In e ne o Things 27 (2024) 101269
A ailable online 27 June 2024
2542-6605/© 2024 The Au ho s. Published by Else ie B.V. This is an open access a icle unde he CC BY-NC-ND license
(h p://c ea i ecommons.o g/licenses/by-nc-nd/4.0/).
Con en s lis s a ailable a ScienceDi ec
In e ne o Things
jou nal homepage: www.else ie .com/loca e/io
Resea ch a icle
C-SLA-MLO: Enhancing SLA Compliance in Indus ial Wi-Fi h ough
Coope a i e Mul ilink Ope a ion
Suneel Kuma a,∗, Daniel Camps-Mu a, Edua d Ga cia-Villegas b
ai2CAT Founda ion, Ba celona, Spain
bUni e si a Poli ecnica de Ca alunya (UPC), Ba celona, Spain
ARTICLE INFO
Keywo ds:
Mul ilink ope a ion
IEEE 802.11be
SLAs
Wi-Fi 7
ABSTRACT
In he dynamic landscape o Indus y 4.0, obus wi eless connec i i y eme ges as a c i ical
enable o Indus ial In e ne o Things (IIoT) applica ions. Wi h he ad en o Wi-Fi 7’s
mul ilink ope a ion (MLO), he demand o schedule s capable o mee ing di e se Se ice Le el
Ag eemen (SLA) equi emen s wi hin indus ial en i onmen s becomes pa amoun . Add essing
his need, his pape in oduces Coope a i e SLA-MLO (C-SLA-MLO), a no el dis ibu ed
mul ilink schedule ha balances da a ac oss MLO channels o ul ill he SLA o STAs ope a ing
wi hin he same BSS. We p o ide a simula ion-based e alua ion whe e we compa e C-SLA-
MLO wi h o he benchma k app oaches, in e ms o SLA compliance when a ying he SLA
cha ac e is ics, he le el o QoS, o backg ound a ic, showcasing an a e age educ ion o
90% in SLA de ia ion in he indus ial use case. This s udy con ibu es p ac ical insigh s
o enhancing wi eless connec i i y in IIoT, o e ing po en ial bene i s o indus ial ne wo k
op imiza ion by exploi ing he MLO ea u e b ough by he new IEEE 802.11be.
1. In oduc ion
Indus ial communica ion echnologies a e unde going a signi ican ans o ma ion o ealize he ision o Indus y 4.0, wi h he
Indus ial In e ne o Things (IIoT) playing a pi o al ole in p o iding scalable, sus ainable, and in elligen solu ions [1,2]. IIoT is
de ined as he in e connec ed ne wo k o in elligen indus ial componen s ha a e s a egically deployed o enhance p oduc ion
e iciency while lowe ing ope a ional cos s. This is achie ed h ough eal- ime moni o ing, e ec i e managemen , and con ol o
indus ial p ocesses, asse s, and ope a ional imelines [3]. IIoT imposes di e se equi emen s including a ailabili y [4], ul a-
low la ency, eliabili y [5], localiza ion accu acy [4], ene gy e iciency, and secu i y [6]. Cu en IIoT communica ions can be
di ided in o hose used o c i ical ope a ions (e.g. sa e y, con ol, and moni o ing applica ions), which a e o en implemen ed
using cus om indus ial E he ne echnologies like P o ine and E he ca [7], and hose used o non-c i ical ope a ions (e.g. ale ing,
supe iso y con ol, and da a-logging asks), ypically implemen ed using Wi-Fi o o he wi eless IoT echnologies [8]. The u u e
ision o IIoT communica ions consis s o ans o ming communica ions o c i ical ope a ions in wo s eps. Fi s , eplacing cus om
indus ial E he ne echnologies wi h s anda dized E he ne Time Sensi i e Ne wo king (TSN) echnology, hus educing cos s.
Second, ex ending E he ne TSN echnology wi h wi eless connec i i y o gain agili y in e-de ining manu ac u ing layou s [9].
Bo h 5GNR and Wi-Fi 8 a e candida e enabling echnologies o achie e he ision o wi eless TSN.
Ex ending E he ne TSN concep s o he wi eless domain poses challenges in e ms o ne wo k a chi ec u e and pe o mance
gua an ees. In e ms o ne wo k a chi ec u e, in eg a ion o IEEE 802.11 and E he ne TSN is s aigh o wa d, as bo h echnologies
∗Co esponding au ho .
E-mail add esses: [email p o ec ed] (S. Kuma ), [email p o ec ed] (D. Camps-Mu ), [email p o ec ed] (E. Ga cia-Villegas).
h ps://doi.o g/10.1016/j.io .2024.101269
Recei ed 17 May 2024; Recei ed in e ised o m 16 June 2024; Accep ed 23 June 2024
In e ne o Things 27 (2024) 101269
2
S. Kuma e al.
a e based on IEEE 802.1 [10]. In he case o 5G, 3GPP has de ined an in eg a ion model based on TSN-T ansla o unc ions, whe eby
a se o Use Equipmen (UEs) and he co e ne wo k beha e as a single E he ne TSN swi ch [11]. The p ima y challenge is o
deli e compa able pe o mance assu ances h ough wi eless echnologies as hose o e ed by E he ne TSN. Fo example, he TSN
amewo k includes Time-Awa e Scheduling (TAS), as de ined in IEEE 802.1Qb [12], which allows bounded end- o-end la encies by
de ining a se o scheduled ga es a each swi ch po . Reliabili y is also enhanced using edundan pa hs, using he F ame Replica ion
and Elimina ion o Reliabili y (FRER) mechanism de ined in IEEE 802-1CB [13]. S udies ha e in es iga ed he implemen a ion o
hese ideas wi hin he con ex o 5G applied o indus y 4.0 [14].
In he case o Wi-Fi hough, gi en i s con en ion-based access and he use o unlicensed spec um, he p og ess owa ds a
de e minis ic la ency and eliabili y has been mo e limi ed. Howe e , a ious Wi-Fi s anda ds ha e been de eloped o add ess
hese challenges. Among hese, Wi-Fi HaLow (IEEE 802.11ah) [15] s ands ou , designed speci ically o he IoT a ena. I p io i izes
low-powe consump ion in long- ange ansmissions in scena ios con aining many connec ed de ices. Al hough i s Res ic ed Access
Window (RAW) mechanism could educe delay a ia ions, ha delay will, in ac , inc ease because s a ions mus wai o hei
assigned window o ansmi . The p ima y goal o he RAW mechanism is o educe ene gy consump ion by allowing s a ions o
s ay in powe -sa ing mode o longe pe iods. In con as , he IEEE 802.11be, ce i ied as Wi-Fi 7, has se he goal o ex emely high
h oughpu , and he ecen ly o med IEEE 802.11bn g oup, he basis o he u u e Wi-Fi 8, has se enhanced ul a-high eliabili y as
i s main goal. Mo i a ed by IEEE 802.11be and bn, he ocus o his pape is o ad ance he s a e-o - he-a o eliable, and (quasi)
de e minis ic Wi-Fi o IIoT applica ions.
Wi-Fi eme ges as a cos -e ec i e solu ion o indoo applica ions, con as ing wi h 5G’s s eng hs in ou doo en i onmen s,
making i well-sui ed o indus ial se ings. The IEEE 802.11be amendmen in oduces mul ilink ope a ion (MLO) as a key ea u e
o enhance pe o mance. No ably, an ad anced ea u e like Mul i AP coo dina ion is pos poned un il he subsequen IEEE 802.11bn
amendmen . As discussed h oughou his pape , MLO plays a pi o al ole in enhancing imeliness o Wi-Fi ansmissions by
simul aneously le e aging mul iple links, he eby educing la ency. Fu he mo e, MLO con ibu es o imp o ed eliabili y by
enabling he ansmission o duplica ed packe s ac oss a ious links, and he agg ega ion o mul iple links esul s in a no ewo hy
boos in o e all h oughpu . These cha ac e is ics collec i ely posi ion he MLO ea u e as a compelling solu ion o add essing he
s ingen equi emen s in indus ial applica ions such as IIoT.
In his wo k, we aim o inc ease eliabili y and educe la ency o indus ial applica ions such as PLCs, SCADA sys ems, obo ic
a ms, indus ial came as, and IoT ga eways ha do no ha e signi ican ene gy cons ain s. IIoT applica ions exhibi di e se Se ice
Le el Ag eemen (SLA) equi emen s o ime-c i ical applica ions. These ange om sa e y applica ions demanding a de e minis ic
delay bound a ound 10 ms (e.g. eme gency ac ions o leak de ec ion), o con ol and moni o ing applica ions wi h mo e lexible
equi emen s spanning om 10 ms o 100 ms [8]. This di e si y o SLAs unde sco es he c i ical necessi y o wi eless s anda ds
capable o accommoda ing he a ied delay- ela ed demands p e alen in indus ial scena ios.
