Demons a ion o Viewpo -Awa e Hyb id
B oadcas -Unicas S eaming o Volume ic Video
Caspe Haems, Ma hias De F ´
e, Tim Wau e s, Filip De Tu ck
Depa men o In o ma ion Technology
Ghen Uni e si y - imec
Ghen , Belgium
caspe [email p o ec ed], [email p o ec ed], [email p o ec ed], [email p o ec ed]
Abs ac —Real- ime s eaming o olume ic ideo equi es
low la ency and high bandwid h, making i ha d o scale
e icien ly wi h unicas deli e y alone. This wo k p esen s a
mul i-pa h anspo amewo k ha combines b oadcas and
unicas o enable scalable, bandwid h-e icien deli e y o in-
e ac i e olume ic con en . A base laye is ansmi ed ia
b oadcas using File Deli e y o e Unidi ec ional T anspo ,
while enhancemen laye s a e e ched on demand ia Dynamic
Adap i e S eaming o e HTTP based on he iewe ’s iewpo .
The app oach educes edundan ansmissions and enables ine-
g ained adap a ion wi hou comp omising in e ac i i y. An end-
o-end implemen a ion is demons a ed wi h eal- ime iewpo
acking, dynamic quali y swi ching, and li e moni o ing unde
a ying bandwid h cons ain s, imp o ing eliabili y, achie ing
sub-40 ms la ency and educing pe -clien unicas load by
18 Mbps (15%) pe objec a he highes quali y.
Index Te ms—Volume ic ideo, hyb id b oadcas -unicas ,
mul i-pa h anspo , eal- ime s eaming, imme si e media
I. INTRODUCTION
Real- ime imme si e applica ions such as holog aphic con-
e encing and ee- iewpoin elep esence equi e high da a
a es and ul a-low la ency [1]. Howe e , unicas -based imme -
si e s eaming does no scale e icien ly o la ge audiences and
o en s uggles unde ne wo k impai men s. To add ess his
challenge, a mul i-pa h deli e y amewo k o eal- ime ol-
ume ic media is in oduced. This a chi ec u e decouples base
and enhancemen con en ac oss b oadcas and unicas pa hs,
educing edundancy and enabling pe sonalized, bandwid h-
awa e adap a ion. The demons a ion ocuses on a hyb id
ins ance o his app oach, using b oadcas ia File Deli e y
o e Unidi ec ional T anspo (FLUTE) o deli e a base scene
and unicas ia Dynamic Adap i e S eaming o e HTTP
(DASH) o iewe -speci ic enhancemen laye s. Viewpo -
awa e logic ac i a es highe -quali y ep esen a ions when a
ele an objec en e s he use ’s ield o iew (FOV), ensu ing
e icien bandwid h use wi hou comp omising pe cei ed qual-
i y. The end- o-end pipeline suppo s eal- ime isualiza ion,
o e ing bo h a i s -pe son iew and a global o e iew wi h
posi ion acking and isualiza ion o he use ’s us um (i.e.,
isible 3D egion). A li e dashboa d displays sys em pe o -
mance, including la ency, bandwid h, and poin cloud esolu-
ion. Ne wo k condi ions a e con igu able a un ime wi hin
Minine , allowing o s ess- es ing and alida ion o adap i e
beha io [2]. Figu e 1 summa izes he sys em. The se up is
open sou ce and suppo s ep oducible expe imen a ion1.
II. RELATED WORK
Resea ch on eal- ime imme si e media deli e y spans
p o ocol design, adap i e sys ems, and hyb id a chi ec u es.
