Measu emen 234 (2024) 114792
A ailable online 27 Ap il 2024
0263-2241/© 2024 The Au ho s. Published by Else ie L d. This is an open access a icle unde he CC BY-NC-ND license (h p://c ea i ecommons.o g/licenses/by-
nc-nd/4.0/).
Con en s lis s a ailable a ScienceDi ec
Measu emen
jou nal homepage: www.else ie .com/loca e/measu emen
Cha ac e izing in-band ull-duplex in b oadcas communica ions: F om ield
ials o a loopback channel model
Iñigo Bilbao a, Eneko I adie a,∗, Ma a Fe nandez a, Jon Mon alban a, Pablo Anguei a a,
Zhihong Hun e Hong b, Yiyan Wu b
aUni e si y o he Basque Coun y (UPV/EHU), Spain
bCommunica ions Resea ch Cen e Canada, On a io, Canada
ARTICLE INFO
Keywo ds:
Channel modeling
Cha ac e iza ion
Delay sp ead
Field ials
Full-duplex
IBFD
IDL
ITCN
k- ac o
Loopback
Sel -in e e ence
Signal isola ion
ABSTRACT
In-band ull duplex (IBFD) communica ions a e a po en ial solu ion o spec um sca ci y. IBFD communica ions
o e g ea e spec al e iciency han adi ional hal -duplex communica ions by ansmi ing and ecei ing on
he same equency channel. Howe e , IBFD ope a ion equi es o e coming he challenge o elimina ing he
sel -in e e ence coupled om he ansmi an enna o he ecei e subsys em. Knowledge o he cha ac e is ics
o he loopback p opaga ion channel makes i easie o cancel ou he sel -in e e ence. Ne e heless, complex
ield ials a e equi ed o adequa ely cha ac e ize loopback channels, which is s ill lacking in he ecen
li e a u e. This pape p oposes a measu emen campaign in a eal and ongoing b oadcas ansmission cen e
o cha ac e ize he main cha ac e is ics o loopback channels. The wo k p oposes a se o loopback channels
based on ield ials and an empi ical analysis o he mos ele an pa ame e s o he channel model, such as
he Dopple spec um, he delay sp ead, and he K- ac o .
1. In oduc ion
The inc eased demand o new applica ions and he augmen ed
consump ion o high-da a- a e mul imedia se ices d i e he pe manen
e olu ion o wi eless communica ion sys ems. Wi hou a doub , many
o he ecen ad ances a e ema kable, such as coding [1,2] o mul i-
plexing schemes [3,4]. The success o many o hese ad ances is closely
ela ed o he eliabili y o labo a o y es s and ield ials. Bo h o es
hei pe o mance in eal-wo ld en i onmen s and o lea n as much as
possible om he en i onmen in which hey a e deployed.
Conce ning he physical laye (PHY), al hough adi ional signal
p ocessing has con inuously imp o ed he pe o mance, i is al eady
e y close o he Shannon limi , so he imp o emen ma gin is mini-
mal [5]. In addi ion, due o spec um sca ci y, he limi a ions o using
he adio spec um mus also be conside ed when a ge ing sys em
e iciency [6]. Techniques such as in-band ull-duplex (IBFD) commu-
nica ions can be decisi e in mee ing u u e spec um needs [7–9].
An IBFD sys em simul aneously ansmi s and ecei es signals wi hin
he same adio equency (RF) channel. Thus, heo e ically, IBFD
could double he spec al e iciency i compa ed wi h hal -duplex
sys ems [10]. The inc eased spec al e iciency opens he doo o new
use cases and bands limi ed by he capaci y o cu en hal -duplex
∗Co esponding au ho .
E-mail add ess: [email p o ec ed] (E. I adie ).
sys ems. Se e al au ho s ha e al eady p oposed and e alua ed IBFD in
combina ion wi h di e en wi eless communica ion s anda ds. In pa -
icula , se e al measu emen campaigns ha e shown he applicabili y
and challenges o IBFD in di e se applica ions, such as Wi-Fi [11–
14], sa elli e communica ions [15,16], wea able applica ions [17], and
5G/6G [18–21].
One o he s anda d amilies ha ha e e ol ed signi ican ly o e
he las decade is e es ial b oadcas ing [22]. T adi ional DTT s an-
da ds a e one-way (i.e., downlink-only mode), limi ing he numbe o
possible se ices. This limi a ion has been add essed by pushing o
inco po a ing new IBFD-based solu ions. Recen examples include he
in-band dis ibu ion link (IDL) sys em and he in e - owe communi-
ca ions ne wo k (ITCN) p oposal. IDL is a one-way wi eless con en
dis ibu ion sys em ha eplaces he s udio- o- ansmi e link (STL)
ia mic owa e o ibe [23,24]. ITCN aims o in e connec all b oadcas
owe s o c ea e a communica ion ne wo k o con ol, moni o ing, da a
communica ion, localized da acas , and b oadcas se ices [25]. IBFD
communica ions will be used o manage he simul aneous ansmission
o he ongoing b oadcas con en , ITCN, and IDL da a and will open
he doo o u u e a chi ec u es [26]. Fig. 1 shows a ep esen a ion
o an ITCN implemen a ion using wo ansmission owe s. As he
h ps://doi.o g/10.1016/j.measu emen .2024.114792
Recei ed 20 Feb ua y 2024; Recei ed in e ised o m 8 Ap il 2024; Accep ed 25 Ap il 2024
Measu emen 234 (2024) 114792
2
I. Bilbao e al.
Fig. 1. Rep esen a ion o an ITCN applica ion combining DTT con en ansmission
and ITCN communica ions be ween TX1 and TX2.
igu e shows, wo se ices a e combined: ITCN con en ansmission
ha in e connec s he ansmi e s (i.e., TX1 and TX2) and each owe ’s
DTT con en deli e y o he su ounding use s.
Al hough IBFD-based applica ions ha e a p omising u u e, cancel-
la ion o he loopback signal o sel -in e e ence (SI) mus be add essed
in ealis ic en i onmen s. The loopback sel -in e e ence is caused by
he signals leaking om he node’s ansmission in o i s ecep ion chain.
As shown in Fig. 1, each ansmi e is simul aneously ansmi ing
(i.e., DTT con en ) and ecei ing (i.e., ITCN con en ). Some locally
ansmi ed DTT powe is leaked in o he ITCN an enna ecei e , gene -
a ing in e e ence. Gene ally, IBFD communica ions need cancella ion
modules o dec ease he impac o he loopback signal [27–29]. Se e al
cancella ion echniques a e conside ed blind o semi-blind, no equi -
ing es ima ing he channel impulse esponse, [30–32]. Ne e heless,
mos o hem inc ease hei pe o mance when he loopback channel
is known [33,34].
As au ho s in [35] desc ibed, SI cancella ion echniques can be
classi ied in o h ee main g oups: p opaga ion, analog, and digi al
domain cancella ion. Fi s , p opaga ion domain echniques a e ela ed
o he in as uc u e and he elemen s in ol ed in he communica ions
and can be di ided in o passi e and ac i e echniques. On he one hand,
he passi e echniques ocus on op imizing he an enna sys ems and
minimizing sel -in e e ence coupling [36]. On he o he hand, some
o he mos common ac i e echniques a e impedance uning, an enna
coupling ne wo ks, o c oss-pola iza ion con olling de ices. Al hough
ield ials ha e been conduc ed in a ious en i onmen s [37–40], all
s udies highligh he di icul y o measu emen due o complex access
and lack o cha ac e iza ion. In addi ion, he wo k in he li e a u e
has only cha ac e ized he powe isola ion be ween ansmi ing and
ecei ing an ennas.
