INL/JOU-14-33556

Accepted Manuscript

OFDM Inspired
Waveforms for 5G

Behrouz Farhang-Boroujeny
Hussein Moradi

May 2016

This is an accepted manuscript of a paper intended for publication in a

journal. This document was prepared as an account of work sponsored by
an agency of the United States Government. Neither the United States
Government nor any agency thereof, or any of their employees, makes any
warranty, expressed or implied, or assumes any legal liability or
responsibility for any third party’s use, or the results of such use, of any
information, apparatus, product or process disclosed in this report, or
represents that its use by such third party would not infringe privately owned
rights. The views expressed in this paper are not necessarily those of the
United States Government or the sponsoring agency.

Prepared for the U.S. Department of Energy

Office of Nuclear Energy
Under DOE Idaho Operations Office
Contract DE-AC07-05ID14517
 1

OFDM Inspired Waveforms for 5G

Behrouz Farhang-Boroujeny† and Hussein Moradi‡
† ECE Department, University of Utah and ‡ Idaho National Laboratory

Abstract
As the standardization activities are being formed to lay the foundation of 5G wireless networks, there is a
common consensus on the need to replace the celebrated OFDM by a more effective air interface that better serves
the challenging needs of 5G. To this end, in the recent past, a number of new waveforms have been introduced in the
literature. Interestingly, and at the same time not surprising, these methods share a common fundamental principle
with OFDM: each data packet is made up of a number of tones that are modulated by information symbols. In this
tutorial article, we build a common framework based on the said OFDM principle and derive these new waveforms
from this point of view. This derivation provides a new prospective that facilitates straightforward understanding of
channel equalization and the application of these new waveforms to MIMO channels. It also facilitates derivation
of new structures for more efficient synthesis/analysis of these waveforms than those that have been reported in the
literature.

I. I NTRODUCTION

In the past, orthogonal frequency division multiplexing (OFDM) has enjoyed its dominance as the most popular

signaling method in broadband wired, [1], [2], and wireless, [3], [4], channels. OFDM has been adopted in the broad

class of DSL standards, as well as in the majority of wireless standards, e.g., variations of IEEE 802.11 and IEEE

802.16, 3GPP-LTE, and LTE-Advanced. OFDM is known to be a perfect choice for point-to-point and downlink

communications. It offers a minimum complexity and achieves good bandwidth efficiency. The low complexity of

OFDM is a consequence of the fact that each OFDM symbol is synthesized as a summation of a number of pure

tones that are modulated by a set of quadrature amplitude modulated (QAM) data symbols. Moreover, channel

distortion is taken care of trivially by adding a cyclic prefix (CP) to each OFDM symbol.

It has been noted that OFDM has to face many challenges when considered for adoption in more complex

networks. For instance, the use of OFDM in the uplink of multiuser networks, known as OFDMA (orthogonal

frequency division multiple access), requires full synchronization of the users’ signals at the base station input. Such

synchronization has been found very difficult to establish, especially in mobile environments where Doppler shifts

of different users are hard to predict/track. Morelli et al. [5] have noted that “carrier and timing synchronization

represents the most challenging task in OFDMA systems.” To combat the problem, in LTE and LTE-Advanced

resource intensive close-loop procedures have been included to achieve the tight synchronization requirements. On
This manuscript has been authored by Battelle Energy Alliance, LLC under Contract No. DE-AC07-05ID14517 with the U.S. Department
of Energy. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United
States Government retains a nonexclusive, paid-up, irrevocable, world-wide license to publish or reproduce the published form of this
manuscript, or allow others to do so, for United States Government purposes. STI Number: INL/JOU-14-33556
 2

the other hand, some researchers have relaxed on the need for a close to perfect carrier synchronization among

users, and have proposed multiuser interference cancellation methods, [6]–[9]. These methods are generally very

complex to implement. Their implementation increases the receiver complexity by orders of magnitude [10]. These

solutions are either too slow and/or too resource intensive to be of interest in may applications of 5G where low

cost devices need to communicate.

Another limitation of OFDM appears when attempt is made to transmit over a set of non-contiguous frequency

bands. The poor response of the subcarrier filters in IFFT/FFT filter banks of OFDM introduces significant out of

band egress noise to other users and also picks up significant ingress noise from them. The same problem appears

if one attempts to adopt OFDM for filling in the spectrum holes in cognitive radios. Methods of reducing OFDM

spectral leakage proven to be very limited in performance and may add significant complexity to the transmitter. For

instance, the side lobe suppression techniques, like those proposed in [11]–[13], can achieve an out of band emission

suppression of only 5 to 10 dB, while they may substantially increase the transmitter complexity. Furthermore, they

add significant complexity to the transmitter and they will incur some loss in bandwidth efficiency.

Filter bank multicarrier (FBMC) is an alternative transmission method that resolves the above problems by using

high quality filters that avoid both ingress and egress noise. Also, because of the very low out of band emission of

the subcarrier filters, application of FBMC in the uplink of multiuser networks is trivial [14], [15]. Accordingly,

FBMC can be deployed without synchronization of mobile user nodes. In the application of cognitive radios, the

filter bank that is used for multicarrier data transmission can also be used for spectrum sensing [16]–[21]. On the

other hand, compared to OFDM, FBMC falls short in handing multiple-input multiple-output (MIMO) channels,

although a few solutions to adopt FBMC in MIMO channels have been reported in the literature, e.g., see [22]–

[24], and research in this domain still continues. Moreover, the ramp-up and ramp-down of an FBMC signal at the

beginning and the end of each packet, respectively, can lead to a significant loss in bandwidth efficiency when the

data packets are short. Unfortunately, this can be problematic, at least, in part of 5G applications where machine-to-

machine (M2M) communications and internet of things (IoT) involve exchange of very short messages and possibly

with a very short latency.

Following the above arguments, it is obvious that none of the existing waveforms are sufficient to serve all

aspects of the 5G. Noting this, in recent past, a number of attempts have been made to propose waveforms that

resolve the problems of both OFDM and FBMC. Generalized frequency division multiplexing (GFDM) that was first

introduced in 2009 by Fettweis and his co-workers, [25], is an improvement to OFDM in which filtering is imposed
 3

on each subcarrier band to minimize the overlapping among subcarriers, thus facilitate multiuser application without

worrying about accurate synchronization of the users. Moreover, the waveform design in GFDM is such that one

may choose to use one CP for many symbols or only one CP for the whole packet, assuming that channel variation

over each packet remains negligible. This, clearly, improves the bandwidth efficiency of the packet significantly. The

down side of these changes is that adjacent subcarriers suffer from some level of interference whose compensation

adds some complexity to the receiver and may incur some loss in performance. The term non-orthogonal waveform

is often used to refer to this property of GFDM, [26]. Since its invention in 2009, a broad set of papers have

explored different aspects of GFDM, e.g., [27]–[32]. A summary of these works have been recently collected and

reported in [26].

More recently, it has been noted that the concept based on which GFDM waveform is built can be extended

to FBMC, [33]–[36]. The waveform that results from this extension may be called circular FBMC (C-FBMC),

because of the reasons that become obvious later. Interestingly this extension preserves the orthogonality of the

subcarrier symbols, hence, no complicated intercarrier interference (ICI) cancellation processing would be necessary

and, consequently, any loss in performance is avoided. Another interesting outcome of C-FBMC waveform is that

its extension to MIMO channels is found straightforward, in the same way as GFDM. Although, there are multiple

publications in the literature that talk about extension of GFDM/C-FBMC to MIMO channels, e.g., see [26], the

trivial solution that is presented in this article is from one our recent publications [37].

Another relevant work that should be acknowledged here is the so called cyclic block filtered multitone (CB-

FMT), [38]–[40]. This method also builds its waveform following the same principle as GFDM and C-FBMC.

However, unlike GFDM and C-FBMC which allow some overlap of adjacent subcarrier bands, hence achieve full

bandwidth efficiency, CB-FMT is constructed such that adjacent subcarrier bands do not overlap. This, obviously,

results in some bandwidth efficiency loss. To keep the content of this article concise, no further details of CB-FMT

is pursued here. Lin and Siohan, [36], have already provided a detailed comparison of CB-FMT and C-FBMC

methods.

Another contender for 5G that has recently been proposed by a number of researchers from the Alcatel-Lucent

Bell Laboratories is universal filtered multicarrier (UFMC), [41]–[45]. UFMC may be seen as a modified form

of OFDM, where a filtering operation is applied to each group of consecutive subcarriers (e.g., a given allocation

of a single user) in order to reduce the out-of-band (OOB) interference. This, when the timing synchronization

is relaxed, minimizes the potential of multiple user interference. Moreover, the CP is removed. Instead, the guard
 4

interval that is associated with CP in OFDM is consumed by the transient of the filters in UFMC. In the absence

of channel, UFMC holds the orthogonality of the subcarriers of each user perfectly, and approximately among

different users. However, as the time spreading of underlying channels increases, the orthogonality assumption

among subcarriers of each user as well as among different users become less accurate. In light of this property,

UFMC may also be categorized as a non-orthogonal multicarrier technique. However, the level of non-orthogonality

in UFMC is significantly lower than GFDM and it varies with the channel. As a result, one may find that the name

quasi-orthogonal better suits the particular features of UFMC.

