PERFORMANCE ANALYSIS
OF
SINGLE-SLAVE BLUETOOTH PICONETS UNDER CO-
CHANNEL INTERFERENCE
D.
Cascado,
J.
L.
Se illann,
S.
Vicen e,
F.
Diaz
del
Rio,
G.
Jimhez, A. Lina es, A. Ci i -Balcells.
ETS lngenie ia ln o ma ica. Uni e sidad de Se illa. A da. Reina Me cedes
sin.
41012, Se illa (Spain).
Abs ac In his pape , we p esen an analy ical model o
single-sla e Blue oo h picone s when he coexis ence o
mul iple in e e ing de ices p oduces collisions. Closed-
o m exp essions o he channel h oughpu and he mean
packe delay a e ob ained, including he ime spen wai ing
on he node's queue. The e ec o p opaga ion losses and
asynch onous picone s a e also discussed. The esul s o he
analysis a e alida ed h ough simula ion.
Keywo ds
Blue oo h, co-channel in e e ence, pe o mance
analysis, picone , WGil sys ems.
I. INTRODUCTION
Blue oo h is a global s anda d o a low-powe , sho - ange
Wi eless Pe sonal A ea Ne wo k (WPAN). A ypical
Blue oo h sys em is composed o a small numbe o de ices
ha o m a wi eless ne wo k called a
picone .
Since
connec ions a e es ablished ad-hoc, a a iable numbe o
picone s may coexis in he same a ea. A ypical scena io
may consis o a numbe o people wi h po able (and
mo ing) de ices en e ing a common a ea and connec ing o
ixed ne wo ks and mayhe
o
o he po able e minals ( o
ins ance o sha e documen s in
a
mee ing)
[l].
The e o e, a si ua ion whe e se e al people
in
p oximi y
ha e open Blue oo h connec ions is e y common: ai po s,
con e ence halls, o ice, e c. In hese cases, channel
in e e ence
is
no negligible and in his pape we
y
o
model i s e ec on pe o mance. Blue oo h ope a es in he
ISM (Indus ial, Scien i ic
&
Medical) band,
om
2.402GHz
o
2.480GHz. This is cu en ly a e y popula
band o wi eless sho - ange de ices [2]. Howe e , we only
ocus on he in e e ence gene a ed by Blue oo h de ices,
al hough he e a e many possible sou ces o in e ie ence
(mic owa e o ens, 802.1
1
LANs, e c.).
When a Blue oo h uni es ablishes a picone , i becomes a
mus e
so
ha he o he uni s (sla es) synch onize wi h i .
Any uni may hnc ion as a mas e o as a sla e ( his ole is
main ained only o he du a ion o he picone ), hu
al hough
i
may pa icipa e as sla e
in
mul iple picone s i
may only he a mas e in
one
picone . The Blue oo h channel
is di ided in o slo s o leng h 625~s
so,
he ime slo s a e
al e na i ely used by he mas e and he sla es. Al hough
mul i-slo packe ansmissions a e allowed (e.g., DH3 and
DH5 packe s), in his pape we assume ha each packe
occupies a single slo . Fo a gi en picone (wi h a mas e
0-7803-8523-3/04/$20.00
02004
IEEE.
and
up
o
I
sla es) communica ion is Time Di ision
Duplex: ansmi e and ecei e al e na e hei
ansmissions in sepa a e slo s. The mas e o he picone
pe o ms a polling among all he sla es. The scheduling
policy is no speci ied by he cu en Blue oo h speci ica ion
because he bes policy depends on he applica ion. Se e al
al e na i es ha e been s udied in he li e a u e mainly
h ough simula ion [3], wi h e y sca ce analy ical s udies
[4]. The ac ual scheduling should be implemen ed a he
middlewa e le el. Howe e , a common use o cu en
Blue oo h de ices is simply a poin - o-poin link be ween
he mas e and a single sla e, ha is, a "single sla e
picone ". In his pape , we a e especially conce ned wi h his
simple case.
Wha e e polling scheme is used, wi hin a picone collisions
do no occu . The e o e, in e e ence may only occu among
de ices connec ed
o
di e en logical channels (independen
picone s o sca e ne s). In o de
o
educe his in e e ence,
and also o se e o he pu poses like educing mul ipa h
e ec s, Blue oo h uses F equency Hopping. The main
hopping sequence, called
Connec ed
Mode,
used in all usual
ansmissions, dis ibu es he hops e enly o e all ca ie s.