The co e challenge add essed in his pape is o e ec i ely mee hese di e se demands o a ious indus ial SLA lows. MLO
s ands ou as he sui able candida e echnology o achie ing his goal. Speci ically, we aim o de elop a schedule ha explici ly
conside s he SLAs o indus ial lows. As o now, he e exis s no p oposed IEEE 802.11be MLO schedule speci ically ailo ed o
handle he di e se SLAs encoun e ed in indus ial con ex s. Exis ing schedule s o en employ a g eedy app oach, p io i izing minimal
la ency o a single ime-c i ical applica ion. Howe e , his s a egy may no be op imal o indus ial scena ios wi h a ange o SLA
equi emen s. In [16], we p oposed SLA-MLO, he i s SLA-d i en MLO schedule ha dynamically selec s he link based on he
SLA o a pa icula low. This app oach dis inguishes i sel by no adhe ing o a g eedy s a egy and ins ead aims o align wi h he
di e se na u e o IIoT applica ion equi emen s, anging om s ingen and ime-sensi i e o highly lexible ime cons ain s. The
essen ial Key Pe o mance Indica o (KPI), op imized by SLA-MLO, is he measu emen o SLA de ia ion, ep esen ing he ex en o
which a low de ia es om i s p esc ibed limi s, speci ically he delay bound alue.
Building upon his ounda ion, he p esen s udy p esen s a coope a i e SLA-MLO, ha allows STA in he same BSS o exchange
in o ma ion abou he SLA compliance s a us o hei lows, o op imize he esou ces alloca ed o lows su e ing om SLA
de ia ions. C-SLA-MLO is able o u he dec ease SLA iola ions up o 90% in indus ial use cases. The main con ibu ions o
he p esen wo k a e as ollows:
•This wo k p o ides a comp ehensi e axonomy o mul ilink schedule s p oposed o da e, analyzing hei easibili y in speci ic
scena ios.
•This wo k p oposes an ex ension o SLA-MLO [16] called Coope a i e SLA-MLO (C-SLA-MLO), ha u he imp o es he SLA
de ia ion o ime-c i ical indus ial lows by enabling coo dina ion among STAs, he eby alloca ing mo e esou ces o lows
ha a e no mee ing hei SLAs.
•This wo k implemen s C-SLA-MLO in he NS-3 simula o , and benchma k i s pe o mance in e ms o SLA de ia ion agains
SLA-MLO and Leas Conges ion Con ol (LCC) [17].
The es o he pape is o ganized as ollows. In Sec ion 2, we i s discuss he basics and a chi ec u e o MLO, and p esen
di e en mul ilink schedule s p oposed in he pas . In Sec ion 3, we discuss he design o C-SLA-MLO. In Sec ion 4, we p o ide
a comp ehensi e pe o mance e alua ion h ough di e en simula ion se ups ha jus i y ou design choices. Las ly, Sec ion 5
summa izes ou conclusions and indings.
2. Backg ound and ela ed wo k
In his sec ion, we discuss he mo i a ion, basic concep s, and a chi ec u al de ails behind MLO. In a second subsec ion, we
p o ide a ela ed li e a u e e iew.
In e ne o Things 27 (2024) 101269
3
S. Kuma e al.
Fig. 1. Taxonomy o MLO schedule s p oposed in he li e a u e.
2.1. Mul ilink ope a ion
The c owded ISM bands pose limi a ions on wide-channel ansmissions, which impac s he maximum po en ial o echnologies
like Wi-Fi 7 suppo ing up o 320 MHz bandwid h. Ne e heless, h ough s a egic spec um alloca ion ac oss a ious bands, we
can capi alize on oppo uni ies o b oade channels, leading o enhanced h oughpu and a signi ican educ ion in la ency. MLO
plays a c ucial ole in enabling hese ad ancemen s by acili a ing da a ansmission and ecep ion ac oss mul iple adio in e aces,
e ec i ely op imizing spec um u iliza ion ac oss he 2.4 GHz, 5 GHz, and he newly alloca ed 6 GHz bands.
The a chi ec u e o Mul ilink De ices (MLD), is dis inc om legacy de ices, ea u ing wo main ypes: AP-MLD and non-AP
MLD (i.e. clien STAs) ha con ain Uppe and Lowe MAC laye s. The Uppe MAC is uni o m ac oss links, managing asks like
sequence numbe ing, and common unc ions. The Lowe MAC, eplica ed o each a ailable link, handles asks like channel access
and packe ansmission. Each link uns an independen physical laye . The cen al conce n in MLD is op imizing mul ilink usage
wi hin a de ice, pa icula ly h ough scheduling, as discussed in he nex sec ion.
Depending on he capabili ies o he de ice, MLD can be ca ego ized in di e en ways, as discussed in [18].
•Mul ilink Single Radio (MLSR): whe e a de ice is capable o ansmi ing and ecei ing da a o e one link a a ime.
•Enhanced Mul ilink wi h Single Radio (EMLSR): whe e a de ice is capable o ecei ing da a o e mul iple links using di e en
MIMO RF chains, while ansmi ing o e one link a a ime by acking channel a ailabili y ac oss all links.
•Mul ilink Mul i-Radio (MLMR): whe e a de ice con ains mul iple adios and is able o ope a e on mul iple links concu en ly.
MLMR can be u he ca ego ized as:
–Non-Simul aneous T ansmi and Recei e (NSTR): whe e a de ice is capable o ei he ecei ing o ansmi ing on mul iple
links a he same ime, equi ing alignmen o de e al o physical laye p o ocol da a uni s o p e en o e lap and ensu e
smoo h ope a ion.
–Simul aneous T ansmi and Recei e (STR): whe e a de ice is capable o concu en ansmission and ecep ion o da a
asynch onously, bu equi es ca e ul managemen o p e en in e -de ice in e e ence o ensu e smoo h ope a ion.
•Enhanced Mul ilink Mul iple Radio (EMLMR): whe e a de ice ea u es he capabili y o econ igu e MIMO chains on each link
o imp o ed pe o mance and lexibili y.
The i s a ailable MLO-capable Wi-Fi 7 de ices a e expec ed o suppo single- adio MLO, le e aging exis ing Wi-Fi 6 ha dwa e.
Howe e , as Wi-Fi 7 ad ances, new STAs will ha ness he ad an ages o inco po a ing mul iple adios. Besides, in he IIoT domain,
de ices equi ing eliable connec i i y, such as obo s o PLCs, a e o en powe connec ed and a e no esou ce-cons ained. Hence,
we can assume ha MLO will be in oduced h ough MLMR-STR de ices, which suppo simul aneous ansmission and ecep ion
h ough mul iple adios.
The MLO schedule is a c ucial componen in mul ilink de ices, playing a key ole in hei pe o mance. I usually wo ks a
he (Uppe ) MAC laye and con ols he ansmission o packe s among di e en adio in e aces. The design o he MLO schedule
in luences how e ec i e, adap able, and eliable MLDs a e, especially in challenging IIoT en i onmen s.
2.2. MLO schedule s
Recen ly, a ious me hods ha e eme ged o implemen ing mul ilink scheduling wi hin IEEE 802.11be ne wo ks, aimed a
enhancing he la ency, eliabili y, and h oughpu o ime-sensi i e da a. Fig. 1 depic s a axonomy o mul ilink schedule s p oposed
In e ne o Things 27 (2024) 101269
4
S. Kuma e al.
Table 1
O e iew o p oposed schedule s o IEEE 802.11be’s mul i-link ope a ion and hei applicabili y.
Sn. Me hod Scena io TSN applica ion KPIs Type SLA-
Awa eness
1 Load balancing (LB)
[19]
High ne wo k load Cyclic indus ial a ic o
mo ion con ol wi h 100 and
1400 by e packe size and
in e -packe gap 1 o 10 ms
La ency S a ic No
2 Packe duplica ion(PD)
[19]
Low ne wo k load S a ic No
3 Packe spli ing(PS)
[19]
Low ne wo k load S a ic No
4 Hyb id (LB+PS+PD)
based on global
in o ma ion [20]
Medium and dynamic
ne wo k load
Dynamic No
5 Leas conges ion con ol
[17]
Dynamic a ic in
con olled scena io
Gene al a ic lows 2–8
Mbps
Channel
occupancy and
h oughpu
Dynamic No
6 Conges ion awa e load
balancing [17]
Uncon olled and
dense scena ios
Dynamic No
7 Sepa a e da a and
signaling channels [21]
La ge and medium
sized packe wi h
high p io i y
Voice a ic Delay quan ile
and e iciency
S a ic No
8 Alloca ing a dedica ed
channel [21]
Long ansmission S a ic No
9 Spli ing dedica ed
channel in o
sub-channel [21]
Long ansmission
wi h in ensi e and
smalle size packe
S a ic No
10 SLA awa e scheduling
called SLA-MLO [16]
Indus ial use case Indus ial scena io ha ing
s ingen SLAs o delay
bound o 1 ms and lexible
SLAs o up o 10 ms
SLA de ia ion Dynamic Yes
in he li e a u e. Each o hese app oaches is well-sui ed o a speci ic scena io. These schedule s ha e been ca ego ized in o wo
ca ego ies, s a ic and dynamic. In a s a ic app oach, he scheduling mechanism emains ixed h oughou he ne wo k’s li e ime,
ega dless o a ying a ic condi ions. Con e sely, in a dynamic app oach, he scheduling unde goes dynamic changes o e ime.
