Professional Documents
Culture Documents
Signl Processing
Signl Processing
Signl Processing
2169-3536
2018 IEEE. Translations and content mining are permitted for academic research only.
19610 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 6, 2018
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
J. Nadal et al.: Design and Evaluation of a Novel Short Prototype Filter for FBMC/OQAM Modulation
FIGURE 1. System description of the PPN and FS implementations of the FBMC receivers.
implementations of the FBMC receiver, respectively referred filtering steps but also to the applied modifications to the
to as PPN-FBMC and FS-FBMC in this paper, are considered. modulator/demodulator architecture. However, the usage of
It is shown that: digital polyphase filter bank structures [5], [7], together with
1) the NPR1 PF achieves improved robustness against the rapid growth of digital processing capabilities in recent
all the considered channel impairments when the years have made FBMC a practically feasible approach.
FS-FBMC receiver is considered, In the literature, two types of implementation for the
2) the FS-FBMC receiver offers improved robustness FBMC modulation exist, each having different hardware
against timing offset and multipath impairments when complexity and performance. The first one is the PPN imple-
compared to the PPN-FBMC receiver, mentation [8], illustrated in Figure 1, which is based on an
3) by exploiting the different properties of the PFs, IFFT and a PPN for the filtering stage, and enables a low
a substantial reduction in hardware complexity can be complexity implementation of the FBMC transceiver.
achieved for the FS-FBMC receiver. The second type of implementation is the FS imple-
Particularly, it is shown in this paper that the hardware mentation (Figure 1), proposed in [9] and [10] for the
complexity of the FS-FBMC receiver is lower than the Martin–Mirabassi–Bellange PF with an overlapping factor
PPN-FBMC receiver when the NPR1 PF is considered. equal to 4 (MMB4), considered for FBMC during PHYDYAS
The rest of the paper is organized as follows. Section II project [11]. The original idea was to shift the filtering stage
provides a technical description of the FBMC modulation into the frequency domain, in order to enable the use of a
with different types of implementation. Section III is ded- low-complexity per-sub-carrier equalizer as in OFDM. The
icated to the presentation of the proposed novel short PF hardware complexity is supposed to be higher than the com-
along with State-of-The-Art (SoTA) existing ones. Section IV plexity of the PPN implementation, at least for long PFs.
evaluates and compares the performance of all considered In fact, it requires one FFT of size L = KM per FBMC
PFs with several channel impairments. Section V presents symbol, where K is the overlapping factor of the PF, and
the proposed hardware architecture of the FS filtering stage M is the total number of available sub-carriers. However,
and illustrates the complexity reduction with respect to the in the short PF case (K = 1), the size of the FFT is same
PPN-FBMC architecture. Finally, Section VI concludes the as for the PPN implementation.
paper. The rest of the section provides a mathematical background
of the PPN-FBMC transceiver and the FS-FBMC receiver.
II. FBMC/OQAM SYSTEM DESCRIPTION
FBMC is a multicarrier transmission scheme that introduces A. POLYPHASE NETWORK-BASED IMPLEMENTATION
a filter-bank to enable efficient pulse shaping for the signal If M is the total number of available sub-carriers and an (m)
conveyed on each individual subcarrier. This additional ele- the Pulse-Amplitude Modulation (PAM) symbol at subcarrier
ment represents an array of band-pass filters that separate index m and time slot n, then the baseband signal s(k) can be
the input signal into multiple components or subcarriers, mathematically decomposed as follows:
each one carrying a single frequency sub-band of the orig-
+∞
inal signal. As a promising variant of filtered modulation X M
s(k) = g k −n xn (k), (1)
schemes, FBMC, originally proposed in [4] and also called 2
n=−∞
OFDM/OQAM [5] or staggered modulated multitone (SMT)
M −1
[6], can potentially achieve a higher spectral efficiency than X km
xn (k) = (−1)nm an (m)φn (m)ei2π M , (2)
OFDM since it does not require the insertion of a Cyclic-
m=0
Prefix (CP). Additional advantages include the robustness
against highly variant fading channel conditions and imper- with i2 = −1. To keep the orthogonality in the real field,
fect synchronization by selecting the appropriate PF type and φn (m) must be a quadrature phase rotation term. In the lit-
coefficients [3]. Such a transceiver structure usually requires erature, it is generally defined as φn (m) = in+m , as in [3].
