Adv. Radio Sci.
, 4, 165–169, 2006
www.adv-radio-sci.net/4/165/2006/
© Author(s) 2006. This work is licensed
under a Creative Commons License.
Multicarrier systems and filter banks
D. S. Waldhauser and J. A. Nossek
Institute for Circuit Theory and Signal Processing, Munich University of Technology, Germany
Abstract. Multicarrier systems, especially orthogonal fre-
quency division multiplex (OFDM), are already established
in a lot of current communications standards. Moreover, they
are very prominent candidates for future systems in mobile
communications.
Filter bank based multicarrier systems can be designed to
provide a better spectral shaping than OFDM systems. That
way, they have some advantages against OFDM, which will
be shown in this work and motivate further research in the
direction of channel equalizers with low complexity. This
leads to optimum bandwidth efficient multicarrier systems
without guard interval, i. e. cyclic prefix.
1 Introduction
Optimum adaptability to the time and frequency selectivity
of the propagation channel is crucial for mobile communica-
tion systems (Kammeyer, 1992). Multicarrier systems pro-
vide the possibility to find a trade-off between robustness
against frequency selectivity and time selectivity by adjust-
ing the number of subcarriers.
Especially orthogonal frequency division multiplex
(OFDM) resp. discrete multitone modulation (DMT) are
widely recognized for high data rate communications.
They are already established in a lot of current standards
like Wireless LAN, HiperLAN, WiMAX, DAB, DVB-T,
and xDSL (Akansu et al., 1998). Moreover, they are very
prominent candidates for the mobile communication systems
of the 4th generation (4G) at least in the downlink, i. e. from
the base station to the mobile stations. Even the extension of
the system of the 3rd generation (3G+) is likely to be based
on multicarrier systems.
Correspondence to: D. S. Waldhauser
(waldhauser@tum.de)
Published by Copernicus GmbH on behalf of the URSI Landesausschuss in der Bundesrepublik Deutschland e.V.
Advances in
Radio Science
OFDM resp. DMT multicarrier systems used in current
standards transmit some kind of redundancy — the cyclic
prefix (CP) with a length of at least the equivalent discrete-
time channel impulse response — for transforming the time-
invariant multipath channel into a periodically time-varying
additive white Gaussian noise (AWGN) channel (Ivrlac,
2005). The time-invariance is assumed for the length of at
least one OFDM symbol. This represents an elegant way of
avoiding channel equalizers, which remove the intersymbol
interference (ISI) and adjacent channel interference (ACI), at
the expense of reduced bandwidth efficiency up to 25%.
Linear time-domain equalizers are state of the art in
OFDM / DMT systems in order to shorten the length of the
equivalent discrete time channel impulse response and, there-
fore, the necessary length of the cylic prefix. In this way
some bandwidth efficiency is saved (Melsa et al., 1996).
Moreover, some research has been done in the field of
bandwidth efficient OFDM transmission without any cyclic
prefix (Toeltsch and Molisch, 2000; Trautmann and Fliege,
2002; Sinn et al., 2002).
Generally, multicarrier systems can be realized by filter
banks, called transmultiplexer filter banks (TMUX) or sim-
ply transmultiplexers (Tzannes et al., 1994).
The most important implementation uses one prototype fil-
ter H0 (z), from which all other transmit and receive filters
Hk (z), k = 0, 1, . . . , M − 1, are derived by complex modu-
lation according to Eq. (1) , which represents a modulation
with the equidistant carrier frequencies (c. f. Fig. 1).
Ideal bandwidth efficiency can be achieved by filters sat-
isfying an overall transfer function according to the first
Nyquist criterion. Offset quadrature amplitude modulation
(OQAM) (Kammeyer, 1992) or staggered quadrature ampli-
tude modulation (SQAM) (Bingham, 1990) have been pro-
posed in the literature to avoid adjacent channel interfer-
ence (ACI).
 synthesis part (polyphase decomposition)

1662. DFT-transmultiplexer
Fig. D. S. Waldhauser and J. A. Nossek: Multicarrier systems and filter banks

x0 [n] Table 1. Real multiplications per symbol (synthesis part of FB).