We note that the goal of this article is not to critically compare the various waveforms that are being under study

for 5G. Such comparison requires a broad set of theoretical studies and numerical simulations that fall beyond

the scope of this article. Our presentation is to provide a review of the candidate waveforms and, in particular,

present GFDM and C-FBMC from a point of view that reveals their close similarity to OFDM. We note that both

waveforms are synthesize by adding a finite number of tones, exactly in the same way that OFDM waveform is

built. The difference is that while in OFDM each symbol modulates a single tone, in GFDM and C-FBMC each

symbol modulates multiple tones. And it is the combination of the respective tones that (when done properly)

confines each subcarrier to the intended bandwidth. The point of view that is presented in this article also reveals a

close similarity of GFDM and C-FBMC and the fact that any advancement in one may be extended to the other. In

addition, we review some interesting features of UFMC which facilitate its comparison with GFDM and C-FBMC.

Moreover, our presentation allows us to explain the extension of GFDM, C-FBMC and UFMC to MIMO channels

straightforwardly.

This paper is organized as follows. In Section II, we highlight the fundamental principle based on which OFDM

waveform is constructed. Next, in Section III, we introduce a less understood, but relevant concept: frequency

spreading design of digital filters. We combine the results of Sections II and III to introduce a new construction

for GFDM waveform in Section IV. The transmitter and receiver implementations of GFDM are also discussed in

this section. Moreover, we show the same principle based on which MIMO communication is formed in OFDM is

applicable to GFDM as well. In Section V, we show the approach that was used in Section IV is readily applicable

to the construction of the respective FBMC waveform, viz., C-FBMC. UFMC is reviewed in Section VI. We find

that due to the removal of CP, UFMC deviates from the said OFDM principle. That is, from the receiver detector

point of view, data symbols may no longer be carried over pure tones. Nevertheless, data symbols separation and

MIMO processing, as in OFDM, is possible by making use of an FFT with twice the size of the one in OFDM.
 5

The concluding remarks and suggestions for further works are presented in Section VIII.

Notations: Our presentation in this article is through a mix of continuous time and discrete time signals. We use

(t) to denote the continuous time variable and [n] to denote discrete time index. Subcarrier index is denoted by

subscript k. Discrete time index, n, when has to be added to a continuous time function is put as a subscript, e.g.,

xn (t). Bold lower case is used for column vectors and bold upper case for matrices. All vectors are in column

form. The vector and matrix transpose and Hermitian are indicated by the superscripts ‘T’ and ‘H’, respectively.

We use F N to denote discrete Fourier transform (DFT) matrix of size N . We also assume that F N is normalized,
−1
such that F N F H H
N = IN , where IN denotes the identity matrix of size N . Hence, F N and F N are equal and both

denote the inverse DFT (IDFT) matrix of size N , and these may be used interchangeably. The terms FFT and IFFT

refer to the fast implementations of DFT and IDFT, respectively.

II. T HE OFDM P RINCIPLE

Each OFDM symbol carries N QAM symbols sk [n], k = 0, 1, · · · , N − 1, and is constructed as

N
X −1
2kπ
xn (t) = sk [n]ej T
t
. (1)
k=0

Related to (1), there are a few points that should be noted:

1) xn (t) is a summation of N tones, weighted by the data symbols sk [n].

2) These tones are spaced at F = 1/T and are located at the frequencies f = 0, 1/T , 2/T , · · · , (N − 1)/T .

3) xn (t) is periodic with a period of T .

4) The Fourier series coefficients of xn (t) are the data symbols sk [n], k = 0, 1, · · · , N − 1, and the construction

of xn (t) may be viewed as an inverse Fourier series.

5) If xn (t) is passed through a channel with the transfer function H(f ), the channel output, excluding its transient

response, is obtained as
N −1  
X k 2kπ
yn (t) = H sk [n]ej T t . (2)
T
k=0

6) Obviously, yn (t) is also periodic with a period of T .

7) The Fourier series coefficients of yn (t) are H Tk sk [n], k = 0, 1, · · · , N − 1. Hence, the transmitted data

symbols sk [n] can be extracted from samples of yn (t), by applying a Fourier series analysis and equalizing

the results by the inverse of channel gains at the respective frequencies.

In practice, where a digital circuitry or a software radio is used for implementation, (1) is implemented through

an inverse discrete Fourier transform, or equivalently and conveniently, through an IFFT. IFFT output delivers only
 6

s0 [0] s0 [1] s0 [2] s0 [M − 1]

s1 [0] s1 [1] s1 [2] s1 [M − 1]
s2 [0] s2 [1] s2 [2]

cyclic prefix

cyclic prefix

cyclic prefix

cyclic prefix
s2 [M − 1]

.. .. .. ··· ..
. . . .
sN −1 [0] sN −1 [1] sN −1 [2] sN −1 [M − 1]

Fig. 1. An OFDM packet.

one cycle of the sampled version of xn (t); say xn [m], for m = 0, 1, · · · , N − 1. Similarly, the data extraction

at the receiver is performed by applying an FFT to the samples of a single period of yn (t), say, yn [m], for

m = 0, 1, · · · , N − 1, and equalizing the results by multiplication to the inverse of the channel gains, as noted in

item 7), above.

Clearly, it will be resource inefficient to transmit xn (t) for any period of time beyond the minimum duration that

one needs for correct extraction of the data symbols sk [n] at the receiver. This minimum duration is T (one period

of xn (t)) plus the duration of the channel impulse response, Tch . The latter is needed to absorb the channel transient

response. To serve this purpose, in digital implementation of the transmitter, a number of samples, equivalent to

Tcp ≥ Tch , from the end of the IFFT output are prefixed to its beginning. The prefixed samples, because of obvious

reasons, are called cyclic prefix (CP). At the receiver, the FFT is applied after removing the CP samples.

From the above discussion, it follows that each OFDM symbol has a duration of T + Tcp , and a data packet

consisting of M OFDM symbols may be expressed, in continuous time, as

M
X −1
x(t) = xn (t − Tcp − n(T + Tcp ))
n=0
M
X −1
2π
(t−Tcp −n(T +Tcp ))
= sk [n]ej T . (3)
n=0

To facilitate our discussions in the subsequent parts of this paper, we have presented in Fig. 1 the structure of

an OFDM packet. As follows from the above discussion, each OFDM symbol (consisting of N QAM symbols)

needs to be cyclic prefixed.

 7

III. F REQUENCY S PREADING D ESIGN OF D IGITAL F ILTERS

Matrin, [46], suggested a method of designing a class of finite impulse response (FIR) square-root Nyquist (N )

filters whose impulse response is expressed as

 P −1

c0 + 2 Ml=1 cl cos
2πln
, 0≤n≤L−1
p[n] = L (4)
0, otherwise

where L = M N is the filter length, N is the spacing between zero-crossings of Nyquist pulse-shape q[n] =

p[n] ⋆ p[−n], and the parameter M is referred to as overlapping factor. Mirabbasi and Martin, [47], later, presented

an improved design over that of [46].

We also note that (5) may be rearranged as

L−1
X 2πln
p[n] = c̃l ej L (5)
l=0

where 
 cl , 0≤l ≤M −1
c̃l = 0, M ≤l ≤L−M (6)

cL−l , L − M + 1 ≤ l ≤ L − 1.
This observation implies that in the frequency domain, over a grid of L frequency bins, p[n] is characterized by

the 2M − 1 non-zero coefficients c̃l or, equivalently, 2M − 1 complex frequency bins/tones. According to [46] and

[47], the choices of M ≥ 4 usually lead to very good designs, with stopband attenuations of over 70 dB.

Next consider the case where p[n] is used as the prototype filter to realize a filter bank consisting of N subbands

centered at the normalized frequencies f0 = 0, f1 = M

L = 1
N, f2 = 2M
L = 2
N, · · · , fN −1 (N −1)M
L = N −1
N . The k-th

modulated filter of this filter bank is characterized by the impulse response

pk [n] = p[n]ej2πfk
L−1
X 2π(M k+l)n
= c̃l ej L

l=0
L−1
X 2πln
= c̃(l−M k)L ej L . (7)
l=0

where (·)L means modulo L. The result in (7) shows that each of the modulated filters, in the filter bank, also, is

characterized by 2M − 1 non-zero coefficients/tones. These coefficients are obtained through circular shift of the

coefficients [c̃0 , c̃1 , · · · , c̃L−1 ] of the prototype filter p[n] in (5). We will make use of this property of the prototype

filter p[n] in the rest of this paper to propose effective implementations for the transmitter and receiver of GFDM

and C-FBMC.
 8

s0 [0] s0 [1] s0 [2] s0 [M − 1]

s1 [0] s1 [1] s1 [2] s1 [M − 1]
s2 [0] s2 [1] s2 [2]

cyclic prefix
s2 [M − 1]

.. .. .. ··· ..
. . . .
sN −1 [0] sN −1 [1] sN −1 [2] sN −1 [M − 1]

Fig. 2. A GFDM block/packet.

Relevant to our discussions in the rest of this article, it follows from (7) that the DFT coefficients of the time

sequence pk [n] (equivalently, the corresponding subband frequency respond) are all zero, except those that lie within

its passband.

IV. GFDM
A. Derivation

As opposed to OFDM, packet construction in GFDM is such that only one CP is needed to take care of the

channel transient response. Fig. 2 presents the structure of a GFDM packet, as given in [25]–[32]. For the reasons

that are explained below, we refer to the collection in Fig. 2 as a GFDM block, and leave the term ‘packet’ for

a collection of blocks. The data symbols are spread across time and frequency, as in OFDM. However, the data

stream in each subcarrier is controlled through a filter which confines its frequency response to a limited bandwidth.

Similar to OFDM, the data symbols (the columns of a GFDM block) are transmitted at an interval T and subcarrier

spacing is set equal to F = 1/T .

