Mathematical Analysis of Signal Propagation in

Ultra-Wideband Transceiver System with Frequency

Offset Correction

Debarati Sen, Saswat Chakrabarti
G. S. S. School of Telecommunications
Indian Institute of Technology Kharagpur
Kharagpur-721302, India
debarati@gssst.iitkgp.ernet.in, saswat@ece.iitkgp.ernet.in
R. V. Raja Kumar
Dept. of Electronics & Electrical Communication Engg.
Indian Institute of Technology Kharagpur
Kharagpur-721302, India
rkumar@ece.iitkgp.ernet.in


AbstractThis work explores theoretically the unique aspects of
different signal processing stages in a Multi-Band (MB)
Orthogonal Frequency Division Multiplexing (OFDM) based
system in realistic Ultra-Wideband (UWB) channel. In the
process of analysis, it emphasizes the significant aspects of signal
processing steps in UWB receiver design comparing a narrow-
band system. The analysis preliminary considers perfect timing
and frequency synchronization which is extended later with a
frequency offset error and its correction. Performance of the
offset estimator in MB-OFDM system via computer simulation is
provided in order to support the mathematical analysis.
Keywords- MB-OFDM, Signal analysis, Transceiver, UWB

I. INTRODUCTION
In commercial high-data rate Wireless Personal Area
Networks (WPAN), the Multi-Band Orthogonal Frequency
Division Multiplexing (MB-OFDM) based transmission
scheme for Ultra-Wideband (UWB) has proved to be more
attractive than code-division multiple access-based and
impulse-radio-based communication.
Different signal processing stages in a UWB transceiver
bear significant differences compared to a narrow-band system.
It is interesting to explore theoretically the crucial aspects of
different signal processing blocks in a UWB system. Detailed
mathematical studies on UWB transceiver can light upon in
understanding the important aspects of an UWB receiver
design which is the main aim of this work. To the best of our
knowledge no such analysis is presented in literature so far for
MB-OFDM UWB system.
In this paper, we start with analyzing a MB-OFDM UWB
transmitter-receiver under realistic UWB channel environment
considering perfect timing and frequency offset estimation.
Then, the study is extended further to evaluate the performance
of the system incorporating frequency offset errors through
mathematical analysis supported by computer simulation.
The rest of this paper is organized as follows: Section II
deals with the signal analysis in MB-OFDM transceiver. The
ultra-wideband channel is modeled in Section III. Signal
analysis in a MB-OFDM receiver is presented in Section IV.
Performance of the system with frequency offset error
correction with Cramer Rao Lower Bound (CRLB) of
estimated error is analyzed in Section V. Section VI figures out
simulation results with some relevant discussion. Important
concluding remarks with the summary of the paper are
included in Section VII.
II. SIGNAL ANALYSIS IN A MB-OFDM TRANSMITTER
A. MB-OFDM System Description
The MB-OFDM system splits the 7.5GHz unlicensed
spectrum into fourteen (14) bands [1]. OFDM symbols are
transmitted by frequency hopping over the bands. UWB
transmitter and receiver block diagrams considered for this
study are shown in Fig. 1 and Fig. 2.
Let {X(k)} be the QPSK modulated complex sequence of
symbols to be transmitted using OFDM (= T
6
signal point in
transmitter in Fig. 1). A complex base band OFDM symbol can
be obtained using N point Inverse Fast Fourier Transformation
(IFFT) represented as,
1
2
0
1
( ) ( )
N
j fkt
k
x t X k e
N

=
=

for 0t<T, = 0 elsewhere (1)
where, f = sub carrier spacing. Taking samples at t = nT
S,
the n-
th sample of the m-th OFDM symbol can be shown as,
( )
( )
1
2
0
1
( , ) ,
k
N
j n mN
N
k
x n m X k m e
N

=
=
; 0mZ-1 (2)
where, N = no. of samples over duration T and Z is the no. of
OFDM symbols transmitted.
B. Cyclic Prefix Insertion
In MB-OFDM transmission, Zero-Padded bits (ZP) are
added in place of cyclic prefix to compensate the multipath
effect. Let, N
Z
number (no.) of ZP bits and N
g
no. of guard bits
(for inter band synchronization in MB-OFDM) are added with
N points IFFT output to construct one MB-OFDM symbol.
The total no. of samples N
s
in one OFDM symbol is given as,
N
s
= N+ N
Z
+ N
g
.
978-1-4244-2644-7/08/$25.00 2008 IEEE
















