You are on page 1of 4

A Novel Approach of implementing an FFT

Algorithm on OFDM system

Nafiz Imtiaz, Anwarul Azim, Rishad Ahmed Shafik, Muhammad


Department of Electrical and Electronic Engineering
Islamic University of Technology (IUT)
Gazipur-1704 , Bangladesh
nafiziut@gmail.com, azim1866@gmail.com, rishad09@gmail.com, muh14_bd@yahoo.com

Abstract—In this paper, we make insightful assessment of the unique frequency and combined together using the Inverse
computational performance of orthogonal frequency division Fast Fourier Transform (IFFT) to yield the time domain
multiplexing (OFDM) system in terms of complex calculations
waveform to be transmitted [3]. After IFFT is done, the time
required using different Fourier transform techniques. We
briefly introduce the different transform technique, viz. discrete domain signals are then converted to serial data and cyclic
Fourier transform (DFT) and various types of fast Fourier extension is added to the signal. Then the signal is transmitted
transform (FFT) as 2-radix FFT, 4-radix FFT etc. We have also .At the receiving side we do the reverse process to get original
discussed about a new FFT approach that can be applied in an data from the received one. In case of deep fade several
OFDM system and shown comparison of it with other FFT symbols in single carrier is damaged seriously, but in parallel
techniques. transmission each of N symbol is slightly affected. So though
the channel is frequency selective sub-channel is flat or
Keywords—- Orthogonal frequency division multiplexing
(OFDM); Discrete Fourier transform (DFT); Fast Fourier slightly frequency selective. This is why OFDM provide good
transform (FFT); protection against fading.

I. INTRODUCTION In an OFDM system there are N numbers of sub-channel. If N


is high then it will be very complex to design a system with N
OFDM is a modulation scheme that allows digital data to be modulators and demodulators. Fortunately, it can be
efficiently and reliably transmitted over a radio channel, even implemented alternatively using DFT / FFT to reduce the high
in multi-path environments. [1] OFDM transmits data by using complexity. A detailed system model for OFDM system is
a large number of narrow bandwidth carriers. These carriers are shown in Figure 1
regularly spaced in frequency, forming a block of spectrum.
The frequency spacing and time synchronization of the carriers
is chosen in such a way that the carriers are orthogonal,
meaning that they do not cause interference to each other. In
OFDM system DFT/FFT is used instead of modulators. In this
paper we have implemented a novel FFT to reduce
computational complexity of an OFDM system.

II. OFDM SYSTEM MODEL

OFDM is a kind of FDM (frequency division multiplexing)


technique in which we divide a data stream into a number of
bit streams which are transmitted through sub-channels [2].
The characteristic of these sub-channels is that they are
orthogonal to each other. As the data that are transmitted
through a sub-channel at a particular time are only a portion of
the data transmitted through a channel so bit rate in a sub-
channel can be kept much low. After splitting the data in N
parallel data streams each stream is then mapped to a tone at a
Figure 1: OFDM system

1-4244-1312-5/07/$25.00 © 2007 IEEE 57


Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on June 17, 2009 at 11:53 from IEEE Xplore. Restrictions apply.
2
III. FOURIER TRANSFORM ALGORITHMS But WN = WN / 2
so the expression can be written
( N / 2 ) −1 ( N / 2 ) −1
A. Discrete Fourier Transform (DFT) Algorithms
For a sequence of data x (n) of length N, the DFT may be
X (k ) =
m =0
1
( m )W Nkm/ 2 +W Nk
m=0
∑ f 2
( m )W Nkm/ 2 (8) ∑ f
expressed as [4] = F1 ( k ) + F 2 (k ); k = 0, 1& N -1 , (9)
N −1 Where, F (k) and F (k) are the N/2–point DFT of the
X(k) = ∑ x (n )W kn 0 ≤ k ≤ N − 1
1 2
(1) sequences f1(m) and f2(m) respectively. Since F1 (k) and
N F2 (k) are periodic, with period N/2, we can evaluate:
n=0 2 π
−j
Where, WN = e N (2) X (k ) = F1 ( k ) + F 2 (k ) (10)
DFT may be expressed in matrix form as: X (k + N / 2) = F1 (k ) - F 2 (k ) (11)
X N = WN * x N (3) For Equations (6) , (7) k=0, 1 … N/2-1. So there is reduction
1 1 1 : : 1  of number of multiplications from N2 to (N2 /2 + N/2), which
1 W 
( N −1)
is about a factor of 2, evaluating the significance of radix-2
 N W N2 : : W N  algorithm for efficient computation. So this algorithm can
1 W N2 W N4 : : W N

