BER Performance Improvement in OFDM System with ZFE and MMSE Equalizers

Bhasker Gupta, Gagan Gupta, and Davinder S. Saini

Department of Electronics and Communication Engineering,
Jaypee University of Information Technology, Waknaghat, Distt. Solan, Himachal Pradesh – 173215
[email protected], [email protected], and [email protected]

Abstract— In this paper, an Orthogonal Frequency Division spectral usage by allowing subcarrier overlapping in the
Multiplexing (OFDM) system with equalizers is modeled. Two frequency domain. The conversion of parallel low rate
different equalizers, namely Zero Forcing (ZF) and Minimum frequency domain signals to time domain at the transmitter
Mean Square Error (MMSE), along with different modulations and time domain signals to frequency domain at the receiver
are used. The modulation with multicarrier is employed, which is performed, using inverse fast fourier transform (IFFT)
provides advantages like inter symbol interference (ISI) and fast fourier transform (FFT) operations. The IFFT and
reduction, high reliability, and better performance in multi- FFT operations are computationally efficient. The various
path fading. These equalizers are adopted to remove the ISI subcarriers in OFDM are synchronized in time and
generated in the transmitted data under various fading
frequency, representing a single block of spectrum. This is
environments. The results show that, with MMSE and ZFE
equalizers, the bit error rate (BER) performance is improved.
to ensure that the orthogonal nature of the structure is
Further, the BER performance of MMSE is superior to ZFE maintained.
equalizer.
II. OFDM SYSTEM MODEL
Keywords:-OFDM; Equalizer; ZFE; MMSE; The schematic model for OFDM system used in the
paper is shown in Figure 1. The basic and main building
I. INTRODUCTION blocks of model shown in Figure 1 are transmitter, channel
In modern high data rate communication systems, e.g. and receiver.
digital video broadcast (DVB), OFDM multicarrier A. Transmitter and receiver section.
modulation technique [1] is preferred. In OFDM, a single
high rate data stream is divided into multiple low rate data This segment consists of following blocks.
streams, and each stream is modulated with subcarrier 1) S/P converter and P/S converter: This block is used at
which is orthogonal to the remaining ones. OFDM provides both, transmitter and receiver. These blocks convert random
an efficient transmission over limited bandwidth. The generated data with high rate into low data rate and vice
OFDM systems exhibits delay spread tolerance and efficient versa.

OFDM Transmitter

Data to Zero Padding &

Random Cyclic
S/P Symbol Inverse Fast
Data Prefix P/S
Mapper Fourier
Generator Insertion
Transform (IFFT)

AWGN
Noise Rayleigh/
Rician Multi-
Path Channel

OFDM Receiver

Cyclic Symbol
Time Zero To Data P/S
S/P Prefix Fast Fourier
domain Padding De-
Removal Transform
Equalization Removal Mapper
(FFT)

Figure 1. OFDM Simulation Model

___________________________________
978-1-4244 -8679-3/11/$26.00 ©2011 IEEE

193
 2) Data to symbol mapper and De-mapper: This block interference occur. For linear channels, cyclic prefix is used
provides modulation. The input data bits are grouped to convert linear convolution of data into circular
decrease data rate. When input stream is of low rate, a convolution [6, 7, 8]. For both the cases, we need to use
simple BPSK modulation works well; otherwise, QPSK or equalization. For the model used in the paper, two different
16/64-QAM is used. The selection of modulation scheme data symbol before appended zero padding
applied to each sub channel depends solely on the 1

compromise between the data rate requirement and 0.8

transmission robustness. 0.6

3) Zero padding/removal and IFFT/FFT: Zero padding
0.4
increase sampling rate for better resolution of signal in
frequency domain. While zero padding provides spectral 0.2

amplitude
interpolation in time domain, the IFFT converts frequency 0

domain data into time domain, maintaining the -0.2

orthogonality of subcarriers. Before IFFT, numbers of zeros -0.4
are inserted in the input to make its length equal to IFFT bin
-0.6
size (say L, L is 64 in this paper). This addition of zeros is
called zero padding, and is used only when the subcarriers -0.8

are less than bin size. Generally, in OFDM system, we -1

0 1 2 3 4 5 6 7 8 9 10
select the number of data symbols equal to 52 giving 12 frequency

zero symbols. Zero padding removal does the reverse at Figure 2.The modulated data before appended zero padding
receiver. Further, the zero padding bits are located at the
edges of the spectrum shown below in Figures 2-3. equalization techniques, namely, ZF equalization and
MMSE equalization both in time domain are investigated. A
Let Xp(k) is the input vector to IFFT block and k varies
from 0 to N-1. The output of IFFT block is given by 1
data symbol after appended zero padding

