Resea ch A icle Vol. 17, No. 9 / Sep embe 2025 / Jou nal o Op ical Communica ions and Ne wo king 757
T a ic-awa e ONU g ouping and downs eam
bandwid h alloca ion o mul icas se ices
in lexible- a e PONs
Xiang Lu,1,2Xinshui Wei,1Zhiyuan Zhong,1Jun Li,1,*AND Luis Velasco2
1School o Elec onic and In o ma ion, Soochow Uni e si y, Suzhou 215031, China
2Ad anced B oadband Communica ions Cen e (CCABA), Uni e si a Poli ècnica de Ca alunya (UPC), 08034 Ba celona, Spain
*[email p o ec ed]
Recei ed 22 Ma ch 2025; e ised 20 July 2025; accep ed 21 July 2025; published 18 Augus 2025
In lexible- a e passi e op ical ne wo ks (PONs), op ical ne wo k uni s (ONUs) wi h di e en channel condi-
ions can achie e di e en da a a es by implemen ing lexible ansmission pa ame e s, e.g., modula ion o ma ,
he eby enhancing sys em capaci y. Co espondingly, he downs eam ame is di ided in o mul iple sub ames,
each being ecei ed and p ocessed only by ONUs in i s a ge ed g oup. Howe e , o downs eam mul icas se -
ices, he da a need o be duplica ed and encapsula ed in o mul iple sub ames o ONUs belonging o di e en
g oups, esul ing in da a edundancy and deg ada ion o e ec i e h oughpu (i.e., h oughpu wi hou edun-
dan da a). To imp o e esou ce u iliza ion, ONU g ouping should be dynamically adjus ed acco ding o bo h
channel condi ions and ime- a ying ne wo k ac o s, e.g., a ic loads and mul icas membe ships. Fo his,
we i s enhance he cu en downs eam scheduling p o ocol o suppo dynamic ONU g ouping in a mul icas
scena io and p opose a a ic-awa e ONU g ouping (TAOG) algo i hm o imp o e e ec i e h oughpu , which
op imizes ONU g ouping by conside ing ime- a ying ne wo k condi ions. As ONUs belonging o di e en
g oups ha e di e en da a a es, we u he p opose a g oup-based downs eam ime slo alloca ion (GBDTA)
algo i hm o adjus ime slo s o each se ice by conside ing hei demands and ONU da a a es. Exhaus i e sim-
ula ion esul s show ha he in eg a ed TAOG-GBDTA scheme adap s e ec i ely o dynamic ne wo k condi ions
and, compa ed o con en ional schemes, i e ec i ely imp o es e ec i e h oughpu , educes edundancy, and
achie es lowe packe la ency unde a ious mul icas scena ios. © 2025 Op ica Publishing G oup. All igh s, including
o ex and da a mining (TDM), A i icial In elligence (AI) aining, and simila echnologies, a e ese ed.
h ps://doi.o g/10.1364/JOCN.562901
1. INTRODUCTION
Time di ision mul iplexing-passi e op ical ne wo ks (TDM-
PONs), ope a ing on a poin - o-mul ipoin (P2MP)
a chi ec u e, a e among he mos commonly-used access ne -
wo k echnologies. No ably, In e na ional Telecommunica ion
Union-Telecommunica ion S anda diza ion Sec o (ITU-T)
PONs, e.g., Gigabi PON (GPON) and XG(S)-PON, a e
widely deployed, especially in he scena io o ibe o he home
(FTTH) [1]. To mee he s ingen quali y o se ice (QoS)
equi emen s o eme ging new b oadband se ices, ITU-T has
s anda dized he nex -gene a ion highe speed TDM-PON
[2–7]. In a ypical ITU-T TDM-PON, a single op ical line
e minal (OLT) is connec ed o mul iple op ical ne wo k uni s
(ONUs) ia an op ical dis ibu ion ne wo k (ODN). In he
ups eam di ec ion, each ONU ansmi s i s da a o he OLT
in a bu s mode wi hin alloca ed ime slo s. In he downs eam
di ec ion, he OLT agg ega es da a o all ONUs in o a ame
wi h a ixed leng h o 125 µs, e e ed o as he downs eam
PHY ame, and b oadcas s i con inuously o all ONUs. Each
ONU iden i ies and e ie es i s da a om he downs eam
PHY ame. Fu he mo e, bo h ups eam and downs eam
ansmissions ope a e a a uni o m, cons an peak da a a e
ac oss all ONUs [8,9]. In his ega d, hese TDM-PONs a e
ca ego ized as ixed- a e PONs.
Howe e , he channel condi ions be ween ONUs and he
OLT can a y signi ican ly due o op ical pa h loss (OPL),
in luenced by ac o s such as he physical dis ance and he
numbe o spli e s [10–13]. In a ixed- a e PON, he con-
s an peak da a a e is cons ained by he wo s -case op ical
budge , a he han conside ing he ac ual channel condi ions
expe ienced by each ONU [10,13]. As a esul , ONUs wi h
be e channel condi ions ope a e below hei po en ial capac-
i y, he eby limi ing he o e all sys em capaci y. To add ess
his issue, a p omising app oach is o di e en ia e he peak
da a a es o ONUs by allowing ONUs wi h be e channel
condi ions o ope a e a highe da a a es while main aining
compa ibili y wi h hose unde poo e condi ions. TDM-
PONs ha allow ONUs o ope a e a a ying peak da a a es
1943-0620/25/090757-14 Jou nal © 2025 Op ica Publishing G oup
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 09:44:35 UTC om IEEE Xplo e. Res ic ions apply.
758 Vol. 17, No. 9 / Sep embe 2025 / Jou nal o Op ical Communica ions and Ne wo king Resea ch A icle
a e e e ed o as lexible- a e PONs [12], and his concep is
also unde discussion in he ongoing wo k o ITU-T S udy
G oup 15 (SG15) on Ve y High Speed PON [14].
In lexible- a e PONs, ONUs wi h simila channel con-
di ions can be g ouped o sha e he same ansmission
pa ame e s, e.g., modula ion o ma s and o wa d e o
co ec ion (FEC) coding, enabling hem o ope a e a a speci ic
peak da a a e. Fu he mo e, he peak da a a e can be lexibly
adjus ed by modi ying ansmission pa ame e s wi hin he
cons ain o he powe budge [11–20]. Ups eam mul i- a e
ecep ion is suppo ed no only in mul i-gene a ion PON
coexis ence sys ems [21,22] bu also na i ely in ad anced
s anda ds such as NG-PON2 and 50 G-PON [4,5]. Since
he OLT can ecei e da a encoded wi h di e en modula ion
o ma s, he adi ional ups eam scheduling p o ocol can
s ill be used a e in oducing lexible a es [23]. No ably, he
di e ences in peak da a a es among ONUs mus be consid-
e ed du ing ime slo alloca ion o ensu e e icien scheduling
[15,24]. In con as , in he downs eam di ec ion, since ONUs
may employ di e en ansmission pa ame e s, he adi-
ional b oadcas -based downs eam scheduling p o ocol in
he ixed- a e PON becomes ine ec i e. To add ess his, he
downs eam PHY ame is di ided in o mul iple sub ames,
each occupying mul iple speci ic ime slo s [10,16]. The sub-
ame di e s in modula ion and coding pa ame e s, which a e
assigned o a g oup o ONUs wi h simila channel condi ions
and da a a e equi emen s. Di e en om ixed- a e PONs,
whe e ONUs need o p ocess he whole downs eam ame,
ONUs in lexible- a e PONs only ecei e and p ocess hei
own downs eam sub ame o educe compu ing complexi y
and ene gy consump ion [10,18,25]. Compa ed o ixed- a e
PONs, lexible- a e PONs ha e g ea e po en ial o imp o ing
h oughpu e iciency by exploi ing a ailable link ma gin.