Then, analog cancella ion echniques e e o he signal p ocessing
ca ied ou in he analog domain, mainly on he RF sec ions o he
ecei e chain. They can be de eloped in equency and ime domains
and o e a help ul al e na i e o digi al domain cancella ion mod-
ules wi h la ge dynamic ange equi emen s [41]. Typically, analog
cancella ion echniques imp o e he ansmi ed– ecei ed a io o a
equency band o a ound 15–40 dB. They a e usually implemen ed
wi h passi e cancella ion echniques o inc ease he pe o mance a e
o 80 dB [42–44]. Ne e heless, he success o he analog in e e ence
cancella ion algo i hms lies in he p ecise knowledge o he channel
impulse esponse [45–47]. The e o e, p ope channel es ima ion and
ep esen a i e e e ence channel models a e c i ical.
Digi al domain echniques e e o channel modeling and digi al
cancella ion. Gene ally, digi al cancella ion echniques ou pe o m ana-
log me hods [48]. So wa e-based p ocesses ha e ewe ope a ion limi s
han ha dwa e-based ones, and se e al s udies ha e demons a ed ha
i is possible o o e come he 50 dB pe o mance ba ie using only
digi al cancella ion [23,49]. Ne e heless, as in he case o analog
cancella ion, he accu acy o he loopback channel es ima ion is i al
o high-pe o ming in e e ence cancella ion sys ems [50–52].
Conside ing all he abo e, he e is a need o cha ac e ize he loop-
back channel o IBFD ansmi e s. Al hough some IBFD- ela ed e-
sea ch wo ks ha e used loopback channel models [27,30], he e is no
an accu a e cha ac e iza ion o he mos ele an aspec s o a wi eless
p opaga ion channel, such as he delay sp ead, K- ac o , and Dopple
spec um. Only ansmi e and ecei e sys em isola ion has been
s udied in p e ious wo ks [40]. The p esen s udy p oposes a comple e
cha ac e iza ion o a loopback channel model based on empi ical da a
ob ained om a measu emen campaign and combining a ious p opos-
als in p e ious wo ks [53–56]. To he au ho s’ knowledge, his pape
p esen s he i s cha ac e iza ion o a loopback channel model based
on expe imen al measu emen s in an ac ual and ongoing ansmission
in as uc u e. The wo k p esen s he me hodology ollowed du ing he
ield ials, he esul s ob ained when cha ac e izing he channel, and
an analysis o di e en pa ame e s ha cha ac e ize he channel, such
as he Dopple spec um, he delay sp ead, o he K- ac o . In summa y,
he echnical con ibu ions o his pape include he ollowing:
•A new me hodology o measu ing and cha ac e izing b oadcas
loopback channels.
•Cha ac e izing he loopback channel in e ms o he ampli ude
and he delay o he pa hs.
•An analysis o he ob ained delay sp ead and K- ac o alues.
•Cha ac e izing he Dopple spec um o he loopback channel in
b oadcas ansmi e s.
•Pe o mance e alua ion o he p oposed channel models unde an
ATSC 3.0 complian anscei e chain.
The es o he pape is o ganized as ollows. The nex sec ion
desc ibes he sys em model o he IBFD sys ems and he mul ipa h chan-
nels. Then, Sec ion 3p esen s he measu emen me hodology ollowed
du ing he ield ials. In Sec ion 4, he ob ained channel model is cha -
ac e ized, analyzed, and discussed. Sec ion 5p esen s he pe o mance
e alua ion o he p oposed channel models. Finally, he conclusions o
his pape a e summa ized in Sec ion 5.
2. Sys em model
This sec ion p esen s he IBFD sys em model p oposed in his wo k.
Fi s , he gene al concep s o he sys em model a e desc ibed, and la e ,
he apped delay line (TDL) p opaga ion channels a e p esen ed om
he pe spec i e o in-band ull-duplex communica ions.
2.1. Gene al concep s
In-band ull-duplex communica ions simul aneously handle da a
lows in bo h uplink and downlink di ec ions. The e o e, he nodes pa -
icipa ing in ull-duplex communica ions a e ecei e s and ansmi e s
simul aneously and wi hin he same bandwid h. Pa o he ansmi ed
signal will be leaked om he ansmi ing adia ion sys em o he
ecei ing an enna, gene a ing sel -in e e ence. F om he e on, we will
e e o he sel -in e e ence signal as he loopback signal (LBS). Thus,
he ecei ed signal comp ises wo pa s, one ela ed o he desi ed
Measu emen 234 (2024) 114792
3
I. Bilbao e al.
o o wa d signal (FWS) and he o he o he loopback signal. This
si ua ion in he equency domain can be desc ibed as ollows:
𝑌(𝑘) = 𝑋𝐹 𝑊 𝑆 (𝑘)⋅𝐻𝐹 𝑊 𝑆 (𝑘) + 𝜂⋅𝑋𝐿𝐵𝑆 (𝑘)⋅𝐻𝐿𝐵𝑆 (𝑘) + 𝑁0(𝑘),(1)
whe e 𝑋𝐹 𝑊 𝑆 (𝑘)and 𝑋𝐿𝐵𝑆 (𝑘)a e he desi ed o wa d signal and he
loopback signal, espec i ely. 𝐻𝐹 𝑊 𝑆 (𝑘)and 𝐻𝐿𝐵𝑆 (𝑘)a e he desi ed
o wa d signal channel esponse and he loopback channel esponse
in he 𝑘 h subchannel, espec i ely, 𝜂is he powe a io be ween he
o wa d and loopback signals, and 𝑁0is he Addi i e Whi e Gaussian
Noise (AWGN) noise.
The 𝑋𝐹 𝑊 𝑆 in o ma ion is decoded only a e elimina ing he loop-
back componen . Se e al signal cancella ion echniques exis o loop-
back elimina ion in analog and digi al domains. Howe e , hese me h-
ods necessi a e an accu a e es ima ion o he loopback channel o
achie e op imal pe o mance [27–29,33,34]. In gene al, a model o he
loopback signal is gene a ed, and he channel e ec s (such as a enu-
a ion, delay, and phase shi ) a e applied. Each empo al componen
( ap) o he c ea ed signal o cancel can be desc ibed as ollows:
𝑠(𝑡) = 𝛼⋅𝑥𝐿𝐵𝑆 (𝑡−𝜓)𝑒𝑗𝜙,(2)
whe e 𝛼is he ampli ude, 𝜓is he ime delay and 𝜙is a phase shi e .
