|Views: 1,317
|Likes: 4

Published by Moataz Fouad Yousef

See more

See less

Digital Object Identifier 10.1109/MSP.2011.940267 Date of publication: 19 April 2011

1053-5888/11/$26.00©2011IEEE

[

Behrouz Farhang-Boroujeny

]

[

Development of broadbandcommunication systems

]

s of today, orthogonal frequency division multiplexing (OFDM) has been the dominant technology forbroadband multicarrier communications. However, in certain applications such as cognitive radiosand uplink of multiuser multicarrier systems, where a subset of subcarriers is allocated to each user,OFDM may be an undesirable solution. In this article, we address the shortcomings of OFDM inthese and other applications and show that filter bank multicarrier (FBMC) could be a more effec-tive solution. Although FBMC methods have been studied by a number of researchers, and some even before theinvention of OFDM, only recently has FBMC been seriously considered by a few standard committees. The goal of this article is to bring this upcoming trend to the attention of the signal processing and communications commu-nities and to motivate more research in this important area.

INTRODUCTION

OFDM has been researched and deployed for broadband wired and wireless communications for the past twodecades. OFDM is widely adopted because of a number of advantages that it offers:

■

Orthogonality of subcarrier signals that allows• trivial generation of transmit signal through an inverse fast Fourier transform(IFFT) block• trivial separation of the transmitted data symbols at the receiver through afast Fourier transform (FFT) block• trivial equalization through a scalar gain per subcarrier• trivial adoption to multiple-input multiple-output (MIMO) channels.

A

IEEE SIGNAL PROCESSING MAGAZINE

[92]

MAY 2011

■

Closely spaced orthogonal subcarriers partition the available bandwidthinto a maximum collection of narrow subbands.

■

Adaptive modulation schemes can be applied to subcarrier bands to maxi-mize bandwidth efficiency/transmission rate.

IEEE SIGNAL PROCESSING MAGAZINE

[93]