↑M H0 (z) bank. The subband filters are realized as IDFT operation and
.. ..
subsequent polyphase
M subcarriers: 128 component
256 filters at1024
512 the low symbol
. . rate. Due to the evolving IDFT operation, which is advan-
xk−1 [n] OFDM calculated
tageously 1792by an4096
inverse9216 20480 transforma-
fast Fourier
↑M Hk−1 (z)
tionRRC-TMUX
(IFFT), those 6912
transmultiplexer filter 61440
banks are denoted
14336 29696
as DFT transmultiplexers (Fliege, 1993).
xk [n] s[m]
↑M Hk (z) At this point it becomes obvious, that orthogonal fre-
subcarriers M for the
quency division multicarrier
multiplex systemisisa varied
(OFDM) specialbetween
case of these
128, 256, 512 and 1024.
general DFT-transmultiplexer filter banks with the following
xk+1 [n]
↑M Hk+1 (z) Both in thefilter
prototype OFDM and the general DFT-transmultiplexer
case with the RRC prototype filter (RRC-TMUX), the same
.. ..
. . number of real H0OFDM
multiplications
(z) = 1 +per z −1symbol
+ ···+ 2M z −(M−1)
log2 (M). is (2)
xM−1 [n] needed for the IFFT operation, because it only depends on
↑M HM−1 (z) theThe
number of subcarriers.
polyphase Additionally,
components the(z),
H0,0 H0,1 (z), . . . ,
RRC polyphase
filters require
H0,M−1 4Kbecome
(z) real multiplications
unity in the percasesymbol.
of OFDM. This is the
computational overhead arequired
In the following squarefor theraised
root improved cosinespectral
FIR filter
Fig.
Fig.1.1.Synthesis
SynthesisPart
part of
of aa Transmultiplexer Filter
transmultiplexer filter Bank
bank.
shaping.
(RRC) of length M K is assumed for the the prototype fil-
Because
ter H0 (z) of in
thethe
duality
generalof the synthesis part and
transmultiplexer case thebecause
analy- of its
2 Complexity estimation
are deduced from one prototype filter H0 (z), which deter- sis part of the filter bank, the same observations
beneficial properties as square root Nyquist filter. The num- are sincerely
mines the spectral shaping of the subbands. Subband
The subband filters Hk (z), k = 0, 1, . . . , M − 1, of complex fre- true forof
ber thesubcarriers
receiver. M for the multicarrier system is varied
quency
modulatedresponses can be designed
transmultiplexer which
filter banks withoverlap only with
M subcarriers One
betweenadditional issue512
128, 256, should
and be
1024.taken into consideration:
directly adjacent subbands and provide separation
are deduced from one prototype filter H0 (z), which deter- from other Especially in OFDM systems
Both in the OFDM and the general with cyclic prefix, a large num-
DFT-transmultiplexer
subbands by a high attenuation characteristic.
mines the spectral shaping of the subbands. Subband fre- The complex ber of subcarriers M is desired in order
case with the RRC prototype filter (RRC-TMUX), to keep the propor-the same
modulation
quency responses can be designed which overlap only with tionnumber of real multiplications per symbol 2M e.g.
of the cyclic prefix small. If no other reasons, log2a(M ) is
directly adjacent subbands and provide separation from other very strongfor
needed frequency
the IFFT diversity, prevent
operation, the reduction
because it only of the
depends on
k
H k (z) = H 0 (W z), W M = exp (−j
subbands by a high attenuation characteristic. The complex
M 2π/M ) (1) number of subcarriers in the RRC-TMUX case, the num-
the number of subcarriers. Additionally, the RRC polyphase
ber of subcarriers can be reduced compared to OFDM in or-
inmodulation
