Documente Academic
Documente Profesional
Documente Cultură
6, JUNE 2007
AbstractIn this paper, with reference to the problem of joint inexpensive detection schemes based on the Fast Fourier Trans-
equalization and narrowband interference (NBI) suppression in form (FFT) followed by one-tap frequency equalization (FEQ).
orthogonal frequency-division multiplexing (OFDM) systems, Moreover, a suitable number of virtual carriers (VCs) are
synthesis and analysis of both unconstrained and constrained
optimum equalizers are carried out, based on the maximum usually inserted into the OFDM signal [1], which are aimed at
signal-to-noise-plus-interference (SINR) criterion. Specifically, a simplifying the design of transmitting and receiving filters.
comparative performance analysis is provided from a theoretical In many scenarios, multicarrier systems operate in the pres-
point of view, either when the second-order statistics (SOS) of the ence of severe narrowband interference (NBI), e.g., in wireless
received data are exactly known at the receiver, or when they are
estimated from a finite number of data samples. Relying on the
systems operating in overlay mode or in non-licensed band,
results of this analysis, a three-stage constrained maximum-SINR or in wireline ones, wherein the transmission cables might
equalizer is then proposed, which outperforms existing receivers be exposed to crosstalk or radio-frequency interference. The
and, in comparison with its unconstrained counterpart, exhibits simple FFT-based receiver exhibits very poor performances in
a significantly stronger robustness against errors in the estimated the presence of NBI, since it merely nullifies interblock inter-
SOS. Moreover, a computationally efficient adaptive implemen-
tation of the three-stage equalizer is derived, and in connection ference (IBI) and interchannel interference (ICI) without taking
with it, a simple and effective NBI-resistant channel estimation any specific measure to counteract noise and NBI effects, i.e.,
algorithm is proposed. Finally, numerical simulations are per- it acts as the simplest form of data-independent zero-forcing
formed that aim to validate the theoretical analysis carried out (ZF) receiver. A viable strategy to jointly counteract channel
and compare the performances of the considered equalizers with impairments and NBI in wireline DMT systems is the adoption
those of existing approaches.
of bit-loading techniques [3] at the transmitter, whose use,
Index TermsConstrained maximum signal-to-interfer- however, is problematic in wireless systems due to the rapid
ence-plus-noise ratio (SINR) optimizations, narrowband in-
changes in channel characteristics. Therefore, a preferred
terference (NBI) suppression, orthogonal frequency-division
multiplexing (OFDM) systems. solution for OFDM-based wireless systems is to devise simple
interference suppression algorithms at the receiver side. To
perform this task, one might exploit different types of redun-
I. INTRODUCTION dancy, which are present in the OFDM signals, such as the
temporal redundancy induced by CP insertion, or the frequency
redundancy associated with the presence of VCs.1
I N many applications, such as high-speed Internet access,
wireless networking, digital audio, and video broadcasting,
the increasing need to integrate heterogeneous services has
Several reception strategies targeted at pure CP-based
systems (i.e., ) exploit temporal redundancy by
led to very high data-rate transmission requirements, thereby processing the portion (so-called unconsumed) of the CP not
making intersymbol interference (ISI), which is induced by contaminated by the channel, provided that the CP length
channel dispersion, one of the main performance limiting exceeds the discrete-time channel order . The resulting win-
factor. To counteract ISI, several physical layer solutions em- dowing receivers [5], [6] build (with different ad hoc criteria)
ploy multicarrier schemes [1], [2], such as discrete multitone a data-dependent window to be used before the FFT, which
(DMT), orthogonal frequency-division multiplexing (OFDM), aims to reduce noise and NBI contributions without modifying
and multicarrier code-division multiple-access (MC-CDMA). the desired signal component. In particular, [5] and [6] carry
OFDM schemes cope with ISI by inserting a cyclic prefix (CP) out window designs based on the minimum mean-square error
of length at the beginning of each transmitted symbol, (MMSE) criterion; however, the design constraints (mainly
which is discarded at the receiver, thus allowing the use of aimed at reducing receiver complexity) do not allow one to
fully exploit the temporal redundancy contained in the CP, thus
Manuscript received December 29, 2005; revised October 5, 2006. This work leading to equalizers with limited NBI suppression capabilities.
was supported in part by Italian National project Wireless 8O2.16 Multi-antenna For the same pure CP-based systems, an MMSE equalizer
mEsh Networks (WOMEN) under Grant 2005093248. The associate editor co- has been proposed [7], which achieves a stronger robustness
ordinating the review of this manuscript and approving it for publication was
Dr. Franz Hlawatsch. against NBI effects by processing all the CP samples. On the
D. Darsena is with the Dipartimento per le Tecnologie, Universit Parthenope, other hand, when the channel order is very high, in order
I-80133, Napoli, Italy (e-mail: darsena@unina.it).