We deno e he numbe o hop came s as M, al hough he
Blue oo h speci ica ion de ines 79 di e en equency hands
a IMHz spacing
[SI.
The
slo
leng h
o
625ps comes om a
hop a e o
1600
hops pe second ( he e a e no equency
hops wi hin a slo ).
When se e al independen (bu in e e ing) Blue oo h
de ices coexis in he same a ea, we can assume ha hey
ansmi using andomly chosen equencies. O cou se,
he e is a possibili y ha se e al de ices choose he same
hop ca ie . In ha case. a collision occu s and he packe
will he ecei ed inco ec ly. The sende is no i ied o his
e o in
he
slo di ec ly ollowing he unsuccess ul
ansmission using
a
as -ARQ scheme
[SI.
The packe is
e ansmi ed a he nex oppo uni y (in al e na e slo s) un il
he packe is success ully ecei ed.
11.
PREVIOUS WORKS
Se e al ecen s udies ha e add essed he p oblem o he
analysis
o
channel in e e ence in Blue oo h sys ems.
El-
Hoiydi
[6]
s udies he wo s case, ha
is,
he conside s 100%
a ic. As a esul , his model p o ides uppe hounds on he
packe e o a e (p obabili y o collisions) as well as lowe
hounds on he h oughpu . On he o he hand, Lim e al. [7]
954
do no limi hei s udy o he wo s case, bu hey p o ide an
app oxima e h oughpu s. o e ed load analysis. In bo h
e e ences he di e en cases o synch onized and
unsynch onized picone s a e conside ed. O cou se, in he
Blue oo h speci ica ion picone s a e no synch onized, bu
equa ions o he synch onous case a e much simple and
hey may p o ide accep able p edic ions
in
some cases. Lim
e al. also conside mul i-slo packe ansmissions.
Packe s delays a e also an impo an pe o mance pa ame e .
El-Hoiydi and Deco ignie
[SI
p esen delay hounds o wo-
way (DATA-ACK) ansac ions unde co-channel
in e e ence. Also in a p e ious wo k
[9],
we ob ained
simple exp essions o he mean delay ha packe s su e
due o possible collisions wi h o he Blue oo h de ices. We
modelled he di e en e ec s o new and e ansmi ed
packe s, and also discussed he di e ences be ween he
synch onous and asynch onous cases.
Howe e none o hese wo ks include he e ec o he
wai ing ime in he de ice bu e . This delay can be
modelled using queuing heo y. Fo ins ance, a ecen wo k
models se e al polling policies in Blue oo h picone s using
MIGII
queues wi h aca ions
[4],
bu igno ing he e ec s
o
in e e ences wi h o he de ices. Howe e , in his pape we
a e in e es ed in modelling hese in e e ences. Since he
scheduling policy
is
no speci ied by he Blue oo h
speci ica ion, we will only conside he case o single-sla e
picone s in o de o exclude he polling delay om ou
analysis. I is clea ha a single-sla e picone can be
modelled as an
WGil
sys em, wi h he Blue oo h channel
conside ed as a single-se e sys em wi h an a bi a y
se ice ime dis ibu ion.
In
all
o hese e e ences he mi iga ion e ec s o
p opaga ion losses o adio wa es a e igno ed, ha is, i is
assumed ha he in e e ence o
jus
one bi is enough
o
des oy he packe s. I is no easy o ake in o accoun he
e ec s o p opaga ion o adio wa es in
a
building,
including cap u e e ec s, posi ion o e minalslobs acles,
en i onmen geome y, e c. Fo ins ance, Kamik and Kuma
[IO]
p esen an analy ical model o he packe
loss
p obabili y, ob ained by limi ing he numbe
o
in e e ing
picone s o h ee because i is e y complica ed o ob ain
e en app oxima e esul s o a highe numbe . P obably, one
o he mos in e es ing app oaches o ou pu poses is gi en
by Mazzenga e al.
[ll].
They p o ide a simple way o
include hese e ec s in he analysis o channel in e e ence,
and we will use hei app oach in ou model.
In he ollowing sec ions, we i s p esen
ou
model
assuming synch onised in e e ing picone s, beginning wi h
he compu a ion o he p obabili y o a success ul
ansmission and hen de i ing impo an pa ame e s like
channel h oughpu . Also, al hough he e ec o p opaga ion
losses a e no explici ly conside ed, we show how hey
could be included in
ou
model ollowing
[ll].