Subsequen ly, he cancella ion signal is combined wi h he ecei ed
signal in he ime domain:
𝑦′(𝑡) = 𝑦(𝑡) − 𝑠(𝑡),(3)
whe e 𝑦′is he emaining signal a e cancella ion. In an ideal scena io,
he objec i e o he cancella ion signal is o elimina e he impac o he
loopback signal (exp essed as 𝜂⋅𝑥𝐿𝐵𝑆 (𝑡) ∗ ℎ𝐿𝐵𝑆 (𝑡)). Howe e , seeking
an exac eplica o he coupled signal is imp ac ical. Consequen ly, he
signal ha emains a e he cancella ion p ocess can be cha ac e ized
as ollows:
𝑦′(𝑡) = 𝑥𝐹 𝑊 𝑆 (𝑡) ∗ ℎ𝐹 𝑊 𝑆 (𝑡) + 𝑛𝐶(𝑡) + 𝑛0(𝑡),(4)
whe e 𝑛𝐶is he non-ideal cancella ion esidue. I is impo an o no e
ha 𝑛𝐶 ep esen s an in e e ing e m and does no ha e he same
e ec as 𝑛0. This means ha 𝑛𝐶does no s ic ly ollow a Gaussian
dis ibu ion. Speci ically, 𝑛𝐶con ains a po ion o he loopback signal
ha has no been ully canceled, al hough i is a enua ed compa ed o
he p e ious s ep (see Eq. (1)). Howe e , depending on he e ec i eness
o he cancella ion echnique, i may all below he noise loo .
In summa y, o achie e op imal loopback cancella ion pe o mance,
minimizing he cancella ion e m (𝑛𝐶) is necessa y, which is closely
ied o he e ec i eness o he associa ed signal p ocessing. The o e all
pe o mance a ies based on he speci ic algo i hm [41], bu a consis-
en ule applies he mo e accu a e he channel es ima ion, he be e
he pe o mance. The e o e, ob aining a ealis ic cha ac e iza ion o he
loopback channels emains c ucial.
2.2. TDL channels
Time- a ying and mul ipa h p opaga ion channels a e gene ally de-
sc ibed using TDL models. A TDL model consis s o disc e e pa hs wi h
speci ic ampli ude and delay. In addi ion, each pa h has an associa ed
Dopple spec um modeling he ime- a iabili y o he pa h ampli ude.
The gene ic TDL model can be desc ibed as [57]:
ℎ(𝑡, 𝜏) =
𝑁−1
∑
𝑘=0
𝑎𝑘(𝑡)𝛿(𝜏−𝜏𝑘),(5)
whe e 𝑁is he numbe o pa hs, 𝑎𝑘 ep esen s he ime- a ying signal
𝑘 h e lec ed pa h, and 𝜏𝑘is he ela i e delay o he e lec ed pa h.
In his case, 𝑘= 0 ep esen s he di ec signal om he ansmi e , so
he e is no ela i e delay (i.e., 𝜏0= 0).
𝐻𝐹 𝑊 𝑆 and 𝐻𝐿𝐵𝑆 can be modeled ollowing he ma hema ical s uc-
u e o TDL models (as in Eq. (5)). Howe e , he di ec and he loopback
channels ha e se e al di e ences due o he na u e o he use case.
In pa icula , due o he in as uc u e o ansmission owe s, 𝐻𝐹 𝑊 𝑆
will ha e a mo e subs an ial mul ipa h e ec han 𝐻𝐿𝐵𝑆 . This e ec
is ypically app ecia ed in he oo mean squa e (RMS) delay sp ead
(𝜏𝑅𝑀𝑆 ) since i measu es he channel’s a e age delay sp ead. The e o e,
he di e ence be ween he 𝜏𝑅𝑀𝑆 measu ed o bo h channels, 𝐻𝐹 𝑊 𝑆
and 𝐻𝐿𝐵𝑆 , is as ollows:
𝜏𝑅𝑀𝑆,𝐹 𝑊 𝑆 ≫ 𝜏𝑅𝑀𝑆,𝐿𝐵𝑆 .(6)
I we analyze 𝜏𝑅𝑀𝑆 in mo e de ail, we can conclude ha he
delay and he ampli ude o he ecei ed pa hs de ine he inal 𝜏𝑅𝑀𝑆
alue. Consequen ly, he di e ence be ween 𝐻𝐹 𝑊 𝑆 and 𝐻𝐿𝐵𝑆 can be
desc ibed as:
𝑁𝐹 𝑊 𝑆 ≫ 𝑁𝐿𝐵𝑆 ,(7)
whe e 𝑁𝐹 𝑊 𝑆 and 𝑁𝐿𝐵𝑆 a e he numbe o pa hs in he o wa d and
loopback channels, espec i ely. Fu he mo e, he maximum delay o
he e lec ed pa hs will be highe in he o wa d channel:
max𝐹 𝑊 𝑆 (𝜏𝑘)≫max𝐿𝐵𝑆 (𝜏𝑘).(8)
Las ly, one o he pa icula cha ac e is ics o he loopback channel
in TDLs is he i s pa h ecei ed (k=0). Since al hough in bo h 𝐻𝐹 𝑊 𝑆
and 𝐻𝐿𝐵𝑆 , he i s pa h is he beam wi h he highes ene gy, hey do
no ha e he same cha ac e is ics. Gene ally, in he di ec channel, he
i s pa h is he line-o -sigh componen , while in 𝐻𝐿𝐵𝑆 , i is no line-
o -sigh since he ansmission and ecep ion adia ing sys ems a e in
he same owe bu diso ien ed.
3. Measu emen sys em
This sec ion p esen s he me hodology p oposed o empi ical da a
eco ding. In pa icula , he ield ials’ basic design and ha dwa e
equipmen a e p esen ed. This measu emen campaign was ca ied ou
on a ansmi e si e o a ne wo k ope a o (I elazpi) [58]. This com-
pany p o ides adio, ele ision, and public communica ions se ices.
Fig. 2 shows a concep ual diag am o he measu emen me hodology
ollowed du ing his wo k, while Fig. 3 p esen s he ac ual implemen-
a ion a he ansmission cen e . The i s elemen in Fig. 2 is signal
gene a ion. The e e ence sou ce is a DVB-T2 signal. Some con igu able
pa ame e s o he DVB-T2 emission a e c i ical o ge ing he bes
channel cha ac e iza ion. QPSK 1/2 was used as he modula ion and
coding a e o gene a e a obus ansmission signal. Then, we selec ed
a high FFT size (i.e., 32 K) o ha e he highes possible numbe o
subca ie s and, he e o e, o inc ease he channel equency de ini ion.
Table 1 p esen s he es o he wa e o m gene a ion pa ame e s. The
measu emen sys em c ea es a p e- eco ded ile wi h In-phase and
Quad a u e (IQ) samples o be ed in o he ansmi ing an enna using
a Dek ec DTU-215 (ou pu bandwid h 8 MHz, ou pu powe −49 o
−18 dBm). A e passing h ough he ansmi e si e loopback channel,
he ecei ed signal IQ componen s a e eco ded using a ec o signal
analyze . The modula o and he gene a ion so wa e a e iden i ied in
Fig. 3 using a ed dashed box.
An addi ional ampli ica ion s age and channel il e ing a e equi ed
be o e eeding he es signal in o he si e ansmission sys em. The
ampli ie is a P omax DT-730 [59] wi h a gain equal o 52 dB, whe eas
he channel il e ing is uned o channel 27 on he UHF band. The
ampli ie is highligh ed in yellow Fig. 3.
Además, hab ías que des aca que
As shown in Fig. 4, he measu emen s we e ca ied ou in a eal and
ongoing ansmission cen e wi h wo adia ing sys ems. The highes
ansmission sys em (i.e., TX 1, in Fig. 4) is ins alled a he heigh o
25 m, whe eas he second ansmission sys em (i.e., TX 2, in Fig. 4) is
23 m high. Each ansmission sys em can be used in a single equency.
