Robustness of Fractional Order for FRFT-OFDM Systems

Pengfei Shi1, Dongjie Qi, Miao Li and Hao Huan

School of Information and Electronics
Beijing Institute of Technology, Beijing, 100081, China
Email: [email protected]

Abstract—Fractional Fourier transform (FRFT) has been This paper is organized as follows: the fractional Fourier
widely used in orthogonal frequency division (OFDM) systems transform and FRFT-OFDM systems are introduced in
to combat the doubly dispersive channels. The sensitivity of the section II. In section III, the analysis of the proposed
fractional order is a serious problem. In this paper, we derive diagonal principle for optimal order p is presented with the
the closed form of the variable order and its optimal value
based on the mathematical expression in the FRFT-OFDM order popt 0.5 . The simulation for the optimal order is
systems. The theoretical results prove that the fractional order given in section IV. Section V makes the conclusions.
p 0.5 is the global optimal fractional order for the FRFT-
OFDM system in doubly dispersive channels. Furthermore, II. DFRFT-OFDM SYSTEM
simulation results verify the robustness of fractional order 0.5 A. Fractional Fourier transform
in the FRFT-OFDM systems.
The fractional Fourier transform (FRFT) is the
Keywords-optimal fractional order, FRFT-OFDM systems, generalization of the conventional Fourier transform (FT).
doubly dispersive channel. D pS / 2 , p is the FRFT order) of the signal x(t ) ,
I. INTRODUCTION denoted as Fp (u ) [7]

Orthogonal frequency division multiplexing (OFDM) f

has been proved to be a promising method in wireless {Fp [ x  t ]} u  ³ x t  K t, u  dt

p (1)
communications with strong anti-fading capability. In slow f

fading channel, it will provide high-speed data transmission. and its inverse FRFT is
However, when channel impulse response changes distinctly
in an OFDM symbol period, or there is large Doppler shift, f

the orthogonality among different subcarriers will be x t  ³ F u  K t , u  dt

p
*
p (2)
destroyed in traditional OFDM systems [1]-[3]. To address f

this problem, Martone put forward using the chirp basis, where the transform kernel K p  t , u  of the FRFT is given
instead of sine basis, as the subcarrier signal to match with
the variable channels [4]. Then, Yang proved that the by
OFDM systems based on fractional Fourier transform
(FRFT) is robust against fast-fading channel in [5]. ­ 1  j cot D t2  u2
° exp( j cot D  jut csc D ), D z kS
Furthermore, the idea was enhanced in [6] with multiple ° 2S 2
° (3)
fractional orders at different Doppler shift. However, the K p t, u  ®G  t  u  , D 2kS
channel impulse response with fast fading changes over °
time irregularly, which results in sufficient prior information °G  t  u  , D (2k  1)S
°̄
to search the right fractional order to match with the
environment. So, for the fractional order p changes from 0 Without loss of generality, it is assumed that the angle D
to 1, it is difficult to obtain the right order in time. Thus, it is of FRFT is satisfied with 0  D  2S , corresponding to
vital to find the global optimal order popt in FRFT-OFDM 0  p  2 . Then the sampling discretization-type discrete
systems. fractional Fourier transform (DFRFT) is
In this paper, based on the mathematical expression for sin D  j cos D 2j cot D ˜m2 ˜'u2 N 1 2j cot D ˜n2 ˜'t 2 j 2S N˜n˜m
X p  m ˜e u ¦e ˜e ˜ x (n)
FRFT-OFDM in doubly-dispersive channel, diagonal N n 0
principle for the optimal order is proposed. Then, we obtain (4)
the optimal order popt 0.5 for the general channel
where D z kS and 't is the sampling interval of signal
environment through theoretical derivation. The simulation
results verify its robustness against fractional offset and x ( n) .
provide the suitable fractional order for the FRFT-OFDM
systems.
____________________________________
978-1-47-- /1/$31.00 ©201 IEEE

