Cb25464467

B.NIHAR, POORNIMA PADARAJU / International Journal of Engineering Research and
Applications (IJERA) ISSN: 2248-9622 www.ijera.com
Vol. 2, Issue 5, September- October 2012, pp.464-467
SNR & BER Optimization For Pre-DFT Combining In Coded
SIMO-OFDM Systems
B.NIHAR (M.Tech), POORNIMA PADARAJU(M.Tech)
(Department of ECE, ASTRA, Hyderabad-40)
(Department of ECE, ASTRA, Hyderabad-40)

Abstract
For coded SIMO-OFDM systems, pre- and limitation of two previously-proposed criteria for
DFT combining was previously shown to provide obtaining the pre-DFT combining weights, i.e.,
a good trade-off between error-rate performance maximization of the sum of subcarrier signal-to-noise
and processing complexity. Max-sum SNR and ratio (SNR) values (called max-sum SNR hereafter)
max-min SNR are two reasonable ways for [1] and maximization of the minimum subcarrier
obtaining these combining weights. In this letter, SNR value (called max-min SNR hereafter) [2]. Our
we employ multi objective optimization to further results show that neither max-sum SNR nor maxmin
reveal the suitability and limitation of these two SNR is universally good. Furthermore, for better error
criteria. Our results show that: 1) Neither rate performance, the means for weight calculation
maxsumSNR nor max-min SNR is universally should be adapted according to the capability of the
good; 2) For better error-rate performance, the error-correcting code used, and multiobjective
means for weight calculation should be adapted optimization can help in the determination. Monte
according to the capability of the error-correcting Carlo simulations are finally provided to verify the
code used, and multi objective optimization can correctness of these sayings. Throughout the
help in the determination. letter, we use boldface letters, boldface letters with
over bar, lower-case letters, and upper-case letters to
Index Terms— SIMO, OFDM, pre-DFT denote vectors, matrices, time-domain signals, and
combining, convex optimization, multiobjective frequency domain signals, respectively. Besides, (⋅)T ,
optimization. (⋅)H, trace(⋅), rank(⋅), and diag{⋅} are used to
represent the matrix transpose, matrix Hermitian,
I. INTRODUCTION matrix trace, matrix rank calculation, and diagonal
ORTHOGONAL frequency division matrix with its main diagonal being the included
multiplexing (OFDM) combined with multiple vector, respectively.
receive antennas, namely, single-input multiple-
output (SIMO) OFDM, has recently been II. PRE-DFT COMBINING IN SIMO-OFDM
investigated for use in wireless communication SYSTEMS
systems. It can provide high spectrum efficiency and We consider an SIMO-OFDM system with
high data rate for information transmission. On one 𝑀 receive antennas. Define an N  1 signal vector
hand, OFDM divides the entire channel into many S( 𝑘) = [S( 𝑘𝑁) S( 𝑘𝑁+1) ⋅ ⋅ ⋅ S( 𝑘𝑁 + 𝑁 − 1)]T as the
parallel sub channels which increases the symbol 𝑘th OFDM data block to be transmitted, where 𝑁 is
duration and therefore reduces the inter-symbol the number of subcarriers. This data block is first
interference (ISI) caused by multipath propagation. modulated by the inverse DFT (IDFT). With matrix
Besides, since the subcarriers are orthogonal to each representation, we can write the output of the IDFT
other, OFDM can utilize the spectrum very as s( 𝑘) = [ 𝑠( 𝑘𝑁) 𝑠( 𝑘𝑁 +1) ⋅ ⋅ ⋅ 𝑠( 𝑘𝑁 + 𝑁 −1)]T = F
efficiently. On the other hand, SIMO along with H
S(k), where F is an 𝑁 × 𝑁 DFT matrix with
combining techniques takes advantage of receive
spatial diversity and therefore further enhances the elements [ F ] 𝑝, 𝑞 = (1/ N ) exp (−j2 𝜋𝑝𝑞/ 𝑁) for 𝑝, 𝑞
performance. = 0, 1, ⋅ ⋅ ⋅ , 𝑁 − 1 and 𝑗 =  1 . A cyclic prefix (CP)
is inserted afterwards and its length (Lcp) is chosen to
It is known that subcarrier-based maximum be longer than the maximum length of the multipath
ratio combining (MRC) performs the best for coded fading channel ( 𝐿). Also define an 𝑁 × 1 vector hm =
SIMO-OFDM systems; however, it requires high [ℎm(0) ℎm(1) ⋅ ⋅ ⋅ ℎm( 𝐿 − 1) 0 ⋅ ⋅ 0]T , where ℎm( 𝑙)
processing complexity. Pre-discrete Fourier represents the 𝑙th channel coefficient for the 𝑚th
transform (DFT) combining was then developed, in receive antenna, with 𝑙 = 0, 1, ⋅ ⋅ ⋅ , 𝐿 − 1 and 𝑚 = 0,
which only one DFT block is necessary at the 1, ⋅ ⋅ ⋅ , 𝑀 − 1. Collecting all channel vectors from the
receiver [1]. It was previously shown to provide a
𝑀 different receive antennas, we construct an 𝑁 × 𝑀
good trade-off between error-rate performance and
processing complexity. In this letter, we employ channel matrix h = [h0 h1 ⋅ ⋅ ⋅ h 𝑀−1], and its
multiobjective optimization to reveal the suitability frequency response as