Figure 1. Multiband OFDM UWB Transmitter

Hence, from (1) we can derive,
x (n) = 0, for 0 n N
z
-1
= x(n-N
z
), for N
z
n N+N
z
-1
= 0, for N+N
z
n N+N
z
+N
g
-1 (3)
Similarly from (2) (at T
8
in Fig. 1),
( )
( )
1
2
0
0 1
1
( , ) , ;
0
z s
k
N
j n N mN
N
s
k
m Z
x n m X k m e
n N N

=

=

(4)
C. Signal Passing through DAC
Digital to Analog Conversion (DAC) block involves first
the conversion of the samples ( , ) x n m into a sequence of
impulses given as,
( ) ( ) ( , )
b s s
n
n m t nT x t x

=
=
(5)
Then it is filtered by a reconstruction filter, which is an ideal
Low Pass Filter (LPF) having impulse response of,
h
r
(t) = sin (t/T
s
)/ t/T
s
(6)
Hence, the signal at the output of DAC (at T
9
in Fig.1) is,
( )
( )
( )
9
sin /
/
( , )
s s
s s n
t nT T
T
t nT T
x t x n m

= =

(7)
The above conversion is performed when sampling of the
signal follows Nyquist sampling theorem.
D. Signal Upconversion
Now, the analog real valued signal ( ) x t is multiplied by
time frequency kernel of exp(j2f
c
t) to convert it to RF signal
from base band. Therefore, the transmitted signal (at T
10
in
Fig.1) may be given as,
( ) ( ) ( ) ( ) ( )
10
cos 2 sin 2
tx c c
T x t x t f t jx t f t = = +
(8)
III. THE ULTRA-WIDEBAND CHANNEL MODEL
The proposed multipath channel model for UWB [2] by
IEEE 802.15.3 channel modeling sub-committee is a modified
form of indoor channel model proposed by A. Saleh and R.
Valenzuela (S-V) [3] that models multipath arrivals in the form
of clusters and rays within clusters. S-V model distinguishes
between cluster arrival rate and ray arrival rate.























Figure 2. Multiband OFDM UWB Receiver

Mathematically, impulse response of channel is,

1 1
, ,
0 0
( ) ( )
L I
i l l i l
i l
h t t T

= =
=

(9)
where,
i,l
= channel coefficient for the i-th ray of the l-th
cluster which is log-normally distributed; T
l
=delay of the l-th
cluster;
i,l
= delay of the i-th ray related to the l-th cluster
arrival time;
0,l
= 0, by definition. The large bandwidth of
UWB significantly increases the resolving capability of UWB
receiver. Here, the no. of reflections from channel arriving
within short period (0.167nsec. for 6GHz B.W. and 0.133nsec.
for 7.5GHz B.W.) of impulse is too small. As a result, the
central limit theorem is not valid further in UWB systems.
However, we shall continue our analysis considering the
S-V channel model given by (9) in the next Section.
IV. SIGNAL ANALYSIS IN MB-OFDM RECEIVER
The transmitted signal (8) after passing through UWB
channel and added with w(t) Additive, white, Gaussian noise,
received at the receiver front end given by (at R
1
in Fig. 2),
( ) ( )
1
1 ,
1
,
0 0
l i l
L I
tx i l
l i
R t x t T w t

= =
= +

(10)
A. Preselect Filter
In preselect filter, the out of band signals are chopped off
and signals of desired bands are selected. Functionally, this is
a wideband Band Pass Filter (BPF). In practice, transfer
function of narrow-band band pass filter is given by,

( )

( ) ( ) Re exp 2
c
h t h t j f t

=

(11)
where,

( ) h t is complex impulse response of band pass system

given by,

( ) ( ) ( )
I Q
h t h t jh t = + , wherein, h
I
, h
Q
are equivalent
LPF coefficients.
Scra-
mbler
T
2

T
1

Convolutio-
nal Encoder
T
3

Punctu-
rer
T
4
Bit
interlea-
ver
T
5

QPSK
modulator
IFFT
T
6
T
7

Null
Suffix
insertion
T
8

DAC
T
9

( ) exp 2
c
j f t
Time
frequency
Kernel
T
10

Input
data
Insert
pilots add
prefix GI
Pre
select
filter
LNA
LPF ADC
LPF ADC
Synchr-
oniza
-tion;

