OCATA Op ical Mul iband Time Domain Digi al Twin:
Suppo ing Ligh pa h P o isioning wi h NLI Mi iga ion in
Mul iband Op ical Ne wo ks
Sadegh Ghas izadeh1, Ma c Ruiz1, and Luis Velasco1,*
1 Op ical Communica ions G oup (GCO), Uni e si a Poli ècnica de Ca alunya (UPC), Ba celona, Spain;
*e-mail: luis. ela[email p o ec ed]du
ABSTRACT1
Mul iband (MB) op ical ansmission en ails la ge nonlinea impai men s (NLI) a ec ing op ical channels
di e en ly. The e o e, accu a e quali y o ansmission (QoT) es ima ion du ing ligh pa h p o isioning is s ic ly
equi ed o ensu e he easibili y o he op ical ansmission o e he compu ed ligh pa hs. In his pape , we e iew
he OCATA MB op ical ime domain digi al win o p o ide as and accu a e QoT es ima ion, while educing he
complexi y associa ed wi h NLI. In addi ion, we show how NLI noise can be mi iga ed by op imizing pa ame e s
a he ecei e side du ing ligh pa h p o isioning o u he op imize ne wo k pe o mance.
Keywo ds: Mul iband op ical ansmission, Op ical ne wo k digi al win, Nonlinea impai men s mi iga ion.
1. INTRODUCTION
Op ical anspo ne wo ks a e essen ial o add ess he g owing capaci y demands d i en by beyond 5G (B5G)
echnologies. To his end, he ITU-T launched a ocus g oup o de ine a ne wo k oadmap o 2030 [1]. Mul iband
(MB) op ical ne wo ks a e conside ed a key solu ion o ex end legacy ne wo k capaci y by exploi ing addi ional
spec al bands beyond C and L [2]. Howe e , MB ansmission inc eases nonlinea impai men s (NLI), pa icula ly
due o in e -channel s imula ed Raman sca e ing (ISRS), which complica es Quali y o T ansmission (QoT)
es ima ion [3]. T adi ional Rou ing and Spec um Alloca ion (RSA) me hods, while conside ing QoT cons ain s,
ace scalabili y challenges in MB scena ios due o ISRS e ec s and he high numbe o channels. Machine
Lea ning (ML) echniques, pa icula ly Digi al Twins (DTs), o e a p omising solu ion. The OCATA DT, based
on deep neu al ne wo ks (DNNs), has been p oposed as a eliable and low-complexi y app oach o QoT es ima ion
and NLI mi iga ion, accu a ely p edic ing he p e-FEC BER o op ical connec ions [4]. A key challenge emains
he gene a ion o la ge aining da ase s, as expe imen al MB da a is sca ce and simula ions a e ime-consuming.
To add ess his, an e icien me hod o sol e he nonlinea Sch ödinge equa ion has been p oposed, signi ican ly
educing compu a ion ime o da ase gene a ion [5]. Building on ou p e ious wo k [6], whe e OCATA was used
o QoT es ima ion and nonlinea mi iga ion du ing p o isioning, his pape ex ends he me hodology o suppo
MB ansmission. We p esen OCATA-MB, which in eg a es scalable QoT es ima ion and NLI mi iga ion in o he
RSA p ocess, suppo ing ou e and channel assignmen wi hin SDN con olle s.
In his pape , we o e a synopsis o he algo i hms and he p ocedu es in oduced in ou p e ious wo k in [7].
2. MB Op ical Ligh pa h P o isioning
Ligh pa h p o isioning, in ol ing ou e and channel selec ion, is c ucial o au oma ed ne wo k ope a ion while
ensu ing equi ed QoT. This sec ion p esen s he MB scena io, in oduces he OCATA-MB digi al win o model
accu a ely model he ligh pa h conside ing ISRS e ec s and enhance QoT in MB op ical ansmission. Fig. 1 (a)
depic s he mul iband (MB) op ical ansmission scena io conside ed, whe e ansponde s a si es A and B can
ope a e ac oss mul iple bands (C, L, S). Signals om hese ansponde s a e mul iplexed and ansmi ed h ough
op ical ibe s, wi h sepa a e op ical ampli ie s (OAs) pe band—such as EDFAs o C and L bands and TDFAs o
he S band—alongside wa eband (de)mul iplexe s. This s uc u e enables he expansion o ne wo k capaci y using
MB ansmission. Howe e , ISRS a ec s QoT by une enly ans e ing powe om highe o lowe equencies,
esul ing in a ying le els o nonlinea impai men s ac oss channels. As shown in Fig. 1 (b), channels in he S
band gene ally expe ience highe p e-FEC BER han hose in he C and L bands, limi ing hei each. Hence, while
C and L bands a e mo e sui able o long-dis ance connec ions, o e using hem isks deple ing a ailable esou ces
and inc easing u u e blocking p obabili ies.
