Beruflich Dokumente
Kultur Dokumente
Hao Lin
Orange Labs, Cesson Sévigné, France
Abstract—Filter Bank Multi-Carrier (FBMC) with Offset modulation side [5], it however cannot relieve the pressure on
Quadrature Amplitude Modulation (OQAM) has been actively the demodulation. On the other hand, the modem complexity
investigated for decades. Now it is also a candidate for 5G air of FS-FBMC is more than 2K times higher than OFDM
interface waveform. Nevertheless, when it goes to standardization [4], where K is the prototype filter overlapping factor whose
consideration, people still prefer Orthogonal Frequency Division common value for FS-FBMC system is 4. The SP complexity
Multiplex (OFDM), in spite of its identified drawbacks for 5G.
is directly linked to compatibility. Here the compatibility
Such conservative attitude comes from the fact that people are not
sure about how costly it would be to adapt all the existing OFDM- does not mean that an FBMC receiver is able to decode the
based solutions to FBMC systems, in particular for Multiple- LTE signals, but rather means that FBMC should be able to
Input-Multiple-Output (MIMO) techniques. For FBMC, these reuse the existing LTE techniques, e.g. CE and MIMO pre-
can be interpreted as compatibility and complexity issues. In /decoding, in a straightforward manner. This requires that any
this paper, we intend to solve these two issues by introducing a technique adaptation should not call for additional re-design
new way of looking at FBMC, where we challenge the traditional effort. However, in the FBMC literature, if we refer to either
transmission fashion. Eventually, our study brings us to a more CE (e.g. [6], [7]) or equalization (e.g. [8], [9]) or MIMO design
concise and compatible format which we call it Filter-Bank (e.g. [10], [11]), technique adaptation is always needed and
OFDM. such design is not straightforward. Moreover, each tailored
technique always introduces additional SP complexity. Thus, a
I. I NTRODUCTION doubt arises: Is this worth trading the FBMC advantages with
Research activities on FBMC have been conducted for such additional design effort? Therefore, we think that it is
decades. Now it is also a candidate for 5G air interface necessary to revisit these two issues for FBMC.
waveform [1], [2]. The main benefits brought by FBMC in the To effectively solve these issues, we decided to identify the
5G respect are two lines: good spectrum containment and re- fundamental source problem and tried to break out from the
laxing synchronization condition [1]. The good spectrum con- conventional concept. Eventually, this investigation leads us to
tainment can enable flexible service offloading, e.g. between a new FBMC solution which is more concise and flexible. It
5G and white space bands, which is helpful for emergency can almost solve the mentioned issues and we call this new
communications [1]. Whilst, the capability of allowing relaxed solution Filter-Bank OFDM (FB-OFDM). We cannot claim yet
synchronism is of paramount important in future Machine that it is the optimum solution but, indeed, it points out a
Type Communications (MTC), not only allowing for signaling new direction for 5G air interface investigation. This paper
overhead reduction but also guaranteeing a stringent latency is organized as follows: In Sec. II, we revisit the FB theory
[2]. These advantages make FBMC an interesting alternatives and identify the source problem. In Sec. III, we present FB-
to Long Term Evolution (LTE) waveform, i.e., OFDM. How- OFDM concept and the solution in detail. In Sec. VI, we
ever, since the best effort Mobile Broad-Band (MBB), which discuss the efficiency of the FB-OFDM solution, in terms of
seeks for more throughput with advanced MIMO solutions its compatibility and complexity. Finally, some conclusions are
[3], will continue playing a major role in future 5G systems, drawn in Sec. V.