a higher implementation complexity related not only to the The impulse response of the PF is g, with g(l) = 0 when
l 6 ∈ [0, L − 1], where L is the length of the PF. In practice, to obtain the Xn signal in frequency domain (6). Then, it intro-
the PPN-FBMC transmitter is implemented using an IFFT duces a filtering stage in frequency domain, as described
of size M followed by a PolyPhase Network. When a short in (7): the frequency response of the PF G is convoluted with
PF is used (K = 1), this latter can be seen as a windowing the Xn signal, for instance using a Finite Impulse Response
operation: the outputs of the IFFT are simply multiplied by (FIR) filter. Finally, the recovered PAM symbols are obtained
the PF impulse response g. Consequently, the complexity by extracting the real part of quadrature phase rotated and
overhead introduced by the PPN is limited. Note that, due down-sampled Yn samples (8).
to the OQAM scheme, the obtained FBMC symbol over- These operations are summarized and illustrated
laps with both the previous and next symbols on half of in Figure 1. The FS-FBMC implementation seems highly
the symbol length. Therefore, for practical implementation, complex, however G has a lot of zero coefficients due to its
2 FBMC symbols may be generated in parallel. It is however frequency localization. Therefore, it can be truncated down
possible to avoid the use of two IFFT blocks at the trans- to NG coefficients. Then, by defining 1 = (NG − 1)/2 (NG
mitter side through the use of the pruned FFT algorithm. is considered an odd number), (7) becomes:
This leads to a reduced-complexity implementation presented
in [12] and [13]. 1−1
X
Receiver side implementation applies dual operations with Yn (m)(k) = G(l)Xn (m − l) (9)
respect to the ones performed by the constituent blocks of l=−1
the transmitter. The IFFT must be replaced by an FFT, and
the operations order must be reversed: PPN (windowing if
III. PROPOSAL OF A NOVEL SHORT PROTOTYPE FILTER
K = 1), FFT then OQAM demapper, as shown in Figure 1.
Current literature often focuses on FBMC using a PF with
If r is the received signal and ân (m) are the recovered PAM
a duration 4 times larger than an OFDM symbol (K = 4),
symbols, then we have:
like MMB4 or Isotropic Orthogonal Transform Algorithm 4
K −1
(IOTA4) [16]. However, a shorter PF can also be applied,
X M
un (k) = g(k + lM ) r k + (2l + n) . (3) as proposed in [17] with the Time-Frequency Localization 1
2
l=0 (TFL1) PF. Another example is the Quadrature Mirror Fil-
M −1
X km ter 1 (QMF1) [18] which was recently applied to FBMC
Un (m) = Cn (m) un (k)e−i2π M . (4) leading to a variant denoted by Lapped-OFDM modulation
k=0
and presented in [19]. In the rest of this paper, PFs with
ân (m) = < φn∗ (m)Un (m) . (5) a duration larger than one OFDM symbol will be referred
to as long PFs (e.g. MMB4, IOTA4), and the ones with a
where ∗ represents the complex conjugate operation, and
duration of one OFDM symbol as short PFs (e.g. TFL1,
Cn (m) is the Zero Forcing (ZF) equalizer coefficient to com-
QMF1). When compared to long PFs, short PFs provide
pensate the impairments introduced by the channel. Note that,
lower latency, higher robustness against Carrier Frequency
contrary to transmitter side, doubling the FFT processing
Offset (CFO) and higher spectral efficiency due to short-
cannot be avoided using the pruned FFT algorithm. The main
ened transition between two successive transmission frames.
reason is that the equalization term Cn (m) introduces complex
In this context, finding a new short PF with good performance
valued coefficients.
