessage unc ion, we un Algo i hm 1 o adjus and p une he his o y leng h. We o e ide he pushPacke unc ion o implemen he so ing algo i hm. Duplica e de ec ion and d opping a e i s TABLE I LINK DELAY PHASES OVER TIME Time (ms) Fas Link Slow Link 0–9 0 ms 0 ms 10–19 0 ms amps 1 →10 ms in 1 ms s eps 20–29 0 ms holds a 10 ms 30–39 0 ms amps 11 →20 ms in 1 ms s eps 40–49 0 ms holds a 20 ms 50–59 down (∞) holds a 20 ms 60–69 0 ms amps 20 →10 ms in 1 ms s eps 70–79 0 ms holds a 10 ms 80–89 0 ms amps 9 →0 ms in 1 ms s eps 90–100 0 ms 0 ms handled by delega ing o S eamMe ge . I enableReo de ing is se o ue, he unc ion ei he bu e s ou -o -o de packe s o o wa ds he nex expec ed packe , inc emen ing i s coun e . A e wa d, he emi InO de unc ion is called o pe o m shap- ing. I pe iodicEmission is ue,emi InO de schedules he eleaseTime o igge a e sende T ansmissionIn e al, pacing packe deli e y o ma ch he sende ’s ansmission a e. IV. EXPERIMENT SETUP This sec ion p esen s he expe imen al se up o DHL, be- ginning wi h he e alua ion scena io, ollowed by he key measu emen me ics and simula ion se ings. A. Simula ion Se up and Link Beha io The simula ion opology adhe es o he s aigh o wa d s uc- u e illus a ed in Fig. 1 wi h wo E he ne links be ween Swi ch1 and Swi ch2. We emula e lossy wi eless links by using wo pa allel E he ne connec ions be ween Swi ch 1 and Swi ch 2, whose la encies and ou ages a e d i en by he Scena ioManage (con igu ed ia an XML ile). Ou es ing scena io consis s o a o al simula ion ime o 100 ms, du ing which he sende ansmi s packe s pe iodically a 1 ms in e als ∆ h ough wo physical links. The condi ions o hese links a e desc ibed in Table I. Fo he as link, he connec ion emains s able bu expe iences a b eakdown o 10 ms in he middle o he simula ion. In con as , he slow link’s la ency inc eases and emains a a pla eau o wo phases un il i eaches 20 ms. A e ha , he la ency dec eases in wo phases un il i ul ima ely eaches ze o. Acco ding o ou design scena io, we se Jequals 10 ms, ep esen ing he empi ically measu ed one-way delay a ia ion obse ed on he link. B. Pe o mance Me ics His o y leng h is conside ed as main pe o mance me ic as i in luences bo h he Duplica e Ra io (Dup) and he Ou - o -O de Ra io (OoO). The Dup measu es he pe cen age o packe s whose sequence numbe s ma ch hose seen ea lie (i.e., duplica es). Meanwhile, he OoO quan i ies he ac ion o ad- jacen packe pai s whose sequence numbe s do no inc emen by exac ly one. In he speci ic case discussed in Sec ion II, he ou -o -o de issue canno be a oided solely h ough he 012345678910 Ji e (ms) 0 20 40 60 Ra io (%) Baseline OoO. Baseline Dup. DHL OoO. DHL Dup. So ing OoO. So ing Dup. Fig. 3. The ou -o -o de a io and duplica e a io a e p esen ed. Since he esul s om he so ing plus shaping algo i hm a e he same as hose om he so ing algo i hm, we did no include hem in his plo . The Ji e Jis only used in he DHL. use o his o y leng h. The e o e, we also equi e an addi ional bu e , e e ed o as he so ing bu e , o ensu e in-sequence ansmission. We will also p esen he leng h o he so ing bu e in ou