potential waveform candidate must prove that, besides its
advantages, its usage will not create severe negative impact II. BALIAN -L OW T HEOREM AND S OURCE P ROBLEM
on the MBB service. Unfortunately, the usage of FBMC in
MBB, compared to OFDM, might be less appealing due to For a MC system, denote F0 as the spacing between sub-
two issues: complexity and compatibility with LTE. carriers and T0 as the symbol duration. The modulated signal
can be written as a linear combination of a Gabor family,
The complexity can further be divided into two parts: the i.e. the triplet (f ,T0 ,F0 ), where f is the waveform pulse or
modem complexity and the Signal Processing (SP) complexity. prototype function. Then, there are three factors, namely spec-
The former refers to the modulation and demodulation for a tral efficiency, orthogonality and prototype function flexibility
Multi-Carrier (MC) system. The latter is related to the other are conditioned by the Balian-Low theorem (BLT) [12]. A
complementary components, e.g. Channel Estimation (CE), common understanding of BLT is that it is not possible to get a
equalization, MIMO pre-/decoding, etc., which are necessary modulation scheme holding at the same time the property of (i)
to establish a complete transceiver. The modem complexity is orthogonality in the complex field, (ii) a well-localized shape
implementation-dependent. In the FBMC literature, there are in frequency domain, and (iii) reaching maximum spectral
two implementation fashions, i.e. PolyPhase Network-based efficiency by keeping Nyquist rate transmission [13]. When
(PPN-FBMC) and Frequency Spreading based (FS-FBMC) [4]. OFDM satisfies (i and iii) it cannot obtain a pulse different
The modem complexity of PPN-FBMC w.r.t. OFDM is more from Rectangular. Thus, FBMC chose to relax the complex
than doubled. Although, wise algorithm can be used on the orthogonality to real field only in order to meet (ii and iii).
1078
IEEE ICC 2015 - Workshop on 5G & Beyond - Enabling Technologies and Applications
E
CM ×K Symbol CM ×2K−1 XM ×2K−1 yM K×1 M K-point sM K×1
Filtering Mapping
Extension IFFT
E
ĈM ×K Symbol ĈM ×2K−1 X̂M ×2K−1 ŷM K×1 M K-point
Filtering Demapping
Recovery FFT
Finally, to recover the transmitted QAM symbol, we naturally Algorithm 1 Construct C̄E
M ×2K−1 from CM ×K
take the dual operation to the extension process which is
equivalent to the following expression 1: Arbitrarily select M symbols from CM ×K and map them
to the central column of C̄EM ×2K−1 .
2: if K is an odd number then
ĈM ×K = ĈE T
M ×2K−1 EK×2K−1 . (13)
3: Take the remaining symbols from CM ×K in an arbitrary
order, and form them to K − 1 column vectors.
Now the remaining task is to find the filter coefficients
4: For each column vector, pair it with its conjugate,
hf1×2K−1 and the symbol extension algorithm. forming K − 1 paired-column vectors.
5: Map K−1 paired-column vectors to the remaining 2K−
2 columns of C̄E M ×2K−1 , on the paired positions of
B. Symbol extension algorithm
(14).
Interestingly, we observe that our system is indeed also a 6: end if
frequency spreading system. Thus, it should share the same 7: if K is an even number then
Nyquist conditions that were used for FS-FBMC [14, (5)], i.e. 8: Take the remaining M × (K − 2) symbols from CM ×K
in an arbitrary order, and form them to K − 2 column
f 2 vectors.
f (h0 ) = 1
hk = (14) 9: For each column vector, pair it with its conjugate,
(hfk )2 + (hfK−k )2 = 1, for k ∈ [1, K − 1].
forming K − 2 paired-column vectors.
10: Map K−2 paired-column vectors to the 2K−4 columns
This greatly simplifies our task as the only remaining item is to
of C̄EM ×2K−1 , on the paired positions of (14).
find the symbol extension matrix EK×2K−1 to satisfy a perfect
11: Take the last M symbols from CM ×K in an arbitrary
reconstruction, i.e. ĈM ×K = CM ×K . Omitting the derivation,
order, and form a paired-column vector by separating
we found that the symbol extension pattern is highly coherent
the real and imaginary parts
with (4) and (14). From (4) we see that the filter coefficients are
12: Map them to the remaining 2 columns of C̄E M ×2K−1 ,
symmetric w.r.t. its central coefficient. Moreover, from (14), it
multiplying with a scale factor.