G. Gelli, L. Paura, and F. Verde are with the Dipartimento di Ingegneria Elet- 1As a matter of fact, besides temporal and frequency redundancy, constel-
tronica e delle Telecomunicazioni, Universit Federico II, I-80125, Napoli, Italy lation redundancy [4] can also be exploited to improve NBI rejection, taking
(e-mail: gelli@unina.it; paura@unina.it; f.verde@unina.it). advantage of symmetry properties exhibited by many constellations in digital
Digital Object Identifier 10.1109/TSP.2007.893946 communications.
to avoid the large overhead arising from the insertion of a long ceivers can suffer from a significant performance degradation
CP, a common solution is to employ a time-domain equalizer with respect to their ideal counterparts. Another limitation of
(TEQ) before the FFT, aimed at shortening the channel impulse all the aforementioned techniques is that the desired channel
response. In the DMT context, relying on the idea that the TEQ impulse response is assumed to be exactly known at the re-
and the demodulating FFT can be interchanged, the authors ceiver; however, in the presence of strong NBI, channel esti-
in [8] proposed a frequency-domain filter, so-called per-tone mation is quite a challenging task and cannot be performed by
equalizer (PTEQ) that outperforms classical TEQs at the ex- resorting to standard algorithms [19]. In this paper, we tackle the
pense of a higher memory cost by separately designing a linear NBI suppression problem by casting it in a more general frame-
MMSE equalizer for each subcarrier. It has been shown in [9] work and addressing explicitly the channel estimation problem
that the combination of windowing and per-tone equalization in the presence of NBI. First, we consider in Section II a general
leads to the synthesis of a windowing PTEQ (WPTEQ) receiver, OFDM signal model, accounting for (possibly combined) inser-
which exhibits increased robustness against NBI, compared to tion of CP and VCs, and encompassing, as particular cases, pure
a TEQ-based receiver. Recently, the results of [8] and [9] have CP-based systems, pure VC-based systems and hybrid CP/VC-
been extended in [10] to account for infinite impulse response based systems. Relying on the maximum signal-to-interference-
(IIR) channel models. plus-noise ratio (SINR) criterion, we synthesize in Section III
Instead of (or in addition to) the temporal redundancy con- interference-resistant IBI-free receivers, which do not make use
tained in the CP, one can exploit frequency redundancy arising of the conventional FFT-based preprocessing, and we analyze
from VC insertion to synthesize NBI-resistant receivers. The their disturbance rejection capabilities, both when the SOS of
use of frequency redundancy for the synthesis of ZF general- the received data are ideally known at the receiver, as well as
ized FEQ-DMT equalizers (operating in the absence of NBI) has when the equalizers are synthesized starting from SOS esti-
been already proposed in [11][14]. More precisely, receivers mates. In particular, we analytically show that, although the
targeted at pure VC-based system (i.e., with and maximum-SINR equalizer is capable of ideally achieving sat-
) were considered in [11], whereas ZF design techniques for an isfactory ICI-plus-NBI suppression under certain conditions, it
hybrid CP/VC-based system (i.e., with and ) suffers from a significant performance degradation when is esti-
were proposed in [12][14]. However, all these ZF receivers mated from data. Capitalizing on the results of our analysis, we
lack of any NBI suppression capability. With reference to a pure design in Section IV a three-stage constrained maximum-SINR
VC-based system, in [15] it is also proposed an MMSE version equalizer, which generalizes our previous formulation [17] and
of the FEQ-DMT receiver [11], which might also be used to offers improved robustness against finite sample-size effects.
counteract the NBI (even though this feature was not explicitly For such a receiver, a theoretical performance analysis is pro-
mentioned in [15]). A different MMSE approach to NBI rejec- vided, in both cases of known and estimated SOS. Moreover, a
tion, which can be applied to hybrid CP/VC-based systems with low-complexity adaptive implementation of the proposed equal-
, is proposed in [16]; it is based on a linear interfer- izer is devised, which allows one to estimate the desired channel
ence canceler that estimates, in the MMSE sense, the NBI at impulse response, even in the presence of strong NBI. Section V
the receiving side and subtracts it from the received signal. The provides numerical results that aim to corroborate the results of
receiver of [16] is synthesized under the assumptions that the the theoretical analysis and assess the performances of both un-
VCs are located in the frequency domain close to the NBI spec- constrained and constrained maximum-SINR equalizers in dif-
tral position, and the second order statistics (SOS) of the NBI are ferent operative scenarios and in comparison with existing re-
known at the receiver; if any or both assumptions are not exactly ceivers. Finally, concluding remarks are drawn in Section VI.