Then, in
o de o model he Blue oo h de ice as an
WG/I
queue we
us de i e he mean and a iance o he “se ice ime”,
which
in
a single-sla e picone is jus he delay due
o
collisions, and de i e an exp ession o he wai ing ime in
he queue. In he subsequen sec ion we p esen some
simula ion esul s in o de o alida e he model, and inally
we p esen ow conclusions.
111.
ANALYSIS
Le
N
be he numbe o in e e ing picone s. We i s assume
ha all he picone s a e synch onized among hem (la e in
his sec ion we discuss his issue u he ). Le
he he
no malized load o e e e y picone . We assume a
homogeneous a ic,
so
is he same o all he picone s.
Collisions may occw i de ices in di e en independen
picone s use he same came . In ha case, packe s a e
e ansmi ed in al e na e slo s. Fo a gi en picone , which is
ansmi ing in a slo
in
a
pa icula came , he p obabili y o
a collision wi h ano he picone is
/M,
ha is, he
p obabili y ha a ansmission a emp occu s in he o he
picone
and
ha he same pa icula ca ie is also chosen. A
collision does no occu wi h p obabili y
1- /M.
The e o e,
he p obabili y ha
n
ou o he in e e ing picone s p oduce
a collision in a slo is
whe e
,l%
a e coe icien s ha include p opaga ion losses and
cap u e e ec s
[I I].
Thei compu a ion is no easy because
hey depend on he geome y and p opaga ion cha ac e is ics
o he en i onmen , he posi ion o obs acles and Blue oo h
de ices o e e y in e e ing picone , e c. A possible
solu ion o his issue is p oposed in
[
141.
Anyway, a success ul ansmission occu s only i he e a e
no collisions in
wo
consecu i e (and independen ) slo s: one
o he packe and he o he one o he acknowledgemen
(ACK). I he ACK
is
no ecei ed, he packe is
e ansmi ed
as
i
a
collision had occu ed. The e o e, he
p obabili y o a success ul ansmission is gi en by
wi h
q(n)
gi en by eq.
(1).
No e ha we assume ha all
picone s de ices (mas e o sla e) p oduce he same load
o e he picone .
I
we igno e p opaga ion e ec s ( ha
is,
/3,,=1
o all
n),
hen eq.
(2)
becomes:
which coincides wi h he esul in
[7].
This esul can be
easily in e p e ed as he p obabili y ha none o he o he
N-
1
picone s ansmi wi h he same ca ie han ow e e ence
955
picone in wo consecu i e slo s. Now we a e able o ob ain
he
h uughpi
( he a e o success ully ansmi ed packe s)
simply as
S= Ps,
ha
is,
he p obabili y ha a ansmission
a emp occu s
and
ha his a emp is success ul.
Equa ion (2) was ob ained assuming synch onized picone s.
The e m o consideing asynch onous picone s is always small
because in he Blue oo h s anda d a single slo packe is only
o du a ion
366ps,
while he ime-slo leng h
is
625~s. In a
p e ious wo k
[9],
we showed ha a simple exp ession o
Ps
o he asynch onous
ase
is
.-(
=
-
(2R )
Ay-''
,
which
is
simila
o
eq.
(3)
excep o he
ac o
2R.
He e
we ake in o
accoun ha
/M<<1(#79
in mos
cases,
and 4)
so
we neglec
he
e m
( /h.02.
In o he wo ds, he equa ions o he
synch onous case p o ide an accep able app oxima ion o he
asynch onous case
jus
assuming in
ou
equa ions an inc ease
o
ZR
(17%)
in he o e ed load.
As
discussed in sec ion
I.
in his pape we a e also
in e es ed in he es ima ion
o
message esponse imes. Le
us
i s compu e he mean delay ha
a
packe su e s due o
possible collisions wi h o he Blue oo h de ices, igno ing
he wai ing ime in he queues. The was ed slo s due o
collisions co espond o a sequence o Be noulli ials wi h
p obabili y o success
Ps
[9].
The numbe o was ed slo s is
he e o e geome ically dis ibu ed,
so
i s mean is:
(4)
Conside ing ha any de ice ansmi in al e na e slo s, he
mean delay including he success ul slo
is
exp essed in slo s o leng h
625p.