can be seen that the coefficients on the right (or left) side of
13: end if
hf0 can form (K − 1) paired coefficients which are placed on
(K − 1) so-called paired positions. As a result, our designed
symbol extension pattern also forms a paired symbols which
are then mapped on the paired positions. Our found solution
E
following. For even K, C̄m,k is obtained by, for k ∈ [0, K
2 −2],
tells that the extension process (1) should be rewritten as E E ∗
√ ) = Cm,k
C̄m,k = (C̄m,K−2−k
CE = JM ×2K−1 ⊙ (CM ×K EK×2K−1 ) (15) E
C̄m,(K/2)−1 = 2ℜ C
M ×2K−1
| {z }
E E
√ m,(K/2)−1
C̄E
M ×2K−1 C̄m,k = C̄m,(K/2)−1+K = 2ℑ Cm,(K/2)−1
E
C̄m,K−1
= Cm,K/2
where JM ×2K−1 is a necessary permutation which can be
C̄ E E ∗
expressed as m,k+K = (C̄m,2K−2−k ) = Cm,k+K/2+1
(17)
E
While for odd K, C̄m,k is obtained by, for k ∈ [0, K−1
2 − 1],
JM ×2K−1 = [jm
1×2K−1 ], for m ∈ [0, M − 1] (16)
E E ∗
where jm is a row vector with all elements the same C̄m,k = (C̄m,K−2−k ) = Cm,k
√
1×2K−1
E E
value of ( −1)m except for the center element whose value C̄m,k = C̄ = Cm,(K−1)/2
E m,K−1
E E
is 1. The construction of the extended symbol C̄m,k , for m ∈ C̄m,k+K = (C̄m,2K−2−k )∗ = Cm,k+(K−1)/2+1
[0, M − 1] and k ∈ [0, 2K − 1], should follow the following (18)
algorithm.
Note that the symbols to be selected from CM ×K to
This algorithm can be mathematically expressed in the C̄E
M ×2K−1 can be arbitrarily chosen. Therefore, (17) and (18)
1079
IEEE ICC 2015 - Workshop on 5G & Beyond - Enabling Technologies and Applications
are not the unique expression1 . On the other hand, (17, 18) −1
10
can be alternatively translated in a matrix form. Taking two OFDM
examples of K = 4 and K = 3, EK×2K−1 is expressed as FB−OFDM
1 √0 1∗ 0 0 √0 0
0 2ℜ 0 0 0 2ℑ 0
EK×2K−1 |(K=4) =
0 0 0 1 0 0 0
0 0 0 0 1 0 1∗
BER
−2
10
and " # 4QAM
1 1∗ 0 0 0 16QAM
64QAM
EK×2K−1 |(K=3) = 0 0 1 0 0
0 0 0 1 1∗
where the element 1∗ stands for conjugation-taken operation,
i.e. Cm,k 1∗ = Cm,k
∗
; while ℜ and ℑ stand for real/imaginary-
√
taken operation, i.e. Cm,k ℜ = ℜ {Cm,k }. The factor of 2 is −3
10
−5 0 5 10 15 20
used to keep constant power. SNR (dB)
At the receiver, the symbol recovery process (13) follows a Fig. 2: BER comparison in AWGN channel with different
dual operation to symbol extension process (17, 18), yielding constellation orders: FB-OFDM vs. OFDM with same spectral
!
ĈM ×K = J∗M ×2K−1 ⊙ ĈE ETK×2K−1 ℑ7→jℜ efficiency and transmit power.
M ×2K−1
| {z }
ČM ×2K−1
where ℑ 7→ jℜ indicates
√ that if there is ℑ, we use ℜ instead; IV. D ISCUSSIONS
then multiply with −1. We can alternatively use linear
system representation to express symbol recovery process. For In this section, we discuss the efficiency of FB-OFDM
even K, Ĉm,k can be recovered by, for k ∈ [0, (K/2) − 2], and how to use it in practice. Due to the page limit, not
all our results can be shown in this paper. Only illustrative
∗
Ĉm,k = Čm,k +Čm,K−2−k hints are provided. Further elaborations will be given in future
√ n o n o
publications.
Ĉm, K −1 = 2 ℜ Čm, K −1 + jℜ Čm, K −1+K
2 2 2
Ĉ
m,K/2 = Čm,K−1
A. User isolation
Ĉm,k+K/2+1 = Čm,k+K + (Čm,2K−2−k )∗
(19) In a practical MC system, not all the subcarriers are used.
While for odd K, Ĉm,k can be recovered by, for k ∈ [0, K−1
2 − For instance in LTE (5 MHz) system, only 300 subcarriers are
1], used for transmitting payload over the overall 512 subcarriers.