As will be shown later, the detection of data symbols that are carried within each GFDM block is only possible

after receiving the complete set of samples of the block. Hence, when dealing with long packets of data, it may

be unreasonable to put the whole packet in one block, as this will introduce a long latency in the detection of the

transmitted information at the receiver. Moreover, long blocks can be affected by time-variation of the channel and

lead to a significant degradation of the receiver performance. To avoid these problems, a large packet of data should

be divided into a number of segments, each of which will be synthesized in one GFDM block. Hence, a complete

packet in GFDM may be a concatenation of multiple blocks, as multiple symbols in OFDM. Short packets, like

those that may be encountered in M2M communications, may consist of only one block.

Before we begin our derivation of GFDM, the reader should be reminded that the derivation that follows builds on

the frequency spreading design of digital filters that was presented above. This derivation is different from what has
 9

appeared in [25]–[32] where GFDM was derived based on circular convolutions. The use of circular convolutions

instead of linear convolutions was to construct a periodic signal and, thus, to allow the addition of CP as dictated

by the OFDM principle. Our derivation, [37], also leads to the same periodic signal, however, at the same time,

reveals some interesting properties of the GFDM waveform which otherwise were hard to see.

From the discussion presented above, for OFDM, one may deduce that to allow the use of a single CP per

GFDM block, the modulated signal that carries all the data symbols in a GFDM block should be a single cycle of a

periodic signal with the length of M T . This is indeed the case in GFDM and such a periodic signal is constructed,

following the OFDM signal synthesis, according to the formulas that follow. The contribution from the n-th data

column in the GFDM block/packet of Fig. 2 is synthesized as

N
X −1 L−1
X 2πl
xn (t) = sk [n]c̃(l−M k)L ej M T (t−nT ) . (8)
k=0 l=0

Note that this synthesizes the data symbols s0 [n] through sN −1 [n], over the N subcarrier bands, by following the

frequency spreading method of the previous section. We also note that xn (t), effectively, is the sum of L = M N

tones at the frequencies 0, 1/M T , 2/M T , · · · . Since all these tones may be considered as periodic with a period

of M T , xn (t) is also periodic with the same period.

The complete waveform carrying all the data symbols in the GFDM block of Fig. 2 is obtained by summing up

the result of (8) over 0 ≤ n ≤ M − 1. That is

M
X −1
x(t) = xn (t). (9)
n=0

Obviously, since the components xn (t) are periodic with a period M T , x(t) is also periodic and has the same

period. The GFDM block is thus obtained by taking a segment of x(t), of (9), over the interval −Tcp ≤ t ≤ M T .

Following the above concept, synthesizing a block of GFDM in discrete time can be performed according to

the schematic diagram shown in Fig. 3. Here, each data symbol sk [n] is fed into an IFFT block through a number

of tones which construct the corresponding subcarrier filter according to the method of frequency spreading (the

frequency spreading coefficients c̃l ). The frequency spreading coefficients, in Fig. 3, are signified by the filter

responses that indicate the frequency responses of different subcarrier bands. The “circular shift and accumulate”

block adds up the results of IFFT for each set of data symbols, after applying a circular shift. The circular shift

is to take care of the time delay between the successive data columns, as indicated in the block/packet format of

Fig. 2. The delay associated with the symbol set at the time index n in continuous time is nT . In discrete time, this

corresponds to nN sample delay/circular shift. Note that this implies there are N samples of x(t) is each symbol
 10

s0 [n] s1 [n] s2 [n] sN −1 [n]

···
IFFT (size M N )

···
circular shift and accumulate

··· x[n]
add CP and serialize to output

Fig. 3. An implementation of GFDM transmitter.

period T . Note that here delay and circular shift are equivalent because of the periodic form of the synthesized

signal, x(t).

B. Transmitter Implementation

Direct implementation of GFDM transmitter according to the schematic diagram of Fig. 3 requires a total of

M IFFT operations of size M N , plus additional operations prior and after IFFT. This complexity can be reduced

significantly, by some rearrangement of the operations as explained next. The detailed formulations that are presented

below were originally developed and presented by the authors in [37].

The equation that relates the input data symbols sk [n] and the IFFT output, denoted by the length M N column

vector x[n], may be expressed as

x[n] = F −1
M N Cse [n] (10)

where se [n] = [s0 [n] 0 s1 [n] 0 · · · sN −1 [n] 0]T , 0 is a row of M − 1 zeros, and
 
c0 c1 · · · cM −1
 c1 c 0 · · · cM −2 
 
 .. .. .. .. 
 .
 . . . 

 cM −1 cM −2 ··· 0 
 
C= 0 cM −1 ··· 0 

.
 .. .. .. .. 
 .
 . . . 
 cM −1 0 · · · cM −2 
 
 .. .. .. .. 
 . . . . 
c1 c2 ··· c0
 11

We note that C is a circular matrix of size M N × M N .

Direct implementation (10) is performed in two steps:

1) The circular convolution of the first column of C and the vector se [n] is performed to obtain Cse [n].

2) An IFFT of size M N is applied to the result of step 1) to obtain x[n].

The complexity of this procedure, which should be repeated M times for each packet, is dominantly determined

by Step 2); an IFFT of size M N .

Alternatively, x[n] may be calculated by arranging (10) as

 −1 
x[n] = F −1 −1
M N F M N (F M N c1 ) ⊙ (F M N se [n]) (11)
| {z }
Cse [n]

where c1 is the first column of C, ⊙ denotes point-wise multiplication, and the circular convolution of the vectors

c1 and se [n] is performed through point-wise multiplication of their respective IDFTs and then applying a DFT

to the result. Here, we have chosen to use IDFT domain instead of the common approach of using DFT domain,

because this, as explained next, gets us to a low complexity implementation of the transmitter.

Obviously, (11) reduces to

x[n] = p ⊙ (F −1
M N se [n]) (12)

where p = F −1
M N c1 is the vector of prototype filter coefficients of GFDM, in the time domain. We note that the

computational complexity of (12) is still dominantly determined by the complexity of the operation F −1
M N se [n],

i.e., computation of IFFT of se [n]. However, since se [n] is the M -fold expanded version of the length N column

vector s[n] = [s0 [n] s1 [n] · · · sN −1 [n]]T , F −1 −1

M N se [n] can be obtained by M repetitions of F N s[n], in a column.

Hence, following (12), x[n] can be calculated through an IFFT of size N .

Next, we note that the final content of the circular shift and accumulate block in Fig. 3, upon including all the

data symbols of the GFDM block, will be the vector

M
X −1
xgfdm = circshift(x[n], nN ) (13)
n=0

where circshift(·) means downward circular shift. After adding the CP to xgfdm , we have a complete GFDM block

to transmit.

The results presented in (12) and (13) have the following interpretation. According to (12), each set of data

symbols modulate the set of carrier tones, added together, and the result is shaped by a well designed window

defined by the prototype filter p. The results for different choices of time index n are then (circularly) shifted and

added together to obtain the vector xgfdm .

 12

h
s[n] IFFT (N) circular shift add CP x[n]
and ⊙ and and
expand accumulate serialize

Fig. 4. A simplified implementation of GFDM transmitter.

Now considering the results of (12) and (13) and the discussions following these equations, the simplified

transmitter block diagram of Fig. 4 is obtained. Following this block diagram, the computation of each GFDM

block involves M FFT operations, each of size N , M dot multiplication of p by the expanded output of IFFT,

and M vector additions. For convenience, as a measure of complexity, each complex multiplication followed by

a complex addition is referred to as one operation. Accordingly, the number of operations for IFFTs will be

(M N/2) log 2 N . We note that the elements of p are real-valued and it has a length of M N . We count each real

by complex multiplication as half a complex multiplication. Hence, point-wise multiplications by p and additions

to the circular shift and add block are counted as M 2 N/2 operations. Adding our results, we find that the number

of operation to calculate each GFDM block will be

MN M 2N
C1 = log2 N + operations. (14)
2 2

Taking the formula (7) of [30] (a previous paper on efficient implementation of the transmitter in GFDM) and

assuming that the number of active subcarriers is equal to N (i.e., in formula (7) of [30] we have replaced K by

N ) to match our assumption here, the reported result in [30] will be1

3M N
C2 = log2 M N + 2M N operations. (15)
2

A quick comparison of (14) and (15) reveals that, for typical situations where M is not a large number (typically

3 or 5), the implementation proposed here is about three times less complex than that of [30].

At this point, it is important that a comment be made on the impact of an abrupt start of a GFDM packet at the

beginning of the preamble and an abrupt stop to the packet at the end of the payload. This is equivalent of applying

a rectangular window to the underlying periodic waveform that the GFDM packet originates from. Application of

a rectangular window in time translates to a convolution of the spectrum of the periodic waveform with a ‘sinc’

pulse in the frequency domain. This, in turn, spreads the subcarriers’ spectra over the whole of the transmission

bandwidth and effectively nullifies the filtering operation that was applied earlier to confine the spectrum of each
1
In [30], the number of operations to perform an FFT of size N is assumed to be N log2 N . Here, we have replaced it with the more
accurate figure (N/2) log2 N , [48].
 13

subcarrier to a limited bandwidth. It also results in a significant OOB emission. The solution to this problem is

trivial. The rectangular window is replaced by a window function with smooth corners at the two sides, e.g., raised

cosine shapes at the beginning and at the end of the packet is done as in filtered OFDM, [19], [20]. Alternatively,

a bandpass filter may be applied to each synthesized GFDM signal, as done for OFDM signals in LTE.