The s a ic schedule includes load balancing (LB), which dis ibu es load among a ailable links, o educe la ency [19], packe
duplica ion (PD), which duplica es he same packe s on o mul iple links o imp o ing eliabili y [19], packe spli ing (PS), in ol es
b eaking down packe s in o smalle chunks and ansmi ing hem in pa allel ac oss mul iple links, he eby enhancing da a a e [19].
Addi ionally, channel seg ega ion [21], is ano he echnique, which can be ca ego ized acco ding o he channel alloca ion. This
includes alloca ing dis inc signaling channels o indi idual da a channels, designa ing a dedica ed channel speci ically o ime-
c i ical da a, o di iding his dedica ed channel u he in o smalle sub-channels o mo e g anula da a managemen . A dedica ed
channel se es well o long ansmissions wi h s ingen la ency needs, while spli ing i u he p o es bene icial o smalle ,
high- olume a ic. Fo scena ios wi h a la ge numbe o lows ha ing a ying a ic p o iles, a dedica ed esou ce scheme does
no scale well.
On he o he hand, dynamic app oaches adap link-selec ion policies using in o ma ion a ailable a he STA, including bo h global
(ne wo k-wide) in o ma ion, and STA o low-speci ic. Dynamic load balancing (LB) schedule s o mul i- adio access echnology
(RAT) mul i-connec i i y (MC) sys ems a e in oduced in [22,23]. In [20], au ho s p opose a schedule o imp o ing he eliabili y
and la ency o indus ial cyclic a ic by in oducing a hyb id app oach ha selec s links by applying PB, PS, o LB acco ding
o global in o ma ion. In [17], he au ho s compa e di e en policies ha a e upda ed based on a link conges ion me ic. These
policies include leas conges ion con ol (LCC) ha selec s links wi h lesse conges ion, as well as bo h s a ic and dynamic load
balancing app oaches ha ake in o accoun he conges ion le el a each link, namely Conges ion-Awa e Load Balancing (CALB).
The au ho s pe o med simula ions conside ing con olled and uncon olled a ic scena ios. They concluded ha LCC pe o ms well
in a con olled en i onmen , while CALB pe o ms be e in uncon olled and dense scena ios. Howe e , no e ha hese policies
we e ixed o he en i e li e o a low. In [24], he same au ho s ex ended hei p io wo k and sugges ed ha he abo e policies
can be u he imp o ed in e ms o la ency i hey a e adap ed dynamically.
Along wi h hese ca ego ies, machine lea ning echniques ha e also been applied o in es iga e mul ilink scheduling. In [25], he
Mul i-Headed Recu en So -Ac o C i ic (MH-RSAC), a Rein o cemen Lea ning (RL) algo i hm o a ic dis ibu ion, is in oduced.
Compa a i e analysis agains non RL baselines, LCC and LB, highligh s MH-RSAC’s supe io i y, in e ms o Th oughpu D op Ra io.
In [26] au ho s in oduced a Fede a ed Rein o cemen Lea ning (FRL) amewo k o op imizing link ac i a ion in MLO. By enabling
In e ne o Things 27 (2024) 101269
5
S. Kuma e al.
Fig. 2. Ne wo k model conside ed o his wo k. Backg ound STA1 o STA3 ep esen legacy de ices wi h single-link capabili ies, while he p oduc ion line
(SLA-STA1) and collabo a i e obo s (SLA-STA2) a e equipped wi h mul ilink capabili ies, ea u ing SLA lows. The lowe diag am illus a es he packe handling
p ocess a SLA-STAs.
collabo a i e lea ning among neighbo ing APs based on eal- ime pe o mance eedback, i imp o es h oughpu ai ness and
eliabili y in dense deploymen scena ios. Howe e , hose s udies lack ime- ela ed pe o mance igu es.
Despi e he as po en ial o his g oundb eaking echnology in di e se se ings, he e is a no able gap in he li e a u e ega ding
he p ac ical applica ion o IEEE 802.11be’s MLO in indus ial and ac o y-like en i onmen s. Speci ically, he e is a lack o p oposals
ha comp ehensi ely s udy MLO in scena ios in ol ing mul iple lows wi h a ying equi emen s o SLAs, which a e common in
IIoT en i onmen s. Fo ins ance, a collabo a i e obo may equi e a igh delay bound o 5 ms [27], while a human-machine
in e ace (H2M) may ha e a mo e elaxed delay bound o 50–200 ms [28]. The e o e, a di e en app oach is needed o ca e o
he speci ic equi emen s imposed by di e en a ic lows. I is wo h no ing he e ha MLO schedule s p oposed in he li e a u e
do no explici ly de ine an SLA, i.e. hey y o imp o e pe o mance in a bes -e o way. Howe e , in [29], he au ho s emphasize
a ious applica ions, a ic p o iles, and connec i i y equi emen s c i ical om an IIoT pe spec i e. These include isoch onous
lows, in ol ing egula da a ansmission wi h ixed bandwid h and iming equi emen s, and cyclic a ic, cha ac e ized by a
epea ing pa e n o e en s bu wi h a iable in e als. Such a ic pa e ns a e undamen al o modeling IIoT en i onmen s, such
as machine- o-machine and human- o-machine in e ac ions using XR (Ex ended Reali y) [30], o ins ance.
In [16] we p oposed he i s SLA-based schedule called SLA-MLO, which selec s links, based on he SLA o pa icula lows on a
pe -packe basis, using locally a ailable delay s a is ics. I was concluded ha , in indus ial use cases, i.e. p esence o mul iple lows
ha ing di e en SLAs, he SLA de ia ion o SLA-MLO imp o ed by almos 50% when compa ed wi h LCC. Table 1 summa izes he
sui abili y o p oposed schedule s, KPIs o which hey ha e been es ed, and TSN applica ion con igu a ion.
3. Design o C-SLA-MLO
3.1. Ne wo k model
In Fig. 2, we depic he sys em model conside ed in his s udy. The con igu a ion includes wo ypes o IEEE 802.11be STAs and
a mul ilink-enabled Access Poin (AP).
•Backg ound STAs: gene a e non-SLA lows (i.e. non ime c i ical a ic) on legacy Wi-Fi de ices wi h single-link connec ions.
The pu pose o hese STAs is o c ea e backg ound a ic. In Fig. 2, STA1 o STA3 a e backg ound STAs.
•SLA-based STAs: MLD-STAs esponsible o ansmi ing a di e se se 𝐹o lows, each deno ed as 𝑓∈𝐅, and associa ed wi h
hei speci ic SLA (such as delay bound and allowed pe cen age o SLA b each). In Fig. 2, p oduc ion line and collabo a i e
obo s a e using MLD-STAs o ansmi ing da a lows o he MLD-AP
The lowe segmen o Fig. 2 p o ides insigh in o he in e nal a chi ec u e o an MLD de ice. The ope a ional sequence ini ia es
wi h he ecep ion o packe s a he MAC laye o he MLD-STA coming om he uppe laye s. Following his, packe s a e classi ied

In e ne o Things 27 (2024) 101269
6
S. Kuma e al.
based on hei associa ed low 𝑓∈𝐅. Unde he in luence o he speci ic SLA linked o ha low, he schedule s ee s he packe
owa ds a designa ed link 𝑖∈𝐋wi h a p obabili y deno ed as 𝑃𝑓,𝑖. The p ima y objec i e o he SLA-based schedule is o dynamically
and adap i ely de e mine he alues o 𝑃𝑓,𝑖 o each low 𝑓and link 𝑖. I is impo an o no e ha all MLD STAs depic ed in Fig. 2
a e assumed o be STR-capable. Consequen ly, Ca ie Sense Mul iple Access wi h Collision A oidance (CSMA/CA) ules a e applied
independen ly in all enabled links. In his wo k, we conside uplink a ic.
3.2. SLA de ini ion
In he ealm o IIoT applica ions, he synch oniza ion o da a and con ol signal lows wi h he manu ac u ing p ocess’s cycle
ime s ands as a c i ical conside a ion. The e iciency and pe o mance o he p ocess a e signi ican ly impac ed by la ency, wi h
cycle imes anging widely — om less han 1 ms o ield bus communica ions o se e al hund ed milliseconds o con olle s
communica ing wi h edge de ices [28]. We de ine SLAs as ollows:
Delay Th eshold (𝐷_𝑇 𝐻𝑓): maximum pe missible delay om when a packe belonging o low 𝑓is ecei ed om he ing ess
queue un il i ecei es a success ul acknowledgmen .
In en i onmen s cha ac e ized by he use o unlicensed spec um and con en ion-based access (e.g. Wi-Fi ne wo ks), de ining SLAs
encoun e s signi ican obs acles. As a esul , an exclusi e eliance on s ingen delay bounds o SLA de ini ion p o es imp ac ical
in his con ex . Hence, we added ano he concep in SLA, de ined as ollows:
E o Th eshold (𝐸𝑟𝑟𝑜𝑟_𝑇 𝐻𝑓): pe cen age o packe s in low 𝑓 ha a e pe mi ed o exceed he delay bound 𝐷_𝑇 𝐻𝑓. This
includes bo h packe s ha expe ience a delay abo e he delay bound, and packe s ha a e los .