satisfied, its performance may degrade significantly. Moreover,
the same receiver can also operate in a system without VCs, but A. Notations
in this case, it undergoes [16] a significant performance degra- The fields of complex, real, and integer numbers are denoted
dation. Recently, an NBI-resistant receiver has been proposed in with , and , respectively; matrices [vectors] are denoted
[17], whose synthesis is based on the minimum mean-output-en- with upper [lower] case boldface letters (e.g., or ); the field
ergy (MMOE) criterion [18]. In comparison with windowing of complex [real] matrices is denoted as ,
receivers (e.g., [6]), the approach of [17] leads to a better ex- with used as a shorthand for or
ploitation of the temporal redundancy; however, the resulting indicates the th element of matrix ,
receiver is targeted at pure CP-based systems and, hence, does with and ; the
not exploit the redundancy induced by VC insertion. superscripts , and denote the conjugate, the trans-
From the previous discussion, it appears that the NBI sup- pose, the Hermitian (conjugate transpose), the inverse, and
pression techniques proposed in [5], [6], [8], [9], [16], and [17] the Moore-Penrose generalized inverse [20] (pseudo-inverse)
do not fully exploit the temporal and/or frequency redundancy of a matrix, respectively; denotes Hadamard (elementwise)
of the OFDM signal. Furthermore, in many of these papers, the product of two matrices and is the Euclidean norm of
performance studies carried out, either theoretically or experi- and denote the null
mentally, have been based on the idealized assumption that the vector, the null matrix, and the identity matrix, respectively;
SOS of the received data are exactly known at the receiver. In , and denote the null space, the range
practice, SOS must be estimated from a finite number of sam- (column space), and the orthogonal complement of the column
ples of the received signal, and the resulting data-estimated re- space of in ; when applied to a
3034 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 6, JUNE 2007
vector is the diagonal matrix with , th IBI-free received data block , with
whereas when applied to a matrix is the vector , can be expressed (see [2] and [22]) as
with ; finally, , and denote ensemble
averaging, convolution and integer ceiling, respectively.
(1)
II. OFDM SYSTEM MODEL AND
MAXIMUM-SINR OPTIMIZATION where the matrix dis-
cards the first entries of the received vector
Let us consider an hybrid CP/VC-based OFDM system , which collects all the
with subcarriers, of which are utilized, whereas the samples , whereas is
remaining are VCs. At the transmitter, the Toeplitz channel matrix, whose first column and row are
the information data stream is converted given by and ,
into parallel substreams , where respectively, depending on the discrete time channel
refers to the subcarrier. By assuming , which is a causal finite impulse re-
for now that the VCs are inserted at the end of the th data sponse (FIR) filter of order , i.e., for
block , one , with ; finally,
obtains, after VC insertion, the new symbol block and
, which can be expressed represent the NBI and the noise vectors, respectively,
as , where where , and
is tall and full-column rank. This relation can be gener- , with
alized to allow for VCs insertion2 in arbitrary positions and . It is
by introducing a row-permu- worth noting that unlike many equalization techniques for pure
tation matrix [21] , i.e., , with CP-based systems (i.e., ), wherein IBI elimination
. Subsequently, the block is subject to is achieved by assuming that and discarding the
the Inverse Discrete Fourier Transform (IDFT), obtaining the first samples of (i.e., the entire CP), it is sufficient
vector , where to remove only samples of the CP to achieve perfect IBI
, with , repre- suppression.4 This strategy is pursued by windowing receivers
sents the unitary symmetric IDFT matrix.3 Then, a CP of [5], [6], [8], [9], and as we will see in the sequel, it allows
length is inserted at the beginning of , thus ob- one to gain additional degrees of freedom, which can also be
taining , with exploited for NBI suppression. Interestingly, as pointed out in
, which can be expressed as , [11][15], perfect IBI elimination can be obtained also when
where is the full-column rank (insufficient CP length), or even when
precoding matrix, with , and (pure VC-based system); in both cases, we will show that VC
is obtained from by picking its last insertion is also mandatory to mitigate NBI effects.
rows. Vector undergoes parallel-to-serial conversion, and In the rest of the paper, the following customary assump-
the resulting sequence feeds a digital-to-analog tions are considered: a1) The information symbols are
converter (DAC) operating at rate , where and modeled as a sequence of zero-mean independent and identi-
denote the sampling and the symbol period, respectively. cally distributed (i.i.d.) circular random variables, with variance
After up-conversion, the continuous-time signal at the ; a2) the interference vector is modeled
DAC output is transmitted over a multipath channel, which is as a zero-mean complex circular wide-sense stationary (WSS)
modeled as a linear time-invariant (LTI) system with impulse random vector that is statistically independent of ; a3) the
response . noise vector is modeled as a zero-mean complex circular
After antialiasing filtering, the received baseband signal is white Gaussian random vector that is statistically independent
given by , where of both and , with autocorrelation matrix
is the impulse response of the analog-to-digital (ADC) an- .
tialiasing filter, whereas and account for the interfer- After partial CP removal, a linear zeroth-order5 FIR equal-
ence and thermal noise at the output of the ADC filter. Denote izer is employed in
with the impulse response of the composite channel (en- order to jointly mitigate the interchannel interference (ICI),
compassing the cascade of the DAC filter, the physical channel, NBI and noise. The th entry of the equalizer output
and the ADC filter), which spans sampling periods,
4Note that partial CP removal requires, in principle, exact knowledge of the
that is, for . After ideal carrier-fre-
channel order L . In practice, when only an upper bound L L is avail-
L
rn
quency recovery, sampling with rate and removing only able, one must resort to a suboptimal solution by discarding the first sam-
samples of the CP to achieve perfect IBI suppression, the ples of ~( ). Although the proposed method can work also when L is em-
ployed in the equalizers synthesis (see Section V), for the sake of clarity, all the
2Observe that VCs are commonly inserted at the edges of the spectrum to subsequent mathematical derivations are derived by assuming L = L .
