Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
7, JULY 2009
Abstract—Most OFDM analyses assume idealized conditions assumption of a CP longer than the channel delay spread
when it comes to the waveform at the receiver in terms of time is used in studies ranging from multi-antenna techniques
and frequency synchronization. In addition, often a cyclic prefix [2][5][6] to channel estimation [4].
(CP) longer than the channel delay spread is assumed, and thus
the impact of the possible inter-symbol interference (ISI) from Performance of OFDM systems with various imperfections
adjacent OFDM symbols is not a consideration. have been considered in references such as [7-24]. References
This work provides a consolidated framework for study and [7-10] analyze the impact of inter-carrier interference (ICI)
analysis of OFDM systems with imperfections. The imperfections
due to a time-varying channel within the OFDM symbol time
include time and frequency synchronization errors, and channel
delay spread beyond the CP duration. Transmit as well as receive span, while [11] studies the impact of ICI due to frequency
filtering operations are explicitly modeled, and different receiver error. The impact of a channel delay spread longer than the
detection criteria are considered. CP has been studied in [12-24].
The use of an OFDM waveform for UWB applications is
known to be a promising choice to simplify the receiver design Reference [12] derives the carrier-to-interference ratio (C/I)
and to maximize the bandwidth scalability, as well as the at each subcarrier, accounting for the effects of ICI and ISI
adaptation to various spectrum allocations. We present bit error due to a channel delay spread longer than the CP and to
rate (BER) sensitivity analysis to key system parameters for residual frequency error. The authors of [12] advocate the use
OFDM-based UWB applications. From these sensitivity analyses, of channel coding and adaptive equalization to compensate
we see graceful performance degradation with decreasing CP
length and increasing residual frequency error. The performance the effects of ICI/ISI. In [13], the power of the ICI/ISI
degradation can be much more critical for timing synchroniza- term is derived, and a sensitivity analysis of the CP length
tion errors. is performed for a channel model corresponding to a hilly
Index Terms—OFDM, UWB, ICI, ISI. environment. In [14], a method to cancel the residual ISI is
presented to alleviate the performance degradation at demod-
ulation. Reference [15] studies the combined impact of the
I. I NTRODUCTION channel variation within the OFDM symbol time span and
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:41 from IEEE Xplore. Restrictions apply.
MONTOJO and MILSTEIN: EFFECTS OF IMPERFECTIONS ON THE PERFORMANCE OF OFDM SYSTEMS 2061
ˆ
To the best of our knowledge, the only publication that char- 2 cos(2 fˆ0t ) sampler
from
acterizes the ICI and ISI terms due to a channel delay spread channel Cyclic
hrx t Prefix
larger than the CP and imperfect synchronization is [24]. In Removal
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:41 from IEEE Xplore. Restrictions apply.
2062 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 57, NO. 7, JULY 2009
n k=0 −∞ +ynoise
RxF
(n Tf + k Tc + τ ) (9)
−j2πΔf u
·e · htx−ch
lp (t − u − kTc − nTf ) · du. (5) where k ∈ [0, NT − 1]. Note that we use again the dis-
crete representation of signals with two arguments, n and
Given that the receive filter has a bandwidth on the k, representing time indexing OFDM blocks and samples
order of N/T = 1/Tc = BWsignal , then the relevant therein, respectively. Since the time span of the effective
range for u in (5) is |u| ≤ ϑTc , where ϑ is an in- channel impulse response, h(t), is, by design, much smaller
teger number meant to capture the number of Tc inter- than the OFDM block duration, when receiving the OFDM
∞
vals so that −∞ hrx (u)e−j2πΔf u htx−ch (t − u − kTc − block with index n , only three blocks can contribute to the
ϑTc rx lp
∼
nTf )du = −ϑTc h (u)e −j2πΔf u tx−ch
hlp (t − u − kTc − final test statistic. These are the block with index n itself,
nTf )du. Looking at the product (Δf u), we see that the block preceding it and the block following it. Therefore,
(Δf ϑTc ) = (Δf ϑ/BWsignal ). Since, for all practical pur- the summation in n above is only relevant for the values
poses, BWsignal >> (Δf ϑ), we have (Δf ϑ)/BWsignal << n = n , n = (n − 1), n = (n + 1), and hence
1, and thus e−j2πΔf u ∼= 1 for the values of u that are relevant RxF,(n) RxF,(n−1)
y RxF [n, k] = ysignal [n, k] + ysignal [n, k]
to the integral. Note that the same approximation is made in
RxF,(n+1)
[7]. Therefore, +ysignal [n, k] + ynoise
RxF
[n, k] (10)
T −1
N where we have defined
RxF
ysignal (t) ∼
= ej2πΔf t u[n, k]
ysignal [n, k] ej2πΔf (nTf +kTc +τ )
RxF,(n)
n
∞
k=0
T −1
N
hrx (u) · htx−ch
lp (t − u − kTc − nTf ) · du (6) · u[n, ξ] · h((k − ξ)Tc + τ ), (11)
−∞
ξ=0
h(t−kTc −nTf )
[n, k] ej2πΔf (nTf +kTc +τ )
RxF,(n−1)
where we have introduced the effective channel impulse ysignal
response (incorporating the transmit and receive filtering op- T −1
N
erations), defined as h(t) hrx (t) ∗ htx−ch (t) = hrx (t) ∗ · u[n − 1, ξ] · h(Tf + (k − ξ)Tc + τ ) (12)
lp
hlp (t) ∗ h (t).