464 | P a g e

H = [H0 H1 ⋅ ⋅ ⋅ H 𝑀−1] = F h (1) Both criteria are reasonable for obtaining the pre-
DFT combining weights. Nevertheless, two questions
with H 𝑚 = F h 𝑚 .In an ordinary OFDM are naturally raised: 1) Is one of the two criteria
signal reception process, after CP removal and DFT strictly superior to the other? 2) Can we further
demodulation, the resultant 𝑁 × 1 signal vector from improve the error-rate performance with pre-DFT
the 𝑚th receive antenna, denoted by R 𝑚( 𝑘), can be combining? We try to answer these questions through
shown to be the use of multiobjective optimization in the
following.
R( 𝑘) = diag{S( 𝑘)}H 𝑚 +N 𝑚( 𝑘) (2)
where N 𝑚( 𝑘) is an 𝑁 × 1 complex Gaussian
II. MULTIOBJECTIVE OPTIMIZATION FOR
noise vector with zero mean and equal variance for
each element. For the considered SIMO scenario, we
PRE-DFT COMBINING
Although max-sum SNR and max-min SNR
can collect the 𝑀 received signal vectors and form an
are both practical, they are normally in conflict with
𝑁 × 𝑀 received signal matrix as
each other, i.e., an improvement in one leads to
deterioration in the other, which will be shown later
R ( 𝑘) = [ R0( 𝑘) R1( 𝑘)⋅⋅ ⋅ R 𝑀− 1( 𝑘)]. ( 3) in this section. This motivates the use of
multiobjective optimization for gaining further
Let w = [ 𝑤0 𝑤1 ⋅ ⋅ ⋅ 𝑤 𝑀−1] be an 𝑀 × 1 weight vector. insight into the two problems for the case at hand can
be stated as follows:
max g(w) =  1
g (w) 
, subject to w 𝐻w = 1 (7)
 g (w)
w
 2 