K K
W W W
avoid aliasing problems at the receiver [1]. 5The extension to a th-order FIR equalizer, which jointly elaborates + 1
3Its inverse = defines the DFT matrix. OFDM symbols, is straightforward.
DARSENA et al.: CONSTRAINED MAXIMUM-SINR NBI-RESISTANT RECEIVER FOR OFDM SYSTEMS 3035
requires only upper bounds (rather than the exact knowledge) Vandermonde vectors [18], we can infer that this happens iff
on the channel order and the NBI rank (i.e., the NBI , which, in turn,
bandwidth ). In particular, it has been experimentally imposes that . Remarkably, note
verified in [16] that for practical values of , the effective rank that this last condition is violated when the tone interference
of the NBI turns out to be slightly greater than , and a quite is located exactly on a used subcarrier, i.e., , for
conservative rule of thumb has been derived for determining . Hence, when a tone interference is lo-
the effective rank of the NBI, which can be used in (5). cated exactly on a used subcarrier, a maximum-SINR equalizer
With regard to condition c2), it should be observed that in the cannot completely suppress the disturbance (i.e., ICI-plus-NBI),
case of pure CP-based systems, it reformulates the well-known even in the absence of noise. As we will show in Section V, max-
result [22] that FIR-ZF equalizers do not exist if the channel imum-SINR equalizers exhibit this behavior for a nonnull-band-
transfer function exhibits one or more zeros located in width NBI signal as well when the NBI frequency offset is
correspondence of the subcarrier frequencies. On the other placed near to an used subcarrier.
hand, for systems with VCs, as it was also recognized in [11], In the sequel, we assume that conditions c1)c3) are ful-
condition c2) infers that VC insertion leads to milder restric- filled. In practice, the synthesis of the maximum-SINR equalizer
tions on channel-zero locations, since possible zeros of at (4) requires that the disturbance autocorrelation matrix
the unused subcarrier frequencies do not affect the existence of be consistently estimated from , which contains also the
FIR-ZF equalizers. contribution of the desired signature . On the other
Condition c3) is a technical requirement and imposes that the hand, it is well-known [24] that a maximum-SINR receiver can
two subspaces and must be nonoverlapping or also be expressed in terms of the autocorrelation matrix
disjoint, which is less restrictive [26] than simple orthogonality of , which can be estimated from the re-
between the same subspaces. To gain more insight about condi- ceived data more easily than . An equalizer belonging to
tion c3), let us characterize the subspace more explicitly. the maximum-SINR family is the well-known MMSE one [23],
It is shown in Appendix B that the matrix can be parameter- which is the solution of the optimization problem
ized as
(6)
(16)
(11)
Since and are equivalent, in the sense that both By virtue of Theorem 1, it is interesting to note that for
maximize the output SINR at each subcarrier, we consider in practical values of and for , it results that
the sequel the MMOE equalizer, because its analysis is sim- , and thus, expression (16) be-
pler. Specifically, our aim is to investigate the SINR degrada- comes , which shows that due
tion when the filtering matrix is synthesized by using to the effects of the finite sample-size , the SINR at the th
the sample correlation matrix of , estimated over subcarrier saturates. This saturation effect gives rise to unac-
symbol intervals, rather than . In this situation, the weight ceptable bit-error-rate (BER) floors, when the value of is not
vector (10) of the MMOE equalizer is significantly larger than . For
instance, to achieve a SINR value of 10 dB at the th subcarrier
in the high SNR region, a sample size is
(12) required. On the other hand, for OFDM systems employing a
large number of subcarriers, supposing that
where, accounting for (1) requires very large values of the sample size which, in wire-
less scenarios, may lead to a packet duration exceeding the
coherence time of the channel. To make the MMOE equalizer
(12) robust against finite sample-size effects, one can exploit
the eigenstructure of , thus obtaining a subspace-based
implementation [29] of the receiver or, alternatively, resort to a
(13)
diagonal loading approach [24], which consists of replacing in
with (12) matrix with , where denotes a di-
, and rep- agonal loading factor. However, the former approach increases
resenting sample estimates of the symbol variance , the the receiver complexity and is not suited to simple adaptive
cross-correlation between the disturbance vector and implementations, whereas the optimal choice of in the latter
the desired symbol , and the autocorrelation matrix of approach is not a simple task, since it is scenario-dependent. In
, respectively. In this case, the weight vector Section IV, we propose instead a constrained maximum-SINR
is random, and thus, the expectations in (2) must also be receiver with channel estimation capabilities, which achieves
evaluated with respect to , that is, a convenient tradeoff between ideal SINR performances and
, where we have addition- finite sample-size robustness.
ally taken into account the constraint . By In conclusion, it should be observed that the computational
assuming7 that is statistically independent of complexity of the MMOE (or MMSE) equalizer is essentially
and resorting to the conditional expectation rule, one obtains dominated by the inversion , which requires flops,
. with . This computational burden might
It is shown in [28] that for moderate-to-high values of the be prohibitive for OFDM systems employing a large number
sample size, i.e., , the predominant cause of SINR of subcarriers and/or operating in time-varying NBI envi-
degradation is represented by . Thus, after inserting (13) ronments. In these scenarios, it is customary to resort to the re-
cursive least square (RLS) algorithm [24], [30], which assures
in (12) and then replacing with , the weight
a fast symbol-by-symbol updating of the receiver, with
vector can be approximated [28] as
computational complexity per iteration.