The objec i e o he schedule s is hen o ensu e ha he pe cen age o packe s in low 𝑓exceeding 𝐷_𝑇 𝐻𝑓 emains below
𝐸𝑟𝑟𝑜𝑟_𝑇 𝐻𝑓wi hin measu emen window 𝑇𝑆𝐿𝐴.
3.3. Wo king o SLA-MLO and coope a i e SLA-MLO
3.3.1. SLA-MLO
In [16], we p oposed SLA-MLO, designed o MLD-STAs, ope a es on a pe - low, pe -packe basis o op imize communica ion
in alignmen wi h p ede ined SLAs. When a packe is gene a ed a an STA, he algo i hm calcula es link p obabili ies (𝑃𝑓,𝑖). Two
c i ical pa ame e s, he delay bound and SLA b each, guide decision-making. I a low 𝑓adhe es o he delay bound (𝐷_𝑇 𝐻𝑓)and
s ays below he allowed SLA b each le el (𝐸𝑟𝑟𝑜𝑟_𝑇 𝐻𝑓), all links ecei e he same p obabili y. Con e sely, i he SLA b each is abo e
he h eshold alue, links showing a delay wi hin he delay bounds a e p io i ized. Howe e , i none o he links adhe es o he
delay bound and he SLA b each exceeds he h eshold, he p obabili y 𝑃𝑓,𝑖 becomes a unc ion o link delay, ensu ing adap abili y
o di e se ne wo k condi ions and minimizing he impac o conges ion.
Algo i hm 1ou lines he ope a ion o he MLD-STA schedule o a speci ic low 𝑓. Inpu pa ame e s include he SLA de ini ion
o low 𝑓deno ed by (𝐷_𝑇 𝐻𝑓, 𝐸𝑟𝑟𝑜𝑟_𝑇 𝐻𝑓). In e nal a iables a e main ained o each low, including STA delay (𝑆𝑇 𝐴_𝐷𝑒𝑙𝑎𝑦𝑓),
which deno es he o al du a ion o ansmi ing a packe and ecei ing an acknowledgmen , he link delay (𝐷𝑒𝑙𝑎𝑦𝑓,𝑖), expe ienced
by packe s o low 𝑓in link 𝑖, Exponen ially Weigh ed Mo ing A e age (EWMA) o link delays (𝐴𝑣𝑔𝐷𝑒𝑙𝑎𝑦_𝐿𝑖𝑛𝑘𝑓,𝑖), and SLA b each
me ic (𝑆𝐿𝐴_𝐵𝑟𝑒𝑎𝑐ℎ𝑓). The algo i hm ou pu s p obabili ies 𝑃𝑓,𝑖 o ansmi ing he nex packe o low 𝑓o e link 𝑖.
The algo i hm begins wi h he Main Schedule Loop() (line 35) when a packe a i es om he uppe laye . Be o e queuing,
p obabili ies o each link a e calcula ed using P ocedu e 3 (line 22), in ol ing 𝑆𝐿𝐴_𝐵𝑟𝑒𝑎𝑐ℎ𝑓compu a ion om P ocedu e 1. In
P ocedu e 1 (line 1), o calcula e he SLA b each, bo h 𝑆𝑇 𝐴_𝐷𝑒𝑙𝑎𝑦𝑓and 𝐷𝑒𝑙𝑎𝑦𝑓,𝑖 a e measu ed i s . Those delays a e upda ed
when he acknowledgmen o he las ansmi ed packe is ecei ed. The a e age link delay (𝐴𝑣𝑔𝐷𝑒𝑙𝑎𝑦_𝐿𝑖𝑛𝑘𝑓,𝑖) is calcula ed
using an Exponen ially Weigh ed Mo ing A e age (EWMA) il e wi h 𝛼= 0.8(line 4). Two a iabls, namely SLA_No Followed and
SLA_Followed a e upda ed o ack he numbe o packe s ha ei he exceed o mee he delay h eshold 𝐷_𝑇 𝐻𝑓, espec i ely. These
coun s a e hen used o calcula e he SLA b each (𝑆𝐿𝐴_𝐵𝑟𝑒𝑎𝑐ℎ𝑓) me ic (line 10).
P ocedu e 3 hen calcula es p obabili ies based on wo s a es:
•S a e 1: i he SLA b each is below 𝐸𝑟𝑟𝑜𝑟_𝑇 𝐻𝑓, all links ecei e equal p obabili ies. This is achie ed by calling he
𝐴𝑙𝑔𝑜𝑟𝑖𝑡ℎ𝑚𝑆𝑒𝑙𝑒𝑐𝑡𝑖𝑜𝑛() unc ion. In case o SLA-MLO, he 𝐵𝑒𝑎𝑐𝑜𝑛_𝑆𝐿𝐴_𝐵𝑟𝑒𝑎𝑐ℎ lag is always alse, hence, all links a e assigned
equal p obabili ies (line 18).
•S a e 2: I SLA b each is abo e 𝐸𝑟𝑟𝑜𝑟_𝑇 𝐻𝑓(line 27 o 31).
–Case 1: assigns equal p obabili ies o links s ill ha ing delay below 𝐷_𝑇 𝐻𝑓as shown in 𝑃 𝑟𝑜𝑐𝑒𝑑𝑢𝑟𝑒3(), line 27.
–Case 2: assigns p obabili ies in e sely p opo ional o link’s a e age delay, as shown in 𝑃 𝑟𝑜𝑐𝑒𝑑𝑢𝑟𝑒3(), line 30.
A e p obabili y calcula ion, he Main Schedule Loop() gene a es a andom numbe 𝑅𝑛(line 39). I he numbe is less han
he cumula i e p obabili y, he packe is o wa ded o he link’s ou bound queue, p io i izing highe p obabili y links (line 41 o
line 45). The algo i hm’s compu a ional complexi y is gene ally 𝑂(𝐿), bu gi en ha he numbe o links L is ypically 2 o 3 in
eal-wo ld implemen a ions, his can be conside ed O(1) in p ac ical scena ios. Hence, we a gue ha ou pe -packe scheduling is
easible e en wi h he high h oughpu expec ed in Wi-Fi 7 de ices.
In summa y, SLA-MLO dynamically adap s o ne wo k condi ions, p io i izing SLA compliance and a o ing links wi h lowe
delays du ing b eaches. I o e s a lexible app oach o eal- ime a ic dis ibu ion based on ne wo k s a e.
In e ne o Things 27 (2024) 101269
7
S. Kuma e al.
Algo i hm 1 Algo i hm o he SLA-MLO and C-SLA-MLO schedule . Decision a an MLD-STA o low 𝑓.
