The 2016 IEEE Asia Pacific Conference on Wireless and Mobile (APWiMob)

Performance Analysis of Generalized

Frequency Division Multiplexing in Various
Pulse-shaping Filter for Next Generation
Communication Systems
Sumarsana, Ali Muayyadi, Dharu Arseno

subcarriers. Ideally, this characteristic would give

Abstract—Currently, the technology of multicarrier system advantage in term of the subcarriers will not disturb each
used by the 4th generation of digital communications system other at the point of sampling. Thus, the OFDM signals
is based on Orthogonal Frequency Division Multiplexing will survive from the Inter Symbol Interference (ISI).
(OFDM). On one hand, this system produces multicarrier
signal that is robust in mitigating the destructive effect of
frequency – selective fading by dividing a wideband into
many small/tiny frequency bands to get a flat fading
response. In addition, they are also inherited the
orthogonality characteristic that not only protects the system
from Inter Symbol Interference (ISI) but by utilizing certain
amount of overhead called cyclic prefix also can reduce the
impact of Inter Carrier Interference (ICI).
On the other hand, OFDM systems have downsides in
emitting a high out of band (OOB) signal and high peak to
average power ratio (PAPR). These characteristics are very
harmful and so do not suitable in low power and higher data Fig. 1 Orthogonal subcarriers in OFDM
rate environment in the next generation communication
systems.
In the real world, the implementation of a perfect
This paper exploring the impact of filters in reducing the
rectangular pulse will cost a complicated system. In the
OOB, PAPR and signal received performance by utilizing
many types of pulse (which is rectangular pulse in OFDM frequency domain, an unlimited bandwidth as depicted in
systems). This is the more general model of frequency the figure of orthogonal OFDM subcarrier (the 𝑠𝑖𝑛𝑐
division multiplexing, the GFDM (generalized frequency function) seems to appear forever in the f-axis. And so, the
division multiplexing), will not take the advantage of signal’s rectangular pulse will require a high bandwidth to
orthogonality. Otherwise, it will use a type of pulse that has implement the ideal OFDM system. This also means the
more effectives characteristic in both time and frequency
domain (time-frequency localization – TFL) to improve the
overall power transmitted will be very high and more
system performance. Three types of filter will be exercised in specific in the out of band emission (OOB).
this simulation i.e. raised cosine, root raised cosine, and Therefore, in the implementation of this system, this
Gaussian filter. pulse has to be truncated and so will not give us a perfect
rectangular pulse anymore (in the time domain). Pulse
Index Terms—pulse – shaping, generalized frequency truncation will not only degrade the signal performance
division multiplexing, Inter Symbol Interference, Inter but also give more complexity in the system formulation.
Carrier Interference, next generation communication system
Instead, we can design a type of pulse to improve
bandwidth effectiveness and lower transmitted power
I. INTRODUCTION (OOB).
This is one of the objectives of recent research to find
N EXT generation communication systems demand
high quality and robust methods in multiplexing
signal to create a reliable high data rate systems. One
out a kind of pulse that gives us a zero-crossing at the
point of period and faster-degraded ripples. This is the
main feature of pulse that characterized by time-frequency
method used as standard today is orthogonal frequency localization. This characteristic is very important not only
division multiplexing or OFDM. It utilizes rectangular in contribution of outside band emission but also can
pulse to shape a multicarrier signal that grant the benefit of reduce overall transmitted power.
orthogonality of 𝑠𝑖𝑛𝑐 function in the frequency domain as There are numbers of pulse type to be the candidates of
indicating in Fig. 1 that is a series of orthogonal the filter to perform such qualities or specifications.
Jinfeng Du [5] suggests a method in calculating the
Sumarsana (sumarsono1@gmail.com), Ali Muayyadi
(alimuayyadi@telkomuniversity.ac.id), Dharu Arseno
parameter Heisenberg 𝜉 dan direction 𝜅 to represent the
(darseno@yahoo.com) are in Faculty of Engineering, Telkom University, time-frequency localization (TFL) of a pulse:
Jl. Telekomunikasi Terusan Buah Batu Bandung 40257 Indonesia
(www.telkomuniversity.ac.id).

