Sie sind auf Seite 1von 5

GFDM with Different Subcarrier Bandwidths

Yuta Akai†, Yuka Enjoji† Yukitoshi Sanada†, Ryota Kimura††, Ryo Sawai††
†Dept. of Electronics and Electrical Engineering,
Keio University, Yokohama, Japan
Email: yakai@snd.elec.keio.ac.jp, yenjoji@snd.elec.keio.ac.jp, sanada@elec.keio.ac.jp
††Innovative Technology Development Division, System R&D Group,
RDS Platform, Sony Corporation,
Email: Ryota.Kimura@jp.sony.com, Ryo.Sawai@jp.sony.com

Abstract—This paper proposes a generalized frequency divi- cialy on a downlink, mobile terminals with various conditions
sion multiplexing (GFDM) modulation scheme that transmits are connected to a base station. Thus, the subcarriers with
a signal with different subcarrier bandwidths. In a receiver, different bandwidths should be transmitted in the same GFDM
the GFDM signal is demodulated by using a zero forcing (ZF)
algorithm or a minimum mean square error (MMSE) algorithm block.
and the BER performance of these algorithms is related to the Thus, this paper proposes a GFDM modulation scheme
condition number of a modulation matrix. This matrix can be that uses subcarrier filters with different bandwidths. A single
optimized by adjusting the roll-off factor of subcarrier filters. GFDM block then consists of subcarriers with different band-
It is shown that the performance of the proposed GFDM is widths. In the GFDM receiver, a zero forcing (ZF) algorithm
about 0.02 dB better than that with a roll-off factor of 0 at
a BER of 10−3 on an AWGN channel. On the other hand, on the or a minimum mean square error (MMSE) algorithm is applied
multipath fading channels, the BER performance improves as the and both of the receiver use inverse matrices that are calculated
subcarrier bandwidth increases because of frequency diversity. from subcarrier filter coefficients. This paper examines the
relation of the bit error rate (BER) performance and the
I. I NTRODUCTION condition number of the inverse matrix on different channel
Recently, the directions of the evolution of wireless com- models.
munication systems are not only to increase data rates, but This paper is organized as follows. Section 2 describes the
also to realize lower latency, lower energy consumption, and system model. Section 3 presents the simulation conditions
higher spectrum efficiency. The fifth generation (5G) mobile and the numerical results obtained though computer simula-
communication system is expected to provide the various kinds tion. Section 4 gives our conclusions.
of wireless networks such as machine-to-machine (M2M)
II. S YSTEM D ESCRIPTION
communications [1], the Tactile Internet [2], or wireless re-
gional area networks (WRANs) [3]. To realize these services, The block diagram of a transmitter is shown in Fig. 1.
new radio wave techniques have been investigated to substi- Information bits are encoded with a turbo encoder, rearranged
tute orthogonal frequency division multiplexing (OFDM) [4], by an interleaver, and punctured to realize a specific coding
which has been standardized in the fourth generation (4G) rate. A QAM mapper maps P encoded bits to a 2P QAM
mobile communication. One of these techniques is general- symbol. It is assumed that the narrow and wide bandwidth
ized frequency division multiplexing (GFDM) [5] [6]. In the subcarriers are modulated and co-existed in one GFDM block.
GFDM, the spectrum of the signal on each subcarrier is shaped The narrow band subcarriers are replaced by a subcarrier with
by a filter in order to suppress the out-of-band (OOB) radiation a C times wider bandwidth. The number of subsymbols on
for better frequency efficiency. Moreover, the peak-to-average a narrow band subcarrier is Mn = M while the number of
power ratio (PAPR) of the GFDM is smaller than that of the subsymbols on a wide band subcarrier is Mw = CMn in order
OFDM and the power consumption is expected to be lower [7]. not to decrease the transmission throughput. Furthermore, the
In addition, in the GFDM, the parameters such as the number number of the narrow band subcarriers is Kn , the number
of subcarriers or the number of subsymbols are easily modified of the wide band subcarriers is Kw , and the total number of
and they can be adapted to various applications. In [8], the subcarriers is K = Kn + CKw . For example, the subcarrier
difference in the OOB radiation has been compared using assignment in the case of C = 3, Kw = 1, and K = 16 is
different filters. [9] has shown the computational complexity shown in Fig. 2.
of the GFDM in terms of the number of subsymbols. In [10], A. Conventional Transmission Signal Model
the difference of demodulation error rates characterized by the
A conventional GFDM signal consists of subcarriers with
conditions of the numbers of subcarriers and subsymbols as
a unified bandwidth. The transmit symbol vector on the k-th
well as the roll-off factor of the subcarrier filters is mathemat-
subcarrier, dkM = [dk,0 · · · dk,m · · · dk,M −1 ]T , is converted
ically analyzed.
into a frequency domain by a Fourier transform matrix as
However, the previous researches have focused on the cases
with unified bandwidth subcarriers in a GFDM block. Espe- DkM = WM dkM (1)