241
B. FRFT-OFDM systems ª 0 N q u( N  N q ) CNquN º
where TPcp « » and
q
In doubly-dispersive channel, the channel impulse
response changes distinctly in an OFDM symbol period ¬« I N u N ¼» Pu N
q

with large Doppler shift. The orthogonality in different j

 
2
cot v[( N  Nq  n )2   Nq  n ]'t 2
subcarriers is destroyed in traditional OFDM systems. To CN q u N q diag[e 2 ,n 0,1, , N q  1] is
solve this problem, Martone used the chirp basis, instead of a diagonal matrix with N q u N q .
sine basis, as the subcarrier signal in the OFDM system,
called FRFT-OFDM systems. The changes of fractional When the modulated signal is transmitted by the doubly
order p are used to reduce the ICI for Doppler shift in the dispersive channel, the received signal is
new systems.
rp RCP HsPcp  K p RCP HTPcp ˜ s p  K p (8)
The FRFT-OFDM system is shown in Fig. g 1 with the
transmitted symbol d [d (0), d (1), , d ( N  1)]T . The where K Pcp is the original white complex Gaussian noise
after modulation and H is the channel model that satisfies
with
Constellation Add Chirp
mapping
S/P IDFrFT Cyclic Prefix
P/S D/A
§ h( N q , 0) 0 0 0 ·
¨ ¸
¨ h (  N q ,1) h (  N q  1, 0) 0 0 ¸
Channel
¨ h( N q  2,1) ¸
¨ ¸
H ¨ h( N q  L, L  1) ¸
Constellation
demapping P/S DFrFT
Remove
Chirp CP
S/P A/D ¨ 0 h( N q  L, L  1) ¸
¨ ¸
¨ 0 0 ¸
¨ h(( N  2,
h 2 0)) 0 ¸
¨ ¸
¨ 0 0 h( N  1,1) h( N  1, 0) ¸¹ Pu P
©
Figure1. Construction of FRFT-OFDM systems. (9)
And then, the received signal after removing the CP is
modulation process can be expressed as
rp RCP HsPcp  K p RCP HTPcp ˜ s p  K p (10)
sp F p ˜ d p FpH ˜ d p (5)
where rp [rp (0), rp (1), , rp ( N  1)]T ,
where s p [s p (0), s p (1), , s p ( N  1)]T is the OFDM
RCP : [0N u Nq I N ]N uP and K p is a AWGN vector with
symbol after digital modulation and Fp denotes the kernel
of DFRFT, length of N.
At last, by using inverse FRFT to realize the FRFT-
Fp
OFDM systems, we obtain the mathematical expression for
§ j
 cot D ˜12 ˜'u 2
j
 cot D ˜( N 1)2 ˜'u 2 ·
the whole modulation and demodulation process as follows
¨ 1 e2 e 2
¸
¨  j cot D ˜12 ˜'t 2 2S ˜11˜
y p Fp ˜ rp  K p Fp ˜ RCP HTPcp ˜ FpH ˜ d p  K p H p ˜ d p  K p
¸
j j j 2 2 j 2 2S ˜1˜( N 1)
 cot D ˜12 ˜'t 2  cot D ˜12 ˜'u 2  j 2
 cot D ˜1 ˜'t  cot D ˜( N 1) ˜'u  j
¨e 2 e2 2 N
e2 2 N ¸ (11)
¨ ¸
¨ ¸ where H p is defined as the channel transform matrix in
¨  j cotD˜( N 1)2˜'t2  j cot D ˜( N 1)2 ˜'t2  j cot D ˜12 ˜'u2  j 2S ˜( N 1)˜1 j j 2S ˜( N 1)˜( N 1) ¸
¨e 2 e2 2 N
 cot D ˜( N 1)2 ˜'t 2  cot D ˜( N 1)2 ˜'u 2  j
e2 2 N ¸ fractional Fourier Domain, which is equal to
© ¹NuN
(6) Hp Fp H p F p in (12)

To reduce or eliminate inter symbol interference (ISI), Above all, the process of FRFT-OFDM system is
the fractional order cycle prefix (CP) is added to the FRFT- presented at great length.
OFDM systems. It is assumed that the CP length is N q and
III. DIAGONAL PRINCIPLE FOR THE OPTIMAL ORDER
the channel order is L. Then, the total length of signal with
CP is P N  Nq . The process of adding fractional order According to the detailed analysis of the channel
transform matrix H p in section 2, we can see that the
CP is given by
diagonal elements of H p represent the strength of useful
sPcp TPcp ˜ s p (7)

