TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.5, May 2014, pp
. 3761 ~ 37
6
8
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i5.5105
3761
Re
cei
v
ed
No
vem
ber 1
1
, 2013; Re
vi
sed
De
cem
ber 2
6
,
2013; Accep
t
ed Jan
uary 9
,
2014
Signal Detection Based on Particle Swarm Optimization
for MIMO-OFDM System
Chao
Qun Wu
1
, Dan Zhao
2
, JingPeng Gao
*
3
1
School of Aut
o
mob
ile a
nd T
r
affic Engin
eeri
ng, Hei
l
on
gji
a
n
g
Institute of
T
e
chn
o
lo
g
y
,
Harbi
n
, 150
00
1, Chin
a
2,3
College of Information and Co
mmu
ni
ca
tion
En
g
i
n
e
e
r
i
n
g
,
H
a
rb
in
En
gi
nee
ri
ng
U
n
i
v
e
r
si
ty
,
Harbi
n
, 150
00
1, Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: gaoj
ing
p
e
n
g
@
hrbe
u.ed
u.cn
A
b
st
r
a
ct
In order to over
come the defe
c
ts of the slow
c
onver
genc
e rate of the tradi
tion
al Genetic A
l
gorit
h
m
and
basic P
a
r
t
icle Sw
arm O
p
timi
z
a
ti
on dr
o
p
s into
l
o
ca
l
opti
m
u
m
easi
l
y
, an i
m
prov
e
d
Particl
e
Sw
arm
Optimi
z
a
t
i
o
n
a
l
gorit
hm
bas
ed
on hybr
id a
l
g
o
rith
m is pr
op
osed
and
app
l
i
ed to the s
i
g
nal d
e
tectio
n for
MIMO-OF
D
M system. T
he al
g
o
rith
m opti
m
i
z
e
s
the basic
Par
t
icle Sw
arm Optimi
z
a
ti
on al
g
o
rith
m an
d so
me
prob
le
ms w
e
re
solve
d
by
me
ans of P
a
rticle
Sw
arm Opti
mi
z
a
t
i
o
n
co
mbin
ed w
i
th Gen
e
ti
c Algor
ith
m
fo
r
sign
al d
e
tectio
n. T
h
roug
h the
theoretic
al an
alysis a
nd the
simulati
on res
e
arch, this i
m
pr
oved a
l
g
o
rith
m is
super
ior to b
a
s
i
c Particle Sw
a
r
m Opti
mi
z
a
t
i
o
n
al
gorith
m
0.5
d
B un
der th
e same n
u
m
ber
of iteratio
ns an
d
is
better than traditional Gene
tic Al
go
ri
thm
0
.
5d
B un
de
r th
e sa
me
b
i
t e
r
r
o
r r
a
te. T
h
is
alg
o
ri
thm
i
m
prov
es t
h
e
system
signal
detection perform
a
nce
effectively with less
iteration
and
reduces the bit error rate. It has
rapi
d spee
d of conver
genc
e a
nd str
ong ca
pa
bility of gl
ob
al search.
Ke
y
w
ords
: MIMO, O
F
DM, particle swarm
optim
i
z
at
io
n, hybr
id al
gorith
m
, si
gna
l detecti
on
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
The com
b
ina
t
ion
of
O
r
tho
gonal
Freque
ncy Division
Multiplexing (OFDM
)
a
nd Multiple
Input Multipl
e
Outp
ut (MI
M
O) i
s
one
of key te
ch
n
o
logie
s
i
n
th
e fourt
h
ge
n
e
ration
of m
obile
telecom
m
uni
cation [1]. OFDM can imp
r
ove the sp
e
c
tral efficie
n
cy and spatial
multiplexing
gain
of a system
can be extra
c
t
ed thro
ugh M
I
MO techni
qu
e. In this way,
the com
b
inat
ion of them can
improve the
spe
c
tral
effici
ency an
d dat
a tran
smi
ssio
n
rate
s. Mea
n
whil
e, introd
ucin
g MIMO
to
OFDM
can
achi
eve bro
a
dban
d OFDM system, whi
c
h is ad
opting tran
smitter array whi
c
h
con
s
i
s
ts
of a
l
a
rge
nu
mbe
r
of low-po
we
r
transmitte
rs t
o
elimin
ate th
e shad
ow effect a
nd
achie
v
e
f
u
ll cov
e
rag
e
.
MI
MO
sy
st
e
m
ca
n
comb
at
mult
i-pat
h
fading, but it
is not
suita
b
l
e
for frequ
en
cy
sele
ctive fadi
ng ch
ann
el in the
MIMO system, whi
c
h
can be over
co
me by OFDM
[2]. The Guard
Interval (GI
)
can
eliminate
inter-symbol
interfer
en
ce
(ISI) and
Cyclic Prefix (CP
)
ca
n elimin
a
t
e
Inter-
ca
rrie
r
Interfer
en
ce (I
CI).
