Anda di halaman 1dari 5

Outperforming Conventional OFDM and SEFDM

Signals by Means of Using Optimal Spectral Pulses


and the M-BCJR Algorithm
Aleksandr Gelgor, Van Phe Nguyen
Higher School of Applied Physics and Space Technologies
Peter the Great St. Petersburg Polytechnic University
St. Petersburg, Russia
a_gelgor@mail.ru, nvphe1905@gmail.com

Abstract— In the paper, we compared spectral efficiency resource elements in LTE) among users, with that the multiple
between OFDM signals, reputed SEFDM signals, and recently access is achieved.
proposed RRC-SEFDM and PR-SEFDM signals. The last three
are different types of signals with non-orthogonal frequency Despite the above advantages of OFDM signals, there is
division multiplexing, i.e. signals with intersymbol interference still interest to finding new waveforms, which can provide
(ISI). All signals were modulated with QPSK and 16-QAM. For better performance, especially with regard to the spectral
signals with ISI demodulation was performed by the suboptimal efficiency. There are a plenty of publications, where the
M-BCJR algorithm with M = 8 paths survived at each step. For authors propose non-orthogonal signals, i.e. signals with
construction PR-SEFDM signals we used optimal spectral pulses controlled inter-symbol interference (ISI). The effectiveness of
with length L = 12 taps. The simulation results were shown that, introducing ISI was apparently first proved in [1], where single
at first, the maximum gain in spectral efficiency is achieved only carrier signals with linear modulation and sinc-pulse were
with simultaneous introducing ISI and increasing the size of considered, i.e. relatively narrowband signals. These signals
signal constellation. At second, PR-SEFDM signals with optimal were called faster than Nyquist (FTN) signals, because they
spectral pulses provide a significant gain in spectral efficiency exceed the Nyquist limit to increase spectral efficiency. Many
with respect to SEFDM signals and RRC-SEFDM signals at fixed years later in [2] it was proposed the idea of forming SEFDM
energy consumptions and fixed complexity of the demodulation (Spectrally Efficient Frequency Division Multiplexing) signals,
algorithm. Also, for QPSK and 16-QAM, with respect to OFDM
which is virtually completely repeat the idea of forming FTN
signals our proposed PR-SEFDM signals provide up to 27% gain
signals, but it is implemented for relatively broadband signals.
in spectral efficiency by additional expenses up to 0.6 and 1.3 dB
respectively. The difference is that FTN signals are processed in the time
domain while SEFDM signals are processed in the frequency
Keywords—SEFDM signals, ISI, RRC-pulses, optimal pulses, domain. Specifically, SEFDM signals are similar to OFDM
M-BCJR algorithm signals, but adjacent subcarriers are located closer in frequency
domain than is required to ensure their orthogonality. The
spectra of subcarriers in OFDM signals have sinc-shape.
I. INTRODUCTION
Therefore, the reducing the frequency spacing between
Currently, OFDM (Orthogonal Frequency-Division subcarriers or, which is the same, between spectral pulses in
Multiplexing) signals with the cyclic prefix (CP) are used in SEFDM signals, is similar to the reducing time spacing
majority of wireless broadband communication systems. between time-pulses in FTN signals.
Probably, the most popular systems that use OFDM signals are
LTE, WiFi (standards n, ac, ax) and DVB-T2. OFDM signals There are many publications on SEFDM signals, a detailed
provide following significant advantages. Firstly, under some overview of which is presented in [3]. GFDM, FBMC, UFMC
conditions it is possible to use simple one-tap equalizer and SEFDM signals (the last name is the same as above, but
regardless of the complexity of the channel characteristic, one the signals are different) proposed in [4–7], also can be
need only the knowledge of channel state. If the coherence time considered as non-orthogonal frequency division multiplexing.
of the channel exceeds the duration of several OFDM-symbols, However, they differ markedly from SEFDM signals in [2].
as it is in WiFi, then one can use preamble to estimate channel In works [8-9], it is noted that the demodulation of SEFDM
state once per packet. If the coherence time is about the signals can be impossible even with a not very strong ISI. In
duration of one OFDM-symbol, as it is in LTE, then one can [9] it was shown that in some cases a bit error rate (BER) does
use pilot subcarriers to update channel state estimation for each not fall below a certain level, i.e. there is a saturation effect
OFDM-symbol. Secondly, using OFDM signals allows to arises. To resolve this problem, SEFDM signals are usually
effectively distribute time-frequency resources (named coupled with forward error correction, and the iterative
processing technique (turbo equalizer) is utilized at the receiver
This work was supported by the Ministry of Education and Science of the
Russian Federation (the state contract #8.2880.2017/ПЧ).
[8-9]. Note that turbo equalizer is also used to improve
processing OFDM signals [10]. Additionally, in [8-9] authors