combination with a polyphase decomposition — if the filters require 4K real multiplications per symbol. This is the
k designed as finite impulse response filter
der to compensate for the additional multiplications of the
prototype
Hk (z) = filter
H0 (WisM z), W M = exp (−j 2π/M) (1) computational
polyphase filters byoverhead required for of
saving multiplications thetheimproved
IFFT oper- spectral
(FIR) or infinite impulse response filter (IIR) with special ations. shaping.
Table 1 shows the number of real multiplications per
in combination with a polyphase decomposition – if the
properties — leads to extremely efficient implementations
prototype filter is designed as finite impulse response filter
of symbol Because of the duality
for the OFDM and the of the synthesis
RRC-TMUX casepart and the analy-
for different
the transmultiplexers. Figure 2 shows the resulting
(FIR) or infinite impulse response filter (IIR) with special struc- sis part of the filter bank, the same observations
numbers of subcarriers and an RRC polyphase filter length are sincerely
of
ture of the the synthesis part of the transmultiplexer
properties – leads to extremely efficient implementations of filter K true
= 10. forThethe receiver.
additional complexity for the RRC polyphase
the transmultiplexers. Figure 2 shows the resulting struc- filters can be seen by the difference in each column. The re-
ture of the the synthesis part of the transmultiplexer filter duction of the number of subcarriers in the RRC-TMUX case
bank. The subband filters are realized as IDFT operation and by a factor of 4 to M/4 leads to an overall complexity even
subsequent polyphase component filters at the low symbol smaller than that of an OFDM system with M subcarriers.
rate. Due to the evolving IDFT operation, which is advan- The main difference is that for OFDM with cyclic pre-
tageously calculated by an inverse fast Fourier transforma- fix no additional channel equalizer is needed as long as the
tion (IFFT), those transmultiplexer filter banks are denoted cyclic prefix is sufficient in length. A comparison at the ba-
as DFT transmultiplexers (Fliege, 1993). sis of the achievable data rates shows that OFDM is able
At this point it becomes obvious, that orthogonal fre- to transmit M log2 (MO ) bit per symbol, whereas the trans-
quency division multiplex (OFDM) is a special case of these multiplexer transmits (M + Mcp ) log2 (MT ) bit per symbol,
general DFT-transmultiplexer filter banks with the following where MO resp. MT denote the modulation alphabets for
prototype filter OFDM resp. the RRC-TMUX. Moreover, the same modula-
tion alphabets are assumed for all subcarriers. Equal rates
H0OFDM (z) = 1 + z−1 + . . . + z−(M−1) . (2)
are obtained by compensating the cyclic prefix length by
The polyphase components H0,0 (z), H0,1 (z), . . ., the modulation scheme. If 25% cyclic prefix is assumed,
H0,M−1 (z) become unity in the case of OFDM. i.e. Mcp /M = 1/3, 256 QAM resp. 16 QAM is necessary for
In the following a square root raised cosine FIR filter OFDM against 64 QAM resp. 8 QAM for a TMUX without
(RRC) of length MK is assumed for the the prototype filter CP. This means that more transmit power is necessary for
H0 (z) in the general transmultiplexer case because of its ben- OFDM in order to get the same (uncoded) bit error rates for
eficial properties as square root Nyquist filter. The number of the higher modulation schemes.

