Sunteți pe pagina 1din 7

GC'12 Workshop: The 8th Broadband Wireless Access Workshop 1

A Comparative Study of FBMC Prototype Filters in


Doubly Dispersive Channels
Alphan Şahin1 , İsmail Güvenç2,∗ , and Hüseyin Arslan1
1 Department of Electrical Engineering, University of South Florida, Tampa, FL, 33620
2 DOCOMO Innovations, Inc., 3240 Hillview Avenue, Palo Alto, CA, 94304

Email: alphan@mail.usf.edu, iguvenc@ieee.org, arslan@usf.edu

Abstract—Due to its numerous advantages, orthogonal fre-

Remove Cyclic Prefix (OFDM)


Polyphase Network (FBMC)

Polyphase Network (FBMC)


Insert Cyclic Prefix (OFDM)
quency division multiplexing (OFDM) has been the broadband
wireless access technology of choice for many wireless standards

Parallel-to-Serial
Serial-to-Parallel

Serial-to-Parallel

Parallel-to-Serial
ܰ-point IFFT

ܰ-point FFT
over the last decade. Recently, filterbank multicarrier (FBMC)

Channel
ܺ ܺ෩
technique is emerging as one of the alternatives to OFDM for

or

or
next generation broadband wireless access systems. While there
are several studies on FBMC in the literature, its potential
merits over OFDM and possible implementation challenges in
practical environments require further investigations. Purpose of
݄ሺ߬ǡ ‫ݐ‬ሻ
the present paper is to review and emphasize the key benefits of
FBMC technology, and provide a comparative study of different
FBMC prototype filter designs in the literature under practical
channel environments. Fig. 1: Transmitter and receiver block diagrams in OFDM and
FBMC.
Index Terms—5G, ambiguity function, beyond 4G, doubly
dispersive channels, FBMC, future radio access, offset QAM
(OQAM), OFDM, prototype filter, PSD.
Weinstein’s DFT method in [3] neglected the baseband
filters while modulating or demodulating the subcarriers, but
I. I NTRODUCTION
instead used a time windowing technique to cope with the
Orthogonal frequency division multiplexing (OFDM) has inter-carrier interference (ICI). In [4], Hirosaki showed that
been the technology of choice for many wireless standards FBMC may also be digitally implemented through DFT pro-
within the last decade, such as WiFi, WiMAX, DVB-T/DAB- cessing by using a ployphase network (PPN) for capturing
T, WRAN 802.22, and LTE. Its advantages such as simple prototype filter characteristics. Several other developments
equalization in frequency domain, efficient implementation over the last two decades have demonstrated low complexity
through fast Fourier transformation (FFT) techniques and and efficient implementations of FBMC, paving the way for its
the multiuser frequency diversity make OFDM an appealing consideration in next generation wireless standards (see e.g.,
technology for many wireless systems. On the other hand, [5]–[8], and the references listed therein).
OFDM also has several important shortcomings, such as high FBMC has two main advantages over OFDM. First, as
sidelobes, susceptibility to carrier frequency offset (CFO), opposed to OFDM, which may only use various time-domain
cyclic prefix (CP) overhead, and high peak to average power windowing techniques [9] to combat with inter-symbol inter-
ratio (PAPR). Due to such shortcomings of OFDM, waveform ference (ISI) and ICI, FBMC may use any prototype filter
design for next-generation broadband wireless access (BWA) design using PPN. Therefore, the transmission/reception may
systems carries critical importance. be better adapted to channel characteristics and other design
Filter bank multicarrier (FBMC) is one of the BWA options constraints [10]. This flexibility of choosing any prototype
that may address the shortcomings of OFDM. Earlier works filter to combat with ISI and ICI makes FBMC an attractive
related to FBMC actually date back to 1960s [1], [2], which choice over OFDM. Secondly, even if same filters are used in
utilize a bank of filters for parallel data transmission. In [3], FBMC and OFDM, FBMC enjoys less ISI/ICI than OFDM,
through the use of discrete Fourier transformations (DFTs), since it allows the use of well-localized prototype filters in
Weinstein and Ebert eliminated the banks of subcarrier oscil- both time and frequency domains. The reason for this is the
lators and coherent demodulators present in [2], which were sparser structure of FBMC’s effective time-frequency grid,
unreasonably expensive and complex for large number of data such as the one in lattice for offset quadrature amplitude mod-
channels at that time. The approach in [3] has been later named ulation (OQAM) instead of quadrature amplitude modulation
as OFDM, and it became more and more popular after 1980s (QAM) [8], [11], [12]. Therefore, CP can be dropped in FBMC
due to its efficient implementation through FFT techniques. to improve spectral efficiency, with some minimal performance
∗ As of August 2012, Ismail Guvenc is an assistant professor in the
degradation in multipath environments.
Department of Electrical and Computer Engineering of Florida International As shown in Fig. 1, transmitter/receiver block diagrams
University, Miami, FL. of OFDM and FBMC are very similar to each other. At
978-1-4673-4941-3/12/$31.00 ©2012 IEEE 197
2