978-1-5090-2767-5/16/$31.00 ©2016 IEEE 41

 The 2016 IEEE Asia Pacific Conference on Wireless and Mobile (APWiMob)

2
TABLE I can be described in a matrix as follows:
TFL GRADE OF VARIOUS TYPE OF PULSE
s+,+ s+,- s+,. ⋯ s+,01-
s-,+ s-,- s-,. ⋯ s-,01-
2 ALVES et al: PERFORMANCE
S = .,+ .,- s OF s
GFDM OVER s ⋯ s.,01- , CHANNELS
FREQUENCY-SELECTIVE
.,. (1)
⋮ ⋮ ⋮ ⋱ ⋮
s41-,+ s41-,- s41-,. ⋯ s41-,01-
levelhigher
The achieved
values by of OFDM.
parameters 𝜉 and 𝜅 mean better TFL K subcarriers transmits M data symbols per GFDM frame.
The aim of this paper is to present
characteristics of a pulse. This pulses should give better in the analysis of the Since the mappers are independent, different constellation
where rows of the matrix representing subcarriers and
performance of GFDM in frequency-selective
various measures and performance in the new GFDM channels con- orders can be used in each stream, allowing for dynamic bit
column for subsymbols. Therefore, each element of the
sidering
signal. different channel profiles. To the best of the authors’ loading mapping according to the channel conditions for each
matrix s8,9 , k = 0, 1, 2, … , K − 1, m = 0, 1, 2, … , M − 1
knowledge, this is a novel analysis
The new system called generalized frequency division and, thereby, it is the subcarrier [29]. Because GFDM transmits M data symbols in
main contribution of this paper. Three types of receivers will be fed into mapper for next processing steps.
are
multiplexing or GFDM in short, like implied by the term is each subcarrier using M time-slots, the data symbols can be
considered: ZFR (Zero Forcing Receiver), MFR (Matched The best part
organized in of
a the GFDM
frame structure systemgiven is by
the flexibility in
a generalized-form of frequency division multiplexing. By
Filter Receiver) and Matched choosing the
⎡ type of filter 𝑔(𝑛) to be used for pulse- ⎤
utilizing a better TFL’s pulse for Filter
shapingReceiver with DSIC
the subcarriers, sprevious sworked, sMichailow . . . [1] sutilized
(MFR-DSIC). shaping. In the 0,0 0,1 0,2 0,M −1
we can play theAll results areofcompared
multiplexing information withsignal
the performance
in both ⎢ filter
s s s . . . s ⎥
of anandOFDM system. All to simulation two different ⎢ 1,0 raised 1,1 cosine (RC)
1,2 and root-raised
1,M −1 ⎥
time frequency domain fulfill theresults
demandthat have been
of higher ⎢ sto s2,1 the effect s1,2 of s2,M −1 ⎥
. . . inter-symbol
obtained using Matlab are compared with theoretical cosineS (RRC)
curves, = ⎢ 2,0 compare ⎥,
data rate. ⎢ . .. interference. .. . He .. to a ⎥
which allow one to conclude that the symbol error probabilityinterference⎣ and ..inter-carrier . . . . came . ⎦
expression proposed for OFDM can be used to estimate conclusion that
the RC filter gave more ISI but less ICI than
II. GENERALIZED FREQUENCY DIVISION MULTIPLEXING s K−1,0 sK−1,1 sK−1,2 . . . sK−1,M −1
performance of GFDM. RRC. (1)
The remaining of this paper is organized as follows: Section For where addition
the k-th to therow flexibility
represents inthechoosing symbols better transmitted TFL in the
Generalized Frequency Division Multiplexing (GFDM)
II presents the generation of GFDM symbols, whereas Sectioncharacteristic k-th subcarrier for filter,
and the GFDM m-th system
column also represents utilizing the asymbols
is multicarrier system, introduced by Fettweis [3] as a robust method in mitigating the effect of time-spreading.
III presents three techniques used to recover the transmitted transmitted in the m-th time-slot.
system that combine M-number of subsymbols and K- This phenomenon usually occurs when the system is used
information. Section IV contains the performance analysis of Each data symbol sk,m is up-sampled by zero-padding
number of subcarriers to build a frame. This is the first by mobile user. The sender or receiver or both terminal
GFDM considering AWGN and Section V evaluates the per- M N − 1 zeroes, resulting in the sequence
thing that make it different to Orthogonal Frequency moves faster than the speed of data transmission will
formance over frequency-selective channels. Finally, Section
Division Multiplexing (OFDM) in the case that it just spread the symbol and sk,mthus
(n) can = sk,moverlapδ(n − itsmN ),
neighbor. The (2)
VI concludes the paper.
utilizes one subsymbol and K-number of subcarriers in a method called tail-biting will provide last chunk of frame
frame. where N is the number of samples used to represent a time-
II. G ENERATION OF THE GFDM S IGNAL to be place
slot. This in sequence
the next frame is applied as redundant
to a transmit data.filter It actually
with impulse
A. GFDM
GFDM is Transmitter implemented in the system by using a circular convolution
a flexible multi-carrier modulation scheme that response g(n) of length L = M N . If conventional linear
in each subsymbol.
convolution is used, likeVOL. in the Filter Bank Multi-carrier
hasThe construction
been introduced by of Fettweis
the transmitter
et al [22]a and little bit interesting
it has more REVISTA TELECOMUNICAÇÕES, 15, Nº02, OUTUBRO DE 2013
complex
features than
for in CROFDM as proposed
applications. Figure by Bruno M. Alves
1 depicts the etblock (FBMC) [30] schemes, the guard time interval between the
all in [2]: of the GFDM transmitter.
diagram GFDM frames should be larger than the channel delay spread
sk,0
plus the filter spreading in order g(n) to avoid IFI (Inter Framewher
Time Slot #1 Time Slot #2 Time Slot #3