Con en ional s eaming p o ocols such as DASH and HTTP
Li e S eaming (HLS) in oduce mul i-second delays due o
segmen bu e ing and T ansmission Con ol P o ocol (TCP)
o e head [3], [4]. E en low-la ency a ian s such as Low
La ency DASH (LL-DASH) and Low-La ency HLS (LL-HLS)
emain abo e 1 second [5]. In con as , Web Real-Time
Communica ion (WebRTC) and Media o e QUIC T anspo
(MOQT) use Use Da ag am P o ocol (UDP) o achie e sub-
second deli e y, hough MOQT emains in ea ly s ages o
adop ion [6], [7]. Volume ic s eaming sys ems such as six
deg ees-o - eedom (6DoF) DASH and WebRTC-based con-
e encing pla o ms adap con en o he use ’s posi ion and
FOV, bu ely on unicas -only deli e y [8], [9]. Hyb id ans-
po a chi ec u es educe pe -use load by b oadcas ing sha ed
con en while unicas ing pe sonalized e inemen s [10]. A p io
demons a ion in 2D s eaming used FLUTE wi h Fo wa d
E o Co ec ion (FEC) o educe la ency and bandwid h [11].
Howe e , no open demons a ion exis s o eal- ime hyb id
deli e y o olume ic media. This demons a ion b idges ha
gap wi h a no el and unique ep oducible sys em ha com-
bines b oadcas and unicas anspo o imme si e con en ,
ea u ing eal- ime adap a ion based on iewpo dynamics.
III. SYSTEM ARCHITECTURE
This sys em demons a es he hyb id mul i-pa h a chi ec u e
in oduced by Haems e al. [1]. The sys em ollows a modula
a chi ec u e ha suppo s eal- ime olume ic media s eam-
ing o e mul iple anspo pa hs wi h mul iple p o ocols. Con-
en can be pa i ioned based on quali y laye s, objec seman-
ics, o spa ial egions, enabling lexible mappings on o b oad-
cas and unicas deli e y channels. This design gene alizes o
a ious imme si e use cases, including scalable elep esence
and in e ac i e mul i-use scena ios. Mos ele an componen s
a e implemen ed in Rus 2using a mul i- h eaded design wi h
p o ocol-agnos ic in e aces. Each anspo p o ocol connec s
o a sha ed pipeline, allowing di e en deli e y mechanisms o
1. h ps://gi hub.com/idlab-disco e /Mul i-pa h-XR
2. h ps://www. us -lang.o g/
Sou ce XR se e
Rou e Rou e
5G b oadcas
ecei e
5G b oadcas
ansmi e
Volume ic
cap u e de ice
360 deg ee
came a
Clien
VR headse
MR headse
AR phone
Fig. 1. O e iew o a hyb id end- o-end deli e y sys em o eal- ime olume ic media. Con en is cap u ed using ei he a 360° came a o a olume ic
cap u e sys em gene a ing poin cloud ideo. A e p ocessing on an XR se e , he media a e dis ibu ed using wo complemen a y pa hs: a con en ional
In e ne P o ocol b oadband ne wo k and a 5G b oadcas channel. The esul ing scene can be ende ed on Vi ual Reali y headse s, Mixed Reali y de ices, o
sma phones wi h Augmen ed Reali y capabili ies.
be combined wi hou al e ing he co e logic. The modula a -
chi ec u e suppo s un ime adap a ion, con en p io i iza ion,
and p o ocol swi ching. On he ecei e side, decoded con en
is exposed h ough a sha ed in e ace, enabling in eg a ion
wi h di e se ende ing clien s. Fo ins ance, a headless ecei e
can suppo analy ics o s o age, while a g aphical on -end
may ende he scene in eal ime. The sys em sepa a es
con en acquisi ion, deli e y, and ende ing conce ns, allowing
indi idual componen s o e ol e independen ly and be eused
ac oss deploymen con ex s.
IV. DEMONSTRATION
The demons a ion showcases eal- ime olume ic s eam-
ing using a hyb id unicas -b oadcas deli e y pipeline and
p o ides quan i a i e e alua ion me ics. A base scene wi h
wo objec s is ansmi ed o e FLUTE using IP mul icas o
e icien b oadcas ing, while one o he objec s is enhanced
on demand ia DASH based on he iewe ’s iewpo . All
componen s un on a single hos wi h ne wo k cons ain s
emula ed using Minine .