the transmitter, inverse fast Fourier transformation (IFFT) where ℜ{·} is the real part of the argument,
operation is followed by CP insertion in OFDM (to combat Z

with ISI), while it is followed by a PPN in FBMC. In the hy(t), ǵnl (t)i , y(t)ǵnl (t)dt
same manner, CP removal is substituted with PPN in FBMC to t
X
extract the information symbols at the receiver. Note that PPN = Xnl Hnlnl + Xmk Hnlmk , (6)
can be flexibly used to capture any type of prototype filter; e.g., (m,k)6=(n,l)
applying filter coefficients for rectangular prototype function
to PPN structure yields the conventional OFDM (without a and
Z Z Z
CP). Therefore, OFDM can also be implemented as a special ∗
Hnlmk = H(τ, ν) gmk (t − τ )ǵnl (t)ej2πνt dtdνdτ ,
case of FBMC. τ ν t
In this paper, due to their critical effect on the system per- (7)
formance, we focus on the review and evaluation of different
FBMC prototype filters under practical channel environments. which captures the interference from the symbol (m, k) to the
We consider some of the well-known prototype filters in the desired symbol (n, l). In (1) and (5), gmk (t) and ǵnl (t) are
literature, such as rectangular (as in OFDM), optimal finite derived from the prototype filter ptx (t) and prx (t) shifted in
duration pulse (OFDP) [13], root-raised cosine (RRC) [14], both time and frequency domains depending on (m, k) and
half cosine [11], PHYDYAS [15], Gaussian [16], Hermite (n, l), respectively. In other words, gmk (t) and ǵnl (t) denote
[17], and isotropic orthogonal transform algorithm (IOTA) [5] a bank of filters at the transmitter and receiver, generated
filters. Average signal-to-interference ratio (SIR) performance through time/frequency shifted replicas of the prototype filters
of these prototype filters are compared under CFO, timing ptx (t) and prx (t).
offset, and ITU channel models [18]. Impact of filter truncation Since the purpose of the paper is to investigate the SIR
on the performance and power spectral densities (PSDs) of the performance of the filters in doubly dispersive channels, noise
FBMC transmission synthesized with different prototype filters and other interference sources are not taken into account.
are also studied. Then, SIR can be written as
σS2
II. S YSTEM M ODEL SIR = , (8)
σI2
In a multicarrier scheme with N subcarriers per symbol as
in Fig. 1, analytical expression of the transmitted baseband where σS2 and σI2 are the power of the desired part and the
signal can be written as interference leaking from other symbols, respectively. The
∞ N −1 analytical expressions of σS2 and σI2 will be given explicitly for
OFDM/QAM and OFDM/OQAM in the following sections.
X X
x(t) = Xmk gmk (t) , (1)
m=−∞ k=0 Additionally, (n, l) is set to (0, 0) in the rest of the paper, since
the impact of ICI and ISI on each subcarrier is statistically
where Xmk is the symbol located on the subcarrier pointed by
equal to each other by assuming all subcarriers are utilized.
the time index m, and the frequency index k and gmk (t) is the
synthesis function. After the signal passes through the linear
time-varying wireless channel h(τ, t), the received signal is III. SIR FOR OFDM/QAM
obtained as
Z For OFDM/QAM, gmk (t) and ǵnl (t) can be expressed as
y(t) = h(τ, t)x(t − τ )dτ
Zτ Z gmk (t) = ptx (t − mτ0 ) ej2πkν0 t , (9)
j2πlν0 t
= H(τ, ν)x(t − τ )ej2πνt dνdτ , (2) ǵnl (t) = prx (t − nτ0 ) e , (10)
τ ν
where H(τ, ν) is the Fourier transform of h(τ, t). Statistical respectively, where τ0 and ν0 are the symbol spacings in time
characteristics of the channel are described with wide-sense and frequency, respectively, and τ0 ν0 = 1. Then, by inserting
stationary uncorrelated scattering (WSSUS) assumption [19] (9) and (10) into (7), Hnlmk can be written as
given by