measu emen s. Finally, he in e -a i al in e al is a c i ical me ic ha highligh s he bene i s o ou shaping ea u e in mi iga ing bu s ansmissions. V. EVALUATION RESULTS In his sec ion, we examine he pe o mance o a ious algo i hms. A. Impac s on In-o de Deli e y In Fig. 3, we p esen he ou -o -o de and duplica e a ios compa ison be ween di e en algo i hms, including baseline, ou p oposed DHL, so ing, and so ing plus shaping. The baseline con igu a ion uses he de aul MRA algo i hm wi h a ixed his o y leng h o 5. The alue o Jin he DHL algo i hm signi ican ly in lu- ences pe o mance, as i di ec ly a ec s he his o y leng h. An inc ease in Jleads o a educ ion in bo h he OoO and he Dup. In con as , he o he algo i hms, which do no u ilize J, main ain cons an a ios. When Jis se o 10 ms, ou p oposed DHL algo i hms demons a e a no able imp o emen in he Dup, educing i om 39 % o 0 % compa ed o he baseline algo i hm. Addi ionally, he ou -o -o de a io dec eases om 61 % o 21 % when compa ed o he baseline. Howe e , when Jis below 5 ms, he pe o mance o he DHL algo i hm is in e io o he baseline. This decline occu s because, du ing he i s 10 ms o he simula ion, he packe delays do no di e , which leads o he DHL algo i hm upda ing he his o y leng h o less han i e. As a esul , he algo i hm s uggles o elimina e duplica ion e ec i ely. The so ing algo- i hm achie es bo h an OoO and Dup o 0 % due o i s s ic acking o sequence numbe s. Since he so ing plus shaping algo i hm yields he same esul s as he so ing me hod, we did no include i in ou compa ison. In addi ion, he so ing and so ing plus shaping algo i hms can u he dec ease bo h he OoO and he Dup o ze o. The p ima y aim o hese algo i hms is o ensu e ha packe s a e 10 210 1100101 In e al (ms) 0.0 0.2 0.4 0.6 0.8 1.0 Empi ical CDF Baseline DHL So ing So ing+Shaping Fig. 4. The empi ical cumula i e dis ibu ion unc ion (ECDF) o in e - a i al in e als a he ecei e is used o compa e he pe o mance o he ou algo i hms. o wa ded in o de . Since bo h he so ing and so ing plus shaping algo i hms yield he same ze o esul s, we ha e only p esen ed he esul s o he so ing algo i hm he e. B. Impac s on La ency Fig. 4 illus a es he packe in e -a i al in e als o a ious algo i hms. The expec ed in e -a i al a e is 1 ms, which highligh s he ad an age o he so ing plus shaping algo i hm, as i main ains mos packe s wi hin he 1 ms in e al. In con as , he o he algo i hms expe ience bu s ansmissions. No ably, he DHL algo i hm ou pe o ms he so ing algo i hm by ansmi ing ewe bu s y packe s. This obse a ion indica es ha app oxima ely 10 % o packe s om he DHL algo i hm and a ound 20 % om he so ing algo i hm ha e in e -a i al in e als o less han 1 ms. To illus a e he ex en o la ge ou lie s, we epo he 99 h pe cen ile (P99) o in e -a i al in e als. This P99 indica es he maximum in e al below which 99 % o all packe s a i e. The measu ed P99 alues a e as ollows: Baseline a 2.0 ms, DHL a 1.3 ms, So ing a 1.6 ms, and So ing wi h Shaping a 5.6 ms. No ably, DHL shows a 35 % educ ion compa ed o he Baseline. Howe e , he e is a 10 ms ail due o a dis up ion