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE


deal with only QPSK and/or with a small number of carrier signals can be achieved only by simultaneously
subcarriers. introducing the ISI and increasing the size of the signal
constellation. Therefore, we expect that similar results will be
Considering the above imperfection of SEFDM signals, in obtained for multicarrier signals.
[11] we proposed modified RRC-SEFDM signals. The
distinction of these signals from origin SEFDM signals is the The rest of the paper is organized as follows. Section II
replacing the spectral pulse of sinc shape by the spectral pulse provides a brief description of RRC-SEFDM signals and PR-
of RRC shape (Root Raised Cosine is actually the shape of SEFDM signals, also a spectra comparison is provided. Section
corresponding time pulse). This provides a decrease the deep III contains the parameters for simulation. Section IV presents
and the strength of ISI. the results of simulation; conclusions are provided in section V.
In [12] we proposed PR-SEFDM signals, which are another
variation of origin SEFDM signals. Spectral pulses of PR- II. NON-ORTHOGONAL FREQUENCY MULTIPLEXING
SEFDM signals are, at first, finite, that reduce the
computational complexity of demodulation algorithm. At A. RRC-SEFDM Signals
second, they are optimal by the criterion of the maximization of A baseband RRC-SEFDM signal proposed in [11] can be
the free Euclidean distance for a fixed duration of time pulse. described by
Using the maximal free Euclidean distance provides minimum
energy consumptions by fixed spectral efficiency, i.e. we can N / 2 -1
totally control ISI strength and thus BER. x(t ) = å å Ck( n ) g ( t - kTb ) exp ( j 2p nDf (t - kTb ) ), (1)
k n =- N / 2
The examination of SEFDM, RRC-SEFDM, PR-SEFDM
signals with QPSK in [12] showed that PR-SEFDM signals where g(t) is the following finite time pulse:
provide better spectral efficiency by fixed energy
consumptions. This result was obtained for the Viterbi
algorithm with the fixed computational complexity at the ì 1, t £ T (1 - b ) / 2;
demodulator: it was used trellis with 256 states, which ï
corresponds to the length of the spectral pulse L = 8. Also, it
g (t ) = í
(
ïï 0.5 ëé1 + cos p ( t - T (1 - b ) / 2 ) / b T ûù , ) (2)
was turned out that all of these signals with QPSK lose to the
ï T (1 - b ) / 2 < t £ T (1 + b ) / 2;
conventional OFDM signals with 16-QAM, and provide an ï
insignificant gain with respect to OFDM signals with QPSK. ïî 0, T (1 + b ) / 2 < t ;
The goal of the paper is to find new optimal spectral pulses for
PR-SEFDM signals to provide a gain with respect to the 0 ≤ β ≤ 1 is the roll-off factor; Tβ = (1 + β)T is the RRC-
OFDM signals under assumption of reasonable computational SEFDM symbol duration (the part of the duration (1 – β)T
complexity of the demodulation algorithm. In this context, the equals to zero and is not transmitted); N is the number of
paper can be considered as the evolution of [11] and [12]. subcarriers; Ck(n) is the modulation symbol transmitted on the
According to [13] the longer the pulse duration the smaller n-th subcarrier in the k-th RRC-SEFDM symbol; ∆f = α/T is the
the additional gain. In [13] the difference between L = 12 and frequency distance between adjacent subcarriers; α is the
L = 14 is negligible. Thus, in the paper, we propose to search bandwidth compression factor, 0 < α ≤ 1. Note that if α = 1 and
for optimal spectral pulses no longer than L = 12 taps. To β = 0, then (1) is equal to conventional OFDM signals. The
decrease computational complexity of the demodulator we RRC-SEFDM signals with β = 0 correspond to the origin
propose to utilize the M-BCJR algorithm [14], which is sub- SEFDM signals [2], which will be further referred to as sinc-
optimal version of the BCJR algorithm [15]. At each step of SEFDM signals. The spectral pulse corresponding to the time
forward pass of the M-BCJR algorithm only the best M paths pulse (2) has the RRC-pulse shape:
are surviving. In contrast to [16] in the backward pass, we
propose using only those paths that survived in the forward
pass, as this provides the best results. The important advantage sin{(1 - b )p fT } + cos{(1 + b )p fT }4 b fT
aRRC ( f , b ) = . (3)
of the BCJR algorithm and its sub-optimal version the M- T {1 - (4 b fT ) 2 }p fT
BCJR as compared with the Viterbi algorithm is that they
provide soft decisions at the output, which improves the This fact led to the proposed name of RRC-SEFDM signals. In
efficiency of subsequent FEC decoders. The advantage of the [11], the idea of additional truncating the time pulse to a