Z Z
Hmk = Hnlmk = H(τ, ν)×
E [H(τ, ν)] = 0 , (3) l=0 τ ν
n=0

E [H(τ, ν)H (τ1 , ν1 )] = S (τ, ν) δ(τ − τ1 )δ(ν − ν1 ) , (4)
Z
e−j2πkν0 τ ptx (t − mτ0 − τ ) p∗rx (t) ej2π(kν0 +ν)t dtdνdτ .
where S (τ, ν) is the scattering function. t
(11)
Depending on the applied modulation (i.e., OFDM/QAM or
OFDM/OQAM), received symbol X̃nl located on time index In the literature, e.g. [16], [20], (11) is conventionally ex-
n and frequency index l is obtained by the projection of the pressed in terms of ambiguity function Aℜ (φ, ψ) given by
received signal on analysis function ǵnl (t) as Z ∞    
∗ φ φ −j2πψt
Aℜ (φ, ψ) ,
(
hy(t), ǵnl (t)i for OFDM/QAM ptx t − prx t + e dt .
X̃nl = , (5) −∞ 2 2
ℜ{hy(t), ǵnl (t)i} for OFDM/OQAM (12)
198
3

߬ ߥ …
߬ ߥ …
ߥ …
߬
߬଴ ߬଴ ߬଴
ܵሺ߬ǡ ߥሻ ܵሺ߬ǡ ߥሻ ܵሺ߬ǡ ߥሻ
ߥ଴ ߥ଴ ߥ଴

݂ ݂ ݂
ʹߥ଴ ʹߥ଴ ʹߥ଴

ߥ଴ ߥ଴ ߥ଴

ȁ‫ܣ‬ȁ Ͳ ȁԸ ‫ ܣ‬ȁ Ͳ ȁԱ ‫ ܣ‬ȁ Ͳ

െߥ଴ െߥ଴ െߥ଴

െʹ߬଴ െ߬଴ Ͳ ߬଴ ʹ߬଴


‫ݐ‬ െʹ߬଴ െ߬଴ Ͳ ߬଴ ʹ߬଴
‫ݐ‬ െʹ߬଴ െ߬଴ Ͳ ߬଴ ʹ߬଴
‫ݐ‬

(a) OFDM/QAM; ICI/ISI from carriers (b) OFDM/OQAM; ICI/ISI from real carriers (c) OFDM/OQAM; ICI/ISI from imaginary carriers

Fig. 2: SIR Calculations for OFDM/QAM and OFDM/OQAM.

Z Z
σS2 = E |ℜ {X00 H00 }|2 = S (τ, ν) |ℜ {A (τ, ν)}|2 dνdτ ,
 
(20)
τ ν
 2  X Z Z n o 2
σI2 = E X̃00 − X00 H00 = S (τ, ν) ℜ j m+k A (mτ0 + τ, kν0 + ν) dνdτ

{m,k}6={0,0} τ ν
 
X Z Z X Z Z
S (τ, ν) |ℜ {A (mτ0 + τ, kν0 + ν)}|2 dνdτ + S (τ, ν) |ℑ {A (mτ0 + τ, kν0 + ν)}|2 dνdτ 
 
=
  .(21)
(m,k)6=(0,0) τ ν (m,k)6=(0,0) τ ν
m+k∈Even m+k∈Odd

In this study, in order to express (11) in a simple form, circles) are equal to zero in single-tap channel with no Doppler
Aℜ (φ, ψ) is modified by introducing additional phase shift spread. Due to multipath and Doppler spread (captured by the
ρ as illustrated scattering function), the desired symbol (red circle)
Z ∞ observes ISI/ICI from neighboring grid points.
−j2πρ
A (φ, ψ) ,e ptx (t − φ) p∗rx (t) ej2πψt dt , (13)
−∞

where A (φ, ψ) is the modified ambiguity function1. Therefore, IV. SIR FOR OFDM/OQAM
selecting ρ = kν0 τ , (11) is simplified to
Z Z For OFDM/OQAM, which is one of the possible ways to
Hmk = H(τ, ν)A (mτ0 + τ, kν0 + ν)dνdτ . (14) implement of FBMC, gmk (t) and ǵnl (t) can be expressed as
τ ν
Then, assuming that Xmk are independent identically dis- gmk (t) = j m+k ptx (t − mτ0 ) ej2πkν0 t , (17)
n+l j2πlν0 t
tributed with zero mean and hptx (t) , prx (t)i = 1, the power ǵnl (t) = j prx (t − nτ0 ) e , (18)
of desired part and the power of interference from other
symbols are obtained as respectively, where τ0 ν0 = 1/2. Using (17) and (18), (7) can
h i Z Z be expressed as
2 2
σS2 = E |X00 H00 | = S (τ, ν) |A (τ, ν)| dνdτ ,
τ ν
Hmk =Hnlmk
(15) l=0
 2  Z n=0
Z
σI2 = E X̃00 − X00 H00 =j m+k H(τ, ν)A (mτ0 + τ, kν0 + ν)dνdτ .