Suffix
remo-
val &
CP
additio
FFT
QPSK
Demodu-
lation
CE
AGC ( ) cos 2
c
f t
FEQ
Remove
Pilots
Carrier Phase,
Time
Tracking
De-
Interle-
aver
Viterbi
decoder
De-
Scramb-
ler Received
data
R
5

R
1
R
2
R
3
R
4
R
6

R
7

R
8
R
9
R
10
R
11
R
12
R
13
R
14
R
15
R
16
R
17
I
Q
( ) sin 2
c
f t
The above transfer function of narrow-band BPF is not valid
here for UWB. The receiver front end signal passes through
BPF with center frequency f
c
and bandwidth B, which is wider
then or equal to the 10dB B.W. of the transmitted signal. For
convenience of analysis, we assume that the BPF passes the
received desired signal part perfectly without distortion and
the filter bandwidth B equals an integer multiple of 1/2T
b
(T
b
=
bit duration) such that negligible inter symbol interference and
intra symbol interference results. The filtered signal waveform
can be expressed (at R
2
in Fig. 2) as,
( )

( )
2
1 1
0 0
, ,
L I
l i
i l l i l tx
R t x t T w t

= =

= +

(12)
where

( ) w t is filtered noise given by [4],

( ) ( ) ( ) ( ) ( ) 2 cos 2 sin 2
c c s c
w t w t f t w t f t =

(13)
wherein, w
c
(t) and w
s
(t) are uncorrelated Gaussian random
process with the autocorrelation function at BPF output,
( )

( )

( )
{ }
( )
( )
0
sin 2
cos 2
2
C r
B
R E w t w t N B f
B



= + = (14)
Power spectral density of each noise component w
c
(t) and w
s
(t)
at the BPF output is given by,
( ) ( )
0
0
C S
W W
N f B
S t S t
otherwise
<

= =

(15)
B. Signal Passed Through LNA and Downconverted
The filtered signal amplified by wideband low noise
amplifier (LNA) with a 3dB flat gain of (at R
3
in Fig. 2) is,
( ) ( )

( )
, ,
,
1 1
3
0 0
2
i l l i l
l i l
L I
l i
c
j f t T
R t x t T e w t






= =

= +

(16)

The amplified signal can also be written as,
( ) ( ) ( ) { }

( )
1 1
3 , ,
0 0
,
cos sin
L I
l i l c c i l
l i
i l
j
R t x t T t jx t t e w t

= =

= + +

(17)
where,
i,l

, c i l
= and
, i l
= delay of i-th ray of l-th cluster
measured with respect to the 1st

ray of 1st cluster.
Here,
i,l
are essentially independent and uniformly
distributed over (0, 2). In MB-OFDM, the OFDM symbol
period is =1/f =242.42nsec. and sample period is 1.89nsec.
The maximum delay spread obtained from measurement and
supported by simulation for the worst channel here is 25nsec.
(CM4). It is to be noted that the phase component
c
=
can not be neglected here like narrow-band system as the
maximum delay in S-V channel model (25nsec.) is
comparable to 2/
c
. For example, 2/
c
= 0.3nsec. (for band
1, with f
c
=3432MHZ) and is 0.09nsec. (for band 3, with f
c

=3960MHZ). It is obvious from (17) that the phases
i,l
are
independent (each is a function of its respective path).
Assuming that the receiver generates the coherent reference
(perfect carrier phase estimate) we get in-phase and q-phase
components that can be expressed as (at R
4
and R
5
in Fig. 2),
( )
( ) ( ) { }

( )
2 2
1 1
4 , , ,
0 0
,
,
(18)
cos sin
cos
c
l
L I
c i l l i l l i l
i l
i l
i l
j
c
R t x t T t jx t T t e
w t T t

= =

= +
+

( ) ( ) ( ) { }

( )
2 2
5
1 1
, , ,
0 0
,
,
(19)
cos sin
sin
l
L I
c c i l l i l i l
l i
c l i l
i l
j
R t x t T t jx t T t e
w t T t

= =

= +
+

C. I-Q Signals at the Low-Pass Filter (LPF) Outputs
Both the above down-converted signals after passing
through LPFs (which are basically Nyquist filters), give the
base band signals expressed as (at R
6
and R
7
in Fig. 2),
( ) ( ) ( )
6 1
1 1
, ,
0 0
,

l
L I
i l i l
l i
i l
j
R t x t T e w t

= =

= +
(20)