7Since g
^ is estimated from fr(`)g , it is statistically independent whose solution is
n
from d ( ), provided that nK . . It is apparent that
3038 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 6, JUNE 2007
(19)
while linear constraints are imposed in (11) for preserving
the desired symbols at each used subcarrier, by
treating the ICI in (1) in the same way as NBI and noise, where . It is worth noting
the optimization problem (17) imposes linear constraints, that accounting for (1), vector turns out [24] to be the
which not only preserve the desired symbols but also assure solution of the following constrained optimization problem:
deterministic ICI cancellation at each used subcarrier. The
constrained optimization problem (17) can be reformulated
as an unconstrained one by resorting to an extension of the (20)
generalized sidelobe canceller decomposition [24], which was
proposed in the array processing context. Specifically, fol- where the goal of the first linear constraint is to preserve the de-
lowing [17], it can be shown that admits the canonical sired symbol, whereas the remaining linear constraints
decomposition assure perfect ICI suppression at the th subcarrier. Solution
(20) is also known [24] as the linearly constrained minimum
variance (LCMV) beamformer. As a consequence of the con-
straints in (20), the output SINR (2) of the CMMOE equal-
izer at the th subcarrier can be written as
(18) , where accounting for (19)
with
, where
represents the minimum-norm (in the Frobenius sense) solu-
tion of the ICI-free constraint imposed in (17), whereas the
signal blocking matrix satisfies the relation
. The choice of is not unique and (21)
will be discussed in Subsection IV-C. In the sequel, without represents the disturbance (NBI-plus-noise) power at the equal-
loss of generality, we only impose that . izer output. Observe that from a mathematical point of view, be-
The second equality in the expression of is a direct cause of the additional linear constraints im-
consequence of the signal blocking property of , with posed in (20), the output cannot be larger than the
representing the autocorrelation maximum value , i.e., .
matrix of the NBI-plus-noise vector . In other words, when the SOS of the received data are ideally
Decomposition (18) leads to the two-stage structure sketched known at the receiver, the CMMOE equalizer cannot perform
in Fig. 1, which shows that the second stage is the better than the MMOE (or MMSE) equalizer. However, the fol-
difference of a fixed (i.e., data-independent) term lowing Theorem proves that, under conditions (c1), (c2) and
and a free or adaptive term . Remarkably, in the ab- (c3), similarly to a maximum-SINR equalizer, the CMMOE de-
sence of NBI, i.e., when , the adaptive part of tector is able to completely reject the NBI as the noise variance
the CMMOE equalizer vanishes, and reduces to its vanishes, that is, in the high SNR region (ap-
data-independent part. proximatively) attains the maximum value .
Since the CMMOE equalizer (17) has been obtained by Theorem 2: In the limiting situation of vanishingly small
adding further constraints to the matrix optimization problem noise, the CMMOE equalizer (17) enables perfect NBI sup-
(11), in the ideal situation when is perfectly known, the pression for each used subcarrier, i.e.,
CMMOE equalizer does not maximize the output SINR for each , iff conditions c1)c3) hold.
subcarrier. Our goal is to show that, with respect to the MMOE Proof: See Appendix C.
equalizer, the ICI-free constraint gains robustness against finite Besides confirming the statement of Theorem 2 for mod-
sample-size effects (Section IV-B), without compromising the erate-to-high values of the SNR, the simulation results of
ideal NBI suppression capabilities in the high SNR region (Sec- Section V will show moreover that with respect to the MMOE
tion IV-A). Furthermore, we show that the CMMOE equalizer equalizer, the performance penalty paid by the CMMOE
DARSENA et al.: CONSTRAINED MAXIMUM-SINR NBI-RESISTANT RECEIVER FOR OFDM SYSTEMS 3039
equalizer is extremely small in the low SNR region. As a since in OFDM systems of practical interest . This
side comment, observe that, similarly to a maximum-SINR implies that to achieve a given target SINR as approaches
equalizer, the CMMOE cannot completely suppress a tone zero, the CMMOE equalizer (22) allows one to use a consid-
interference located exactly on a subcarrier, even in the absence erably smaller sample size in com-
of noise, since in this case, condition c3) is violated. This parison with that required by the MMOE equalizer (12). For
is in agreement with the performance analysis carried out in instance, with reference to the HIPERLAN/2 context (channel
[17] for pure CP-based OFDM systems. Furthermore, as is model A) [31], wherein and
shown in Section V, similarly to a maximum-SINR receiver, , it results that , i.e.,
the CMMOE equalizer is not able to satisfactorily reject a must be four times larger than to achieve the
nonnull-bandwidth NBI signal when its frequency offset is same performance. It is shown in Section V that the CMMOE
placed near to a used subcarrier. equalizer (22) outperforms the MMOE one (12) also for low
values of the SNR since, while is significantly smaller
B. Performance Analysis for Finite Sample-Size than in practice, is only slightly inferior to
In this subsection, we evaluate the SINR degradation when in the low-SNR region.