τ ν
(19)
Z Z
S (τ, ν) |A (mτ0 + τ, kν0 + ν)|2 dνdτ ,
X
=
τ ν
(m,k)6=(0,0) Therefore, σS2 and σI2 are obtained as in (20) and (21), respec-
(16)
tively, where ℑ{·} expresses the imaginary part of argument.
respectively, where E [·] is the expected value operator. The By calculating ℜ{A (φ, ψ)} and ℑ{A (φ, ψ)}, the impact of
illustration of (16) is given in Fig. 2(a), where the absolute the interference from real (i.e. mod {m + k, 2} = 0) and
value of the ambiguity function at the grid points (black imaginary (i.e. mod {m + k, 2} = 1) carriers in a doubly
dispersive channel environment can be visualized on time-
1 A (φ, ψ) can be derived as A (φ, ψ) = A∗ (φ, ψ)

ℜ ℜ ρ=ψφ/2
. frequency grid as in Fig. 2(b) and Fig. 2(c).
199
4

V. P ROTOTYPE F ILTERS The truncation length K is fixed to 16 in Fig. 3(a), Fig. 3(b),
In this study, we assume that same prototype filter is used and Fig. 4, unless otherwise stated.
at the transmitter and the receiver, i.e., ptx (t) = prx (t). Time-frequency lattice is arranged considering the filter lo-
In particular, rectangular, extended rectangular, OFDP, RRC, calization and the modulation type. For example, symbols are
PHYDYAS, Gaussian, Hermite, and IOTA filters are consid- placed to the grid where τ0 = T and ν0 = F for OFDM/QAM
ered. While half-cosine and half-sinc are derived from the with rectangular filter. On the other hand, the symbols √ are
RRC by changing αr factor, extended rectangular filter is placed √ to the time-frequency grid where τ0 = 2T /2 and
obtained from rectangular filter to achieve conventional OFDM ν0 = 2F/2 for Hermite, IOTA and Gaussian filters. Also,
with CP. PHYDYAS filter and OFDP are considered since they the grid is constructed with τ0 = T /2 and ν0 = F for
provide very good stopband performance for given filter length RRC (αr = 0.22), half sinc (i.e., RRC αr = 0), half cosine
and they are commonly used in the literature.. Basically, the (i.e., RRC αr = 1) for OFDM/OQAM, PHYDYAS filters and
aim of OFDP filter is to achieve a Nyquist pulse in both time OFDP. The filter lengths for PHYDYAS filter and OFDP are
and frequency with the maximum energy in the main lobe set to 4 × T . Also, we consider windowing [9] in Fig. 3(a) and
for a given bandwidth and filter length. By representing the Fig. 3(b) and extended rectangular filter in Fig. 4 to compare
pulse as the linear combinations of the prolate spheroidal wave conventional OFDM with other filters. It is worth to note that
functions (PSWFs) and formulating the Nyquist criterion as a employing CP+windowing, i.e. TCP = TW = 0.14T results in
constraint, the weights (ak ) for kth PSWF (ψk,prolate (t)) are a more than 20 percent less spectral efficient scheme compared
obtained through an optimization procedure. Note that since to the OFDM/OQAM. In Fig. 4, SIR plots are obtained by
only even indexed prolate functions are even-symmetrical, taking four channel scattering models into account; 1) Carrier
only a2k are considered. frequency offset, 2) Timing offset, 3) ITU Vehicular A with
Among the filters considered, only the Gaussian filter does Jake’s Doppler spreading, and 4) ITU Pedestrian A with Jake’s
not satisfy the Nyquist criterion. It is included to show Doppler spreading. For ITU channel models, modified Jake’s
the impact of the orthogonality loss and to compare the Doppler spectrum which also includes the time variation of
performances with the IOTA and Hermite filters. In Hermite the phase components of reflected radio waves [21] is applied
filter, by deforming the Gaussian filter with other modes, zero in order to prohibit the infinite density at the spectrum. In this
crossings in time are provided to satisfy Nyquist criterion. model, correlation between phase components is defined with
Similar to the Hermite, the IOTA filter targets to achieve the B parameter which is a positive constant. In the simulations,