( ) ( ) ( )
7 2
1 1
, ,
0 0
,

l
L I
i l i l
l i
i l
j
R t x t T e w t

= =

= +
(21)
where,
( )

( ) 1
,
cos
l c
i l
w t w t T t =
and ( )

( )
2 ,
sin
c l i l
w t w t T t =

are Nyquist filtered noise.
D. Samples at A/D Converter Output with Frequency Offset
Both the above signals are converted in digital domain
taking samples at t = nT
s
given as (at R
8
and R
9
in Fig. 2),
( )
,
1 1
8 , 1
0 0
( ) ( )
i l
L I
j
i l z
l i
R n x n l i N e w n

= =
= +

(22)

( )
,
1 1
9 , 2
0 0
( ) ( )
i l
L I
j
i l z
l i
R n x n l i N e w n

= =
= +

(23)
E. ZP Removal and FFT Implementation
In the next step, the ZP is removed by overlap and add
method assuming that the transmitted and received signals are
perfectly matched. So, from (22) after ZP removal we get,
( ) ( )
( )
( )
10 1
1 1 1
,
0 0 0
,
2 1

L I N
z i l
l i k
i l
j j k N n l i
R n X k w
N
e e n N



= = =

= + +

(24)
Next, the N-point FFT is performed to recollect the frequency
domain transmitted samples. Using
, , i l i l
= ,
( ) [ ]

[ ]
11
,
1 1 1
1
,
0 0 0
2 2 2
.
i l
L I N
i l z
l i n
j lq j iq j nq
j
N N N
f
Signal
noise Ch Coef
R q x q e e e w n N e

= = =

= + +

. .
(25)
where,
1
2 ( ) /
0
( ) ( )
N
j n l i q N
n
x q x n l i e

=
=
(26)
It is seen from (25) that frequency domain recovered samples
are associated with phase errors due to the multipath delay
associated with corresponding paths of ray as well as clusters.
F. Channel Estimation and Equalization
UWB channel is estimated from the channel estimate
symbols of the preamble [1]. Assuming that the channel is
slow fading, its coefficients are considered to be constant
during one OFDM symbol. The channel frequency response
estimated over one OFDM symbol by least square estimator is,
( )
( )
( )
1
,
2
1
2 2
1 1
0
,
0 0

i l
j nq
N
N
j lq j iq
L I
n N N
i l
l i
j
w n e
H q e e e
x q

=
= =

= +

(27)

Final estimated channel frequency response can be obtained
by taking the average of estimated channel coefficients over
channel estimation symbols available per band (C
b
).

( ) ( )
0
1

( )
b
final
C
r
r
b
H q H q
C
=
=

(28)

In the above explanation it is clearly vivid that the error in
channel estimates i.e. the distortion coming due the addition of
noise in path (second part of (27)) and the comparable
multipath delay with respect to 2/
c
reflected as
2 ( ) / j l i q N
e
+
in the first term of (27). After equalization, the
estimated received samples are given by,

( )
( ) ( )

( ) ( )
1
1 1 1
,
0 0 0
,
2
2 2 /

L I N
i l
l i n
final
i l
j nq
j
j lq N j iq N
N
est
x q e e e w n e
q
H q
x

= = =


+
=



where, q = 0, 1, 2, . 128. (29)
G. QPSK Demodulation and Detection
At the output of the demodulator the transmitted symbol is
recovered back taking decision over the signed amplitude of
the x
est
(q). The QPSK demodulator input symbol x
est
(q) is,
( ) q
est
est inphase est qphase
x x x



=
(30)
V. SIGNAL ANALYSIS IN RECEIVER WITH CARRIER
FREQUENCY OFFSET CORRECTION AND CRLB CALCULATION
Let the normalized carrier frequency offset (CFO) is
expressed in terms of sub carrier spacing f as / f f f = , where,
f is the CFO. After ZP removal, signal obtained with CFO is,
( )
( )
( )
( )
,
1 1 1
, 1
0 0 0
10
2
2
( )
1
(31)
i l
L I N
i l z
l i k
z
j j k n l i N
j f n N N
n
X k e e w n N
N
R e