and low hardware complexity becomes a challenging task of
high interest. In the following, after a short description of the
B. FS-FBMC RECEIVER DESCRIPTION
two existing short PFs in the literature, we present a novel
The FS-FBMC implementation is generally considered at the short PF design that shows significant advantages in terms of
receiver side, since it enables a low-complexity and efficient performance and complexity.
equalization scheme [14], [15]. It remains perfectly compati-
ble with the PPN implementation at the transmitter side. The A. TFL1 AND QMF1 PROTOTYPE FILTERS
received symbols are expressed as follows:
The TFL1 PF was the first attempt to specifically design a
L−1 time and frequency localized short PF for the FBMC modu-
X M −i 2π km
Xn (m) = Cn (m) r(k + n )e L (6) lation [17]. It is the most known short PF in the literature.
2
k=0 Indeed, it is already integrated into proof-of-concept hard-
L
2 −1
X ware platforms [20], [21]. The analytical expression of the
Yn (m) = G(l)Xn (m − l) (7) TFL1 PF [22] is given by:
l=− L2 π
g(k) = (1 − d) + γ0 t + 2t(t 2 − 1)(β1 + 4β2 t 2 ),
2
ân (m) = < Yn (Km)φn∗ (m) (8)
2k+1
where d = M with k ∈ [0, M2 − 1], t = 2d − 1 and
where G is the frequency response of the PF. The FS-FBMC
receiver first applies an IFFT of size L on the part of the 1
γ0 (k) =
r signal containing the FBMC symbol to demodulate in order D0 + D1 M2
TABLE 1. Dl coefficients of the analytical expression of TFL1 PF. In this case, we have r(k) = g(k), and Un (m) in (4) becomes:
M −1
X M −i 2π km
Un (m) = g(k)g(k + n )e M . (11)
2
k=0
FIGURE 3. PSD evaluation of the FBMC short PFs in a 4.5 MHz bandwidth.
TABLE 3. NG values needed to reach target SIR for different PFs. If − LCP
2 < ld < LCP 2 , where LCP is the length of the
cyclic prefix, then orthogonality is perfectly restored, since a
circular shift in time domain represents a linear phase rotation
in frequency domain. In 4G/LTE, LMCP = 7% for OFDM.
ld
Thus, orthogonality is still guaranteed if | M | < 3.5%, where
|.| represents the absolute value operator.
Due to the absence of CP in a FBMC system, timing
offset will result in unavoidable performance degradation.
truncation of 55 dB may be sufficient since channel impair- However, depending on the use of PPN or FS implementa-
ments already degrade the resulting SIR, as illustrated in the tion, results are different due to the application of different
next sub-sections. For the rest of the paper, NG is chosen timing offset compensation techniques. For the FS implemen-
so that each PF has the same SIR of 55 dB, enabling a fair tation, the compensation step lies between the FFT and the
comparison. Therefore, we have: FS filtering stage, whereas in the PPN case it is performed
after the PPN and the FFT.
• NG = 31 for TFL1.
For the PPN-FBMC case, the SIR expression in (16) can be
• NG = 7 for NPR1.
adapted to obtain the expression of the recovered PAM sym-
• NG = 41 for QMF1.