excellent time-frequency localization of the Gaussian pulse. heuristically, B is set to 0.5fd where fd is the maximum
Starting with the Gaussian function, orthogonalized pulse is Doppler shift.
obtained as in TABLE I where F and F −1 are the operators 1) Rectangular Filter: Rectangular prototype filter corre-
for Fourier transform and its inverse, respectively, and Oa sponds to sinc functions in frequency domain. Therefore,
is an orthogonalization operator on a function. Analytical OFDM has high sidelobes as given in Fig. 3(a) and Fig. 3(b).
expressions and the parameters of the all prototype filters It is possible to suppress the side-lobes of OFDM by applying
considered are summarized in the TABLE I. additional windowing period. On the other hand, sidelobes are
Ideally, infinite duration in time domain is required for naturally suppressed in FBMC approaches without sacrificing
the prototype filters to achieve perfect symbol reconstruction. the spectral efficiency. In terms of SIR, extended rectangular
However, in practice, prototype filters have to be truncated at filter, which corresponds to conventional OFDM, provides
the transmitter and receiver. Among the filters in TABLE I, the best performance for the scenarios with lower Doppler
only the rectangular, extended rectangular, PHYDYAS, and spread because it is more robust to time dispersion at the
OFDP are not affected by the filter truncation because of expense of CP overhead. However, FBMC approaches might
their definitions. However, SIR performances of the rest of be more robust to higher frequency dispersion depending
the considered prototype filters are affected by the filter length on the filter used, such as Hermite, IOTA, OFDP, half-
Tfilter = KT , where T is the OFDM/QAM symbol spacing, cosine and PHYDYAS filters in Fig. 4(a), Fig. 4(c), and
and K is the truncation length as a multiple of symbol spacing. Fig. 4(d). Note that when CP is not used, OFDM loses its
advantages in multipath environments. Additionally, Fig. 4
shows that since OFDM/OQAM allows more room for time-
VI. S IMULATION R ESULT frequency spreading of prototype filters [11], OFDM/OQAM
In this section, comparisons of the aforementioned prototype provides superior performance than OFDM/QAM when the
filters are presented in terms of side lobes suppression and same prototype filter is employed (e.g. rectangular/QAM ver-
SIR in doubly dispersive channels considering filter truncation. sus rectangular/OQAM).
First, we compare the PSDs of FBMC and conventional 2) PHYDYAS Filter and OFDP: PHYDYAS filter and
OFDM for 5 MHz and 20 MHz bandwidth utilizations in OFDP are limited in time by their definitions as in TABLE I.
Fig. 3(a) and Fig. 3(b). Symbol spacing T , which is 1/F , Therefore, they provide significant advantage in terms of
is set to 66.7 µs. While 300 and 1200 subcarriers are utilized complexity which is directly proportional to the filter length.
for ν0 = F , √ proportionally, 424 and 1697 subcarriers are used Also, as given in Fig. 3(a) and Fig. 3(b), the out-of-band
for ν0 = F/ 2 in Fig. 3(a) and Fig. 3(b). Then, SIR in (8) are leakage performance of these filters are also superior than
plotted for both OFDM/OQAM and OFDM/QAM for different conventional windowing approaches without degrading the
prototype filters considering truncation as in Fig. 4 and Fig. 5. spectral efficiency of the system. Additionally, both OFDP and
200
5

TABLE I: Analytical expressions of the prototype filters considered in the simulations


Prototype Filter Analytical Model in Time Domain τ0 ν0 Simulation Parameters
(
√1 , |t| ≤ T
Rectangular T2 T F
0, otherwise
(
√1 , − T2 − TCP ≤ t ≤ T
Ext. rectangular T 2 T +TCP F TCP = 0.14T
0, otherwise
Np
X
T
OFDP [13] a2k ψ2k,prolate (t) 2
F K=4
k=0
αr + 4 απr , 

1 −h
     i t = 0
αr
1 + π2 sin 4α π
+ 1 − π2 cos 4α π T

√ , t = ± 4α


2 r r r T
RRC [14]  
r t cos (1+α ) πt
 F αr = 0.22
 sin (1−αr ) πt
T
+ 4α
T r T
2
, otherwise