2 ( N −1)
compute N- point FFT in N/2 cycles.
Where WN =   (4)
: : : : : :  C. Radix-4 FFT Algorithm
: : : : : :  In case of N-data points expressed as power of 4V, we can
 N −1 2 ( N −1) ( N −1)( N −1)  employ radix-4 algorithm instead of radix-2 algorithm for
1 W N WN : : WN 
more efficient estimation. If we consider a column-wise
mapping for x(n) as n = Ml + m and row-wise mapping as
k=Mp + q and N be factored as N= LM, then we can write[4],
The expression for Inverse Discrete Fourier Transform (IDFT)
M −1 L −1 ( Mp + q )(mL +1) (12)
in matrix form is xN = WN−1 X N (5) X ( p, q ) = ∑ ∑ x(l , m)WN
m =0 l =0
We observe that for each value of k direct computation of After simplification, we get
X(k) involves N complex multiplications and N-1 complex L −1
  M −1 
additions. So, to compute N values of the DFT, N2 complex X ( p, q) = ∑ WNlq  ∑ x(l , m)WMmq  WLlp (13)
multiplications and (N2-N) complex additions are needed.  m =0 
l =0 
Direct computation of DFT is inefficient because of inability
to use symmetrical and periodical properties of phase factor Let us consider L=4, M=N/4, so it results in,
WN for computation. These properties are symmetry 3
property W K + N / 2 = −W K Periodicity property:
N N
X ( p, q ) = ∑ W
l =0
lq  lp
N F ( l , q )  W4 (14)
K +N
W N =W K
. Direct computation of DFT results in 4N2
N
Where, l = 0,1,2,3, q= 0,1,2, … , N/4-1 and p= 0,1,2,3
real computations and 4N (N-1) real additions. So different
( N / 4 ) −1
DFT algorithms optimize each of these computational
processes in unique ways. And F (l , q ) = ∑ x(l , m)W
m =0
mq
N /4 . (15)

B. Radix-2 FFT Algorithm So, the four N/4 points DFT are obtained from the above
For speed demanding applications which requires expression, which are combined in previous expression to
implementation of FFT algorithm, FFT algorithms are evaluate N-point DFT. So this algorithm results in (3N/8)
decimated in terms of time or frequency. If we consider N log2N complex multiplications and (3N/2) log2N complex
points of data where it can be factored as N = rv. In such case, additions. So the number of multiplications is reduced by
DFTs are of r size and the number r is called radix of the FFT 25%, but the number of addition is increased by 50%.
algorithm. So, in radix-2 FFT N= 2V .
If N-point data sequence is decimated by a factor of 2, it is IV. DEVELOPMENT OF A NOVEL FFT
called radix-2 algorithm. Now from expression of N-point
DFT, we can get In an OFDM system we use modulation techniques such as
BPSK, 4-QAM, 16-QAM, 64-QAM etc. For a particular
X (k ) = ∑ x (n)WN + ∑ x (n)WN
kn kn (6)
QAM technique fixed symbols are sent. This paper converses
even odd
( N / 2 ) −1 ( N / 2 ) −1
an FFT algorithm that works faster for different QAM
techniques.
= ∑ x(2m)W
m =0
2 mk
N + ∑ x(2m + 1)W
m=0
( 2 m +1) k
N
(7)

1-4244-1312-5/07/$25.00 © 2007 IEEE 58


Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on June 17, 2009 at 11:53 from IEEE Xplore. Restrictions apply.
A. BPSK We can Dot multiply each of these symbols with WN matrix
In BPSK modulation technique we send two kinds of and store in Wa, Wb, Wc, Wd matrices.
symbol. They are (1) & (-1). On the sending end IDFT and on Let Wa= (1+j).* WN
the receiving end DFT operation is done upon these symbols. Similarly: Wb = (1-j).* WN; Wc = (-1+j).* WN &
The formulas 3, 5 are generalizing formulas for DFT & IDFT Wd = (-1-j).* WN .
operation. The process is similar to that of the BPSK. we have to identify
the value of xN matrix in every position and according to that
But we want to concentrate only the symbols that we are we have to pull required value from Wa , Wb, Wc ,Wd matrices
sending. As in BPSK only two symbols are send so modulated and do the required additions.
data xN will have a combination of these two symbols. As data
sending is a continuous process so we are sending these C. Generalized M-QAM
symbols repeatedly and in DFT/IDFT operation we are doing In M-QAM modulation technique we are sending M
repeated complex multiplications. But if we store the result of symbols. We have done Dot multiplication each of these
all complex multiplications that can occur for these two symbols with WN matrix where N=M and stored these
symbols in each position of xN we can evade these repeated matrices. Again we have to identify the value of xN matrix in
complex multiplications. This idea is illustrated by the every position and according to that we have to pull the
following example. required value from stored matrices and do the required
additions.
Let N =4. At first we consider xN has only 1 symbol in it.
We dot multiply xN with WN and the result is stored in Wa V. COMPUTATIONAL PERFORMANCE
matrix. Now we consider xN has only -1 symbol. We dot ANALYSES
multiply xN with WN and the result is stored in Wb matrix. So
all kind of complex multiplications that can occur are already In order to compare the computational complexities among the
stored in Wa & Wb matrices. Now we arbitrarily choose our different Fourier transforms & novel FFT on OFDM, we have
data as xN =[ 1 -1 -1 1] . performed the calculations based on the OFDM block sizes
Here, xN (1,1)= 1; xN(1,2)= -1 ; xN(1,3)= -1 ; & xN(1,4)= 1. given in Table I. We have done all the mathematical
Let the DFT of xN is to store in DFT_ xN matrix. So, comparisons according to multiplications & additions done.
In novel FFT whatever is the total number of OFDM block,
DFT_ xN (1,1) = Wa(1,1)+ Wb(2,1)+ Wb(3,1)+ Wa(4,1); number of multiplications is always fixed.