978-1-5090-1701-0/16/$31.00 ©2016 IEEE


The baseband signal is then written as
N
 −1
vp (t) = un pt (t − nTs ) (8)
n=−NGI

where pt (t) is the impulse response of the transmit filter, Ts


is the sampling interval of the GFDM symbol, and NGI is the
GI length. In the receiver side, the received signal is given by
N
 −1
y(t) = un pc t − nTs ) + v(t) (9)
n=−NGI

Fig. 1. Block diagram of transmitter. where n(t) is the noise and pc (t) is the impulse response that
is calculated by
pc (t) = pt (t) ⊗ h(t) ⊗ pr (t) (10)
where h(t) is the impulse response of the channel, pr (t) is
the impulse response of the receive filter, and ⊗ denotes the
convolution operation.
Fig. 2. Example of frequency assignment of GFDM block (C=3).
B. Proposed GFDM Signal Model
In order to generate the subcarrier signals with different
where dk,m denotes the transmit symbol on the k-th subcarrier bandwidths, M is substituted by Mn or Mw for a narrow
of the m-th subsymbol. It is upsampled L times and multiplied band subcarrier or a wide band subcarrier. Suppose that C
with a filter matrix, ΓM , that has diagonal elements as ΓM = is an odd integer, from the (k  − C−1  C+1
2 )-th to (k − 2 )-th
diag[γ0 γ1 · · · γLM −1 ]. Generally L is set to 2 [10]. After narrow band subcarriers are replaced by a subcarrier with C
the filtering, the signal is assigned to the k-th subcarrier by times larger bandwidth, Eq. (4) is redefined as fllows:
multiplying the following assignment matrix. (I) For the narrow band subcarrier:
(I) if k = 0,
 T Sk = Pk ΓMn RMn WMn dkMn , (11)
IU 0U,U (K−1)
P0 = . (2) (II) For the the wide band subcarrier:
0U,U (K−1) IU
(II) if 1 ≤ k ≤ K − 1, Sk  = P k ΓMw RMw WMw dk Mw , (12)
 T
0U,U k IU 0U,U (K−k−1) where
Pk = . (3)  T
0U,U (k−1) IU 0U,U (K−k)
P k = 0CU,U (k −1) ICU 0CU,U (K−k −C+1)
, (13)
Note U = LM/2, 0a,b denotes the (a × b) zero matrix, and 0CU,U (k −C−1) ICU 0CU,U (K−k +1)
Ia denotes the (a × a) unit matrix. The transmit signal on the and Eq. 6 is rewritten as
k-th subcarrier is ⎧⎛  C+1 ⎞ ⎫
⎨ k−2 K−1
 ⎬
Sk = Pk ΓM RM WM dkM (4) x = WN ⎝
H
Sk + Sk ⎠ + Sk . (14)
⎩ ⎭
k=0 k=k + C+1
2
where
T C. Demodulation of Received Signal
RM = [IM · · · IM ] . (5)
   In a receiver, the received signal is converted to digital
L
samples by an A/D converter at the rate of Ts . Therefore,
After summing all K subcarriers, the transmit signal vector in the received digital sample is represented as
the GFDM transmitter is [7]
K−1 yn = y(nTs ). (15)

H
x = WN Pk ΓM RM WM dkM (6) After removing the GI and taking the DFT of N samples, the
k=0 signal on the l-th symbol is expressed as
H
where WN is the (N × N ) inverse Fourier transform matrix. N −1  
 −2πl
A cyclic prefix is then added as a guard interval (GI) by Yl = yn exp j
replicating the last part of the GFDM symbol. The transmit n=0
N
signal with the GI is given as = Hl Xl + Vl (16)

xn+N (−NGI ≤ n < 0) where Hl is the frequency response, Xl is the signal com-
un = . (7)
xn (0 ≤ n ≤ N − 1) ponent, and Vl is the noise on the l-th symbol, respectively.
Then, in order to apply frequency domain equalization (FDE) Furthermore, vn and wn are the correction coefficients to
as well as the removal of inter-carrier interference (ICI), a ZF the symbol candidate and the noise variance owing to the FDE
algorithm or a MMSE algorithm is applied. The demodulation [11]. In the case of ZF, they are given as
matrix of ZF or MMSE is given as N