ch tx ξ=0
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:41 from IEEE Xplore. Restrictions apply.
MONTOJO and MILSTEIN: EFFECTS OF IMPERFECTIONS ON THE PERFORMANCE OF OFDM SYSTEMS 2063
can be identified to be
αn
N/2−1
[n, k] = ej2πΔf (nTf +kTc +τ )
RxF,(n−1) CP,(n−1)
ysignal ysignal [n, k] = √ s[n − 1, m] · ej2πmCP/N
⎧ LU N
ν=LL u[n − 1, k − ν + NT ]h(νTc + τ ),
m=−N/2
⎪
⎪
⎨
LU
for k = 0 . . . LL − 1 if LL > 0
· LU (16) ·ej2π(m+Δf )k/N h(νTc + τ )e−j2πmν/N (19)
⎪
⎪ ν=k+1 u[n − 1, k − ν + NT ]h(νTc + τ ),
⎩ ν=k+CP +1
for k = max(0, LL ) . . . LU − 1
which is non-zero only if LU > CP and applies for k =
and 0 . . . (LU − CP − 1), and
αn
N/2−1
[n, k] = ej2πΔf (nTf +kTc +τ ) s[n + 1, m] · e−j2πmCP/N
RxF,(n+1) CP,(n+1)
ysignal ysignal [n, k] = √
N
k−N m=−N/2
· u[n + 1, k − ν − NT ] · h(νTc + τ ) (17)
k−N
ν=LL ·ej2π(m+Δf)k/N h(νTc + τ )e−j2πmν/N , (20)
ν=LL
for k = NT + LL . . . NT − 1 if LL < 0. which is non-zero only if LL < 0 and applies for k = (N +
The expressions in (15)-(17) show the non-zero contribu- LL ) . . . (N − 1), respectively.
tions of the different terms at each sample point within the The noise term, after discarding the CP, becomes
OFDM symbol (index k). Note that a non-zero contribution
of (17), which is the term representing the ISI from the
CP
ynoise [n, k] = ynoise
RxF
[n, k + CP ] for k = 0 . . . (N − 1). (21)
next OFDM symbol, requires LL < 0. This is equivalent to
Therefore, the overall DFT input is y CP [n, k] =
τ ≥ (Dmin + Tc ) and hence a "late sampling". We will see CP,(n) CP,(n−1) CP,(n+1)
ysignal [n, k]+ysignal [n, k]+ysignal [n, k]+ynoise
CP
[n, k],
now how some of the contributions in (16), which is the term
and we define the DFT output to be
representing ISI from the previous OFDM symbol, vanish after
discarding the CP. N −1
1 CP
After discarding the CP, we obtain the following signal- y DF T [n, l] = √ y [n, k] · e−j2πlk/N , (22)
N k=0
dependent term depending on the current OFDM symbol:
for l = −N/2 . . . (N/2 − 1). Since the DFT is a linear
CP,(n)
ysignal [n, k] = (18) operation, we may characterize the DFT output y DF T [n, l]
⎧ α N/2−1 j2π(m+Δf )k/N
in terms of the DFTs of each of the terms in y CP [n, k]. Using
⎪ m=−N/2 s[n, m] · e
√n
⎪
⎪
N the same notation as before, the DFT output dependent on the
⎪
⎪
⎪
⎪
k+CP
· ν=L h(νTc + τ ) · e−j2πmν/N , for k = 0... current OFDM symbol is given by
⎪
⎪
L
⎪
⎪ (LU − CP − 1) if LU > CP
⎪
⎨ N/2−1 j2π(m+Δf )k/N αn
N/2−1
m=−N/2 s[n, m] · e
DF T,(n)
αn
√ H(m), ysignal [n, l] = s[n, m]·
N N
⎪
⎪ for k = max(0, LU ) . . . min(N + LL − 1, N − 1) m=−N/2
⎪
⎪ N/2−1
⎪
⎪ j2π(m+Δf )k/N if LU >CP
m=−N/2 s[n, m] · e
αn
⎪ √
⎪
⎪ N
LU −CP −1
⎪
⎪ LU
· ν=k−(N h(νTc + τ ) · e−j2πmν/N , for k =
k+CP
⎪
⎩ T −1) e j2π(Δf +m−l)k/N
h(νTc + τ ) · e−j2πmν/N
(N + LL ) . . . (N − 1) and if LL < 0 k=0 ν=LL
min(N−1,N+LL −1)
where we have defined αn ej2πΔf (nTf +CP ·Tc +τ ) , the + ej2π(Δf +m−l)k/N H(m)+ (23)
k=max(0,LU −CP )
normalized frequency error as Δf Δf /(1/T ), which is the
residual frequency error normalized by the subcarrier
N−1
LU
U spacing, ej2π(Δf +m−l)k/N h(νTc + τ ) · e−j2πmν/N
and the channel frequency response H(m) L i=LL h(τ + k=N+LL ν=k−(N−1)
−j2πmi/N
iTc )e . Note that the first term in (18) is non-zero if
if LL <0
LU > CP , or, equivalently, if (Dmax − τ ) > (CP · Tc ). This
is the condition for ISI from the previous OFDM symbol after The DFT output that depends on the previous OFDM symbol
discarding the CP as can be seen from realizing that it is the is nonzero if the CP duration is smaller than the channel delay