Time Slot #1 Interference), as depicted g(n) in Figure 2 for N = 8, M = 3 and

0 QAM
Upsampler
MN
g (n) an arbitrary impulse response g(n). Such a large guard time
is the
− j0

1 QAM
Upsampler
g (n)
e
interval
5 10would 15 be 20 a considerable drawback, 5
causing
10 15
throughput
20

MN sk,1
reduction, leading to a poor spectrum efficiency. However, this
g(<n-N>NM-1)
n
− j 2π
. . . . e N
Sample NM-1
. . . .
. . . . problem can be easily avoided g(n) by using a technique called
K-1 QAM
Upsampler
g (n)
tail-biting [22]. In this technique, the mN last samples at the
MN
e
− j2π
(K−1)n
N s k,2
output
5
of
10
the15
filter
20
are shifted to the first 5
mN10
positions,
15 20
as
g(<n-2N>NM-1)
Time Slot #2 illustrated in Figure 3. This process can be made by circular
convolution [31].
Serial/Parallel Converter

K QAM
Upsampler
g (n-N) g(n)
MN
− j0

K+1 QAM
Upsampler
g (n-N)
e
5 10 15 20 5 10 15
Filter
20
Filter
is the
MN s
Tak
n
. . . . e
− j 2π
N
k,0
g(n) Time Slot #1 Time Slot #2 Time Slot #3 Spreading Spreading
Data
Bits
.
.
.
.
.
.
.
. CP
Fig. 3 Creating GFDM g(n)Symbol using Circular Sample # NM-1 repre
Upsampler
Σ Fig. 3. GFDM symbol obtained Convolution
by circular convolution.
2K-1 QAM MN
g (n-N) 5 10 15 20 5 10 15 20
( K−1)n
− j 2π
e N sk,1 g(n-N)
. Finally, each sub-streamg(n)of symbol will be fed into wher
.
.
symbols transmitted
subcarrier in a given subcarrier. In [26], the author
complex converter:
Time Slot #M presents
s
5a deep
10 analysis
15 20
k,2
about the H
influence of Raised Cosine 5 10 15 20 25 30

g(n-2N)
Upsampler
(RC) and Root Raised Cosine1F.G8 (RRC) filters
(M-1)K QAM MN
g [n-(M-1)N]
e − j0 p 8 n = e g(n)
I , (2) in the performance
(M-1)(K+1) QAM
Upsampler
g [n-(M-1)N] of GFDM systems. The impact of the roll-off factor is analyzed
MN where N5 is number of sampling in a time slot. The output
as well. The 10major conclusions of this analysis are: i) if RCs
n
. . . . − j 2π
15 20 5 10 15 20 25 30 35
e N
. . . . frame will be the summation of every sub-stream from
. . . . are used on the transmitter and receiver sides there will be
each
largersinglewhen
subsymbol and subcarrier.
Fig.ISI
2. GFDM compared withbythe useconvolution.
of RRC because the
Upsampler
MK-1 QAM MN
g [n-(M-1)N] symbol obtained linear
Nyquist criterion is not satisfied, however the ICI will be
(K−1)n