the filtering matrix is synthesized by using the sample
correlation matrix of , given by (13). In this case, the C. Three-Stage Implementation, Computational Complexity,
estimate of the weight vector (19) can be written as and Channel Acquisition
Before discussing computational complexity issues, we
would like to highlight a further nice property of the CMMOE
(22) equalizer. By resorting to the parameterization (6) of , one
Reasoning as in Section III for , and more- obtains that due to the nonsingularity of the diagonal matrices
over, taking into account the additional constraints and , the fixed term of in (18) can be
, one obtains written as . More-
. By substituting (13) over, parameterization (6) leads to a channel-independent syn-
in (22) and accounting for the signal blocking property of , thesis of the signal blocking matrix . Indeed, under condition
we obtain, after tedious but straightforward matrix algebra c2), the null space of spans the column space of , which,
in turn, coincides with the column space of , whose structure
does not depend on the channel coefficients. Thus, the condition
is equivalent to ,
that is, the synthesis of does not require knowledge of
the channel vector ,
(23)
and thus, together with the Moore-Penrose inverse of , can
with representing the sample be carried out off-line by resorting to any orthonormalization
estimate of the autocorrelation matrix of . Equation (23) algorithm (e.g., QR or singular value decomposition), without
evidences that is composed of two terms: The former requiring real-time extra computations. On the basis of these
represents an estimate of [see (19)], whereas the latter observations, we can rewrite (18) equivalently as follows:
is the perturbation resulting from the nonzero sample cross-cor-
relation vector . Along the same lines of Section III, we
resort in (23) to the approximation
(26)
(24) with .
that is, we replace the sample autocorrelation matrix Remarkably, synthesis of the matrix is completely
with the exact one . Relying on (24) and accounting for blind in the sense that it can be done relying only on the re-
, it is easily proven that ceived data, without requiring knowledge of the channel vector
. Additionally, it can be verified that turns out to be
(25) the solution of the following MOE-based criterion:
(31)
(28) is a nonsingular diagonal matrix collecting all the pilot sym-
bols, whereas the entry of the matrix
where is , for and . To
is the overall gain vector, and allow to be full-column rank, we assume that the number of
, with and denoting the estimate, pilot symbols is larger than channel memory, i.e., .
at iteration , of and the forgetting factor It is worth noting that, due to filtering carried out in the second
of the recursive algorithm, respectively. According to the usual stage through the matrix , accounting for the assump-
initialization strategy for the RLS algorithm, we set tion a3), the complex circular Gaussian noise vector turns
and , where is a out to be colored with zero-mean and autocorrelation matrix
DARSENA et al.: CONSTRAINED MAXIMUM-SINR NBI-RESISTANT RECEIVER FOR OFDM SYSTEMS 3041
, where is obtained from CMMOE equalizer, which fully exploits the available degrees
by picking up its rows at the pilot locations . There- of freedom, the receiver of [6] exhibits a substantial BER per-
fore, the maximum likelihood estimation (MLE) of is given formance degradation and outperforms the conventional ZF re-
by (see [32]) ceiver only slightly.