condition for the term (19) to be non-zero. Similarly, the last spread minus the timing error, i.e., if CP < LU , in which case
term in (18) is non-zero if LL < 0, which is equivalent to
αn
N/2−1
τ ≥ (Dmin + Tc ), as seen before. This is the condition for ISI DF T,(n−1)
ysignal [n, l] = s[n − 1, m] · ej2πm·CP /N (24)
from the next OFDM symbol, as can be seen from realizing N
m=−N/2
that is the condition for the term (20) to be non-zero. Further, LU −CP −1
LU
note that discarding the CP does not remove the contribution · ej2π(Δf +m−l)k/N h(νTc + τ )e−j2πmν/N .
from the next OFDM symbol. k=0 ν=k+CP +1
Similarly, the signal-dependent terms depending on the The DFT output depending on the next OFDM symbol is
previous and next OFDM symbols after discarding the CP nonzero if the timing error, τ ≥ (Dmin + Tc ) , in which
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:41 from IEEE Xplore. Restrictions apply.
2064 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 57, NO. 7, JULY 2009
case where we can clearly identify the ICI and ISI terms at the
N/2−1 DFT output.
αn
s[n + 1, m] · e−j2πm·CP/N
DF T,(n+1)
ysignal [n, l] = For the data detection process, it is convenient to put
N (32) in matrix form. Matrices are represented by boldface
m=−N/2
N −1
k−N capital letters, e.g., A, and vectors are represented by boldface
· ej2π(Δf +m−l)k/N h(νTc + τ )e−j2πmν/N (.25) lowercase letters, e.g., v. Therefore, we define the matrices
k=N +LL ν=LL
Ψx [ζx [l, m]]m,l=−N/2,...,(N/2−1) (33)
The noise component at the DFT output may be written as
N −1
1 CP with x = 1, 2, 3, 4, 5. Note that, from (32), Ψ1 is non-zero
DF T
ynoise [n, l] = √ ynoise [n, k] · e−j2πlk/N (26) only if there is ISI from the previous OFDM symbol, Ψ2 is
N k=0
the matrix characterizing the useful signal term as well as the
where ynoise
CP
[n, k] is defined in (21). To shorten notation, we ICI term due to frequency error, and Ψ3 is non-zero only if
can define the following variables: there is ISI from the next OFDM symbol. Further, note that
(LU −CP −1) Ψ4 and Ψ5 fully characterize the ISI from the previous and
ζ1 (l, m) e−j2π(l−Δf −m)k/N next OFDM symbol, respectively. With the matrices defined
k=0
in (33), we can write (32) in matrix form as follows:
(k+CP )
A(n) Apre (n)
· h(τ + iTc )e−j2πmi/N (27)
αn αn
i=LL ysignal (n) =
DF T
(Ψ1 + Ψ2 + Ψ3 ) ·s(n) + Ψ4 ·s(n − 1)
N N
αn
ζ2 (l, m) H(m) + Ψ5 ·s(n + 1) (34)
min(N +LL −1,N −1)
N
Anext (n)
· e−j2π(l−Δf −m)k/N (28)
k=max(LU −CP,0) where s(n), s(n − 1) and s(n + 1) are column vectors of
length N with information symbols at the nth , (n − 1)th and
−1
N
(n + 1)th OFDM symbols, respectively, and where we have
ζ3 (l, m) e−j2π(l−Δf −m)k/N defined the channel state information matrices, A(n), Apre (n)
k=N +LL and Anext (n). Note that there is no “time” dependency in
LU
either ζx (l, m) or Ψx because we assume a particular channel
· h(τ + iTc)e−j2πmi/N (29) realization that remains constant over the OFDM symbol
i=k−N +1
duration. This condition requires a channel coherence time
(LU −CP −1) Tcoh >> T or, equivalently, 1/Tcoh << 1/T , which means
ζ4 (l, m) ej2πmCP/N e−j2π(l−Δf −m)k/N that the subcarrier spacing (1/T ) is much greater than the
k=0
Doppler spread (1/Tcoh ∼ = fd ). The time dependency in the
LU newly defined matrices A(n), Apre (n) and Anext (n) comes
· h(τ + iTc )e−j2πmi/N (30) from the time-varying phasor αn and we will drop it from the
i=k+CP +1 development to shorten notation.