0.8
1 N −1
x p (n) = IFFT [ X p (k )] = ¦ Xp(k )e − j 2πkn / N (1) 0.6
N k =0
0.4

This applies to all the parallel branches, while FFT is 0.2

amplitude

taken at the receiver side in the parallel branches. 0

-0.2
N −1
X p (k ) = FFT [ x p (n)] = ¦ x p (n)e − j 2πkn / N (2) -0.4
n=0
-0.6

Where n varies from n=0, 1, 2……N-1 -0.8

4) Cyclic Prefix and cyclic prefix removal: The cyclic -1

0 1 2 3 4 5 6 7 8 9 10
prefix is used to mitigate ISI effect in original OFDM frequency
symbol. This is achieved by adding partial symbol
information of each cycle to the beginning of the symbol. Figure 3. The modulated data after appended zero padding.
Higher is the delay spread, higher is the length of cyclic
time domain equalizer use shorter length of cyclic prefix
prefix, and in this paper, the CP length is chosen 1/4 of the
which does not affect efficiency of system.
symbol period. After CP insertion, the symbols from parallel
paths are combined to make a serial data. In OFDM, one a) Zero Forcing Equalizer: It is a linear equalizer, which
frame length is inverts the frequency response of the channel. The ZFE
transfer function is the inverse of that of channel’s response.
T = TS + TCP The name Zero-Forcing, signify bringing down the ISI to
zero in a noise free case. This is useful when ISI is
where TS = NT, N is number of carriers, and TCP is cyclic significant compared to noise. Let the discrete time
prefix duration. transmitted symbols, the channel response, and the set of
filter coefficients are represented as s(k), h(k), and c(k)
5) Equalization in Time domain: Considering non linear respectively as shown in Figure 4. According to the
channel (or when signaling is not done according to nyquist definition of ZFE algorithm we have to find filter
criterion), there is a great chance that intersymbol coefficients which can mitigate the channel effects such that

194
 Noise
RXY = h(k) and RYY = E(h*hT) + E(n2) (10)
s[k] Channel h[k] y[k] Equalizer c[k]
+ B. Channel description
Adder In generally there are various fading channels, which
includes AWGN, Rician, Rayleigh and Nakagami-m, and in
Figure 4. General Equalizer Model this paper OFDM is simulated in three type of fading
channels i.e. AWGN, Rayleigh and Rician.
1) AWGN channel: The most common channel is additive
h(k ) ⊗ c(k ) = δ (k ) (3) white Gaussian noise (AWGN) channel [10] which is static
in nature. In wireless communication, the AWGN channel is
In (3), symbol ⊗ represents circular convolution. These one in which the signal is distorted with wideband noise of
filter coefficients c(k) are then convolved with received time constant spectral density. This type of channel does not
domain discrete signal y(k) to equalize y(k) as given below. produce any type fading or interference. In practical
scenario, AWGN does not work well and thus more
Y ZF = c(k ) ⊗ y(k ) (4) practical channels are opted.
2) Rayleigh channel: In a Rayleigh fading channel model,
Equation (4) can be further simplified as it is assumed that that there is no direct path [4, 5, 11]
between transmitter and receiver out of all multiple
Y ZF = c(k ) ⊗ [s(k ) ⊗ h(k ) + n] (5) reflective paths. The output of such channel can be
expressed as
Where n is AWGN noise. Equation (5) can be further
reduced to R(n) = ¦ h(n, τ ) S (n − m) + Z (n) (11)

Y ZF = s(k ) + c(k ) ⊗ n (6)