In pa icula , he cen e equency o TX 1 is 522 MHz, while TX 2 uses
514 MHz. The adia ing sys em ins alled on he owe comp ises ou
sec o s (i.e., panels loca ed a 50◦, 140◦, 230◦, and 320◦), and each sec-
o uses wo panels pe bay, imp o ing di ec i i y on he e ical plane.
Measu emen 234 (2024) 114792
4
I. Bilbao e al.
Fig. 2. Block diag am o he comple e ansmission- ecep ion chain implemen ed o he channel measu emen campaign.
Fig. 3. T ansmission- ecep ion chain implemen a ion using ha dwa e equipmen . The
h ee p incipal s ages o channel measu ing a e highligh ed: signal gene a ion,
ampli ica ion, and eco ding.
Panels a e Rymsa 6xAT15-250, which a e composed o ou dipoles.
Each panel has a 3 dB beamwid h o 61◦and 27◦in he ho izon al and
e ical planes, espec i ely, and a gain o 11.1 dB e e enced o he 𝜆/2
dipole. These panels ha e a dimension (i.e., wid h ×heigh ×dep h) o
483 ×264 ×983 mm and a weigh o 13 kg. I should be highligh ed
ha as a ully ope a ional b oadcas si e, each adia ing sys em deli e s
a se o speci ic RF channels. In ou case, we used TX 1 as he main
ansmi e ; he e o e, he wo king equency was 522 MHz. One o he
main challenges o measu ing in eal b oadcas ing acili ies is ha ing
an adap able ecei e . In his case, we used a single panel o he signal
ecep ion (see Rx in Fig. 4). The dis ance be ween he ansmission
Table 1
Summa y o he con igu a ion pa ame e s used o signal
gene a ion.
Pa ame e Value
FFT size 32 k
Gua d in e al 1/16
Bandwid h 8 MHz
Pilo pa e n PP2
Modula ion QPSK
Code a e 1/2
L1 Modula ion BPSK
Ro a ed cons. No
#Supe ames 1
#F ames pe supe ame 2
#Da a symbols 63
Supe ame leng h 488 ms
sys em used (i.e., TX 1) and he ecei e an enna was 10 m, meaning
ha he an enna was ins alled a 15 m. All he adia ing sys ems ha e
ho izon al pola iza ion.
The signal passed h ough a bandpass il e on he ecep ion side
o emo e po en ial ex e nal co-channel in e e ences. A ec o signal
analyze (An i su MS2690 A) sampled he ecei ed signal. This s age
is iden i ied by a dashed blue line in Fig. 3. The measu ed IQ da a
a e eco ded on a 32-bi loa bina y o ma . An au oma iza ion p o-
cess is applied o ake pe iodical eco dings o he ecei ed signal.
Fi e seconds o he signal a e eco ded each minu e, and o e all, he
measu emen p ocess las s one hou .
The measu emen s we e aken wice o eco d da a in wo scena ios
and UHF channels. In he i s case, he ecei ing an enna was o i-
en ed o an open a ea (ca pa king a ea). This an enna con igu a ion
is displayed in Fig. 4. The second measu emen was eco ded a e
o a ing he ecei e an enna 180◦on he ho izon al plane ( he an enna
poin s o a moun ainous a ea). This way, wo di e en p opaga ion
channels o he loopback pa h should be discussed. F om now on,
he p opaga ion pa h whe e he an enna is acing he ca pa king will
be called Loopback Channel Model 1 (LCM1), and he channel acing
he moun ain a ea will be e e ed o as Loopback Channel Model 2
(LCM2).
The eco ded da a unde wen o line pos -p ocessing using special-
ized so wa e o demodula ing and decoding he DVB-T2 signal [60].
This p ocess yielded de ailed insigh s in o all decoding s ages, including
channel es ima ion based on sca e ed pilo s in bo h ime and equency
domains. The so wa e pla o m o e s a ious ime and equency
in e pola ion me hods, wi h linea in e pola ion being he p e e ed
choice ecommended by he DVB-T2 implemen a ion guidelines [61].
The selec ion o he pilo pa e n signi ican ly in luences he g anula i y
Measu emen 234 (2024) 114792
5
I. Bilbao e al.
Fig. 4. The ongoing b oadcas ansmission owe du ing he measu emen campaign.
The ansmission adia ing sys em (TX 1 and TX 2) and he po able ecei e (RX) a e
iden i ied.
Fig. 5. Measu ed signal isola ion using TX 1 and TX 2 and di e en an enna sepa a ion
be ween ansmi e and ecei e .
o he es ima ed channel impulse esponse and associa ed Dopple
mode. In ou case, we con igu ed he ansmission signal wi h he
PP2 pilo pa e n (see Table 1), which in ol es a low sepa a ion o
pilo -bea ing ca ie s (𝐷𝑥= 6) and a low sequence leng h in symbols
(𝐷𝑦= 2). This con igu a ion p o ides he highes channel in o ma-
ion (app oxima ely 8.33% sca e ed pilo o e head) while minimizing
he numbe o equi ed symbols among all possible pilo pa e ns.
Fo u he de ails abou he DVB-T2 ecei e pla o m, e e o he
sou ce [60].
Finally, we ca ied ou some signal isola ion measu emen s o es
he accu acy o he designed me hodology in he ansmission owe .
The ansmission powe was se o 30 dBm, and he signal isola ion
was calcula ed as he di e ence be ween he ansmi ed signal powe
and he ecei ed powe in he channel bandwid h. Fig. 5 shows he
Fig. 6. Rep esen a ion o he a e age loopback channel model LCM1 and LCM2.
Algo i hm 1 Channel modeling algo i hm.
1: De ine: 𝑛𝑝. The numbe o pa hs o he esul ing channel model;
2: De ine: 𝑁. The numbe o samples in he a e age channel esponse;
3: De ine: ℎ𝑓=ℎ𝑓,1, ..., ℎ𝑓 ,𝑁 . Se o 𝑁samples in he equency
domain esponse;
4: Ini ializa ion: Calcula e ℎ𝑓 om he DVB-T2 ecei e pla o m
channel es ima ions;
5: ℎ𝑡=𝑖𝑓𝑓 𝑡(ℎ𝑓);
6: while 𝑁 > 𝑛𝑝do
7: 𝑚𝑖𝑛𝑝𝑜𝑠 =Posi ion[min(|ℎ𝑡|)];
8: ℎ𝑡(𝑚𝑖𝑛𝑝𝑜𝑠 − 1) = ℎ𝑡(𝑚𝑖𝑛𝑝𝑜𝑠 − 1) + ℎ𝑡(𝑚𝑖𝑛𝑝𝑜𝑠);
9: emo e(ℎ𝑡(𝑚𝑖𝑛𝑝𝑜𝑠));
10: 𝑁=𝑁− 1;
11: end while
esul s using TX 1 and TX 2 and di e en TX-RX an enna sepa a-
ions. Minimum signal isola ion is close o 70 dB, while he maximum
eaches 85 dB. Mo eo e , inc easing he an enna sepa a ion inc eases
he isola ion, indica ing co ec beha io . Addi ional in o ma ion on
he me hodology o calcula ing signal isola ion, signal powe s eng h
le els, and a iabili y can be ound in [40].