Overall, at the DFT output, we have
N −1
ζ5 (l, m) e−j2πmCP/N e−j2π(l−Δf −m)k/N yDF T (n) = As(n) +
k=N +LL
DT F
yusef ul and ICI (n)
k−N
· h(τ + iTc )e−j2πmi/N (31) Apre s(n − 1) + Anext s(n + 1) + w(n) (35)
i=LL DT F (n)
yISI DT F (n)
ynoise
Therefore, the signal-dependent output of the DFT may be
written as When there is no ISI, the expression in (35) reduces to
if LU >CP yDF T (n) = A · s(n) + w(n). The condition for no ISI
αn
N/2−1 can be seen from observation of the terms in (32), i.e.,
DF T,(n)
ysignal [n, l] = s[n, m] ζ1 (l, m) (32)
N LU ≤ CP and LL ≥ 0. For this case, the matrix Ψ2 is
m=−N/2
the only non-zero matrix, and the term ζ2 (l, m) takes the
if LU >CP N −1
form ζ2 (l, m) = H(m) k=0 e−j2π(l−Δf −m)k/N . Therefore,
if LL <0
α
N/2−1 the matrix Ψ2 canbe written as Ψ2 = Δ · DH , where
n N −1
+ζ2 (l, m) + ζ3 (l, m) + s[n − 1, m] · ζ4 (l, m) Δm+N/2,l+N/2 k=0 ej2π(l−Δf −m)k/N , and
N
m=−N/2 ⎡ ⎤
H(−N/2) 0 ··· 0
αn
N/2−1
⎢ ⎥
+ s[n + 1, m] · ζ5 (l, m) ⎢ 0 H(−N/2 + 1) · · · 0 ⎥
N DH ⎢ .. .. .. ⎥
m=−N/2 ⎣ . . ··· . ⎦
if LL <0 0 0 · · · H(N/2 − 1)
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:41 from IEEE Xplore. Restrictions apply.
MONTOJO and MILSTEIN: EFFECTS OF IMPERFECTIONS ON THE PERFORMANCE OF OFDM SYSTEMS 2065
Since Δ is circulant, it can be shown that it can be decom- of different OFDM symbols, and hence, independence among
posed as Δ = N F · Mφ · FH , where the ICI and the two possible ISI terms.
⎡ ⎤ Defining Rs (n) E{s(n)sH (n)} and Rw (n)
1 0 ··· 0
⎢ 0 ej2πΔf /N · · · ⎥ E{w(n)wH (n)}, we can express the SINR of the lth sub-
⎢ 0 ⎥
Mφ ⎢ . ⎥ carrier at the data detector output as
.. .. ..
⎣ .. . . . ⎦
|[B]l,l |2 E{|s(n, l)|2 }
0 0 · · · ej2πΔf (N −1)/N γ(n, l) , (44)
d2 (n, l)
and F is the Fourier matrix of size N × N . As a result, for
where
the ISI-free case,
αn d2 (n, l) [BRs (n)BH ]l,l − 2{[BRs (n)]l,l [BH ]l,l }
yDF T (n) · Δ · DH · s(n) + w(n). (36)
N +|[B]l,l |2 E{|s(n, l)|2 } + [Bpre Rs (n − 1)BH
pre ]l,l
Further, if there is no residual frequency error, the matrix Δ
+[Bnext Rs (n + 1)BH
next ]l,l + [MRw (n)M ]l,l
H
(45)
becomes the identity matrix multiplied by N and αn = 1.