No e ha his exp ession
ep esen s he mean delay ha packe s su e due o possible
collisions wi h o he Blue oo h de ices, excluding wai ing
imes in he queues. Also no e ha we add one slo o he
success ul ansmission while a single slo packe is only
o
du a ion
R=366/625=0.5856
slo s. This di e ence should be
aken in o accoun mainly o small loads
(W<<).
A.
Conside ing he ime spen
in
he queue.
So
a ,
we ha e only conside ed he delay ha he packe
su e s due o collisions wi h o he Blue oo h de ices. In
single-sla e picone s, he o he sou ce o delay is he ime
spen wai ing on he node's queue. We model he de ice
queue
as
an
M/G/I
sys em. We assume ha messages a i e
ollowing a Poisson dis ibu ion, wi h mean
h
messages pe
second. The Blue oo h channel
is
conside ed a single-se e
sys em wi h an a bi a y se ice ime dis ibu ion. Fo such a
WGII
sys em, he a e age numbe o cus ome s in he
queue is gi en by he Pollaczek-Kinchin mean alue
o mula [12]
whe e a [q and
T
a e he a iance and he mean o he
se ice ime dis ibu ion, espec i ely, and
p
is he
u ilimlion uc o .
Now using Li le's law, we ge he mean
wai ing ime in he queue
The second exp ession comes om he ac ha
p=hT.
In
o de lo ela e his exp ession wi h he model de eloped
abo e, we use
( he o e ed load) ins ead
o
p,
as a measu e
o he p obabili y ha he sys em is busy.
I in e e ence om o he Blue oo h de ices is igno ed, hen
se ice ime
is
a cons an ( a [q=O) and he sys em
becomes an
M/D/I
sys em. Wi h se e al in e e ing picone s
he mean se ice ime should include he
slo s
was ed due o
collisions, and he e o e is gi en by 2W+2, wi h Wgi en by
eq.
(4).
No e ha he queue is emp ied only in al e na e slo s
and he e o e he se ice ime o he
WG/I
queue includes
wo slo s pe ansmission a emp . The a iance can be
ob ained since he a iance o he geome ic dis ibu ion o
he numbe o was ed slo s Wis:
By de ini ion V~I[T]=T~-(T~,
so
om eq.
(5)
we ge ha
he a iance
o
he se ice ime dis ibu ion is
a [T]=4 a [W].
As
a inal esul , he o al ime ha a packe
spends in he sys em can be ob ained as he sum o he ime
in he queue and he packe se ice ime i sel [eq.
(5)]:
This equa ion is exp essed in slo s o leng h
. =625ps.
A
inal obse a ion is ha he mean queuing delay
o
a no mal
WG/1
queue and ha o an
WG/I
queue wi h clocked
se ice di e s by hal he clock in e al
[
131.
The e o e, due
o he slo ed ime assump ion, we should modi y eq.
(9)
by
inc easing he mean delay by one slo o 625~s, since he
clock in e al o
OUT
M/G/l
sys em is 2. ( wo slo s pe
ansmission a emp ).
IV. VALIDATION
In o de o alida e he abo e model we pe o med se e al
simula ions wi h a de ailed Blue oo h simula o de eloped
using Ma lab 6.1.
Fo
he compa ison be ween analysis and
simula ion, we conside ed a a iable numbe o in e e ing
single-sla e picone s, igno ing p opaga ion losses o adio
wa es and o he e ec s like o wa d e o co ec ion. Tha
is, we assume ha a cochannel in e e ence will
always
956
des oy all packe s. Picone s a e synch onized and packe s
a e gene a ed wi h exponen ial in e -a i al imes. Finally,
since we a e only in e es ed in he compa ison be ween he
models, we assume single-slo packe s.
Figu e
1.
Channel h oughpu s. O e ed load o
a
a iable
numbe o in e e ing picone s.
In
Figu e
1
we show he h oughpu -o e ed load
cha ac e is ics o 2,
5
and
10
in e e ing picone s, om bo h
he simula ion and he analy ical model [using eq. (3)]. I
can be obse ed ha he es ima ion using he analysis is
e y close o he simula ion esul s. No e ha
as
he numbe
o in e e ing picone s g ows, he maximum h oughpu
dec eases because o he slo s was ed due o collisions.