1: p ocedu e 1: Upon Recei ing Ack ⊳In oked when ACK is ecei ed
2: Ge Ack o he las success ul packe
3: Measu e 𝑆𝑇 𝐴_𝐷𝑒𝑙𝑎𝑦𝑓&𝐷𝑒𝑙𝑎𝑦𝑓,𝑖
4:
𝐴𝑣𝑔𝐷𝑒𝑙𝑎𝑦_𝐿𝑖𝑛𝑘𝑓 ,𝑖 ←𝐴𝑣𝑔𝐷𝑒𝑙𝑎𝑦_𝐿𝑖𝑛𝑘𝑓,𝑖 ⋅𝛼
+𝐷𝑒𝑙𝑎𝑦𝑓,𝑖 ⋅(1 − 𝛼)⊳Calcula e EWMA o delay o each link
5: i 𝑆𝑇 𝐴_𝐷𝑒𝑙𝑎𝑦𝑓≤𝐷_𝑇 𝐻𝑓 hen
6: 𝑆𝐿𝐴_𝐹 𝑜𝑙𝑙𝑜𝑤𝑒𝑑 ←𝑆𝐿𝐴_𝐹 𝑜𝑙𝑙𝑜𝑤𝑒𝑑 + 1
7: else
8: 𝑆𝐿𝐴_𝑁𝑜𝑡𝐹 𝑜𝑙𝑙𝑜𝑤𝑒𝑑 ←𝑆𝐿𝐴_𝑁𝑜𝑡𝐹 𝑜𝑙𝑙𝑜𝑤𝑒𝑑 + 1
9: end i
10: 𝑆𝐿𝐴_𝐵𝑟𝑒𝑎𝑐ℎ𝑓←𝑆𝐿𝐴_𝑁𝑜𝑡𝐹 𝑜𝑙𝑙𝑜𝑤𝑒𝑑
𝑆𝐿𝐴_𝐹 𝑜𝑙𝑙𝑜𝑤𝑒𝑑+𝑆𝐿𝐴_𝑁𝑜𝑡𝐹 𝑜𝑙𝑙𝑜𝑤𝑒𝑑 ⊳SLA B each calcula ion o each low
11: end p ocedu e
12:
13: p ocedu e 2: Algo i hm Selec ion ⊳Selec ing be ween SLA-MLO and C-SLA-MLO Algo i hms
14: i 𝐁𝐞𝐚𝐜𝐨𝐧_𝐒𝐋𝐀_𝐁𝐫𝐞𝐚𝐜𝐡 =𝑡𝑟𝑢𝑒 hen ⊳The lag is ue when Algo i hm is C-SLA-MLO
15: 𝑘=a gmax𝑖∈𝐿{𝐴𝑣𝑔𝐷𝑒𝑙𝑎𝑦_𝐿𝑖𝑛𝑘𝑓 ,𝑖}⊳Finding link ha ing maximum delay/la ency
16: 𝑃𝑓,𝑘 = 1
17: else
18: 𝑃𝑓,𝑖 ←1
|𝐋|
19: end i
20: end p ocedu e
21:
22: p ocedu e 3: Calcula e P obabili y ⊳Calcula e p obabili y o each link o low
23: 𝐋𝐛𝐞𝐥𝐨𝐰 ←{𝑖∈𝐋|𝐷𝑒𝑙𝑎𝑦𝑓,𝑖 < 𝐷_𝑇 𝐻𝑓}⊳Links ha ing delay less han delay h eshold alue
24: 𝐋𝐚𝐛𝐨𝐯𝐞 ←{𝑖∈𝐋|𝐷𝑒𝑙𝑎𝑦𝑓,𝑖 ≥𝐷_𝑇 𝐻𝑓}⊳Links ha ing delay g ea e han delay h eshold alue
25: i 𝑆𝐿𝐴_𝐵𝑟𝑒𝑎𝑐ℎ𝑓≤𝐸𝑟𝑟𝑜𝑟_𝑇 𝐻𝑓 hen
26: 𝐴𝑙𝑔𝑜𝑟𝑖𝑡ℎ𝑚𝑆𝑒𝑙𝑒𝑐𝑡𝑖𝑜𝑛()
27: else i |𝐋𝐛𝐞𝐥𝐨𝐰|≠0 hen
28: 𝑃𝑓,(𝑖|𝑖∈𝐿𝑎𝑏𝑜𝑣𝑒)←0⊳Assigning ze o p obabili y o link abo e delay h eshold
29: 𝑃𝑓,(𝑖|𝑖∈𝐿𝑏𝑒𝑙𝑜𝑤)
←1
|𝐿𝑏𝑒𝑙𝑜𝑤|⊳Assigning equal p obabili ies o links below delay h eshold
30: else
31: 𝑃𝑓,(𝑖|𝑖∈𝐿)←
1∕ 
𝐴𝑣𝑔𝐷𝑒𝑙𝑎𝑦_𝐿𝑖𝑛𝑘𝑓,𝑖
∑𝑖∈𝐿
1∕𝐴𝑣𝑔𝐷𝑒𝑙𝑎𝑦_𝐿𝑖𝑛𝑘𝑓,𝑖
⊳No malized link p obabili y based on espec i e delays
32: end i
33: end p ocedu e
34:
35: p ocedu e Main Schedule Loop ⊳The s a poin o Algo i hm
36: while ue do
37: Wai o packe assigned o low 𝑓
38: 𝑃𝑓,𝑖 ←𝐶𝑎𝑙𝑐𝑢𝑙𝑎𝑡𝑒𝑃 𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦()
39: 𝑅𝑛 ←𝑢𝑛𝑖𝑓𝑜𝑟𝑚(0,1) ⊳Random numbe c ea ion
40: 𝑝𝑟𝑜𝑏 ←0⊳Ini ializing p ob a iable wi h ze o
41: o 𝑖∈𝐋do
42: 𝑝𝑟𝑜𝑏 ←𝑝𝑟𝑜𝑏 +𝑃𝑓,𝑖
43: i 𝑅𝑛 < 𝑝𝑟𝑜𝑏 hen ⊳Checking andom numbe s wi h cumula i e p obabili y
44: o wa ds he packe o link 𝑖
45: end i
46: end o
47: end while
48: end p ocedu e
In e ne o Things 27 (2024) 101269
8
S. Kuma e al.
Table 2
O e iew o gene al simula ion pa ame e s.
Fea u e Simula ion pa ame e
F equency band 6 GHz (channel numbe 15 & 47)
Channel wid h 160 MHz
S anda d IEEE 802.11ax
Agg ega ion Mode OFF
MCS Mins el
Backg ound STAs h oughpu 1500 B/ms
Simula ion ime 30 s (pe simula ion)
Packe queue size 500p (pe link)
T a ic ype Uplink
P opaga ion model Log dis ance p opaga ion loss model
3.3.2. Coope a i e SLA-MLO
We ha e enhanced SLA-MLO by in oducing modi ica ions in he algo i hm, esul ing in C-SLA-MLO. When a leas one MLD
STA is he ansmi e o a low ailing o mee i s SLA, i no i ies he o he MLD-capable STAs. Said no i ica ion could be
accomplished in di e en ways, by le e aging di e en p ocedu es ou lined in he IEEE s anda d [31], such as he T ansmi
S eam/Ca ego y Measu emen epo o Mul icas Diagnos ics epo . To elabo a e, when an MLD STA de ec s ha one o i s lows
is no mee ing he SLA, i in o ms he AP using a igge ed epo , by se ing a ce ain lag. When SLA iola ion anishes, i again
no i ies he AP by means o ano he igge ed epo . Upon ecei ing hese no i ica ions, he MLD-capable AP would igge a lag
(𝐵𝑒𝑎𝑐𝑜𝑛_𝑆𝐿𝐴_𝐵𝑟𝑒𝑎𝑐ℎ) in he beacon, indica ing ha a leas one STA is ailing o mee i s SLAs. All MLD STAs moni o he beacon
o be no i ied abou he s a us o o he STAs. I he 𝐵𝑒𝑎𝑐𝑜𝑛_𝑆𝐿𝐴_𝐵𝑟𝑒𝑎𝑐ℎ is se in his beacon ield, hen all MLD STAs ac i a e C-
SLA-MLO. The key modi ica ion can be ound in p ocedu e 2 o Algo i hm 1. In SLA-MLO, when 𝑆𝐿𝐴_𝐵𝑟𝑒𝑎𝑐ℎ𝑓is below he allowed
h eshold, 𝐵𝑒𝑎𝑐𝑜𝑛_𝑆𝐿𝐴_𝐵𝑟𝑒𝑎𝑐ℎ is se as alse and hence all links ecei e he same p obabili y (line 18). Howe e , in C-SLA-MLO, as
discussed ea lie , 𝐵𝑒𝑎𝑐𝑜𝑛_𝑆𝐿𝐴_𝐵𝑟𝑒𝑎𝑐ℎ is se o be 𝑡𝑟𝑢𝑒 by AP, and o all lows ha a e adhe ing hei SLAs, he link expe iencing
he highes conges ion is selec ed wi h p obabili y 1 (line 14 o 16). This adjus men aims o imp o e SLA-MLO’s pe o mance by
di ec ing mo e lexible a ic lows o links suppo ing mo e conges ion, he eby c ea ing addi ional space o lows wi h s ingen
equi emen s. I is wo h men ioning he e ha he complexi y o Algo i hm 1 emains unchanged, as C-SLA-MLO in p ocedu e 2
in ol es inding a link wi h a maximum delay ha has a compu a ion complexi y o 𝑂(𝐿).
The undamen al dis inc ion be ween SLA-MLO and C-SLA-MLO a ises when a ne wo k encoun e s si ua ions whe e lows wi h
s ingen equi emen s ail o mee hei SLAs, while lexible lows adhe e o hei speci ied SLAs (i.e. delay bounds and SLA b each
a e below he h eshold). In such ins ances, SLA-MLO uni o mly alloca es lows ha ing lexible SLA ac oss all links. In con as ,
C-SLA-MLO akes a mo e s a egic app oach by di ec ing his a ic speci ically o conges ed links. This s a egic adjus men aims
o op imize he o e all pe o mance by enabling less conges ed links o handle mo e a ic om lows wi h s ingen SLAs. On he
o he hand, C-SLA-MLO also equi es an es ablished no i ica ion mechanism o igge he solida i y o lexible lows, whe eas in
SLA-MLO, STAs ope a e au onomously wi hou equi ing any signaling exchange.
4. Pe o mance e alua ion
To e alua e he pe o mance o di e en mul i-link schedule s, we used NS-3 (V3.36) [32], a packe -le el simula o . This e sion
o NS-3 does no ully implemen he mul ilink ope a ion ea u e in oduced in Wi-Fi 7 (as pe IEEE 802.11be), so we modi ied i
o inco po a e his capabili y. Speci ically, adap a ions we e made a he MAC le el o NS-3 o accommoda e MLO. The MAC-le el
implemen a ion o NS-3 is s uc u ed in o uppe MAC and lowe MAC laye s. The uppe MAC encompasses common unc ionali ies
such as sequence numbe assignmen , ac i e p obing, and associa ion- ela ed mechanisms. The lowe MAC le el consis s o h ee
modules:
1. Channel Access Manage (CAM): his module manages channel access, including Dis ibu ed Coo dina ion Func ion (DCF)
and Enhanced Dis ibu ed Channel Access (EDCA).