in he as link, du ing which he nex alid packe is ecei ed only a e he link esumes. Addi ionally, bo h he so ing and so ing wi h shaping algo i hms exhibi a 20 ms ail because hey bu e ou -o -o de packe s un il he sequence numbe is co ec ly inc emen ed. C. Beha io o His o y Leng h Dynamics Fig. 5 illus a es he link delay ( igh y-axis) de i ed om he S eamMe ge So e module in Swi ch 2. The link delay co esponds o ou con igu a ion ou lined in Table I. The nea ly ze o alues esul om packe s ecei ed o e he as link, while he wo phases o inc easing and dec easing delays o igina e om he slow link. I ’s impo an o no e ha du ing simula ion ime om 60 ms o 70 ms , he e is a g adual educ ion in la ency om 20 ms o 10 ms. This esul s in ea lie packe s expe iencing highe delays, while la e packe s ace lowe delays, wi h bo h a i ing simul aneously a he 80 ms ma k. 0 10 20 30 40 50 60 70 80 90 100 Time (ms) 0 10 20 30 40 Size Link delay (ms) Baseline DHL So ing So ing+Shaping 0 10 20 30 40 La ency (ms) Fig. 5. This igu e displays he link delay ecei ed om he algo i hms, along wi h he his o y leng h om he baseline and DHL algo i hms. Addi ionally, i shows he bu e size o he so ing and so ing plus shaping algo i hms. The baseline his o y leng h is se o 5, and he DHL acks he link delay di e ences wi h an upda e equency o 10 ms. As he la ency o he slow link inc eases, he his o y leng h also inc eases. I is impo an o no e ha be ween 50 ms and 60 ms o he simula ion ime, he as link ails. Consequen ly, he packe s a e hen only coming om he slow links, which leads o a educ ion in he la ency di e ence obse ed in he packe s. As a esul , he DHL dec eases in he ollowing 10 ms, om 60 ms o 70 ms o simula ion ime. Addi ionally, Fig. 5 p o ides in o ma ion abou he bu e size, as bo h he so ing and so ing plus shaping algo i hms c ea e a new bu e o s o e ou -o -o de packe s. S a ing a 60 ms in o he simula ion, bo h algo i hms bu e he ou -o - o de packe s. The so ing algo i hms subsequen ly o wa d hese packe s in a bu s a ound 80 ms. In con as , he so ing plus shaping algo i hm eleases packe s a he same a e as he sende ’s ansmission in e al. A he same ime, i ecei es highe sequence numbe s and s o es he packe s in he bu e . As a esul , he bu e o he so ing plus shaping algo i hm emains s able. In summa y, we can see ha he his o y leng h adap s acco ding o he link la ency and he usage o bu e size in he so ing and so ing plus shaping algo i hms in Fig. 5. D. Impac s on Packe Sequence In Fig. 6, he e olu ion o he sequence numbe o e ime du ing he simula ion is shown. This p o ides insigh s in o how di e en algo i hms handle he sequence numbe , including he elimina ion o duplica es, wai ing o he co ec sequence numbe o a i e, and he smoo h elease o packe s. Since he sende ’s ansmission in e al is 1 ms, he sequence numbe should ideally inc ease by one e e y 1 ms du ing he simula ion. The baseline algo i hm s a s accep ing ou - o -o de packe s a e 20 ms because i s his o y leng h o 5 is insu icien o handle la e-a i ing packe s wi h lowe sequence numbe s. This limi a ion esul s in equen " ee