−1 H
BZF = (HA) (17) wn = (WN BZF )n,n , (26)
n =1
or vn = 1, (27)
2 H H −1 H H
BM M SE = (σ + A H HA) A H (18) while they are also given for the case of MMSE as
where N

⎛ ⎞ H
k − C+1
 2 K−1

wn = (WN BM M SE )n,n , (28)
A=⎝ Pk ΓMn RMn + Pk ΓMn RMn ⎠ n =1

k=0 k=k + C+1
2
(19) vn = wn , (29)
+P k ΓMw RMw
where (·)n,n is the (n, n )-th element of the matrix. The LLR
is the modulation matrix and H = [H0 · · · HN −1 ]. Therefore, is calculated as
the demodulated signal in the frequency domain with ZF or Q1p
MMSE is L(bp |dˆk,m ) = log (30)
Q0p
D̂ = BZF Y (20)
where L(bp |dˆk,m ) is the LLR of the pth bit of the m-th
or subsymbol on the k-th subcarrier.
D̂ = BM M SE Y (21) D. Condition Number and BER performance of GFDM
where It is known that the BER performance of the ZF and
D̂ = [D̂T0Mn , · · · D̂T(k −(C+1)/2)Mn , D̂Tk Mw , MMSE algorithms are related to the condition number of the
(22) modulation matrix [12]. The BER performance is better if
D̂T(k +(C+1)/2)Mn , · · · D̂T(K−1)Mn ]T the condition number is smaller. In the conventional GFDM,
is the demodulated signal vector in a frequency domain, the condition number can be determined mathematically be-
D̂kM = [D̂k,0 · · · D̂k,M −1 ]T is the demodulated signal vector cause modulation matrix can be regarded as a cyclic matrix.
in a frequency domain, and Y = [Y0 · · · YN −1 ]T is the However, in the proposed GFDM with different subcarrier
received signal vector. Thus, the demodulated vector on the bandwidths, the matrix A is not a cyclic matrix. Therefore,
k-th subcarrier in a time domain is the condition number has to be numerically calculated and
H
the BER characteristics of the proposed GFDM scheme has
d̂kM = D̂kM WM (23) to be evaluated. It is shown that the optimum roll-off factor
where d̂kM = [dˆk,0 · · · dˆk,M −1 ]T is the demodulated signal of the subcairrer filter is different depending on the GFDM
vector, and dˆk,m denotes the demodulated signal of the m-th modulation parameters as shown in Fig. 3. While the condition
subsymbol on the k-th subcarrier, respectively. For the case number in the conventional GFDM becomes smaller when the
of the proposed modulation scheme, M is set to M = Mn roll-off factor approaches to 0, the proposed GFDM has the
or M = Mw so as to extract the corresponding subsymbols optimum value.
on the narrow band subcarrier or the wide band subcarrier. A III. N UMERICAL R ESULTS
log-likelihood ratio (LLR) is calculated for each coded bit in
the systematic part of the codeword. The likelihood values for A. Simulation Conditions
the pth bit of the symbol is given as follows: Simulation conditions are shown in Table I. A turbo code
 with 8 state memory is applied as the error correction code.
Q1p = exp(− wn1σ2 dˆk,m −vn Ŝp1 2 ), (24) The code rate is 1/3 through puncturing and the interleaver
Ŝp1 ∈{S}bp =1 size is fixed to 4800. The decoding algorithm is log-MAP
 and the number of decoding iterations is 8. Each signal
Q0p = exp(− wn1σ2 dˆk,m −vn Ŝp0 2 ), (25) stream is modulated with QPSK and multiplexed through
Ŝp0 ∈{S}bp =0
GFDM. The numbers of subcarriers and subsymbols are set as
[K, C, Mn , Mw ] = [16, 3, 7, 21] or [16, 7, 7, 49]. The sampling
where σ 2 is the noise variance, Ŝp1 and Ŝp0 are candidate interval is 1/20 MHz. The upsampling rate is two and the
symbols in which the pth coded bit of a 2P QAM symbol is length of the GI is 32 samples. Raised cosine (RC) filters are
”1” and ”0”, {S}bp =1 and {S}bp =0 is the set of the candidate used as subband filters. The number of filter taps are set to
symbols in which the pth coded bit of the 2P QAM symbol is 2Mn or 2Mw and the roll-off factor is selected from 0.9, 0.1, 0,
”1” and ”0”, and Q1p and Q0p are the sum of the likelihood or the optimum value. The optimum value is α = 0.04736 for
values for the pth coded bit of ”1” and ”0”. C = 3 or α = 0.02 for C = 7 from Fig. 3. As channel models,
0
3 10
M =7 (Conventional GFDM)
w
2.8 M =21 (3 times bandwidth subcarrier)
w
2.6 Mw=49 (7 times bandwidth subcarrier)
−1
2.4 10
Condition Number