DFT_ xN (1,2) = Wa(1,2)+ Wb(2,2)+ Wb(3,2)+ Wa(4,2); TABLE I. COMPARISON OF DIFFERENT FOURIER TRANSFORM

DFT_ xN (1,3) = Wa(1,3)+ Wb(2,3)+ Wb(3,3)+ Wa(4,3); OFDM Complex Complex Complex Complex Complex
Block, multiplicatio multiplicatio multiplicatio multiplicat multiplicat
n n n ion ion
DFT_ xN (1,4) = Wa(1,4)+ Wb(2,4)+ Wb(3,4)+ Wa(4,4); N In DFT In FFT In novel In novel In novel
algorithm FFT FFT FFT
Now, if xN =[1 1 -1 1] . Then xN(1,1)= 1 ; xN(1,2)= 1 ; N2 algorithm algorithm algorithm
(N/2)log2 N (4QAM) (16QAM) (64QAM)
xN(1,3)= -1 & xN(1,4)= 1;
4 16 4 16 256 4096
then DFT_ xN (1,1) = Wa(1,1)+ Wa(2,1)+ Wb(3,1)+ Wa(4,1) ;
8 64 12 16 256 4096
DFT_ xN (1,2) = Wa(1,2)+ Wa(2,2)+ Wb(3,2)+ Wa(4,2); 16 256 32 16 256 4096
32 1024 80 16 256 4096
DFT_ xN (1,3) = Wa(1,3)+ Wa(2,3)+ Wb(3,3)+ Wa(4,3);
64 4096 192 16 256 4096
DFT_ xN (1,4) = Wa(1,4)+ Wa(2,4)+ Wb(3,4)+ Wa(4,4) ; 128 16384 448 16 256 4096
256 65536 1024 16 256 4096
So, in this way by storing Wa & Wb we can avoid repeated
complex multiplications. All we have to do is to identify what
512 262144 2304 16 256 4096
value of xN is in what position of xN matrix and according to 1024 1048576 5120 16 256 4096
that we have to pull required value from Wa & Wb matrices 2048 4194304 11264 16 256 4096
and do the required additions.

B. 4-QAM
Figure 2 gives the comparison between DFT, Radix-2 FFT and
In 4-QAM modulation technique we are sending
our FFT in terms of complex multiplications done.
four symbols. They are:(1+j), (1-j),(-1+j)&(-1-j) .

1-4244-1312-5/07/$25.00 © 2007 IEEE 59


Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on June 17, 2009 at 11:53 from IEEE Xplore. Restrictions apply.
Figure 2: Comparison between DFT, Radix-2 FFT and our FFT (Complex multiplications)

VII. CONCLUSIONS
[4] Proakis J.G., Manolakis D.G “Digital Signal Processing :Principles,
Algorithms And Applications” Third edition, chap.-6, Pages 448-475,
The performance of an OFDM system depends on DFT/FFT 2002.
as in an OFDM system DFT/FFT works as an modulator. If
the computational complexity of DFT/FFT decreases, then the [5] Langton C. “Orthogonal frequency divisional multiplex (OFDM)
speed of OFDM system will increase. This novel FFT tutorial” at www.complextoreal.com [Last accessed: July 13, 2006].
algorithm decreases computational complexity by discarding
repeated calculations. As a result it increases the speed as well [6] Wirastuti N., Noras J.M., Jones S.M.R. ”Evaluation of the very fast
as the performance of OFDM system. fourier transform applied to OFDM” , 2nd IEE-EURASIP DSPEnabled
Radio Conference, 19-20Sept, 2005, Southampton, UK
ACKNOWLEDGEMENT
[7] Zalewski M. , Schuupp S.”FFT implementations” RPI, NY
The authors would like to thank Mr. Shahriar Rahman &
Mr. Riaz Raihan Haque for their extended assistance with [8] Goldsmith, A., Wireless Communications, first edition, Cambridge
University Press, 2005.
lecture notes of Professor Sheng Chen during the time of this
work.
[9] Shafik, Rishad A. Technical Report, “Personal Multimedia
Communication- Simulations and Analyses”.
REFERENCES
[10] Prasad H. “Simulation and Software Radio for Mobile Communication”,
[1] Eric Phillip Lawrey B.E. “Adaptive Techniques for Multi-user OFDM”, Chap-4, pp. 165-169.
pp.33-34,2004.

[2] Chen S., Lecture note“ Fast Fourier Transform” in lectures of Radio
Communications Networks and Systems, 2005.

[3] The International Engineering Consortium WEB ProForum Tutorial


“OFDM for mobile data communications” pp. 1-16 at www.iec.org.
[Last accessed:13th july,2006]

1-4244-1312-5/07/$25.00 © 2007 IEEE 60


Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on June 17, 2009 at 11:53 from IEEE Xplore. Restrictions apply.

You might also like