C. Receiver Implementation

The key idea for development of any receiver implementation for GFDM is to recall that after removing the CP,

the received signal vector y, of size M N × 1, may be thought as one period of a periodic signal whose spectral

content are those of the transmit signal vector xgfdm , scaled by the channel gains at the respective frequencies

(i.e., the transmitted tones). Channel equalization at the receiver, thus, becomes a trivial task. An FFT is applied

to the received vector y and each element of the result will be divided by the channel gain at the frequency of the

respective tone. The result will be the frequency domain equalized vector yf,e = xgfdm,f + vf , where xgfdm,f and

vf are the FFTs of xgfdm and the channel noise vector v, respectively.

The procedure presented in [28] is a computationally efficient method for extracting the data symbols from yf,e .

This procedure works as follows. The portion of the yf,e that corresponds to each subcarrier is extracted through

a frequency domain filtering; by applying a frequency mask. The extracted portion is shifted to the baseband and

converted from the frequency domain to the time domain through an IFFT (of size M ). The results for the kth

subcarrier band are the estimates of transmitted data symbols sk [0], sk [1], · · · , sk [M − 1]. A point that needs our

attention here is that these estimates, even in the absence of channel noise, are not perfect. Because of the overlapping

of the adjacent subcarrier filters (see Fig. 3), the estimates are subject to ICI among adjacent subcarriers. This point

has been acknowledged in [28] and an iterative method for successive interference cancellation (SIC) of ICI has

been proposed. The successive ICI cancellation clearly has some drawbacks that put GFDM in a disadvantage when

compared with C-FBMC that is introduced in a later part of this article.

In [28], it has also been noted that the iterative ICI cancellation may be avoided through a single step ZF

or MMSE solution which follows by noting that the transmit vector xgfdm is related to the vector sgfdm =

[sT [0] sT [1] · · · sT [M − 1]]T of all the transmitted symbols, within the packet, as

xgfdm = Asgfdm (16)

where A is a non-singular matrix that is determined by the packet parameters; e.g., see [28] or [26] for details.

Also, see [27] for the necessary condition for A not to be singular. One point that should be quoted from [27] is
 14

that for A to be non-singular, M has to be an odd integer. Hence, the ZF detector that is presented below is strictly

applicable to the cases where M odd. The MMSE detector should also be applied to the cases where M is odd,

as when M is even, the receiver will suffer from a larger noise enhancement.

The channel clearly has some impact. The received vector, after stripping CP from it, has the following form

y = HAsgfdm + v (17)

where H is a circular matrix arising from the channel impulse response and v is the channel additive noise vector.

The ZF solution to (17) is, thus, obtained as

szf −1
gfdm = (HA) y. (18)

On the other hand, the MMSE solution to (17) is calculated as

smmse H H −1 H H
gfdm = (Rvv + A H HA) A H y (19)

where Rvv = E[vvH ].

The direct solutions (18) and (19) may be prohibitive in complexity as the size of underlying matrices may

be a few thousands in practice. Beside, as shown in [26] they both may incur some loss in performance, arising

from some possible weak conditioning of H and/or A. Taking note of these, the general consensus seem to be

utilization of an SIC method, [25]–[32]. Nevertheless, we note that, recently, algorithms that reduce the complexity

of implementation of (18) and (19) have been proposed in [49].

D. Extension to MIMO Channels

The key idea that has made application of OFDM to MIMO channels trivial relies on the fact that each subcarrier

signal is a pure tone (after removing the CP) and thus for each channel link it will be affected by a complex gain.

For instance, if we consider a MIMO channel with Nt transmit antennas and Nr receive antennas that multiplexes

Nt data streams in space, the equation relating an Nt × 1 transmit data vector sk of the kth subcarrier and the

corresponding received signal vector (of size Nr × 1) is obtained as

uk = Hk sk + vk , for k = 0, 1, · · · , N − 1 (20)

where uk is the received signal vector, vk is the channel noise vector and Hk is the channel gain matrix at the kth

subcarrier frequency. Applying a zero-forcing (ZF) or a minimum mean square error (MMSE) equalizer to (20),

will allow one to obtain an estimate of sk .

 15

Noting that GFDM waveform is made up of M N tones, the same concept can be applied here too. An FFT

of size M N is applied to the CP stripped received vector y of each received antenna and the results of the same

indexed outputs are collected to obtain an equation similar to (20). Then, a ZF or MMSE equalizer is applied for

each tone separately. This process gives estimates of the kth elements of the signal vector xgfdm for all the space

multiplexed signals. Collating the elements corresponding to each space multiplexed signal gives the estimate of

the corresponding xgfdm which then may be processed according to [28], to extract the data symbols.

For space-time coded (STC) transmission, particularly, the Alamouti codes [50], also one can benefit from the

fact that each GFDM block is made of a summation of the tones and accordingly design an effective signaling

scheme that benefit from the diversity gain provided by multiple antennas. Such a system has been developed and

presented in [26]. The proposed method follows the time-reversal single-carrier approach that was first reported in

[51] and later was adopted for FBMC systems (with linear convolution) in [52]. The method proposed in [26], for

a MIMO setup with Nt = 2 and Nr = 1, is implemented as follows. For other choices of Nt and Nr , following

the same line of thoughts, the respective designs can be made, straightforwardly.

The first antenna transmits a block of GFDM that is characterized by the frequency domain vector xgfdm,f,1 .

Simultaneously, the second antenna transmits another block of GFDM that carries a different set of data symbols

and is characterized by the frequency domain vector xgfdm,f,2 . Next, both antennas transmit a second pair of GFDM

blocks that are characterized by the frequency domain vectors −x∗gfdm,f,2 and x∗gfdm,f,1 , transmitted from the first

and the second antennas, respectively. If the channel gains at the tones that form GFDM signal are presented by

the diagonal matrices H1 and H2 , the received signal blocks, after being CP stripped and passed through an FFT,

result in

y1 = H1 xgfdm,f,1 + H2 xgfdm,f,2 + v1 (21)

and

y2 = −H1 x∗gfdm,f,2 + H2 x∗gfdm,f,1 + v2 , (22)

for the first block and the second block, respectively. Using (21) and (22) and following the Alamouti method [51],

estimates of xgfdm,f,1 and xgfdm,f,2 are calculateded as

−1

x̂gfdm,f,1 = H1 H∗1 + H2 H∗2 H∗1 y1 + H2 y2∗

−1

= xgfdm,f,1 + H1 H∗1 + H2 H∗2 H∗1 v1 + H2 v2∗ (23)
 16

and

−1

x̂gfdm,f,2 = H1 H∗1 + H2 H∗2 H∗2 y1 − H1 y2∗

−1

= xgfdm,f,2 + H1 H∗1 + H2 H∗2 H∗2 v1 − H1 v2∗ , (24)

respectively. Once x̂gfdm,f,1 and x̂gfdm,f,2 are available, the method proposed in [28] may be used to extract the

transmitted data symbols.

V. C IRCULAR FBMC

In this section, we show how the waveform construction that was developed in Section IV, for GFDM, can be

also used to construct alternative waveforms based FBMC signaling. To this end, we first present a short review

of FBMC methods. These presentations follow the conventional FBMC systems where convolutions are linear.

Extensions to GFDM-like waveforms, where linear convolutions are replaced by circular ones, will then follow.

The prefix “C” is added whenever reference is made to the implementations that are based on circular convolution.

A. Historical notes

Filter bank multicarrier (FBMC) methods were first proposed about 50 years ago, by Chang, [53], and Saltzberg,

[54]. This was before the first introduction of OFDM in 1971, [55]. The method proposed by Chang considers

transmission of a set of sequences of pulse amplitude modulated (PAM)/real-valued data symbols through a number

of vestigial side-band (VSB) subcarriers. The subcarriers spectra are overlapped such that, following the Nyquist

definition, a bandwidth efficiency of 100% is achieved. Saltzberg rearranged the real and imaginary parts of each

symbol in a set of quadrature amplitude modulated (QAM) symbol sequences such that they could be also transmitted

with a bandwidth efficiency of 100%. Here, the real and imaginary parts of each QAM symbol are time staggered

by half a symbol interval. In a recent work, [56], we have shown that the methods presented in [53] and [54] are

effectively the same and the names cosine modulated multitone (CMT) and staggered modulated multitone (SMT),

respectively, have been suggested to refer to them.

Although SMT (often referred to as OQAM-OFDM) has been more popular in the literature, consideration of

CMT for certain implementations may lead to more effective receivers. In particular, in some recent works, [57],

[58], we have shown that the blind equalization capability of CMT which has been introduced in [59] may prove

very useful in overcoming the problem of pilot contamination (a serious problem) in massive MIMO systems, e.g.,

see [60]. Here, when it comes to the details of implementation of transmitter and receiver, we follow the norm in the
 17

f
.. .. .. .. ..
7
. . . . T
.
f=
4T π π π
··· 0 0 ···
2 2 2
1
5 F =
f= 2T
4T π π
··· 0 0 0 ···
2 2

3
f=
4T π π π
··· 0 0 ···
2 2 2

1
f=
4T π π
··· 0 0 0 ···
2 2

−2T −T T 2T t
n = −2 n = −1 n=1 n=2

Fig. 5. The CMT time-frequency phase-space lattice.

literature and provide the details for SMT. Extensions to CMT are not much different and should be straightforward

for the interested readers.