2.2

BER
−2
10
1.8

1.6
RCn (α=0.9)
1.4 RCw (α=0.9)
−3
10
RCn (α=0.04736)
1.2 RCw (α=0.04736)
RCn (α=0)
1
RCw (α=0)
−2 −1 0
10 10 10 −4
10
Roll−off Factor α 0 0.1 0.2 0.3 0.4
Eb/N0 [dB]

Fig. 3. Condition number verses roll-off factor.


Fig. 4. BER vs. Eb /N0 (AWGN, ZF, [K, C, Mn , Mw ] = [16, 3, 7, 21]).
TABLE I
S IMULATION C ONDITIONS 0
10

Error Correction Coding Turbo Code ɹ


Code Rate 1/3
Decoding Algorithm Log-MAP −1
10
Number of
Decoding Iterations 8
interleaver size 4800
Modulation Scheme QPSK
BER

−2
10
GFDM parameters [K, C, Mn , Mw ] = [16, 3, 7, 21]
[K, C, Mn , Mw ] = [16, 7, 7, 49]
Sampling Interval 1/20 MHz RCn (α=0.9)
Over Sampling Rate 2 −3 RCw (α=0.9)
10
Guard Interval 32 Samples RCn (α=0.04736)
Subcarrier Filter RCw (α=0.04736)
RCn (α=0)
(narrow/wide) Raised Cosine(2Mn /2Mw Taps) RCw (α=0)
Roll-Off Rate (α) 0.9ɼ0.1ɼ0, and Optimum
−4
Channel Model AWGN 10
0 0.1 0.2 0.3 0.4
Raician(K-factor=10) Eb/N0 [dB]
Indoor-Residential A
Channel Estimation Ideal
Demodulation Scheme ZF, MMSE Fig. 5. BER vs. Eb /N0 (AWGN, MMSE, [K, C, Mn , Mw ] =
[16, 3, 7, 21]).

additive white Gaussian noise (AWGN), Rician (K-factor=10), performance by 0.06 dB in Fig. 4, by 0.02 dB in Fig. 5, by
and Indoor Residential A are assumed and channel estimation 0.01 dB in Fig. 6 at the BER of 10−3 as compared to those
in the receiver is assumed to be ideal. As mentioned, the with a roll-off factor of 0. The optimization of the condition
ZF algorithm or the MMSE algorithm is applied as the number is effective in both the ZF and MMSE algorithms. The
demodulation scheme. In the figures, ”RCn ” means the narrow optimization works even on the Rician channel as shown in
band subcarrier and ”RCw ” implies the wide band subcarrier. Fig. 7. It improves the performance of the wide band subcarrier
by 0.03 dB as compared to that with a roll-off factor of 0.
B. BER Performance on AWGN and Rician Channels
The BER performance of the ZF algorithm and the MMSE IV. C ONCLUSIONS
algorithm versus Eb /N0 is shown in Figs. 4-7. On the AWGN This paper has proposed the GFDM scheme with dif-
channel, the BER performance in the proposed GFDM dete- ferent subcarrier bandwidths. In the proposed scheme, the
riorates by increasing the subcarrier bandwidth owing to the GFDM block consists of different bandwidth subcarriers. The
interference from adjacent subcarriers as observed in Figs. 5 BER performance degrades when the subcarrier bandwidth
and 6. However, by using the optimum roll-off factors, the increases and it is possible to suppress the amount of the
BER improves and the wide band subcarrier achieves better performance deterioration by optimizing the roll-off factors
0
10
[2] G. P. Fettweis, ʠ The Tactile Internet: Applications and Challenges, ʡ
IEEE Veh. Technol. Mag., vol. 9, no. 1, pp. 64-70, Mar. 2014.
[3] N. Tadayon and S. Aissa, ʠ Modeling and Analysis of Cognitive Radio
Based IEEE 802.22 wireless regional area networks, ʡ IEEE Trans. on
−1
10
Wireless Commun., vol. 12, no. 9, pp. 4363-4375, Sep. 2013.
[4] J. Bingham,ʠ Multicarrier Modulation for ɹ Data ɹ Transmission: An
Idea Whose Time Has Come, ʡ IEEE Commun. Mag., vol. 28, no. 5,
pp. 5-14, May 1990.
[5] G. Fettweis, M. Krondorf, and S. Bittner, ʠ GFDM Generalized Fre-
quency Division Multiplexing, ʡ69th IEEE Vehicular Technology Con-
BER