A. Con en and Adap i i y Se up
The demons a ion uses he longd ess and ed and
black objec s om he MPEG Poin Cloud Comp ession
(PCC) da ase . In his demons a ion, only he longd ess
objec is enhancemen -enabled. Each objec consis s o 300
ames, co esponding o 10 seconds o con en a 30 ames
pe second (FPS), which is played in a con inuous loop. All
ames a e encoded using he D aco codec3. Each objec is
downsampled o 100k poin s pe ame. Wi hou conside ing
ne wo k p o ocol o e head, his esul s in app oxima e bi a es
o 92 Mbps o longd ess and 84 Mbps o ed and
black. To enable adap i i y, each objec is spli in o h ee
non-o e lapping subse s o 60k, 25k, and 15k poin s, ol-
lowing a Mul iple Desc ip ion Coding (MDC) s uc u e. This
se up allows he clien o compose addi ional i ual quali ies
om one objec by combining mul iple subse s. Fo example,
s eaming bo h he 15k and 25k subse s yields an e ec i e
40k quali y, while combining all h ee es o es he o iginal
100k quali y. These combina ions p o ide ine g anula i y
wi hou equi ing addi ional encoding. In he hyb id scena io,
wo deli e y pa hs a e used. Pe objec , he 15k base laye is
ansmi ed o e FLUTE, ensu ing ha all ecei e s ob ain a
minimal ep esen a ion. Enhancemen laye s a e deli e ed o e
DASH as 25k and 60k subse s, as well as a p e-me ged 85k
subse ha combines he 25k and 60k subse s. When displayed
wi h he 15k b oadcas ed base laye , hese enhancemen laye s
enable adap i e quali ies up o 40k, 75k, o 100k, depending
on iewpo isibili y and a ailable bandwid h. Figu e 2 shows
he unicas h oughpu o he longd ess objec unde
a ying bandwid h cons ain s. The FLUTE s eam o he
15k baseline (longd ess only) consumes oughly 18 Mbps
and is cons an . Depending on he a ailable unicas bandwid h
and selec ed quali y, he hyb id se up (in ed) educes DASH
h oughpu by up o 18 Mbps (15%) compa ed o a pu e
unicas scena io (in blue). Figu e 3 displays he co esponding
end- o-end la ency. While he cu en ly used DASH imple-
men a ion lacks smoo hing and does no coo dina e wi h he
FLUTE s eam, he se up emains consis en . Hyb id deli e y
achie es lowe unicas load and compa able o imp o ed
la ency. Compa ed o Haems e al. [1], he adap i e bi a e
s eaming logic has been e ined, leading o imp o ed s abili y
and ewe oscilla ions unde a ying bandwid h condi ions.
B. In e ac i e Beha io and Viewpo Logic
The demons a ion s eams he enhancemen -enabled
longd ess objec while also s eaming a second objec
( ed and black) ha is no enhancemen -enabled. This
con as allows use s o obse e how he sys em esponds
when he hyb id longd ess objec en e s he iewpo , ig-
ge ing he DASH enhancemen s eam, while he ed and
black objec emains a he 15k base quali y ega dless o
3. h ps://google.gi hub.io/d aco/
Fig. 2. A e age unicas h oughpu as a unc ion o he unicas bandwid h
cap. The 18 Mbps FLUTE s eam o 15k o he longd ess objec is shown
as a e e ence.
Fig. 3. A e age la ency as a unc ion o unicas bandwid h cap. La ency is
in luenced by he selec ed ep esen a ion and a ailable capaci y.
isibili y. The clien ende s he scene om wo synch onized
pe spec i es: a i s -pe son iew (use iewpo ) and a hi d-
pe son o e iew ha displays he use ’s posi ion and iewing
us um. An example is shown in Figu e 4. Viewpo acking
de e mines which objec s in e sec wi h he use ’s us um.
When he enhancemen -enabled objec becomes isible, he
DASH playe e ches addi ional quali y laye s o augmen he
FLUTE base. Once he objec exi s iew, he DASH s eam is
paused, e e ing o he baseline.