To add ess his, a mo e balanced channel assignmen s a egy in ol es selec ing channels wi h accep able p e-
FEC BERs, e en i sligh ly highe , o op imize o e all esou ce u iliza ion. Howe e , his equi es e alua ing he
QoT o all a ailable channels, which can be compu a ionally in ensi e. Mo eo e , some connec ions may s ill be
blocked i no channel mee s he QoT h eshold. The ollowing sec ions p opose e icien solu ions o op imal
1 The esea ch leading o hese esul s has ecei ed unding om he Eu opean Commission MSCA-DN NESTOR (G.A.101119983) and he
Ho izon Eu ope SEASON (G.A. 101096120); and om he ICREA Ins i u ion.
979-8-3315-9777-1/25/$31.00 ©2025 IEEE
2025 25 h Anni e sa y In e na ional Con e ence on T anspa en Op ical Ne wo ks (ICTON) | 979-8-3315-9777-1/25/$31.00 ©2025 IEEE | DOI: 10.1109/ICTON67126.2025.11125019
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 08:58:47 UTC om IEEE Xplo e. Res ic ions apply.
channel assignmen and mi iga ion echniques o enhance he QoT o channels ha would o he wise ail o mee
he equi ed pe o mance.
# Channel
R1
R2
1
S C L
|Ch|
P e-FEC BER
TP TP
demux
mux
mux
mux
MB OA
demux
mux
MB OA
(a)
(b)
Si e A Si e B
ch1 ch2
Tx
Rx
Tx
Rx
DNNs
RC1
Fea u e
composi ion
(c) OCATA MB
Ligh pa h Ta ge Ch
QoT
Es ima ion
RC2
RCn Map o NLI
mi iga ion
FeX
[Y]
RC
selec ion
Cons .
Recons .
Ta ge Ch
Fig. 1 O e iew o he conside ed MB scena io (a) illus a i e pe o mance o MB op ical ansmission (b) and Main building blocks o he
OCATA-MB ime-domain digi al win, adop ed om [7].
The classical OCATA a chi ec u e o C-band, uses DNNs ained on a e e ence channel (RCh) o model signal
p opaga ion by ex ac ing key noise- ela ed ea u es om cons ella ion poin s (CPs), enabling e icien QoT
es ima ion. Howe e , applying his di ec ly o MB scena ios is challenging due o he ISRS e ec , which impac s
channels di e en ly and would equi e an imp ac ically la ge numbe o DNN models. To add ess his, he
OCATA-MB a chi ec u e (Fig. 1 (c)) in oduces a ew ca e ully selec ed RChs ep esen ing di e en spec al
egions, balancing model complexi y and accu acy. A Fea u e Composi ion block es ima es CP ea u es o any
channel ac oss C-, L-, and S-band by le e aging he p opaga ed ea u es o hese RChs, ensu ing scalabili y. As in
he classical app oach, a Cons ella ion Recons uc ion block econs uc s ea u es o non-p opaga ed CPs,
allowing accu a e QoT p edic ion in MB op ical ne wo ks.
Imp o ing he quali y o ansmission (QoT) in mul iband op ical ne wo ks equi es adap ing de ec ion a eas o
cons ella ion poin s in he ecei e o mi iga e NLI. Howe e , inding op imal de ec ion a eas o all poin s is
compu a ionally demanding. OCATA-MB ackles his by e icien ly compu ing nea -op imal de ec ion maps
du ing ligh pa h p o isioning, enhancing QoT and allowing connec ions ha would o he wise be ejec ed. The
algo i hms ela ed o NLI mi iga ion has been accu a ely in es iga ed in [4]. Wi h e icien QoT es ima ion and
NLI mi iga ion, a channel selec ion algo i hm is designed o iden i y he bes channel assignmen o a gi en ou e.
