International Journal of Distributed Sensor Networks 3
where Γ = diag{1,
𝑗2𝜋]/𝑁
,...,
𝑗2𝜋](𝑁−1)/𝑁
}.eentriesof
Fourier matrix F
𝑁
are [F
𝑁
]
𝑛,𝑘
=1/
−𝑗2𝜋𝑛𝑘/𝑁
,where0≤
, ≤−1. AWGN noise vector at the th Rx branch
is k
𝑗
=[V
𝑗
(0),V
𝑗
(1),...,V
𝑗
(−1)]
𝑇
.enoisesamples
are independent circularly symmetric zero-mean complex
Gaussian random variables with variance 1/2
2
V
per dimen-
sion.
Due to the presence of CP, channel linear convolution is
converted into circular convolution. e noise-free received
signal vector is x
𝑗
=
∑
𝑁
𝑇
𝑖=1
S
𝑖
h
𝑖,𝑗
,whereS
𝑖
is transmit signal
matrix at the th antenna, S
𝑖
=F
𝐻
𝑁
diag{c
𝑖
}F
𝑁,𝐿
.ereceived
signal power is dened as
2
𝑥
=1/∑
𝑁−1
𝑛=0
|
𝑗
()|
2
.
We denote S =[S
1
,S
2
,...,S
𝑁
𝑡
], h =[h
𝑇
1
,h
𝑇
2
,...,h
𝑇
𝑚
,...,
h
𝑇
𝑁
𝑟
]
𝑇
, h
𝑚
=[h
𝑇
1,𝑚
,h
𝑇
2,𝑚
,...,h
𝑇
𝑁
𝑡
,𝑚
]
𝑇
,andk =[k
𝑇
1
,k
𝑇
2
,...,
k
𝑇
𝑁
𝑟
]
𝑇
. Aer stacking all the received vectors from dierent
Rx branches r
𝑗
, 1≤≤
𝑟
, the new received vector, r =
[r
𝑇
1
,r
𝑇
2
,...,r
𝑇
𝑁
𝑟
]
𝑇
,isgivenby
r =I
𝑁
𝑟
⊗ΓSh +k =Ah +k, (2)
where A =I
𝑁
𝑟
⊗ΓS.emeanofk is zero and the covariance
matrix of k is
2
V
I
𝑁𝑁
𝑟
.
3. ML Carrier Frequency Offset
Estimation Algorithms
For known A and h,thereceivedvectorr is Gaussian with
mean Ah and covariance matrix
2
𝑤
I
𝑁𝑁
𝑟
. e likelihood
function of r is a
𝑟
dimension complex Gaussian function:
Λ
(
r;h,]
)
=
1
2
V
𝑁𝑁
𝑟
exp −
r −Ah
2
2
V
.
(3)
ML algorithm searches for the optimum value of h and
], which can maximize the likelihood function in (3). e
maximization is equivalent to the minimization of the cost
function below
(
r;h,]
)
=
r −Ah
2
=
𝑁
𝑟
𝑗=1
r
𝑗
−ΓSh
𝑗
2
.
(4)
Assuming ] isxed,wetaketherstorderpartialderivativeof
(r;h,])with respect to h andsetittozero.eleastsquare
(LS) estimation of CIR becomes [18]
h
𝑗
=(ΓS)
+
r
𝑗
,1≤≤
𝑟
,
(5)
where X
+
is the Moore-Penrose generalized inverse of X.
Substituting (5)into(4) and discarding the irrelevant terms,
(4)becomes
(
r;]
)
=
𝑁
𝑟
𝑗=1
r
𝐻
𝑗
(
ΓS
)(
ΓS
)
+
r
𝑗
=
𝑁
𝑟
𝑗=1
r
𝐻
𝑗
ΓBΓ
𝐻
r
𝑗
,
(6)
where B = SS
+
is called project matrix. Notice that the pro-
jection matrix B is Hermitian symmetrical and (r;])can be
further derived as
(
r;]
)
=
(
0
)
−2R
𝑁−1
𝑙=0
(
)
−𝑗2𝜋]𝑙/𝑁
,
(7)
where ()is given by
(
)
=
𝑁
𝑟
𝑗=1
𝑁−1
𝑘=𝑙
B
𝑘−𝑙,𝑘
𝑗
(
)
∗
𝑗
(
−
)
.
(8)
Compared with SISO systems, there are diversity eect in
(). e ML CFO estimator for nonperiodic preambles is
] =arg max
]
R
𝑁−1
𝑙=0
(
)
−𝑗2𝜋]𝑙/𝑁
.
(9)
e last term of right hand of (7)hasthesamestructure
as Discrete Fourier Transform (DFT). So this term can be
implemented by FFT eectively.
3.1. MLE-MA. As we know, for system with periodic pream-
bles, its packet detectors and CFO estimator have simpler
form and lower complexity [9, 10]. A periodic preamble has
multiple identical short slots and each slot has a length of .
For simplicity, the length of the periodic sequence is rst set to
, the same as the FFT size. e periodic sequence includes
=/identical short slots. Actually, more repeated slots
maybeusedinpractice,suchasin802.11a/n[8], the short
preamble contains ten identical short slots, and each slot has
a length of 16. To generate periodic preambles, nonzero pilots
are inserted with a xed spacing ;namely,theentriesofc
𝑖
are
c
𝑖
(
)
=
𝑖
, =
,
0, otherwise,
(10)
where pilot
𝑖
(
) is usually constant power modulation
symbols with |
𝑖
(
)| =
1/(
𝐷
𝑡
),and
∈{:
[1,
𝐷
/2]∪[−
𝐷
/2,−1]},wherethenumberofpilots
𝐷
is an even number and
𝐷
≤
𝑢
/. In 802.11n, the
frequencydomainpilotsequencesusedineachTxantenna
are the same but with dierent cyclic delay. Now the pilot
sequences c
𝑖
is represented as c
𝑖
= c
1
u,whereu =
[1,
𝑗2𝜋𝜏
𝑖
/𝑁
,...,
𝑗2𝜋(𝑁−1)𝜏
𝑖
/𝑁
]
𝑇
.etimedelayoftheth Tx
branch is
𝑖
.Wehave
1
=0.Inthecaseofmultiple(>)
repeated slots, assuming that timing uncertainty existed,
but there are enough samples to assure the receiving vector
not containing other parts of the signal, thus the received
signal vector at the th Rx branch is given by
r
𝑗
=Γ
𝑁
𝑡
𝑖=1
F
𝐻
𝑁
Γ
𝑖
+diag c
1
F
𝑁,𝐿
h
𝑖,𝑗
+k
𝑗
,
(11)
where Γ(
𝑖
+)=diag{1,
𝑗2𝜋(𝜏
𝑖
+𝜇)/𝑁
,...,
𝑗2𝜋(𝜏
𝑖
+𝜇)(𝑁−1)/𝑁
}.
Now, the transmit signal matrix at the th antenna becomes
S
𝑖
=F
𝐻
𝑁
Γ(
𝑖
+)diag{c
1
}F
𝑁,𝐿
.