Towa ds E icien T anspo o Real-Time
Imme si e Applica ions o e Hyb id Ne wo ks
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 —Imme si e elep esence demands high da a a es
and low la ency, ye no single comme cial da a pa h eliably
mee s hese needs. Fine-g ained con en selec ion also emains
unde de eloped. This wo k p oposes a hyb id, mul i-pa h deli -
e y amewo k combining b oadcas and unicas in o a single
se ice. A ligh weigh base scene is b oadcas ia File Deli e y
o e Unidi ec ional T anspo (FLUTE), ensu ing no iewe e e
sees a ully blank scene, while iewe -speci ic enhancemen s
a e s ee ed o e unicas . An open-sou ce es bed is eleased
o in es iga e he impac o ne wo k impai men s, ins umen
common p o ocols, and enable ep oducible expe imen s. On
high-quali y olume ic ideo (up o 100k poin s pe ame a
30 ames pe second), he hyb id design (i) keeps la ency below
40 ms while scaling quali y wi h unicas bandwid h, (ii) educes
se e and ne wo k load compa ed o pu e unicas , and (iii)
masks ypical wi eless loss pa e ns wi h only 15% Fo wa d
E o Co ec ion (FEC) o e head. These indings show ha
ea ing b oadcas and unicas as complemen a y channels is
c ucial o scalable Ex ended Reali y (XR) se ices.
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
Imme si e communica ion, anging om olume ic elep-
esence and ee- iewpoin spo s o collabo a i e Ex ended
Reali y (XR), aims o s eam dynamic poin cloud ideo o
head-moun ed displays wi h low la ency and high quali y.
While 5G/6G oadmaps en ision such holog aphic se ices
wi h slicing and i ualiza ion, eal-wo ld deploymen s e-
main limi ed by bandwid h-hung y, la ency-sensi i e s eams
a e sing lossy, conges ed ne wo ks [1], [2].
Exis ing solu ions each add ess only pa o he p oblem:
Dynamic Adap i e S eaming o e HTTP (DASH) and HTTP
Li e S eaming (HLS) o e adap i i y bu incu high la ency;
Web Real-Time Communica ion (WebRTC) achie es sub-
200 ms delay bu s ill consumes subs an ial bandwid h; File
Deli e y o e Unidi ec ional T anspo (FLUTE) b oadcas s
e icien ly a scale bu lacks pe sonaliza ion [3]–[7]. Wo k on
hyb id b oadcas -unicas a chi ec u es shows p omise o 2D
ideo bu has ye o be ex ended o olume ic media. This
wo k closes he gap wi h a hyb id, mul i-pa h anspo ha
b oadcas s a base scene o e FLUTE and unicas s iewe -
speci ic enhancemen s ia DASH. The goal is o main ain a
ne e -blank scene in eal ime while scaling o many iewe s
and adap ing quali y o a ailable bandwid h. This pape con-
ibu es: (i) an open-sou ce es bed o e alua e eal- ime ol-
ume ic s eams o e hyb id pa hs1, (ii) a uni ied me ic sui e
cap u ing la ency, h oughpu , quali y, and ame d ops, (iii)
a head- o-head e alua ion o WebRTC, WebSocke , DASH,
and FLUTE unde ne wo k impai men s, (i ) and a hyb id,
mul i-pa h design ha balances scalabili y and pe sonaliza ion.
Figu e 1 summa izes he concep : a base desc ip ion is b oad-
cas o all, while enhancemen s a e deli e ed on demand o e
unicas . The ollowing sec ions desc ibe ela ed wo k, sys em
a chi ec u e, me ics and me hodology, expe imen al esul s,
conclusion, and u u e wo k.
II. RELATED WORK
Th ee lines o esea ch ha con e ge on he goal o a
hyb id, mul i-pa h anspo o eal- ime imme si e media a e
su eyed in his sec ion: s eaming p o ocols, imme si e-media
deli e y sys ems, and hyb id unicas -b oadcas a chi ec u es.