B. CMT

As noted above, in CMT, data symbols are from a PAM alphabet. To establish a transmission with the maximum

bandwidth efficiency, PAM symbols are distributed in a time-frequency phase-space lattice with a density of two

symbols per unit area. This is equivalent to one complex symbol per unit area. Moreover, a 90 degree phase shift

is introduced to the respective carriers among the adjacent symbols. These concepts are presented in Fig. 5. VSB

modulation is applied to cope with the carrier spacing F = 1/2T . The pulse-shape used for this purpose at the

transmitter as well as for matched filtering at the receiver is a square-root Nyquist waveform, p(t), which has been

designed such that q(t) = p(t) ⋆ p(−t) be a Nyquist pulse with regular zero crossings at 2T time intervals. Also,

p(t), by design, is a real-valued even function of time, hence, p(t) = p(−t). This implies that P (f ) is also a

real-valued even function of the frequency.

Fig. 6 presents a set of magnitude responses of the modulated versions of the pulse-shape p(t) for the data

symbols transmitted at t = 0 and t = T . The colors used for the plots follow those in Fig. 5, to reflect the

respective phase shifts.

 18

t=0 ···
1 3 5 7
f
4T 4T 4T 4T

t=T
···
1 3 5 7
f
4T 4T 4T 4T

Fig. 6. Magnitude responses of the CMT pulse-shaping filters at various subcarriers and time instants t = 0 and t = T .

f
.. .. .. .. ..
. . . . T /2
.
3
f=
T π π
0
π
··· 0 ···
2 2 2
1
F =
2 T
f=
T π π
··· 0 0 0 ···
2 2

1
f=
T π π π
··· 0 0 ···
2 2 2

f =0 π π
··· 0 0 0 · ·t ·
2 2
−T −T /2 T /2 T
n = −2 n = −1 n=1 n=2

Fig. 7. The SMT time-frequency phase-space lattice.

C. SMT

SMT may be thought as an alternative to CMT, [56]. Its time-frequency phase-space lattice is obtained from that

of CMT (Fig. 5) through a frequency shift of the lattice points down by 1/4T , scaling the time axis by a factor

1/2 and hence the frequency axis by a factor 2, [56]. This leads to the time-frequency phase-space lattice that

is presented in Fig. 7. The magnitude responses of the SMT pulse-shaping filters at various subcarriers and time

instants t = 0 and t = T /2 are presented in Fig. 8.

D. Transmitter implementation

The transmitter implementation for both C-CMT and C-SMT follows the same procedure as GFDM. Minor

changes are required to consider the phase adjustment of π/2 among adjacent symbols. Also, some adjustment to
 19

t=0 ···
1 2 3
f
T T T

t = T /2
···
1 2 3
f
T T T

Fig. 8. Magnitude responses of the SMT pulse-shaping filters at various subcarriers and time instants t = 0 and t = T /2.

the width of subcarrier bands and their position have to be noted, in the case of C-CMT. In C-SMT, on the other

hand, it should be noted that phase and quadrature parts of QAM symbols have a time offset of T /2. In the rest

of this section, we only discuss C-SMT and leave the C-CMT details for an interested reader to work out.

In C-SMT, a block/packet is constructed by taking the following steps:

1) For m = 0, 1, · · · , M − 1, let

sR [n] = [s0,R [n] js1,R [n] · · · j N −1 sN −1,R [n]]T

and

sI [m] = [s0,I [n] js1,I [n] · · · j N −1 sN −1,I [n]]T .

2) Construct the length M N vectors xR [n] = p ⊙ (F −1 −1

M N sR,e [n]) and xI [n] = p ⊙ (F M N sI,e [n]), where sR,e [n]

and sI,e [n] are the M -fold expanded versions of sR [n] and sI [n], respectively.

3) Compute
M
X −1 M
X −1
xc-smt = circshift(xR [n], nN ) + j circshift(xI [n], nN + N/2) (25)
n=0 n=0

4) Add a CP to xc-smt

Note that in Step 2) F −1 −1

M N sR,e [n] and F M N sI,e [n] can be obtained efficiently by stacking M copies of the vectors

−1
FN sR [n] and F −1
N sI [n], respectively, at the top of each other. Moreover, further reduction in complexity is possible

by removing the factors of j in sR [n] and sI [n], calling the results s̃R [n] and s̃I [n], calculating F −1
N s̃R [n] and
−1
FN s̃I [n], and cyclicly shifting the result by N/4 samples, to get F −1 −1
N sR [n] and F N sI [n], respectively. This

procedure simplifies the calculations, since the FFT of the pair of real vectors s̃R [n] and s̃I [n] can be performed

with a complexity that is approximately equal to that of one FFT with a complex input, [48].
 20

The implementation procedure discussed here has a complexity that is comparable to its GFDM counterpart.

This implementation has some similarity with the one presented in [34] and [36], however, these works did not

mention the simplification steps that are noted above.

E. Receiver implementation

The receiver implementation in C-FBMC systems can also follow the steps mentioned in Section IV-C. Here, as

opposed to GFDM, the embedded data symbols in yf,e are carried by a set of bases functions that are orthogonal

with one another, where orthogonality is the real orthogonality defined in the FBMC literature, e.g., see [61], [19],

and [20]. This, in turn, implies that the procedures mentioned in the second paragraph of Section IV-C will result

in perfect recovery of data symbols free of intersymbol interference (ISI) and ICI. In other words, here, no SIC is

required.

F. Extension to MIMO Channels

C-FBMC extension to MIMO channels is not different from that of GFDM. ZF or MMSE equalizer is applied

in the frequency domain to separate the tones that correspond to different spatial sequences. Subsequently, the

data extraction for each spatial stream is carried out separately according to the procedure that was discussed in

the previous subsection. Alamouti STC can also be applied to C-FBMC following a similar procedure to the one

discussed in [26] for GFDM. Detailed implementation of the Alamouti STC in C-FBMC has been reported in [62].

VI. UFMC
A. Transmitter implementation

Fig. 9 presents a block diagram of the UFMC transmitter. A set of data symbols that corresponds to a given

allocation of a single user are modulated over the assigned subcarriers through an IFFT processing block. The

unused (zero) inputs to the IFFT block in Fig. 9 are reserved for other users in the network. The IFFT output is

serialized and passed through a bandpass filter before transmission. The bandpass filter removes the OOB spectral

leakage that arises due to truncation of data carrying tones to the length of IFFT. By taking this filtering step, we

are assured that different users in the network remain (approximately) orthogonal even under the conditions that

they are not perfectly synchronized.

Unlike OFDM (as well as GFDM and C-FBMC), UFMC does not rely on a CP to absorb the channel transient

response. Instead, the output samples of IFFT are padded with a number of zeros equal to the duration of the

impulse response of the bandpass filter. Moreover, the bandpass filter is designed such that the duration of its
 21

IFFT
bandpass output
and
filter
serialize

data symbols of the intended user

zero inputs (reserved for other users)

Fig. 9. Block diagram of the UFMC transmitter.

impulse response be equal to the common CP length used in OFDM, [41]–[45]. This choice, although not necessary,

equalizes the length of UFMC and OFDM symbols, hence, makes any comparison between the two straightforward.

For the discussions that follow, we let the size of IFFT in Fig. 9 be N , and assume the duration of the impulse

response of the bandpass filter be equal to N ′ + 1, where N ′ = NCP is the CP length in OFDM. Accordingly, each

UFMC symbol will have a length of N + N ′ . Following [41]–[45], we assume successive UFMC symbols of this

length are concatenated and transmitted, as in the case of OFDM.

B. Key points

A few key points that help us in understanding the main features of UFMC are in order. First, we note that a

UFMC symbol that has been generated according to the procedure in Fig. 9, before going through the channel,

preserves the orthogonality of the intended user’s symbols perfectly. This fundamental point is the basis for the

receiver structure that has been proposed in UFMC literature, e.g., [41]–[45], and is presented below. An explanation

to this observation in provided in the next subsection, below. Second, the presence of channel expands the duration

of each UFMC symbol beyond the length of N + N ′ samples. This concept is depicted in Fig. 10, where a stream

of UFMC symbols (S1, S2, · · · ) is depicted at the channel input and channel output. Obviously, channel introduces

ISI among successive symbols. Also, if the processing at the receiver is made based on the segments of N + N ′

samples of the received signal, some loss in the orthogonality of the intended user’s data symbols will incur. By

increasing the length of the segments, this loss of orthogonality can be recovered at the cost of an introduction

of some ISI. It may be argued that although, strictly speaking, in the presence of a channel, UFMC has to be

categorized as a non-orthogonal multicarrier system, the loss of orthogonality in many practical situations may

remain insignificant. Thus, more often, UFMC is categorized as a quasi orthogonal signaling method.
 22

S1 S2 S3 ···
time
(a)

S1 S2 S3 ···

(b)

Fig. 10. A UFMC symbol stream (a) before channel, and (b) after channel.

C. UFMC orthogonality

The IFFT output in Fig. 9 is the summation of set of sine-waves/tones that are time limited to an interval T

seconds (equivalently, N samples) and spaced in frequency by 1/T Hertz. In the frequency domain, these truncated

tones are characterized by a set of sinc pulses, as depicted in Fig. 11. The orthogonality of the tones is characterized

by the fact that they do not interfere with one another as along as they are sampled in the frequency domain at the

middle of the sinc pulses; the vertical dashed lines in Fig. 11.

f
T
no ICI

Fig. 11. Demonstration of the orthogonality of a set of tones that are time-limited to an interval T seconds and spaced by 1/T Hz.

