The BER Performance of OFDM Systems using Non-Synchroniz

Samphg
Thierry Pollet(a), Paul S p r ~ y t ‘and
~ ) Marc Moeneclaey(a)

(a) Communication Engineering Lab, University of Ghent (b) Alcatel Bell, Research Center
Sint I’ietersnieuwstraat 41, 9000 Ghent, Belgium Francis Wellesplein 1,2018 Antwerp, Belgium
tel: i--32-(0)926434 12,fax: i-32-(01926442 95 tel: +-32-(0)32407743,fax: +-32433240 99 32
E-mail: 7;c,iery.Po llet @ lei.rug.ac.be E-mail:pspr @rc.be 1. a lcateE. be

Abstract -In fully digital receivers, carrier and timing infor- this iigure, H(f) represents the transfer function of the
mation is derived from samples of the (anti-aliasing-filtered) transmission channel and r(t) is the (continuous-time) signal
received continuous-timesignal. In case of synchronized sam- at the input of the receiver.
pling, this information is used to align the sampling clock of In section 2, we calculate the received signal at. the output
the receiver with the remote transmit clock. In non-
of the FFT, given a frequency offset between transmitter
synchronized sampling systems, the sampling at the receiver
and receiver clock. In section 3, we compute for a given
is performed by means of a fixed free-running clock, and
additional post-processing is necessary to perform timing offset, the degradation of the BER, expressed in dB. Finally,
correction in the digital domain. In this paper, we investigate in section 4,we compare the performance of systems using
the effect of non-synchronized sampling on the BER perfor- synchronized and non-synchronized sampling.
mance of OFDM systems. We calculate the BER degradation
caused by a given frequency offset between receiver and
transmitter clock, as compared with the case of ideal sam-
pling. The obtained resulb arc compared with the perfor-
mance of synchronized sampling systems.
decision FFT

I. mTRODUCTION

In a digital OFDM (Orthogonal Frequency Division Mul-

tiplexing) receiver, the received continuous-time signal is
e-]........:.:.:.:...:.:./:.
..............
.............
:.:.qua;:;:;
............
:.:.:.:.:.:.::.
...........

sampled at instants determined by the receiver clock. At the

receiver, demodulation of the OFDM symbol consists of an
FFT (Fast Fourier Transform) operation performed on a Fig. la. OFDM receiver with VCXO (synchronizedsampling)
sequence of samples from the received OFDM symbol.
In synchronized sampling systems the timing algorithm
controls a VCXO (Voltage Controlled Crystal Oscillator)
in order to align the receiver clock wiih the transmitter clock
[11. In case of non-synchronized sampling, the sampling n (1)
rate remains fixed. Therefore, an offset between receiver
and transmitter clock rate will result in additional distorticn
of the signal at the outpt‘t of the FFT performed by the
receiver. The effect of a clock frequency offset is twofold. fusd
xla
First, the useful signal component is rotated and attenuated.
The rotation of the received symbol is compensated for in
the digital domain by a frequency domain equalizer
(ROTOR). In addition, the impairment gives rise to Iater
Carrier Interference (ICI), i.e. signal components caused by
carriers other than the considered carrier. Figure 1 depicts Fig. lb. OFDM receiver with fixed Crystal
the difference between the two synchronizing structures. In (non-synchronizedsampling)
253
0-7803-1820-XI94$4.00 0 1994 IEEE

Authorized licensed use limited to: Oxford University Libraries. Downloaded on March 31,2010 at 05:24:05 EDT from IEEE Xplore. Restrictions apply.
 11. THE OFDM RECEIVER Defining
-k k jZn-(mN+6)-
k fS

Essentially, an OF'DM signal is the sum of a large number am=rq,.e N fs+L?If

of QAM (or QPSK) modulated carriers (sub-channels).