  
2 2

πt 1− 16αr t


T T2
T
Half sinc Root raised cosine (αr = 0) 2
F
T
Half cosine [11] Root raised cosine (αr = 1) 2
F
L=4
a0 = 1
( PL−1  
kt
1+2 k=1 ak cos 2π Lt , − LT
2
≤t≤ LT
2
, T
PHYDYAS [15] 2
F a1 = −0.97195983
0, otherwise a2 = 0.70710681
a3 = −0.23514695
2 T F
Gaussian [16] (2αg )1/4 e−παg t √ √ αg = 1
2 2
PNH NH = 4, a0 = 1.1850899
k=0 √ (t)
a4k ψ4k,hermite  , 2 a4 = −1.9324881 × 10−3 × a0
2πt e−πt , T F
Hermite [17] ψn,hermite (t) = Hn √ √ a8 = −7.3110588 × 10−6 × a0
2 2
2 dn
−t2 a12 = −3.1542096 × 10−9 × a0
Hn (t) = (−1)n et dt ne
a16 = 9.6634138 × 10−13 × a0
F −1 Oτ0 F Oυ0 pgaussian (t, αg )
x(t)
Oa x(t) = q P∞ , x(t) ∈ R T F
IOTA [5] a k=−∞ kx(t−ka)k2 √ √ αg = 1
Z Z 2 2
F −1 X(f ) = X(f )e−j2πf t df , F x(t) = x(t)ej2πf t dt

0 PHYDYAS (K = 4) 0 PHYDYAS (K = 4)
OFDP (K = 4) OFDP (K = 4)

−50 −50

−100 −100
Power (dB)

Power (dB)

−150 −150

Rect. Rect.
Ext. Rect. (TCP = 0.14T, TW = 0) Ext. Rect. (TCP = 0.14T, TW = 0)
−200 −200
Ext. Rect. (TCP = 0.14T, TW = 0.14T) Ext. Rect. (TCP = 0.14T, TW = 0.14T)

−250 Gaussian RRC (α = 0.22) −250 Gaussian RRC (α = 0.22)


r r
Hermite Hermite
Half Sinc Half Sinc
IOTA IOTA
Half Cosine Half Cosine
−300 −300

0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18
f/F f/F

(a) 5 MHz bandwidth utilization. (b) 20 MHz bandwidth utilization.

Fig. 3: Comparison of the power spectral densities of OFDM and FBMC with different prototype functions.

PHYDYAS filter have remarkably good SIR performances in apparently increases the sidelobes of FBMC. In all plots in
Fig. 4 beside Hermite and IOTA filters. Fig. 4, half-cosine filter approaches to PHYDYAS and OFDP
3) RRC Family: As opposed to rectangular filter, RRC are in terms of SIR performances due to the lower sidelobes
theoretically band-limited in frequency domain. However, in in time domain. However, decreasing αr degrades the SIR
practical conditions, the power spectrum of FBMC with these performance of RRC, since the impact of the filter truncation
filters will deviate from ideal frequency localization as in on SIR becomes dominant due to the increasing sidelobes
Fig. 3(a) and Fig. 3(b) because of the filter truncation in in time domain. In Fig. 5, the impact of truncation on SIR
time domain. Additionally, using small αr (e.g., half sinc) is investigated for different prototype filters. Apparently, for
201
6

80 80
Gaussian, OQAM Rect., QAM Gaussian, OQAM Rect., QAM
Hermite, OQAM Rect., OQAM Hermite, OQAM
70 Ext. Rect. (TCP = 0.14T, TW = 0), QAM 70 IOTA, OQAM Rect., OQAM
IOTA, OQAM

PHYDYAS (K = 4), OQAM PHYDYAS (K = 4), OQAM


60 OFDP (K = 4), OQAM 60 OFDP (K = 4), OQAM

Half Cosine, OQAM


50 50 RRC (αr = 0.22), OQAM
SIR (dB)

SIR (dB)
Half Sinc, OQAM

40 40

30 30

20 20
Impact of Half Cosine, OQAM Impact of
truncation RRC (αr = 0.22), OQAM
truncation
10 10
Half Sinc, OQAM

0 100 200 300 400 500 0 0.5 1 1.5 2


fd (Hz) Timing Offset (µs)

(a) Carrier Frequency Offset. (b) Timing offset. Extended rectangular is not included since it satisfies the
orthogonality in simulated timing offset range.