In [16], the authors considered an OFDM system with CP
(i.e., ), possibly employing VC insertion at the
(32) transmitting side.8 The receiver proposed in [16] is a two-stage
IBI-free receiver, where the first stage removes the entire
Remarkably, it results [32] that is unbiased (i.e., CP by performing a fixed time-domain windowing. After
) and consistent (i.e., attains the CramerRao lower bound), some straightforward manipulations, it can be shown that the
and hence, it represents the minimum variance unbiased (MVU) second stage of [16] can be decomposed as
estimator. After estimating through (32), the third stage per- , where9 the fixed matrix satisfies
forms, in practice, the FEQ reported in Fig. 2 by resorting to the the ICI-free equation, whereas the synthesis of requires
inverse of the diagonal matrix , whose entries are given by knowledge of the autocorrelation matrix of the NBI. Although
the transfer function of the estimated channel , evaluated at the receiver of [16] can work in a mismatched mode, its per-
. A final remark concerns the computational com- formance may degrade significantly if is not accurately
plexity of the MLE estimator (32); since , the com- known. In contrast, the proposed CMMOE strategy overcomes
putational complexity of (32) is essentially dominated by the this drawback, since it does not require any a priori information
inversion of , which requires flops. about the NBI, but rather, it implicitly estimates its SOS on
the basis only of the received data. Furthermore, by virtue of
D. Relationships With Existing NBI-Resistant Receivers
the above decomposition of , it can be observed that the
Receiver windowing [5], [6], [8], [9] is a low-complexity ZF equalizer of [16] admits a parallel implementation, which is
equalization technique proposed for pure CP-based OFDM sys- similar to that reported in Fig. 1. More specifically, the fixed
tems (i.e., and ). We focus attention here filtering matrix performs substantially the same operations
on the windowing receiver proposed by Redfern [6], since rela- as in (18). Thus, similarly to the CMMOE equalizer,
tionships with other windowing-based techniques can be estab- the upper branch of the receiver [16] contains both the OFDM
lished with similar reasonings. After some algebra, the equalizer signal and the disturbance. On the other hand, unlike
of [6] can be expressed as , in (18), the adaptive filtering matrix does not exhibit
where , which is of the form of (33), shown at the in general the signal blocking property. Indeed, for a system
bottom of the page, can be interpreted as a generalized win- employing VC insertion at the transmitter (i.e., ),
dowing matrix, with and under the assumption that the subcarriers are located in
. Note that instead of the parallel imple-
the proximity of the NBI carrier frequency, the matrix
mentation depicted in Fig. 1, matrix leads to a serial
behaves as a blocking matrix, which removes the OFDM signal
decomposition of the ZF receiver and performs, in the given
component in the lower branch. On the other hand, when no
order, generalized windowing, DFT and one-tap FEQ on the
VC insertion is carried out at the transmitter (i.e., ),
IBI-free received signal . The windowing strategy of [6],
matrix does not block the OFDM signal, and thus, in
as depicted in [6, Fig. 2], can be obtained from (33) by im-
addition to the disturbance contribution, the lower branch also
posing and ,
contains the OFDM signal component. This is a very undesired
with . In [6], vector is chosen to suppress
situation since subtraction between the upper and lower branch
the disturbance contribution in the MMSE sense, and its syn-
outputs leads to partial OFDM signal cancellation, which, as
thesis does not require a priori knowledge of the disturbance
observed in [16], implies a substantial performance degrada-
autocorrelation function. It is thus apparent that in comparison
tion, in comparison with the case where VCs are available. On
with the CMMOE equalizer, although the technique of [6] al-
the contrary, the proposed CMMOE equalizer does not suffer
lows one to reduce the implementation complexity (in terms
of this problem.
of number of complex multiplications needed), the resulting
equalizer has a diminished interference suppression capability, 8Although the equalizer of [16] can jointly elaborate multiple consecutive
since the structure imposed to and , before carrying out OFDM symbols, for the sake of comparison, we consider here a zeroth-order
equalizer, which is the case examined in depth in the simulations reported in
MMSE optimization, leads to suboptimal exploitation of the [16].
available degrees of freedom for disturbance suppression. It has 9For the sake of conciseness, we defer directly to [16] for the explicit expres-
been experimentally shown in [17] that in comparison with the sion of G and G .
(33)
3042 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 6, JUNE 2007
Fig. 3. ABER versus SNR (SIR = 10 dB, K = 500 symbols). Fig. 5. ABER versus K (SNR = 24 dB, SIR = 10 dB).
Fig. 4. ASINR versus SNR (SIR = 10 dB, K = 500 symbols). Fig. 6. ASINR versus K (SNR = 24 dB, SIR = 10 dB).
markably, the ABER curve of the CMMOE-dmi (channel es- the sample size , the assumption that the predominant cause
timated) equalizer strictly follows those of the CMMOE-dmi of SINR degradation in (23) is represented by is better
(channel known) and WPTEQ (ideal) for all the considered verified for the CMMOE receiver.
SNR values. Observe that the experimental behaviors of the Example 2ABER and ASINR versus sample size : In
data-estimated MMOE and CMMOE receivers (with channel Fig. 5, we have reported the ABER performance only for
known) are in good agreement with the results of the theoretical the considered data-estimated receivers, as a function of the
performance analysis carried out in Sections III and IV-B. sample size . The SNR and SIR have been kept constant to
To further corroborate this analysis, we have also reported in 24 and 10 dB, respectively. Results show that, with respect
Fig. 4 the ASINR at the output of the MMOE-dmi (channel to the MMOE equalizers, the CMMOE receivers can assure
known) and CMMOE-dmi (channel known) equalizers; in a significant performance gain for a wide range of the values
the same plot, the simulation results (referred to as simula- of the sample size . In particular, it can be observed that to
tion) are compared with the corresponding theoretical curves achieve an ABER value of , the MMOE-dmi (channel
(referred to as theoretical) [see (16) and (25)]. Results show known) equalizer requires a sample size equal to
that the theoretical expression (25) for the CMMOE equalizer symbols, i.e., three times that required by the CMMOE-dmi
agrees very well with the simulation results for all values of (channel estimated) receiver, whereas, to achieve the same per-
SNR, whereas the theoretical expression (16) for the MMOE formance, the MMOE-dl (channel known) and MMOE-sub
equalizer is not as accurate as (25). Indeed, since the CMMOE (channel known) equalizers require a sample size of about 700
equalizer provides a stronger robustness than the MMOE one symbols. Additionally, we have reported in Fig. 6 the ASINR
against finite sample-size effects, for the considered value of at the output of the data-estimated DMI-based CMMOE and
3044 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 6, JUNE 2007
Fig. 7. ABER versus SIR (SNR = 24 dB, K = 500 symbols). Fig. 8. ABER versus = M T f (SNR = 24 dB, SIR = 10 dB, K = 500
symbols).