= = =

+
+ +
=


After FFT, the signal is presented as,
( ) ( ) ( )
( )( )
( )
,
1 1
11 ,
0 0
1
1
0
2
2
2
(32)
z
i l
L I
i l
l i
N
z
n
j f N N
j j l i q f N
j nq N
R q e x q f N
e e w n N e

= =

+

=
+ +

It is observable from (32) that there will be gradual
performance decrease for frequency offset f<f. But if f >f,
there will be a complete detection error. In (32), f causes a
phase rotation of 2 ( ) /
z
f n N N + . If remains uncorrected, it
causes both rotation and spread of QPSK constellation points.
Frequency offset synchronization may be performed in
receiver in time domain either by cyclic extension or by using
special training symbols. Synchronization technique based on
cyclic extension is not suitable for high rate packet
transmission (like MB-OFDM) because : (i) accurate
synchronization needs averaging over large (>10) nos. of
OFDM symbols to attain distinct no. of peaks and a reasonable
signal-to-noise ratio (SNR) (ii) synchronization time needs to
be as short as possible. In Training symbol based
synchronization technique, the incoming signal is correlated
with the complex conjugate of the known training signal.
From the correlation peaks in the matched filter output signal,
symbol frequency offset can be estimated. We continue here
with our frequency offset estimation scheme [5]. Considering,
( )
( )
1 1 1
,
0 0 0
2 1
( )
L I N
r i l
l i k
j k n l i N
x n X k e
N

= = =

=
(33)
The received signal R
10
(n) for the n-th sample of the m-th

OFDM symbol (0 m Z-1) can be expressed as,
( )
( )
1 1
,
0 0
10
1
2
( , ) ( , )

L I
c i l
l i
z s
z s
r
j
j f n N mN N
R n m x n m e e
w n N mN

= =

+ +
=
+ + +


( ) { }
( )
1
2
( , ) ,
z s
r
j f n N mN N f
x n m e w n m
+ +
= + (34)
where, ( )
1
, w n m is the AWGN added with the n-th sample of
m-th OFDM symbol and
1 1
,
0 0
L I
i l
l i


= =
=
.
The phase difference between n-th sub-carriers of 2
consecutive OFDM symbols m & (m+1) is 2{(fN
S
)/N - f}.
Phase offset between 2 consecutive OFDM symbols is [5],
( ) ( ) ( ) ( )
( ) ( ) { }
( )
2
(35)
, , , 1
,
z s s z s
r r
j f n N mN N N f f n N mN N f
Z n m x n m x n m
e w n m

+ + + + + +
== +
+
where,
( ) ( )
( ) { }
( ) ( ) ( )
( ) ( )
( ) ( )
( ) ( ) ( )
1 1
1 1
2
2 1

(36)
, , , 1 ,
, 1 , , 1
z s
z s
r
r
j f n N mN N f
j f n N m N N f
w
w
w n m x n m e n m w n m
x n m e w n m n m

+ +
+ + +
= + +
+ + +

Estimated frequency offset from P pairs of L successive
OFDM symbols [5],
`
( )
1 1
1 0
1 1
,
1

( )
2
L P
noise
m n
s
n m
L p
f L f
N N

= =

= +

(37)
The n-th noise sample of m-th OFDM symbol is Gaussian
distributed with zero mean with the variance
2
/3.
Correspondingly, the variance calculated,
( )

Var f L



=
2
N /{
( )
2
12 1
s
P L N } (38)
A. CRLB for Estimated Frequency Offset
The expression for complex conjugate multiplication of n-th
pair of m-th and (m+1)-th OFDM symbols from (35) is as,
( ) ( ) ( )
( ) ( ) ( )
2
, cos 2 sin 2
; where, , , ,
s s
r
in qp
Z n m A f N N jA f N N
A x n m w n m j w n m
= +
= + +
(39)
Now, let us take,

( ) ( ) ( )
( ) ( ) ( )
, cos 2 ,
, sin 2 ,
in
s
qp
s
U n m A f N N w n m
V n m A f N N w n m

= +

= +

(40)
And
( ) ( )
( ) ( )
, cos 2
, sin 2
s
s
u n m A f N N
v n m A f N N

(41)
We have to estimate the unknown parameter vector,
[ ]
T
A f = (42)
The joint probability density function of the sample vector
Z(n, m) (when unknown parameter vector is ) is given by,
( ) ( ) ( ) ( ) { }
1
2
2
2
0
1 1
, ; exp , ,
2
2
p
p
n
f Z n m U n m u n m