A. P o ocols o S eaming
Con en ional HTTP Adap i e S eaming (HAS) p o ocols
(DASH, HLS) di ide ideo in o mul i-second segmen s and
ely on T ansmission Con ol P o ocol (TCP), incu ing li e
la encies abo e 5 s [3]–[5], [8]. E en low-la ency a ian s
(Low La ency DASH (LL-DASH), Low-La ency HLS (LL-
HLS)) emain abo e 1 s due o TCP se up, segmen bu e ing,
and head-o -line blocking [9], [10]. Addi ionally, WebSocke
and hype ex ans e p o ocol (HTTP)/2 Push s eam media
o e pe sis en TCP connec ions o educe la ency [11], [12].
Use Da ag am P o ocol (UDP)-based p o ocols add ess TCP’s
limi a ions. Secu e Reliable T anspo adds Au oma ic Repea
Reques o sub-second li e s eaming in p o essional se ups,
hough web b owse s do no suppo i [13]. WebRTC is he
de- ac o choice o sub-second deli e y, combining Real- ime
T anspo P o ocol aming wi h applica ion-le el conges ion
con ol [14]. Al hough o iginally designed o pee - o-pee ,
se e -side Selec i e Fo wa ding Uni s a chi ec u es enable
scaling o many clien s [15]. Finally, QUIC and he eme ging
Media o e QUIC T anspo (MOQT) combine UDP’s low de-
lay wi h mul iplexing and publish/subsc ibe abs ac ions [16],
[17]. Ea ly p o o ypes achie e WebRTC-like la ency, bu spec-
i ica ions and suppo emain nascen [10].
1h ps://gi hub.com/idlab-disco e /Mul i-pa h-XR
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. End- o-end hyb id deli e y concep o eal- ime olume ic media. Con en is cap u ed ei he by a 360° came a o by a olume ic cap u e de ice
capable o p oducing poin -cloud ideo. A e p ocessing on an XR se e , he media a e dis ibu ed o e wo complemen a y pa hs: he con en ional In e ne
P o ocol b oadband ne wo k and a 5G b oadcas link. The s eamed scene can hen be consumed on Vi ual Reali y headse s, Mixed Reali y headse s and
phones wi h Augmen ed Reali y unc ionali y.
B. Imme si e Media Deli e y Sys ems
Real- ime olume ic s eaming equi es much highe band-
wid h han 2D ideo while mee ing simila la ency con-
s ain s. Ea ly wo k adap ed 2D s eaming, an de Hoo
e al. encoded each objec a mul iple quali ies and deli -
e ed hem wi h DASH-based six deg ees-o - eedom (6DoF)
s eaming, which selec s quali y based on use posi ion, ield
o iew (FOV) and bandwid h [18]. Howe e DASH incu s
highe delay han UDP-cen ic anspo s, mo i a ing o he
designs. VR2Ga he enables mul i-pa y elep esence wi h
mul iple TCP-based anspo s [19]. De F ´
e e al. imp o ed
upon his concep wi h an open-sou ce con e encing pla o m
using D aco-encoded poin clouds o e WebRTC, achie ing
∼160 ms end- o-end la ency while using Mul iple Desc ip ion
Coding (MDC) [6], [20]. Li eVV sepa a es backg ound and
o eg ound, applies poin decima ion and olume ic-awa e
adap i e bi a e s eaming (ABR), sus aining 24 ames pe
second (FPS) wi h ≤350 ms la ency [21]. LL-Spa se p edic s
he FOV o p e- e ch only isible da a [22]. New da ase s
such as UVG-VPC and MazeLab enable ealis ic e alua ion
o such echniques [23], [24]. Toge he , hese wo ks highligh
how p o ocol choice, con en op imiza ion, and p edic i e de-
li e y enable 6DoF s eaming. Howe e , none exploi mul iple
deli e y pa hs concu en ly.