B. GFDM Receiver
− j2π
e N

Fig. 2 Transmitter of GFDM System smaller

The In than
order
receiver thetopart
oneuseobtained
theGFDM
of withsystem
tail-bitting RRC because
technique, of the
the
is composed sharper
filter
byimpulse
Fig. 1. Block diagram of the GFDM transmitter. frequency
response response
must
subsystems as follows: of
allow the
for RC
circularand; ii)
shifts the
of Nsmaller
samples,the asroll-
shown
To get the data structure ready for multiplexing, the first off in
factor
Figurethe3better
[22] [23].the system performance because of the is the
stepThe input bits are
is converting converted
serial into M
data stream into data streams
K parallel that feedreduction
form that Since of g(n)
the ICI.can Clearly there is a trade-off
have non-rectangular pulsebetween ISI
shape, GFDM
M K independent J-QAM mappers. Each mapper converts aand subcarriers ICI in the choice
can be ofnon theorthogonal
RC or RRC. to each other, which can
block of q bits into a data symbol sk,m , k = 0, 1, 2, . . . , Once lead the filterAdditionally,
to ICI. impulse response is chosen,
the transmit each
filter sub-stream
impulse response
K − 1, m = 0, 1, 2, . . . , M − 1. Therefore, each of theis up-converted can cause ISI by(Intersymbol
a complex subcarrier Interference) givenamong
by the M data
n
pk (n) = e−j2πk N . (3)
978-1-5090-2767-5/16/$31.00 ©2016 IEEE 42 At this point, it is important to notice that in GFDM the
 III. R ECEPTION OF THE GFDM S IGNAL approach, A is the inverse of matrix A and req is the
Figure
The 20165IEEE
shows the basic
Asia Pacific block ondiagram
Conference Wireless of
and aMobile
GFDM equalized signal vector.
(APWiMob)
receiver. Matrix A has order KM x N M , N ≥ K, which means 3 that
it is not necessarily
all the parameter calculations. square. Therefore, the inversion operation
may not be suitable for this matrix. In this case, it is possible
B. toA use
Shaping Pulse Construction
the pseudoinverse matrix of A, defined by
A shaping pulse required in this # simulation$−1 has to be
constructed in impulse A =Aresponse
+ H and to
AA H be provided in a
, (19)
matrix. Upsampling factor 𝐿 = 𝑀𝑁 have to be considered
Fig. 2 Receiver of GFDM System while constructing the filter matrix with the dimension of
Fig. 5. Basic block diagram of a GFDM receiver. where AH is the Hermitian (conjugate and transpose) matrix
1𝑥𝐿.
Detector match filter and slicer obviously is a complex of A. Notice that A+ A = IN M where IN M is the identity
Belows are the discreet formulation of the shaping pulse
converter to from
get back the sub-stream of GFDMtosymbol matrix of order N M .
The signal the antenna is down-converted base-bandused in the simulation:
using multiplication
and sampled, resulting n discrete
of inp8the ∗
. Finally, it follows
received signalbyrCP
the(n). The ZFR is capable of completely removing the ICI resulted
circular convolution and demapper to channel
get the with
estimated from the n 1,
non-orthogonality n = 0, 1, 2, … ,the
between N −subcarriers.
1 However,
In this paper a time-invariant multipath impulse 𝑔QRS = (4)
data bits. h(n) has been considered, leading to since A +
has 0, n
high values, this procedure ≥ N enhances the influ-
response
H
ence of the noise in the detected symbols, which increases the
rCP (n) ∗= xCP
p8 n = e KF.G8
(n) I∗, h(n) +
(3) w(n) (16) BER. πn
sin ( ) cos(πnβ/N)
𝑔 n = N (5)