with 𝑔1(w) = max  n w 2and 𝑔2(w) = (wH HH H
n
w)/in which 𝑔2(w) is normalized for
convenience during numerical calculation. With (7),
Fig. 1 Block diagram of OFDM diversity receiver we can generally look for some good trade-offs,
with pre-DFT combining rather than a single solution of either max-sum SNR
or max-min SNR. For this problem, a solution is
With (1)-(3), the pre-DFT combining operation and optimal if there exists no other solution that gives
the resultant 𝑁 × 1 signal vector can be expressed as enhanced performance with regard to both 𝑔1(w) and
𝑔2(w) - Pareto optimizers. The set of Pareto
Y ( 𝑘) = R ( 𝑘) w = diag{S(𝑘)} H w+ N ( 𝑘)w (4) optimizers is called the Pareto front [3].However;
with N ( 𝑘) = [N0( 𝑘) N1( 𝑘) ⋅ ⋅ N 𝑀−1( 𝑘)]. Fig. 1 is the there is no systematic manner to find the Pareto front
block diagram of a simplified OFDM receiver in (7). Instead, we use a simple and popular way, i.e.,
performing pre- DFT combining. In [1], w was the weighted-sum method, to approach to the solution
set. This essentially converts the multiobjective
calculated based on max-sum SNR. For that case, the
optimization problem into a single objective problem.
optimum w can be shown to be the solution of the
Mathematically speaking, the objective function in
following optimization problem:
this circumstance is changed to be a linear
combination of the two objectives as
max wH H H H w subject to w 𝐻w = 1 (5)
w
max  𝑔1(w) + (1 −  ) 𝑔2(w), subject to w 𝐻w = 1 (8)
in which wH H H H w indicates the sum of the w
signal power in all 𝑁 subcarriers. As an alternative, where  ∈ [0, 1] is a parameter determining
pre-DFT combining based on max-min SNR was the relative importance between max-sum SNR and
proposed in [2]. Define a 1 × 𝑀 vector  n as the 𝑛th max-min SNR. Solving (8) yields the solution that
gives the best compromise for a typical  Next, we
row of the channel matrix H given in (1), with 𝑛 = 0, show that (8) can be efficiently evaluated via convex
1, ⋅ ⋅ ⋅ , −1. For that approach, the optimization of w optimization techniques. Without loss of generality,
can be described as we can recast the optimization problem in (8) to be
max max  n w 2subject to w 𝐻w =1 (6) max  [ min trace( n W) ]+ (1 −  ) [trace (QW)]
w n W
subject to trace(W) = 1, rank(W) = 1, W  0 (9)
in which  n w 2 indicates the signal power of the with n =  n H  n and Q = H H H
𝑛th subcarrier after pre-DFT combining. In (9), W is an 𝑀 × 𝑀matrix to be
It is understood that while max-sum SNR tends to determined and the inequality W  0 means that W
help the good, max-min SNR tends to help the bad. is symmetric positive semi definite. Instead of

465 | P a g e

solving the above nondeterministic polynomial-time phase-shift keying (QPSK) is used for modulation.
hard (NP-hard) problem directly, we seek an Besides, 𝑁 = 64, 𝐿cp = 16, and 𝐿 = 2(independently
approximation of the solution. generated with the Rayleigh distribution) are set.
By dropping the nonconvex rank-one constraint, this Convolution codes with different error-correcting
weighted sum objective function can be relaxed to capabilities (different minimum free distance 𝑑free)
max  [ min trace ( n W) ]+ (1 −  ) [trace (QW)] are used for error protection. At the receiver, the
W Viterbi algorithm with hard decision is employed for
subject to trace(W) = 1, W  0. (10) decoding. Figs. 3 and 4 present the corresponding
Let 𝑧1 and 𝑧2 be two scalars. The relaxation is BER performance. From these figures, we have the
equivalent to following observations: For the case of higher error-
max  z1+ (1 −  ) z2 correcting capability (Fig. 3), max-sum SNR performs
W slightly better than max-min SNR. Note that
subject to trace ( n W)  z1, trace (QW)  z2 maxsumSNR generally focuses on the good and
subject to trace (W) = 1, W  0. ignores the bad. With the relatively large amount of
error protection, the low sub carrier SNR values may
which becomes convex. It is not difficult to be compensated. Together with the “boosted” high-
see that (11) can be categorized to be a semi definite SNR subcarriers, max-sum SNR provides better BER
programming problem. The optimal choice of W, i.e., performance in this case. On the contrary, for the
Wopt, can be obtained systematically using the case of lower error-correcting capability (Fig. 4),
efficient interior point method [4], and then a maxminSNR outperforms max-sum SNR, especially in
randomization step is used to produce an the high SNR region. The small amount of error
approximated solution to (7). In general, the protection makes each subcarrier equally essential.
complexity from weight calculation can be ignored as Max-min SNR usually does a good job in balancing
compared with the complexity saving from the the subcarrier SNR values, and thus gives better BER
reduction of DFT components [1], [2]. performance. Moreover, it is interesting to note that
in either Fig. 3 or Fig. 4, the weighted-sum method
which successfully captures the advantages of both
max-sum SNR and max-min SNR is superior to these
two previously-proposed criteria. By varying 𝜆, there
exist some cases in which a lower BER can be
achieved. That is to say, multiobjective optimization
can be employed to form some better pre-DFT
combining weights over the pure max-sum SNR and
max-minSNR. By means of exhaustive simulations,
we find that the effect of max-min SNR is more
substantial than that of maxsumSNR in most
circumstances.