C. Hyb id Unicas -B oadcas A chi ec u es
Hyb id s eaming educes pe -use bandwid h by o loading
common da a o a b oadcas channel, while ese ing unicas
o pe sonalized e inemen s. Fo example, a scene’s s a ic
backg ound o base laye can be b oadcas , while dynamic
o eg ounds o enhancemen s use unicas . Haems e al. in e-
g a e and demons a e 5G e es ial b oadcas wi h o e - he-
op, 2D linea ideo s eaming, achie ing sub-second li e la-
ency and signi ican ly educing unicas a ic ia FLUTE wi h
Fo wa d E o Co ec ion (FEC) [7], [25]. Ad anced Tele i-
sion Sys ems Commi ee 3.0 and 3 d Gene a ion Pa ne ship
P ojec adop compa able schemes [26], [27]. These s udies
Clien
Se e
Ing ess
Files
Gene a o
WebRTC
Ing ess
FLUTE
WebSocke
DASH
WebRTC
Decode
PLY
DRACO
Decode
PLY
DRACO
P ocesso
Scaling
Ro a ion
Sampling
Encode
PLY
DRACO
Eg ess
FLUTE
WebSocke
DASH
WebRTC
Bu e
One bu e pe
ep esen a ion pe
objec
Playback
VR headse
Sc een
Fig. 2. Modula pipeline o eal- ime olume ic media.
show ha hyb id a chi ec u es e ec i ely educe bandwid h
and la ency o 2D sha ed con en . Howe e , hey s op sho o
olume ic media, and no open, ep oducible amewo k exis s
o hyb id olume ic s eaming, a gap his wo k add esses.
III. SYSTEM ARCHITECTURE
Figu e 2 depic s he modula pipeline, wi h mos ly sym-
me ical se e (blue) and clien (g een) pa hs. Bo h consis o
in e changeable blocks o codecs, p ocessing, and anspo .
I needed, he decode , p ocesso and encode o he se e
can be skipped al oge he .
a) Ing ess and Decode : The se e inges s ei he p e-
encoded da ase s ( om iles) o li e s eams (WebRTC) and
decodes hem in o poin clouds. Bo h se e and clien use he
same decode , suppo ing Polygon File Fo ma and D aco.
b) P ocesso : This op ional block scales, o a es, down-
samples, o combines poin clouds. I can also spli he s eam
in o mul iple disjoin poin clouds o ABR o MDC.
c) Encode : P ocessed ames a e encoded wi h a se-
lec able codec be o e ansmission.
d) Eg ess (Se e ): Encoded ames a e mul iplexed
ac oss one o mo e p o ocols: (i) FLUTE b oadcas s wi h
block-le el FEC o e UDP using IP mul icas , (ii) WebSocke
(Socke .IO) pushes con en o e TCP, (iii) DASH, as he con-
en ional ABR app oach, pulls ames as mp4 segmen s using
HTTP, and (i ) WebRTC p o ides a unicas UDP anspo
as a eal- ime app oach. The design is p o ocol-agnos ic, and
XR se e
Rou e s
Clien s
Fig. 3. Ne wo k opology: each box is a node. The numbe o se e s, clien s,
and ou e s is con igu able.
u u e wo k could add MOQT o ine-g ained con en and
quali y p io i iza ion.
e) Ing ess (Clien ): Clien s mi o he se e eg esses:
hey open WebSocke , WebRTC, FLUTE, o DASH sessions
and eed decoded ames in o a minimal playback bu e .
Unlike con en ional ABR, his DASH ing ess uses only
an Exponen ially Weigh ed Mo ing A e age-based es ima o
wi hou bu e -based smoo hing o educed la ency.
This pa h-awa e implemen a ion, w i en in Rus , suppo s
hyb id expe imen s by allowing any anspo combina ion a
un ime. Mul i h eading enables ame n+1 o decode while
decoding o ame nis s ill in p og ess, p e en ing conges ion
on one pa h om blocking o he s.
IV. EVALUATION METRICS AND METHODOLOGY
This sec ion ou lines he expe imen al es bed, ollowed by
he me ics used o e alua e each anspo .