The carrier spacing equals the OFDM symbol rate 1/T [2,3].
OFDM also has been referred to as discrete multitone
(DMT) [4]. In case the transmission channel characteristic
is slowly varying in time, the size of the QAM constellations
can be chosen according to the signal-to-noise ratio (SNR)
at each of the sub-channels, in order to obtain the same BER
performance for each of the QAM sub-channels [2,3].
During the m-th symbol period of duration T, the complex
envelope of the transmitted OFDM signal can be expressed
as
N-1
k=O
where a h denotes the m-th transmitted symbol on the carrier
k (at carrier frequency k/T) and N is the total number of
carriers. In a practical system, samples of the OFDM sym-
bol (1) are generated by means of an N-point FFT on the
symbols {ah), k E [O,N-11 [51.
At the receiver, the signal
+m
m-m
is sampled at a sampling rate fs+Af, where n(t) denotes the
(filtered) additive white Gaussian noise (AWGN), fs=N/T
and Af is the sampling frequency offset between receiver
and transmitter. The receiver performs an FFT on consecu-
tive blocks of N samples, in order to detect the OFDM
symbols.
Because of the frequency mismatch, the OFDM symbol
duration at the receiver, N/(fs + Af), differs from the one at
the transmitter, N/fs. Hence, symbol synchronization must
be performed by a timing algorithm. The algorithm ensures
that the blok of N samples is well aligned within the cor-
responding symbol period. This requires that, at regular
intervals, samples are robbed (if Af > 0) or stuffed (Af < 0).
Let us define 6k as the number of samples of the received
sample sequence to be robbed or stuffed during the recep-
tion of the k-th OFDM symbol. For each OFDM symbol k,
& can take the values { - 1,0,1}.The N consecutive samples
belonging to the m-th symbol that are fed to the FFT can be
expressed as
m.N+n+6
rk = r( ) n E [O,N-I]
fs+Af
where
we can rewrite ( 3 ) as
N-1
kn fs
k jZn-- (6)
rk=Ciim.e N fs+af +&
k=O
where n k denotes the AWGN contribution.
In reality, the transmitted signal will be distorted during
transmission over the channel. To avoid Inter Carrier Inter-
ference (ICI) and Inter Symbol Interference (ISI) at the
receiver, each symbol is extended with a cyclic prefix (i.e.
a repetition of the last samples) before transmission [6].At
the receiver, the prefix is removed before FFI'. It can be
shown [6] that the use of a cyclic prefix ensures or-
thogonality between the carriers and the results obtained
hereafter remain valid if in (6) a' k is replaced by a k.Hk,
where Hk is the FFT of the channel impulse response. The
guard time combined with an appropriate symbol timing
algorithm guarantees that all samples r% (n E [0, N-11)
belong to the signal transmitted during the m-th symbol
period. As a result, no IS1will appear. However, in order to
keep the expressions tractable, the notations do not reflect
the effect of a guard time.
The FFT operates on {rk}, n E [O,N-11, to produce Rh.
After some elementary calculations, one obtains (without
considering the AWGN contribution),
We observe that each output of the FFT,R ,: consists of
several contributions :
- The first term in (7) denotes the useful component: the
transmitted symbol, ,a: is rotated and attenuated. Both
effects depend on the considered carrier frequency n/T. The
(3) angle over which the symbol a k is rotated equals:
n fs
N [m.N+6]- f d A f + arg[I,,,].
8;=2x-
(4)
Remark that the angle 0: is proportional to the tone index
n and increases linearly for successive OFDM symbols as

254

Authorized licensed use limited to: Oxford University Libraries. Downloaded on March 31,2010 at 05:24:05 EDT from IEEE Xplore. Restrictions apply.
long as no samples are robbed or stuffed (8 = 0). Conse- (9)
quently, the received points of a QAM constellation are
\
rotating at a velocity depending linearly on the carrier index. where
The rotation is compensated for by means of a frequency
domain equalizer (ROTOR). The equalizer consists of one (10)
complex-valued tap for each FFT output and is updated
every OFDM symbol. Robbing or stuffing a sample (&I =
+1) results into an additional rotation over an angle ( A 0 3
equal to m.RESULTSAND DISCUSSION
jZn &E f,
A€&=e ~ f $ + ~ f
In a non-synchronized OFDM system, for a sufficiently
The ROTOR values should be corrected accordingly. large Es/No, we see from (8) that the degradation, Dn, for
The useful component is attenuated by a factor IIn,nl I1. the carrier n/T mainly results from ICI.
- The second term denotes the ICI, i.e. a contribution of
2.OEc5
signal components resulting from other carriers than the
considered one. This interference is considered as addition-
al noise. 1.5E+5