80
80
Gaussian, OQAM Rect., QAM
Gaussian, OQAM Rect., QAM
Hermite, OQAM Rect., OQAM
Hermite, OQAM Rect., OQAM
70 70 IOTA, OQAM Ext. Rect. (TCP = 0.14T, TW = 0), QAM
IOTA, OQAM Ext. Rect. (TCP = 0.14T, TW = 0), QAM

PHYDYAS (K = 4), OQAM


PHYDYAS (K = 4), OQAM
60 60 OFDP (K = 4), OQAM
OFDP (K = 4), OQAM

50 50
SIR (dB)
SIR (dB)

40 40

30 30

20 20
Impact of Half Cosine, OQAM
Impact of Half Cosine, OQAM
truncation RRC (αr = 0.22), OQAM
truncation RRC (α = 0.22), OQAM
10 10 r
Half Sinc, OQAM Half Sinc, OQAM

0 100 200 300 400 500 0 100 200 300 400 500
fd (Hz) fd (Hz)

(c) ITU Vehicular A. (d) ITU Pedesterian A.

Fig. 4: SIR performance of the prototype filters for different channel scattering functions

lower αr , the impact of truncation on SIR performance is 5) Hermite and IOTA Filters: When SIR, throughput and
highly dominant. For example, SIR is limited to 20.5 dB for sidelobes are considered, Hermite and IOTA filters has the
K = 16 for half-sinc filter even if the channel dispersion best characteristics because of two fundamental reasons: 1)
is ignored. Considering the practical channels, less time- very good joint time-frequency localization and 2) being
frequency dispersive channel, i.e. ITU Pedestrian A, results an orthogonal filter. Both of them provide very high time-
in better SIR performance for higher αr . Additionally, RRC frequency dispersion immunity as well as CFO and timing
and half-cosine filters approximately have the same immunity misalignment as given in Fig. 4(a) and Fig. 4(b). Additionally,
to timing misalignment given in Fig. 4(b) especially for higher the impact filter truncations on SIR is not as dominant as other
fd when K = 16. filters. For example, Hermite and IOTA filters with K = 8
is sufficient to yield more than 80 dB SIR as in Fig. 5. On
4) Gaussian Filter: Since Gaussian filter does not satisfy
the other hand, these filters lose their advantages for low
the orthogonality, the channel dispersion does not directly
Doppler scenarios compared with the extended rectangular, i.e.
affect the performance of the Gaussian filter as much as other
conventional OFDM. According to Fig. 4(c), and Fig. 4(d),
prototype filters. Even if Gaussian filter has the worst SIR
while FBMC with Hermite and IOTA filters are better than
results, the required equalization effort for OFDM/OQAM
conventional OFDM for fd > 135 Hz for ITU Vehicular A, it
with Gaussian filter might be less than other prototype filters
has better characteristics for frequencies as low as fd = 15 Hz
since Gaussian prototype filter has the best time-frequency
for ITU Pedestrian A.
localization. Additionally, Gaussian filter yields better sidelobe
performance compared to OFDM as in Fig. 3(a) and Fig. 3(b).
202
7

[6] P. Siohan, C. Siclet, and N. Lacaille, “Analysis and design of


120 OFDM/OQAM systems based on filterbank theory,” IEEE Trans. Sig.
Proc., vol. 50, no. 5, pp. 1170–1183, May 2002.
[7] M. Bellanger, “Specification and design of a prototype filter for filter
100 bank based multicarrier transmission,” in Proc. IEEE Int. Conf. Acous-
tics, Speech, Sig. Proc. (ICASSP), vol. 4, Salt Lake City, UT, May 2001,
pp. 2417–2420.
80 [8] B. Farhang-Boroujeny, “OFDM versus filter bank multicarrier,” IEEE
Gaussian, OQAM
Sig. Proc. Mag., vol. 28, no. 3, pp. 92–112, 2011.
SIR (dB)