bol when a timing offset of ld samples is applied, as follows:
The TFL1 and QMF1 PFs require more than 30 non-zero
coefficients to obtain this SIR target for a FS implementation. Q M −1
Such high number of coefficients may not be acceptable if a X X M
ân (m) = < CTO (m)φn∗ (m) g(k)g k + q +ld
low complexity receiver is targeted. For the NPR1 PF, only 2
q=−Q k=0
7 coefficients are required to achieve a SIR up to 55 dB,
making it better suited for the FS implementation. It is
M km
× xn−q k + n + ld e−i2π M
worth noting that a convolution operation using 7 coefficients 2
may appear too complex to implement in practice. There-
X
M
fore, a low-complexity hardware architecture is proposed in = < i−q−p Fg (p, q + ld )
2
Section V to address this aspect. (p,q)∈
pld
× an−q (m − p)e−i2π M ,
C. ROBUSTNESS TO TIMING OFFSET
Timing offset impairment occurs when the transmitter and and the expression of the SIR becomes:
receiver baseband samples are not perfectly aligned in time. 2
It is always the case in practice, since the channel introduces < Fg (0, ld )
a propagation delay. Therefore, timing synchronization algo- SIRPPN (ld ) = h pl i2
rithms must be employed. In LTE uplink case, the timing syn- p+q F (p, q M + ld)e−i2π Md
P
(p,q)∈0 < i g 2
chronization is realized using time advance mechanism [26]
to compensate the propagation delay of each User Equip- Note that the number of FBMC symbols acting as interfer-
ment (UE) located at different geographical distance from ence denoted by Q must be set to 2. Indeed, when a timing
the base station. However, new highly demanding scenarios offset is considered, the FBMC symbols at q = −M and
like massive machine communications are considered in 5G. q = M are now acting as interference. Therefore, the obtained
To reduce energy consumption and to improve spectral numerical values are similar to that obtained by simulation
usage, time advance mechanism should be avoided and in Figure 5. Concerning the FS-FBMC receiver, it has been
relaxed synchronization should be supported, where the prop- evaluated in [27], where the following expression is obtained:
agation delay of each UE is not compensated. Therefore, 1
synchronization errors appear, which causes two types of SIRFS (ld ) = 2 (25)
σINT P d −1 2
impairments: σa + 2 ll=0 g (l)
1) Linear phase rotation for each sub-carrier due to the
additional delay. This effect can be totally compen- where σINT
2 is the power of the residual interference when
sated after channel estimation and equalization. Indeed, ld = 0. In our case, the residual interference comes from the
if ld is the time offset in number of samples, then the NPR nature of the PF and the truncation applied on the filter
frequency domain compensation term is expressed as coefficients. Therefore, we have:
2 π m ld σ 2
CTO (m) = e−i M .
h i2
INT
X
= < ip+q Fg,gt (p, q)
2) Inter-Symbol and Inter-Carrier Interference (ISI and σa 0
(p,q)∈
ICI) due to PF misalignment between the transmitter
and the receiver. Figure 5 shows SIR values for each considered short PF
It is considered, in this paper, that the OFDM signal is using the FS-FBMC receiver. The NG parameters used for
synchronized (ld = 0) at the middle of its cyclic prefix. the PFs are those defined in Sub-section IV-B. It is clear
FIGURE 5. Timing offset evaluation in terms of measured SIR for OFDM FIGURE 6. SIR evaluation in presence of CFO.
and FBMC with the considered short PFs. The effect of different
implementations is also evaluated.
FIGURE 10. SIR as function to the sub-carrier index for FS-FBMC and
PPN-FBMC with different short PFs, for an ETU channel.
all the considered PFs, particularly at higher SNR values. V. COMPLEXITY EVALUATION
A difference of at least one decade of BER can be observed The FS-FBMC receiver is known to be more complex than
at Eb /N0 = 28 dB. Furthermore, the FS implementation the PPN-FBMC receiver. This is mainly due to the required
with TFL1 and QMF1 PFs shows comparable performance convolution operation with NG truncated coefficients, com-
to OFDM with CP. FS implementation with NPR1 offers pared to the simple windowing operation of the PPN-based
implementation when short PFs are used. However, addi- since half of the circular convolution outputs are discarded