I checks i any channel mee s QoT equi emen s, selec s he op imal one i possible, o applies mi iga ion
echniques o imp o e QoT when no channel ini ially sa is ies he h eshold.
3. P oposed Me hod
OCATA-MB p e- ains DNN models o selec ed RChs, hen du ing p o isioning, i es ima es a ge channel
ea u es, op imizes de ec ion a eas o NLI mi iga ion, and selec s he bes channel assignmen ensu ing QoT. The
RCh selec ion p ocess analyzes symbol dispe sion ea u es (σI, σQ), which a e he a iance o symbols o I and Q
axes, i s piece-wise linea unc ions o iden i y key a ia ion poin s, clus e s hese cu -poin s, selec s cen oids as
candida e RChs, and i e a i ely adjus s he numbe o RChs un il BER es ima ion accu acy mee s a a ge h eshold
wi hou excessi e model complexi y. The ea u e composi ion algo i hm iden i ies he wo adjacen RChs o a
a ge channel and linea ly in e pola es hei p opaga ed ea u es o es ima e he a ge channel’s ea u es. The NLI
mi iga ion me hod op imizes CP de ec ion a eas by di iding he IQ plane in o small squa es, compu ing symbol
p obabili ies o each CP, and assigning each squa e o he CP wi h he highes p obabili y. The esul ing mapping
is sen o he Rx o e icien symbol decoding du ing ligh pa h p o isioning.
The p oposed heu is ic add esses he online MB-RSA p oblem o ligh pa h p o isioning by i e a ing o e
candida e ou es and a ailable channels o ind a easible solu ion assis ed by OCATA-MB. Fo each ou e, i
e ie es a ailable channels, selec s he op imal one using p e- ained DNN models, and compu es de ec ion a eas
o NLI noise mi iga ion i equi ed. The channel selec ion p ocess builds end- o-end DNN models om e e ence
channels, p opaga es inpu ea u es, and es ima es p e-FEC BER o each candida e channel. Channels a e
explo ed based on p e-compu ed BER cu es, using ea u e composi ion and econs uc ion o e ine BER
es ima ion. I a channel mee s he h eshold, i is selec ed; o he wise, he sea ch con inues. When no channel
sa is ies he QoT equi emen , he bes channel is enhanced h ough NLI mi iga ion. A g id-based de ec ion a ea
op imiza ion is applied, assigning de ec ion egions o CPs o minimize e o s. The p e-FEC BER is e-es ima ed,
and i wi hin limi s, he channel and de ec ion a eas a e selec ed. I no , no easible solu ion is ound. The lowcha
o he p ocedu es is shown in Fig. 2. De ailed explana ions o he p ocedu es can be ound in [7].
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 08:58:47 UTC om IEEE Xplo e. Res ic ions apply.
Ligh pa h eques
<s c, des , capaci y>
Selec m(and BER h )
ha ensu es capaci y
Compu e ksho es
pa hs P<s c, des >
wi h a ailable channels
Is P={}? Block eques
YES
NO
Ge and emo e
sho es pa h p om P
Find ch wi h min BER
A={de aul }
BER(ch) ≤
BER h ? NO
YES
RSA ound
<p, ch, A>
Find op imal
de ec ion a eas A
o <p, ch>
BER(ch, A)
≤ BER h ?
NO
YES
DT-assis ed
NLI noise
mi iga ion
DT-assis ed MB-RSA
Fig. 2 P oposed DT-assis ed MB-RSA p ocedu e; adop ed om [7]
4. EVALUATION
This sec ion e alua es OCATA-MB and he MB-RSA algo i hm, co e ing DNN aining, RCh selec ion,
nonlinea mi iga ion, and i s in eg a ion. A MATLAB-based simula o [5] gene a ed 16QAM@32GBd IQ
cons ella ions o a C+L+S WDM sys em wi h 337 channels spaced a 50 GHz. Signals p opaga e h ough 70-100
km ibe spans wi h 0 dBm launch powe , including ISRS and ibe nonlinea i ies, modeled ia he nonlinea
Sch ödinge equa ion using Runge-Ku a. EDFAs/TDFAs a e idealized wi h ixed gain and noise igu es. A o al
o 1,000 samples we e gene a ed o DNN aining, alida ion, and es ing. DNN models use 20 inpu ea u es,
wo hidden laye s (12 anh neu ons), and 20 ou pu s, wi h speci ic models ained pe RCh and link con igu a ion.
