979-8-3315-9777-1/25/$31.00 ©2025 IEEE
Demons a ing Ligh pa h Ope a ion wi h he
OCATA Digi al Twin in Mul iband Op ical
Ne wo ks
P. Gonzalez
Op ical Communica ions G oup
(GCO)
Uni e si a Poli ècnica de Ca alunya
(UPC)
Ba celona, Spain
pol.gonzalez.p[email p o ec ed]
H. Shakespea -Miles
Op ical Communica ions G oup
(GCO)
Uni e si a Poli ècnica de Ca alunya
(UPC)
Ba celona, Spain
[email p o ec ed]
S. Ghas izadeh
Op ical Communica ions G oup
(GCO)
Uni e si a Poli ècnica de Ca alunya
(UPC)
Ba celona, Spain
mohammadsadegh.ghas [email protected]
M. Ruiz
Op ical Communica ions G oup
(GCO)
Uni e si a Poli ècnica de Ca alunya
(UPC)
Ba celona, Spain
ma c. uiz- ami [email protected]
S. Ba zega
Ba celona Supe compu ing Cen e
(BSC)
Ba celona, Spain
[email p o ec ed]
L. Velasco
Op ical Communica ions G oup
(GCO)
Uni e si a Poli ècnica de Ca alunya
(UPC)
Ba celona, Spain
[email p o ec ed]
Abs ac – In his wo k, he ou line o a li e
demons a ion o he mul iband OCATA
digi al win is p esen ed. This digi al win
enables in elligen unc ions such as quali y
o ansmission (QoT) es ima ion and
adap i e model uning, among o he
unc ionali ies. This demons a ion
highligh s he in eg a ion o deep neu al
ne wo ks (DNNs) o simula e end- o-end
mul iband ligh pa hs ac oss C+L+S spec al
bands. A li e deploymen on a i ualized
es bed a UPC, Ba celona, showcases he
OCATA DT's abili ies in wo main scena ios.
A endees will engage wi h an in e ac i e
dashboa d o explo e how he DT es ima es
QoT and e ines i s in e nal models using
eleme y da a. This demons a ion aims o
illus a e he p ac ical bene i s o digi al
wins o au onomous ne wo k ope a ion,
pe o mance assu ance, and cos -e icien
scaling o nex -gene a ion op ical
in as uc u e.
Keywo ds: Op ical digi al win, ne wo k au oma ion,
quali y o ansmission es ima ion
I. OVERVIEW
Digi al Twins (DTs) a e popula ools o he
ope a ion and au oma ion o op ical ne wo ks. A DT
is essen ially a high- ideli y i ual model o a
physical sys em [1]. They can be used o many
di e en pu poses wi hin he op ical ne wo k
including he p edic ion o pe o mance, de ec ion
o anomalies, and help in o m decision-making o
ope a ion. They a e especially aluable due o he
na u e o he complex beha iou s hey a e able o
model. As ne wo ks o en use mul iple spec al
bands (C+L+S) o maximize bandwid h, he mo e
channels added and ansmission dis ances
inc eased, signal impai men s become mo e
p onounced which can lead o deg aded
pe o mance.
Among he a ious app oaches o DT modelling
op ical ansmission sys ems, machine lea ning
(ML) has shown pa icula p omise o DT
de elopmen due o i s abili y o app oxima e
complex beha iou s. The OCATA DT [1] is one ha
has been p oposed o using ML echniques o
a ious ne wo k ope a ion applica ions such as ligh
pa h p o isioning and ailu e managemen [3]. One
o OCATA DT’s applica ions is o p ecisely p edic
he p e- o wa d e o co ec ion (FEC) bi e o a e
(BER) in mul iband (MB) op ical ansmission
sys ems, whe e non-linea impai men s (NLI) such
as in e -channel s imula ed Raman sca e ing plays a
signi ican ole.
OCATA DT simula es he p opaga ion o op ical
signals in he ime domain, speci ically he in-phase
(I) and quad a u e (Q) componen s ac oss mul iple
op ical spans. The app oach employs DNNs
g ounded in i s -p incipal models, which a e p e-
ained and s o ed in a dedica ed model da abase.