Passing the summation of the tones through the bandpass filter at the UFMC transmitter output attenuates the

amplitude of the side-lobes that fall out of the band of interest, but, as depicted in Fig. 12, has no impact on the

orthogonality of the subcarriers. This, clearly, shows that the filtering step in Fig. 9 does not affect the orthogonality

of the subcarriers in UFMC. The presence of channel, as demonstrated in Fig. 10, extends the length of each UFMC

symbol beyond N + N ′ samples. Obviously, this does not affect the orthogonality of subcarriers and, hence, the

data symbols can be recovered free of ICI if processing at the receiver is performed based on the extended length

of the received signal samples. Nevertheless, as noted above, the detected symbols will suffer from ISI. One may

choose to process a length of the received signal samples to minimize the sum of ISI and ICI. One may also note
 23

that both ISI and ICI can be completely removed if the length of each UFMC symbols is extended to N + N ′

plus the maximum during of the channel impulse response, by appending a set of zero samples at the end of each

UFMC symbol. This is equivalent of adding a guard interval between successive UFMC symbols.

f
T
no ICI

Fig. 12. The impact of the filtering in UFMC on the spectra of subcarrier tones.

D. Filtering options

Any bandpass filter that covers the subcarriers of the user of interest may be used. The proposers of UFMC have

emphasized on the use Dolph-Chebyshev filters, [41], but have noted that other choices are also possible, [42].

Methods to design filters that provide improved robustness to carrier frequency offset and timing offset are also

have been reported, [63].

Early works on the implementation of UFMC assumed direct convolution of the IFFT output with the bandpass

filter; as in Fig. 9. This implementation may be too complex as it requires about N N ′ multiplications and additions.

A recent result, [64], suggests a method of reducing this complexity significantly, through a frequency domain

implementation of the filtering operation. Another possibility that we suggest here is to apply a separate filter to

each subcarrier, as is done in the early developments of filtered OFDM, e.g., in IEEE 802.11a and g, [65]. Here,

all one needs to do is to apply a window with smooth roll-offs at the two sides to the IFFT output. It is equivalent

of replacing the rectangular prototype filter in IFFT filter bank by a window with smooth transitions at the two

sides. The advantage of this method is a very low complexity. It only requires 2N ′ multiplications. Nevertheless,

it loses some flexibilities that direct filter design offers. A tutorial review of the filtered OFDM and a method of

optimizing the corresponding window function is available in [19].

E. Receiver implementation

At the receiver side, the symbol boundaries are identified as part of the synchronization process. To recover the

QAM data symbols carried in each UFMC symbol, the respective segment of the received signal, usually consisting
 24

of N + N ′ samples, is taken and appended with sufficient number of zeros to be extended to a vector of length

2N . The result is then passed to an FFT block. The FFT outputs at the respective subcarriers (a set of odd or

even indexed outputs) carry the desired symbol, scaled by channel gains at the respective frequencies. As in the

conventional OFDM, the channel gains are compensated through a set of single tap equalizer; one per subcarrier.

F. Extension to MIMO channels

From the above discussions, we note that the presence of channel may result in the loss of orthogonality

among the subcarriers. This potentially can limit the application of UFMC in MIMO channels. However, since for

channels with moderate spreading in time, the loss of orthogonality among subcarriers remain insignificant, MIMO

processing techniques may be applied to separate subcarriers as in OFDM, albeit within some approximation. Such

approximation may be avoided by adding a guard interval between the successive symbols of UFMC. To emphasize,

UFMC can be used in for both STC and MIMO space multiplexing with the same level of flexibility as OFDM.

This follows from the fact that, in UFMC, each data symbol modulate a single tone, as in OFDM.

VII. M ULTIUSER C OMMUNICATIONS

As was noted earlier, the main motivation in proposing the new 5G contender waveforms is to reduce the

interference among different users’ signals when they are not perfectly synchronized. In this section, we make

some comments on effectiveness of GFDM, C-FBMC, and UFMC in handling multi-user signals when these

signals may not be synchronized both in symbol timing and carrier frequency.

A. GFDM/C-FBMC

Fig. 13 depicts a communication scenario where a set of users are transmitting in an uplink of a network. The

case presented is applicable to both GFDM and C-FBMC, but not to UFMC. The case of UFMC is discussed

separately later. Each user transmits a sequence of GFDM/C-FBMC blocks. Each block has a structure similar

to the one in Fig. 2. Each user channel introduces spreading in time to the respective GFDM/C-FBMC sequence

and as a result the successive blocks may overlap. Also, with a relaxed synchronization among the users, the data

blocks from different users may not be perfectly time aligned. Near perfect extraction of users’ information may

be achieved through a receiver signal processing algorithm as long as a correct time window of the received signal

is chosen. A correct time window is the one that does not pick any user signal that suffer from blocks overlap. In

Fig. 13, the time window B is a correct choice, but A and C both suffer from some inter-block interference. If the

time window is selected incorrectly, the portions of the users’ waveforms that belong to the transient parts will no
 25

A B C

frequency
!"#$ !"#$ !"#$
MT MT MT

user 1

user 2

user 3

time

CP payload block overlap

Fig. 13. A UFMC symbol stream (a) before channel, and (b) after channel.

longer be a sum of pure tones. Thus, the balance among the tones in each data stream in the respective user signal

which results in well designed filters that were discussed in Section III will be lost and, as a result, OOB spectral

leakage will appear. This, clearly, leads to multi-user and inter-block interference.

A correct choice of a window (of length M T ) of the received signal that was discussed above avoids ISI, but

cannot resolve the problem of inter-user interference, if the users are not perfectly carrier synchronized. Here,

an attempt is made to clarify this point through couple of numerical examples without digging into the detailed

mathematical equations. More on the subject of multi-user interference in GFDM and C-FBMC can be found in

[66], [67].

Fig. 14 presents two samples of the prototype filter in a GFDM/C-FBMC for the case where N = 16 and M = 4.

These filters have been designed using the Matrin and Mirabbasi method, [46], [47]. The prototype filter hb (t) is

for shaping the data symbols at the beginning of the packet, and hm (t) is for shaping the data symbols at the middle

of the packet. Similar filters are used at the receiver for matched filtering. We also recall that these prototype filters

are modulated to the various subcarriers before being amplitude scaled by the respective data symbols.

Looking at the plots in Fig. 14, one may note that hm (t) is a proper pulse shape as commonly seen in digital

communication systems and hb (t) is obtained by circularly shifting hm (t) within the a window of length M T . The

magnitude of the Fourier transforms of hb (t) and hm (t) are presented in Fig. 15. Here, we note that while Hm (f )

has a common lowpass response that we often see for pulse shaping filters, the magnitude response of Hb (f ) is
 26

0.8
hb (t) hm (t)
0.6

0.4

0.2

-0.2
0 1 2 3 4
t/T

Fig. 14. Impulse responses of two samples of the prototype filter in a C-FBMC system for shaping the data symbols at the beginning of a
packet (red) and the data symbols at the middle of the same packet (blue).

vastly different. However, Hb (f ) and Hm (f ) have the following common features. (i) Hb (0) = Hm (0) = 1. (ii)
1 1
Hb (f ) and Hm (f ) share the same null frequencies. These null frequencies are at the positions T, T + M1T , 1
T + M2T ,

· · · . We also note that the less seen form of Hb (f ), particularly, its high side lobes beyond the point f T = 1,

arises because of the sharp cut-offs at the beginning and the end of hb (t).

Next, if we think of the modulated versions of hb (t) and hm (t) to the subcarriers center frequencies that are
1
located at integer multiples of T, and recall that the sinusoidal tones synthesizing each subcarrier band are located
1
at integer multiples of MT , one immediately finds that the position of the null points in Hb (f ) and Hm (f ) implies

that, in GFDM/C-FBMC, any pair of non-adjacent subcarriers remain orthogonal to each others if they are perfectly

carrier synchronized. Clearly, if the pair of subcarriers are originated from the same transmitter/user, they will be

synchronized and, thus, have no interference with one another. However, if they originate from different transmitters

and, thus, there may be some mismatch between their respective carrier frequencies, multi-user interference can

occur. For the data symbols at the middle of each packet such interference will remain insignificant, because of high

attenuation in the stop-bands of the respective filters/pulse shapes. But for the data symbols at or near the beginning

and the end of each packet, the level of interference can be significant, obviously, because of poor responses of the

respective filters.

Different measures can be taken to avoid multiuser interference in GFDM/C-FBMC. Among these, [67] has

adopted the received signal windowing that was originally proposed in [68]–[72], to avoid out-of-band interference
 27

!
|Hm (f )| |Hb (f )|
&*!

&"!

&)!

dB
|
&#!
|
&(!

&$!

&'!
! " # $ %
fT

Fig. 15. Magnitude response of the prototype filters whose impulse responses are presented in Fig. 14.

in digital subscriber lines (DSL) modems. This method requires an extension of the CP at the transmitter, and at the

receiver a window of the received signal with a length larger than M T is used for signal processing and extraction

of data symbols of different users. Interested readers are referred to [67] for the details.

B. UFMC

Fig. 16 presents a similar scenario to that of Fig. 13, for the case where transmission is based on UFMC. Each

UFMC symbol has a transient at the beginning and a transient at the end. Each transient has a duration that is equal

to the sum of the durations of the channel impulse response and the impulse response of the bandpass filter that is

used at the transmitter to contain each user signal within the desired bandwidth, as well as any other processing in

the transmitter/receiver chain. Here, the choice of the time window should include the transients on both sides of

each UFMC symbol, for all the users, to benefit from filtering operation applied at the respective transmitters to

resolve multi-user interference. Also, complete coverage of the transients assures the orthogonality of subcarriers

for each user. However, as also noted earlier, an extended time window has the drawback of introducing ISI. Hence,

in practice, a balanced decision has to be made on the size of the time window. The presented time window Tw ,

in Fig. 16, covers the transient periods of the middle symbol for the three users. However, because of the time

misalignments of the users, this choice results in extra ISI from the adjacent symbols beyond the transients of

the individual users. As noted earlier, by adding guard intervals between successive UFMC symbols, ISI can be

avoided at a cost of some reduction in the transmission efficiency.