Al hough no shown in Figu e
1,
a hump can he obse ed
wi h e y la ge N, ha is, maximum h oughpu is achie ed
a medium loads and hen h oughpu dec eases o highe
alues o he o e ed load [7]. The eason is ha in his
si ua ion he channel is almos “idle” (because o he was ed
slo s due o collisions) while he e a e packe s in he sys em
wai ing o he ansmi ed. Howe e his e y high numbe o
in e e ing picone s
(N>40
[7]) is no common in a eal
scena io. Fu he mo e, when
N
becomes la ge ,
a
highe
dis ance sepa a es Blue oo h de ices. As
a
esul , he e ec
on
he p obabili y o collision o de ices si ua ed a away
becomes almos negligible. Tha is, in eq.
(I)
he assump ion
o
,&=I
o all
n
is
no
longe alid. This in ui i e easoning
is
con i med
h ough
simula ion in [SI, whe e i is shown
ha he packe e o p obabili y emains almos cons an
as
N
inc eases. Ou model would he able o include his e ec
i we could es ima e adequa e alues o
p,,.
This is he
ma e o u he esea ch. Howe e , in his pape we
main ain he assump ion ha co-channel in e e ence always
des oy he packe s and o his eason we only conside
ela i ely low alues o
N.
Figu e 2a, 3b and 3c show he mean delay e sus o e ed
load cha ac e is ics o
N=2, N=5
and
N=10
espec i ely,
using a loga i hmic scale because o he e y high alues
o
he mean delay a high loads. The simula o coun s
366p
o he ansmission ime
o
one slo packe , while eq. (9)
includes one ull slo o
625ps
o he success ul
ansmission. This di e ence be ween he models would
p oduce small de ia ions limi ed o he zone o small loads.
Ins ead, we p e e o modi y eq.
(9)
by using he ac ual
ansmission ime o
R=366/625=0.5856
slo s ins ead o one
(success ul) slo . Wi h his change, he compa ison be ween
he models con i ms he accu acy o he analysis. No e ha
he mean delay inc eases e y apidly wi h he o e ed load.
I queuing delay is no conside ed, he
mean
delay inc eases
slowly wi h he o e ed load
as
expec ed om eq.
(5)
(see
[9]).
On
he o he hand, i is
a
known esul ha wai ing
imes in
a
queue depend on he i s and second momen o
he se ice ime dis ibu ion [12]. The e o e, he highe he
a iance o he se ice ime, he wo se he pe o mance. In
ou case he mean wai ing ime in he queue inc eases wi h
he a iance o he numbe o was ed slo s [eq. (7)]. This
a iance
is
gi en by eq.
(S),
whe e i can he seen ha i
g ows wi h
much mo e apidly han he delay due o
collisions i sel , since
Ps
is
a
dec easing unc ion o
.
This
esul is in ui i e since e ansmi ed packe s ha e he same
oppo uni ies han new packe s,
so
i may occu ha some
unlucky packe s su e e y la ge delays. As
a
esul o his
high a iabili y o he se ice imes, he queue size (and hus
he ime spen in he queue) g ows e y apidly wi h
,
especially when he numbe o in e e ing picone s is la ge.
This is he eason o he high alues o he mean delay in
Figu e
2.
Finally, an impo an obse a ion is ha he o e ed load
includes bo h new packe s
as
well
as
he e ansmission
a emp s o p e iously collided packe s. The e o e, he
ac ual o e ed load o e he channel may he much highe
han he “o iginal” load due o he highe -le el p o ocols
(see
[9]
o nume ical esul s). As
a
esul , he alues o he
o e ed load o which he mean delay suddenly inc eases
may
be
ela i ely no unusual.
V.
CONCLUSIONS
When many Blue oo h de ices coexis in he same a ea, he
p oblem o channel in e e ence may become o high
impo ance. In his pape we ha e p esen ed an analysis ha
adequa ely models he e ec o his in e e ence in
pe o mance. Pa icula ly, we ob ain simple exp essions o
he channel h oughpu and he mean packe delay. The
analysis has been alida ed h ough simula ion. We ha e
also discussed how o include he e ec o p opaga ion
losses
and he ac ha
in
gene al picone s
a e
no
synch onised. Al hough he delay due o
he
polling policy
in mul i-sla e picone s is no conside ed,
ou
esul s could
he use ul o any Blue oo h connec ion unde co-channel
in e e ence.
957
:
18'
I
.
",
Figu e
2.
Mean packe delay s. O e ed load.