In ord
e
r to
achi
eve excellent tra
n
sm
i
ssi
on p
e
rfo
r
mance fo
r
MIMO-OF
D
M
system,
pre
c
ise
sign
al
dete
c
tion i
s
very ne
ce
ssa
r
y bef
o
r
e d
e
m
odulatio
n [3
]. In MIMO-O
FDM
system,
the
perfo
rman
ce
and the com
p
lexity of signal dete
c
tion
algorithm of
system rece
iver will directly
affect the qu
ality of the entire commu
n
i
cation
sy
ste
m
. Signal det
ection
algo
rithm with ex
ce
llent
detectio
n
pe
rforman
c
e i
s
often accom
panie
d
by
hi
gher
co
mple
xity;
the real
ization
of hi
gh
compl
e
xity algorithm i
s
often limited by
hardwar
e processing
capability. Theref
ore, developi
n
g
an alg
o
rithm
with optimal
sign
al dete
c
ti
on pe
rform
a
n
c
e a
nd mo
de
rate compl
e
xity is the key
to
achi
eve satisfacto
ry pe
rforma
nce of
receiv
er fo
r MIMO-OF
D
M sy
stem.
Signal det
ection
method
s
gen
erally i
n
clu
d
e
linear dete
c
ti
on a
n
d
nonli
n
ear
dete
c
tion
in the
field
of
sign
al d
e
tecti
o
n
for MIMO-OF
D
M
system
so far. T
he
re
search
of line
a
r
dete
c
tion
m
a
inly focuses
on Ze
ro
Fo
rci
ng
(ZF) dete
c
tio
n
[4], Minimu
m Mea
n
Sq
u
a
re
Error (M
MSE) dete
c
ti
on [5], an
d L
i
near Minim
u
m
Mean
Squa
re
Error
(LM
M
SE) dete
c
tion
[6]. The
de
sign of li
nea
r
detectio
n
al
g
o
rithm i
s
sim
p
le
and
ea
sy to
be impl
emen
ted, but the
detectio
n
p
e
rf
orman
c
e
is
worse, it i
s
n
o
t suita
b
le to
be
applie
d sepa
rately in practi
cal
system. T
he res
earch i
n
nonli
nea
r d
e
tection
main
ly focuses
on
QR de
co
mpo
s
ition dete
c
ti
on algo
rithm
[7], Serial Interferen
ce Ca
ncell
a
tion det
ection al
gorit
hm
[8] and V
-
BL
AST dete
c
tio
n
alg
o
rithm
[9
].The ab
ove
algorith
m
s de
tect laye
r by l
a
yer to
elimin
ate
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3761 – 37
68
3762
the inte
rfere
n
c
e, thei
r
pe
rforma
nc
e i
s
li
mited by th
e
first lin
ear de
tection
accu
racy. Th
e e
r
ro
r of
the first linea
r dete
c
tion re
sults
will in
crease the er
ro
r pro
bability
of interfere
n
ce can
c
ell
a
tio
n
in
the se
con
d
d
e
tection a
nd
bring
cum
u
lat
i
ve erro
r.
Many intelligent optimization algorithms
have bee
n put forward, these
ki
nds of detection
algorith
m
s are p
r
opo
se
d
based
on th
e lo
cal
optim
iz
ation
of int
e
lligen
ce
alg
o
rithm. Intelli
gent
algorith
m
ca
n solve the
optimizatio
n
probl
em of
functio
n
s, such as
cal
c
ulati
ng the maxi
mum
value or the
minimum val
ue of a functi
on in a
sol
u
tion sp
ace .In the field of signal dete
c
tio
n
,
intelligent alg
o
rithm takes the maximum likeliho
od
function a
s
the obje
c
tive function a
n
d
all
possibl
e tran
smit sig
nal
vector a
s
th
e sol
u
ti
on
space. The a
l
gorithm imit
ates the
nat
ural
pro
c
e
s
ses of
intelligent
o
p
timization
in
col
onial
org
anism
s to
fin
d
the
optimal
tran
smit
sig
nal.
These si
gnal
detection m
e
thod
s first
appe
are
d
in
the CDMA
system multi
-
user d
e
tecti
on.
Kech
riotis a
pplied
Ho
pfield n
eural n
e
twork
to
th
e CDMA
sy
stem to
solve the m
a
ximum
likeliho
od d
e
tection
whi
c
h
is a NP p
r
o
b
l
e
m [10]. Jian
g Ming an
d
Han
z
o p
r
o
p
o
s
ed a
multi-u
s
er
detectio
n
alg
o
rithm ba
sed
on Geneti
c
Algorithm for M
I
MO-OF
D
M systems [11]. Lou
w and Botha
prop
osed sp
here d
e
tectio
n based on
Hopfiel
d
neu
ra
l netwo
rk fo
r MIMO syst
em [12]. Higuchi
and Kaiwa propo
sed a MI
MO system
si
gnal dete
c
ti
o
n
algorith
m
b
a
se
d on tabo
o sea
r
ch [13].
In this p
ape
r, we p
r
e
s
ent
an imp
r
ove
d
Particl
e
S
w
arm Optimi
zation
sig
nal
detectio
n
algorith
m
wit
h
good p
e
rfo
r
mance and lo
wer
com
p
lexi
ty with the same numb
e
r o
f
iterations an
d
bit error
rat
e
, whi
c
h i
s
com
b
ining
hybrid
genet
ic alg
o
rithm
with ba
si
c Particle
Swarm
Optimizatio
n
to minimize the negative effects of t
he e
r
ror diffusio
n
, its pe
rform
ance is su
pe
rior t
o
other t
r
aditio
nal alg
o
rithm
s
. The
re
st o
f
the pap
er
i
s
org
ani
zed
a
s
follo
ws.
Firstly, we d
e
scribe
the syste
m
model of MI
MO-O
FDM.