where xCP (n) is the transmitted signal with the cyclic prefix QS
πn 1 − 4β. n. /N .
and w(n) is a vectorIII. of gaussian
SYSTEM noise samples with zero mean
SIMULATION
andThevariance σ 2
. B. Matched Filter Receiver
simulation is composed into two separated portion,
n
1 − β πn
The toreceived
firstly build thesignal
systemisconsists
used for of asynchronization and to
transmitter, channel sinparallel single frequency
The MFR can be 1 + seen
β πn as K N
estimate the channel impulse response. Subsequently, the CP cos +
and a receiver using MatLab code. Link performance will receivers processing theN equalized signal 4βnreq (n). Since only
is removed. It is assumed that the CP length is larger than the
be exercise with the present of AWGN channel (frequency the time 2β samples n = mN are of interest, N the MFR (6)can be
channel delay spread, which means that there is no interference 𝑔QQS (n) =
selective and time variant channel will be exercise in an implementedπ N as a correlator receiver 4βn .[33], as shown in Figure
among GFDM frames. 1−
advanced system), while out of band emission and peak 6. N
Afterwards, the signal must be equalized to compensate for
power ratio don’t need to be exercise in a simulation The symbol received at a given subcarrier and at a given
the influence of the channel frequency response in the received
channel. time-slot is
signal. The channel frequency response can be considered flat 1 hi
Second part of the simulation is the construction of the 𝑔`abcc n = 𝑒
1 i
.j (7)
for each subcarrier if K is large enough to make the subcarriers N% M −1
shaping pulse using rectangular, raised cosine, root raised 2𝜋. 𝜎
bandwidth smaller than the channel coherence bandwidth. In ŝ ′ ,m′ = req (n) [gm′ (n)pk′ (n)] .
∗
(20)
cosine, and Gaussian. A couple of factors to considered k
this case, the received signal can be equalized in the frequency
while constructing the filter are roll-off factor for raised where β is roll-off factor, and 𝜎 is standard deviation.
n=0
domain using an Zero-forcing equalizer. Assuming the receiver
cosine and root raised cosine, while Gaussian pulse have In this simulation, rectangular pulse will be used as base
is able to estimate the channel impulse response, the equalized If the influence of the noise and multipath channel is
to consider standard deviation factor. for comparing the performance of OFDM system to
sequence can be obtained from disregarded, then req (n) = x(n), which leads to
GFDM system using the various shaping pulses.
A. A GFDM System Model
A multicarrier modulation system in GFDM is IV. RESULTS
comprised by two signal component, one in time domain
In OFDM based system, where each subcarrier is
and another in frequency domain. The time component is a
shaped with a rectangular pulse in time domain, the first
sequence of timeslots that represent subsymbols and the lmH n
frequency component is a sequence of subcarriers of side lob of the corresponding frequency domain pulse
n
GFDM. The construction of GFDM frame can be decay fairly slow. This will introduce a higher power
illustrated as follows: spectrum outside the signal bandwidth compared to the
one of GFDM system about 10 – 15 dB [1]. In this
Time (m) simulation, the GFDM system will produce PSD side lobe
lower than in OFDM system by 20 dB (RRC pulse-
s11 s12 S13 … s1M shaping), 30 dB (RC pulse-shaping), and 5 dB (Gaussian
Frequency (k)