242
 Hp Fp H p F p
§ L 1 N 1 L 1 N 1 L 1 N 1
·
¨ ¦¦ h(n, l ) Fp (0, n) F p ( N  l  n, 0) ¦¦ h(n, l ) F (0, n) F p p ( N  l  n,1) ¦¦ h(n, l ) F (0, n) F p ( N  l  n, N 1) ¸
p
¨ l 0n 0 l 0 n 0 l 0 n 0 ¸ (12)
¨ L 1 N 1 L 1 N 1 L 1 N 1
¸
¨ ¦¦ h(n, l ) Fp (1, n) F p ( N  l  n, 0) ¦¦ h(n, l ) F (1, n) F p p ( N  l  n,1) ¦¦ h(n, l ) Fp (1, n) F p ( N  l  n, N  1) ¸
¨ l 0n 0 l 0 n 0 l 0 n 0 ¸
¨ ¸
¨ L 1 N 1 L 1 N 1 L 1 N 1
¸
¨ h(n, l ) Fp ( N  1, n) F p ( N  l  n, N  1) ¸¸
¨ ¦¦ h(n, l ) Fp ( N  1, n) F p ( N  l  n, 0) ¦¦ h(n, l ) F ( N  1, n) F p p ( N  l  n,1) ¦¦
©l 0n 0 l 0 n 0 l 0 n 0 ¹N uN

signal. So, we are trying to maximize the value and obtain

its absolute value. It is assumed that the SNR is large
enough, so that the elements in the diagonal are positive.
The summation of elements in diagonal shows the total
recovery’s power. For law of conservation of energy, the 400

elements except the diagonal are becoming lower when we

300
enlarge the diagonal elements. So, enhancing the elements
in the diagonal is to realize the process. 200

Pp
We use tr ( H p ) to represent the diagonal sum of matrix 100

H p , and then denote tr ( H p )D as the derived function of

0
tr ( H p ) with respect to D . According to the expression of 1
1
0.8
H p in equation (11), we have the diagonal sum f (D ) that 0.5
0.4
0.6
-10
x 10
satisfies, p 0 0
0.2
'p

f (D ) tr ( H p )
L 1 N 1 L 1 N 1

¦¦ h(n, l ) F (0, n) F p p ( N  l  n, 0)  ¦¦ h(n, l ) Fp (1, n) F p ( N  l  n,1)

l 0 n 0 l 0 n 0 Figure 2. Relationship among fractional order p , order derivation
L 1 N 1
  ¦¦ h(n, l ) Fp ( N  1, n) F p ( N  l  n, N  1) 'p and chirp rate P p .
l 0 n 0
L 1 N 1 N 1

¦¦¦ h(n, l ) F (k , n) F
l 0 n 0k 0
p p ( N  l  n, k )

(13) We get the general derivation of f (D ) , remarked as

f (D ) ,
'
And then, taking the elements of Fp in equation (3) into
the above equation (13), we obtain the expression of f (D ) ,
S2 2S l
S2 L 1 N 1
( N  l  2n)( N  L) j 2 sin(2D )˜( N  l  2 n )( N  l )  j
f ' (D ) cos(2D )¦¦ ˜ h(n, l )e 't N

't 0 sin(2S ( N  l ) / N )
2
l 0 n

L 1 N 1 N 1 (16)
f (D ) ¦¦¦ h(n, l ) F (k , n) F p p ( N  l  n, k )
l 0 n 0k 0
L 1 N 1 N 1 j j 2S ˜ kn j j 2S ˜ ( N  l  n ) ˜ k
For f (D ) is the complex data, the derivation of f (D )
 cot D ˜ k 2 ˜'t 2  cot D ˜ n2 ˜'u 2  j cot D ˜ k 2 ˜'t 2  cot D ˜( N  l  n )2 ˜'u 2  j
¦¦¦ h(n, l )e
l 0 n 0k 0
2 2 N
˜e2 2 N
is given by dividing the real part R( f ' (D )) and imaginary
L 1 N 1
j j
cot D ˜( N  l  2 n )( N  l )˜'u 2  j
2S l
I ( f ' (D )) of f ' (D ) , respectively. It is assumed that
¦¦ 2sin(2S ( N  l ) / N ) h(n, l )e 2 N

l 0 n 0
R( f ' (D )) 0 and I ( f ' (D )) 0 . Both of the stable extreme
(14)
point of real and image f (D ) are D S / 4 , which means
As is known, the chirp period in fractional Fourier the optimal order popt 0.5 , where the function f (D )
domain is 'u 2S sin D / 't , submitted into equation (14) satisfies,
and yielding to
2S l
L 1 N 1
j j
cot D ˜( N  l  2 n )( N  l )˜'u 2  j
f (D ) ¦¦ 2sin(2S ( N  l ) / N ) h(n, l )e 2 N
S L 1 N 1
j j
( N  l  2 n )( N  l )˜'u 2  j
2S l

l 0 n 0 f( )
4
¦¦ 2sin(2S ( N  l ) / N ) h(n, l )e 2 N

S2 2S l
l 0 n 0 (17)
L 1 N 1
j j 2 sin(2D )˜( N  l  2 n )( N  l )  j
¦¦
l 0 n 0 2sin(2S ( N  l ) / N )
h(n, l )e 't N L 1 N 1