Mo eo e , hey educe powe consump ion and lowe ONU
compu a ional complexi y, and hus a e wo hy o u he
in es iga ion.
Despi e hese ad an ages, lexible- a e PONs s ill ace
se e al open issues, such as he challenge o main aining
s able downs eam synch oniza ion unde lexible modula-
ion and he po en ial inc ease in sys em cos . Among hese,
downs eam scheduling emains a key challenge, especially
in mul icas scena ios. On he one hand, mul icas scena ios
in oduce addi ional edundancy in o he cu en schedul-
ing mechanisms, he eby impac ing sys em pe o mance.
The g owing demand o high-bandwid h ideo se ices,
e.g., li e ideo s eaming, pe sonalized sho - o m ideos, and
ul a-high-de ini ion ideo-on-demand se ices, has signi i-
can ly inc eased mul icas a ic, u he s aining scheduling
e iciency [26,27]. Based on b oadcas mode, downs eam
se ices can be classi ied in o unicas and mul icas se ices.
Unicas se ices, e.g., ideo-on-demand (VoD) s eams, equi e
dedica ed da a lows o indi idual use eques s o ensu e pe -
sonalized con en deli e y. Thus, a unicas a ic low is sen
exclusi ely o i s co esponding ONU in a ixed- a e PON.
In con as , a a ic low om a mul icas se ice, e.g., li e
s eaming, may be b oadcas simul aneously o mul iple
ONUs. The se o ONU membe s associa ed wi h a speci ic
mul icas se ice is e e ed o as i s mul icas membe ship.
Nowadays, mul icas deli e y has al eady been implemen ed
in FTTx b oadband access ne wo ks [28–30]. In GPON and
EPON sys ems, pa ching-based mul icas mechanisms a e
commonly used: he i s use ini ia es a mul icas s eam, and
subsequen use s ecei e he ongoing con en ia mul icas
and he missed po ion h ough sho unicas pa ches [28].
This me hod signi ican ly educes edundan ansmissions
and imp o es bandwid h u iliza ion. Mo eo e , hese mul-
icas se ices o en ollow a Zip dis ibu ion, whe e a small
numbe o highly popula con en s accoun o a la ge po -
ion o o al a ic. As a esul , ope a o s can p e-cache hese
popula con en s in use -side bu e s du ing o -peak hou s,
he eby imp o ing deli e y e iciency [29]. In addi ion, o li e
s eaming scena ios, an SDN-aided NG-EPON a chi ec u e
has been p oposed o dynamically manage mul icas g oups
based on use iewpo s and QoS demands [30]. Howe e , in
lexible- a e PONs, each ONU g oup can only p ocess i s own
sub ame. Consequen ly, when he membe ships o a mul icas
se ice a e dis ibu ed ac oss mul iple ONU g oups, he mul-
icas da a mus be duplica ed and encapsula ed in o sepa a e
sub ames. The amoun o duplica ed da a inc eases wi h he
numbe o ONU g oups associa ed wi h he se ice, leading o
signi ican edundancy and limi ing e ec i e h oughpu . A
s aigh o wa d app oach o educing edundancy is o min-
imize he numbe o ONU g oups. Howe e , his may lowe
he da a a es o ONUs wi h be e channel condi ions, which
educes he o e all sys em capaci y. Consequen ly, despi e he
dec ease in edundancy, e ec i e h oughpu may s ill deg ade.
The e o e, op imizing ONU g ouping equi es balancing he
ade-o be ween minimizing edundancy and main aining
high da a a es o maximize o e all e ec i e h oughpu .
On he o he hand, in mul icas scena ios, alloca ing
downs eam bandwid h o sa is y he bandwid h equi e-
men s o se ices wi h di e en b oadcas modes in oduces
addi ional challenges. In [31], a session-based p io i y band-
wid h alloca ion mechanism was p oposed o dynamically
alloca e bandwid h among ONUs in TDM-PONs, add ess-
ing ine iciencies caused by mul iple ONUs eques ing he
same mul icas con en . On his basis, Kim e al. [32] p o-
posed a mul icas -sha e weigh ed ai queuing (MS-WFQ)
mechanism o a ge in e -session ai ness, which dynamically
adjus s he se ice weigh s o mul icas se ices o op imize
h oughpu while main aining ai ness ac oss hese se ices.
Howe e , such schemes a e di icul o apply in lexible- a e
PONs ha employ ONU g ouping me hods, as hey do no
accoun o mul icas edundancy and he a ying da a a es o
ONUs. Consequen ly, hey s uggle o mee he equi emen s
o se ices, which a e c ucial o main aining high e ec i e
h oughpu in a lexible- a e PON. The e o e, downs eam
bandwid h alloca ion schemes should be enhanced o conside
bo h mul icas edundancy and da a a e di e ences, he eby
ully mee ing se ice equi emen s and imp o ing scheduling
e iciency.
To add ess hese challenges, his pape i s enhances he
downs eam scheduling p o ocol o lexible- a e PONs o
suppo mul icas scena ios. Gi en he limi a ions o exis -
ing ONU g ouping and bandwid h alloca ion schemes, we
in oduce a no el downs eam a ic-awa e ONU g ouping
(TAOG) algo i hm, complemen ed by a g ouping-based
downs eam ime slo alloca ion (GBDTA) algo i hm.
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 09:44:35 UTC om IEEE Xplo e. Res ic ions apply.
Resea ch A icle Vol. 17, No. 9 / Sep embe 2025 / Jou nal o Op ical Communica ions and Ne wo king 759
Simula ion esul s show ha he combina ion o he p o-
posed algo i hms, e e ed o as he TAOG-GBDTA scheme,
can e ec i ely suppo mul icas scena ios and signi ican ly
imp o e sys em pe o mance in e ms o e ec i e h oughpu
and la ency.
The es o he pape is o ganized as ollows. Sec ion 2
in oduces an enhanced downs eam scheduling p o ocol
and analyzes he e ec i e h oughpu o lexible- a e PONs.
Sec ion 3p esen s he p oposed TAOG-GBDTA scheme based
on he p oposed scheduling p o ocol. Sec ion 4p esen s and
analyzes he simula ion esul s o he p oposed algo i hm.
Finally, Sec ion 5concludes he pape .