BCJR algorithm in relation to the sphere decoding (SD) duration shorter than (1 + β)T was proposed and it is
algorithm [17] is the linear dependence of the computational implemented by zeroing NZS samples at the beginning and
complexity on the length of the processed sequence, while the (NZS – 1) samples at the end of the original pulse of 2T length.
SD algorithm is virtually impossible to process sequences of On the one hand, it allows to additionally increase the spectral
length more than a few dozen [18]. In case of PR-SEFDM efficiency by the reduction of the duration of RRC-SEFDM
signals, this means that with the SD demodulator it is symbol. At the same time, it leads to decrease the peak factor
impossible to process a large number of subcarriers, i.e. to due to truncating the edge parts of the time pulse having small
process broadband signals. Additionally, we propose to amplitude. On the other hand, it also distorts the spectral pulse,
consider the possibility of using 16-QAM. In [13] we have reduces the noise immunity and, consequently, reduces the
shown that the maximum gain in spectral efficiency for single
energy efficiency. However, this approach is effective in spectral pulse a(f) obtained by solving the optimization
general. problem. The criterion of optimization problem is the
maximization the free Euclidean distance under constraining
The disadvantage of using RRC-SEFDM signals is the need the time pulse duration Tε, containing ε percents of pulse
to use FFT block with K times more size than is required for energy. The technique for solving optimization problem is
similar OFDM signals. The purpose is to over-sample spectrum similar to [19]. Thanks to the ISI introduced between
and thus make trellis homogeneous [11]. The value of the subcarriers, the length of the useful part of PR-SEFDM symbol
coefficient K depends on the compression factor α. In general, (i.e. with relatively high amplitude values) is less than the
the stronger the compression of the subcarriers, i.e. the smaller duration of corresponding OFDM symbol, i.e. Tε < T.
value of α, the greater value of K. Therefore, the idea of zeroing NZS samples at the beginning and
A simplified block diagram for generating and processing (NZS – 1) samples at the end of the original pulse of T length is
RRC-SEFDM signals is shown in Fig. 1. At the transmitter, also applied to time pulses of PR-SEFDM signals.
stream of QAM symbols from mapper is divided into N A simplified block diagram for generating and processing
substreams. Zeros are added to each block of N QAM symbols PR-SEFDM signals is shown in Fig. 2. Unlike the transmitter
at the beginning, at the end, and between each pair of QAM of RRC-SEFDM signals, zeros are added only at the beginning
symbols. Time samples obtained from IFFT block are and at the end of block of N QAM symbols. Inserting zeros
elementwise multiplied with the time pulse. Because edge between modulation symbols is not required, since over-
samples will not be transmitted, are simply dropped. sampling is not performed for PR-SEFDM signals, i.e. K = 1.
At the receiver of RRC-SEFDM signals, the procedures are
opposite to the transmitter in general. However, an equalizer is C. Spectra of RRC-SEFDM and PR-SEFDM Signals
added at the beginning, and the demodulator is added at the To compare spectra of RRC-SEFDM and PR-SEFDM
end. There are spectral samples processed in the demodulator. signals with OFDM signals we used vector signal generator
Agilent E8267D and handheld spectrum analyzer N9342C.
B. PR-SEFDM Signals The common parameters of signals are following: carrier
As the development of the idea of RRC-SEFDM signals, in frequency 2.6 GHz, sampling rate 30.72 MHz, number of
[12] we proposed PR-SEFDM signals, where optimal spectral subcarriers 1200 (all according to LTE standard).
pulses are used as the shape of the spectrum of subcarriers. A
baseband PR-SEFDM signal can be described by: From Fig. 3 it follows that, at first, the transition from
OFDM signal to sinc-SEFDM signal with α = 0.5 led to
bandwidth compression at 23.71/12.75 = 1.86 times. It is close
N / 2 -1
x(t ) = å å Ck( n ) g ( t - kTe ) exp ( j 2p n ( t - kTe ) / T ), (4) to the theoretical value 2 achieved for N → ∞. At second, the
k n =- N / 2 bandwidth of RRC-SEFDM signal with β = 1 is less than the
bandwidth of sinc-SEFDM signal with β = 0 (but RRC-
where g(t) is the time pulse corresponding to the optimal finite SEFDM symbol is longer than sinc-SEFDM symbol). At third,