h" pa e ns in he baseline algo i hm. The DHL algo i hm add esses mos o hese ou -o -o de issues, bu be ween 70 ms and 80 ms, i s ill s uggles. This is because o iginal packe s (no duplica es) coming om bo h as 0 10 20 30 40 50 60 70 80 90 100 Time (ms) 0 20 40 60 80 100 Sequence Numbe Baseline DHL So ing So ing+Shaping Fig. 6. This igu e illus a es how he sequence numbe changes o e ime, de ailing he ope a ional beha io s o di e en algo i hms. and slow links need o be accep ed and o wa ded. In oducing a so ing algo i hm allows he sys em o bu e sequence numbe s om 60 ms un il nea ly 80 ms, a which poin i eleases he bu e ed packe s in a bu s once i ecei es he packe s ha a i ed be o e 60 ms in sequence. Simila ly, he so ing plus shaping algo i hm eleases he so ing bu e a an in e al o 1 ms, in line wi h he sende ’s ansmission in e al. This helps mi iga e he bu s ansmission issue. VI. CONCLUSIONS AND FUTURE DIRECTIONS This s udy in oduced DHL, an enhanced FRER elimina ion algo i hm, ha join ly adap s i s his o y leng h sliding window based on li e link-delay measu emen s. Addi ionally, we im- plemen a s a e-o - he-a algo i hm o s ic sequencing ia so ing and shaping. Ou OMNeT++/INET ex ension anspa - en ly augmen s he exis ing S eamMe ge module, enabling on- he- ly his o y leng h esizing, ou -o -o de packe bu e ing, and paced emission wi hou al e ing exis ing FRER p imi i es in INET. In e alua ion scena ios wi h dynamic la ency amps on he slow-links un il 20 ms and a 10 ms as -link ou age, ou DHL alone educed duplica e ames and ou -o -o de ames by 39 % and 40 %, espec i ely. Addi ionally, DHL ou pe o ms he so ing and shaping algo i hm ega ding he 99 h pe cen ile o in e -a i al in e als, which is 1.3 ms compa ed o 5.6 ms. This demons a es he ad an age o DHL in limi ing ji e o he sende ’s 1 millisecond pacing. This wo k pa es he way o explo ing mul i-s eam FRER, ha dwa e-awa e ade-o s, and 6G scheduling in eg a ion [15]. Nex s eps include p o o yping on Sma NICs, adding mul i- low p io i iza ion wi h shaping, and es ing unde ealis ic 6G–FRER a ic [16], [17]. By eleasing ou sou ce-code h ps://gi hub.com/TUDComne sTSN/FRER_so ing_shaping, we aim o suppo and ad ance communi y e o s owa d achie ing de e minis ic, eliable communica ions in he 6G–TSN e a. ACKNOWLEDGMENT This wo k was unded by he Ge man Fede al Minis y o Economic A ai s and Clima e Ac ion (BMWK), p ojec s “TICCTEC” – g an 01MC22007A, “5G-OPERA” – g an 01MJ22008A, and “s ic5G” – g an 01MJ22018C, and by he Ge man Resea ch Founda ion (DFG, Deu sche Fo schungs- gemeinscha ) as pa o Ge many’s Excellence S a egy – EXC 2050/1 – P ojec ID 390696704 – Clus e o Excellence “Cen e o Tac ile In e ne wi h Human-in- he-Loop” (CeTI) o Technische Uni e si ä D esden. REFERENCES [1] P. Schwen eck, G. T. Nguyen, H. Boche, W. Kelle e , and F. H. P. Fi zek, “6G pe spec i e o mobile ne wo k ope a o s, manu ac u e s, and e icals,” IEEE Ne w. Le ., ol. 5, no. 3, pp. 169–172, 2023. [2] H. K. Naza i, J. Abich , S. Senk, H.