Therefore, (36) becomes Note that the expressions in (44) and (45) are valid for any
y DF T
(n) DH · s(n) + w(n) (37) linear receiver, and will be specialized below for different
receiver criteria.
which is the traditional set of N parallel flat fading channels 2
For the special case of white noise, i.e., Rw (n) = σw IN
that OFDM systems are regularly associated with. and Rs (n) = Rs (n − 1) = Rs (n + 1) = IN , (44) takes the
following form:
III. DATA D ETECTION
|[B]l,l |2
We assume linear processing at the receiver, so that γ(n, l) (46)
d3 (n, l)
z(n) = M · yDF T (n) (38) with
where the matrix M will take different forms depending on
d3 (n, l) [BBH ]l,l − |[B]l,l |2 + [Bpre BH
pre ]l,l
the data detector type. Applying (35) to (38) yields 2
+[Bnext BH
next ]l,l + σw [MM ]l,l
H
(47)
z(n) = MAs(n) + MApres(n − 1) + MAnext s(n + 1) + Mw(n)
(39)
Denoting B MA, Bpre MApre and Bnext MAnext , A. Data Detection with Single-Tap Equalizer (ST)
we can re-write (39) as
For the ST receiver, the detection operation is given by
z(n) = Bs(n) + Bpre s(n − 1) + Bnext s(n + 1) + Mw(n),
(40) zST (n) = Diag(AH )yDF T (n) (48)
which we can express as z(n) = zsignal (n) + w (n), and Note that this receiver does not attempt to equalize the
where zsignal (n) Bs(n)+Bpre s(n−1)+Bnexts(n+1) and received signal and only the useful part of the received
w (n) Mw(n). Clearly, the first summand in (40) contains signal, i.e., Diag(A), is considered, while ignoring the ICI,
the desired term as well as the ICI term, and therefore we can ISI and noise terms. For this case, B = Diag(AH )A,
re-write it as Bpre = Diag(AH )Apre and Bnext = Diag(AH )Anext , and
z(n) = Diag(B)s(n) + Diag(B)s(n) + Bpre s(n − 1) (44) and (45) take the following forms, respectively:
+Bnext s(n + 1) + Mw(n) (41) |[Diag(AH )A]l,l |2 E{|s(n, l)|2 }
γST (n, l) , (49)
where Diag(A) represents a diagonal matrix containing only dST (n, l)
the diagonal elements of the matrix A and Diag(A) =
A−Diag(A) represents a matrix with zero diagonal elements dST (n, l) [Diag(AH )ARs (n)AH Diag(A)]l,l (50)
containing all but the diagonal elements of A. −2{[Diag(AH )ARs (n)]l,l [AH Diag(A)]l,l }
By inspection of (41), we can obtain the SINR of the lth
+|[Diag(AH )A]l,l |2 E{|s(n, l)|2 }
subcarrier at the data detector output as
+[Diag(AH )Apre Rs (n − 1)AH pre A)]l,l
E{|[Diag(B)s(n)]l |2 }
γ(n, l) (42) +[Diag(AH )Anext Rs (n + 1)AH
next A)]l,l
d1 (n, l)
+[MRw (n)MH ]l,l
where
d1 (n, l) E{|[Diag(B)s(n)]l |2 } + E{|[Bpre s(n − 1)]l |2 }
B. Data Detection with Zero-Forcing Equalizer (ZF)
+E{|[Bnext s(n + 1)]l |2 } + E{|[Mw(n)]l |2 } (43)
For the ZF receiver, the detection operation is given by
and where the expectations are done over the random noise and
the random information symbols. Note that we have assumed zZF (n) Rs (n)AH ARs (n)AH + Apre Rs (n − 1)AH pre
independence between the noise and the signal terms, and −1
we have assumed independence of the modulation symbols +Anext Rs (n + 1)AH next · yDF T (n) (51)
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:41 from IEEE Xplore. Restrictions apply.