 28

frequency
Tw

user 1

user 2

user 3

time

transient payload symbol overlap

Fig. 16. A UFMC symbol stream (a) before channel, and (b) after channel.

VIII. C ONCLUSION

In this paper, we presented a novel point of view of GFDM and C-FBMC modulation techniques; two new

contender waveforms of the upcoming fifth generation (5G) of wireless communication systems. We showed both

GFDM and C-FBMC waveforms can be synthesized on the same principle that OFDM waveform is built. The

constructed waveforms can be viewed as a summation of a number of tones that at the receiver can be separated

through application of an FFT. We showed that this property of GFDM/C-FBMC is the key to simple (single-

tap) equalization in the frequency domain and straightforward adoption of these waveforms to MIMO channels.

Moreover, we reviewed UFMC, the third contender waveform that has recently been proposed for 5G, and showed

that this waveform also preserves the key features of OFDM. The application of GFDM/C-FBMC and UFMC to

multiuser networks was also discussed briefly. It was noted that special receiver signal processing is needed to

avoid multiuser interference in GFDM/C-FBMC. UFMC was found to be more relaxed to such needs. It was noted

that the non-orthogonality of GFDM is a disadvantage that make this waveform less attractive than C-FBMC and

UFMC. However, more research is needed to weight the advantages and the disadvantages of C-FBMC and UFMC

as progress is being made in their development for 5G applications.

R EFERENCES
[1] W. Y. Chen, DSL Simulation Techniques and Standards Development for Digital Subscriber Line Systems. Indianapolis, Indiana:
Macmillan Publishing, 1998.
[2] T. Starr, J. Cioffi, and P. Silverman, Understanding Digital Subscriber Line Technology. Upper Saddle River, NJ: Prentice Hall, 1999.
[3] R. V. Nee and R. Prasad, OFDM for Wireless Multimedia Communications. Boston: Arthec House, 2000.
 29

[4] Y. Li and G. Stüber, Eds., Orthogonal Frequency Division Multiplexing for Wireless Communications. New York, NY: Springer, 2006.
[5] M. Morelli, C. C. J. Kuo, and M. O. Pun, “Synchronization techniques for orthogonal frequency division multiple access (OFDMA):
A tutorial review,” Proc. of the IEEE, vol. 95, no.7, pp. 1394–1427, July 2007.
[6] H. Defeng and K. B. Letaief, “An interference-cancellation scheme for carrier frequency offsets correction in OFDMA systems,” IEEE
Trans. Commun., vol. 53, no.7, pp. 1155–1165, July 2005.
[7] K. Lee and I. Lee, “CFO compensation for uplink OFDMA systems with conjugated gradient,” Proc. of the IEEE ICC ’11, pp. 1–5,
June 2011.
[8] K. Lee, S.-R. Lee, S.-H. Moon, and I. Lee, “MMSE-based CFO compensation for uplink OFDMA systems with conjugate gradient,”
IEEE Trans. Wireless Commun., vol. 11, no.8, pp. 2767–2775, Aug. 2012.
[9] A. Farhang, N. Marchetti, and L. Doyle, “Low Complexity LS and MMSE Based CFO Compensation Techniques for the Uplink of
OFDMA Systems,” Proc. of the IEEE ICC ’13, June 2013.
[10] H. Saeedi-Sourck, Y. Wu, J. Bergmans, S. Sadri, and B. Farhang-Boroujeny, “Complexity and Performance Comparison of Filter Bank
Multicarrier and OFDM in Uplink of Multicarrier Multiple Access Networks,” Signal Processing, IEEE Transactions on, vol. 59, no. 4,
pp. 1907–1912, 2011.
[11] S. Brandes, I. Cosovic, and M. Schnell, “Sidelobe suppression in OFDM systems by insertion of cancellation carriers,” in Vehicular
Technology Conference, 2005. VTC-2005-Fall. 2005 IEEE 62nd, vol. 1, 2005, pp. 152–156.
[12] Z. Yuan, S. Pagadarai, and A. M. Wyglinski, “Cancellation carrier technique using genetic algorithm for OFDM sidelobe suppression,”
in Military Communications Conference, 2008. MILCOM 2008. IEEE, 2008, pp. 1–5.
[13] A. Selim, I. Macaluso, and L. Doyle, “Efficient sidelobe suppression for ofdm systems using advanced cancellation carriers,” in
Communications (ICC), 2013 IEEE International Conference on, June 2013, pp. 4687–4692.
[14] S. Premnath, D. Wasden, S. Kasera, B. Farhang-Boroujeny, and N. Patwari, “Beyond ofdm: Best-effort dynamic spectrum access using
filterbank multicarrier,” in Communication Systems and Networks (COMSNETS), 2012 Fourth International Conference on, 2012, pp.
1–10.
[15] S. Premnath, D. Wasden, S. Kasera, N. Patwari, and B. Farhang-Boroujeny, “Beyond ofdm: Best-effort dynamic spectrum access using
filterbank multicarrier,” Networking, IEEE/ACM Transactions on, vol. 21, no. 3, pp. 869–882, 2013.
[16] B. Farhang-Boroujeny and R. Kempter, “Multicarrier communication techniques for spectrum sensing and communication in cognitive
radios,” Communications Magazine, IEEE, vol. 46, no. 4, pp. 80–85, 2008.
[17] B. Farhang-Boroujeny, “Prolate Filters for Nonadaptive Multitaper Spectral Estimators With High Spectral Dynamic Range,” Signal
Processing Letters, IEEE, vol. 15, pp. 457–460, 2008.
[18] ——, “Filter Bank Spectrum Sensing for Cognitive Radios,” Signal Processing, IEEE Transactions on, vol. 56, no. 5, pp. 1801–1811,
2008.
[19] ——, “OFDM Versus Filter Bank Multicarrier,” IEEE Signal Processing Magazine, vol. 28, no. 3, pp. 92–112, 2011.
[20] ——, “Filter bank multicarrier modulation: A waveform candidate for 5G and beyond,” Advances in Electrical Engineering, vol. 2014,
2014.
[21] A. Sahin, I. Guvenc, and H. Arslan, “A survey on multicarrier communications: Prototype filters, lattice structures, and implementation
aspects,” pp. 1–27, 2013.
[22] Bellanger, “FBMC physical layer: a primer,” PHYDYAS, Tech. Rep., 06/2010.
[23] M. Payaró, A. Pascual-Iserte, and M. Najar, “Performance comparison between fbmc and ofdm in mimo systems under channel
uncertainty,” in Wireless Conference (EW), 2010 European, 2010, pp. 1023–1030.
[24] T. Ihalainen, A. Ikhlef, J. Louveaux, and M. Renfors, “Channel equalization for multi-antenna fbmc/oqam receivers,” Vehicular
Technology, IEEE Transactions on, vol. 60, no. 5, pp. 2070–2085, 2011.
[25] G. Fettweis, M. Krondorf, and S. Bittner, “GFDM - generalized frequency division multiplexing,” in Vehicular Technology Conference,
2009. VTC Spring 2009. IEEE 69th, April 2009, pp. 1–4.
[26] N. Michailow, M. Matthe, I. Gaspar, A. Caldevilla, L. Mendes, A. Festag, and G. Fettweis, “Generalized frequency division multiplexing
for 5th generation cellular networks,” Communications, IEEE Transactions on, vol. 62, no. 9, pp. 3045–3061, Sept 2014.
[27] M. Matthe, L. Mendes, and G. Fettweis, “Generalized frequency division multiplexing in a Gabor transform setting,” Communications
Letters, IEEE, vol. 18, no. 8, pp. 1379–1382, Aug 2014.
[28] I. Gaspar, N. Michailow, A. Navarro, E. Ohlmer, S. Krone, and G. Fettweis, “Low complexity GFDM receiver based on sparse frequency
domain processing,” in Vehicular Technology Conference (VTC Spring), 2013 IEEE 77th, June 2013, pp. 1–6.
[29] M. Matthe, L. L. Mendes, and G. Fettweis, “Space-time coding for generalized frequency division multiplexing,” in European Wireless
2014; 20th European Wireless Conference; Proceedings of, May 2014, pp. 1–5.
[30] N. Michailow, I. Gaspar, S. Krone, M. Lentmaier, and G. Fettweis, “Generalized frequency division multiplexing: Analysis of an
alternative multi-carrier technique for next generation cellular systems,” in Wireless Communication Systems (ISWCS), 2012 International
Symposium on, Aug 2012, pp. 171–175.
[31] R. Datta, N. Michailow, S. Krone, M. Lentmaier, and G. Fettweis, “Generalized frequency division multiplexing in cognitive radio,”
in Signal Processing Conference (EUSIPCO), 2012 Proceedings of the 20th European, Aug 2012, pp. 2679–2683.
[32] M. Danneberg, R. Datta, A. Festag, and G. Fettweis, “Experimental testbed for 5G cognitive radio access in 4G LTE cellular systems,”
in Sensor Array and Multichannel Signal Processing Workshop (SAM), 2014 IEEE 8th, June 2014, pp. 321–324.
[33] M. J. Abdoli, M. Jia, and J. Ma, “Weighted circularly convolved filtering in OFDM/OQAM,” in Personal Indoor and Mobile Radio
Communications (PIMRC), 2013 IEEE 24th International Symposium on, Sept 2013, pp. 657–661.
[34] H. Lin and P. Siohan, “An advanced multi-carrier modulation for future radio systems,” in Acoustics, Speech and Signal Processing
(ICASSP), 2014 IEEE International Conference on, May 2014, pp. 8097–8101.
[35] M. Schellmann, Z. Zhao, H. Lin, P. Siohan, N. Rajatheva, V. Luecken, and A. Ishaque, “FBMC-based air interface for 5G mobile:
Challenges and proposed solutions,” in Cognitive Radio Oriented Wireless Networks and Communications (CROWNCOM), 2014 9th
International Conference on, June 2014, pp. 102–107.
 30