2. F ame Exchange Manage (FEM): esponsible o handling IEEE 802.11 amendmen -speci ic ame exchange sequences.
3. TXOP (T ansmission Oppo uni y): along wi h i s subclass QOS-TXOP, i manages queueing unc ionali ies.
Fo he implemen a ion o he MLO ea u e in NS-3, sepa a e channel and ame exchange manage s pe link a e necessa y. A uni ied
𝑇 𝑋𝑂𝑃 wi h dis inc queues o each MLO link was cons uc ed. When he Uppe MAC decides o send a packe o a link, based
on he schedule decision, he 𝑇 𝑋𝑂𝑃 assigns he co esponding packe o he co esponding queue. A e placing he packe in he
designa ed queue, he co esponding 𝐹 𝐸𝑀 and 𝐶𝐴𝑀 a e in oked. Following he execu ion o lowe MAC unc ionali ies, he 𝐹 𝐸𝑀
assigns he ame o a pa icula Wi-Fi Phy (Wi-Fi physical laye en i y).
4.1. Gene al simula ion se up
Unless speci ied o he wise, he simula ed scena ios p esen ed in his sec ion sha e he ollowing cha ac e is ics. Each scena io
con ains 𝑁numbe o STAs and 1 AP. To make he simula ion mo e ealis ic, nodes we e andomly deployed wi hin a 15-me e
In e ne o Things 27 (2024) 101269
9
S. Kuma e al.
adius o he AP in e e y se up, and by de aul , MLD nodes included wo adios in he 6 GHz band. In IIoT en i onmen s, o he
bands a e used o non-c i ical ope a ions, which is why we ocus on he g een ield 6 GHz band o p o ide gua an eed pe o mance.
The e a e wo ypes o STAs, as discussed in he ne wo k model depic ed in Fig. 2, i.e. legacy de ices ansmi ing non-SLA lows
and MLD-STAs ansmi ing lows wi h speci ic SLAs.
To ep esen he he e ogenei y o SLAs ac oss lows in IIoT en i onmen s, wo ypes o MLD STAs we e employed, each wi h
dis inc SLA equi emen s.
•The 𝑆𝑡𝑟𝑖𝑛𝑔𝑒𝑛𝑡 ype has a s ic delay bound, below 5 ms, used o example in isoch onous lows o con ol loops [29], and a
5% o SLA b each h eshold (STA1).
•The 𝐹 𝑙𝑒𝑥𝑖𝑏𝑙𝑒 ype has a mo e elaxed delay bound o up o 50 ms, used o example in moni o ing applica ions o
indus y [8,29], and a 10% o SLA b each h eshold.
Each scena io includes one s ingen STA alongside mul iple backg ounds and lexible STAs. All simula ions adhe ed o he sys em
model ou lined in Fig. 2. Each simula ion has been un o a leas 10 di e en seeds.
To benchma k he pe o mance o C-SLA-MLO and SLA-MLO, we use a dynamic app oach o conges ion con ol, known as Leas
Conges ion Con ol (LCC), p oposed in [17]. This me hod keeps ack o he conges ion le els o selec he leas conges ed link.
Key Pe o mance Indica o (KPI): based on he SLA de ini ion p o ided in Sec ion 3.2, we e alua e he le el o SLA compliance
o a low in e ms o SLA-de ia ion. The calcula ion o SLA-de ia ion is ou lined as ollows:
Fo each SLA low 𝑓, we measu e he pe cen age o packe s wi h a delay su passing 𝐷_𝑇 𝐻𝑓a egula in e als o 𝑡_𝑠𝑎𝑚𝑝𝑙𝑒,
esul ing in a ime- a ying signal 𝐸𝑟𝑟𝑜𝑟_𝑓(𝑡). To assess SLA compliance, we compu e he mo ing a e age o 𝐸𝑟𝑟𝑜𝑟_𝑓(𝑡)wi hin a
window o du a ion 𝑇𝑆𝐿𝐴 using a con olu ion ope a ion:
A g_e o _ (𝑡) = E o _ (𝑡) ∗ (1
TSLA )Sq(𝑡)(1)
He e, 𝑆𝑞(𝑡) ep esen s a squa e signal wi h a du a ion o 𝑇𝑆𝐿𝐴 seconds. Subsequen ly, he SLA de ia ion o low 𝑓is compu ed
as:
SLA_de _ ( ) = max(A g_e o _ ( ) − E o _TH ,0) (2)
I is impo an o emphasize ha he me ic 𝑆𝐿𝐴_𝑑𝑒𝑣_𝑓(𝑡)only inc eases when low 𝑓su passes i s SLA’s e o h eshold. I he
low emains below he e o h eshold, he me ic emains unchanged, as he e a e no disce nible bene i s om he applica ion
pe spec i e.
4.2. E alua ion scena ios
We s uc u e ou pe o mance e alua ion in six di e en scena ios ha answe he ollowing esea ch ques ions.
•Scena io 1: in he i s scena io, we show he dynamics o p oposed schedule s.
•Scena io 2: he second scena io s udies how inc easing he numbe o lexible lows in luences he pe o mance o s ingen
lows unde di e en schedule s.
•Scena io 3: in he hi d scena io, we s udy he e ec o SLA he e ogenei y, on he pe o mance o he schedule s. The goal is
o s udy how sensi i e SLA-MLO and C-SLA-MLO gains a e, o he di e ence be ween he s ic and he elaxed SLAs.
•Scena io 4: he ou h scena io in ol es he e alua ion o di e en QoS se ings. The goal is o s udy he e ec o MLO
scheduling combined wi h EDCA p io i iza ion.
•Scena io 5: in he i h scena io, we in oduce a hi d link on MLD de ices. The mo i a ion behind his expe imen is o check
how MLO schedule s beha e when we inc ease he numbe o links.
•Scena io 6: In he inal scena io, we ep oduce ealis ic indus ial scena io ea u ing 6 indus ial lows and andom backg ound
a ic. To subs an ia e ou asse ion ega ding he e ec i eness o a mul ilink schedule ailo ed o IIoT en i onmen , i is
impe a i e o conduc a pe o mance e alua ion wi hin a se ing closely esembling a p ac ical IIoT en i onmen .
Table 2 shows he gene al pa ame e s o he simula ion se up.
4.2.1. Scena io 1: Unde s anding he dynamics o SLA-MLO & C-SLA-MLO
This expe imen is in ended o show he beha io o bo h SLA-MLO and C-SLA-MLO unde di e en a ic condi ions.
Simula ion se up: One s ingen low wi h 0.750 ms o delay bound and an e o h eshold o 5% and 5 lexible lows wi h 50 ms
o delay bound and 10% o e o h eshold. SLA lows a e ansmi ing 64B packe s e e y 2.5 ms, i.e. 204.8 kbps, and backg ound
STAs a e ansmi ing a he a e o one 1500B packe pe ms, i.e. 12 Mbps.
Discussion:inFig. 3, we plo he ou pu o Algo i hm 1i.e. p obabili ies o links o s ingen lows (Fig. 3(a) o C-SLA-MLO
and Fig. 3(b) o SLA-MLO) and o lexible lows (Fig. 3(d)). Fig. 3(c) shows he SLA b each o he s ingen lows. The simula ion
is di ided in o 3 phases.
•Du ing he i s 10 s o simula ion, he e is no backg ound a ic on ei he link, esul ing in nea -ze o SLA b each o bo h
s ingen and lexible lows in bo h schedule s. Consequen ly, p obabili ies o selec ing link 1 o link 2 in he s ingen low
a e 50% o bo h links (Algo i hm 1, line 18).
In e ne o Things 27 (2024) 101269
16
S. Kuma e al.
Table 6
Scena io 5 simula ion se up.
(a) Flow se ings in 3 Links se up
Flow # E o h eshold Delay Bound
1 5% 2 ms
Relaxed lows (2 onwa ds) 10% 50 ms
(b) Gene al se ings
Fea u e Simula ion pa ame e
SLA lows Da a a e 64 B/5 ms
Backg ound STAs a io a s a 10:10:5 (Link1:Link2:Link3)
S ingen Vs Relaxed Flows a s a 1:10 (S ingen : Flexible)
F equency Band 6 GHz (channel numbe 15, 47 & 79)
Access Ca ego y o SLA Flows AC_VI - CW(15,63)
Access Ca ego y o Backg ound lows De aul (AC_BK)
Table 7
Scena io 6 simula ion se up.
(a) FLow se ings o indus ial scena io
Flow # E o h eshold Delay Bound
1 5% 1 ms
2 o 6 10% 10 ms
(b) Gene al se ings
Fea u e Simula ion pa ame e s
F equency band 6 GHz & 5 GHz
Channel wid h 160 MHz & 20 MHz
Ra e Con ol Algo i hm Mins el
Flow1 Mo ion Con ol (64 B/1 ms) [33]
Flow 2 o 6 Close loop Con ol (64 B/10 ms) [33]
Backg ound low Videos a ic (1500 B packe wi h da a a e o 40 and
46.5 Mbps in o al on link1 and link2) [33]
Backg ound STAs in S a 5:9 (Link1:Link2)
S ingen Vs Relaxed Flows 1:6 (S ingen : Flexible)
4.2.6. Scena io 6: Realis ic indus ial scena io
In he p e ious scena ios, we ha e conside ed a i icial a ic pa e ns ha ha e allowed us o be e unde s and he ac o s
go e ning he pe o mance o he di e en MLO schedule s unde s udy. Howe e , in an indus ial o ac o y-like scena io, mul iple
machines wi h di e en a ic pa e ns and QoS equi emen s need o in e ac seamlessly o ensu e smoo h and e icien ope a ions.