4. Channel model cha ac e iza ion and analysis
This sec ion shows he main esul s ob ained om he loopback
channel cha ac e iza ion. Fi s , he ob ained channel model is de-
sc ibed, and hen some ele an channel cha ac e is ics a e de i ed:
Dopple spec um, delay sp ead, and K- ac o .
4.1. Pa h model
An a e aging p ocess has been ollowed o dec ease he impac o
addi ional noise added o he channel inpu esponse cap u ed by he
DVB-T2 ecei e pla o m. In addi ion, as will be shown in Sec ion 4.2,
he a e aging p ocess is easible because o he low channel a iabili y
o he measu emen s ( he Dopple e ec is negligible). Fig. 6 shows he
esul s o bo h LCM1 (blue line) and LCM2 ( ed line). As expec ed, he
channel impulse esponses concen a e mos ene gy in he i s pa h.
La e , bo h channels show e lec ed pa hs be ween 0.3 μs and 1 μs.
Then, he channel esponse alls, and a e 1.5 μs, a la esponse can
be assumed. Mo eo e , he ampli ude di e ence a e 1.5 μs wi h he
maximum (𝑡= 0) is be ween 20–25 dB.
Fig. 6 displays a channel model associa ed wi h each case unde
analysis (i.e., LCM1 and LCM2). The s udy by Candel e al. [62] has
Measu emen 234 (2024) 114792
6
I. Bilbao e al.
Fig. 7. Combined ep esen a ion o he a e age channel models LCM1 and LCM2 in linea uni s and he ob ained channel app oxima ions using 6 and 12 aps o each channel
model.
Table 2
Desc ip ion o he ob ained channel models in e ms o ap delay and ampli ude.
Taps Tap delay (μs) Tap ampli ude (dB)
LCM1 6 0, 0.11, 0.22, 0.44,
0.55, 0.66
0, −6.6, −6.7, −8.0,
−8.1, −4.2
LCM1 12 0, 0.11, 0.22, 0.33, 0.44,
0.55, 0.66, 0.77, 0.98, 5.47,
6.13, 6.78
0, −6.6, −8.7, −10.9, −8.0,
−8.1, −6.0, −11.1, −10.0, −10.6,
−14.1, −10.3
LCM2 6 0, 0.11, 0.22, 0.55,
0.77, 0.88
0, −5.3, −5.7, −5.3,
−9.0, −4.2
LCM2 12 0, 0.11, 0.22, 0.33, 0.44,
0.55, 0.77, 0.88, 2.19, 4.16,
8.10, 12.69
0, −5.3, −6.8, −11.9, −9.4,
−7.5, −9.0, −8.1, −12.7, −11.3,
−12.2, −12.5
been used as a e e ence o ansla e he loopback channel unc ion in o
a channel model wi h a de e mined numbe o pa hs, ampli ude, and
phase alues. A pseudo-code summa y o he algo i hm is p esen ed in
Algo i hm 1.
The algo i hm s a s de ining some ele an pa ame e s, such as
he numbe o pa hs used o model he loopback channel (i.e., 𝑛𝑝),
he numbe o samples ha compose he a e age channel esponse
(i.e., 𝑁), and he complex alues o he equency domain a e age
channel (i.e., ℎ𝑓). Du ing he ini ializa ion s age, we compu e he
a e age channel in he equency domain by aking he mean o all
channel es ima ion alues ob ained om he DVB-T2 ecei e pla o m.
Each channel sample in he equency domain is a e aged o pe o m
his ope a ion. The cos o his ope a ion is de e mined by he numbe
o samples in he equency domain (𝑁) and he numbe o channels
ob ained du ing he es ima ion phase (𝑀). As a esul , he a e aging
cos is 𝑂(𝑁⋅𝑀). A e wa d, he i s ope a ion o he channel model
is o ob ain he ime domain esponse (i.e., ℎ𝑡) by applying he in e se
as Fou ie ans o m (IFFT) o he equency domain esponse. Nex , a
loop is execu ed while he numbe o channel samples (i.e., 𝑁) is highe
han he numbe o pa hs o he loopback channel model (i.e., 𝑛𝑝). In
each loop lap, he algo i hm inds he posi ion o he ℎ𝑡sample wi h he
lowes ampli ude (see Line 7) and adds he minimum sample alue o
he adjacen sample (see Line 8). Then, i emo es he sample wi h he
lowes ampli ude om he a e age channel esponse (see Line 9) and
dec eases he numbe o samples in he a e age channel esponse due
o he emo al (see Line 10). Finally, when he while loop inishes, ℎ𝑡
con ains a se o 𝑛𝑝samples wi h he mos ep esen a i e ime domain
esponse in o ma ion.
The algo i hm’s cos p ima ily depends on 𝑁, as his pa ame e
de e mines he numbe o i e a ions in Algo i hm 1. Consequen ly, he
algo i hm’s complexi y can be exp essed as 𝑂(𝑁). Howe e , bo h com-
ponen s mus be combined when inco po a ing he loopback channel
a e aging as pa o he o al cos . Thus, he o e all algo i hmic cos
can be exp essed as 𝑂(𝑁⋅𝑀+𝑁).
As desc ibed in Algo i hm 1, he numbe o aps used o model
he channel is a p ede ined alue. The loopback channel has a low
mul ipa h e ec due o i s cha ac e is ics, so 𝑛𝑝can be a low alue. To
be su e o choose a ep esen a i e alue, we ha e used o he exis ing
and widely used channels in he li e a u e as a e e ence. In pa icula ,
se e al models such as Typical U ban-6 (TU-6), Po able Indoo (PI),
and Po able Ou doo (PO) channels [63] ha e been cha ac e ized
wi h 6 o 12 aps and hei pe o mance accu acy has been widely
demons a ed. Consequen ly, Fig. 7 shows he channel app oxima ions
ob ained wi h Algo i hm 1: LCM1 in Fig. 7(a) and LCM2 in Fig. 7(b).
Mo eo e , Fig. 7 also shows wo app oaches o each channel using 6
and 12 aps.
In addi ion, based on he ob ained channel impulse esponses in
Fig. 7, mos o he channel ene gy is in he i s μs, which implies ha
ewe app oxima ion aps a e equi ed han in high mul ipa h e ec
channel models. The main eason o using wo di e en app oaches is
he complexi y in ol ed in each o hem. Using he model based on six
aps in any o he communica ion simula o is less expensi e han he
12 aps model, especially om he compu a ional cos poin o iew.
Howe e , he mo e aps used o model, he mo e closely i esembles
he ac ual channel. Fo his eason, we ha e p oposed wo di e en
models o he loopback channel ha can be adap ed o he needs and
condi ions o each pa icula case.
The speci ic alues o he channel aps a e in Table 2. The able
desc ibes each ap’s ime loca ion ( ap delay in μs) and ampli ude
( ap ampli ude in dB). LCM1 and LCM2 p esen mul ipa h componen s
wi h signi ican delay. In pa icula , he las aps o LCM1 a e loca ed
be ween 5-7 μs and a ound 12 μs o LCM2. Howe e , hese aps a e
less ele an han he ene gy ecei ed wi h a sho delay and become
no iceable only when he numbe o modeling aps ises om 6 o 12.
Fu he mo e, sho delay echoes a e ela ed o di ec loopback leakage.