Secon
d
ly,
the propo
sed
sign
al dete
c
t
o
r of MIMO
-OFDM
system
ba
se
d on
Parti
c
le
Swa
r
m
Opti
mization
si
gn
al dete
c
tion
algorith
m
i
s
i
n
trodu
ce,
an
d the
simulatio
n
re
sults a
r
e obta
i
ned. Finally, some
con
c
lu
sions a
r
e d
r
a
w
n in this se
cti
on.
2. Sy
stem Model
The sy
stem
block dia
g
ra
m of MIMO-OFDM
i
s
sh
own
as Fi
gu
re 1. Th
e transmitted
sign
als go th
roug
h co
de mappin
g
. before pa
ssing the se
rial-to
-
p
a
rallel
conve
r
sio
n
and IF
FT
modulatio
n. At last, a cyclic prefix is ad
ded to
them. They are tra
n
s
mitted via different tran
smit
antenn
a re
sp
ectively. Afte
r goin
g
thro
u
gh a freq
uen
cy-sele
c
tive cha
nnel, the
receivers firstly
move the
cyclic p
r
efix from
the sig
nal
s.
Seco
n
d
ly, the
re
ceived
sig
nals
are
tran
sform
ed by F
FT
,and then pa
ss throu
gh a p
a
rallel
-
to-se
r
i
a
l conve
r
si
on
. Finally, the r
e
ceive
d
sign
a
l
s go thro
ugh
a
decode
r and t
he re
ceive
r
s
get the re
covered o
r
igi
nal data.
Mod
u
la
ti
o
n
Bi
na
r
y
da
ta
MIMO
Encoder
OFDM
Mo
du
l
a
tion
OFDM
Mo
du
l
a
tion
OFDM
Mo
du
l
a
tion
...
Se
ri
a
l
P
a
ralle
l
IFFT
CP
Ins
e
rt
ion
Pa
r
a
llel
Se
ri
al
OF
D
M
M
o
du
l
a
t
i
on
OF
DM
D
e
m
odul
at
ion
OF
DM
D
e
m
odul
at
ion
OF
DM
D
e
m
odul
at
ion
MI
MO
Dec
oder
Dem
odul
ati
o
n
Bi
n
a
r
y
da
ta
...
Seria
l
Pa
r
a
l
l
e
l
FF
T
CP
Re
m
ova
l
Pa
r
a
l
l
e
l
Se
r
i
a
l
OFD
M
Dem
odula
t
i
o
n
Figure 1. Block
Diag
ram o
f
a MIMO-OF
D
M System
Model
For a
MI
M
O
sy
st
em
wit
h
Nt
transmitting a
n
tenna
s
and
Nr
re
ceiving
a
n
tenna
s, hi
gh-
spe
ed d
a
ta st
ream
s g
o
through
a seri
al
-to-p
a
rall
el co
nversi
on, the
r
e is
d
N
data in
each g
r
oup.
The vecto
r
form of the
n
-th set of data in the
t
i
-th tran
smitting anten
na is a
s
follo
ws:
T
d
i
i
i
i
n
N
d
n
d
n
d
n
d
t
t
t
t
)
,
1
(
,
),
,
1
(
),
,
0
(
)
(
(1)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Signal Dete
ct
ion Base
d on
Particle Swarm
Optim
i
zation for MIMO-O
FDM…
(Cha
oQun
Wu
)
3763
In ord
e
r to eli
m
inate the
ef
fects
of ISI and ICI,
t
i
d
is
coupled
with a Cyclic P
r
efix whose
length i
s
CP
N
after the IFFT transfo
rmatio
n, then we
ca
n
get the time-domain
OF
DM symbol.
The vecto
r
form of the
n
-th OFDM
symb
ol in the
t
i
-th transmitting a
n
t
enna is a
s
fo
llows:
T
CP
d
i
i
i
i
n
N
N
x
n
x
n
x
n
x
t
t
t
t
,
1
,
,
1
,
,
0
)
(
(2)
Whe
r
e,
1
0
2
exp
)
,
(
)
,
(
d
t
t
N
i
d
CP
i
i
N
N
m
l
j
n
l
d
n
m
x
(3)
The re
ceivin
g
antenna
rem
o
ve the cycli
c
pref
ix from the re
ceived
signal
s then carry out
the FFT tran
sform a
nd g
e
t the initial
data of the tran
smitter in
freque
ncy do
main. The
n
-t
h
OFDM
sig
nal
at the
k
-th su
bca
rri
er re
cei
v
ed
by
the
r
i
-th re
ceiving
a
n
tenna
ca
n b
e
de
scribe
d
as:
)
,
(
)
,
(
)
(
)
,
(
1
0
,
n
k
W
n
k
X
k
H
n
k
Y
r
t
t
t
t
r
r
i
N
i
i
i
i
i
(4)
Then, we
can
get the matrix form of t
he
MIMO-OF
D
M
transmi
ssion
system mod
e
l as:
)
,
(
)
,
(
)
(
)
,
(
n
k
W
n
k
X
k
H
n
k
Y
(5)
Whe
r
e,
T
Nr
n
k
Y
n
k
Y
n
k
Y
n
k
Y
)]
,
(
,
),
,
(
),
,
(
[
)
,
(
1
1
0
rep
r
e
s
ent
s
re
ceive
d
si
gnal
vect
or
at the
k
-th sub
c
arrie
r
;
T
Nt
n
k
X
n
k
X
n
k
X
n
k
X
)]
,
(
,
),
,
(
),
,
(
[
)
,
(
1
1
0
rep
r
e
s
ent
s tran
smi
tted
sign
al vecto
r
at the
k
-th sub
c
a
rri
e
r
;
T
Nr
n
k
W
n
k
W
n
k
W
n
k
W
)]
,
(
,
),
,
(
),
,
(
[
)
,
(
1
1
0
rep
r
e
s
ent
s th
e noi
se
vecto
r
.