The objec i e o his expe imen is o mimic a ealis ic indus ial scena io by simula ing eal a ic models, da a a es, and
in e e ence. This expe imen is speci ically ailo ed o assess he e ec i eness o he p oposed schedule in an indus ial con ex .
Simula ion Se up: in Table 7, we summa ize mos ele an pa ame e s. We ha e chosen six indus ial lows, whe e low 1
ep esen s mo ion con ol a ic and all o he s ep esen closed-loop con ol a ic ( h ough mobile-connec ed I/O ga eways), as
p oposed in [33]. I is no ed he e ha low 1 p esen s he mos s ingen SLA. In eal indus ial scena ios, backg ound a ic can be
pe iodic o ape iodic. Hence, we andomly assigned backg ound a ic on and o ime, i.e. each backg ound STA will s a and s op
a backg ound low andomly, bu a a ixed packe in e -a i al ime. To es SLA de ia ion, we inc eased he numbe o backg ound
STAs while keeping Link 1 sligh ly less conges ed han Link 2 ( 40 Mbps and 46 Mbps, espec i ely). On a e age, we ensu ed ha
backg ound lows las o a leas 18 s ou o he 30 s o simula ion ime.
Discussion: The SLA de ia ion o he mo ion con ol (s ingen ) low in an indus ial scena io is depic ed in Fig. 9. No ably,
SLA-MLO and C-SLA-MLO exhibi ema kable pe o mance compa ed o LCC in scena ios wi h dynamic backg ound a ic in he
indus ial se ing. Ini ially, he imp o emen is subs an ial, wi h C-SLA-MLO showing an a e age imp o emen o 73% compa ed o
LCC and 48% compa ed o SLA-MLO. Addi ionally, SLA-MLO demons a es a 60% imp o emen compa ed o LCC.
The inc ease in he numbe o backg ound STA leads o an inc ease in SLA de ia ion o all SLA lows. As a consequence, elaxed
lows (closed loop con ol) s a ailing o mee hei SLAs hus mo ing packe s o he as e link, lea ing less oom o s ingen
low ansmissions. In such ins ances, he imp o emen o C-SLA-MLO diminishes o 36% compa ed o LCC and 17% compa ed o
SLA-MLO. Fo SLA-MLO, he imp o emen dec eases o 22% compa ed o LCC.
The imp o emen in SLA de ia ion can be unde s ood by looking a he low pe cen age g aph in Fig. 9(b). Ini ially, o C-SLA-
MLO, he a e age pe cen age o elaxed lows h ough he conges ed link is high and dec eases wi h an inc ease in backg ound
STAs. The numbe o s ingen lows passing h ough he conges ed link is ini ially low bu g adually inc eases o e ime, when he
elaxed lows mo e away om he conges ed link. A simila end is obse ed o SLA-MLO, whe e he pe cen age o lexible lows

In e ne o Things 27 (2024) 101269
17
S. Kuma e al.
is a ound 50% on a e age on he conges ed link ini ially, bu his pe cen age dec eases a e eaching a poin o 12–16 backg ound
STAs. I is wo h no ing ha he e SLA de ia ion o lexible lows is 5% o 10% o C-SLA-MLO and up o 2.5% o 7% o SLA-MLO.
In summa y, hese esul s demons a e he abili y o C-SLA-MLO o be e p o ec ime-sensi i e lows in ealis ic indus ial
scena ios. Howe e , all schedule s su e when he amoun o backg ound load in he channel exceeds he channel capaci y.
Admission con ol, limi ing he numbe o backg ound lows, would be equi ed o limi SLA de ia ion in all si ua ions.
5. Conclusion
In indus ial In e ne o Things (IIoT) scena ios, ensu ing SLA compliance is pa amoun o main aining ope a ional e iciency and
eliabili y, especially conside ing he di e se se o lows wi h a ying SLA equi emen s p e alen in indus ial en i onmen s. In his
s udy, we in oduced C-SLA-MLO, a coope a i e mul ilink scheduling algo i hm aimed a enhancing SLA adhe ence in indus ial Wi-
Fi ne wo ks le e aging he new MLO ea u e in oduced in IEEE 802.11be. Th ough igo ous e alua ion, we demons a ed signi ican
pe o mance imp o emen s, wi h C-SLA-MLO achie ing up o a 90% enhancemen in SLA compliance compa ed o o he app oaches
in indus ial use cases.
Ou assessmen encompassed six scena ios, each ailo ed o add ess speci ic esea ch ques ions and assess he e ec i eness
o C-SLA-MLO in di e se ne wo k condi ions. In Scena io 1, we showcased he dynamic adap abili y o he p oposed schedule ,
highligh ing i s abili y o p io i ize lows based on SLA equi emen s. Scena io 2 explo ed he e ec o augmen ing he numbe
o lexible lows on s ingen low pe o mance, esul ing in an imp o emen o up o 50% in SLA de ia ion. Meanwhile, Scena io
3 del ed in o SLA he e ogenei y and i s impac on schedule gains, wi h enhancemen s o up o 35% in SLA de ia ion obse ed.
Fu he mo e, Scena io 4 sc u inized he e alua ion o di e se Quali y o Se ice (QoS) se ings, showing he combined e ec o MLO
scheduling wi h EDCA p io i iza ion o op imize pe o mance. Expanding ou analysis, Scena io 5 in oduced a hi d link on MLD
de ices o examine schedule beha io in he con ex o inc eased link di e si y, leading o a 30% imp o emen in SLA de ia ion.
Finally, in Scena io 6, we simula ed a ealis ic indus ial en i onmen ea u ing mul iple indus ial lows and andom backg ound
a ic, ea i ming he e ec i eness o C-SLA-MLO in p ac ical IIoT se ings. In his case up o 90% o imp o emen in SLA de ia ion
was obse ed.
As o u u e wo k, we en ision ex ending C-SLA-MLO o mul i-AP scena ios o add ess la ge -scale indus ial ne wo ks and u -
he enhance i s adap abili y o e ol ing ne wo k dynamics. Addi ionally, explo ing op imiza ion echniques o esou ce alloca ion
and a ic managemen could o e addi ional insigh s in o imp o ing SLA compliance in complex indus ial en i onmen s.
CRediT au ho ship con ibu ion s a emen
Suneel Kuma : W i ing – o iginal d a , Visualiza ion, Valida ion, So wa e, Resou ces, Me hodology, In es iga ion, Fo mal
analysis, Da a cu a ion. Daniel Camps-Mu : W i ing – e iew & edi ing, W i ing – o iginal d a , Valida ion, Supe ision, Resou ces,
Me hodology, In es iga ion, Funding acquisi ion, Fo mal analysis, Concep ualiza ion. Edua d Ga cia-Villegas: W i ing – e iew
& edi ing, W i ing – o iginal d a , Valida ion, Supe ision, Resou ces, Me hodology, In es iga ion, Funding acquisi ion, Fo mal
analysis, Concep ualiza ion.
Decla a ion o compe ing in e es
The au ho s decla e ha hey ha e no known compe ing inancial in e es s o pe sonal ela ionships ha could ha e appea ed
o in luence he wo k epo ed in his pape .
Da a a ailabili y
No da a was used o he esea ch desc ibed in he a icle.
Acknowledgmen s
This wo k is suppo ed by he EU’s H2020 5GSma Fac p ojec unde he MSCA, Spain g an ag eemen ID 956670. This wo k
was also pa ly suppo ed by he Spanish he MCIN/AEI, Spain/ 10.13039/501100011050 h ough p ojec PID2019-106808RA-I00
and MINECO, Spain and EU PRTR UNICO I+D numbe TSI-063000-2021-15-6GSMART-EZ.
Disclous e ins u ions
Du ing he p epa a ion o his wo k, he au ho s used Cha GPT 3.5 o enhance he eadabili y o ce ain pa ag aphs wi hin his
pape . A e using his ool, he au ho s e iewed and edi ed he con en as needed and ook ull esponsibili y o he con en o
he publica ion.
In e ne o Things 27 (2024) 101269
18
S. Kuma e al.
Re e ences
[1] S.N. Anbalaga, M. Schwa z, R. Bem huis, P. Ha inga, Assessing ac o y’s Indus y 4.0 eadiness: A p ac ical me hod o IIoT senso and ne wo k analysis,
P ocedia Compu . Sci. 232 (2024) 2730–2739.
[2] Z. Fa ima, A.U. Rehman, R. Hussain, S. Ka im, M. Shaki , K.A. Soom o, A.A. Lagha i, Mobile c owdsensing wi h ene gy e iciency o con ol oad conges ion
in in e ne cloud o ehicles: a e iew, Mul imedia Tools Appl. 83 (18) (2024) 53949–53974.
[3] W.Z. Khan, M.H. Rehman, H.M. Zango i, M.K. A zal, N. A mi, K. Salah, Indus ial in e ne o hings: Recen ad ances enabling echnologies and open
challenges, Compu . Elec . Eng. 81 (2020).