2066 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 57, NO. 7, JULY 2009
This receiver attempts to suppress the signal-dependent inter- IV. P ERFORMANCE C HARACTERIZATION
ference, i.e., ICI and ISI. Assuming Rs (n) = Rs (n − 1) =
Rs (n + 1) = IN , (51) becomes Let us express the detector output for the lth subcarrier from
(40) as
−1
zZF (n) AH AAH + Apre AH pre + Anext Anext
H
N/2−1
−1 −1
+ [Bpre ]l,m · s(n − 1, m)
+[(A A)
H
A Rw A(A A)
H H
]l,l (55) m=−N/2
N/2−1
For the ISI-free case, A = αn /N ΔDH , as can + [Bnext ]l,m · s(n + 1, m) + w (n, l) (59)
be seen from Equation (36), and therefore AH A = m=−N/2
1/N 2 DH H ΔH ΔDH . Using the expression for Δ derived in which constitutes the test statistic for a given receiver. Defining
Section II, we can further simplify AH A to yield AH A = the signal amplitude as
DH H F(Mφ )H FH FMφ FH DH = DH H DH , and therefore
AH A = (56) αRxT
(n,l)
ype
(A, Apre , Anext , sl (n), s(n − 1), s(n + 1)) [[B]l,l
⎡ ⎤
|H(−N/2)|2 0 ··· 0
N/2−1
⎢ 0 |H(−N/2 + 1)|2 ··· 0 ⎥ + [B]l,m · s(n, m) (60)
⎢ ⎥
⎢ .. .. .. .. ⎥.
⎣ . . . . ⎦ m=−N/2,m=l
noise contributions.
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:41 from IEEE Xplore. Restrictions apply.
MONTOJO and MILSTEIN: EFFECTS OF IMPERFECTIONS ON THE PERFORMANCE OF OFDM SYSTEMS 2067
0
10 CM1 ensemble performance
0
10
CP0
CP16
CP32
CP48
−1
10
−1
10
BER
BER
−2
10
−2
10 Test stat − no ISI. Eq. (63)
Gauss. approx − no ISI
Simulated − no ISI 10
−3
Figure 2 shows
the average BER, 10
−2
pBP
n
SK
(error|A, Apre , Anext ) for two channel realizations,
one incurring ISI and the other not incurring ISI. A residual
frequency error of 5% of the subcarrier frequency spacing −3
10
producing ICI is assumed in both cases. We further assume
perfect timing, i.e., τ = 0, and data detection using the ST
receiver. The average BER is obtained from the test statistics
(Equation (63)), from simulations and from approximating 10
−4
−10 −5 0 5 10 15 20 25 30
the interference component of the final test statistic as a SNR (dB)
Gaussian random variable. The BER assuming Gaussian
interference
√ uses the SINR expressions in Section III and
Fig. 4. BER sensitivity to CP duration for CM2.
pe = Q( 2γ) is plotted. Note that the number of subcarriers
for the channel realization not incurring ISI was chosen to
be N=16, and for the channel realization incurring ISI was 0% roll-off and bandwidth 528MHz. Note that CP durations of
chosen to be N=8. Further, note that, despite the low value 0, 16, 32, 48, 64, and 80 samples constitute 0%, 11.11%, 20%,
of N chosen, the Gaussian approximation, as also found in 27.27%, 33.33%, and 38.46% fixed overhead. The results
[8], provides a good match with simulations and the exact shown in Figures 3-6 assume perfect time and frequency
prediction using the test statistic, and therefore, will be used synchronization, an ST receiver, and result from averaging
in the sequel. the performance over 1000 channel realizations. Note that for
Similar to other studies [8][12][17] for certain channel each of the channel models, at some point a longer CP does
models, Figures 3-6 show the sensitivity to the CP duration not improve the performance; this is the point where the ISI
of an ensemble of 1000 channel realizations for the UWB is not significant.
channel models CM1, CM2, CM3, and CM4 defined in Figure 7 shows the sensitivity to frequency errors for a given
[25]. CM1 is based on (0-4m) line-of-sight (LOS) channel CM1 channel realization. The results are obtained assuming a
measurements with a root-mean-square (RMS) delay spread system with N=128, 20% CP overhead, perfect time synchro-
of 5ns. CM2 is based on (0-4m) non-line-of-sight (NLOS) nization, a Nyquist filter with 0% roll-off and a bandwidth of
channel measurements with an RMS delay spread of 8ns. CM3 528MHz, and an ST receiver. This figure contains analytical
is based on (4-10m) NLOS channel measurements with an (distinguished by lines with inserted characters, e.g., ’-*-’)
RMS delay spread of 15ns. Finally, CM4 was generated to fit and simulation based (distinguished by just characters, e.g.,
a 25ns RMS delay spread. ’*’) results for frequency errors ranging from 0 to 5% of the
All the results assume a system with a subcarrier spacing of subcarrier frequency spacing. Assuming a ±20ppm frequency
4.125MHz, N=128 subcarriers [1], and a Nyquist filter with tolerance at the transmitter and at the receiver, the maximum
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:41 from IEEE Xplore. Restrictions apply.