In con as , long delay echoes a e c ea ed by mul ipa h e lec ions o
he loopback signal due o he en i onmen . In addi ion, i should be
Measu emen 234 (2024) 114792
7
I. Bilbao e al.
Fig. 8. Compa ison be ween he Dopple powe spec al densi y ob ained om he empi ical da a and he exponen ial i ing cu e co esponding o each channel model ollowing
Eq. (9).
no ed ha , in gene al, bo h channels p esen sho -dis ance echoes.
In pa icula , esul s a e aligned i we compa e he esul s wi h he
channel models used in o de DTT s anda diza ion p ocesses. O he
channel models, such as DVBT-P [61] o he TU-6, p esen mos o he
ecei ed powe concen a ed wi hin he i s ew μs. Channel model
p oposals like he ones de eloped o DVB-H [63] show he same
beha io , whe e, in addi ion o he TU-6 channel model, wo addi ional
channels we e p oposed (i.e., Po able Indoo and Ou doo channels, PI
and PO), cha ac e ized by 12 aps, and concen a e mos o he channel
ene gy in he i s aps.
4.2. Dopple spec um
A ull cha ac e iza ion o he channel model equi es a desc ip ion
o he ime e olu ion o he ecei ed signal componen s. Se e al wo ks
ha e modeled he Dopple powe spec al densi y (PSD) in di e en
applica ions such as mobile- o-mobile [64], ehicle- o- ehicle [65],
and o he sys ems. Se e al classical models a e widely accep ed o
model PSDs in ansmission/ ecep ion chain simula ions (e.g., Jakes,
Laplacian, Gaussian, o la Dopple spec ums) [66].
Classical models do no always i he eal- ime e olu ion o he
p opaga ion channel. In hose cases, au ho s usually model hei sys em
wi h empi ical da a. Fo example, au ho s in [53,54] p esen ed hei
PSD model wi h empi ical measu emen s. Based on he cha ac e is ics
ound in he ob ained da a, he au ho s de ined hei PSD wi h a Di ac
del a o 𝑓= 0 Hz and exponen ial i ings o bo h he nega i e and
posi i e equencies. In pa icula , he PSD model was ma hema ically
de ined as ollows:
𝑆(𝑓) = ⎧
⎪
⎨
⎪
⎩
𝑎exp(𝑏𝑓) − 𝑐, 𝑓𝑚𝑖𝑛 ≤𝑓 < 0
𝛿(𝑓), 𝑓 = 0
𝑑exp(−𝑒𝑓) − 𝑔, 0< 𝑓 ≤𝑓𝑚𝑎𝑥
(9)
whe e 𝑎,𝑏,𝑐,𝑑,𝑒and 𝑔a e posi i e cons an s, 𝑓is he Dopple
equency (Hz), and 𝑓𝑚𝑎𝑥 and 𝑓𝑚𝑖𝑛 a e he maximum and minimum
equency alues. 𝑏and 𝑒 ep esen he exponen ial decay ( he lowe
alues, he b oade spec al cha ac e is ics). 𝑐and 𝑔a e ela ed o
he asymp o ic alues o in ini e nega i e and posi i e equencies,
espec i ely. Then, 𝑎and 𝑑accoun o he ela i e alue o he cu e
o he 𝑦-axis conce ning he asymp o ic alues o in ini e equencies
gi en by 𝑐and 𝑔. This way, he PSDs a e ob ained in dB/Hz a e
no maliza ion conce ning he main componen , which is ob ained o
𝑓= 0 Hz.
The PSD models desc ibed by Eq. (9) accu a ely ep esen he
Dopple spec um o pseudo-s a ic en i onmen s. This is because he
s a ic componen (i.e., 𝑓= 0 Hz) con ains mos o he powe , and
as we mo e away om ha poin , he powe d ops apidly. Fo his
wo k, Eq. (9) was used as a e e ence PSD model because b oadcas
en i onmen s exhibi hese cha ac e is ics. The Dopple spec um ex-
hibi s minimal a iabili y in en i onmen s wi h ew mobile elemen s
and highly di ec ional communica ions. As a esul , he 𝑓= 0 Hz
componen is he mos signi ican o he spec um.
Since he PP2 pilo pa e n has been selec ed, we ha e a channel
es ima ion o e e y wo symbols (i.e., 𝐷𝑦= 2). So, he channel
es ima ion pe iod can be calcula ed as ollows:
𝑇𝑒𝑠𝑡 =𝐷𝑦⋅(𝑇𝑈+𝑇𝐺)=2⋅(3.584 + 0.224) = 7.616 ms, (10)
whe e 𝑇𝑈and 𝑇𝐺a e he symbol ime and he gua d in e al, espec-
i ely. The exac alues a e de ined in [61]. The e o e, he Dopple
equency ange can be calcula ed as ollows:
𝑓𝐷= ± 1
2
⋅1
𝑇𝑒𝑠𝑡
= ±65.65 Hz,(11)
whe e he ends o he ange a e he maximum and minimum Dopple
equencies (i.e., 𝑓𝑚𝑎𝑥 = +65.65 Hz and 𝑓𝑚𝑖𝑛 = -65.65 Hz).
Fig. 8 shows he PSDs and he exponen ial i ing co esponding o
he measu ed da a o LCM1 (see Fig. 8(a)) and LCM2 (see Fig. 8(b)). As
can be obse ed in he igu e, he measu ed da a bes i in he cen e
o he 𝑥-axis (i.e., o 𝑓= 0 Hz) in bo h cases. The main eason o his
e ec is ha he measu ed da a a iabili y is highe when he Dopple
equency inc eases han in he p incipal componen . In addi ion, he
speci ic i ing pa ame e s o hese wo examples a e included in he
PSD de ini ions o he ollowing Eqs. (12) and (13):
𝑆𝐿𝐶𝑀1(𝑓) = ⎧
⎪
⎨
⎪
⎩
13.9exp(0.13𝑓) − 42.1, 𝑓𝑚𝑖𝑛 ≤𝑓 < 0
𝛿(𝑓), 𝑓 = 0
13.3exp(−0.12𝑓) − 41.9,0< 𝑓 ≤𝑓𝑚𝑎𝑥
(12)
𝑆𝐿𝐶𝑀2(𝑓) = ⎧
⎪
⎨
⎪
⎩
13.5exp(0.08𝑓) − 38.6, 𝑓𝑚𝑖𝑛 ≤𝑓 < 0
𝛿(𝑓), 𝑓 = 0
13.7exp(−0.09𝑓) − 38.2,0< 𝑓 ≤𝑓𝑚𝑎𝑥
(13)
LCM2 p esen s a lowe exponen ial decay (see 𝑏and 𝑒) han LCM1.
Focusing on he mean alue o 𝑐and 𝑔pa ame e s, which gi es an idea
o he powe densi y o he highes and lowes equencies, i is sligh ly
highe in LCM2.
In summa y, he esul s p esen ed in his sec ion indica e ha mos
o he Dopple spec um densi y is concen a ed in 𝑓= 0 Hz. In
ac , in bo h Dopple PSDs (i.e., LCM1 and LCM2), inc easing he
Dopple equency o 𝑓= 1 Hz implies a powe all o 20 dB, and
hey ha e an asymp o ic beha io a a ound 40 dB. The e o e, we can
assume ha he obse ed loopback channel models a e s a ic. This
conclusion may be gene alized o o he b oadcas cen e s due o he
simila i ies in he applica ion use cases (i.e., ixed ansmission and
Measu emen 234 (2024) 114792
8
I. Bilbao e al.
Table 3
Empi ical alues o Delay Sp ead and K- ac o .