tional complexity reduction is possible for the FS-based due to the OQAM scheme. In this case, we have:
receiver thanks to the properties of the PF and the OQAM 1+1
X
scheme. After detailing the proposed complexity reduction â02n (2m) = (−1) (n+m)
G0 (l)< X2n
0
(2m, l) ,
approach, a hardware architecture is proposed for the FS filter l=0
stage and its complexity is evaluated and compared to the 1+1
X
PPN filter stage. â02n (2m + 1) = (−1) (n+m+1)
G0 (l)= X2n
0
(2m+1, l) ,
l=0
A. COMPLEXITY REDUCTION OF THE FILTER STAGE 1+1
X
When a short PF is considered (K = 1), the FFT size becomes â02n+1 (2m) = (−1) (n+m+1)
G0 (l)= X2n+1
0
(2m, l) ,
equal to M , and there is no down-sampling step after the filter l=0
stage. Therefore, the complexity overhead comes from the 1+1
X
circular convolution operation, which depends on the number â02n+1 (2m + 1) = (−1)(n+m+1) G0 (l)< X2n+1
0
(2m, l) ,
of filter coefficients NG after truncation. A circular convo- l=0
lution typically requires NG Complex Multiplications (CMs) (30)
and 2(NG − 1) Real Additions (RAs) per sample. However,
where â0n (m) being the re-scaled ân (m) signal. Therefore,
these resources can be reduced by exploiting the properties of
the PF and the OQAM scheme. First, if g(k) = g(N − k) and there is no need to process = Xn0 (m, l) or < Xn0 (m, l)
=(g) = 0, verified for the NPR1 PF (14), then we have: depending on the parity of n and m. Only 1 RMs and 21 RAs
/2−1
NX
per sample are now required. Additionally, only the outputs
2π 2π
(N −k)l corresponding to the allocated sub-carrier indexes can be
G(l) = g(k)ei N kl
+ g(N − k)ei N
considered. If the number of allocated sub-carriers is denoted
k=0
by Nc , this gives 1Nc RMs required per FBMC symbol.
/2−1
NX
2π Regarding the PPN stage, it requires 2M RMs per FBMC
= 2g(k)<(ei N kl ). (26)
symbol for short PFs. Thus, when considering only multipli-
k=0
cation operations, the complexity ratio between FS and PPN
This expression shows that G is real valued. Therefore, the fil- filter stages is given by RRM = α1/2, with α = Nc /M ≈
ter stage requires now only 2NG Real Multipliers (RMs) per 0.586 in 4G/LTE. For QMF1 (1 = 20) and TFL1 (1 = 15)
sample. Furthermore, G(−l) can be expressed as follows: PFs, the complexity (in number of RMs) is multiplied by 5.86
−1
NX ∗ and 4.395 respectively. This confirms that these PFs are not
2π
G(−l) = g(k)e−i N kl suitable for the FS implementation. However, for NPR1 PF
k=0 (1 = 3), the FS filter stage is 12% less complex than the
= G(l). (27) PPN stage.
It is worth noting that the PPN stage requires a LUT
Consequently, the output of the filter stage Yn (m) becomes: memory of depth M to store the PF coefficients. Furthermore,
1+1
X the FS filter stage still require 21Nc additions per FBMC
Yn (m) = G(0)Xn (m)+ G(l) Xn (m − l) + Xn (m + l) symbols. When targeting hardware implementation, registers
l=1 must also be considered to store the input signal Xn0 (k) due to
1+1
X the iterative processing of the circulation convolution oper-
= G(l)Xn (m, l), ation. Finally, this operation uses constant filter coefficients
l=0 which do not change during the processing iterations. This
(28) particularity can be taken into account to further reduce the
where Xn (m, l) = Xn (m − l) + Xn (m + l) if l > 0 and complexity when considering hardware implementation.
Xn (m, 0) = Xn (m). One RM can be removed by re-scaling the Therefore, the above ratio may not be accurate enough to
PF coefficients by G0 (l) = G(l)/G(0). Then, (28) becomes: reflect the comparative hardware complexity since it only
considers the number of multipliers. For an accurate compar-
1+1
X ison, we propose a detailed hardware architecture for the FS
Yn (m) = G0 (l)Xn0 (m, l). (29) filter stage.
l=0
with Xn0 (m, l) = G(0)Xn (m, l). This scaling factor G(0) can B. PROPOSED HARDWARE ARCHITECTURE FOR
be integrated without complexity increase in the equalizer THE FS FILTER STAGE
coefficients. Alternatively, it can be taken into account in the The circular convolution operation can be efficiently imple-
decision stage (QAM demapper). The re-scaled PF coeffi- mented in hardware using a typical FIR filter architecture.