2. FLEXIBLE-RATE PON ARCHITECTURE
AND PROTOCOL
In his sec ion, we i s p esen a ep esen a i e a chi ec u e
o a lexible- a e PON. To suppo dynamic ONU g ouping
in a mul icas scena io, an enhanced downs eam scheduling
p o ocol is p oposed. Then, we u he analyze he downs eam
e ec i e h oughpu in a lexible- a e PON.
A. A chi ec u e
Figu e 1illus a es he high-le el a chi ec u e o a lexible- a e
PON, which consis s o an OLT and mul iple ONUs. The
ONUs adop di e en ansmission pa ame e s, e.g., modu-
la ion o ma , acco ding o hei OPL, which is in luenced by
ac o s such as physical ibe dis ance and he numbe o op i-
cal spli e s along he ODN. Fo example, ONU1in Fig. 1is
connec ed o he OLT ia a single op ical spli e , hus expe i-
encing a low OPL, which enables u ilizing he 4-le el pulse
ampli ude modula ion (PAM4) o ma o achie e highe da a
a es. In con as , ONU2and ONU3a e connec ed ia wo
op ical spli e s, hence expe iencing a highe OPL. The e o e,
hey employ he non- e u n- o-ze o (NRZ) modula ion o ma
o main ain eliable pe o mance unde mo e challenging
channel condi ions. Based on hei OPL, hese ONUs a e
g ouped in o wo g oups: ONU1belongs o G oup 1, while
ONU2and ONU3belong o G oup 2. Howe e , he da a a es
o hese ONUs can be adjus ed dynamically, and he g oup
assignmen s may change acco dingly. To suppo his lexibil-
i y, digi al signal p ocessing (DSP) modules a e in oduced in
bo h he OLT and he ONUs, enabling modula ion o ma
adap a ion du ing ansmission and ecep ion. The ope a ion
o he DSP modules is managed by he PON ansmission
con e gence (TC) laye , which consis s o he se ice adap a-
ion sublaye , he aming sublaye , and he PHY adap a ion
sublaye . These sublaye s handle aming, scheduling, and
g oup managemen , among o he s.
In he downs eam ansmission, he OLT di ides he ame
in o mul iple o wa d e o co ec ion (FEC) codewo ds,
i.e., da a blocks encoded wi h FEC o e o esilience, cus om-
ized o di e en ONU g oups. Fo ins ance, FEC codewo ds
modula ed wi h PAM4 a e designa ed o G oup 1, while
hose modula ed wi h NRZ a e di ec ed o G oup 2. Upon
ecei ing he downs eam ame, each ONU iden i ies he
FEC codewo ds co esponding o i s modula ion o ma and
econs uc s i s assigned sub ame. Fo example, G oup 1 only
p ocesses he PAM4-modula ed codewo ds (yellow blocks),
while G oup 2 p ocesses he NRZ-modula ed codewo ds
(g een blocks). This ensu es ha each ONU g oup handles
only i s designa ed da a, he eby educing DSP complexi y.
The lexible- a e PON a chi ec u e de e mines a special
downs eam scheduling scheme ha is dis inc om ha o
con en ional ixed- a e PONs. The downs eam scheduling
p o ocol plays a c i ical ole in implemen ing dynamic ONU
g ouping, especially in mul icas scena ios. In he nex sub-
sec ion, we in oduce he downs eam scheduling p o ocol
o ixed- a e PONs, ollowed by ou p oposed enhanced
downs eam scheduling p o ocol ailo ed o lexible- a e
PONs.
Tx( )
OLT: Op ical line e minal ONU: Op ical ne wo k uni Tx: T ansmi e Rx: Recei e
DSP: Digi al signal p ocessing TC: T ansmission con e gence
Rx( )
Rx( )
ComTC
F aming
sublaye
PHY adap a ion
sublaye
Se ice adap a ion
sublaye
Tx
4:1 Spli e
Rx
PON TC
F aming
sublaye
PHY adap a ion
sublaye
Se ice adap a ion
sublaye
Fil e
DSP
OLT
4:1 Spli e
ime
a e
NRZ NRZ
ime
a e
PAM4
PAM4
ime
a e
NRZ NRZ PAM4
PAM4
Downs eam PHY ame
Rx
Tx
ONU1
PON
TC
DSP
(PAM4)
ONU
2
ONU
3
Tx
Rx
PON
TC
DSP
(NRZ)
G oup 1
G oup 2
Fig. 1. High-le el a chi ec u e o a lexible- a e PON.
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 09:44:35 UTC om IEEE Xplo e. Res ic ions apply.
760 Vol. 17, No. 9 / Sep embe 2025 / Jou nal o Op ical Communica ions and Ne wo king Resea ch A icle
(b)
Flexible- a e PON OLT TC laye ONU TC laye (G oup 1: ONU 1)
SA sublaye
PHY sublaye SA sublaye
FS sublaye PHY sublaye
XGEM ames
FS ame 1
FS ame 2
DS PHY ame
FS ame 2
(a)
Fixed- a e PON OLT TC laye ONU TC laye
SA sublaye FS sublaye PHY sublaye SA sublaye
FS sublaye
PHY sublaye
XGEM ames
FS ame DS PHY ame FS ame
U
1
M
2
U
2
FS Heade
(BWmap +
PLOAMd)
HM
1
U
1
M
2
U
2
FS sublaye
DS bandwid h alloca ion
ONU g ouping and
bandwid h alloca ion
T a ic
schedule
T a ic
schedule
DS PHY ame
T a ic
classi ie
T a ic
classi ie
HM
1
U
1
M
2
U
2
P
H
1
M
1
U
1
PSBd
H
M
1
U
1
M
2
U
2
PHM
1
U
1
M
2
U
2
M
1
U
1
DS PHY ame 1
DS PHY ame 2
T a ic
classi ie
M
1
U
2
FS ame 1
ONU TC laye (G oup 2: ONU 2 and 3)
M
1
U
1
M
2
U
2
M
2
MT
2
UT
2
MT
1
UT
1
Ups eam scheduling
and da a a e adjus ing
OLT: Op ical line e minal ONU: Op ical ne wo k uni TC: T ansmission con e gence MT: Mul icas a ic
UT: Unicas a ic SA: Se ice adap a ion FS: F aming DS: Downs eam
PHY: Physical in e ace adap a ion BWmap: Bandwid h map PSBd: Physical synch oniza ion block downs eam
PLOAMd: Physical laye ope a ion, adminis a ion and main enance downs eam
1
3
2
3
5
46
7
8
5
4
67
8
M
1
U
1
M
2
U
2
MT
2
UT
2
MT
1
UT
1
M
1
1
2
Ups eam scheduling
Ups eam scheduling
and da a a e adjus ing
UT
3
(OMCI) U
3
U
3
U
3
U
3
U
3
U
3
UT
3
(OMCI)
U
3
H
2
M
1
M
2
U
2
U
3
H
2
H
1
M
1
U
1
U
2
PM
1
M
2
U
3
H
1
M
1
U
1
P
H
2
U
2
PM
1
M
2
U
3
H
2
U
2
M
1
M
2
U
3
H
1
M
1
U
1
U
3
FS Heade -G1
FS Heade -G2
ONU
2&3
ONU
2
ONU
1&2
ONU
1
ONU
2&3
ONU
2
ONU
1&2
ONU
1
ONU
2
ONU
2
Fig. 2. (a) Downs eam scheduling p o ocol in a ixed- a e PON; (b) enhanced downs eam scheduling p o ocol in a lexible- a e PON.