Hermite, OQAM
IOTA, OQAM [9] T. Weiss, J. Hillenbrand, A. Krohn, and F. Jondral, “Mutual interference
60
in ofdm-based spectrum pooling systems,” in Proc. IEEE Vehic. Technol.
Half Cosine, OQAM
RRC (αr = 0.22), OQAM
Conf. (VTC), vol. 4, May. 2004, pp. 1873–1877.
K = 16 × T
[10] W. Kozek and A. Molisch, “Nonorthogonal pulseshapes for multicarrier
40 Half Sinc, OQAM
communications in doubly dispersive channels,” IEEE J. Select. Areas
Commun. (JSAC), vol. 16, no. 8, pp. 1579–1589, Oct. 1998.
20 [11] J. Du and S. Signell, “Classic OFDM systems and pulse-shaping
OFDM/OQAM systems,” Technical Report (KTH - Royal Institute of
Technology), pp. 1–32, Feb. 2007.
5 10 15 20 25 30 35 40 45 50 [12] T. Strohmer and S. Beaver, “Optimal OFDM design for time-frequency
Filter length (×T) dispersive channels,” IEEE Trans. Commun., vol. 51, no. 7, pp. 1111–
1122, 2003.
[13] A. Vahlin and N. Holte, “Optimal finite duration pulses for OFDM,”
IEEE Trans. Commun., vol. 44, no. 1, pp. 10–14, Jan. 1996.
Fig. 5: Impact of filter truncation. [14] B. Farhang-Boroujeny, Signal processing techniques for software radios,
2nd ed. Lulu Press Inc., 2010.
[15] K. W. Martin, “Small side-lobe filter design for multitone data-
communication applications,” IEEE Trans. Circuits and Systems–II:
VII. C ONCLUDING R EMARKS Analog and Digital Signal Processing, vol. 45, no. 8, pp. 1155–1161,
Aug. 1998.
In this paper, we investigate the performance of different [16] J. Du and S. Signell, “Pulse shape adaptivity in OFDM/OQAM systems,”
prototype filters in literature. According to the our findings, in Proc. IEEE Int. Conf. Advanced Infocomm Technology (ICAIT), Shen
Zhen, China, July 2008.
even if OFDM benefits from CP utilization yielding simple [17] R. Haas and J. Belfiore, “A time-frequency well-localized pulse for
frequency domain equalization for low Doppler spreads, it multiple carrier transmission,” Wireless Personal Commun., vol. 5, no. 1,
loses its advantage for higher frequency dispersion. On the pp. 1–18, July 1997.
[18] ITU Recommendation, “ITU-R M.1225 Guidelines for Evaluation of
other hand, Hermite and IOTA filter eliminates CP usage, Radio Transmission Technologies for IMT-2000,” 1997.
achieves higher SIR for the environments with higher fre- [19] P. Bello, “Characterization of randomly time-variant linear channels,”
quency dispersion, and inherently provides better sidelobe IEEE Trans. Commun. Syst., vol. 11, no. 4, pp. 360–393, Dec. 1963.
[20] I. Trigui, M. Siala, S. Affes, A. Stéphenne, and H. Boujemâa, “Op-
characteristics. On the other hand, PHYDYAS filter and OFDP timum pulse shaping for OFDM/BFDM systems operating in time
achieve superior SIR performances with finite durations which varying multi-path channels,” in Proc. IEEE Global Telecommun. Conf.
allow less complex structures at the receiver and transmitters. (GLOBECOM), Washington, D.C., Nov. 2007, pp. 3817–3821.
[21] T. Feng and T. R. Field, “Statistical analysis of mobile radio reception:
Also, compared to the other prototype filters, both RRC with An extension of Clarke’s model,” IEEE Trans. Commun., vol. 56, no. 12,
higher αr and Hermite filters are robust to the filter truncation. pp. 2007–2012, Dec. 2008.
As a follow-up of the present work, our future work includes
taking channel statistics and equalization complexity at the
receiver into account in order to obtain more accurate perfor-
mance comparison among different prototype filter designs.

ACKNOWLEDGMENT
This study has been supported by DOCOMO Innovations,
Inc.

R EFERENCES
[1] R. W. Chang, “Synthesis of band-limited orthogonal signals for multi-
channel data transmission,” The Bell System Technical J., pp. 1775–1796,
Dec. 1966.
[2] B. Saltzberg, “Performance of an efficient parallel data transmission
system,” IEEE Trans. Commun. Technol., vol. 15, no. 6, pp. 805–811,
Dec. 1967.
[3] S. Weinstein and P. Ebert, “Data transmission by frequency-division mul-
tiplexing using the discrete fourier transform,” IEEE Trans. Commun.
Technol., vol. 19, no. 5, pp. 628–634, Oct. 1971.
[4] B. Hirosaki, “An orthogonally multiplexed QAM system using the
discrete Fourier transform,” IEEE Trans. Commun., vol. 29, no. 7, pp.
982–989, 1981.
[5] B. Le Floch, M. Alard, and C. Berrou, “Coded orthogonal frequency
division multiplex,” Proceedings of the IEEE, vol. 83, no. 6, pp. 982–
996, 1995.
203

S-ar putea să vă placă și