𝜏𝑚𝑒𝑎𝑛 (ns) 𝜏𝑅𝑀𝑆 (ns) 𝐾- ac o (dB)
LCM1 41.59 151.56 9.93
LCM2 16.73 49.67 13.31
ecep ion an ennas, low e lec i e a ea, o high isola ion). Ne e heless,
he Dopple analysis may di e om Fig. 8 in non-b oadcas use cases
such as b oadband communica ions (i.e., 5G/6G) since hei applica ion
a ge is u ban a eas wi h many mobile use s/objec s. In addi ion,
bad wea he condi ions such as wind and s o ms may also a ec he
Dopple spec um due o he owe swinging. Ne e heless, conduc ing
channel measu emen s in hose wea he condi ions implies ex emely
high isk o he ha dwa e equipmen and echnical wo ke s who climb
h ough he ansmission owe . Consequen ly, channel modeling o
bad wea he condi ions was no easible in his wo k. Howe e , o
u u e wo k, hese complex scena ios can be modeled as demons a ed
in he s udy by Hong e al. [27]. Thei wo k models he impac o
owe swing due o wind on he Dopple spec um. Consequen ly, he
spec um a ia ions p oposed by he au ho s could be in eg a ed in o
he spec a de ined in Eqs. (12) and (13).
4.3. Addi ional channel pa ame e s: delay sp ead and K- ac o
Mos channel models p o ide Delay Sp ead and K- ac o calcula-
ions. Those igu es will desc ibe he dispe sion in ime o he ecei ed
ene gy a he ecei e . In gene al, he Delay sp ead is used o analyze
he ele ance o he mul ipa h p opaga ion o he o al ecei ed ime
ene gy.
The mean Delay Sp ead can be in e p e ed as an a e age me ic o
cha ac e ize he di e ence be ween he ime o a i al o he ea lies
signi ican mul ipa h componen and he ime o a i al o he las mul-
ipa h componen s [67]. In addi ion o 𝜏𝑚𝑒𝑎𝑛, he RMS delay sp ead is
also used (𝜏𝑅𝑀𝑆 ), which is also known as he oo o he second cen al
momen o he no malized delay powe densi y spec um. The e e ence
equa ions o ob aining hese pa ame e s can be ound in [68].
The calcula ion o he Delay Sp ead can be complemen ed wi h he
ading K- ac o . This pa ame e is de ined as he a io o signal powe in
he dominan componen o e he sca e ed and e lec ed signal powe .
Consequen ly, i de e mines he dis ibu ion o he ecei ed signal
ampli ude [69]. The K- ac o calcula ion me hod conside ed he e o
e e ence compa ison pu poses is he one ecommended by ETSI [61].
Table 3 summa izes Delay Sp ead’s and K- ac o ’s empi ical esul s.
In gene al, LCM1 and LCM2 loopback channels p esen sho delay
sp eads (mean delay sp ead below 50 ns in bo h cases). The maximum
RMS delay sp ead is close o 150 ns. These esul s a e compa ible wi h
he sho dis ances be ween he ansmi ing and ecei ing an enna.
In addi ion, i should also be highligh ed ha LCM2 p esen s an e en
sho e mean and RMS delay sp ead alues i compa ed wi h LCM1.
This esul is in line wi h he measu emen s shown in Fig. 7 (solid
blue line) since he channel impulse esponse o LCM1 shows a highe
ampli ude han LCM2 be ween 4 μs and 8 μs.
In bo h channels (LCM1 and LCM2), mos o he ecei ed powe
is concen a ed in he i s pa h. Fu he mo e, i should also be no ed
ha LCM2 has less powe dispe sion han LCM1 since he calcula ed K-
ac o is always mo e signi ican han ha calcula ed o LCM1. These
esul s align wi h he delay sp ead calcula ions, whe e LCM1 has shown
a highe delay sp ead han LCM2 due o he channel impulse esponse
ampli ude be ween 4 μs and 8 μs.
5. Channel model pe o mance e alua ion
The main goal o his sec ion is o measu e he channel models’ pe -
o mance in ealis ic anscei e chains. Hence, we ha e implemen ed
a comple e ATSC 3.0 ansmission- ecep ion chain and es ed all he
Table 4
Pa ame e summa y o he ATSC 3.0 anscei e chain.
Pa ame e Value
Bandwid h 6 MHz
FFT Size 16 k
Gua d In e al 1/16
Ou e FEC BCH
Inne FEC LDPC
Modula ion QPSK, 16-NUC, 64-NUC
Code a e 3/15, 7/15, 12/15
Equaliza ion Ideal
MIMO No
LDM No
SNR s ep 0.05 dB
#Seeds 1000
p oposed channel models o analyze how hey a ec he ansmi ed
signal.
Table 4 summa izes he mos ele an pa ame e s desc ibing he
simula ed ATSC 3.0 anscei e . I should be highligh ed ha ATSC 3.0
uses wo consecu i e Fo wa d E o Co ec ion (FEC) echniques: Bose–
Chaudhu i–Hocquenghem (BCH) as ou e coding and Low-Densi y Pa i y-
Check (LDPC) codes as inne coding. In addi ion, we ha e no con-
side ed op ional con igu a ions such as Mul iple-Inpu Mul iple-Ou pu
(MIMO) o Laye ed Di ision Mul iplexing (LDM) o analyze he pu es
ATSC 3.0 s uc u e and no o in oduce addi ional e ec s o he
ansmission chain. Conce ning he modula ion and he code a e o he
ansmi ed signal, h ee di e en con igu a ions ha e been selec ed o
es channel models combined wi h di e en signal obus ness le els:
QPSK 3/15 (high obus ness and low da a a e), 16-non uni o m
cons ella ion (NUC) 10/15 (mid obus ness and mid da a a e), 64-NUC
12/15 (low obus ness and high da a a e).
Fig. 9 shows he esul s ob ained wi h he ou p oposed channel
models (see Table 2) in e ms o bi e o a e (BER) s. signal- o-noise
a io (SNR). Each modula ion and code a e con igu a ion also includes
he simula ion o he AWGN channel model o use as a e e ence. The
i s e iden akeaway om he esul s is ha he p oposed loopback
channels a ec he ansmi ed signal mo e s ongly han he AWGN
model. This means ha o a ixed SNR alue, all he LCM channels
p esen a lowe BER alue. In pa icula , he di e ence be ween AWGN
and he LCM models inc eases when he equi ed SNR inc eases. While
a a ia ion below one dB is obse ed in Fig. 9(a), he di e ence
inc eases up o h ee dB in Fig. 9(c). Ano he ele an conclusion is ha
using 6 o 12 aps o model he channel sligh ly a ec s he pe o mance.