B. Enhanced Downs eam Scheduling P o ocol
Figu e 2(a) illus a es he downs eam scheduling p o o-
col acco ding o he ITU-T 50G-PON Common TC laye
speci ica ion [5]. In he OLT, se ice da a uni s (SDUs), such
as E he ne ames, a e i s classi ied and added o di e -
en queues based on hei se ice ypes (s ep 1). Then, in
he se ice adap a ion sublaye , SDUs a e encapsula ed in o
XGEM ames using an XG-PON encapsula ion me hod and
assigned a Po -ID, which is included in he heade (s ep 2).
The XGEM Po -ID iden i ies i s co esponding se ice ype
and b oadcas mode. The XGEM Po -ID o a unicas XGEM
ame co esponds o a single ONU, whe eas he XGEM Po -
ID o a mul icas XGEM ame can be ecognized by mul iple
ONUs. Fo illus a i e pu poses, Fig. 2(a) shows wo mul icas
(M1and M2) and h ee unicas (U1,U2, and U3) a ic lows
belonging o di e en se ices. Mul icas XGEM ames a e
ma ked wi h dashed-line blocks, while unicas XGEM ames
a e ep esen ed by solid-line blocks. Among hese, U3 ep e-
sen s he ONU Managemen and Con ol In e ace (OMCI)
a ic, which deli e s con igu a ion and managemen in o ma-
ion om he OLT o each ONU indi idually. One unc ion o
OMCI is o assign o wi hd aw he logical iden i ie s o each
ONU (XGEM Po -IDs), which allows he OLT o con ol
whe he an ONU can ecei e a speci ic unicas o mul icas
a ic low. As a esul , all mul icas XGEM ames need o be
ansmi ed only once, and he e o e, each se ice equi es one
single dedica ed ime slo .
In he aming sublaye , a bandwid h alloca ion module
alloca es downs eam bandwid h o each a ic, de e mining
he XGEM ames ha need o be scheduled (s ep 3). As pe
he bandwid h alloca ion in o ma ion, he a ic schedule
agg ega es hese XGEM ames and encapsula es hem in o
an FS ame by adding an FS heade (s ep 4). The FS heade
con ains bo h he ups eam bandwid h alloca ion in o ma ion
(i.e., he BWmap ield) and he downs eam physical laye
ope a ion, adminis a ion, and main enance (PLOAMd) ield
o ONUs. The PLOAMd messages accoun o ca ying man-
agemen and con ol ins uc ions o ONUs, such as ONU
ac i a ion o anging. Since a mul icas XGEM ame can be
ecognized by mul iple ONUs, each mul icas a ic (M1and
M2) is alloca ed one single dedica ed ime slo in an FS ame.
Subsequen ly, in he PHY adap a ion sublaye , his FS ame
is u he encapsula ed in o a 125 µs downs eam PHY ame
a e FEC encoding, sc ambling, and bi in e lea ing (s ep 5).
A e adding a downs eam physical synch oniza ion block
(PSBd), he downs eam PHY ame is b oadcas o all ONUs
in he lexible- a e PON. Upon ecei ing he PHY ame, each
ONU synch onizes wi h he PSBd ield and ex ac s he FS
ame (s ep 6). The FS ame is hen u he de-encapsula ed
in o XGEM ames a he aming sublaye , whe e he BWmap
ields a e ex ac ed om he FS heade and u ilized by he
ONU o ups eam scheduling (s ep 7). A a ic classi ie hen
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 09:44:35 UTC om IEEE Xplo e. Res ic ions apply.
Resea ch A icle Vol. 17, No. 9 / Sep embe 2025 / Jou nal o Op ical Communica ions and Ne wo king 761
ca ego izes he da a ames and selec s hei co esponding
XGEM ames based on he XGEM Po -IDs, es o ing hem
in o SDUs in he se ice adap a ion sublaye (s ep 8). No ably,
in his p o ocol, e e y ONU mus ecei e and p ocess he
en i e downs eam PHY ame.
In con as o he ixed- a e PON, he downs eam sched-
uling p o ocol in a lexible- a e PON mus accoun o he
a ying ansmission pa ame e s ac oss ONU g oups. No e
ha o a oid po en ial de ec ion e o s and limi DSP com-
plexi y, in his pape , each ONU is es ic ed o belong o only
one g oup a a ime in a lexible- a e PON. Consequen ly,
da a o di e en ONU g oups a e independen ly encapsu-
la ed acco ding o hei espec i e ansmission pa ame e s.
To achie e his, a e p ocessing in he se ice adap a ion sub-
laye , XGEM ames belonging o he same ONU g oup a e
agg ega ed and encapsula ed in o an independen FS ame.
Fo mul icas a ic, i all i s ONU membe s belong o he
same g oup, he co esponding FS ame can di ec ly ca y
i s mul icas XGEM ames wi hou duplica ion. O he wise,
he XGEM ames mus be duplica ed and encapsula ed in o
nFS ames, whe e n ep esen s he numbe o ONU g oups
spanned by he mul icas se ice. This implies ha a mul i-
cas se ice will be assigned a di e en ime slo in mul iple
dis inc FS ames. Figu e 2(b) illus a es he enhanced down-
s eam scheduling p o ocol o suppo mul icas scena ios.
Assume ha ONU1belongs o G oup 1, ONU2and ONU3
belong o ano he g oup. The mul icas XGEM ame M1
is in ended o ONU1and ONU2. The e o e, M1is dupli-
ca ed and encapsula ed in o bo h FS F ame 1 and FS F ame
2 (s ep 4). In con as , mul icas a ic 2 has ONU membe s
ONU2and ONU3in he same g oup, and hus M2is only
encapsula ed in o FS F ame 2. No e ha he duplica ion o
XGEM ames occu s only du ing hei encapsula ion in o FS
ames. As a esul , no addi ional queues a e equi ed in he
OLT o s o e da a ames o di e en g oups, elimina ing he
need o modi ying he queue managemen mechanism. Since
ONU g ouping is de e mined by ansmission pa ame e s,
egula upda es o hese pa ame e s a e necessa y o suppo
dynamic ONU g ouping. To achie e his, each FS ame
includes he ansmission pa ame e in o ma ion wi hin he
PLOAMd ield o he ONUs o he g oup pe iodically. This
in o ma ion, oge he wi h he BWmap, is gene a ed by a no el
ONU g ouping and bandwid h alloca ion module (s ep 3)
and copied o all FS ame heade s by he a ic schedule o
ensu e consis en upda es ac oss all ONU g oups (s ep 4). We
p e e he PLOAMd ield a he han OMCI o deli e ing
g ouping upda es, as PLOAM messages o e lowe signaling
la ency, he eby enabling p omp ONU econ igu a ion. The
FS ames om di e en g oups a e subsequen ly agg ega ed
in he PHY adap a ion sublaye and encapsula ed in o a com-
ple e downs eam PHY ame (s ep 5). Du ing his p ocess,
a codewo d in e lea ing me hod be ween FS ames can be
pe o med o educe DSP complexi y [25]. Upon ecei ing
he downs eam PHY ame, he ONU econs uc s he FS
ame o i s g oup (s ep 6). Subsequen ly, he FS ame is
de-encapsula ed a he aming sublaye (s ep 7). In addi ion o
he BWmap ield used o ups eam scheduling, he PLOAMd
ield in he FS heade p o ides upda ed ansmission pa ame-
e s. To enable dynamic ONU g ouping, he ONU uses his
in o ma ion o adjus i s modula ion o ma o ecei ing
downs eam da a. Meanwhile, he ex ac ed XGEM ames a e
p ocessed a he se ice adap a ion sublaye (s ep 8), whe e he
a ic classi ie de e mines whe he he ames a e in ended o
he ecei ing ONU.