Fig. 1. Block diagram for RRC-SEFDM transmitter and receiver


...

...

...
...

NFFT - IFFT

NFFT - FFT

Equalizer
Mapper

S/P

S/P

P/S
P/S

...
...

...
...

...

...
...

...

...

...

Fig. 2. Block diagram for PR-SEFDM transmitter and receiver


the bandwidth of PR-SEFDM signal is larger than the AWGN channel with ideal estimation of the channel state. In
bandwidth of RRC-SEFDM signal and sinc-SEFDM signal, addition, we used zero-forcing equalizer.
but PR-SEFDM symbol is shorter than RRC-SEFDM symbol To study the dependence of efficiency of the M-BCJR
and sinc-SEFDM symbol. Thus, PR-SEFDM signals can have algorithm on its computational complexity we used several
larger spectral efficiency than RRC-SEFDM and sinc-SEFDM values for number of surviving paths: M = 4, 8, 16.
signals. In addition, it should be noted that the level of out-of-
band emission of RRC-SEFDM and PR-SEFDM signal is
IV. SIMULATION RESULTS
lower than for sinc-SEFDM signal.
Let us define the energy consumptions as the value of
III. SIMULATION PARAMETERS signal-to-noise ratio per modulation bit:
To study spectral efficiency and energy consumptions of
proposed RRC-SEFDM and PR-SEFDM signals we used h 2 = Ebit / N 0 , (5)
following common parameters corresponding to one of the
LTE scenarios: the size of FFT NFFT = 2048, the number of
data subcarriers N = 1200, QPSK or 16-QAM on each We define the spectral efficiency as follows:
subcarrier.

For RRC-SEFDM signals, the value α, the corresponding g = R / W99% , (6)


oversampling coefficient K, and the number of data subcarriers
N are listed in Table 1. For each value α, the following values β where R is the transmission bitrate (bit/s), W99% is the
were considered: from 0 to 1 with the step 0.1. All possible bandwidth comprising 99% energy of signal’s power.
values of NZS were considered in each case. As in [12] the studying characteristics of PR-SEFDM
signals with ε = 0.95, 0.99, 0.999 for QPSK and 16-QAM cases
TABLE I. PARAMETERS OF RRC-SEFDM SIGNALS
showed that the value ε = 0.999 provides the best results.
Α 1 0.9 0.8 0.7 0.6 0.5 Therefore, only the results for PR-SEFDM signals with
K 2 10 5 10 5 2 ε = 0.999 will be presented.
N 1200 1332 1500 1714 2000 2400 The analysis of Fig. 4 shows that as the value of M
For PR-SEFDM signals we used spectral pulses of length decreases the efficiency of the M-BCJR algorithm decreases.
L = 12 taps. The value Tε/T was chosen such that for time pulse However, for h2 ≤ 14 dB, where the ISI is not very strong, the
the number of samples NFFTTε/T containing ε percent of pulse detection quality of M-BCJR algorithm is almost the same for
power is an integer. As with RRC-SEFDM signals, all possible M = 8, 16, and coincides with the one for the “true” BCJR
values of NZS were considered. algorithm (not drawn on Fig. 4). This phenomenon is explained
The aim of the paper is to estimate potential possibilities of by the fact that with not very strong ISI the most energy of the
proposed RRC-SEFDM and PR-SEFDM signals. Therefore, optimal spectral pulse is concentrated at its beginning.
we supposed ideal time and frequency synchronization and The utilizing M-BCJR algorithm for RRC-SEFDM signals
(that is, for sinc-SEFDM too, as sinc-SEFDM is a particular
case of RRC-SEFDM) is not always successful. Since the small