[36] H. Lin and P. Siohan, “Multi-carrier modulation analysis and wcp-coqam proposal,” EURASIP Journal on Advances in Signal
Processing, vol. 2014, no. 1, 2014. [Online]. Available: http://dx.doi.org/10.1186/1687-6180-2014-79
[37] B. Farhang-Boroujeny and H. Moradi, “Derivation of GFDM based on OFDM principles,” in Communications (ICC), 2015 IEEE
International Conference on, June 2015, pp. 2680–2685.
[38] A. Tonello, “A novel multi-carrier scheme: Cyclic block filtered multitone modulation,” in Communications (ICC), 2013 IEEE
International Conference on, June 2013, pp. 5263–5267.
[39] A. Tonello and M. Girotto, “Cyclic block FMT modulation for communications in time-variant frequency selective fading channels,”
in Signal Processing Conference (EUSIPCO), 2013 Proceedings of the 21st European, Sept 2013, pp. 1–5.
[40] ——, “Cyclic block fmt modulation for broadband power line communications,” in Power Line Communications and Its Applications
(ISPLC), 2013 17th IEEE International Symposium on, March 2013, pp. 247–251.
[41] V. Vakilian, T. Wild, F. Schaich, S. ten Brink, and J.-F. Frigon, “Universal-filtered multi-carrier technique for wireless systems beyond
LTE,” in Globecom Workshops (GC Wkshps), 2013 IEEE, Dec 2013, pp. 223–228.
[42] X. Wang, T. Wild, F. Schaich, and A. Fonseca dos Santos, “Universal filtered multi-carrier with leakage-based filter optimization,” in
European Wireless 2014; 20th European Wireless Conference; Proceedings of, May 2014, pp. 1–5.
[43] T. Wild, F. Schaich, and Y. Chen, “5G air interface design based on universal filtered (UF-)OFDM,” in Digital Signal Processing (DSP),
2014 19th International Conference on, Aug 2014, pp. 699–704.
[44] F. Schaich and T. Wild, “Waveform contenders for 5G - OFDM vs. FBMC vs. UFMC,” in Communications, Control and Signal
Processing (ISCCSP), 2014 6th International Symposium on, May 2014, pp. 457–460.
[45] ——, “Relaxed synchronization support of universal filtered multi-carrier including autonomous timing advance,” in Wireless
Communications Systems (ISWCS), 2014 11th International Symposium on, Aug 2014, pp. 203–208.
[46] K. Martin, “Small side-lobe filter design for multitone data-communication applications,” Circuits and Systems II: Analog and Digital
Signal Processing, IEEE Transactions on, vol. 45, no. 8, pp. 1155–1161, 1998.
[47] S. Mirabbasi and K. Martin, “Overlapped complex-modulated transmultiplexer filters with simplified design and superior stopbands,”
Circuits and Systems II: Analog and Digital Signal Processing, IEEE Transactions on, vol. 50, no. 8, pp. 456–469, 2003.
[48] A. V. Oppenheim and R. W. Schafer, Digital Signal Processing. Prentice–Hall, 1975.
[49] A. Farhang, N. Marchetti, and L. Doyle, “Low complexity gfdm receiver design: A new approach,” in Communications (ICC), 2015
IEEE International Conference on, June 2015, pp. 4775–4780.
[50] S. Alamouti, “A simple transmit diversity technique for wireless communications,” Selected Areas in Communications, IEEE Journal
on, vol. 16, no. 8, pp. 1451–1458, Oct 1998.
[51] N. Al-Dhahir, “Single-carrier frequency-domain equalization for space-time block-coded transmissions over frequency-selective fading
channels,” Communications Letters, IEEE, vol. 5, no. 7, pp. 304–306, July 2001.
[52] M. Renfors, T. Ihalainen, and T. Stitz, “A block-alamouti scheme for filter bank based multicarrier transmission,” in Wireless Conference
(EW), 2010 European, April 2010, pp. 1031–1037.
[53] R. Chang, “High-speed multichannel data transmission with bandlimited orthogonal signals,” Bell Sys. Tech. J., vol. 45, pp. 1775–1796,
Dec. 1966.
[54] B. Saltzberg, “Performance of an Efficient Parallel Data Transmission System,” IEEE Transactions on Communication Technology,
vol. 15, no. 6, pp. 805–811, 1967.
[55] S. B. Weinstein and P. M. Ebert, “Data transmission by frequency-division multiplexing using the discrete Fourier transform,” IEEE
Transactions on Communication Technology, vol. 19, no. 5, pp. 628–634, Oct. 1971.
[56] B. Farhang-Boroujeny and C. (George) Yuen, “Cosine modulated and offset QAM filter bank multicarrier techniques: a continuous-
time prospect,” EURASIP Journal on Applied Signal Processing, 2010, special issue on Filter Banks for Next Generation Multicarrier
Wireless Communications, vol. 2010, p. 16 pages, 2010.
[57] A. Farhang, N. Marchetti, L. Doyle, and B. Farhang-Boroujeny, “Filter bank multicarrier for massive MIMO,” in arXiv: 1402.5881, to
appear in the proceedings of VTC Fall, 2014.
[58] A. Farhang, M. N. Aminjavaheri, A., L. Doyle, and B. Farhang-Boroujeny, “Pilot decontamination in CMT-based massive MIMO
networks,” in arXiv: 1402.5881, to appear in the proceedings of ISWC, 2014.
[59] B. Farhang-Boroujeny, “Multicarrier modulation with blind detection capability using cosine modulated filter banks,” IEEE Transactions
on Communications, vol. 51, no. 12, pp. 2057–2070, 2003.
[60] J. Jose, A. Ashikhmin, T. Marzetta, and S. Vishwanath, “Pilot Contamination and Precoding in Multi-Cell TDD Systems,” IEEE
Transactions on Wireless Communications, vol. 10, no. 8, pp. 2640–2651, 2011.
[61] B. Le Floch, M. Alard, and C. Berrou, “Coded orthogonal frequency division multiplex,” Proceedings of the IEEE, vol. 83, no. 6, pp.
982–996, 1995.
[62] D. Kong, X.-G. Xia, and T. Jiang, “An Alamouti coded CP-FBMC-MIMO system with two transmit antennas,” vol. 58, no. 10, pp.
1–6, 2015. [Online]. Available: http://dx.doi.org/10.1007/s11432-015-5345-3
[63] X. Wang, T. Wild, and F. Schaich, “Filter optimization for carrier-frequency- and timing-offset in universal filtered multi-carrier systems,”
in Vehicular Technology Conference (VTC Spring), 2015 IEEE 81st, May 2015, pp. 1–6.
[64] T. Wild and F. Schaich, “A reduced complexity transmitter for UF-OFDM,” in IEEE Vehicular Technology Conference (VTC), Spring
2015, accepted.
[65] B. Farhang-Boroujeny, Signal Processing Techniques for Software Radios. Lulu publication house, second edition, 2009.
[66] A. Aminjavaheri, A. Farhang, A. L. RezazadehReyhani, and B. Farhang-Boroujeny, “Impact of timing and frequency offsets on
multicarrier waveform candidates for 5G,” in IEEE DSP Workshop and Education, Snowbird, Salt Lake City, Aug. 2015 2015.
[67] A. L. RezazadehReyhani and B. Farhang-Boroujeny, “Asynchronous performance of circularly pulse-shaped waveforms for 5G,” in
Communications (ICC), 2016 IEEE International Conference on, submitted. Also, available at arXiv:1511.07910, 2016.
[68] F. Sjoberg, R. Nilsson, M. Isaksson, P. Odling, and P. Borjesson, “Asynchronous zipper,” Proc. of the IEEE ICC ’99, pp. 231–235
vol.1, 1999.
 31

[69] D. Mestdagh, M. Isaksson, and P. Odling, “Zipper VDSL: a solution for robust duplex communication over telephone lines,”
Communications Magazine, IEEE, vol. 38, no. 5, pp. 90–96, 2000.
[70] M. Nava and G. Okvist, “The Zipper prototype: a complete and flexible VDSL multicarrier solution,” Communications Magazine,
IEEE, vol. 40, no. 12, pp. 92–105, 2002.
[71] R. Nilsson, F. Sjoberg, M. Isaksson, J. Cioffi, and S. Wilson, “Autonomous synchronization of a DMT-VDSL system in unbundled
networks,” Selected Areas in Communications, IEEE Journal on, vol. 20, no. 5, pp. 1055–1063, 2002.
[72] “Very-high bit-rate digital subscriber lines (vdsl) metallic interface, part 3: Technical specification of a multi-carrier modulation
transceiver,” Working Group R1E1.4, Savannah, GA, Document T1E1.4/2000-013R3, Nov. 13–17, 2000.