MAY 2011

■

The very special structure of OFDM symbols simplifies thetasks of carrier and symbol synchronizations.These points are well understood and documented in theliterature [1], [2]. On the other hand, multiple-access OFDMor orthogonal frequency division multiple access (OFDMA) hasbeen recently proposed in a number of standards and proprie-tary waveforms (e.g., [3]). Some particular forms of OFDMAhave also been proposed for cognitive radio systems [5]. InOFDMA, each user is allocated a subset of subcarriers. To pre- vent intercarrier interference (ICI), the users’ signals must besynchronized at the receiver input (but not necessarily at thetransmitters’ outputs). Considering this point, one may notethat OFDMA only works well in the network downlink of abase station, where all of the subcarriers are transmitted fromthe same point (the base station) and hence can be easily syn-chronized and undergo the same Doppler frequency shiftbefore reaching each receiver. However, synchronization is nottrivial in the uplink where a number of nodes are transmittingseparately. Since, in practice, perfect synchronization in theuplink of an OFDMA network may not be possible, additionalsignal-processing steps have to be taken to minimize interfer-ence among signals from different nodes. Such steps add sig-nificant complexity to an OFDMA receiver; see [4] and thereferences therein.The problem is worse in a cognitive radio setting where bothprimary (noncognitive nodes) and secondary users (cognitivenodes) transmit independently and may be based on differentstandards. In such a setting, the only way that one may adopt toseparate the primary and secondary user signals is through afiltering mechanism. OFDM is thus a poor fit because the filtersassociated with its synthesized subcarrier signals (at the trans-mitter) and analyzed subcarrier signals (at the receiver) haverelatively large side lobes and such lobes will result in leakage of signal powers among the bands of different users [7]. Althoughsuggestions have been made to improve the side lobes of OFDManalysis and synthesis filters through the use of filtered OFDM[5] (discussed later) and other methods [6], these solutions aregenerally very limited in performance.Furthermore, data transmissions over digital subscriber lines(DSLs) and power line communication (PLC) technologies oftenuse unshielded wires and thus become a source and victim of elec-tromagnetic interference. This is similar to the case of OFDMAand cognitive radios, where each transmission can be establishedover certain portions of a broadband, and spectral activities overthe rest of the band should be avoided to allow coexistence withradio communication activities within the band of interest.The above problems could be greatly alleviated if the filtersthat synthesize/analyze the subcarrier signals had small sidelobes. An interesting, but apparently not widely understood, factis that the first multicarrier technique developed before theinvention of OFDM [10] used filter banks for the synthesis andanalysis of multicarrier signals. Such filter banks can bedesigned with arbitrarily small side lobes and, therefore, are anideal choice in multiple access and cognitive radio applicationsas well as broadband data transmission over unshielded wires.The goal of this article is to present a tutorial review of FBMC techniques and compare them with OFDM in variousapplications. We note that most of the advantages of FBMC orig-inate from the fact that, by design, the nonadjacent subcarriersin this modulation are separated almost perfectly through abank of well-designed filters. OFDM, on the other hand, wasoriginally designed with a great emphasis on a low-complexityimplementation. Much of the low complexity of OFDM is due toa fundamental assumption: subcarrier signals are a set of per-fectly synchronized orthogonal tones. These tones are generatedat the transmitter using an IFFT block, and they are separatedat the receiver through an FFT block. Although this article highlights a number of limitations of OFDM in present and future communication systems, theauthor has no intention of ignoring the many important anddesirable features of OFDM that were itemized at the begin-ning of this section. The intention is to emphasize the factthat OFDM, although widely adopted in the present industry,is not necessarily the best solution in many future communi-cation systems, particularly in multiple access and cognitiveradio networks where FBMC may be found more appealing. At the same time, the limitations of FBMC will be noted. Forinstance, we note that while deployment of a MIMO tech-nique in OFDM is a straightforward task, unfortunately, thedevelopment of MIMO-FBMC systems/networks is nontrivialand may be very limited. Moreover, while for many applica-tions, FBMC may be more complex than OFDM, there arecases where the added steps to undo the undesirable featuresof OFDM may lead to systems that are more complex thantheir FBMC counterparts.This article is organized as follows. To draw a connectionbetween OFDM and FBMC, a unified filter bank formulationthat is applicable to both is presented in the section “A UnifiedFormulation for OFDM and FBMC.” This formulation showsthat the prototype filter based on which an OFDM signal is syn-thesized/analyzed has to obey certain constraints, and it isthese constraints that result in the undesirable performance of OFDM in the applications discussed in [7] and those that wefurther discuss in this article. Filtered OFDM, a modifiedOFDM that has been designed to improve on the spectral con-tainment of subcarrier signals, is then presented and discussedin detail in the section “Filtered OFDM.” This presentationshows some of the serious limitations of a filtered OFDM inrestricting the spectra of desired subcarriers to any band of interest. FBMC systems are reviewed in the section “FilterBank Multicarrier.” This presentation is structured to clearlyshow the spectral advantages of FBMC over OFDM. The appli-cations of multicarrier techniques are reviewed in the section“Applications.” This section is organized to highlight the prosand cons of both OFDM and FBMC systems in a variety of cur-rent and future applications. To present the basic ideas behindOFDM and FBMC methods, without getting too far into thedetails, the presentations in sections “A Unified Formulationfor OFDM and FBMC,” “Filtered OFDM,” and “Filter BankMulticarrier” ignore some parts of the literature. This

IEEE SIGNAL PROCESSING MAGAZINE

[94]

MAY 2011

compromise is made to keep the intended tutorial feature of the article. To fill this gap, the section “Additional Notes” pro- vides a brief overview of some important recent developmentsin the area of FBMC. The section “Conclusions” presents theconcluding remarks of this article.

A UNIFIED FORMULATION FOR OFDM AND FBMC

Figure 1 presents a block diagram that is commonly used todepict an FBMC transceiver. This structure is also applicable toOFDM. We note that although, in practice, the synthesis andanalysis filter banks are implemented in discrete time, we havechosen to present them in terms of continuous time filters, asthis formulation serves our discussion the best.The inputs in Figure 1 are the data signals defined as

s

k

1

t

2

5

a

n

s

k

3

n

4

d

1

t

2

nT

2

, (1) where

s

k

3

n

4

s are the subcarrier data symbols,

k

is a subcarrierindex, and

T

is the symbol time spacing.The difference between OFDM and FBMC lies in the choiceof