We begin by selec ing he RChs used o ligh pa h p o isioning. Fig. 3 illus a es he alues o ea u es o
wo ou e CPs ac oss all channels. A piecewise linea i wi h 6 segmen s p o ides a good app oxima ion, esul ing
in 7 RChs. Howe e , he exac posi ion o cu -poin s a ies o each ea u e and link con igu a ion. Fo ins ance,
he 4 h cu -poin shi s be ween channel indices 153 and 174, as seen in Fig. 3.
0.14
0.16
0.18
0.2
0.22
0.24
0.26
1 51 101 151 201 251 301
0.14
0.16
0.18
0.2
0.22
0.24
0.26
1 51 101 151 201 251 301
0.14
0.16
0.18
0.20
0.22
0.24
0.26
1 51 101 151 201 251 301
(a)
σ
I
, (-3+3i) (b)
σ
Q
,
(-3+3i)
Channel
Fea u e alue
(c)
σ
I
, (+3-3i)
1 segmen
4 segmen s
6 segmen s
337
173
154 153
Fig. 3 Value o selec ed σ ea u es and CPs s channel index and piecewise linea i ing wi h 1, 4 and 6 segmen s.
To assess QoT es ima ion accu acy, Fig. 4 compa es simula ed p e-FEC BER wi h alues es ima ed by
Algo i hm 6 wi hou NLI mi iga ion (using squa e de ec ion a eas). Resul s o RCh 97 and non-p opaga ed
channels 180 and 310 a e shown o e a ying dis ances. The es ima ions closely ma ch simula ions, con i ming
high accu acy.
1E-9
1E-7
1E-5
1E-3
1E-1
1 2 3 4 5 6 7 8 9 10 11 12 13
1E-9
1E-7
1E-5
1E-3
1E-1
1 2 3 4 5 6 7 8 9 10 11 12 13
1E-9
1E-7
1E-5
1E-3
1E-1
1 2 3 4 5 6 7 8 9 10 11 12 13
Pe
-
FEC BER
RCh. 97
(S-band)
Ch. 180
(C-band)
Ch. 310
(L-band)
# spans
Simula ion
Es ima ed
Fig. 4 E olu ion o P e-FEC BER as a unc ion o he numbe o spans o se e al channels.
Fig. 5 shows p e-FEC BER es ima es wi h and wi hou NLI mi iga ion using op imized de ec ion a eas o h ee
channels ac oss di e en spans by di iding he IQ plane in o k = 10,000 small g ids. Resul s con i m p e-FEC
BER imp o emen o all channels, ega dless o dis ance. The maximum ansmission each a ies by channel,
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 08:58:47 UTC om IEEE Xplo e. Res ic ions apply.
highligh ing he need o OCATA-MB o selec sui able channels du ing p o isioning. NLI mi iga ion ex ends
each by a leas one span.
# spans
0.0001
0.001
0.01
0.1
123456789
0.0001
0.001
0.01
0.1
123456789
1.E-4
1.E-3
1.E-2
1.E-1
123456789
Es ima ed P e-FEC BER
w/o NLI mi iga ion
w/ NLI mi iga ion
(a) Ch. 1 (S band) (b) Ch. 150 (C band) (c) Ch. 337 (L band)
Fig. 5 Real and es ima ed p e-FEC BER wi h squa ed and op imized de ec ion a eas s # spans o ch. 1 (a), 150 (b), and 337 (c).