Adv. Radio Sci., 4, 165–169, 2006 www.adv-radio-sci.net/4/165/2006/

D. S. Waldhauser and J. A. Nossek: Multicarrier systems and filter banks 167
2 D. S. Waldhauser and J. A. Nossek: Multicarrier Systems and Filter Banks

↓M H0,0 (z) ↑M RRC D/A

z −1 z −1

↓M H0,1 (z) ↑M Complexity Estimation:

−1 −1
z z
IFFT: 4M/2 log2 (M )

IFFT H0,i (z): 4K

real mult. / symbol
z −1 z −1

↓M H0,M−2 (z) ↑M ⇒ DFT-TMUX

z −1 z −1
OFDM: H0,i (z) = 1
↓M H0,M−1 (z) ↑M
RRC : root raised cosine

S/P P/S
synthesis part (polyphase decomposition)

Fig. 2. DFT-transmultiplexer
Fig. 2. DFT-transmultiplexer.

x0 [n]
bank. The subband filters are realized as IDFT operation and
3 Power spectral ↑density
M H0 (z)
channel
subsequent of the TMUXcomponent
polyphase have only filters
overlap at with
the lowthesymbol
directly
.. ..
. . adjacent
rate. Due subchannels and separate
to the evolving the otherwhich
IDFT operation, subchannels
is advan- by
The power xspectral
k−1 [n] density of the input symbol vector of the large attenuation
tageously calculated characteristic,
by an inverse fast whereas the transforma-
Fourier power spec-
both the OFDM and ↑M the TMUX Hk−1 (z)
 systemH is assumed to be tral
tiondensities
(IFFT), thosefor OFDM show largefilter
transmultiplexer overlap
bankswith all other
are denoted
white with unity power, i.e. E x[n]x[n] = 1M . It is eas- subchannels.
as DFT transmultiplexers (Fliege, 1993).
xk [n]
ily shown, that both for OFDM and the RRC-TMUX s[m] the
↑M Hk (z) At this
The point itRRC
subsequent becomes
pulse obvious,
shaping filterthat –orthogonal
operating atfre- the
transmitted symbol vector – disregarding the cyclic prefix quency
high, division
M-fold multiplex
symbol rate –(OFDM)
has to limitis a special case of these
the bandwidth to the
for OFDM x–k+1keeps
[n] the white power spectral density. With general DFT-transmultiplexer
existing bandwidth restrictions filter banks
while with theintersymbol
avoiding following
OFDM the IDFT operation
↑M is a H
unitary
k+1 (z)matrix transformation prototype filter
interference. In Figs. 3a and 3b a rolloff factor ρ = 0.20 has
which doesn’t change. the spectral .properties of the vector it been used. OFDM The resulting power spectral densities are given
is applied to. The RRC
.. .
filters of. the TMUX are approxi- H (z) = 1 + z −1 + · · · + z −(M−1) . (2)
xM−1 [n] in Figs. 3c 0and 3d. Obviously, the high stopband attenua-
mately square root M-th
↑ M band filters
HM−1 (z)which means that tion
Theof polyphase
the RRC subchannel
components filters
H0,0allows
(z), relatively
H0,1 (z), relaxed
...,
M−1
X requirements
H0,M−1 (z) become on the RRC pulse
unity in shaping
the case filter. The resulting
of OFDM.
2 k bandwidth for the OFDM system
H 1.(zW
Fig. 1. of a Transmultiplexer Filter Bank
M ) = Part
Synthesis (3) In the following a square rootis about
raised 1.5 timesFIR
cosine the filter
band-
k=0 width
(RRC)forofthe TMUX
length M Ksystem,
is assumedwhenfor anthe
attenuation
the prototype threshold
fil-
ofter40HdB
0 (z)isinconsidered.
the general transmultiplexer case because of its
Therefore, the approximately
are deduced M-th band
from one prototype filter spectral power deter-
H0 (z), which den-
beneficial
Therefore, properties
a smallerasrolloff
squarefactor
root Nyquist filter. shaping
for the pulse The num- fil-
sities of each subcarrier add up to a white power
mines the spectral shaping of the subbands. Subband fre- spectral
berisof
ter subcarriers
required M same
for the for the multicarrier
bandwidth demandsystemin an is varied
OFDM
density
quencyat the transmitter
responses output.
can be designed which overlap only with betweenThis 128,increases
256, 512 and 1024.
system. the complexity for the pulse shaping
In practice a certain number
directly adjacent subbands and provideof subcarriers at from
separation the band
other Both
filter andindeteriorates
the OFDM the andpeak-to-average
the general DFT-transmultiplexer
ratio of the trans-
edges is omitted for data transmission as frequency
subbands by a high attenuation characteristic. The complex guard casesignal.
with the RRC prototype filter (RRC-TMUX), the same
mit
band in order to simplify the subsequent signal processing
modulation number of real multiplications per symbol 2M log2 (M ) is
tasks concerning the band k
limitation and D/A conversion. needed for the IFFT operation, because it only depends on
Figures 3a Hkand
(z) = 3bHshow
0 (WMthez), power
WM spectral
= exp (−j 2π/M ) both
densities (1)
the number of subcarriers. Additionally, the RRC polyphase
for in
an combination
OFDM and an RRC-TMUX
with a polyphasemulticarrier
decompositionsystem.
— ifThethe 4filters
Conclusions
require 4K real multiplications per symbol. This is the
number of total subcarriers is M = 32, and M = 8 subcarri-
prototype filter is designed as finite impulseg response filter computational overhead required for the improved spectral
ers (FIR)
are omitted at the
or infinite band edges.
impulse responseTherefore, in with
filter (IIR) both special
cases Complex
shaping. modulated transmultiplexer filter banks with RRC
the properties
total power spectral density is not white any
— leads to extremely efficient implementationsmore. Theof subband
Because filters provide
of the dualityanof improved
the synthesis spectral
part andshaping com-
the analy-
power
the spectral densities ofFigure
transmultiplexers. the output
2 shows signals of each struc-
the resulting sub- pared
sis partto of
OFDM systems.
the filter This
bank, the means
same additionalare
observations costs for the
sincerely
ture of the the synthesis part of the transmultiplexer filter true for the receiver.
www.adv-radio-sci.net/4/165/2006/ Adv. Radio Sci., 4, 165–169, 2006
168 D. S. Waldhauser and J. A. Nossek: Multicarrier systems and filter banks