These DNNs a e hen combined o build an end- o-
end (e2e) ligh pa h model, o med by linking span-
speci ic DNNs acco ding o he ligh pa h’s ou e
cha ac e is ics, such as span leng h and channel. To
educe he numbe o p e- ained models needed a
limi ed se o e e ence channels (RCh) a e selec ed
[4]. A ea u e composi ion me hod is used o
gene alize his model, enabling he es ima ion o
ansmission cha ac e is ics o any channel ac oss
he C+L+S bands by le e aging he ou pu s om he
e e ence models.
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/ICTON67126.2025.11125283
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 08:49:48 UTC om IEEE Xplo e. Res ic ions apply.
Fu he wo ks showed he uning o end- o-end
ligh pa h DNN models as an addi ional building
block o he MB OCATA DT [5]. He e, eleme y
da a was used o he uning p ocess p oposing his
new block in he MB OCATA DT a chi ec u e o he
main enance and accu acy o he p e- ained span
models.
The li e demons a ion o he MB OCATA DT
in eg a es many o he ea u es and unc ionali ies
a o emen ioned o p esen a eal- ime demons a ion
o he OCATA DT and how i unc ions. The
demons a ion aims o showcase he p ac ical
bene i s o DT-d i en ne wo k in elligence,
including how ML-enhanced models can be used o
he managemen o op ical ne wo ks, bu also how
he ine- uning and main enance o he DT i sel is
ca ied ou . As nex -gene a ion se ices place
inc easingly s ingen demands on op ical anspo
sys ems, DTs like OCATA eme ge as essen ial ools
o ensu ing eliabili y and e iciency. The
demons a ion will unde line how in elligen DTs
can assis in p o isioning and pe o mance
assu ance, con ibu ing o a scalable and cos -
e icien ne wo k.
II. INNOVATION
In his demons a ion, we aim o showcase he
capabili ies o he MB OCATA DT. This DT
le e ages machine lea ning o p o ide a i ual
ep esen a ion o op ical ansmission sys ems,
enabling in elligen da a-d i en decision-making
inco po a ing eedback om he ne wo k o he
ine- uning o models.
This demons a ion has been designed o esona e
wi h he ICTON audience, especially hose
in e es ed in au onomous ne wo k ope a ion,
in elligen moni o ing, and he in eg a ion o AI/ML
in o op ical in as uc u e. The li e es bed a UPC
in Ba celona will highligh he dynamic beha io o
he OCATA DT in a eal- ime se ing, illus a ing i s
ole no only in p edic ing ne wo k pe o mance bu
also in main aining i s own accu acy ia con inuous
model uning using eleme y eedback.
III. DEMO CONTENT AND
IMPLEMENTATION
A. Goals
The goals o his demons a ion can be di ided
in o gene al goals and echnical goals.
Fi s ly, he gene al goal o his demons a ion is o
p o ide people wi h an unde s anding o DTs and
how hey can be an impo an oo o op ical
ne wo ks. The demons a ion will allow o people
o in e ac and p omo e con e sa ions su ounding
DTs and how hey can be used in u u e ne wo ks.
The echnical goals o his demons a ion a e o
p o ide people wi h he unde s anding o he
OCATA DT, including he p inciples and
echnologies ha make i unc ion, and how i can be
used in op ical ne wo ks. This will be achie ed by
he li e demons a ion o wo OCATA DT unc ions:
i) QoT es ima ion
ii) OCATA model uning
B. Tes bed Se up
The OCATA DT will un in a Vi ual Machine
(VM) deployed in UPC in as uc u e using
OpenS ack as i ualiza ion pla o m and Ubun u
Se e 24.04.2 LTS as ope a ing sys em. A isual
ep esen a ion o he es bed o be used in he
demons a ion is shown in Fig. 1.