)
(
k
H
mean
s th
e
ch
ann
el freq
uen
cy
respon
se matrix of
the
k
-th
sub
c
a
rri
er, it can b
e
de
scri
bed a
s
follows:
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
(
1
,
1
1
,
1
1
,
0
1
,
1
1
,
1
1
,
0
0
,
1
0
,
1
0
,
0
k
H
k
H
k
H
k
H
k
H
k
H
k
H
k
H
k
H
k
H
Nr
Nt
Nr
Nr
Nt
Nt
）
(6)
3. Dete
ction
Process a
n
d
Optimi
ztion Algorithm
Descriptio
n
3.1. Particle S
w
arm Opti
miz
a
tion Algorithm
In1995, Dr. Eberh
a
rt and
Dr. Kenn
edy prop
osed
Particle Swa
r
m Optimizatio
n
algorith
m
[15]. It is a
n
evolution
a
ry
tech
nolo
g
y, origi
nat
ing
from th
e
re
search
on
bird flock
preying
behavio
r. Particle
swarm
O
p
timization
is
based
on t
h
e
ob
servatio
n
of clu
s
te
r a
c
ti
vity behavior,
it
use
s
th
e info
rmation
sh
arin
g of in
dividual
s in
g
r
ou
p to
make
the
enti
r
e
gro
up
mov
e
from
di
so
rd
er
to ord
e
r in
th
e solvin
g sp
ace
and fin
a
l
l
y obtain
the
optimal
solu
tion. All the particl
es hav
e a
fitness valu
e determi
ned b
y
the optimization f
unctio
n
and a spe
e
d
determin
e
s the dire
ction a
n
d
distan
ce. The
pa
rticle
s adj
ust
the
spee
d
dyna
mica
ll
y acco
rding
t
o
its own flyi
ng exp
e
rie
n
ce a
s
well a
s
the flying experi
e
n
c
e from their
compani
on.
The initiali
zat
i
on is
a g
r
ou
p of ra
ndom
particl
es. T
h
e
n
we
ca
n get
the optimal
solutio
n
throug
h itera
t
ion. Durin
g
each iteratio
n, the
particles upd
ate themselves b
y
trackin
g
two
extreme
s
. Th
e first is the
optimal solution found
by
the parti
cle itself. This
sol
u
tion is calle
d
person
a
l be
st
. The othe
r e
x
treme is
call
ed group
be
s
t, which i
s
the
optimal solut
i
on of the e
n
tire
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3761 – 37
68
3764
grou
p. Assu
me that there
is a comm
un
ity composed
of
N
particl
es i
n
a
D
-dime
n
sio
nal se
ar
ch
spa
c
e, ea
ch
particl
es
rep
r
ese
n
ts a
s
a
D
-dimen
sion
al vector.
N
i
x
x
x
X
iD
i
i
i
,
,
2
,
1
),
,
,
,
(
2
1
(7)
The sp
eed of
the particl
e is a
D
-dimen
sio
n
a
l vector, den
oted as
N
i
v
v
v
V
iD
i
i
,
,
2
,
1
),
,
,
2
1
i
，
（
(8)
Personal b
e
st
is also a
D
-di
m
ensi
onal ve
ctor, de
noted
as
N
i
p
p
p
p
iD
i
i
best
,
,
2
,
1
),
,
,
,
(
2
1
(9)
Whe
n
findi
ng
the two
be
st
values, th
e p
a
rticle
s
upd
ate the
sp
eed
a
nd p
o
sition
a
c
cordi
n
g
to the followin
g
Equation (1
0) and
(11
)
.
)
(
2
2
1
1
1
q
id
q
gd
q
id
q
id
q
id
q
id
x
p
r
c
x
p
r
c
wv
v
(10
)
1
1
q
id
q
id
q
id
v
x
x
(11
)
3.2. Signal Detec
t
ion Pro
cess
Bas
e
d on Particle Sw
a
r
m Optimi
zatio
n
Algori
t
hm
A
sig
nal dete
c
tion sche
me
ba
sed
on
Pa
rticle
Swa
r
m Optimizatio
n
algorith
m
fo
r MIMO-
O
F
D
M
s
y
s
t
em is
d
e
s
i
gn
ed
in
th
is
pa
pe
r
.
In
o
r
d
e
r t
o
achieve a
sufficiently go
od pe
rform
a
n
c
e of
Particle Swarm Optimizati
on sig
nal det
ection al
gor
it
hm, the main
param
eters of Particle S
w
arm
Optimizatio
n
algorith
m
are
desi
gne
d as f
o
llows:
Firstly, species initiali
zati
on.
q
is the
nu
mber
of itera
t
ions;
1
c
,
2
c
is th
e accel
e
ratio
n
coeffici
ent, they usu
a
lly eq
ual 2;
2
,
1
r
is a
ra
ndom
numb
e
r
bet
wee
n
0 t
o
1;
q
id
v
is the
speed i
n
dth-dim
e
n
s
io
n of the qth
it
eration of p
a
rticle i;
id
x
is th
e cu
rrent po
sition of p
a
rti
c
le i in dth
-
dimen
s
ion;
id
p
is the be
st po
sition of pa
rti
c
le i in dth
-
di
mensi
on;
gd
p
is the be
st po
sition of
whol
e gro
up i
n
d
th
-
d
imen
s
i
on
.