pulse-shaping) in average.
s21 s22 s23 … s2M

s31 s32 s33 … s3M

… … … … …

sK1 sK2 sK3 … sKM

Fig. 4 Construction of GFDM Frame

Data stream have to be constructed in two dimension

(time and frequency) structure as illustrated in the above
figure and it’s ready to be fed into a GFDM transmitter. Fig. 5 PSD of OFDM & GFDM with RRC (b=0.1)
The next process will be very straight forward by
providing mapper, upsampling, circular convolution with
the shaping pulse, and applying multiplication by complex
converter. MatLab using matrix operations for conducting

978-1-5090-2767-5/16/$31.00 ©2016 IEEE 43

 The 2016 IEEE Asia Pacific Conference on Wireless and Mobile (APWiMob)

Fig. 6 PSD of OFDM & GFDM with RC (b=0.1) Fig. 9 PAPR of OFDM & GFDM with RC pulse-
shaping

Fig. 7 PSD of OFDM & GFDM with Gaussian pulse-

shaping Fig. 10 PAPR of OFDM & GFDM with Gaussian pulse-
shaping
The strong out of band radiation in OFDM will produce All in all, the better performance in out of band
an interference from a subcarrier to its neighbor subcarrier. emission and peak to average power ratio have to be refer
To mitigate the effect, it has to provide wider cyclic prefix to the link performance of the system. Below is the link
but will reduce the effectiveness of the bandwidth usage. performance of GFDM system in AWGN channel that
This system also very sensitive in term of carrier represent the quality can meet of OFDM system using MF
frequency offset, which requires a very complex or ZR receiver.
synchronization method to guarantee that the orthogonality
is not affected.
The GFDM system will offer lower out of band
radiation because each subcarrier is shaped with a filter of
choice. Raised cosine, root raised cosine, and Gaussian
filter produce faster decay in the first side lob than
rectangular pulse. Furthermore, the system can improve
OOB radiation performance and the effectiveness of
bandwidth usage.
In term of power emission, the superposition of many
subcarriers can create a bulk of power radiation. The
Fig. 11 Link Performance (SER) of OFDM & GFDM with
GFDM system will reduce a significant power by radiating
MF Receiver
a lower side lobe than the one radiated by OFDM system
as indicating by PAPR graphs below:

Fig. 12 Link Performance (SER) of OFDM & GFDM with

ZF Receiver
Fig. 8 PAPR of OFDM & GFDM with RRC pulse-
shaping

978-1-5090-2767-5/16/$31.00 ©2016 IEEE 44

 The 2016 IEEE Asia Pacific Conference on Wireless and Mobile (APWiMob)

5
V. CONCLUSION experience as senior engineer at PT PBI/Bechtel (1998)
and PT Austrian Energy Indonesia (2000).
The GFDM system can prove as an effective system to
support the introduction of next generation communication
system. A much faster data rate and more reliable system
along with low power consumption is mandatory in the
upcoming communication services.
The exercising of variety of pulse to be used in shaping
the subcarriers in GFDM system is intended to get the
more effective way to accomplish the job. Hopefully, we
can find the better pulse property such as roll-off factor
and standard deviation, or more advanced pulse type to be
used in GFDM system. The preliminary results can be
taken that we have to choose a type of another pulse with a
better time-frequency localization than rectangular.

REFERENCES
[1] Nicola Michailow et all, Generalized Frequency
Division Multiplexing for 5th Generation Cellular
Networks, IEEE Transactions on Communications,
Vol. 62, No. 9, September 2014.
[2] Bruno M. Alves et all, Performance of GFDM over
Frequency – Selective Channels, Revista Telecom.
Vol. 15, October 2013.
[3] Gerhard Fettweis et all, GFDM – Generalized
Frequency Division Multiplexing, IEEE, 2009.
[4] Rohit Datta et all, GENERALIZED FREQUENCY
DIVISION M ULTIPLEXING IN COGNITIVE
RADIO, 20th European Signal Processing Conference
(EUSIPCO 2012).
[5] Jinfeng Du, Pulse Shape Adaptation and Channel
Estimation in Generalized Frequency Division
Multiplexing Systems, Licentiate Thesis in
Electronics and Computer Systems, Stockholm,
Sweden 2008.
[6] Nicola Michailow et all, Generalized Frequency
Division Multiplexing: Analysis of an Alternative
Multi-Carrier Technique for Next Generation Cellular
Systems, CONFERENCE PAPER
(www.researchgate.net /publication/261231239),
August 2012.

Sumarsana is undergraduate in telecommunication

engineering in Telkom University (former STT Telkom).
Recently taking graduate in the same field in Telkom
University since 2014. The most professional carrier as
engineer in PT. Telekomunikasi Indonesia since 1997.
Now, has role as a Capex/investment planner in the state
owned company and the largest telco in Indonesia that
operating broadband infrastructures to serve broadband
connectivity (both fixed and mobile).

Ali Muayyadi is currently a lecturer in Faculty of

Electrical & Telecom Engineering, Telkom University. He
holds the titles: BEng in Electrical Engineering, Institut
Teknologi Bandung (1990), MSc in Mobile
Communications, ENST, Telecom Paris (1997), PhD in
area of Digital Communications, University of Plymouth,
UK (2004).

Dharu Arseno is a lecturer in Faculty of Electrical &

Telecom Engineering, Telkom University. He has working

978-1-5090-2767-5/16/$31.00 ©2016 IEEE 45