All he componen s o he OCATA DT ha e been
implemen ed in Py hon 3.12.3. The OCATA DT
exposes a REST API implemen ed using he Flask
lib a y ha o e s he capabili ies o OCATA o
ex e nal en i ies. A Redis ins ance is used as a
message b oke o he communica ion among he
in e nal componen s o OCATA. The Teleme y DB
has been implemen ed by means o an In lux DB
2.7.1 ins ance and he Model DB uns a MongoDB
6.0.23 ins ance. Each componen uns in an
independen Docke con aine and is deployed using
he Docke Compose ool. Bo h Teleme y and
In en o y DB ha e been p e iously popula ed wi h
da a o enhance he ep oducibili y o he
expe imen s. A Py hon clien sc ip simula ing a
So wa e-De ined Ne wo king (SDN) Con olle
has been de eloped o acili a e he in e ac ion
be ween he a endees and he sys em. This clien
ea u es he ollowing ope a ions:
i) es ima ion o he QoT o a ligh pa h,
ii) es ablishmen o a ligh pa h igge ing he
au oma ic model uning
iii) he decommission o an es ablished
ligh pa h.
OCATA DT
Model DB
Teleme y DB
OCATA VM
OCATA
OCATA API
Clien
Fig. 1: OCATA demons a ion es bed se up
C. OCATA
This sec ion p o ides echnical de ails on he
OCATA DT and i s a chi ec u e which a e necessa y
o unde s anding he ou pu s he demons a ion will
p oduce.
OCATA ep esen s IQ cons ella ion samples X as
sequences o symbols x
∈
X, whe e each symbol
co esponds o one o m cons ella ion poin s (CPs)
in an m-QAM signal. These samples a e summa ized
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 08:49:48 UTC om IEEE Xplo e. Res ic ions apply.
in o a se o cons ella ion ea u es Y, wi h each Yi
desc ibing he p ope ies o CP i. The ea u e
ex ac ion (FeX) p ocess uses Gaussian Mix u e
Models (GMMs) o model each CP as a bi a ia e
Gaussian dis ibu ion. Each ea u e ec o Yi =
[µI,µQ,σI,σQ,σIQ] includes he mean posi ions (µI,µQ),
a iances (σI,σQ), and co a iance (σIQ) o he
symbols a ound he mean.
Two samples (X1, X2) can be compa ed by
calcula ing he Euclidean dis ance be ween hei
ea u es Y1 and Y2, as shown by:
1,2=‖1− 2‖2 (1)
Addi ionally, he p e-FEC BER can be es ima ed
om he ea u es Y. The pa ame e Φiou ep esen s
he p obabili y o a symbol om CP i alling ou side
i s de ec ion a ea Ai. This is compu ed as:
= 1 − ⊂ ~ (2)
Finally, he es ima ed p e-FEC BER is ob ained by
a e aging Φiou o e all CPs, assuming G ay coding:
p e-FEC BER ~
∙!"#$∑Φ
' (3)
The scena io in Fig. 2 shows an MB op ical
ne wo k wi h op ical ansponde s and ampli ie s,
whe e EDFAs a e used o C and L bands and
TDFAs o he S band. A SDN con olle handles
ligh pa h p o isioning and eleme y collec ion,
while he OCATA digi al win ope a es alongside o
suppo QoT es ima ion and ailu e managemen . An
example ligh pa h be ween si es A and Z is
illus a ed.
OCATA includes a da abase o p e- ained span
DNN models o quickly build ligh pa h models
which is pe o med in he OCATA MB block. A
de ailed iew o his block and i s componen s is
shown in Fig. 3. The OCATA MB es ima es ea u es
o some o he selec ed CPs using DNN models o
ela ed RChs a ge channels. Then, es ima ed
ea u es a e passed h ough ano he DNN model,
called ea u e composi ion o ob ain he ea u es o
he selec ed CPs o he a ge channel.
Once comple ed he cons ella ion Recons uc ion
block es ima es he ea u es o he non-selec ed CPs
which is ollowed by a NLI mi iga ion p ocedu e.