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
(a) PSD of OFDM: Mch = 32, Mg = 8 (b) PSD of RRC-TMUX: Mch = 32, Mg = 8

0 0

−10 −10
dB

dB
−20 −20

−30 −30

−40 −40
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
(c) PSD of OFDM: Mch = 32, Mg = 8 (d) PSD of RRC-TMUX: Mch = 32, Mg = 8

Fig. 3. Comparison: Power spectral densities of OFDM vs. DFT.

pulse shaping filters, but the polyphase decomposition allows
for a very efficient implementation.
The power spectral density of the transmit signal with a
frequency guard band is nearly band-limited by the subband
filters alone, the RRC pulse shaping filter is mainly needed
to eliminate periodic repetitions. Therefore, the requirements
on the RRC pulse shaping filter are relaxed and a relatively
large rolloff factor can be applied. This leads to less com-
plex pulse shaping filters and a lower peak-to-average ratio
of the transmit signal because of the faster decaying impulse
response of the pulse shaping filter.
The cyclic prefix of OFDM systems is an elegant solu-
tion to cope with the distortions of the multipath channel, but
it reduces the bandwidth efficiency, because the number of
subcarriers cannot be increased arbitrarily due to the time-
selectivity of mobile communication channels. Therefore,
the modulation alphabet has to be increased to achieve the
same data rate as an RRC-TMUX. This requires more trans-
mit power for the same (uncoded) bit error rate. The increase
of the transmit power not only boosts the complexity of the
transmit amplifier but also enhances the inter-cell interfer-
ence in cellular systems like mobile communication systems.
The RRC-TMUX necessitates an additional channel
equalizer to cancel intersymbol and adjacent channel inter-
ference caused by the frequency selective channel. The al-
leged complexity of this equalizer has been the main obstacle
in the past to the application of DFT-transmultiplexer filter
banks with RRC subband filters. The good attenuation prop-
erties of the subband filters, which only show overlap with
directly adjacent subchannels, should reduce the complexity
of the equalization considerably. This will be the subject of
future work.
References
Akansu, A. N., Duhamel, P., Lin, X., and de Courville, M.: Or-
thogonal Transmultiplexers in Communication: A Review, IEEE
Transactions on Signal Processing, 46, 979–995, 1998.
Bingham, J. A. C.: Multicarrier Modulation for Data Transmission:
An Idea Whose Time Has Come, IEEE Communications Maga-
zine, 5–14, 1990.
Fliege, N.: Multiraten-Signalverarbeitung, Informationstechnik,
B. G. Teubner, Stuttgart, 1993.
Ivrlac, M. T.: Wireless MIMO Systems — Models, Performance,
Optimization, Ph.D. thesis, Munich University of Technology,
Shaker Verlag, 2005.

Adv. Radio Sci., 4, 165–169, 2006 www.adv-radio-sci.net/4/165/2006/

D. S. Waldhauser and J. A. Nossek: Multicarrier systems and filter banks
Kammeyer, K. D.: Nachrichtenübertragung, Informationstechnik,
B. G. Teubner Stuttgart, 1992.
Melsa, P. J., Younce, R. C., and Rohrs, C. E.: Impulse response
shortening for discrete multitone transceivers, IEEE Transactions
on Communications, 44, 1662–1672, 1996.
Sinn, C. V., Götze, J., and Haardt, M.: Efficient Data Detection Al-
gorithms in Single- and Multi-Carrier Systems without the Ne-
cessity of a Guard Period, in: IEEE ICASSP, 3, 2737–2740,
2002.
www.adv-radio-sci.net/4/165/2006/
169
Toeltsch, M. and Molisch, A. F.: Efficient OFDM Transmission
without Cyclic Prefix over Frequency-selective Channels, in:
IEEE Int. Symp. on PIMRC, 2, 1363–1367, 2000.
Trautmann, S. and Fliege, N. J.: Perfect Equalization for DMT Sys-
tems Without Guard Interval, IEEE J. on Selected Areas in Com-
munications, 20, 987–996, 2002.
Tzannes, M. A., Tzannes, M. C., Proakis, J., and Heller, P. N.: DMT
Systems, DWMT Systems and Digital Filter Banks, in: IEEE Int.
Conf. Comm., New Orleans, USA, 1, 311–315, 1994.
Adv. Radio Sci., 4, 165–169, 2006