A. Expe imen al Se up
All ials a e conduc ed in a Minine en i onmen on an
In el®Co e™ i7-1365U wi h 32 GB RAM unning Ubun u
24.04 LTS [28]. The emula ed ne wo k opology is shown
in Figu e 3, whe e each box ep esen s one node on he
Minine opology. Linux c en o ces bandwid h, delay, and
packe -loss p o iles indi idually pe link displayed on he
igh side o he ou e s. Each scena io uns o en minu es,
wi h he i s and las minu e disca ded o exclude s a -up
and ea -down a i ac s, lea ing eigh minu es o s eady-s a e
measu emen s. The se e s eams poin clouds as ames in a
loop aken om he longd ess sequence om he MPEG
PCC da ase [29]. Table I lis s he a es o he downsampled
e sions o his sequence. All sequences a e encoded wi h he
D aco codec. The h ee smalle e sions con ain disjoin poin
se s o he 100k e sion, enabling bo h adi ional ABR and
MDC expe imen s. Fo example, b oadcas ing 15k poin s as a
base desc ip ion while unicas ing he 25k desc ip ion combines
TABLE I
BITRATE OF LONGDRESS AT DIFFERENT QUALITIES (DRACO-ENCODED,
30 FPS).
Numbe o poin s Visual quali y Mbps
15k Low 16.57 ±0.05
25k Medium-low 26.38 ±0.09
60k Medium 58.17 ±0.23
100k High 91.96 ±0.40
TABLE II
THROUGHPUT AND LATENCY AT 200 MBPS BANDWIDTH CAP FOR
100K-POINT STREAM.
P o ocol Th oughpu (Mbps) La ency (ms)
DASH 118.58 22.33
WebSocke 117.83 21.39
WebRTC 100.70 17.18
FLUTE 97.66 17.99
in o a 40k econs uc ed desc ip ion. Loss o packe s on a
single pa h he e o e deg ades quali y g ace ully a he han
in e up ing playback.
B. Me ics
Running on a single hos enables clock synch oniza ion
o accu a e la ency measu emen s, a he cos o inc eased
compu a ional load. La ency is measu ed as he ime be ween
he momen a ame is igge ed o be sen using he selec ed
eg ess a he se e and he momen he ame is ully
ecei ed a he clien . This la ency includes packe iza ion,
FEC, anspo and depacke iza ion and excludes he ime o
p ep ocessing, encoding and decoding he poin cloud. As
such, he me ic measu es he ull la ency o he anspo
p o ocol. In addi ion, he ame a e, poin cloud encoding and
decoding ime, CPU and memo y usage, he numbe o poin s
ecei ed and o he ele an me ics, such as he h oughpu pe
link, a e measu ed.
V. EXPERIMENTAL RESULTS
The e alua ion p oceeds in wo s ages. Sec ion V-A quan-
i ies he beha io o anspo s, es ablishing e e ence poin s
o la ency, bandwid h e iciency, and obus ness. Sec ion V-B
hen ac i a es mul iple eg esses simul aneously o demons a e
he bene i s and ade-o s o he p oposed hyb id design.
A. Single-Pa h Baselines
a) T anspo e iciency unde bandwid h ceilings: The
baseline scena io s eams he longd ess sequence in a loop
a 30 FPS wi h a ixed quali y o 100k poin s pe ame
encoded using D aco. This sou ce has a bi a e o 91.96 Mbps.
The links be ween se e and he clien a e h o led o
200 Mbps. Table II summa izes he measu ed h oughpu and
la ency pe p o ocol. UDP-based p o ocols (WebRTC, FLUTE)
incu only ∼8% o e head ela i e o he sou ce bi a e, while
TCP-based p o ocols (DASH, WebSocke ) incu ∼25% mo e.
When dec easing he bandwid h un il below he sou ce a e,
FPS deg ades g adually o TCP (due o conges ion con ol)
TABLE III
LATENCY OF FLUTE WITH DIFFERENT FEC SCHEMES.