cients G0 (l) can be computed during design time, and stored in Such architecture can take one input and generate one output
a Look-Up-Table (LUT). At this step, the filter stage requires per clock cycle in pipelined manner. If a Multiple Constant
21 RMs per sample. This number can be further reduced Multiplier (MCM) is used, multiplier-less FIR architecture
can be designed for fixed-point precision. If GQ are the PF we propose a novel architecture adapted for the FS filter
coefficients quantized on Q-bit to be multiplied by a Q-bit stage.
input XQ , then: In the following, only even FBMC symbol indexes are
Q−1
considered (â02n (m) and X2n
0 (m)). Similar demonstration can
X be applied for the odd indexed symbols. Index 2m of (30) can
GQ × X Q = ck 2k XQ , (31)
be rewritten as follows:
k=0
The above equations show that the FS filter stage can TABLE 4. Required hardware resources for each considered unit.
be separated into two FIR filters, each respectively holding
even and odd indexes of the PF coefficients. Indeed, for each
received X2n0 (2m) sample, its real part is processed by the
complexity evaluation shows that the proposed FS-based [20] M. Lanoiselee et al., ‘‘Comparative evaluation on real-time hard-
FBMC receiver, using the NPR1 PF, requires 11% less hard- ware platforms of coded OFDM/QAM and OFDM/OQAM systems,’’
in Proc. Int. Symp. Wireless Commun. Syst. (ISWCS), Aug. 2012,
ware resources than the PPN-based FBMC receiver. There- pp. 186–190.
fore, combining the proposed NPR1 filter and FS-FBMC [21] J. Nadal, C. A. Nour, A. Baghdadi, and H. Lin, ‘‘Hardware prototyping of
receiver architecture provides an original solution that com- FBMC/OQAM baseband for 5G mobile communication systems,’’ in Proc.
IEEE Int. Symp. Rapid Syst. Prototyping (RSP), Delhi, India, Oct. 2014,
bines complexity reduction and performance improvement pp. 135–141.
with respect to a typical PPN-FBMC receiver. [22] D. Pinchon and P. Siohan, ‘‘Derivation of analytical expressions for flexible
PR low complexity FBMC systems,’’ in Proc. 21st Eur. Signal Process.
Conf. (EUSIPCO), Sep. 2013, pp. 1–5.
REFERENCES [23] C. Jin, S. Hu, Y. Huang, F. Li, J. Zhang, and S. Ma, ‘‘On design of
[1] M. Maternia et al. (Apr. 2016). 5G PPP Use Cases and Performance Eval- conjugated transmission scheme for FBMC/OQAM systems with inter-
uation Models. [Online]. Available: https://5g-ppp.eu/wp-content/uploads/ ference cancellation,’’ China Commun., vol. 14, no. 8, pp. 166–175,
2014/02/5G-PPP-use-cases-and-performance-evaluation-modeling_v1.0. Aug. 2017.
pdf [24] M. Bellanger et al., ‘‘FBMC physical layer: A primer,’’ Phydyas, vol. 25,
[2] M. Schellmann et al., ‘‘FBMC-based air interface for 5G mobile: Chal- pp. 7–10, Jan. 2010.
lenges and proposed solutions,’’ in Proc. 9th Int. Conf. Cognit. Radio [25] S. Jošilo, M. Pejović, B. Ðorević, M. Narandžić, and S. Nedić,
Oriented Wireless Netw. Commun., Jun. 2014, pp. 102–107. ‘‘Multicarrier waveforms with I/Q staggering: Uniform and
[3] H. Lin, M. Gharba, and P. Siohan, ‘‘Impact of time and carrier nonuniform FBMC formats,’’ EURASIP J. Adv. Signal Process.,
frequency offsets on the FBMC/OQAM modulation scheme,’’ Signal vol. 184, no. 1, p. 167, Nov. 2014. [Online]. Available: https://
Process., vol. 102, pp. 151–162, Sep. 2014. asp-eurasipjournals.springeropen.com/articles/10.1186/1687-6180-
[4] B. Saltzberg, ‘‘Performance of an efficient parallel data transmission sys- 2014-167
tem,’’ IEEE Trans. Commun. Technol., vol. COM-15, no. 6, pp. 805–811, [26] Requirements for Support of Radio Resource Management,
Dec. 1967. document 3GPP TS 36.133, 2010. [Online]. Available: http://www.