Di e en ly om o he models in he li e a u e, he analysis
also includes he wai ing imes in he queues. This is
especially impo an in his case because al hough collisions
p oduce
an
inc ease in he mean se ice ime
o
a packe , i
is he
a iance
o
his ime ha ends o g ow e y apidly
wi h he o e ed load and he numbe o in e e ing picone s.
As a esul , he wai ing ime in he queues inc eases apidly
and becomes he mos impo an componen in he packe
delay. This e ec may become a se ious d awback in
applica ions like con inuous oice
o
ideo s eams, eal-
ime con ol, e c. e en in si ua ions whe e he numbe
o
in e e ing Blue oo h de ices
is
ela i ely low.
VI.
ACKNOWLEDGEMENTS
The Spanish Minis y o Science and Technology suppo ed
his wo k unde con ac s TIC2001-1868-'203-02 and
TIC2000-0087-P4-0
1.
REFERENCES
[I] P. lahamsan, M. Kazan zidis, R.
Kapoo ,
M.
Ge la,
Blue oo h:
An
Enablc o Pe sonal
A ea
Ne wo king, IEEE Nc wo k,
15
(2001) 28-
37.
[2] M. HBnnikZinen,
T.
D.
HZmBlainen, M. Niemi,
J.
Saa inen, T ends in
pe sonal
wi eless da a
communica ions,
Compu e Communica ions,
[3]
A.
Capone, M.
Ge la,
R.
Kapoo ,
E icien
plling
schemes o
Blue oo h picacells Communica ions, P oc. IEEE In . Con .
Comm.,
pp.
1990-1994,2001,
141
1.
Misic, V.B. Misic, Modeling Blue oo h picone pe o mance, IEEE
Communica ions
Lc e s
7(2003)
I8
-20.
[5]
J.C.
Haa sen,
The
Blue oo h Radio Sys em,. IEEE Pc sonal
Communica ions 7(2noo) 28-36.
[6]
A.
El-Hoiydi,
In e e ence
Be ween
Blue oo h Ne wo ks-Uppe Bound
on
hc Packe E o Ra e, IEEE Comm.
Le c s
5(2001) 245-247.
[7]
Y.
Lim,
J.
Kim,
S.L.
Min,
J.S.
Ma, Pe o mance E alua ion
o
he
Blue oo h-based Public
In e ne
ACCCSS
Poin .
P oc.
15'
In .
Con .
In o ma ion Nc wa king,
pp.
643-648.
Beppu,
Japan.
2001
[8]
A.
El-Hoiydi,
J.D.
Deco ignie. SoR-Dcadline Bounds o Two-way
T ansac ions in Blue oo h Picane s unde co-channel
In e e ence,
25
(1)
(2002)
pp.
84-99,
IEEE
In .
Con . Eme ging Technologies and Fac o y
Au oma ion, pp.
143-150.
Oc .
2001.
[9] J.L.Se illano
c
al.,
An
analy ical model o In e -channel In e e ence in
Bluc w h-Bascd Syslems.
P oc.
IEEE MWCN'02,
pp.
384-388.
[IO]
A.
Kamik
and
A.
Kuma ,
"Pe o mance o he Bluc oa h Physical
Laye ".
IEEE
h
Con
Pe sonal
Wi eless
Comm.,
pp.
70-74.
Hyde abad, India.
2000.
[I
I]
F. Mazzenga,
D.
Cassioli,
P.
Lo e i,
F.
Va ala o "E alua ion o
Packe
Loss
P obabili y
in
Blue oo h
Ne wo ks".
P oc.
IEEE
In/.
Con
Comm.,pp.
313-317.NVC,
2002.
[I21
B.R.
Ha e ko ,
"Pe o mance o Compu e Communica ion Sys ems".
Wiley, 1998.
[I31
W.
Bun.
"Local
A ea
Subnc wo ks:
A
Pe o mance
Compa ison",
IEEE
~ ans.
Comm.
COM-~~
(1981) 1465-1473.
[I41
D.
Cascado,
S.
Vicen e,
A.
Lina es,
J.
L. Se illano,
G.
Jimhez,
A.
Ci il.
"Inclusion
o p opaga ion
e ec s
on
packe
e o
a e
o
Blue w h
ne wo ks",
IEEE
Wo khop
on
Wi eless Ci cui s
and
Sys ems.
Vancou e , Canada. 2004.
s ockholm,
swede". 20n2.
958