C. Downs eam E ec i e Th oughpu Analysis
To e alua e he po en ial pe o mance imp o emen s b ough
by dynamic g oup op imiza ion, we u he analyze he down-
s eam e ec i e h oughpu unde h ee di e en ONU
g ouping me hods: (a) a ixed- a e PON, (b) a lexible- a e
PON, and (c) a lexible- a e PON wi h ONU g ouping
op imiza ion.
Figu e 3(a) illus a es an example o a downs eam PHY
ame in a ixed- a e PON wi hou ONU g ouping. Fo sim-
plici y, he o e head is omi ed. The colo ed blocks ep esen
he bandwid h alloca ed o di e en se ices, which is de e -
mined by a ixed da a a e 0and ame du a ion 0.Ui
jand Mi
j
indica e he j h unicas and mul icas se ice a ge ing ONU
i, espec i ely. Suppose ha he e a e six ONUs and i e a ic
lows ( om ou unicas se ices and one mul icas se ice) ha
equi e ansmission. Since bo h unicas and mul icas da a
a e ansmi ed only once, no edundancy is gene a ed. In his
scena io, he downs eam e ec i e h oughpu equals he sum
o he bandwid h alloca ed o each se ice.
Figu e 3(b) p esen s an example o a downs eam PHY ame
in a lexible- a e PON wi h a ixed ONU g ouping me hod. In
his case, he sys em achie es i s maximum sys em capaci y. Six
ONUs a e di ided in o h ee g oups: G oup 1, G oup 2, and
G oup 3. ONU1 and ONU2 in G oup 1 achie e he highes
da a a e 1due o a o able channel condi ions, while ONU3
and ONU4in G oup 2 ope a e a a lowe peak da a a e 2.
ONU5and ONU6in G oup 3 expe ience he poo es channel
condi ions and ope a e a he lowes peak da a a e 3= 0.
Assuming he FS ame o each g oup is alloca ed an equal
(b)
(a)
(c)
G1(ONU1&2)
G2(ONU3&4)
G3(ONU5&6)
G1(ONU1&2)
G2(ONU3)
G3(ONU4&5&6)
U3
Fig. 3. Examples o a downs eam PHY ame in (a) a ixed- a e
PON, (b) a lexible- a e PON, and (c) a lexible- a e PON wi h
ONU g ouping op imiza ion.
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 09:44:35 UTC om IEEE Xplo e. Res ic ions apply.
762 Vol. 17, No. 9 / Sep embe 2025 / Jou nal o Op ical Communica ions and Ne wo king Resea ch A icle
ime slo du a ion (i.e., 1= 2= 3), he lexible- a e PON can
ansmi he same da a in less ime (indica ed by he ed dashed
blocks), he eby inc easing he maximum sys em h oughpu .
Howe e , he mul icas da a M1di ec ed o ONUs in h ee
di e en g oups (i.e., ONU1, ONU4, and ONU6) mus be
ansmi ed sepa a ely o each g oup, occupying up o h ee
ime slo s (M1
1,M4
1, and M6
1), which in oduces unnecessa y
edundancy and signi ican ly educes he e ec i e h oughpu .
Figu e 3(c) shows an example a e op imizing he g ouping
by mo ing ONU4 om G oup 2 o G oup 3 (i.e., lowe ing
i s da a a e o 3), so he edundan ansmission M4
1can be
elimina ed. Compa ed o he solu ion in Fig. 3(b), his adjus -
men u he imp o es he e ec i e h oughpu . Howe e ,
mo ing ONU1 o G oup 3 o educe u he edundancy is
no ad isable, as lowe ing i s da a a e would inc ease he ime
slo equi ed o he unicas se ice U1
1, hus dec easing he
o e all e ec i e h oughpu . The e o e, op imizing ONU
g ouping in ol es a ade-o be ween minimizing edundancy
and main aining high ONU da a a es.
Fu he mo e, conside ing ha no only he mul icas a ic
load bu also mul icas membe ships can dynamically change
as ONUs join o lea e a mul icas session, i is necessa y o
design a a ic-awa e algo i hm o op imize ONU g ouping
dynamically. In addi ion, he examples in Figs. 3(b) and 3(c)
adop a simple bandwid h alloca ion s a egy, whe e band-
wid h is e enly alloca ed among ONU g oups. In p ac ice,
downs eam bandwid h alloca ion in lexible- a e PONs is
mo e complex and can signi ican ly impac he sys em e ec-
i e h oughpu . To add ess his, he nex sec ion p oposes
a scheme ha in eg a es op imized ONU g ouping wi h an
ad anced downs eam bandwid h alloca ion algo i hm.
3. TRAFFIC-AWARE ONU GROUPING AND
DOWNSTREAM BANDWIDTH ALLOCATION
Based on he p oposed downs eam scheduling p o ocol, we
u he design a TAOG-GBDTA scheme combined wi h wo
heu is ic algo i hms o lexible- a e PONs. Figu e 4shows
he TAOG-GBDTA amewo k in he TC laye o OLT. In
his amewo k, he p oposed TAOG algo i hm and GBDTA
algo i hm op imize he ONU g ouping and downs eam
bandwid h alloca ion, espec i ely.