[5] P. Siohan, C. Siclet, and N. Lacaille, ‘‘Analysis and design of 3gpp.org/dynareport/36133.htm
OFDM/OQAM systems based on filterbank theory,’’ IEEE Trans. Signal [27] D. Mattera, M. Tanda, and M. Bellanger, ‘‘Performance analysis of some
Process., vol. 50, no. 5, pp. 1170–1183, May 2002. timing offset equalizers for FBMC/OQAM systems,’’ Signal Process.,
[6] P. Sabeti, H. Saeedi-Sourck, and M. Omidi, ‘‘Low-complexity CFO correc- vol. 108, pp. 167–182, Mar. 2015.
tion of frequency-spreading SMT in uplink of multicarrier multiple access [28] D. Mattera, M. Tanda, and M. Bellanger, ‘‘Frequency domain CFO com-
networks,’’ in Proc. 23rd Iranian Conf. Elect. Eng. (ICEE), May 2015, pensation for FBMC systems,’’ Signal Process., vol. 114, pp. 183–197,
pp. 410–415. Sep. 2015.
[7] M. Bellanger and J. Daguet, ‘‘TDM-FDM transmultiplexer: Digital [29] Base Station (BS) Radio Transmission and Reception, document 3GPP
polyphase and FFT,’’ IEEE Trans. Commun., vol. COM-22, no. 9, TS 36.104, 2010. [Online]. Available: http://www.3gpp.org/dynareport/
pp. 1199–1205, Sep. 1974. 36104.htm
[8] B. Hirosaki, ‘‘An orthogonally multiplexed QAM system using the dis- [30] Physical Channels and Modulation, document 3GPP TS 36.211, 2011.
crete Fourier transform,’’ IEEE Trans. Commun., vol. COM-29, no. 7, [Online]. Available: http://www.3gpp.org/dynareport/36211.htm
pp. 982–989, Jul. 1981. [31] W. Cui, D. Qu, T. Jiang, and B. Farhang-Boroujeny, ‘‘Coded auxiliary
[9] M. Bellanger, ‘‘FS-FBMC: An alternative scheme for filter bank based pilots for channel estimation in FBMC-OQAM systems,’’ IEEE Trans. Veh.
multicarrier transmission,’’ in Proc. 5th Int. Symp. Commun. Control Technol., vol. 65, no. 5, pp. 2936–2946, May 2016.
Signal Process. (ISCCSP), May 2012, pp. 1–4. [32] R. M. Hewlitt and E. S. Swartzlantler, ‘‘Canonical signed digit representa-
[10] K. W. Martin, ‘‘Small side-lobe filter design for multitone data- tion for FIR digital filters,’’ in Proc. IEEE Workshop Signal Process. Syst.,
communication applications,’’ IEEE Trans. Circuits Syst. II, Analog Digit. Oct. 2000, pp. 416–426.
Signal Process., vol. 45, no. 8, pp. 1155–1161, Aug. 1998. [33] M. Püschel et al., ‘‘SPIRAL: Code generation for DSP transforms,’’ Proc.
[11] Phydyas Project. Accessed: Feb. 6, 2018. [Online]. Available: IEEE, vol. 93, no. 2, pp. 232–275, Feb. 2005.
http://www.ict-phydyas.org
[12] Y. Dandach and P. Siohan, ‘‘FBMC/OQAM modulators with half complex-
ity,’’ in Proc. IEEE Global Telecommun. Conf. (GLOBECOM), Dec. 2011,
pp. 1–5.
[13] J. Nadal, C. A. Nour, and A. Baghdadi, ‘‘Low-complexity pipelined archi-
tecture for FBMC/OQAM transmitter,’’ IEEE Trans. Circuits Syst. II, Exp.