Sec
o
ndly, parameter
s
e
ttings
. In order to pr
event t
he p
a
rticl
e
s
away from th
e search
space, the
particle velocit
y
of
each di
mensi
on
will
be confined to
]
,
[
max
max
d
d
v
v
, gene
rally
max
max
d
d
kx
v
,
1
1
.
0
k
, each dime
nsio
n use the
same
setting
s.
Finally, the selectio
n of the fitness fun
c
tion.
The fitness functio
n
is the stan
da
rd based
on the
maxim
u
m li
keliho
od
to asse
ssme
nt the
syst
em
dete
c
tion
pe
rforman
c
e
go
o
d
o
r
b
ad, a
n
d
it
is non
-ne
gati
v
e:
]
2
[
max
arg
}
{
min
arg
2
Hx
H
x
y
H
x
X
H
y
T
T
T
T
x
x
x
(12
)
Set the object
i
ve function o
f
the MIMO-O
FDM si
gnal d
e
tector
as:
Hx
H
x
y
H
x
x
T
T
T
T
2
)
(
(13
)
Assu
me th
at the o
b
je
ctive fun
c
tion
achieves maxi
mum valu
e
whe
n
x i
s
b
, then th
e
obje
c
tive function value is
)
(
b
, beca
u
se
)
(
b
ca
n be po
sitive or neg
ative, in orde
r to en
sure
the fitness fu
nction i
s
non
-negative, the fitne
ss fun
c
tio
n
in sign
al de
tection is
set as:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Signal Dete
ct
ion Base
d on
Particle Swarm
Optim
i
zation for MIMO-O
FDM…
(Cha
oQun
Wu
)
3765
))
(
exp(
)
(
b
b
f
(14
)
Whe
r
e
is a
norm
a
l num
b
e
r in th
e exp
e
rime
nt of this pa
pe
r,
is
set to 0.1. So
that
the best d
e
te
ction of the M
I
MO-OF
D
M
signal can
be
descri
bed a
s
a PSO optimal individual, that
is the output
of the detecti
on.
3.3. MIMO-OFDM
Sign
al De
tec
t
io
n Proce
s
s
Ba
sed
on
improv
ed
Particle S
w
a
r
m
Optim
i
za
tion
In order to overcome the p
r
oble
m
of basic
Particle Swarm Optimi
za
tion algorithm
easily
fall into lo
cal
extremum, it
is
requi
re
d to find a
n
alg
o
rithm
with g
ood gl
obal
search
ability, the
Geneti
c
Algo
rithm ca
n effectively prev
ent the sea
r
ch pro
c
e
ss fal
ling into local
extremum [1
6].
This al
gorith
m
is ba
sed
on the pri
n
ci
ple of hybr
id
in Geneti
c
algorith
m
. In each ite
r
ati
on, a
spe
c
ified n
u
m
ber of pa
rticle
s are
sel
e
cted into
th
e hybrid po
o
l
accordi
ng to hybridi
z
ati
on
prob
ability. Particle
s
hybri
d
ize
ra
ndoml
y
in the
po
ol
and the
n
p
r
o
duce the
sam
e
num
ber of filial
gene
ration p
a
rticle
s. After that, filial
generatio
n par
ti
cle
s
su
bstitut
e
the pare
n
tal particl
es.
The
positio
n of filial gene
ration
is cal
c
ul
ated
by the
arithm
etic cro
ssove
r of the pare
n
t
al position
：
q
parent
q
parent
q
child
x
p
x
p
x
2
1
)
1
(
1
(15
)
Whe
r
e
p
is
a
ran
dom
nu
mber
betwee
n
0 to 1. T
he spee
d of
filial gene
ra
tion is
cal
c
ulate
d
by (16
)
.
q
parent
q
parent
q
parent
q
parent
q
parent
q
child
v
v
v
v
v
v
1
2
1
2
1
1
(16
)
The dete
c
tio
n
step
s ba
sed on hybri
d
parti
cle
swa
r
m algo
rithm
optimization
are as
follows
:
Step 1: Initialize the po
sitio
n
x
and sp
eed
v
of each pa
rticle ra
ndo
mly.
Step 2: Evaluate the fitnes
s of each
particl
e and
store the
current po
sition
and the
adaptatio
n value
s
in the
best p
o
sitio
n
of each pa
rt
icle, then
sto
r
e the
be
st individual fitn
ess
value amon
g all pbe
st in the gbe
st.
Step 3: Updat
e the spe
ed
v
and
po
sition
x
of each p
a
rticle.
Step 4: Com
pare th
e fitness value of
each
parti
cl
e as
sho
w
n
in (14
)
with t
he be
st
positio
n it has experien
c
e
d
and take the better one a
s
the curre
n
t best po
sition;
Step 5: Comp
are the
curre
n
t value of
id
p
and
gd
p
then upd
a
t
e the gbest.