FEC scheme La ency (ms)
No FEC 17.99
Rap o 19.36
Rap o Q 18.86
Reed-Solomon GF(28) 25.47
Reed-Solomon GF(28) (Unde Speci ied) 24.30
Fig. 4. F ame- a e esilience o FLUTE wi h 15% FEC unde inc easing
packe -loss a es. Decoding success a e is non-linea and can be in e ed
om he ame- a e cu e.
bu d ops ab up ly o UDP once he pa h sa u a es due o he
lack o conges ion con ol. I is hus ecommended o enable a
conges ion con ol scheme o WebRTC. La ency emains low
and s able o all es ed p o ocols when bandwid h su ices.
Among he p o ocols, only WebSocke implemen s backp es-
su e, keeping i s la ency mo e s able unde conges ion. FLUTE
h oughpu emains cons an unde mul iple clien s, unlike
unicas p o ocols which scale linea ly. Likewise, se e -side
p ocessing demands emain cons an o FLUTE, bu inc ease
wi h he numbe o ecei e s o unicas -based app oaches.
b) Loss esilience wi h o wa d-e o co ec ion: Adding
15% FEC o e head o he 100k FLUTE s eam, inc eases
h oughpu om 97.66 Mbps o 119.94 Mbps, bu signi ican ly
imp o es obus ness o packe loss. Table III shows he la ency
o ou common schemes [30]–[33]. Rap o Q achie ed he
bes ade-o be ween o e head and la ency while ole a ing
up o 15% loss, as shown in Figu e 4.
B. Hyb id Mul i-Pa h E alua ion
Fig. 5. A e age unicas h oughpu o he s eam in he se up as he unicas
bandwid h cap is inc eased. The h oughpu used by FLUTE o ansmi ing
he baseline quali y wi h a bandwid h cap o 200Mbps is shown o e e ence.
Fig. 6. A e age la ency o he se up o DASH as he unicas bandwid h cap
is inc eased. La ency depends on he selec ed quali y ep esen a ion, which
depends on he bandwid h cap.
A hyb id se up is e alua ed combining b oadcas and unicas
pa hs. The scene is spli in o 15k (base), 25k, and 60k quali ies,
wi h an addi ional 85k e inemen quali y (25k+60k). The 15k
base is b oadcas ia FLUTE o ensu e all clien s ecei e
a leas a minimal-quali y scene, while highe quali ies (25k,
60k, 85k) a e deli e ed adap i ely o e unicas using DASH.
This aims o educe pe -clien ne wo k load while enhancing
quali y by combining b oadcas and unicas s eams. This
scena io is compa ed wi h an adap i e, unicas -only s eam
using DASH ha can choose be ween ou quali ies as well
(15k, 25k, 60k, 100k). Figu e 5 shows unicas h oughpu
as he unicas bandwid h cap inc eases, wi h he b oadcas
pa h s eaming a 18.09 Mbps. Hyb id deli e y educes unicas
load by o loading he base laye o b oadcas , especially a
highe bandwid h capaci ies whe e DASH does no need o
s eam he 100k quali y, educing he unicas h oughpu by
∼18 Mbps. As men ioned in Sec ion III, he clien uns a
la ency- i s ABR s a egy, which lacks addi ional heu is ics
such as playback bu e -based ABR smoo hing. Figu e 6
shows ha la ency emains below 40 ms, hough he simplis ic
ABR implemen a ion occasionally oscilla es be ween quali ies
due o i s agg essi e, la ency- i s adap a ion. This is also
isible in Figu e 5 as i some imes ies o s eam a highe
quali y han he a ailable bandwid h allows. Compa ed o he
100k quali y-only s eam in Table II, la ency is mo e a iable
because i depends on which quali y DASH e ches. Using
a la ge playback bu e would allow mo e ABR heu is ics
and s abilize he quali y bu a he cos o highe playback
delay. In conclusion, he hyb id adap a ion mee s i s a ge s:
i keeps la ency below 40 ms, and always deli e s a leas he
15k baseline while upg ading o e all quali y, and signi ican ly
educes o e all ne wo k and se e load.