Briefs, vol. 63, no. 1, pp. 19–23, Jan. 2016.
[14] M. Bellanger, ‘‘FS-FBMC: A flexible robust scheme for efficient
multicarrier broadband wireless access,’’ in Proc. IEEE Globecom
Workshops (GC Wkshps), Mar. 2012, pp. 192–196.
[15] V. Berg, J.-B. Dore, and D. Noguet, ‘‘A flexible FS-FBMC receiver JÉRÉMY NADAL received the Ph.D. degree
for dynamic access in the TVWS,’’ in Proc. 9th Int. Conf. Cognit. from Telecom Bretagne, Brest, France, in 2017.
Radio Oriented Wireless Netw. Commun. (CROWNCOM), Jun. 2014, He was a Research Engineer with the Elec-
pp. 285–290. tronics Department from 2013 to 2014. He is
[16] B. Le Floch, M. Alard, and C. Berrou, ‘‘Coded orthogonal frequency currently a Post-Doctoral Researcher with the
division multiplex [TV broadcasting],’’ Proc. IEEE, vol. 83, no. 6, IMT Atlantique/Lab-STICC laboratory. His gen-
pp. 982–996, Jun. 1995. eral research interests include both algorithm
[17] D. Pinchon, P. Siohan, and C. Siclet, ‘‘Design techniques for orthogonal
development for 5G waveform candidates and cor-
Modulated filterbanks based on a compact representation,’’ IEEE Trans.
responding efficient hardware architecture design.
Signal Process., vol. 52, no. 6, pp. 1682–1692, Jun. 2004.
[18] H. S. Malvar, ‘‘Modulated QMF filter banks with perfect reconstruction,’’ In this context, he proposed several original con-
Electron. Lett., vol. 26, no. 13, pp. 906–907, Jun. 1990. tributions for FBMC/OQAM, FC-OFDM, and UF-OFDM waveforms, asso-
[19] M. Bellanger, D. Mattera, and M. Tanda, ‘‘Lapped-OFDM as an alter- ciated with corresponding hardware proof-of-concept (Exhibition at Mobile
native to CP-OFDM for 5G asynchronous access and cognitive radio,’’ World Congress 2015, European Conference on Networks, and Communi-
in Proc. 81st IEEE Veh. Technol. Conf. (VTC Spring), May 2015, cations EuCNC 2014 and 2017). He has actively participated in METIS and
pp. 1–5. Fantastic-5G European research projects.
CHARBEL ABDEL NOUR received the com- AMER BAGHDADI (SM’15) received the engi-
puter and communications engineering degree neering, M.Sc., and Ph.D. degrees from the
from Lebanese University in 2002, the master’s Grenoble Institut National Polytechnique, France,
degree in digital communications from the Uni- in 1998 and 2002, respectively. He received the
versity of Valenciennes and Hainaut-Cambresis, accreditation to supervise research in sciences and
France, in 2003, and the Ph.D. degree in digital technologies of information and communication
communications from Telecom Bretagne, France, from the University of Southern Brittany, France,
in 2008. Since 2007, he has been a Post-Doctoral in 2012. He is currently a Professor with the
Fellow with the Electronics Department, Telecom IMT Atlantique/Lab-STICC laboratory. His gen-
Bretagne. Since 2011, he has been an Associate eral technical area concerns both theoretical and
Professor with the Electronics Department, IMT Atlantique. His inter- practical aspects, and both algorithm development for digital baseband com-
ests concern the radio mobile communications systems, broadcasting sys- ponents and corresponding hardware/software implementations and digital
tems, coded modulations, waveform design, error correcting codes, MIMO, circuit design. His research activities target mainly digital communication
and iterative receivers. He is involved in several research projects related applications, in addition to other application domains, and more particularly
to broadcasting and satellite communications. Since 2007, he has been the design of flexible digital physical layer for future wireless communi-
active in the digital video broadcasting consortium where he had important cation standards and terminals. He has co-authored over 100 papers on
contributions. scientific journals and proceedings of international conferences. He serves
on the technical program committee for several international conferences.