[4] J. Mus a a, K is ian O. Sands , Niclas E icsson, L. Riz ano ic, Analyzing a ailabili y and QoS o se ice-o ien ed cloud o indus ial IoT applica ions, in:
2019 24 h IEEE In e na ional Con e ence on Eme ging Technologies and Fac o y Au oma ion, ETFA, IEEE, 2019, pp. 1403–1406.
[5] Z. Ma, M. Xiao, Y. Xiao, Z. Pang, H.V Poo , B. Vuce ic, High- eliabili y and low-la ency wi eless communica ion o in e ne o hings: Challenges,
undamen als, and enabling echnologies, IEEE In e ne Things J. 6 (5) (2019) 7946–7970.
[6] R. Nazi , A. Lagha i, K. Kuma , S. Da id, M. Ali, Su ey on wi eless ne wo k secu i y, A ch. Compu . Me hods Eng. (2021) 1–20.
[7] X. Wu, L. Xie, End- o-end delay e alua ion o indus ial au oma ion sys ems based on E he CAT, in: 2017 IEEE 42nd Con e ence on Local Compu e
Ne wo ks, LCN, IEEE, 2017.
[8] E. A e xe, O. Ba ambones, I. Cal o, P. Fe nández-Bus aman e, I. Ma in, J. U alde, Wi eless echnologies o Indus y 4.0 applica ions, IEEE T ans. Ind.
In o m. 16 (3) (2023) 1349.
[9] M. Noo -A-Rahim, J. John, F. Fi yaguna, H.H.R. She azi, S. Kushch, A. Vijayan, E. O’Connell, D. Pesch, B. O’Flynn, W. O’B ien, e al., Wi eless
communica ions o sma manu ac u ing and indus ial IoT: Exis ing echnologies, 5G and beyond, Senso s 23 (2023) 73.
[10] A. Adame, M. Ca ascosa-Zamacois, B. Bellal a, Time-sensi i e ne wo king in IEEE 802.11be: On he way o low-la ency Wi-Fi 7, Senso s 21 (2021) 4954.
[11] L. Ma en o m elde, A. Neumann, L. Wisniewski, L. Sch eckenbe g, Co-con igu a ion o 5G and TSN enabling end- o-end quali y o se ice in indus ial
communica ions, 2021, h ps://openda a.uni-halle.de//handle/1981185920/41505.
[12] Inc. Ins i u e o Elec ical and Elec onics Enginee s, O icial websi e o he 802.1 ime-sensi i e ne wo king ask g oup, 2016, URL h p://www.ieee802.
o g/1/pages/ sn.h ml. (Accessed 01 Janua y 2024).
[13] IEEE s anda d o local and me opoli an a ea ne wo ks– ame eplica ion and elimina ion o eliabili y, in: IEEE S d 802.1CB-2017, 2017, pp. 1–102.
[14] A. La añaga, M.C. Lucas-Es añ, I. Ma inez, I. Val, J. Gozal ez, Analysis o 5G-TSN in eg a ion o suppo indus y 4.0, in: 2020 25 h IEEE In e na ional
Con e ence on Eme ging Technologies and Fac o y Au oma ion, ETFA, Vol. 1, 2020, pp. 1111–1114.
[15] Wi-Fi Alliance, Wi-Fi alliance in oduces low powe , long ange Wi-Fi halow, 2016, www.wi- i.o g.
[16] S. Kuma , E. Ga cia-Villegas, D. Camps-Mu , SLA-MLO: Conges ion-awa e SLA-Based scheduling o mul iple links in IEEE 802.11be, in: 2024 IEEE 21s
Consume Communica ions & Ne wo king Con e ence, CCNC, 2024, pp. 875–880.
[17] A. López-Ra en ós, B. Bellal a, IEEE 802.11be mul i-link ope a ion: When he bes could be o use only a single in e ace, in: 2021 19 h Medi e anean
Communica ion and Compu e Ne wo king Con e ence, MedComNe , IEEE, Ibiza, Spain, 2021, pp. 1–7.
[18] IEEE P802.11 Wo king G oup, IEEE P802.11be™/D5.0, No embe 2023 (amendmen o IEEE P802.11-REVme/™D3.0), D a S anda d IEEE
P802.11be™/D5.0, IEEE Compu e Socie y LAN/MAN S anda ds Commi ee, Th ee Pa k A enue, New Yo k, New Yo k 10016-5997, USA, 2023, D a
S anda d o In o ma ion echnology—Telecommunica ions and in o ma ion exchange be ween sys ems Local and me opoli an a ea ne wo ks—Speci ic
equi emen s Pa 11: Wi eless LAN Medium Access Con ol (MAC) and Physical Laye (PHY) Speci ica ions Amendmen 8: Enhancemen s o ex emely
high h oughpu (EHT).
[19] M.-T. Sue , C. Thein, H. Tchouankem, L. Wol , Mul i-connec i i y as an enable o eliable low la ency communica ions—an o e iew, IEEE Commun.
Su . Tu o . 22 (1) (2019) 156–169.
[20] S.M. The es, C. Thein, H. Tchouankem, L. Wol , Adap i e mul i-connec i i y scheduling o eliable low-la ency communica ion in 802.11be, in: 2022 IEEE
Wi eless Communica ions and Ne wo king Con e ence, WCNC, IEEE, 2022, pp. 102–107.
[21] D.V. Banko , A.I. Lyakho , E.M. Kho o , e al., On he use o mul ilink access me hods o suppo eal- ime applica ions in Wi-Fi ne wo ks, J. Commun.
Technol. Elec on. 66 (2021) 1476–1484, URL h ps://doi-o g. ecu sos.biblio eca.upc.edu/10.1134/S1064226921120056.
[22] L. Diez, A. Ga cia-Saa ed a, V. Valls, X. Li, X. Cos a-Pe ez, R. Agüe o, LaSR: A supple mul i-connec i i y schedule o mul i-RAT OFDMA sys ems, IEEE
T ans. Mob. Compu . 19 (3) (2020) 624–639.
[23] C. Roman, P. Ball, S. Ou, E alua ing he bene i o a sma schedule in a non-coope a i e, mul i-use he e ogeneous wi eless ITS en i onmen , in: 2018
11 h In e na ional Symposium on Communica ion Sys ems, Ne wo ks & Digi al Signal P ocessing, CSNDSP, 2018, pp. 1–6.
[24] A. López-Ra en´
os, B. Bellal a, Dynamic a ic alloca ion in IEEE 802.11be mul i-link WLANs, IEEE Wi el. Commun. Le . 11 (7) (2022) 1404–1408.
[25] P.E. I u ia Ri e a, M. Chenie , B. He sco ici, B. Kan a ci, M. E ol-Kan a ci, RL mee s mul i-link ope a ion in IEEE 802.11be: Mul i-headed ecu en
so -ac o c i ic-based a ic alloca ion, 2023, a Xi p ep in a Xi :2303.08959. URL h ps://a xi .o g/abs/2303.08959.
[26] R. Ali, B. Bellal a, A ede a ed ein o cemen lea ning amewo k o link ac i a ion in mul i-link Wi-Fi ne wo ks, in: 2023 IEEE In e na ional Black Sea
Con e ence on Communica ions and Ne wo king, BlackSeaCom, 2023, pp. 360–365.
[27] S. Sudhaka an, V. Mageshkuma , A. Baxi, D. Ca alcan i, Enabling QoS o collabo a i e obo ics applica ions wi h wi eless TSN, in: 2021 IEEE In e na ional
Con e ence on Communica ions Wo kshops, ICC Wo kshops, 2021, pp. 1–6.
[28] M.K. A iq, R. Muza a , O. Seijo, I. Val, H.-P. Be nha d, When IEEE 802.11 and 5G mee Time-Sensi i e ne wo king, IEEE Open J. Ind. Elec on. Soc. 3
(2021) 14–36.
[29] D. Ca alcan i, C. Co dei o, M. Smi h, A. Rege , Wi-Fi TSN: Enabling de e minis ic wi eless connec i i y o e 802.11, IEEE Commun. S and. Mag. 6 (2022)
22–29.
[30] M. Alsaka i, Wi-Fi 7: Mul i-Link Ope a ion in XR Indus ial Scena ios, (Mas e ’s hesis), KTH Royal Ins i u e o Technology, 2022.
[31] IEEE Compu e Socie y, IEEE S anda d o In o ma ion Technology—Telecommunica ions and In o ma ion Exchange be ween Sys ems Local and
Me opoli an A ea Ne wo ks—Speci ic Requi emen s Pa 11: Wi eless LAN Medium Access Con ol (MAC) and Physical Laye (PHY) Speci ica ions,
LAN/MAN S anda ds Commi ee, 2020, URL h ps://ieeexplo e.ieee.o g/documen /9109952. IEEE S d 802.11TM -2020 (Re ision o IEEE S d 802.11-2016).
[32] NS-3 De elopmen Team, NS-3 Wi-Fi Module Design, URL h ps://www.nsnam.o g/docs/models/h ml/wi i-design.h ml. NS-3 Documen a ion.
[33] P.M. Ros , T. Kolding, Pe o mance o in eg a ed 3GPP 5g and IEEE TSN ne wo ks, IEEE Commun. S and. Mag. 6 (2) (2022) 51–56.