As shown in Fig. 4, bo h he TAOG and GBDTA algo i hms
ecei e a ic pa ame e s om he mul icas in o ma ion man-
age module in he uppe laye du ing each scheduling cycle,
which has a ixed leng h o 125 µs. These pa ame e s include
he a ic load o se ices and he mapping able be ween
se ices and hei mul icas membe ships. Based on hese
pa ame e s, he TAOG algo i hm pe iodically gene a es
op imized ONU g oupings, which a e ed in o he GBDTA
algo i hm and p o ided o he a ic schedule o gene a ing
PLOAMd messages. The in e al be ween hese upda es is
e e ed o as an op imiza ion cycle, which is se as an in ege
mul iple o he scheduling cycle. No e ha he op imiza ion
cycle is con igu able and can be se signi ican ly longe han
he scheduling cycle o ma ch ha dwa e cons ain s and ensu e
enough ime o ONUs o swi ch g oups ia PLOAMd-based
upda es. In Fig. 4, blue a ows indica e he inpu s and ou -
pu s ha occu a each scheduling cycle, while g een a ows
TAOG: T a ic-awa e ONU g ouping
GBDTA: G ouping-based downs eam ime slo alloca ion
ONU g ouping and
bandwid h alloca ion
P ocessed pe scheduling cycle
P ocessed pe op imiza ion cycle
Op imized ONU
g ouping
Time slo
alloca ion
Mul icas in o ma ion manage
T a ic schedule
T a ic pa ame e s
TAOGGBDTA
Op imized ONU
g ouping
Fig. 4. TAOG-GBDTA amewo k.
ep esen hose a each op imiza ion cycle. The GBDTA algo-
i hm is esponsible o pe o ming downs eam bandwid h
alloca ion. When a new ONU g ouping is p o ided, GBDTA
upda es he ele an pa ame e s acco dingly o ensu e op imal
bandwid h alloca ion decisions. The ou pu decisions a e ed
in o he a ic schedule o gene a e FS ames o each ONU
g oup. The de ailed desc ip ions o hese wo algo i hms a e
p o ided in he ollowing subsec ions.
A. T a ic-Awa e ONU G ouping
The TAOG algo i hm is designed o i e a i ely op imize ONU
g ouping in esponse o ime- a ying ne wo k condi ions,
he eby imp o ing e ec i e h oughpu . Speci ically, he algo-
i hm begins wi h an ini ial ONU g ouping, i.e., he cu en ly
adop ed ONU g ouping, and explo es new ONU g oupings by
making inc emen al adjus men s, i.e., eassigning one ONU
o a di e en g oup o modi ying i s da a a e. A each s ep,
he algo i hm e alua es whe he he new solu ion can imp o e
sys em e ec i e h oughpu by an accep ance h eshold. As
he algo i hm p og esses, he h eshold g adually dec eases,
making i inc easingly selec i e and ul ima ely con e ging
on an op imized g ouping solu ion. The TAOG algo i hm is
de ailed as ollows.
A he beginning o each scheduling cycle, a ic- ela ed
pa ame e s a e upda ed. Le 3,, and 0deno e he se s o
all se ices, ONUs, and g oups, espec i ely. Bo h unicas
and mul icas se ices a e included in 3, as a unicas se ice
can be conside ed a special mul icas case wi h a single ONU
membe . Fo each se ice i∈3, he newly a i ed a ic in
he p e ious scheduling cycle is deno ed by di, and he cumu-
la i e a ic load om he end o he las op imiza ion cycle
o he p esen momen is deno ed by pi. These pa ame e s
a e collec i ely ep esen ed by ec o s D= [d1,d2, ... , dNS]
and P= [p1,p2, ... , pNS], espec i ely, whe e NS ep esen s
he o al numbe o se ices. Pis upda ed as (P+D)du ing
each scheduling cycle. Besides, he mul icas membe ship o
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 09:44:35 UTC om IEEE Xplo e. Res ic ions apply.
Resea ch A icle Vol. 17, No. 9 / Sep embe 2025 / Jou nal o Op ical Communica ions and Ne wo king 763
each se ice is upda ed using he mul icas managemen mes-
sages ecei ed by he OLT, such as he JOIN/LEAVE eques s
om he highe -laye In e ne G oup Managemen P o ocol
(IGMP) [33]. Upon ecei ing such eques s h ough he da a
plane, he OLT upda es i s mul icas membe ship able and
econ igu es he XGEM po -ID o a ec ed ONUs acco dingly
ia OMCC. The mapping be ween se ices and hei ONUs
can be desc ibed as a ma ix M, de ined as
M=
m1,1m1,2... m1,NO
m2,1m2,2... m2,NO
.
.
..
.
.....
.
.
mNS,1mNS,2... mNS,NO
,(1)
whe e NOdeno es he o al numbe o ONUs, and
mi,j∈ {0,1}indica es whe he ONU jis a membe o se ice
i. Speci ically, mi,j=1 indica es ha ONU jis a membe o
se ice i, and mi,j=0 o he wise. Simila ly, he cu en ONU
g ouping is desc ibed by a ma ix GC, exp essed as
GC=
g1,1g1,2... g1,NG
g2,1g2,2... g2,NG
.
.
..
.
.....
.
.
gNO,1gNO,2... gNO,NG
,(2)
whe e NGdeno es he numbe o g oups, and gj,k∈ {0,1}
indica es whe he ONU jbelongs o g oup k(i.e., gj,k=1
means ONU jbelongs o g oup k, and gj,k=0 o he wise).
As men ioned in Sec ion 2, each ONU is assumed o belong o
only one g oup a a ime. Acco dingly, each ONU is es ic ed
o a single g oup du ing he op imiza ion, and gi,kmus sa is y
X
k∈0
gj,k=1,j∈.(3)
Then, he OLT checks an in e nal coun down ime τ,
which acks he emaining scheduling cycles be o e he nex
g ouping op imiza ion. I τ=1, he op imiza ion p ocess is
igge ed, and τis ese o τG, which deno es he p ede ined
op imiza ion cycle. O he wise, τis dec emen ed by one, and
he algo i hm e mina es o his scheduling cycle. Du ing
g ouping op imiza ion, he pe o mance o he cu en g oup-
ing GCis e alua ed. Fo his, a ma ix L= [li,k]is gene a ed,
exp essed as
L=MGC,(4)
whe e he elemen li,kindica es he numbe o ONU membe s
o se ice iin g oup k. In he enhanced downs eam schedul-
ing p o ocol, he ONU membe s o a se ice belonging o he
same g oup sha e he same ime slo , and edundancy occu s
when ONU membe s a e om di e en g oups. To accoun
o sha ed ime slo s, each li,kin Lis ans o med in o l0
i,k;
hus, a new ma ix L0is gene a ed. l0
i,kis calcula ed by
l0
i,k=1,i li,k≥1
0,o he wise ,(5)
whe e l0
i,kindica es whe he se ice ihas any ONU membe s
in g oup k. In o he wo ds, each non-ze o l0
i,kco esponds
o one eal a ic low ha mus be scheduled, including he
edundan ones. Consequen ly, he o al numbe o lows
equals he sum o all l0
i,kin L0. Me ix L0is hen used o cal-
cula e he o al a ic load and edundancy o each se ice,
which is deno ed by a ec o B= [b1,b2, ... , bNG]and
exp essed as
B=PL0.(6)
Le he ec o R= [ 1, 2, ... , NG]deno es he da a a es
o he ONU g oups. Using Band R, he o al ime slo s o
g oup kin he las op imiza ion cycle a e calcula ed as (bk/ k).
Fu he mo e, he e ec i e h oughpu unde he cu en ONU
g ouping GCin he las op imiza ion cycle, e e ed o as ETc,
is exp essed as
ETC=Pi∈3pi
Pk∈0
bk
k
.(7)
No e ha he nume a o , which ep esen s he o al amoun
o accumula ed a ic o be se ed, is ixed du ing each op i-
miza ion cycle, while he denomina o e lec s he o al ime
equi ed o se e his a ic unde he cu en ONU g ouping
GC. The e o e, he op imiza ion on e ec i e h oughpu can
be o mula ed by minimizing he denomina o , e e ed o as
he e ec i e se ice ime (ESC).