∗ Moreover, these 300 subcarriers are further separated into 25
Ĉm,k = Čm,k + Čm,K−2−k chunks (each contains 12 subcarriers), so that the chunks can
Ĉ = Čm,K−1 be allocated for multiple users. For FB-OFDM, one drawback
m,(K−1)/2
Ĉm,k+(K−1)/2+1 = Čm,k+K + (Čm,2K−2−k )∗ is that if two adjacent chunks are allocated to different users,
Naturally, if a block interleaver is applied before the symbol their combinational symbols will overlap at the extended
extension process, a de-interleaver should equivalently be symbol layer. This causes inter-user interference for channel
applied after the symbol recovery process. It is worth noting estimation, and will be more cumbersome for MIMO pre-
that when K = 1, the FB-OFDM automatically generates the /decoding. To solve this issue, special care must be taken at the
classical OFDM signal. edge of the allocated bands in order to avoid such overlapping.
One simple solution is to insert null symbols to the first
C. Power increase due to symbol extension (or the last) subcarrier of the extended symbols C̄E N ×2K−1 ,
where N is the number of allocated subcarriers. For example,
It is obvious that the transmit power is increased due to if we insert the null symbols to the first subcarrier of the
the symbol extension. Thus, to keep the same transmit power E
extended symbols, we force C̄0,0···K−1 = 0. Alternatively, we
as in OFDM p case, the FB-OFDM signal can be normalized by can insert the null symbols to the last subcarrier, such that
the factor of (2K − 1)/K. Nevertheless, this does not affect E
C̄N −1,K···2K−1 = 0. The later option, according to (17,18),
the FB-OFDM noise sensitivity due to the addition in (19,20). corresponds to the modification of original symbols as
In Fig. 2 we show the Bit Error Rate (BER) comparison
between FB-OFDM and OFDM under an AWGN channel CN −1,K/2−1 = ℜ CN −1,K/2−1
For even K: (20)
with different constellation orders. Our simulation parameters CN −1,K/2+1···K−1 = 0.
are: (M = 512, K = 4, no coding) and we assume the
same throughput and transmit power for both systems. The For odd K: CN,(K−1)/2+1···K−1 = 0. (21)
results confirm that FB-OFDM has the same noise sensitivity With this solution, the combinational symbols of a given user
as OFDM. are completely isolated from the other user who accesses to
1 One can imagine to apply a random block interleaver on C
M ×K before
the adjacent band. However, transmitting null symbols will
the symbol extension process. naturally reduce the spectral efficiency.
1080
IEEE ICC 2015 - Workshop on 5G & Beyond - Enabling Technologies and Applications
1081
IEEE ICC 2015 - Workshop on 5G & Beyond - Enabling Technologies and Applications
As a reference, the OFDM modem complexity, for processing lieved two issues that we identified for traditional FBMC
CN ×K , symbols is assumed to be scheme, namely complexity issue and compatibility issue with
CoM(Tx) = CoM(Rx) = (M K/2) log2 (M ) OFDM. The FB-OFDM scheme indeed can largely reduce
(26) the additional complexity w.r.t. OFDM (i.e. less than 30%).
CoA(Tx) = CoA(Rx) = M K log2 (M ).
Moreover, this scheme is completely compatible with OFDM.
In the following comparison, we take LTE-DL 20 MHz Actually OFDM can be seen as one special case of FB-
configuration as our baseline, where M = 2048 and the OFDM. Therefore, all the complementary techniques, such as
overall modulated subcarriers are divided into 100 chunks. channel estimation, equalization, MIMO pre-/decoding can be
In what follows, We separately compare the Transmitter (Tx) straightforwardly re-used for FB-OFDM without the needs of
complexity and Receiver (Rx) complexity because the Tx additional re-design effort.
complexity is consumed by Base Station (BS) where full
chunks should be considered. While the Rx complexity is R EFERENCES
on User Equipment (UE) side, i.e. only allocated chunks are
[1] METIS Delivarable, “D2.4: Proposed solutions for new radio access”,
concerned. For UE Rx complexity, we consider two cases: 1) March 2015.
Full-Chunk allocation; 2) 1 Chunk allocation. The comparison [2] G. Wunder et al.,“5GNOW: non-orthogonal, asynchronous waveforms for
results are given in Tab. I with K = 4. future mobile applications”. Communications Magazine, IEEE , vol.52,
no.2, pp.97-105, February 2014.