¦¦
1 S 2S l
j( 
h(n, l )e 2 N
)

l 0 n 0 2sin(2S ( N  l ) / N )
(15)

243
 IV. PERFORMANCE ANALYSIS AND SIMULATION throughout. The simulation results prove its robustness
To facilitate the analysis, we take the IEEE 802.11a against fractional offset and provide the most suitable
protocol in this paper. It is assumed that the symbol duration fractional order for the FRFT-OFDM systems.
is T 3.2P s , the fractional order is p  [0,1] and the order ACKNOWLEDGMENT
derivation is 'p  [0,1] u1012 . The chirp rate P p and its This work was supported in part by the Ph.D. Programs
corresponding fractional order p and order derivation 'p Foundation of Ministry of Education of China (No.
is given in Fig. 2. 20121101130001), CAST key foundation (No.
CAST201215) and Open Research fund Program of Key
In Fig. 2, with the fractional order p 0.5 , the absolute Lab. for Spacecraft TT&C and Communication, funded by
value of chirp rate P p obtains the minimal value, which the Ministry of Education of China.
results in the not obvious phase rotation corresponding to
order offset 'p . However, when the fractional order p is
REFERENCES
apart from 0.5 gradually, the absolute value of chirp rate P p
[1] L. Tomba. “On the effect of wiener phase noise in OFDM systems,”
rises up rapidly, which results in obvious phase rotation. IEEE Trans. Commun., vol. 46, no. 5, pp. 580-583, May 1998.
Then we set the fractional order p 0.5 in Fig. 3. When [2] H. Zhang and Y. Li. “Anti-jamming property of clustered OFDM for
'p is in the magnitude of 106 instead of 1012 , the chirp dispersive channels,” IEEE MILCOM. Conf. Transformation: Power
to the Edge, vol. 1, pp. 336-340, 2003.
rate still keeps in an affording range. However, the large [3] D. W. Chi and P. Das. “Effect of jammer on the performance of
'p will result in high level the absolute value of chirp rate OFDM in the presence of nonlinearity in Rayleigh fading channel
with application to 802.11n WLAN,” IEEE MILCOM. Conf. Sensor,
Pp . Mesh and Ad Hoc Communications and Networks, Washington, USA,
pp. 1-7, 2006.
Above all, we can see that the demodulation for FRFT- [4] M. Martone. “A multicarrier system based on the fractional Fourier
OFDM systems is sensitive with the order offset 'p . The transform for time-frequency-selective channels,” IEEE Trans.
small offset will lead to high phase rotation. Fortunately, Commun., vol. 49, no. 6, pp.1011-1020, Jun. 2001.
when then fractional order is p 0.5 , the FRFT-OFDM [5] Q. Yang, R. Tao and J. Ni. “Anti-jamming performance analysis
performance analysis of FRFT-OFDM,” 2010 1st Int. Conf. Pervasive
systems will be robust against order offset and phase Computing, Signal Processing and Applications, Harbin, China, 2010.
rotation. [6] M. H. Azmy, S. Elgamel and A. M. Aziz. “Fractional Fourier transf
orm OFDM system performance enhancement over doubly dispersive
V. CONCLUSION channel,” 2012 Int. Conf. Engineering and Technology, Cairo, Engpt,
2012.
In this paper, we drive the optimal order p 0.5 for the
[7] L. B. Almeida. “The fractional Fourier transform and time-frequency
general channel environment in the FRFT-OFDM systems. representations,” IEEE Trans. Signal Processing, vol. 49, no. 11,
The results show a fact that there exists a most suitable pp.3084-3091, Nov. 1994.
fractional order instead of the traditional FFT in the
fractional Fourier domain. Meanwhile, by applying the
order p 0.5 to test the new channel environment, we do
not need to use multi orders at all, which saves the data

500

450

400

350

300

250
Pp

200

150

100

50

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
'p -5
x 10

Figure 3. Relationship between order offset 'p and chirp rate P p

with fractional order p 0.5 .
244