The OCATA-assis ed on-line MB-RSA algo i hm was e alua ed in a Py hon-based simula ion using a Spanish
10-node co e ne wo k. A s a ic a ic model was applied, wi h 1,000 p e-gene a ed connec ion eques s p ocessed
sequen ially. Rou es we e selec ed om up o 3 sho es pa hs be ween sou ce and des ina ion nodes, and bo h
wi h and wi hou NLI mi iga ion. Fig. 6 (a) shows he numbe o blocked eques s, and Fig. 6 (b) p esen s he
blocking a io e sus he numbe o eques s. The i s - i app oach led o he highes blockages, wi h a 34%
blocking a io a e 1,000 eques s. Using OCATA-assis ed MB-RSA educed blocked eques s by 54%, lowe ing
he a io o 15.6%. Wi h NLI noise mi iga ion, blockages d opped by an addi ional 56%, achie ing a inal blocking
a io o jus 6.8%, esul ing in o e 80% o al educ ion.
0
50
100
150
200
250
300
350
400
0 200 400 600 800 1000
0%
5%
10%
15%
20%
25%
30%
35%
40%
0 200 400 600 800 1000
w/o NLI mi iga ion
w/ NLI mi iga ion
Fi s Fi
# eques s # eques s
# blocked
Blocking Ra io
56%
54%
34.4%
15.6%
6.8%
(a) (b)
Fig. 6 Numbe o demands blocked (a) and blocking a io e olu ion (b) s demand numbe .
5. CONCLUSIONS
A digi al win-assis ed ligh pa h p o isioning app oach is p oposed o mul iband op ical ne wo ks o ensu e he
equi ed QoT h ough e ec i e channel and ou e assignmen . OCATA-MB uses DNNs o op ical signal
p opaga ion, de ining e e ence channels o educe he numbe o DNN models and in e pola ing ea u es o a ge
channels. Nonlinea mi iga ion op imizes de ec ion a eas a he ecei e o imp o e QoT and educe connec ion
blocking. Simula ion esul s show high accu acy in QoT es ima ion, pa icula ly p e-FEC BER, and signi ican
imp o emen s in pe o mance. The p oposed app oach educed he blocking a io by o e 50% compa ed o he
adi ional i s - i algo i hm, wi h an addi ional 50% imp o emen when nonlinea mi iga ion was applied,
demons a ing i s e ec i eness.
REFERENCES
[1] B. Da and M. Ca ugi, “Rep esen a i e use cases and key ne wo k equi emen s o ne wo k 2030,” Tech. ep., ITU-T FG-
NET2030, 2020.
[2] L. Velasco, F. Cugini, R. Casellas, M. Nakagawa, G. Wellb ock, and X. Chen, “In oduc ion o he JOCN Special Issue
on Ad ances in Mul i-Band Op ical Ne wo ks,” J. Op . Commun. Ne w. 15, AIMON1-AIMON2 2023.
[3] M. Meh abi e al., “Mul i-band elas ic op ical ne wo ks: in e -channel s imula ed aman sca e ing-awa e ou ing,
modula ion le el and spec um assignmen ,” J. Ligh . Technol., ol. 39, pp. 3360-3370, 2021.
[4] M. Ruiz, D. Sequei a, and L. Velasco, “Deep Lea ning -based Real-Time Analysis o Ligh pa h Op ical Cons ella ions
[In i ed],” J. Op . Commun. Ne w., ol. 14, pp. C70-C81, 2022.
[5] P. Kha e, N. Cos a, J. Ped o, A. Napoli, F. A panaei, J. Comellas, M. Ruiz and L. Velasco, “SSMS: A Spli S ep Mul iBand
Simula ion So wa e,” in p oc. IEEE ICTON, 2023.
[6] S. Ghas izadeh, P. Kha e, M. Ruiz and L. Velasco, “Using he OCATA Digi al Twin o Imp o e QoT o Op ical
Connec ions in Mul iband Op ical Ne wo ks,” in p oc. IEEE ONDM, 2024.
[7] Ghas izadeh, S.; Kha e, P.; Cos a, N.; Ruiz, M.; Napoli, A.; Ped o, J.; Velasco, L. Digi al Twin-Assis ed Ligh pa h
P o isioning and Nonlinea Mi iga ion in C+L+S Mul iband Op ical Ne wo ks. Senso s 2024, 24, 8054.
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 08:58:47 UTC om IEEE Xplo e. Res ic ions apply.