−2
10
ference, Apr. 2009.
[6] N. Michailow, M. Matthe, I. S. Gaspar, A. N. Caldevilla, L. L. Mendes,
A. Festag, and G. Fettweis, ʡGeneralized Frequency Division Multi-
RCn (α=0.9) plexing for 5th Generation Cellular Networks, ʡ IEEE Transactions on
−3 RCw (α=0.9)
10 Commun., vol. 62, no. 9, pp. 3045 - 3061, Sept. 2014.
RCn (α=0.02)
[7] N. Michailow and G. Fettweis, ʡLow Peak-to-Average Power Ratio for
RCw (α=0.02)
RCn (α=0)
Next Generation Cellular Systems with Generalized Frequency Division
RCw (α=0)
Multiplexing, ʡIntelligent Signal Processing and Communications Sys-
tems, pp. 651-655, Nov. 2013
−4
10
0 0.1 0.2 0.3 0.4
[8] M. Matthe, N. Michailow, I. Gaspar, G. Fettweis, ʡInfluence of Pulse
Eb/N0 [dB] Shaping on Bit Error Rate Performance and Out of Band Radiation of
Generalized Frequency Division Multiplexing, ʡ2014 IEEE International
Conference on Communications Workshops, pp. 43-48, June 2014.
[9] A. Farhang, N. Marchetti, and L. E. Doyle, ʡLow Complexity GFDM
Fig. 6. BER vs. Eb /N0 (AWGN, MMSE, [K, C, Mn , Mw ] =
Receiver Design: A New Approach, ʡ2015 IEEE International Confer-
[16, 7, 7, 49]).
ence on Communications, pp. 4775 - 4780, June 2015.
[10] M. Matthe, L. L. Mendes, and G. Fettweis, ʡGeneralized Frequency
−1
10 Division Multiplexing in a Gabor Transform Setting, ʡIEEE Commun.
RCn (α=0) Letters, vol. 18, no. 8, pp. 1379-1382, Aug. 2014.
RCw (α=0)
[11] C. Mori, Y. Tanaka, T. Kawamura, N. Miki and M. Sawahashi ʡ
RCn (α=0.04736)
Performance of Turbo Frequency Domain Equalizer Using Iterative
RCw (α=0.04736)
Decision-Directed Channel Estimation for DFT-precoded OFDMA, ʡ
16th International Symposium on Wireless Personal Multimedia Com-
−2
10
munications, pp. 24-27 June 2013.
[12] J.Maurer, G. Matz,D. Seethaler, ʡLow-Complexity and Full-Diversity
MIMO Detection Based on Condition Number Thresholding, ʡIEEE
International Conference on Speech and Signal Processing, pp. 61-64,
BER

April 2007.

−3
10

−4
10
0.5 0.6 0.7 0.8
E /N [dB]
b 0

Fig. 7. BER vs. Eb /N0 (Rician, MMSE, [K, C, Mn , Mw ] = [16, 3, 7, 21]).

in order to reduce the condition number of the demodulation


matrix.
The numerical results obtained through computer simulation
have shown that optimizing the roll-off factors improves the
BER performance about 0.02 dB at the BER of 10−3 in
comparison to the BER with the roll-off factor of 0.

ACKNOWLEDGMENTS
The authors gratefully knowledge the support of Sony
Corporation.

R EFERENCES
[1] J. Kim, J. Lee, J. Kim, and J. Yun, ʠ M2M Service Platforms: Survey,
Issues, and Enabling Technologies, ʡIEEE Commun. Surveys Tuts., vol.
16, no. 1, pp. 61-76, 2014.

Das könnte Ihnen auch gefallen