Step 6: Elect a sp
ecifie
d numb
e
r of
particl
es int
o
the hybrid
pool a
c
cording to
hybridi
z
ation probability, particl
es hybridized
randomly in t
he pool and pro
duce the
same
number
of filial generation parti
cles
(child), then upda
te the
position and ve
locity of the filial
gene
ration by
(15) a
nd (1
6), maintaining
id
p
and
gd
p
unchan
ged.
Step 7: Stop whe
n
con
d
ition whi
c
h
is the
co
m
puting p
r
e
c
ision or the n
u
mbe
r
of
iteration
s
is
met, the sea
r
ch
stop
s an
d
outputs
re
su
lts, otherwise
returns to
st
ep3 to
conti
nue
the sea
r
ch.
Whe
n
the p
r
eset n
u
mbe
r
of iteration
s
and
a
c
curacy
is re
ached, t
he search
st
ops
and
the output re
sult is the optimal detectio
n
signal.
4. Simulation Resul
t
s
In this
pap
e
r
, the
simul
a
tion of the
algorith
m
s m
entione
d b
e
fore
is cond
u
c
ted to
demon
strate
the perfo
rmance of th
e pro
p
o
s
ed
sign
al dete
c
tion meth
o
d
. The
simu
lation
para
m
eter va
lues a
r
e
sho
w
n in Ta
ble 1
.
Assumin
g
th
at the sen
d
in
g and receiving anten
na
s
are
indep
ende
nt.
The simul
a
tions are
pe
rfo
r
med
i
n
a
4×4 MIMO-OF
D
M syste
m
wit
h
one
path.
We
assume th
at the cha
nnel
state informati
on (CSI)
is
known. The
sende
r u
s
e
s
the mod
u
latio
n
of
BPSK and 8QAM. The transmi
ssion
power of the sende
r is
1. Each
noise is
compl
e
x-valued
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
0
46
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3761 – 37
68
3766
additive Ga
u
ssi
an
white n
o
ise, a
nd it is indep
end
ent
and id
entical
l
y distribute
d
Gau
ssi
an
whi
t
e
noise with zero mean; the
MIMO-OF
D
M
system is
sin
g
le path.
Table 1. Tabl
e of Paramet
e
rs Setting fo
r Algorithm Si
mulation
Parameter value
Maximum velocity
3.5
Inertia w
e
ight
[0.4,0.9]
Acceleration coefficient(c1)
2
Acceleration coefficient(c2)
2
Length of CP
512
Points of FFT
2048
Transmitting ant
enna
4
Receiving antenna
4
Figure 2 sho
w
s th
e difference in the
bit er
ror rate
unde
r
the
condition of d
i
fferent
iteration
s
an
d simulatio
n
para
m
eters b
e
twee
n
improved parti
c
le
swa
r
m opti
mization d
e
te
ction
algorith
m
and
maximum likelihoo
d dete
c
tion algorith
m
.
Figure 2. BER of IPSO Algorit
hm in
Different Iterations
The sim
u
latio
n
results sho
w
that the maximum
num
b
er of iteratio
ns of 25 is su
perio
r to
the maximu
m num
ber
of iteration
s
is 2
0
a
n
d
15 u
nde
r t
he imp
r
oved
Particl
e
S
w
arm
Optimizatio
n
algorith
m
. Th
e maximum
numbe
r of iteration
s
2
0
i
s
better th
an
the maximu
m
numbe
r of ite
r
ation
s
1
5
wit
h
abo
ut
0.5d
B, but there
i
s
no
si
gnifica
n
t increa
se i
n
perfo
rma
n
ce
of
maximum nu
mber
of iterations
25 to
20. The
r
e
a
s
on
is
g
a
i
n r
e
du
c
t
io
n du
e
to
th
e
e
r
r
o
r
accumul
a
tion
with
the i
n
crease of
the
n
u
mbe
r
of
iterations. At th
e
sa
me tim
e
, its p
e
rfo
r
ma
nce is
clo
s
er to opti
m
al ML
(Maxi
m
um Li
kelih
o
od)
dete
c
tor
whe
n
the m
a
ximum num
b
er of ite
r
ation
s
i
s
20, but has l
o
we
r co
mputi
ng co
st. We
can
con
c
lu
de
that the proposed algo
rit
h
m can d
e
te
ct
sign
al well an
d has fa
ster
converg
e
n
c
e speed.
Figure 3 an
d Figure 4
give the bit
error
rate pe
rforma
nce cu
rve of the Maximum
Likeli
hood
de
tection alg
o
rit
h
m, Zero
-Fo
r
cing d
e
tect
io
n algo
rithm, Particle S
w
arm Optimizati
on
detectio
n
alg
o
rithm, Ge
ne
tic Algorith
m
detectio
n
a
nd imp
r
oved
Particle
Swarm O
p
timization
detectio
n
al
g
o
rithm
ba
sed
on
hybrid
u
nder the
sa
m
e
sim
u
lation
para
m
eters i
n
the m
odul
a
tion
scheme of B
PSK and 8QAM respectiv
e
ly.