Fig. 2. Pareto front for max-sum SNR and max-min
SNR with SNR=15 dB, 𝑀 = 2, 𝑁 = 64, 𝐿cp = 16, 𝐿 =
2, and  = [0: 0.1: 0.8 0.9: 0.05: 1].

An example of a typical Pareto front solved
via (11) is illustrated in Fig. 2. To obtain the entire
approximation set, the search is repeated with various
values of 𝜆. We clearly see the trade-off between
max-sum SNR and max-min SNR. Besides, the
weighted-sum method along with the convex
formulation can efficiently approach the Pareto front,
as expected.
Fig. 3. BER for 1/2-rate convolutional-coded SIMO-
IV. SIMULATIONS AND DISCUSSION OFDM with generator sequence ([247 371])8 and
A comparison of the bit-error-rate (BER) 𝑑free = 10 [5].
performance with different pre-DFT combining is
made by Monte Carlo simulations carried out
regarding a 1 × 2 coded OFDM system. Quadrature

466 | P a g e

-1
Performance analysis of multi objective SIMO
10
MAX-SUM
MAX-MIN
COFDM,” IEEE Trans. Veh. Technol., vol.
MULTIOBECT LAMDA=0.7
MULTIOBECT LAMDA=0.9 50, pp. 487–496, Mar. 2001.
[2] Y. Lee, “Max-min fair pre-DFT combining
for OFDM systems with multiple receive
antennas,” IEEE Trans. Veh. Technol., vol.
58, pp. 1741–1745, May 2009.
BER

-2
10

[3] W. Stadler, Multicriteria Optimization in
Engineering and in the Sciences. New York:
Plenum Press, 1988.
[4] S. Boyd and L. Vandenberghe, Convex
Optimization. Cambridge Univ. Press, 2004.
10
-3

2 4 6 8 10 12 14 16 18 20
[5] J.-J. Chang, D.-J. Hwang, and M.-C. Lin,
SNR per receiving anteena
“Some extended results on the search for
good convolutional codes,” IEEE Trans. Inf.
Fig. 4 BER for 3/4-rate convolutional-coded SIMO- Theory, vol. 43, pp. 1682–1697, Sept. 1997
OFDM with generator sequence ([1 1 1 0],[3 0 0 1],
[3 2 0 2])8 and 𝑑free = 3 [5].

Simulation result for code rate 1/3 with minimum
distance 15 is also calculated in this calculation 𝜆 is
taken very close to 1 is used which improves the
BER value which is shown in Fig 5

-3
Performance analysis of multi objective SIMO
10
lam=0.99
BER

-4
10

-5
10
0 5 10 15 20 25 30
SNR for receivng anteena

Fig. 5 BER for 1/3-rate convolutional-coded SIMO-
OFDM with generator sequence [117 127 155]8
And 𝑑free = 15[5]

V. CONCLUSIONS
This letter has discussed and compared the
error-rate performance for coded SIMO-OFDM
systems with different pre-DFT combining. Our
results show that multiobjective optimization can be
used to determine some better pre-DFT combining
weights, which are generally superior to both
maxsum SNR and max-min SNR for achieving a lower
BER.

REFERENCES:
[1] M. Okada and S. Komaki, “Pre-DFT
combining space diversity assisted

467 | P a g e

Cb25464467

More Related Content

What's hot (18)

Viewers also liked (20)

Similar to Cb25464467 (20)

Cb25464467