2068 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 57, NO. 7, JULY 2009
−3
10
BER
BER
−2 −4
10 10
−5
10
−3 −6
10 10
−7
10
−4 −8
10 10
−10 −5 0 5 10 15 20 25 30 −10 −5 0 5 10 15 20 25 30
SNR (dB) SNR (dB)
Fig. 5. BER sensitivity to CP duration for CM3. Fig. 7. BER sensitivity to residual frequency error (analysis and simulation).
−1
10
−1 −2
10 10
−3
10
BER
BER
−2 −4
10 10
τ=0
10
−5 τ = −10T
CP0 c
CP16 τ = 4T
c
−3 −6
10 CP32 10 τ = 8T
c
CP48
τ = 12T
CP64 10
−7 c
CP80 τ = 16T
c
−4 −8
10 10
−10 −5 0 5 10 15 20 25 30 −10 −5 0 5 10 15 20 25 30
SNR (dB) SNR (dB)
Fig. 6. BER sensitivity to CP duration for CM4. Fig. 8. BER sensitivity to timing error (analysis and simulation).
frequency error incurred would be ±40ppm, which for a 5GHz to that of a shortened CP. Unlike the frequency errors, the
carrier frequency and 4.125MHz subcarrier spacing equates performance degradation for a positive timing error may be
to a frequency error of 5% of the subcarrier spacing. Note quite abrupt, as seen in Figure 8. The performance degradation
that the frequency error in actual systems would typically be of a negative timing error is much more graceful, as it is
compensated by some frequency compensation mechanism, equivalent to a shortened CP. Figure 9 illustrates this fact
so that these results show the performance at different levels graphically, showing the impact of timing error as it affects
of residual frequency error. From the results, we can see a ISI.
graceful performance degradation as frequency errors increase. Figure 10 shows the sensitivity to the data detection cri-
Figure 8 shows the sensitivity to timing errors for a given terion (receiver type) for a given CM4 channel realization
CM1 channel realization. The same assumptions as for the incurring ISI. We keep the same hypothesis as for Figures
results in Figure 7 are made, with the exception that now we 7 and 8 in terms of number of subcarriers, subcarrier spacing,
assume perfect frequency synchronization in order to concen- filtering, and CP overhead. The results shown assume a 1%
trate on the performance impact of varying the transmit/receive frequency error. However, the interference is dominated by
timing error, τ . A positive value of τ means that the OFDM the ISI/ICI incurred from a channel delay spread larger than
symbol boundary at the receiver is late with respect to the the CP duration. We compare the BER for the data detection
actual symbol boundary, and therefore inevitably there will criteria presented in Section III. The receiver type 1 (ST)
be ISI from the next OFDM symbol. A negative value of τ ignores the ICI, ISI and noise terms, however, there is no
advances the OFDM symbol boundary, but since the CP is interference or noise enhancement at the detector output. The
preceding the OFDM symbol, the actual effect is equivalent receiver type 2 (ZF) tries to minimize the signal-dependent
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:41 from IEEE Xplore. Restrictions apply.
MONTOJO and MILSTEIN: EFFECTS OF IMPERFECTIONS ON THE PERFORMANCE OF OFDM SYSTEMS 2069
R EFERENCES
[1] A. Batra, J. Balakrishnan, G. R. Aiello, J. R. Foerster, and A. Dabak,
“Design of a multiband OFDM system for realistic UWB channel
environments," IEEE Trans. Microw. Theory Tech., vol. 52, no. 9, pp.
2123-2138, Sept. 2004.
[2] W. P. Siriwongpairat, W. Su, M. Olfat, and K. J. Ray Liu, “Multiband-
OFDM MIMO coding framework for UWB communication systems,"
IEEE Trans. Signal Processing, vol. 54, no. 1, Jan. 2006.
[3] W. P. Siriwongpairat, W. Su, and K. J. Ray Liu, “Characterizing
performance of multiband UWB systems using poisson cluster arriving
fading paths," IEEE J. Select. Areas Commun., Special Issue on Ultra-
Wideband Commun., vol. 24, no. 4, pp. 745-751, Apr. 2006.
[4] X. Fan, B. Leng, and G. Bi, “An improved channel estimation algorithm
Fig. 9. ISI impact of transmit/receive timing error. for OFDM UWB," in Proc. 2005 International Conf. Wireless Commun.,
Networking Mobile Computing, vol. 1, pp. 173-176, Sept. 2005.
Sensitivity to Receiver type [5] J. Wang, G. Zhu, and J. Jin, “Optimal power allocation for space-
10
0
time coded OFDM UWB systems," in Proc. 2005 International Conf.