The sim
u
lati
on results
show that: Both
in BPSK
and 8QAM
modulation mode, the
perfo
rman
ce
of the improved Parti
c
le
Swarm
Op
ti
mization
ba
sed on
hybrid
is si
gnifican
t
ly
improve
d
co
mpared with
Zero Fo
rcing
detectio
n
alg
o
rithm, Particle Swarm det
ection alg
o
rit
h
m
and G
eneti
c
Algorithm
with the sa
me
numbe
r of it
e
r
ation. Thi
s
i
s
be
ca
use that the improv
ed
Particle S
w
arm O
p
timization alg
o
rit
h
m ha
s
stronge
r
glo
bal
sea
r
ch a
b
i
lity and faster
conve
r
ge
nce rate. In the case of the
sa
me maxi
mum
iteration time of 20, there is only 0.5
d
B
differen
c
e
be
tween
its
pe
rforma
nce a
nd Maxim
u
m
Likeliho
od
algorith
m, bu
t it has lo
wer
comp
ute co
m
p
lexity.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Signal Dete
ct
ion Base
d on
Particle Swarm
Op
tim
i
zation for MIMO-O
FDM…
(Cha
oQun
Wu
)
3767
Figure 3. BER of Different
Detectio
n
Algorithms
under BPSK
Figure 4. BER of Different
Detectio
n
Algorithm
s
un
der 8QA
M
5. Conclusio
n
In this
pap
er,
we
propo
se
an im
proved
Particl
e
Swarm
Optimiza
tion alg
o
rith
m whi
c
h
combi
n
e
s
Ge
netic Algo
rith
m and Particle Swarm O
p
timization alg
o
rithm to det
ect sig
nal
s for
MIMO-OF
D
M
system.
We assu
me t
hat the
ch
annel information is
acc
u
rate. It is
a
comp
re
hen
si
ve utilization of the good global c
onve
r
gen
ce of Ge
netic Algorith
m
and the quick
learni
ng abili
ty of Particle Swarm O
p
timization
a
l
gorithm, so
that the hybrid optimi
za
tion
algorith
m
h
a
s
b
e
tter
para
llel processin
g
po
we
r
and
faste
r
conv
erge
nce tha
n
the tradition
al
algorith
m
s.
We
sim
u
late
and
analy
z
e t
he al
gorith
m
s mentio
ned
i
n
this pa
pe
r
and
co
mpa
r
e
the
perfo
rman
ce
s between
different
algo
rith
ms. Simul
a
ti
o
n
results
sh
o
w
that
the i
m
proved
Pa
rticle
Swarm
Opti
mization
algo
rithm is
bette
r than tradi
tio
nal Ge
netic
Algorithm a
n
d
Particl
e
Swarm
Optimizatio
n
algorith
m
, its perform
an
ce
is close
to the ideal dete
ction of Maximum Likeli
ho
od
algorith
m
. Signal-to
-
noi
se
ratio loss i
s
on
ly 0.5 dB.
Beside
s, the
co
mplexity is rel
a
tively low. T
h
e
results
dem
onstrate the
effectivene
ss an
d t
he
appli
c
ability
of improve
d
Particle S
w
arm
Optimizatio
n
in sign
al dete
c
tion for MIM
O
-OF
D
M
syst
ems.
Ackn
o
w
l
e
dg
ements
This work was supp
orted
by the Science and T
e
ch
nolo
g
y Proje
c
t of Hei
l
ongjian
g
Educatio
n Departm
ent (1251
1455
)
and Nationa
l
Trainin
g
Prog
ram
s
of Innovation a
n
d
Entreprene
urship for Unde
rgradu
at
es “S
tudy on Train
s
Ope
r
ation
Control Simulation Sys
t
em for
Urb
an
Rail T
r
ansit”. T
he a
u
thors
would
like to th
a
n
k t
h
e pa
per edit
o
r a
nd the
re
viewers fo
r th
eir
valuable
com
m
ents an
d su
gge
stion
s
.
Referen
ces
[1]
Du Yue
lin, Z
h
ang Ji
ng
xi
an. T
he performan
ce of
s
y
nc
hron
izatio
n alg
o
rith
m in real-time
O
F
DM-PO
N
sy
s
t
e
m
.
T
E
LK
O
M
NIKA Indon
esia
n Journ
a
l o
f
Electrical Eng
i
ne
erin
g
. 201
2; 10(7): 178
4-1
794.
[2]
Z
hu Yon
gho
ng
, F
eng Q
i
ng, W
ang Jia
n
h
o
n
g
. Neura
l
net
w
o
rk-bas
ed a
d
a
p
tive pass
i
ve o
u
tput feed
bac
k
control for MIMO
uncertain s
ystem.
T
E
LKOMNIKA Indone
sian Jo
urna
l
of Electrical E
ngi
neer
ing
. 20
12
;
10(6): 12
63-
12
72.
[3]
JingP
eng
G
a
o
,
DanF
en
g Z
h
ao, Ch
aoQ
u
n
W
u
. A Joint
S
y
nc
hro
n
izati
o
n an
d C
h
a
n
n
e
l Estimati
o
n
Algorit
hm for MIMO
-OF
D
M S
y
stem.
T
E
LK
O
M
NIKA Indonesi
an Jo
urna
l
of Electrical
Engi
neer
in
g
.
201
3; 11(1
2
): 7571-
757
9.
[4]
W
an F
,
Z
hu
W
P
, S
w
am
y
MNS. Ch
ann
el
estimati
on
of
pu
lse-sh
ap
ed
multi
p
le-
i
np
ut multi
p
le-
outp
u
t
orthog
on
al freq
uenc
y d
i
visi
on multipl
e
xi
ng
s
ystems.