III. DEGRADATION
,i
'1.OE+5

The degradation of the BER performance, given a certain

BER, can be defined as the increase of SNR at the input of 5.OErl
the decision device needed to compensate for the impair-
ment. In case of a sampling frequency error, the effect of
0.0
this impairment depends on the considered carrier.
In absence of the impairment, the useful signal component 0 32 64 96 128 160 192 224 256

at the output of the FFT equals Hn. :a and the variance of carrier index (n)

the noise is EII nil2] = No, where No denotes the power
spectral density of the AWGN. In presence of a sampling
frequency offset, the useful component and the variance of
the noise become respectively
H, .Jn,n . a- n and No + var [ICI] .
Hence, the degradation Dn (in dB) for the QAM signal
modulating the carrier at frequency n/T, can be expressed
as
Fig. 2, Kn as function of the carrier index n
In order to evaluate Dn, figure 2 depicts Kn as a function
of the frequency index n. The total number of carriers (NI
was chosen to be 256. We observe that over a large area, Kn
is proportional to the square of the carrier index. Hence, if
the argument of the logarithmic function in (9)is close to 1,
the degradation is proportional to the square of the relative
frequency offset and to the square of the carrier index. We
can approximate expression (9) by (appendix B)

- N- I

k#n
except for the smallest and highest carrier indexes.
- 10 log (I In.n12) [dBI
10 (8) In figure 3, we depict the degradation as a function of the
relative frequency offset (expressed in ppm) for the tone
where Es denotes the average symbol energy per QAM
(249D) suffering the most from ICI. The OFDM signal
sub-carrier. The first term in (8) results from IC1 and is
consists of 256 modulated carriers. Figure 4 depicts, for
dominant for large WO.
different values of the relative frequency offset, the maxi-
Assuming an AWGN channel (IHkl=l, Vk), one obtains
mum number of bits per symbol to be transmitted on sub-
for small Af/fs (appendix A) carrier 249D as function of the E D 0 in order to obtain a
BER of [61.
255

Authorized licensed use limited to: Oxford University Libraries. Downloaded on March 31,2010 at 05:24:05 EDT from IEEE Xplore. Restrictions apply.
 100 frequency offset which is determined by the closed loop
bandwidth of the DPLL.
10
The received constellation points which serve as input to
,m-. the DPLL are obtained at the output of the FFT. As the F I 3
9
C
0
operates on a symbol per symbol basis, the VCXO is
._ 1
c
U
updated only once every OFDM symbol period. Hence, the
2
5
l
instantaneous frequency off set remains fixed during each
a,
n 0.1 OFDM symbol. Therefore, the degradation can be ex-
pressed by (9) where (Af)2 is replaced by E[(Af)2].
0.01
V. CONCLUSION
0 20 40 60 80 100

offset (ppm) We conclude that non-synchronous sampling systems are

Fig. 3. Degradation as function of the frequency offset (ppm)
Ln applications such as high speed transmission over the
unshielded twisted pair wire (ADSL and HDSL), the SNR
at each of the sub-carriels is known by the transmitter. The
size of the QAM constellation modulating each carrier
frequency is determined depending on the SNR at the con-
sidered carrier. This ensures the same BER performance for
each received QAM signal. The degradation due to a sam-
pling frequency error for the n-th carrier (Dn) can be ex-
pressed by (8) where No is substituted by the power density
of the noise at carrier n/T.
In an OFDM system with synchronized sampling, infor-
mation on the instantaneous frequency offset between
receiver and transmitter clock is derived from the rotation
of the received constellation points. This information is
used by the DPLL (Digital Phase Locked Loop) to control
the VCXO that determines the sampling instants. This
assures that the average frequency offset between receiver
and transmitter clock is zero. The degradation Dn then
depends on the variance of the instantaneous sampling
much more sensitive to a frequency offset between trans-
mitter and receiver clock, compared with synchronous sam-
pling systems. For non-synchronous sampling systems, we
showed that the degradation due to a frequency sampling
offset depends on the square of the carrier index and on the
square of the relative frequency offset.
ACKNOWLEDGEMENT
The first author expresses his gratitude to Alcatel Bell,
Antwerp, Belgium, for the financial support.
The third author gratefully acknowledges the support from
the Belgian National Fund for Scientific Research
(NFWO).
REFERENCES
[I] T. Jesupret, M. Moeneclaey, G . Ascheid, "Digital
Demodulator Synchronization," ESTEC Contract No.
8437/89/NL/RE, 1991
[2] I. Kalet, "The Multitone Channel," IEEE Trans. on Comm.,
vol. 37, no. 2, Feb. 1989,pp. 119-124.