The channel models based on six aps pe o m be e han hose based
on 12. This e ec is due o he loca ion o he addi ional aps in he 12-
ap models. In ac , when he numbe o aps used o channel modeling
inc eases, he model shows a highe mul ipa h e ec . This issue was
al eady de ec ed in Table 2. Ne e heless, as he mul ipa h e ec can
be conside ed sho (be ween 5-7 μs in LCM1 and a ound 12 μs in
LCM2), he in luence o he numbe o aps in he channel modeling
is always below one dB. The maximum di e ence can be obse ed
when ansmi ing a 64-NUC 12/15 signal h ough LCM2. Finally, he
las ele an conclusion ob ained om Fig. 9 is he di e en pe o -
mances ega ding LCM1 and LCM2. The di e ence be ween LCM1 and
LCM2 inc eases when he equi ed SNR inc eases. Pa icula ly, while
e y simila esul s a e shown in Fig. 9(a), he a ia ion be ween he
models inc eases up o wo dB in he low obus ness con igu a ion (see
Fig. 9(c)). This conclusion aligns wi h he esul s obse ed du ing he
delay sp ead calcula ion (see Table 3), whe e a highe mul ipa h e ec
was de ec ed in LCM1, and, he e o e, he ansmi ed signal is mo e
a ec ed by he channel.
I is impo an o no e ha he pe o mance cu es shown in Fig. 9
assume ideal channel es ima ion. These cu es illus a e pe o mance
and de ia ion ela i e o he AWGN channel, which is solely based on
he channel s uc u e. Howe e , se e al channel- ela ed pa ame e s can
in luence hese cu es. Fo ins ance, i a di e en Dopple spec um
Measu emen 234 (2024) 114792
9
I. Bilbao e al.
Fig. 9. Pe o mance e alua ion o he loopback channel models using an ATSC 3.0 anscei e . AWGN channel model is used as a e e ence.
model is used, i may esul in slowe con e gence o he cu es. This
is pa icula ly ue when mos Dopple powe is no cen e ed a ound
𝑓= 0 Hz. Ano he signi ican di e ence would a ise om using a non-
ideal channel es ima ion. In such cases, e o s a e in oduced, which
can be modeled as addi ional noise in he sys em. Consequen ly, his
ac causes all cases o shi o he igh in he g aph, implying ha
a highe a ailable SNR would be equi ed o a gi en BER alue.
Ne e heless, a comple e ansmission- ecep ion chain is equi ed o
a comple e non-ideal sel -in e e ence cancella ion. This implies a sel -
in e e ence channel es ima ion and cancella ion (non-ideal), ollowed
by a channel es ima ion s age o eco e he ansmi ed signal.
Finally, we ha e de ailed some ecommenda ions o he co ec use
o he measu ed channels based on he da a shown in his sec ion and
on he ield ial condi ions:
•The ou channels p esen ed in his wo k desc ibe loopback chan-
nels exclusi ely.
•Al hough bo h loopback channels p esen a simila delay-ampli
ude pa h ela ion, he measu emen con igu a ion di e ed. The e-
o e, we ecommend using LCM2 o eplica e moun ainous a eas
and LCM1 o he es o he cases.
•Fo mo e ealism, i is ecommended o use he 12- ap-based
channels. Consequen ly, he 6- ap channels a e ese ed o cases
whe e channel complexi y is a c ucial equi emen .
6. Conclusions
This a icle empi ically cha ac e izes he loopback channel o in-
band ull-duplex communica ions. In pa icula , he au ho s p opose
loopback channel models ob ained in an ope a ional b oadcas ans-
mission cen e . Two di e en channel models ha e been p oposed
depending on he o ien a ion o he ecei e an enna. Then, analysis
and discussion a e p esen ed ega ding se e al channel pa ame e s,
namely he Dopple spec um, he delay sp ead, and he K- ac o .
Di e en channel models a e p oposed he e depending on he el-
e an pa hs (i.e., 6- aps and 12- aps). The cha ac e iza ion includes
he de ini ion o each pa h’s ampli ude and delay. Du ing his p ocess,
we ha e concluded ha mos channel ene gy is concen a ed in he
i s mic osecond o he signal. Mo eo e , he las aps o LCM1 a e
loca ed be ween 5-7 μs and a ound 12 μs in he case o LCM2. The delay
sp ead analysis showed shallow alues in bo h cases, al hough LCM1
had highe alues han LCM2. The mean delay sp ead in bo h cases is
below 50 ns, and he maximum RMS delay sp ead is below 150 ns.
A simila conclusion is ob ained om he K- ac o analysis, whe e
LCM1 p esen s a alue 3–4 dB highe han LCM2. Finally, we ha e
also analyzed he Dopple PSD o he loopback channel and concluded
ha he dis ibu ion ha bes i s he eal Dopple spec um is a Di ac
del a o 𝑓= 0 Hz and exponen ial decay unc ions o he emaining
equency. This model means ha mos o he Dopple spec um densi y
is concen a ed in 𝑓= 0 Hz; he e o e, he loopback channel is almos
s a ic.
The analysis, de ini ion, and cha ac e iza ion o he loopback chan-
nel model ill a gap ha he in-band ull-duplex communica ion designs
ha e had du ing he las yea s. In ac , o he bes o he au ho s’ knowl-
edge, his is he i s pape p esen ing he cha ac e iza ion o a loopback
channel model based on expe imen al measu emen s in an au hen ic
and ac i e b oadcas acili y. This s udy enla ges he knowledge o he
ull-duplex esea ch communi y, especially in b oadcas en i onmen s.
In addi ion, he p oposal o hese new channel models p o ides a
new ool o de eloping and alida ing in-band ull-duplex solu ions
since any esea che can use hem du ing he simula ion s age o he
channel emula ion. Doub lessly, using hese channel models inc eases
he ealism o u u e ull-duplex de elopmen s.
Since he s udy was conduc ed a a speci ic ansmission cen e , and
al hough i ep esen s many cu en b oadcas cen e s, u u e wo k is
needed o ensu e he models’ sui abili y o o he scena ios, such as
MIMO en i onmen s. We ha e planned u u e wo k o b oaden he
s udy’s applicabili y. We aim o epea he measu emen s in o he
ansmi e cen e s o compa e and analyze he models’ uni e sali y.
Addi ionally, we will de elop a cha ac e iza ion ha inco po a es he
en i onmen and can be eplica ed using ay acing so wa e.
CRediT au ho ship con ibu ion s a emen
Iñigo Bilbao: Concep ualiza ion, Da a cu a ion, In es iga ion,
Me hodology, W i ing – o iginal d a , W i ing – e iew & edi ing.
Eneko I adie : W i ing – e iew & edi ing, W i ing – o iginal
d a , Me hodology, In es iga ion, Da a cu a ion. Ma a Fe nandez:
Me hodology. Jon Mon alban: In es iga ion, Concep ualiza ion.
Pablo Anguei a: Supe ision, Concep ualiza ion. Zhihong Hun e
Hong: Fo mal analysis. Yiyan Wu: Supe ision, Concep ualiza ion.
Decla a ion o compe ing in e es
The au ho s decla e ha hey ha e no known compe ing inancial
in e es s o pe sonal ela ionships ha could ha e appea ed o
in luence he wo k epo ed in his pape .
Da a a ailabili y
The da a ha has been used is con iden ial.
Acknowledgmen s
This wo k was suppo ed by he Basque Go e nmen , Spain (un-
de he g an IT1436-22 and he g an Elka ek KK-2022/00069) and
by he Spanish Go e nmen unde he g an PID2021-124706OB-I00
unded by MCIN/AEI/ 10.13039/501100011033 and by ERDF A way
o making Eu ope. We wan o hank he company I elazpi o he help,
ime, equipmen , and access o he b oadcas in as uc u e p o ided.
Wi hou hei help, his wo k would no ha e been possible.