Rx type 1 Wireless Commun., Networking Mobile Computing, vol. 1, pp. 189-192,
Rx type 2 Sept. 2005.
Rx type 3 [6] W. Chan, W. Wu, C. Wang, M. Chiu, and C. Chao, “Application
Rx type 4 of space-time codes to OFDM UWB systems with under-sampled
receivers," in Proc. IEEE VTC 2005/Spring, vol. 2, pp. 1191-1195, May
10
−1 2005.
[7] M. Russel and G. L. Stüber, “Interchannel interference analysis of
OFDM in a mobile environment," in Proc. IEEE VTC-1995, vol. 2,
July 1995, pp. 820-824.
BER
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:41 from IEEE Xplore. Restrictions apply.
2070 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 57, NO. 7, JULY 2009
[23] M. Batariere, K. Baum, and T. P. Krauss, “Cyclic prefix length analysis Laurcence B. Milstein (S66, M68, SM77, F85)
for 4G OFDM systems," in Proc. IEEE VTC-2004/Fall, vol. 1, Sept. received the B.E.E. degree from the City College
2004, pp. 543-547. of New York, New York, NY, in 1964, and the M.S.
[24] H. Q. Lai, W. P. Siriwongpairat, and K. J Ray Liu, “Performance analysis and Ph.D. degrees in electrical engineering from the
of multiband OFDM UWB system with imperfect synchronization and Polytechnic Institute of Brooklyn, Brooklyn, NY, in
intersymbol interference," IEEE J. Select. Topic Signal Processing, 1966 and 1968, respectively.
Special Issue Performance Limits Ultra-Wideband Systems, vol. 1, no. From 1968 to 1974, he was with the Space and
3, pp. 521-534, Oct. 2007. Communications Group of Hughes Aircraft Com-
[25] IEEE P802.15-02/490r1-SG3a, “Channel modeling sub-committee re- pany, and from 1974 to 1976, he was a member
port final." of the Department of Electrical and Systems Engi-
neering, Rensselaer Polytechnic Institute, Troy, NY.
Since 1976, he has been with the Department of Electrical and Computer
Engineering, University of California at San Diego, La Jolla, where he is
Juan I. Montojo received the engineering degree
from the Escola Tecnica Superior d’Enginyeria de the Ericsson Professor of Wireless Communications and former Department
Telecomunicacio de Barcelona of the Universitat Po- Chairman, working in the area of digital communication theory with special
emphasis on spread-spectrum communication systems. He has also been
litecnica de Catalunya (UPC), Barcelona, Spain, and
a consultant to both government and industry in the areas of radar and
the Communications Systems Engineer certificate
from Institut Eurecom, Sophia-Antipolis, France, communications.
Dr. Milstein was an Associate Editor for Communication Theory for
both in 1997. He received an M.S. in Electrical En-
the IEEE T RANSACTIONS ON C OMMUNICATIONS, an Associate Editor for
gineering from the University of Southern California
(USC), Los Angeles, CA in 1999. Book Reviews for the IEEE T RANSACTIONS ON I NFORMATION T HEORY,
From 1997, he has been with Qualcomm Inc., an Associate Technical Editor for the IEEE C OMMUNICATIONS M AGAZINE,
and the Editor-in-Chief of the IEEE J OURNAL ON S ELECTED A REAS IN
San Diego, CA, where he has worked as a systems
C OMMUNICATIONS. He was the Vice President for Technical Affairs in
engineer on different wireless systems: Globalstar, cdma2000 1xEV-DO, TD-
SCDMA, WCDMA, LTE, LTE-A. He is currently focused on the design and 1990 and 1991 of the IEEE Communications Society, and is a former Chair
of the IEEE Fellows Selection Committee. He is a recipient of the 1998
specification of the physical layer of 3GPP’s Long Term Evolution (LTE). He
Military Communications Conference Long Term Technical Achievement
participates in the physical layer standardization group of 3GPP (RAN1) as
Qualcomm’s lead delegate for LTE and LTE-A. Award, an Academic Senate 1999 UCSD Distinguished Teaching Award,
Mr. Montojo is working towards a Ph.D. in electrical engineering with an IEEE Third Millennium Medal in 2000, the 2000 IEEE Communication
Society Armstrong Technical Achievement Award, and the 2002 MILCOM
emphasis on Communications Systems in the Electrical and Computer En-
Fred Ellersick Award.
gineering (ECE) Department of the University of California at San Diego
(UCSD). His research is focused on OFDM-based UWB and cognitive radio.
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:41 from IEEE Xplore. Restrictions apply.