IET Commu
n
icati
ons
.
201
0; 4(17): 21
04-2
114.
[5]
Siritea
nu
C, Mi
ya
na
ga Y, B
l
os
tein S
D,et al.
MI
MO
Z
e
ro-F
orcing
Detecti
o
n
Anal
ys
is for C
o
rrelat
ed
an
d
Estimated Ric
i
an F
adi
ng.
IEEE Transactions
on Vehic
u
lar Technology
. 201
2; 61(7): 30
87-
309
9.
[6]
Chisa
g
u
a
n
o
D
J
R, O
k
ada M.
ESPAR antenna assisted MIM
O
-OFDM rece
iver using sub-
m
a
trix divided
MMSE sparse-
S
Q
RD detecti
o
n
. 201
2 Intern
ation
a
l S
y
mp
o
s
ium on C
o
m
m
unic
a
tions
a
nd Informati
on
T
e
chnolog
ies (
I
SCIT
).
Q
L
D. 2012: 19
8-2
03.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3761 – 37
68
3768
[7]
Xi
ao
jun Yu
an,
Li Pin
g
.
Space-
time l
i
ne
ar pre
c
odi
ng an
d iter
at
ive LMMSE d
e
tection for MIMO channe
ls
w
i
thout CSIT
. 20
11 IEEE I
n
ternati
o
n
a
l S
y
mp
osi
u
m o
n
Informatio
n
T
heor
y
Pr
oce
edi
ngs (ISIT
)
.
Petersbur
g. 20
11: 129
7-1
301.
[8]
Kuan
gh
ao Li
n, Chan
g RC, C
h
ie
nli
n
Hua
ng,
et al.
Imple
m
entatio
n of QR deco
m
p
o
sitio
n
for MIMO-
OFDM detection system
s
. 15th IEEE Internatio
nal
Confe
r
ence
o
n
Elect
r
onics, Circ
u
its
and S
y
stem
s
(ICECS). Julie
n. 2008: 5
7
-60.
[9]
Buzzi S, Saturnin
o D. SINR-maximizi
ng sp
read
i
ng co
de
alloc
a
tio
n
for non-li
ne
ar seria
l
interferenc
e
cance
llati
on.
IEEE Transactions on Communications
. 201
0; 58(2): 63
1-6
4
1
.
[10]
Lee
K, Joo
h
w
a
n Ch
un. S
y
mb
ol d
e
tectio
n i
n
V-
BLAST
architectures u
nder
chan
ne
l estim
a
tion
error
s
.
IEEE Transactions on Wireles
s
Communic
a
tions
. 200
7; 6(2)
: 593-59
7.
[11]
W
e
ixin Ga
o, N
an T
ang,
Xi
an
g
y
an
g Mu.
A
n
Algorit
h
m
for
U
n
it Co
mmitmen
t Based
on
H
o
pfiel
d
N
eura
l
Netw
ork
. F
ourth Internatio
na
l Confer
ence
on
Natural
C
o
mp
utation (ICN
C). Ji
nan. 20
08; 2
:
286-29
0.
[12]
Jian
g M, Akht
man J, H
anzo
L. Iterative J
o
int
C
h
a
nne
l Estimation an
d
Multi-
User Detectio
n
fo
r
Multipl
e
-Ante
n
na Aid
ed OF
D
M
S
y
stems.
IEEE Transactions on Wireless
Com
m
unications
. 200
7; 6(8):
290
4-29
14.
[13]
Rad
o
savl
jevic
P, Yuanb
in Gu
o, Caval
l
aro J.
Proba
bil
i
stical
l
y
b
oun
de
d soft sphere
detecti
on for MIMO-
OF
DM receivers: algor
ithm
and s
y
stem
architecture.
IEEE Journal on S
e
lect
ed Areas
in
Co
mmun
icati
o
ns
. 2009; 2
7
(8)
:
1318-1
3
3
0
.
[14]
T
a
spinar N, Ka
linl
i
A, Yild
ir
im
M. Partial T
r
ansmit Seque
nce
s
for PAPR Re
ductio
n
Usi
ng
Parall
el T
abu
Search Al
gorit
hm in OF
DM Systems.
IEEE Comm
unications Letters
. 201
1; 15(9): 97
4-9
76.
[15]
Hsue
hhsi
en
Cha
ng,
L
u
n
g
s
hu
Lin, N
a
nmin
g C
hen,
et al. Pa
rtic
l
e
-S
w
a
rm-Opti
m
izatio
n-Base
d
Noni
ntrusiv
e
D
e
man
d
Mon
i
to
ring a
nd
Lo
ad
Identificati
on
in Smart Mete
rs.
IEEE Transactions
on
Industry Appl
ic
ations
. 20
13; 4
9
(5): 222
9-2
2
3
6
.
[16]
F
e
i Li, W
e
i W
ang.
Qua
n
tu
m genetic
alg
o
ri
thm b
a
sed s
i
g
nal d
e
tectio
n sche
m
e for M
I
MO-OF
D
M
system
. Intern
ation
a
l Co
nfer
ence o
n
Com
m
unic
a
tion S
ystems, Net
w
or
ks and App
lica
t
ions (ICCSNA
).
Hon
g
Kon
g
. 20
10; 1: 298-
301.
Evaluation Warning : The document was created with Spire.PDF for Python.