Where Z(n) is AWGN noise with zero mean and unit
variance, h(n) is channel impulse response which is equal to
Also, we can develop the same model in frequency
domain. If the channel response in frequency domain
h(n) = ¦ α (n)e − jθ ( n ) (12)
response is H[f], filter coefficients C[f] is constructed by C(f)
=1/H(f). Thus combination of channel and equalizer gives a
flat frequency response H(f)C(f) = 1. Where Į(n) and ș(n) are attenuation and phase shift for
nth path. So, we summarize them for all the channel taps.
b) Minimum mean square error (MMSE) equalizer: The The Rayleigh channel can be simulated as flat or frequency
MMSE equalizer minimizes the mean square error (MSE) [9] selective channel depending upon its coherence bandwidth.
between desired signal and equalized signal. Mathematically, If coherence bandwidth is larger compared to signal
bandwidth, the channel is called flat, otherwise, it is called
E (e(k )) = E s(k )−Y ZF
2
2
(7) frequency-selective. In this paper, OFDM is simulated under
Rayleigh frequency-selective channel. The channel is
implemented by choosing number of channel taps (say 18)
Equation (7) can be solved as which should be more than cyclic prefix length (say 16).
The channel h is known at the receiver.
E (e(k )) = E s(k )− c(k )⊗ y(k )
2 2
(8) 3) Rician channel: The Rician fading is similar to that
for Rayleigh fading, except that in Rician fading [12], a
where e(k) is the error at sample time k, s(k) is desired strong dominant component exists. This dominant
signal, c(k) is set of filter coefficients and YZF is equalized component can for instance be the line-of-sight wave.
output. For MMSE criterion we need to find c(k) which Specifically, in Rician model we have exhibits following
minimizes the E(e(k))2, The filter coefficients c(k) will be properties
minimized if a) The dominant wave can be a phasor sum of two or
more dominant signals, e.g., the line of sight plus a ground
c(k ) = RYY
−1
R XY (9) reflection. This combined signal is then mostly treated as a
deterministic (fully predictable) process.
where RXY is cross correlation between input sequence b) The dominant wave can also be subject to shadow
and received sequence, i.e., between s(k) and y(k). RYY is attenuation.
auto correlation of received sequence i.e. y(k). Simplifying The system model for Rician channel is same as
we get Rayleigh channel but with difference in scaling factor.
III. SIMULATION RESULTS

195
 Simulation parameters chosen are listed in Table 1. The Additive White Gaussian noise, Rayleigh channel and
simulation results are plotted for bit error rate (BER) BER for OFDM using EQUALIZER in a Rayleigh fading channel
0
performance of OFDM system Vs Eb/No. Simulation is 10
Theoretical BER for RAYLEIGH
Simulated BER for ZFE due to RAYLEIGH
Table I Simulation Parameters for OFDM transceiver -1 Simulated BER for MMSE due to RAYLEIGH
10
S.No. Parameter Value
Carrier modulation
1 QPSK and 16-QAM
used 10
-2

Bit Error Rate

2 Coding Rate 1/2
12Mbps (QPSK)
3 Data Rate
24Mbps (16-QAM) -3
10
4 52
5 No. of pilot subcarriers None
6 IFFT size 64 10
-4

cyclic extension of the

7 Guard period type
symbol
8 Cyclic prefix length 16 -5
10
9 Window type No windowing used 0 5 10 15 20 25 30 35
Eb/No, dB
10 Bandwidth 20MHz
18 (Freq. selective
Number of channel
11 channel) Figure 6. BER for OFDM using Equalizer in a Rayleigh channel
Taps
10 (Freq. Flat channels)
12
Sub carrier frequency
20 MHz/64 = 0.3125 MHz
Rician channel is plotted. Three plots are shown: 1)
spacing theoretical value of BER for OFDM in AWGN; 2)
13 TFFT : IFFT/FFT period 3.2μsec simulated result using ZFE equalizer; 3) simulated result
TCP : cyclic prefix
14 0.8μsec
duration BER for OFDM using EQUALIZER in a RICIAN Fading Channel
0
Total OFDM symbol 10
15 TFFT + TCP = 4μsec
duration Theoretical BER for RICIAN
Number of carrier/symbol Simulated BER for ZFE due to RICIAN
Symbol Rate Simulated BER for MMSE due to RICIAN
16 duration = 52/3.2 μsec, -1
10
16.2Msps

-2
10
Bit Error Rate

0
BER for OFDM using EQUALIZER in a AWGN Channel
10
Theoretical BER for AWGN
Simulated BER for ZFE due to AWGN -3
10
-1 Simulated BER for MMSE due to AWGN
10

-4
10
-2
10
Bit Error Rate

-5
10
0 5 10 15 20 25 30 35
-3
10 Eb/No, dB

-4
10 Figure 7. BER for OFDM using Equalizer in a Rician channel

using MMSE equalizer under a particular channel condition.