Imp o ing he QoT in mul iband op ical ne wo ks
equi es adap ing de ec ion a eas o mi iga e NLI,
bu his is compu a ionally in ensi e. OCATA MB
e icien ly compu es nea -op imal de ec ion maps
du ing p o isioning, enhancing QoT and educing
blocking. The de ailed p ocedu e and algo i hms can
be ound in [4].
Howe e , using he ini ial cons uc ed models
in oduces some e o s due o de ice a ia ions,
aging e ec s, he use o e e ence channels ha may
no ma ch he ac ual channel, and di e ences in span
leng hs be ween aining and eal ne wo k spans.
While de ice and channel misma ches ha e limi ed
impac on QoT es ima ion a e p o isioning, span
leng h di e ences can no ably a ec model
accu acy.
TP TP
demux
mux
mux
mux
MB OA
demux
mux
MB OA
Tx
Rx
Tx
Rx
Si e Z
Si e A
SDN
Con olle
OCATA Digi al Twin
Teleme y /
Model DBs
Model Tuning Phase
check Tune
Model
S o e
models
Ligh pa h
OCATA
Digi al Twin
Algo i hms
QoT Es ima ion Phase
OCATA
MB
Sandbox
Domain
Fig. 2: OCATA digi al win a chi ec u e
DNNs
RC1 Fea u e
composi ion
Ligh pa h Ta ge Ch
QoT
Es ima ion
RC2
RCn
Map o NLI
mi iga ion
FeX
[Y]
Re e ence
Channel
selec ion
Cons .
Recons .
Ta ge Ch
Inpu Da a
Fig. 3: De ailed iew o OCATA MB block
In MB op ical ansmission, algo i hms a e applied
o he wo e e ence RChs adjacen o he alloca ed
channel, when he alloca ed channel is no i sel an
RCh. The QoT es ima ion is hen ob ained using eq
(3) on ea u es gene a ed h ough a ea u e
composi ion p ocess. Addi ionally, he ligh pa h
model buil om he span models o he nea es RCh
is sa ed in he model da abase. A e deploying a
ligh pa h, i s end- o-end model is es ablished and
equi es uning wi h eleme y da a. Fo his
pu pose, a Model Tuning block pe iodically checks
o signi ican de ia ions and i ound, he ligh pa h
model is upda ed by minimizing he e o de ined by
eq (1). The cause o he de ia ion can hen be
analysed o iden i y po en ial deg ada ions.
D. Implemen a ion
This demons a ion highligh s he co e capabili ies
o he OCATA DT in assis ing an SDN con olle
wi h op ical ne wo k managemen , ocusing
speci ically on QoT es ima ion and model uning.
These wo phases e lec he in elligen decision-
making and adap i e lea ning p ocesses cen al o
he OCATA amewo k.
The i s phase o he demo in ol es he es ima ion
o QoT o a p ospec i e ligh pa h. Upon ecei ing
a eques om he SDN con olle , he OCATA DT
collec s ele an in o ma ion abou he ne wo k
componen s ha o m he end- o-end pa h. This
includes s a ic pa ame e s such as link leng hs,
ampli ie con igu a ions, and o he physical
p ope ies. The DT hen igge s he p ocedu e on
his in o ma ion in o de o p edic whe he he
p oposed ligh pa h can mee he equi ed QoT
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 08:49:48 UTC om IEEE Xplo e. Res ic ions apply.
h esholds. This p edic ion is e u ned o he
con olle o assis in pa h selec ion and se ice
p o isioning decisions. The es ima ed QoT se es as
a p oac i e measu e o a oid se ing up low-quali y
o un easible ligh pa hs.
In he second phase o he demons a ion, he ocus
shi s o he adap i e beha iou o he OCATA DT
h ough model uning. Once he ligh pa h has been
es ablished and is ope a ional, eleme y da a is
collec ed om he ne wo k. This eal-wo ld da a is
hen used o e ine he in e nal DNN model,
allowing i o be e e lec cu en ne wo k
condi ions and imp o e i s u u e p edic ions. The
demo will show a compa ison be ween he ini ial
QoT es ima ion (based on s a ic desc ip o s) and he
upda ed es ima ion (a e he model has been uned
wi h li e eleme y), illus a ing he model’s
lea ning and co ec ion p ocess.