T

and the transmitter and receiver prototype filters

p

T

1

t

2

and

p

R

1

t

2

, respectively. In a conventional OFDM,

p

T

1

t

2

is a rectan-gular pulse of height one and width

T

. The receiver prototypefilter

p

R

1

t

2

is also a rectangular pulse of height one, but its width is reduced to

T

FFT

,

T, where

T

FFT

5

1/B, and

B

is thefrequency spacing between subcarriers. (We note that the nota-tion

T

FFT

is used here since this is equal to the time durationover which the received signal is sampled and passed throughan FFT block [1], [2].) In FBMC systems that are designed formaximum bandwidth efficiency,

T

5

T

FFT

5

1/B, however, thedurations of

p

T

1

t

2

and

p

R

1

t

2

are greater than

T

(usually, aninteger multiple of

T

). Hence, in FBMC, the successive datasymbols overlap.The use of prototype filters with rectangular impulseresponses leads to undesirable magnitude responses that suf-fer from large side lobes in the frequency domain. This fol-lows immediately from the fact that the Fourier transform of a rectangular pulse is a sinc function, and it is well knownthat the side lobes of a sinc pulse are relatively large; the peakof the first side lobe is only 13 dB below the peak of its mainlobe. To combat this problem, in some standards, e.g., in very-high-bit rate DSL (VDSL) [22], [23] and PLCs [48], [49],the rectangular pulse shape, with sharp edges, is replaced bya pulse shape with smooth edges; see also [24]–[27]. In thisarticle, we use the term filtered OFDM to refer to the cases where the rectangular window is replaced by a window withsmooth edges.

FILTERED OFDM

FUNDAMENTAL CONCEPTS

From the structure depicted in Figure 1, the transmit signal isobtained as

x

1

t

2

5

a

n

a

k

[

K

s

k

3

n

4

p

T

1

t

2

nT

2

e

j

2

p

1

t

2

nT

2

f

k

,

(2) where denotes a set of active symbol indices and

j

5

"

2

1.In a conventional OFDM where

p

T

1

t

2

is a rectangular pulse, (2)may be read as follows. For every

n

,

x

1

t

2

is generated by addinga number of time-limited complex-valued tones, whose magni-tude and phase are determined by the data symbols

s

k

3

n

4

s.These tones, when passed through the channel, after the tran-sient period of the channel response, will be the same tonesmodified by the channel gains at the respective frequencies.Separating different subcarriers, (2) can be written as

x

1

t

2

5

a

k

[

K

x

k

1

t

2

, (3) where

x

k

1

t

2

5

a

n

s

k

3

n

4

p

T,

k

1

t

2

nT

2

(4)and

p

T,

k

1

t

2

5

p

T

1

t

2

e

j

2

p

tf

k

.

(5) We note that (4) may be viewed as a filtering operation appliedto a sequence of impulses, modulated by the data symbols

s

k

3

n

4

s. The filter

p

T,

k

1

t

2

is obtained by modulating the proto-type filter

p

T

1

t

2

. Alternatively, (2) may be written as

x

1

t

2

5

a

n

a

k

[

K

s

k

3

n

4

p

T,

k

1

t

2

nT

2

. (6) Assuming an ideal channel, the received signal

y

1

t

2

is the sameas the transmit signal

x

1

t

2

. In that case, the data symbols

s

k

3

n

4

,for

k

[

K

, and all values of

n

will be separable if

8

p

T,

k

1

t

2

mT

2

,

p

R

,l

1

t

2

nT

29

5d

kl

d

mn

, (7) where

8

p

T,

k

1

t

2

mT

2

,

p

R,

l

1

t

2

nT

29

5

3

`2`

p

T,

k

1

t

2

mT

2

p

R,

l

*

1

t

2

nT

2

dt

,(8)* denotes a complex conjugate, and

d

kl

is the Kronecker deltafunction defined as

Transmitter Receiver......

s

0

(

t

)

ˆ s

0

(

n

)

ˆ s

1

(

n

)

s

1

(

t

)

s

N

–1

(

t

)

ˆ s

N

–1

(

n

)

T

Channel

p

T

(

t

)

e

j

2

π

f

0

t

p

T

(

t

)

e

j

2

π

f

1

t

p

T

(

t

)

e

j

2

π

f

N

–1

t

p

R

(

t

)

e

j

2

π

f

0

t

p

R

(

t

)

e

j

2

π

f

1

t

p

R

(

t

)

e

j

2

π

f

N

–1

t

[FIG1]Block diagram of an FBMC transceiver. This structure isalso applicable to OFDM.

Filters

scribd