-5
10
Figure 5 shows simulation results in AWGN channel. It can
0 5 10 15 20 25 30 35 be observed that BER start decreasing monotonically as
Eb/No, dB SNR increases, but still BER using equalizers is more than
the theoretical limit. The BER reduces to zero at SNR of
Figure 5. BER for OFDM using Equalizer in a AWGN channel
around 22dB. If we compare ZFE and MMSE equalizers,
performed for OFDM model given in Figure 1, for flat or initially MMSE outperforms ZFE, but as number of iteration
frequency-selective channel in presence and absence of increases, or SNR increases, both ZFE and MMSE converge
equalizer. Simulation results for the OFDM in presence of with each other.

196
 Figure 7 shows the results in presence of Rayleigh
channel. It shows the same pattern as in Figure 6, but with
more errors as more practical channel is used. In this case,
BER again starts decreasing monotonically with SNR and
reduces to zero at SNR beyond 35dB. Initially MMSE
outperforms ZFE and later, both converge with each other.
Figure 8 shows the simulation results in Rician channel. The
BER pattern is same as that of Fig 5 and 6. Overall, Figures
5-7 implies that simulated BER is still much higher than
theoretical limits, which can be achieved more novel
equalizers.

IV. CONCLUSION
Multi-path propagation leads to ISI in received signal.
To eliminate ISI, we need to equalize the received signal.
Equalizers like ZFE and MMSE can be employed in time
domain for this purpose. The BER performance improves
significantly using these equalizers, but still ISI is not
eliminated completely. The BER is approaching theoretical
limits. Work can be done to further improve BER and
reliability. Work can be further extended in more general
Nakagami-m channels. Also performance can be
investigated in other novel equalizers.
REFERENCES
[1] A. John and C. Bingham, “Multicarrier modulation for data
transmission: An idea whose time has come,” IEEE Commun.Mag.,
vol. 28, no. 5, pp. 5-14, May 1990.
[2] Z. Wang, “OFDM or single carrier block transmission,” IEEE Trans.
on Comm., vol. 52, no. 3, pp. 380-394, Mar 2004.
[3] F. B. Frederiksen and R. Prasad, “An overview of OFDM and related
techniques towards development of future wireless multimedia
communications,” in Proc. IEEE Conf, Radio and Wireless
Communication (RWC 02), Aug. 2002, pp.19-22.
[4] A. Chandra, D. Biswas and C. Bose, “BER Performance of Coherent
PSK in Rayleigh Fading Channel with Imperfect Phase Estimation,”
in Proc. IEEE Conf, International Conference on Recent Trends in
Information, Telecommunication and Computing (ITC 10), pp.130-
134, March 2010.
[5] M. X. Chang and Y. T. Su, “Performance Analysis of Equalized
OFDM Systems in Rayleigh Fading,” IEEE Trans. on wireless
commun, vol. 1, no. 4, pp. 721-732, Oct. 2002.
[6] John Proakis, Digital communication, 4rd ed., vol. 1. Mcraw Hill:
2008.
[7] T. S. Rappaport, Wireless Communications, 2nd ed., vol.1. Pearson
Education, pp. 356–376, 2002.
[8] Y. Sato, “A method of self-recovering equalization for multilevel
amplitude modulation systems", IEEE Trans. on Communications, vol.
23, pp. 679-682, June 1975,
[9] A. Wallace, Martins and P. S. R. Diniz, “Suboptimal Lin ear MMSE
Equalizers With Minimum Redundancy,” IEEE signal process. Lett.,
vol. 17, no. 4, pp. 387-390, Apr. 2010.
[10] S. S. Sarnin, N. Kadri and A. Mahyuni, “Performance Analysis of
BPSK and QPSK Using Error Correcting Code through AWGN,” in
Proc. IEEE Conf, 2010 International Conference on Networking and
Information Technology (ICNIT 10), pp.178-182, June 2010 .
[11] B. Sklar, “Rayleigh Fading Channels in Mobile Digital
Communication Systems Part I: Characterization,” IEEE Comm. Mag.,
vol.35, no.7, pp. 90–100, July 1997.
[12] W. Lindsey, “ Error probabilities for Rician fading multichannel
reception of binary and N-ary signals,” IEEE Trans. Inf. Theory, Vol.
10, no. 4, pp. 339–35. 1964.

197