To minimize he e ec i e se ice ime ESC, an op imized
g ouping solu ion GBshould be ound. To achie e his, he
algo i hm ollows an i e a i e p ocess o explo e po en ial solu-
ions, and he numbe o i e a ions is de e mined by Imax. ESC
and GCa e ini ially se o he minimum e ec i e se ice ime
ESBand he op imal ONU g ouping solu ion GB, espec-
i ely. A each i e a ion, a new ONU g ouping solu ion GNis
gene a ed based on GCby andomly selec ing an ONU and
adjus ing i s da a a e up o down by one le el. No e ha he
adjus men should no exceed he peak da a a e cons ain o
he ONU.
The new e ec i e se ice ime ESNunde GNis hen calcu-
la ed by Eqs. (4)–(7). The di e ence in e ec i e se ice ime,
e e ed o as δ, is exp essed as
δ=ESN−ESC.(8)
I δ < 0, GNis accep ed as he new cu en solu ion.
O he wise, he accep ance o GNis de e mined p obabilis-
ically using he accep ance p obabili y PA, which can be
w i en as
PA=e−δ
AT ,(9)
whe e AT deno es he accep ance h eshold. A highe ini ial
AT allows he algo i hm o explo e a wide solu ion space by
inc easing he p obabili y o accep ing subop imal solu ions,
whe eas a lowe AT esul s in a mo e conse a i e sea ch. To
de e mine whe he GNis accep ed, PAis hen compa ed
agains a andom alue Y anging om 0 o 1. I PA>Y,
GNis accep ed; o he wise, i e ains GC. Following his, AT
is educed acco ding o (AT =γ·AT), whe e γis he decay
ac o con olling he a e o educ ion. A highe γensu es a
slowe decay and mo e ho ough explo a ion, while a lowe γ
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 09:44:35 UTC om IEEE Xplo e. Res ic ions apply.
764 Vol. 17, No. 9 / Sep embe 2025 / Jou nal o Op ical Communica ions and Ne wo king Resea ch A icle
Algo i hm 1. T a ic-Awa e ONU G ouping (TAOG)
Algo i hm
Inpu :τG,τ,AT,Imax ,γ,3,,0,D,P,M,GC,R
Ou pu :GB
1. P←P+D
2. i τ > 1 hen
3. τ←τ−1
4. elsei τ== 1 hen
5. τ←τG
6. Calcula e L0and Bby Eqs. (4)–(6)
7. Calcula e ESCby Eq. (7)
8. GB←GC; ESB←ESC
9. while i≤Imax do
10. Gene a e a new solu ion GN
11. Calcula e ESNbased on GN
12. δ←ESN−ESC
13. i δ < 0 hen
14. GC←GN;ESC←ESN
15. else
16. PA←exp(−δ/AT)
17. Gene a e Y
18. i PA>Y hen
19. GC←GN;ESC←ESN
20. end i
21. end i
22. i ESC<ESB hen
23. GB←GC;ESB←ESC
24. end i
25. AT ←AT ·γ
26. i←i+1
27. end while
28. Rese P
29. end i
accele a es con e gence bu isks ge ing apped in a local op i-
mum. A he end o each i e a ion, he bes -known solu ion
GBis upda ed i ESCis smalle han he p e iously eco ded
ESB; o he wise, GB emains unchanged.
A e comple ing he i e a ions, he bes solu ion GBis
selec ed based on he minimum obse ed e ec i e se ice ime.
The elemen s in Pa e ese o ze o o he nex op imiza ion
cycle. Finally, GBis passed o he GBDTA algo i hm and he
a ic schedule o downs eam bandwid h alloca ion and
a ic scheduling.
The pseudo-code o he p oposed TAOG algo i hm is
shown in Algo i hm 1. S eps (1–5) check whe he g ouping
op imiza ion is equi ed in he cu en scheduling cycle. In
s eps (6–8), he e ec i e se ice ime o he cu en ONU
g ouping in he las op imiza ion cycle is calcula ed and em-
po a ily se as he op imal solu ion. Following his, s eps
(9–26) y o ind an op imized ONU g ouping solu ion by
i e a ion. S eps (10–11) gene a e a new g ouping solu ion, and
s eps (12–21) de e mine whe he o accep he new solu ion.
S eps (22–24) compa e he new solu ion wi h he cu en
op imal solu ion and upda e he op imal solu ion acco dingly.
A e wa d, he algo i hm educes i s accep ance h eshold (s ep
25) and i e a es again. Finally, Pis ese in s ep (28).
B. G ouping-Based Downs eam Time Slo Alloca ion
The GBDTA algo i hm is p oposed o op imize he down-
s eam bandwid h alloca ion in mul icas scena ios, he eby
u he maximizing he e ec i e h oughpu o he lexible- a e
PON. He e, bandwid h speci ically e e s o he ime slo s
alloca ed o da a ansmission.
Fi s , he OLT calcula es he leng h o he a ailable ime
slo s o he en i e downs eam PHY ame (T) be o e he
beginning o each scheduling cycle, which can be exp essed as
T=E−TPSBd −TFS −TP,(10)
whe e Edeno es he o al du a ion o a downs eam PHY
ame, TPSBd deno es he ime slo s occupied by he PSBd,
TFS accoun s o he ime slo s used o FS heade s, and TP
ep esen s he ime slo s alloca ed o pa i y bi s in he FEC
codewo ds. Simul aneously, i he TAOG algo i hm has gen-
e a ed a new ONU g ouping solu ion, GCis upda ed. The
ma ix M, which e lec s he mul icas membe ships o se -
ices, is also upda ed based on he la es ne wo k condi ions.
Whene e ei he GCo Mis upda ed, L0is ecalcula ed using
Eqs. (4) and (5).
Subsequen ly, he OLT alloca es downs eam ime slo s
o each se ice. To p e en he o e use o downs eam band-
wid h by any single se ice due o he a ia ion in bandwid h
demand, each se ice mus be alloca ed a maximum a ailable
ime slo . He e, he maximum a ailable ime slo o se ice iis
de e mined by i s bandwid h weigh ωi, which is de ined as he
a io o he equi ed ime slo o se ice i o he o al equi ed
ime slo s o all se ices, exp essed as
ωi=TSReq
i
Pn∈3TSReq
n
,(11)
whe e TSReq
ideno es he eques ed ime slo o se ice i. Fo a
mul icas se ice, mo e han one ime slo may be equi ed in
di e en FS ames, whe e di e en modula ion o ma s a e
applied du ing ansmission. Thus, TSReq
iis exp essed as
TSReq
i=X
k∈0
l0
i,k
di
k
.(12)
Based on ωi, he maximum a ailable ime slo o se ice
ican be ob ained by (ωiT). I he equi ed TSReq
ican be
accommoda ed wi hin his limi , i is alloca ed as eques ed;
o he wise, i is limi ed o (ωiT). The o al ime slo alloca ed
o se ice i, i.e., TSSe
i, is exp essed as
TSSe
i=TSReq
i,i TSReq
i≤ωiT
ωiT,o he wise. (13)
Following his, TSSe
i o each se ice is u he di ided
and inco po a ed in o mul iple FS ames o i s associa ed
ONU g oups. The ime slo alloca ed o g oup k o se ice i,
deno ed as TSG o
i,k, can be calcula ed as
TSG o
i,k=σi,k·TSSe
i,(14)
whe e σi,k ep esen s he bandwid h weigh assigned o g oup
k o se ice i. Fo a mul icas se ice, he da a encapsula ed in
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 09:44:35 UTC om IEEE Xplo e. Res ic ions apply.