[3] METIS Delivarable, “D6.5: Report on simulation results and evalua-
TABLE I: Modem complexity increase tions”, March 2015.
BS (Full-Chunk) UE (Full-Chunk) UE (1 Chunk) [4] M. Bellanger, “FBMC: physical layer: a primer”, ICT-PHYDYAS Project,
CoM 26% 26% 18.3% pp. 131, 2010
CoA 22% 23.5% 18.2% [5] Y. Dandach and P. Siohan, “FBMC/OQAM modulators with half com-
plexity,” Proceedings Globecom’11 (Houston, USA), Dec. 2011.
The complexity increase on the BS side is 26% and on [6] J.-P. Javaudin, D. Lacroix, A. Rouxel,“Pilot-aided channel estimation for
OFDM/OQAM”, Vehicular Technology Conference (VTC) 2003-Spring,
UE side the increase is between 18% and 26%, which is April 2003.
significantly lower than that of PPN-FBMC (more than ×2) [7] Z. Zhao, N. Vucic, M. Schellmann, “A simplified scattered pilot for
and FS-FBMC (more than ×8). Note that if we compare the FBMC/OQAM in highly frequency selective channels”, International
complexity of a complete receiver processing, this complexity Symposium on Wireless Communications Systems (ISWCS), Aug. 2014.
increase ratio for FB-OFDM will significantly be reduced. [8] T. Ihalainen, T. Hidalgo Stitz, M. Rinne, and M. Renfors, “Channel
Equalization in Filter Bank Based Multicarrier Modulation for Wireless
D. Key difference between FB-OFDM and FS-/PPN-FBMC Communications”, EURASIP Journal on Advances in Signal Processing,
2007(49389), 2007.
There are two key differences between FB-OFDM and [9] H. Lin, P. Siohan, P. Tanguy, J.-P. Javaudin, “An Analysis of the EIC
FS/PPN-FBMC. The first difference is that FS/PPN-FBMC Method for OFDM/OQAM Systems”, Journal of Communications, Vol
systematically separates real and imaginary parts of the QAM 4, No 1 (2009), 52-60, Feb 2009.
symbols and processes them separately. While FB-OFDM [10] R. Zakaria and D. le Ruyet, “On interference cancellation in Alamouti
scheme does not apply such separation in a systematic manner; coding scheme for filter bank based multicarrier systems” International
Symposium on Wireless Communications Systems (ISWCS), June. 2013.
this thus greatly reduces the modem complexity. The second
[11] M. Caus and A. Perez-Neira, “Multi-Stream Transmission for Highly
difference is that, for FS/PPN-FBMC systems, after the IFFT Frequency Selective Channels in MIMO-FBMC/OQAM Systems,” Sig-
transform, there is always an overlapping, in the time domain, nal Processing, IEEE Transactions on, vol. 62, no. 4, pp. 786-796, Feb
between at least two consecutive FBMC symbols. This over- 2014.
lapping indeed makes FBMC weakly compatible with OFDM- [12] H.G. Feichtinger and T. Strohmer, Gabor analysis and algorithms,
based techniques. While FB-OFDM scheme does not have Birkhäuser, 1998.
such overlapping in the time domain. This actually solves the [13] H. Bölcskei, “Orthogonal Frequency Division Multiplexing Based on
compatibility issue. Moreover, FB-OFDM has an additional Offset QAM”, Advances in Gabor Analysis Applied and Numerical
Harmonic Analysis 2003, pp. 321-352, 2003.
advantage, i.e. diversity gain. Due to the symbol extension, it
[14] M. Bellanger.“specification and design of a prototype filter for filter
introduces symbol diversity, i.e. the original QAM symbol and bank based multicarrier transmission”.ICCASP, May 2001.
its conjugate part. Therefore, the channel decoding can indeed [15] H.M. Park, J.S. Han and H.J. Choi, “Enhanced Detection Method for
take advantage of this property to obtain a diversity gain. 3GPP LTE QO-SFBC System in Frequency Selective Rayleigh Fading
Channel Environment,”ICUIMC ’10, NY, USA, 2010.
V. C ONCLUSIONS
In this paper, we presented a new modulation format
named FB-OFDM. This modulation scheme has greatly re-
1082