where . Thus,
letting , matrix
can be expressed
as
(35)
[28] M. Wax and Y. Anu, Performance analysis of the minimum vari- Luigi Paura was born in Napoli, Italy, on February
ance beamformer, IEEE Trans. Signal Processing, vol. 44, no. 4, pp. 20, 1950. He received the Dr. Eng. degree summa
928937, Apr. 1996. cum laude in electronic engineering in 1974 from the
[29] X. Wang and H. V. Poor, Blind multiuser detection: A subspace ap- University of Napoli Federico II.
proach, IEEE Trans. Inf. Theory, vol. 44, no. 2, pp. 677690, Mar. From 1979 to 1984, he was with the Department
1998. of Electronic and Telecommunication Engineering,
[30] S. Haykin, Adaptive Filter Theory. New York: Prentice-Hall, 1996. University of Napoli, first as an Assistant Professor
[31] Channel Models for HIPERLAN/2 in Different Indoor Scenarios, [On- and then as an Associate Professor. Since 1994, he
line]. Available: http://www.etsi.org, ETSI Normalization Committee, has been a Full Professor of telecommunication, first
Norme ETSI, available on with the Department of Mathematics, University of
[32] S. M. Kay, Fundamentals of Statistical Signal Processing: Estimation Lecce, Lecce, Italy, then with the Department of In-
Theory. Englewood Cliffs, NJ: Prentice-Hall, 1993. formation Engineering, Second University of Napoli, and, finally, since 1998,
[33] M. K. Tsatsanis and Z. D. Xu, Performance analysis of minimum vari- he has been with the Department of Electronic and Telecommunication Engi-
ance CDMA receivers, IEEE Trans. Signal Processing, vol. 46, no. 11, neering, University of Napoli Federico II. He also held teaching positions at
pp. 30143022, Nov. 1998. the University of Salerno, Italy, at the University of Sannio, Italy, and the Uni-
[34] G. Marsaglia and G. P. H. Styan, Equalities and inequalities for ranks versity of Napoli Parthenope, Italy. In 19851986 and 1991, he was a Visiting
of matrices, Linear Multilinear Algebra, pp. 269292, Feb. 1974. Researcher at the Signal and Image Processing Laboratory, University of Cali-
fornia, Davis. Currently, his research activities are mainly concerned with sta-
Donatella Darsena (M06) was born in Napoli, Italy, tistical signal processing, digital communication systems, and medium access
on December 11, 1975. She received the Dr. Eng. de- control in wireless networks.
gree summa cum laude in telecommunications engi-
neering in 2001 and the Ph.D. degree in electronic
and telecommunications engineering in 2005, both
from the University of Napoli Federico II. Francesco Verde was born in Santa Maria Capua
Since 2005, she has been an Assistant Professor Vetere, Italy, on June 12, 1974. He received the
with the Department for Technologies, University of Dr. Eng. degree summa cum laude in electronic
Napoli Parthenope. Her research activities lie in the engineering in 1998 from the Second University of
area of statistical signal processing, digital commu- Napoli, Italy, and the Ph.D. degree in information
nications, and communication systems. In particular, engineering in 2002 from the University of Napoli
her current interests are focused on equalization, channel identification, and nar- Federico II.
rowband-interference suppression for multicarrier systems. Since 2002, he has been an Assistant Professor
with the Department of Electronic and Telecommu-
nication Engineering, University of Napoli Federico
II. His research activities lie in the areas of statistical
Giacinto Gelli was born in Napoli, Italy, on July signal processing, digital communications, and communication systems. In
29, 1964. He received the Dr. Eng. degree summa particular, his current interests are focused on cyclostationarity-based tech-
cum laude in electronic engineering in 1990 and niques for blind identification, equalization and interference suppression for
the Ph.D. degree in computer science and electronic narrowband modulation systems, code-division multiple-access systems, and
engineering in 1994, both from the University of multicarrier modulation systems.
Napoli Federico II.
From 1994 to 1998, he was an Assistant Professor
with the Department of Information Engineering,
Second University of Napoli. Since 1998, he
has been with the Department of Electronic and
Telecommunication Engineering, University of
Napoli Federico II, first as an Associate Professor and, since November 2006,
as a Full Professor of telecommunications. He also held teaching positions at
the University Parthenope of Napoli. His research interests are in the fields of
statistical signal processing, array processing, image processing, and mobile
communications, with current emphasis on code-division multiple-access
systems and multicarrier modulation.