=

=





( ) ( ) { }
1
2
2
2
0
1 1
exp , ,
2
2
p
p
n
V n m v n m

=




(43)

Inverse of Fisher Information Matrix [6] gives the variance of
unknown parameter estimates. Unbiased Cramer Rao Lower
Bound (CRLB) of estimations obtained from diagonal
elements of Inverse Fisher Information Matrix I() given as,
( )
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
2 2
2
2 2
2
ln , ; ln , ;
ln , ; ln , ;
f Z n m f Z n m
E E
A f A
I
f Z n m f Z n m
E E
f A f

=









(44)
We know,
2 2
SNR A = , when frequency offset estimation
is carried over L OFDM symbols, the variance is,
( )

Var f L



= N /{ ( ) ( )
2
2 1
s
N L SNR } (45)

VI. SIMULATION RESULTS AND DISCUSSIONS
Simulation is carried out to study the performance of MB-
OFDM systems with frequency offset correction using Time
Frequency Interleaved (TFI) pattern 1 [1] under UWB channel
model CM4 and AWGN. Relevant parameters from ECMA-
368 standard for MB-OFDM are considered for the simulation
study. We have considered 1000 noisy realizations in each of
the 100 UWB channels for our simulation. CFO of the
frequency offset estimation scheme [5] is estimated for an
AWGN channel using several consecutive OFDM symbols (L
= 2, 3 4, 5, and 6) in the preamble for MB-OFDM systems.
Fig. 3 shows change of variance of the frequency offset
estimation error of our scheme vs. SNR for various values of L,
viz. 2, and 6. As expected, the variance of frequency offset
estimation error decreases with increase in L. However, a
lower value of L is desirable in practice to reduce the system
complexity. It is also observed that the error variance as per
our scheme is very close to the respective CRLB.
VII. CONCLUSION
This paper has analyzed the propagation of signal in a MB-
OFDM based UWB transceiver system under realistic channel
environment. The significant differences of UWB signal
processing compared to narrow-band system design are also
identified in the process of analysis. This work also
investigates the effects of synchronization imperfections due
to carrier frequency offset in the receiver and justifies the need
of its correction. The frequency offset is corrected by our
earlier proposed estimation scheme and CRLB for the
estimated offset is obtained.
It is observed from the mathematical analysis supported by
the computer simulations that variance of estimated frequency
offset of our scheme decreases with increased no. of OFDM
symbols used for estimation.
This work can further be extended to the system analysis
considering timing imperfection scenario.
0 5 10 15 20 25 30 35 40
10
-9
10
-8
10
-7
10
-6
10
-5
10
-4
10
-3
SNR in dB
v
a
r
i
a
n
c
e

o
f

f
r
e
q
u
e
n
c
y

o
f
f
s
e
t

e
s
t
i
m
a
t
i
o
n

e
r
r
o
r
CRLB L=6
simulation L=6
CRLB L=2
simulation L=2
a
b
c
d

Figure 3. Variance of frequency offset estimation error: a) CRLB
calculated with L=6; b) simulation with L=6; c) CRLB calculated with
L=2; d) simulation with L=2 for MB-OFDM system in AWGN channel.

REFERENCES

[1] ECMA-368, High Rate Ultra Wideband PHY and MAC Standard, 2nd
Edition, December 2007,
www.ecma-international.org/publications/standards/Ecma-368.htm.
[2] J. Foerster, Channel modeling sub-committee report final, IEEE,
Document IEEE P802.15-02/490r1-SG3a, Feb. 2003.
[3] A. A. M. Saleh and R.A. Valenzuela, A statistical model for indoor
multipath propagation, IEEE J. Selected Area Commn., vol. 5, pp 128-
137, Feb. 1987.
[4] M. K. Simon, S. M. Hinedi, N. C. Lindsey, Digital Communication
Techniques Signal Design and Detection, Prentice Hall PTR, 2003.
[5] D. Sen, S. Chakrabarti and R. V. Raja Kumar, An efficient frequency
offset estimation scheme for multi-band OFDM ultra-wideband
systems, IEEE VTC08-spr., Singapore. pp. 973-977, May 11-14, 2008.
[6] S. M. Kay, Fundamentals of statistical signal processing: estimation
theory, Prentice Hall PTR, 1993.