The expec ed ou pu o he demo includes:
i) A isual compa ison o QoT es ima ions
be o e and a e model uning, demons a ing
imp o ed accu acy.
ii) De ails o inpu da a used in each es ima ion
phase
iii) Indica ions o how eleme y in luences he
in e nal model
i ) Clea e idence o he DT's eedback loop and
adap a ion mechanism.
The demo will be deli e ed ia an in e ac i e
dashboa d. This dashboa d p o ides all he ele an
inpu and ou pu in o ma ion om he OCATA DT.
A endees will see he incoming QoT eques s, he
co esponding es ima ions, he injec ion o
eleme y da a, and how he in e nal model adjus s.
The in e ace will be p esen ed using g aphs, ables,
and s a us indica o s o help cla i y he in e nal
wo kings o he DT wi hou equi ing deep echnical
knowledge.
A endees can in e ac wi h he demons a ion
di ec ly by igge ing new QoT es ima ions and
ini ia ing he model uning phase wi h eleme y
om a ious ne wo k condi ions. By adjus ing
inpu s and obse ing he esul ing changes,
pa icipan s will gain insigh s in o how he OCATA
DT p ocesses in o ma ion, adap s o e ime, and
suppo s sma e op ical ne wo k managemen .
We belie e ha his demons a ion will p o ide a
clea , concise, and impac ul iew o how he
OCATA DT enhances decision-making in op ical
ne wo ks h ough in elligen , da a-d i en p ocesses.
IV. CONCLUSIONS
This demons a ion will show he li e ope a ion o
he OCATA DT h ough a VM deployed a he UPC
p emises in Ba celona, Spain. The demons a ion
ocuses on wo main cases o ope a ion, he i s
being QoT es ima ion, and he second, model ine
uning o he OCATA DT DNN models. The demo
is designed o p omo e con e sa ion abou he ole
o digi al wins in op ical ne wo ks, and p o ide
a endees wi h an unde s anding o he OCATA DT
and how i wo ks h ough he in e ac i e
demons a ion.
AKNOWLEDGEMENTS
The esea ch leading o hese esul s has ecei ed
unding om he Eu opean Union's Ho izon Eu ope
esea ch and inno a ion p og amme SEASON (G.A.
101096120), The Eu opean Commission MSCA-
DN NESTOR (G.A.101119983) and om he
ICREA Ins i u ion.
REFERENCES
[1] D. Wang, Z. Zhang, M. Zhang, e al., “The ole
o digi al win in op ical communica ion: aul
managemen , ha dwa e con igu a ion, and
ansmission simula ion,” IEEE Commun. Mag.
59(1), 133–139 (2021).
[2] D. Sequei a e al., “OCATA: A Deep Lea ning-
based Digi al Twin o he Op ical Time
Domain,” JOCN, ol. 15, pp. 87-97, 2023.
[3] M. De igili e al., “Applica ions o he OCATA
Time Domain Digi al Twin: om QoT
Es ima ion o Failu e Managemen ,” JOCN, ol.
16, 2024.
[4] S. Ghas izadeh e al., “DT-Assis ed Ligh pa h
P o isioning and Nonlinea Mi iga ion in
C+L+S MB Op ical Ne wo ks,” Senso s, ol. 24,
2024.
[5] S. Ghas izadeh, N. Cos a, M. Ruiz, A. Napoli, J.
Ped o, and L. Velasco, “C ea ing and Tuning
Mul iband Op ical T ansmission Digi al Twin
Ligh pa h Models,” accep ed in In e na ional
Con e ence on Op ical Ne wo k Design and
Modeling (ONDM), 2025.
Au ho ized licensed use limi ed o: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on Oc obe 06,2025 a 08:49:48 UTC om IEEE Xplo e. Res ic ions apply.