C. Moni o ing and Adap a ion
A eal- ime dashboa d isualizes key pe o mance me ics,
including la ency ( ypically unde 40 ms), bandwid h usage
pe p o ocol, displayed poin coun pe objec (15k-100k),
and sys em esou ce usage. Bandwid h cons ain s and playe
iewpo can be changed du ing un ime o demons a e he
sys em’s adap abili y. Unde educed bandwid h condi ions,
he DASH enhancemen laye is disabled, main aining a con-
sis en use expe ience h ough FLUTE allback. The sys em
achie es sus ained ame a es o 30 FPS wi h ne wo k la-
encies below 40 ms. Unde cons ained emula ed links, he
DASH s eam disables while he FLUTE laye emains isible
wi hou in e up ion.
V. CONCLUSION
This demons a ion p esen s a eal- ime hyb id deli e y
amewo k o olume ic ideo ha le e ages he comple-
men a y s eng hs o b oadcas and unicas . By decoupling
a eliable base laye om use -speci ic enhancemen s and
Fig. 4. Sc eensho showing he dual isualiza ion: global o e iew wi h he
playe and hei us um, wi h he use iew on he bo om igh . Con en ha
is in he use ’s iewpo is enhanced using DASH.
enabling iewpo -awa e ac i a ion, he sys em achie es sub-
40 ms la ency while educing unicas bandwid h and e i-
cien ly adap ing o bandwid h a iabili y. The se up illus a es
how e en unde cons ained o luc ua ing condi ions, use s
main ain a seamless imme si e expe ience. Li e moni o ing
p o ides isibili y in o he sys em’s pe o mance, o e ing a
anspa en and ep oducible en i onmen o u he esea ch.
The modula design suppo s mul iple p o ocols and un ime
econ igu a ion, making i a lexible open sou ce pla o m
o u u e expe imen a ion in scalable XR deli e y, adap i e
s eaming, and hyb id ne wo k a chi ec u es.
ACKNOWLEDGMENT
This wo k has been unded by he Eu opean Union (SPIRIT
p ojec , G an Ag eemen 101070672, h ps://www.spi i -
p ojec .eu/).
REFERENCES
[1] C. Haems e al., “Towa ds e icien anspo o Real-Time imme si e
applica ions o e hyb id ne wo ks,” in NoF ’25, 2025.
[2] N. Handigol e al., “Rep oducible ne wo k expe imen s using con aine -
based emula ion,” in P oc. ACM CoNEXT ’12, 2012, pp. 253–264.
[3] ISO, “In o ma ion echnology - Dynamic adap i e s eaming
o e HTTP (DASH) - Pa 1: Media p esen a ion desc ip ion
and segmen o ma s,” Tech. Rep., 2022. [Online]. A ailable:
h ps://www.iso.o g/s anda d/83314.h ml
[4] R. Pan os and W. May, “HTTP Li e S eaming,” RFC 8216, 2017.
[Online]. A ailable: h ps://www. c-edi o .o g/in o/ c8216
[5] Z. Gu el e al., “Media-o e -QUIC anspo s. low-la ency dash: a
dea hma ch es bed,” in P oc. ACM MMSys ’24, 2024, pp. 448–452.
[6] B. Jansen e al., “Pe o mance e alua ion o web c-based ideo con e -
encing,” SIGMETRICS Pe o m. E al. Re ., ol. 45, no. 3, pp. 56–68,
2018.
[7] L. Cu ley e al., “Media o e QUIC T anspo ,” 2024. [Online].
A ailable: h ps://da a acke .ie .o g/doc/d a -ie -moq- anspo /03/
[8] J. an de Hoo e al., “Towa ds 6DoF HTTP adap i e s eaming h ough
poin cloud comp ession,” in P oc. ACM MM ’19, 2019, pp. 2405–2413.
[9] M. De F ´
e e al., “Scalable MDC-based olume ic ideo deli e y o
eal- ime one- o-many WebRTC con e encing,” in P oc. ACM MMSys
’24, 2024, pp. 121–131.
[10] C. Haems e al., “Enabling adap i e and eliable ideo deli e y o e
hyb id unicas /b oadcas ne wo ks,” in P oc. ACM NOSSDAV ’24, 2024,
pp. 29–35.
[11] ——, “Real- ime demons a ion o low-la ency ideo deli e y ia hyb id
unicas -b oadcas ne wo ks,” in CNSM ’24, 2024, pp. 1–3.