-H. Liu, T. Scheine , G. T. Nguyen, and F. H. P. Fi zek, “B idging he gap: 5G-TSN in eg a ion o indus ial obo ic communica ion,” in P oc. Eu . Wi eless Con ., 2023, pp. 102–109. [3] K. Zanbou i, M. Noo -A-Rahim, J. John, C. J. S eenan, H. V. Poo , and D. Pesch, “A comp ehensi e su ey o wi eless ime-sensi i e ne wo king (TSN): A chi ec u e, echnologies, applica ions, and open issues,” IEEE Commun. Su eys Tu s., pp. 1–1, 2024. [4] IEEE S anda ds Associa ion, IEEE S anda d o Local and Me opoli an A ea Ne wo ks—F ame Replica ion and Elimina ion o Reliabili y, IEEE S d. IEEE S d 802.1CB-2017, 2017. [5] L. Maile, D. Voi lein, K.-S. Hielsche , and R. Ge man, “Ensu ing eliable and p edic able beha io o IEEE 802.1CB ame eplica ion and elimina ion,” in P oc. IEEE ICC, 2022, pp. 2706–2712. [6] R. Ho mann, B. Nikoli´ c, and R. E ns , “Challenges and limi a ions o IEEE 802.1CB-2017,” IEEE Embed. Sys . Le ., ol. 12, no. 4, pp. 105– 108, 2020. [7] L. Thomas, A. Mi daoui, and J.-Y. Le Boudec, “Wo s -case delay bounds in ime-sensi i e ne wo ks wi h packe eplica ion and elimina ion,” IEEE/ACM T ans. Ne w., ol. 30, no. 6, pp. 2701–2715, 2022. [8] J. Ansa i, T.-s. Hsiao, M. H. Ja a i, B. Va ga, J. Fa kas, I. Moldo án, A. Göppe , and R. H. Schmi , “5G enabled lexible lineless assembly sys ems wi h edge cloud con olled mobile obo s,” in P oc. IEEE PIMRC, 2022, pp. 1419–1424. [9] P. E. Kehl, J. Ansa i, M. Lo in, P. Mohan am, C.-C. E. Liu, J.-L. L. Yeh, and R. H. Schmi , “5G-TSN in eg a ed p o o ype o eliable indus ial communica ion using ame eplica ion and elimina ion o eliabili y,” Elec onics, ol. 14, no. 4, 2025. [Online]. A ailable: h ps://www.mdpi.com/2079-9292/14/4/758 [10] J. Gu, T. Chen, Y. Lu, X. Wu, and R. Wang, “Op imizing ame eplica ion and elimina ion o eliabili y (FRER) p o ocol wi h pigeon- inspi ed op imiza ion algo i hm,” in P oc. ICAIRC, 2024, pp. 842–846. [11] G. Peng, S. Wang, T. Huang, F. Li, K. Zhao, Y. Huang, and Z. Xiong, “Fas s: Enabling aul - ole an and ime-sensi i e scheduling in space- e es ial in eg a ed ne wo ks,” IEEE J. Sel. A eas Commun., ol. 42, no. 12, pp. 3551–3565, 2024. [12] A. A. Syed, S. Ayaz, T. Leinmülle , and M. Chand a, “Faul - ole an dynamic scheduling and ou ing o TSN-based in- ehicle ne wo ks,” in P oc. IEEE VNC, 2021, pp. 72–75. [13] A. Aijaz, “5G eplica es TSN: Ex ending IEEE 802.1CB capabili ies o in eg a ed 5G/TSN sys ems,” in P oc. IEEE CSCN, 2024, pp. 108–112. [14] D. E genç and M. Fische , “Implemen a ion and o ches a ion o IEEE 802.1CB FRER in OMNeT++,” in P oc. IEEE ICC Wo kshops, 2021, pp. 1–6. [15] H. K. Naza i, M. A. Ku , H.-H. Liu, S. Senk, G. T. Nguyen, and F. H. P. Fi zek, “Inc emen al join scheduling and ou ing o 5G-TSN in eg a ion,” in P oc. Eu . Wi eless Con ., 2023, pp. 110–116. [16] H.-H. Liu, S. Senk, M. Ulb ich , H. K. Naza i, T. Scheine , M. Reisslein, G. T. Nguyen, and F. H. P. Fi zek, “Imp o ing TSN simula ion accu acy in OMNeT++: A ha dwa e-aligned app oach,” IEEE Access, ol. 12, pp. 79 937–79 956, 2024. [17] S. Senk, T. Scheine , H. K. Naza i, H.-H. Liu, G. T. Nguyen, and F. H. P. Fi zek, “5G-TSN lex ac: Expe ience a new ouch,” in P oc. IEEE INFOCOM WKSHPS, 2024, pp. 1–3.