VI. CONCLUSION
This pape has p esen ed he i s open, ep oducible ame-
wo k o eal- ime imme si e applica ions o e hyb id ne -
wo ks, e alua ing ou anspo app oaches: WebRTC, Web-
Socke , DASH, and FLUTE. The con ibu ions a e h ee old:
(i) an open-sou ce es bed o poin cloud ideo s eams wi h
con ollable ne wo k impai men s, (ii) a uni ied me ic sui e
co e ing la ency, h oughpu , and o he me ics, and (iii) a
hyb id sende / ecei e ha b oadcas s a base desc ip ion while
unicas ing enhancemen s. Resul s e eal clea ade-o s: UDP-
based WebRTC and FLUTE add unde 10% bandwid h o e -
head and sus ain sub-30 ms la ency when link capaci y mee s
he sou ce a e, whe eas TCP-based WebSocke and DASH
incu ∼25% o e head. A modes 15% Rap o Q FEC o e head
masks andom losses on he b oadcas pa h wi hou no iceable
delay. The hyb id app oach ensu es he scene is ne e blank,
scales quali y wi h a ailable unicas bandwid h, and educes
he linea a ic g ow h o pu e unicas by o loading sha ed
con en o b oadcas . A ca e ully o ches a ed hyb id a chi ec-
u e can mee he dual demands o eal- ime la ency and mul i-
gigabi olume ic s eaming, pa ing he way o p ac ical XR
se ices on o hcoming 5G/6G deploymen s. Taken oge he ,
hese indings demons a e ha hyb id mul i-pa h o ches a ion
is essen ial o scalable imme si e communica ion. By ea ing
b oadcas and unicas as complemen a y, he sys em mee s
bo h he scalabili y and use expe ience demands.
VII. FUTURE WORK
Fu u e wo k will ex end he cu en ame-le el p o o ype o
objec -le el s eaming, whe e indi idual scene elemen s (e.g.,
backg ound, a a a s, p ops) a e assigned p io i ies and mapped
o app op ia e pa hs: high-saliency objec s o e unicas o
pe sonalized quali y, and low-saliency ones o e bes -e o
b oadcas . This ine g anula i y would enable be e ABR
decisions and allow p o ocols such as MOQT o en o ce pe -
objec p io i ies, educing bandwid h was e and imp o ing
pe cep ual quali y. Mo eo e , he Minine -based e alua ion
is limi ed o s a ic opologies, ens o lows and single-hos
compu a ion. Scena ios wi h mobili y, uplink con en ion, and
c oss- a ic emain unexplo ed. Thanks o i s open-sou ce
na u e, he amewo k can be eplayed on la ge es beds o
public clouds. Fu u e wo k will po i o la ge -scale pla o ms
and deploy i in 5G/6G mobili y ials o assess eal-wo ld
pe o mance unde dynamic condi ions.
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] Z. Liu e al., “Poin cloud ideo s eaming: Challenges and solu ions,”
IEEE Ne wo k, ol. 35, no. 5, pp. 202–209, 2021.
[2] R. S. Kalan, S. Clayman, and M. Sayi , “ DANE: Using i ualiza ion
o imp o ing ideo quali y wi h se e and ne wo k assis ed dash,” In .
J. Ne w. Manag., ol. 32, no. 5, p. e2209, 2022.
[3] I. Sodaga , “The mpeg-dash s anda d o mul imedia s eaming o e he
in e ne ,” IEEE Mul iMedia, ol. 18, no. 4, pp. 62–67, 2011.
[4] 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
[5] 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
[6] 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.
[7] 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.
[8] J. an de Hoo e al., “A u o ial on imme si e ideo deli e y: F om
omnidi ec ional ideo o holog aphy,” IEEE Communica ions Su eys &
Tu o ials, ol. 25, no. 2, pp. 1336–1375, 2023.
[9] ISO, “In o ma ion echnology - Mul imedia applica ion o ma
(MPEG-A) - Pa 19: Common media applica ion o ma (CMAF)
o segmen ed media,” Tech. Rep., 2024. [Online]. A ailable:
h ps://www.iso.o g/s anda d/85623.h ml
[10] 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.
[11] B.M. Taba i e al., “Low la ency li e ideo s eaming on and oid de ices
using web-socke ,” in ICCCNT ’17, 2017, pp. 1–6.