(a) (b)

(c) (d) Fig. 4. Comparison of (γ, h2) dependencies for PR-SEFDM signals with
Fig. 3. Spectra of OFDM signal (a), sinc-SEFDM signal with α = 0.5 and L = 12, ε = 0.999; the results are obtained with using the M-BCJR algorithm
β = 0 (b), RRC-SEFDM signal with α = 0.5 and β = 1 (c), PR-SEFDM signal and M = 4, 8, 16. For PR-SEFDM signals we considered many optimal pulses,
with normalized pulse’s length TεT = 0.75 and ε = 0.99 (d). The occupied and each curve consist of points corresponding to those combinations of pulse
bandwidth contained 99% of signal power shape and NZS value which providing the best results.
values at the edges of the RRC pulses, it is sometimes more
effective to apply a shortened central part of origin pulse in the
, bit/s/Hz BER = 10 -4 1.3 dB
demodulator [20]. This fact explains why the values of pulse’s 5 RRC-SEFDM, QPSK
length L = 4, 8, 12 were also considered at the demodulator for RRC-SEFDM, 16-QAM
25%
RRC-SEFDM signals. PR-SEFDM, QPSK
PR-SEFDM, 16-QAM
From Fig. 5 it follows that, firstly, when using the M-BCJR 4
OFDM
algorithm, PR-SEFDM signals with the optimal spectral pulses
16-QAM
provide a gain in spectral efficiency with respect to RRC-
SEFDM signals. Indeed, with a fixed signal constellation the
curves for PR-SEFDM signals are always higher than curves 3 0.6 dB
for RRC-SEFDM signals. Secondly, the maximum increase in 27%
spectral efficiency is achieved with a simultaneous increasing
the size of the signal constellation and with introducing ISI. OFDM h2 , dB
2 QPSK
This follows from the fact that the curves for PR-SEFDM
signals with 16-QAM are higher than the curves for the QPSK. 8.4 9 10 11 12 13 14 15
Thirdly, with QPSK and 16-QAM PR-SEFDM signals provide 2
Fig. 5. Comparison of (γ, h ) dependencies for PR-SEFDM signals with
gains in spectral efficiency up to 27% with respect to OFDM L = 12, ε = 0.999 and RRC-SEFDM signals with L = 4, 8, 12; the results are
signals. In these cases, the energy losses are no more than obtained with using the M-BCJR algorithm and M = 8. For RRC-SEFDM
1.3 dB. signals we observed many combinations of values α, β, and NZS, including
β = 0, NZS = 0 corresponding to sinc-SEFDM. Each RRC-SEFDM curve
consist of points corresponding to those triplets (α, β, NZS) which providing
V. CONCLUSIONS the best results.
In the paper, we have shown the superiority of our [8] Tongyang Xu, Izzat Darwazeh, “Spectrally Efficient FDM: Spectrum
proposed PR-SEFDM signals over the conventional OFDM Saving Technique for 5G?,” in International Conference on 5G for
signals and origin SEFDM-signals. It was shown that the Ubiquitous Connectivity (5GU), 2014, pp. 273-278.
maximum gain in spectral efficiency is achieved only with [9] Rashich, A., Kislitsyn, A., Fadeev, D., Ngoc Nguyen, T., “FFT-based
trellis receiver for SEFDM signals,” in IEEE Global Communications
simultaneously increasing the size of signal constellation and Conference (GLOBECOM), Washington, DC, 2016, pp. 1-6.
with introducing the controlled ISI. All these results are [10] A. Gelgor, A. Gorlov, P. Ivanov, E. Popov, A. Arkhipkin, T. Gelgor,
obtained with the M-BCJR algorithm at the demodulator and “Improving BER Performance of Uplink LTE by Using Turbo
only with M = 8 surviving paths at each step of the algorithm. Equalizer,” Lecture Notes in Computer Science, 2015, pp. 459-472.
[11] A. Gelgor, A. Gorlov and Van Phe Nguyen, “The design and
ACKNOWLEDGEMENTS performance of SEFDM with the sinc-to-RRC modification of