Resea ch A icle Vol. 17, No. 9 / Sep embe 2025 / Jou nal o Op ical Communica ions and Ne wo king 765
Algo i hm 2. G ouping-Based Downs eam Time Slo
Alloca ion (GBDTA) Algo i hm
Inpu :E,TPSBd,TFS,TP,3,,0,M,GC,D,R
Ou pu : TSG o
i,k
1. T←E−TPSBd −TFS −TP
2. Upda e L0by Eqs. (4) and (5)
3. o i∈3do
4. Calcula e TSReq
iand ωiby Eqs. (11) and (12)
5. i TSReq
i≤ωiT
6. TSSe
i←TSReq
i
7. else
8. TSSe
i←ωiT
9. end i
10. o k∈0do
11. Calcula e σi,kand TSG o
i,kby Eqs. (14) and (15)
12. end o
13. end o
14. Gene a e downs eam FS ames based on TSG o
i,k
each FS ame a e iden ical in size, while he ime slo leng h
a ies depending on he da a a e o he g oup. The e o e, σi,k
is exp essed as
σi,k=l0
i,k
kPn∈0
l0
i,n
n
.(15)
Since each ONU in he g oup can independen ly iden i y i s
own da a ames by ecognizing he XGEM heade , TSG o
i,kcan
be sha ed among he ONU membe s in he g oup, and he e
is no need o u he di ision a he ONU le el. Finally, he
a ic schedule gene a es he FS ame o each g oup based
on TSG o
i,k.
The pseudo-code o he GBDTA algo i hm is de ailed in
Algo i hm 2. S ep (1) upda es he pa ame e s. S ep (2) cal-
cula es he maximum a ailable ime slo o each se ice. In
s eps (3–10), he algo i hm alloca es ime slo s o se ices based
on hei demands. Subsequen ly, he ime slo s alloca ed o
each g oup a e calcula ed in s ep (11). Finally, in s ep (14), FS
ames a e gene a ed based on he alloca ed ime slo s o each
g oup.
The compu a ional complexi y o TAOG-GBDTA is ana-
lyzed as ollows. The complexi y o he TAOG algo i hm
depends on he numbe o i e a ions. Fo he GBDTA algo-
i hm, i s complexi y is de e mined by he numbe o se ices
and hei co esponding g oups, exp essed as O(|3| · |0|).
Gi en ha con en ional bandwid h alloca ion algo i hms in
lexible- a e PONs also assign ime slo s o each g oup associ-
a ed wi h a se ice, he ime complexi y emains unchanged.
The e o e, he p ima y sou ce o complexi y inc ease in he
p oposed scheme mainly s ems om he op imiza ion o ONU
g ouping.
4. SIMULATION AND PERFORMANCE
EVALUATION
In his sec ion, we e alua e he pe o mances o ou p oposed
TAOG-GBDTA scheme compa ed o wo benchma ks. In
Benchma k 1, each ONU ope a es a i s maximum a ailable
peak da a a e, while in Benchma k 2, all ONUs a e con ig-
u ed o ope a e a he lowes a ailable da a a e (i.e., 25 Gb/s)
h oughou he simula ion. Fo bandwid h alloca ion, bo h
benchma ks use a classic mul icas scheduling mechanism,
i.e., he mul icas sha ing-based weigh ed ai queuing (MS-
WFQ) mechanism [32]. Fo his, we de elop a simula o in
MATLAB, which inco po a es ou enhanced downs eam
scheduling p o ocol o simula e he downs eam scheduling
p ocedu es in a lexible- a e PON.
A. Simula ion Condi ions
Ou simula ion conside s a lexible- a e PON sys em wi h a
a ying numbe o ONUs: 16, 32, and 64. The ONUs a e
ini ially di ided in o ou g oups based on hei a ailable peak
da a a es, which co espond o 25, 50, 75, and 100 Gb/s,
espec i ely. This a e combina ion can be achie ed by he
ine-g anula i y a e con ol demons a ed in [16]. The co e-
sponding peak da a a es a e assigned based on ypical ecei ed
op ical powe dis ibu ions and a e h esholds epo ed in
exis ing s udies [16,18]. Speci ically, he p opo ion o ONUs
ope a ing a 25, 50, 75, and 100 Gb/s is se o app oxima ely
1%, 31%, 58%, and 8%, espec i ely. A leas one ONU is
included in each g oup. The physical dis ance o each ONU
ollows a Gaussian dis ibu ion, whe e he mean is 10 km
and he a iance is 2 km. The downs eam a ic consis s o
8 unicas and 8 mul icas se ices, each modeled as a ypical
In e ne P o ocol ele ision s eam. The packe a i al imes
ollow a Poisson dis ibu ion, wi h all se ices sha ing he same
a e age packe a i al a e. Each se ice has a minimum a ic
a e o 10 Mbi /s, whe e he XGEM ame size is 1324 by es,
including an 8-by e XGEM heade [5,34]. The bu e size
o each se ice queue is se o 20 MB. Fu he mo e, ONU
join and lea e e en s ( igge ed by mul icas JOIN and LEAVE
eques s) o mul icas membe ships a e conside ed as a Poisson
p ocess. Acco dingly, he OMCI messages sen by he OLT o
assign o wi hd aw mul icas XGEM Po -IDs also ollow
a Poisson dis ibu ion [31–33]. Meanwhile, he numbe o
mul icas eques s o each mul icas se ice ollows a Zip
dis ibu ion. The Zip dis ibu ion is commonly used o model
mul imedia access pa e ns and e lec s he dynamic changes
in mul icas membe ships [31,32]. The eques p obabili y o
mul icas se ice i, deno ed by RPi, is exp essed as
RPi=C
iα,i∈3, (16)
whe e αde e mines he long- ail beha io o he dis ibu ion.
In ou simula ion, αis se o 1 o closely e lec ypical mul i-
cas access pa e ns [31,32]. The no maliza ion cons an Cis
de ined as
C=1
Pi∈3
1
iα
.(17)
Addi ionally, ano he ligh weigh OMCI se ice is also
conside ed, which includes pe iodic Managemen In o ma ion
Base (MIB) Upload and pe o mance moni o ing (PM)
epo s. Speci ically, he OLT b oadcas s a 56-by e MIB
Upload eques (including an 8-by e heade ) e e y 30 s [35].
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 09:44:35 UTC om IEEE Xplo e. Res ic ions apply.