[12] J. an de Hoo e al., “An HTTP/2 push-based app oach o low-la ency
li e s eaming wi h supe -sho segmen s,” JNSM, ol. 26, no. 1, pp. 51–
78, 2018.
[13] M. Sha abayko e al., “The SRT P o ocol,” IETF, In e ne -D a
d a -sha abayko-s -01, 2021, wo k in P og ess. [Online]. A ailable:
h ps://da a acke .ie .o g/doc/d a -sha abayko-s /01/
[14] 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.
[15] H. T. Al es and, “O e iew: Real-Time P o ocols o B owse -Based
Applica ions,” RFC 8825, 2021. [Online]. A ailable: h ps://www. c-
edi o .o g/in o/ c8825
[16] 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/
[17] J. Iyenga and M. Thomson, “QUIC: A UDP-Based Mul iplexed
and Secu e T anspo ,” RFC 9000, 2021. [Online]. A ailable:
h ps://www. c-edi o .o g/in o/ c9000
[18] 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.
[19] I. Viola e al., “V 2ga he : A collabo a i e, social i ual eali y sys em
o adap i e, mul ipa y eal- ime communica ion,” IEEE Mul iMedia,
ol. 30, no. 2, pp. 48–59, 2023.
[20] M. De F ´
e e al., “Demons a ing adap i e many- o-many imme si e
elecon e encing o olume ic ideo,” in P oc. ACM MMSys ’24, 2024,
pp. 453–458.
[21] K. Hu e al., “Li e : Human-cen e ed li e olume ic ideo s eaming
sys em,” IEEE In e ne o Things Jou nal, pp. 1–1, 2025.
[22] J. Ouelle e and A. Ben aleb, “LL-Spa se: Low-la ency 6-do ield o
iew p edic ion,” in P oc. ACM MMSys ’25, 2025, pp. 57–67.
[23] G. Gau ie e al., “U g- pc: Voxelized poin cloud da ase o isual
olume ic ideo-based coding,” in QoMEX ’23, 2023, pp. 244–247.
[24] J. Ouelle e, J. S. Sidhu, and A. Ben aleb, “MazeLab: A la ge-scale
dynamic olume ic poin cloud ideo da ase wi h use beha io aces,”
in P oc. ACM MMSys ’25, 2025, pp. 298–304.
[25] C. Haems e al., “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.
[26] J. Lee e al., “Ip-based coope a i e se ices using a sc 3.0 b oadcas and
b oadband,” IEEE T ans. on B oadcas ing, ol. 66, no. 2, pp. 440–448,
2020.
[27] D. Silha y e al., “3GPP Rel-17 5G media s eaming and 5G b oadcas
powe ed by 5G-MAG e e ence ools,” in P oc. ACM MHV ’23, 2023,
pp. 85–90.
[28] 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.
[29] E. d’Eon e al., “8i Voxelized Full Bodies - A Voxelized Poin
Cloud Da ase ,” ISO/IEC JTC1/SC29 Join WG11/WG1 (MPEG/JPEG)
inpu documen WG11M40059/WG1M74006, 2017, a ailable a
h ps://plenodb.jpeg.o g/pc/8ilabs/.
[30] A. Shok ollahi e al., “Rap o Fo wa d E o Co ec ion Scheme o
Objec Deli e y,” RFC 5053, 2007. [Online]. A ailable: h ps://www. c-
edi o .o g/in o/ c5053
[31] L. Minde e al., “Rap o Q Fo wa d E o Co ec ion Scheme o
Objec Deli e y,” RFC 6330, 2011. [Online]. A ailable: h ps://www. c-
edi o .o g/in o/ c6330
[32] V. Roca e al., “Reed-Solomon Fo wa d E o Co ec ion (FEC)
Schemes,” RFC 5510, 2009. [Online]. A ailable: h ps://www. c-
edi o .o g/in o/ c5510
[33] ——, “Simple Reed-Solomon Fo wa d E o Co ec ion (FEC)
Scheme o FECFRAME,” RFC 6865, 2013. [Online]. A ailable:
h ps://www. c-edi o .o g/in o/ c6865