subcarriers spectrums,” in IEEE International Conference on Advanced
The simulation results were obtained with the use of Technologies for Communications (ATC), 2016, pp. 65–69.
computational resources of the Supercomputing Center of Peter [12] A. Gelgor, A. Gorlov and Van Phe Nguyen, “Performance analynis of
the Great St. Petersburg Polytechnic University SEFDM with optimal subcarriers spectrum shapes,” in Black Sea
Conference on Communications and Nerworking (BlackSeaCom), 2017
(http://www.scc.spbstu.ru). IEEE International, pp. 1–5.
[13] Van Phe Nguyen, A. Gorlov, A. Gelgor, “An Intentional Introduction of
REFERENCES ISI Combined with Signal Constellation Size Increase for Extra Gain in
Bandwidth Efficiency,” Lecture Notes in Computer Science, 2017, pp.
[1] J.E. Mazo, “Faster-than-Nyquist signaling,” Bell System Technical
644-652.
Journal, vol. 54, no. 8, pp. 1451–1462, 1975.
[14] V. Franz and J. B. Anderson, “Concatenated Decoding with a Reduced-
[2] M. Rodrigues and I. Darwazeh, “A spectrally efficient frequency
Search BCJR Algorithm,” IEEE Journals on Select. Area. Commun.,
division multiplexing based communications system,” in Proc. 8th Int.
vol. 16, no. 2, pp. 186-195, 1998.
OFDM Workshop, Hamburg, 2003, pp. 48–49.
[15] L. Bahl, J. Cocke, F. Jelinek and J. Raviv, “Optimal decoding of linear
[3] Izzat Darwazeh, Hedaia Ghannam, Tongyang Xu, “The First 15 Years of
codes for minimizing symbol error rate,” IEEE Trans. Inf. Theory, vol.
SEFDM: A Brief Survey,” 2018 11th International Symposium on
20, no. 2, pp. 284–287, 1974.
Communication Systems, Networks & Digital Signal Processing
(CSNDSP), Year: 2018, Page s: 1–7. [16] A. Plotnikov, A. Gelgor, “Spectral Efficiency Comparison Between
FTN Signaling and Optimal PR Signaling for Low Complexity
[4] N. Michailow, M. Matthe, I. Gaspar, A. Caldevilla, L. Mendes, A.
Detection Algorithm,” Lecture Notes in Computer Science, 2018, pp.
Festag, and G. Fettweis, “Generalized frequency division multi-plexing
191-199.
for 5th generation cellular networks,” Communications, IEEE
Transactions on, vol. 62, no. 9, Sept 2014, pp. 3045–3061. [17] U. Fincke and M. Pohst, “Improved Methods for Calculating Vectors of
Short Length in a Lattice, Including a Complexity Analysis,”
[5] M. Schellmann, Z. Zhao, H. Lin, P. Siohan, N. Rajatheva, V. Luecken,
mathematics of computation, vol. 44, no. 170, pp. 463-471, 1985.
and A. Ishaque, “FBMC-based air interface for 5G mobile: Challenges
and proposed solutions,” in CROWNCOM, 2014 9th International [18] Babak Hassibi and Haris Vikalo, “On the Sphere-Decoding Algorithm I.
Conference on, June 2014, pp. 102–107. Expected Complexity,” IEEE Trans. sig. proc., vol. 53, no. 8, 2005.
[6] V. Vakilian, T. Wild, F. Schaich, S. Ten Brink, and J.-F. Frigon, [19] Said and J.B. Anderson, “Bandwidth-efficient coded modulation with
“Universal-filtered multi-carrier technique for wireless systems beyond optimized linear partial-response signals,” IEEE Trans. Inform. Theory,
LTE,” in Globecom Workshops (GC Wkshps), 2013 IEEE, Dec 2013, vol. 44, no. 2, pp. 701–713, 1998.
pp. 223–228. [20] A. Gorlov, A. Gelgor, V.P. Nguyen, “Root-Raised Cosine versus
[7] S. Zavjalov, S. Volvenko, S. Makarov, "A Method for Increasing the Optimal Finite Pulses for Faster-than-Nyquist Generation”, Lecture
Spectral and Energy Efficiency SEFDM Signals," in IEEE Notes in Computer Science, 2016, pp. 628-640.
Communications Letters, vol. 20, issue 12, pp. 2382-2385, 2016.

Anda mungkin juga menyukai