[3] J.A. Bingham, "Multicarrier Modulation for Data Transmis-

sion: an Idea Whose Time has come," IEEE Communications
magazine, May 1990,pp. 5-14.

b 10- (41J.M. Cioffi, J.S. Chow, and J.T. Aslanis, "Detailed DMT
'S.-3 Transmitter Description for ADSL," ANSI TlEl.4/93-084,
.a April 15,1993.
c

2 5-
[5] S.B. Weinstein, P.M. Ebert, "Data Transmission by Frequen-
cy-Division Multiplexing Using the Discrete Fourier Trans-
A form," IEEE Trans. on Comm., vol COM-19, no. 5, Oct. 1971.
o , " " , ' " " " ' ~ ' " ~ " ' ' ~
[6] J.S. Chow, J.C. Tu and J.M. Cioffi, "A Discrete Multitone
Transceiver for HDSL Applications, "IEEE J. on Selected
Areas in Communications,vol. 9, no. 6, pp. 895-908, August
Fig. 4. Bit assignment as function of EJNo 1991.
256

Authorized licensed use limited to: Oxford University Libraries. Downloaded on March 31,2010 at 05:24:05 EDT from IEEE Xplore. Restrictions apply.
 APPENDIXA

The variance of the IC1 can be expressed as

N-1 sin2 (x(k.-- fs n)) N-1

1 fdAf N-v V sin(~cv0)
=Es-z = 1+ 2 - cos(~c-(0(N-l)-2 n) ____
N2 k=O sin2(&(k --fs N2 N Jc
n)) v= 1 sin(-vO)
kzn N 'fs+Af N

For small Af/f,, this expression can be approximated by For a large range of values of n, n E [O,N-11, and 8 = 1,the
above expression becomes
Af
N-1 sin2(x(k-n)-JckF) N-I
var[ICI]= Es -
1
1 2Jc
1s

x Af
N2 k=o sin (-(k-nk-k-)
k#n N fs

The second term can be expressed, using Taylor expansion,

N-1
Af
sin2 (Jck-) as
fS

N2 k=O sin2(%k-n))
km Af
sin(x.n.---)
N-1 2 1 fs+Af
lIn,nl =- Af
N2 sin(%.-)
N' fs+Af

One observes that the variance depends on the carrier index

n.

APPENDLX B Hence,

Expression (11) can be obtained from (8) (for IHkl=l)

considering that
kzn

For small Af/fs, the degradation due to the attenuation of

the useful signal component and the interference from IC1
km become
The first term of this expression can be written as - 10 loglo(l - f(Jc.n.f)2) and 10 log (1+- 1-& (Jc.n.3
A f 2)
10 3 NO fs

If the arguments of the logarithmic functions are close to

1,the degradations can further be expressed as
N-1 N-1
=-E 1
1 (N-IVI) e
j2nv(k.&n)
N lo
~- Af
(x.n.-) 2 and 10 1 Es (Jc.n.-)Af 2
N2 k=O v=-(N-l) In10 3 fs h l 0 3 NO fs

fS respectively.
where 0 = - It is obvious that when E 2 N o is large, the degradation
f s+Af
caused by IC1 is dominant.

257

Authorized licensed use limited to: Oxford University Libraries. Downloaded on March 31,2010 at 05:24:05 EDT from IEEE Xplore. Restrictions apply.