Research ArticleEvaluation of 5G Modulation CandidatesWCP-COQAM GFDM-OQAM and FBMC-OQAM inLow-Band Highly Dispersive Wireless Channels

Aitor Lizeaga1 Pedro M Rodriacuteguez2 Intildeaki Val2 andMikel Mendicute1

1Signal Theory and Communications Area Mondragon University Mondragon Spain2Department of Communications IK4-IKERLAN Mondragon Spain

Correspondence should be addressed to Aitor Lizeaga alizeagamondragonedu

Received 16 December 2016 Revised 17 March 2017 Accepted 15 August 2017 Published 16 October 2017

Academic Editor Lei Zhang

Copyright copy 2017 Aitor Lizeaga et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

We analyse some of the candidates for modulations for 5G FBMC-OQAM GFDM-OQAM and WCP-COQAM Unlike most ofthe related bibliographies which are oriented to mobile communications our research is focused on 5G in cognitive radio basedindustrial wireless communications According to the ultrareliability and low-latency requirements of industrial communicationswe simulate the aforementioned modulations in low-band transmissions (carrier frequencies below 6GHz and a bandwidthnarrower than 100MHz) through large indoor spaces and severe multipath channels that emulate industrial halls Moreover wegive detailed information aboutWCP-COQAM and how the windowing affects the protection against multipath effect and reducesspectral efficiency compared to GFDM-OQAMWe also compare the aforementioned filtered multicarrier techniques and OFDMin terms of robustness against multipath channels power spectral density and spectral efficiency Based on these results we aim atproviding an approximate idea about the suitability of 5G MCM candidates for industrial wireless communications based on CR

1 Introduction

Multicarriermodulation (MCM) schemes arewidely adoptedin wireless communication systems orthogonal frequencydivision multiplexing (OFDM) being the most extended oneamong them Despite the features and benefits that OFDMoffers in wireless communications it also presents signifi-cant drawbacks such as high out-of-band (OOB) radiationConsidering this fact and the upcoming applications andneeds in wireless communications a lot of research on newwaveforms for the physical (PHY) communication layer ofthe fifth generation (5G) has been carried out in order to over-come OFDMrsquos limitations [1ndash4] As a result several pulse-shaping techniques exist which allow subcarrier filtering inMCM systems These techniques provide good frequencylocalization and reduce the OOB radiation Among the mainPHY candidates for 5G the most extended MCM schemesin the bibliography are filter-bank multicarrier offset quadra-ture amplitude modulation (FBMC-OQAM) [5] generalizedfrequency division multiplexing (GFDM) [6] and universal

filteredmulticarrier (UFMC) [7] FBMC-OQAM andGFDMconsist in filtering each subcarrier in order to get goodfrequency localization in each subcarrier UFMC on thecontrary consists in filtering groups of subcarriers in orderto reduce only the OOB radiation

Although most of the research carried out on PHYproposals for 5G is oriented to mobile communications 5Gis actually meant to cover a wide range of application fields[8] In order to cope with the diversity of services and theflexibility that 5G will demand in [9 10] innovative frame-works allowing multiservice subband filtered multicarriersystems are introducedThese frameworks present a structurein which the available spectrum is divided into subbandsdedicated to different services The division and interferenceavoidance between different services is provided by means ofsubband filtering

Among all the application fields that 5G will coverour research focuses on wireless industrial communications[11] Specifically we investigate robust PHY layers designedfor industrial wireless communications based on cognitive

HindawiJournal of Computer Networks and CommunicationsVolume 2017 Article ID 2398701 11 pageshttpsdoiorg10115520172398701

2 Journal of Computer Networks and Communications

radio (CR) [12 13] Since the existing industrial wirelesscommunication systems based on the IEEE 802154 standardlack robustness against these harsh channel conditions [1415] new schemes for the PHY layer in industrial wirelesscommunications must be proposed This kind of industrialscenarios is characterized by the presence of high impulsivenoise and interference fromother wireless systems and fadingchannels due to severemultipath effect and shadowing causedby the numerous large and metallic obstacles within thepropagation channel In this work we simulate transmissionsthrough large indoor spaces and severemultipath channels inorder to model some of these characteristics

Besides the aforementioned harsh environments require-ments for industrial communications suppose an extra chal-lenge for wireless systems especially in some factory automa-tion (FA) applications Since 5G is also meant to cover theseindustrial communicationsrsquo ultrareliability and low-latencyrequirements (up to 10minus9 packet error rates near 100 120583s over-the-air transmission time andle1ms end-to-end latency) [16ndash18] in this work we simulate and evaluate some of the MCM5Gcandidates in low-band transmissions (carrier frequenciesbelow 6GHz and a bandwidth narrower than 100MHz)

In this article among the aforementioned industrialenvironment characteristics we focus on severe multipatheffect We analyse FBMC-OQAM GFDM-OQAM and alsowindowed cyclic prefix-circular OQAM (WCP-COQAM)modulation systems AlthoughWCP-COQAM [19ndash21] is notas extended as the other two schemes in the bibliography itshares several similarities withGFDMwhichmake it anothercandidate for 5G [22] and an interesting MCM scheme tobe analysed We evaluate their performance in terms of biterror rate (BER) power spectral density (PSD) and spectralefficiency (SE) under highly dispersive channelsWe comparethese results with traditional OFDM and we use the SE torepresent the cost of using cyclic prefix (CP) extensionssubcarrier filtering and windowing schemes in each MCMsystem

Based on those analyses and results we aim at providinga notion about the suitability of 5G MCM candidates forindustrial wireless communications based on CR

From the analysis and the assessment of the aforemen-tioned MCM systems we provide the following contribu-tions

(1) This article complements the work in [19ndash21] bybringing additional details about how windowingaffects the protection against multipath channels andthe SE in WCP-COQAM We state the conditionsto provide full orthogonality in multipath channelsand we show how windowing reduces SE in WCP-COQAM compared to GFDM-OQAM We explainall these details in Sections 23 24 and 3 and providesome results and discussion in Section 43

(2) While most research about 5G is focused on mobilecommunications here we simulate and assess someof the main 5G modulation candidates under differ-ent conditions We simulate low-band transmissionsthrough large indoor spaces and severe multipathchannels in order to model some of the wireless

a bja + bj

T T2 T2

Figure 1 Complex to pure real and imaginary operation

industrial environmentsrsquo characteristics Under theseconditions and by means of BER PSD and SE analy-sis we evaluate the suitability of the aforementionedMCM systems for wireless industrial communica-tions based on CR

The paper is organised as follows in Section 2 we explainthe theoretical background of theMCM systems we are goingto analyse in Section 3 we give more detailed explanationsabout WCP-COQAM and make some clarifications withrespect to the bibliography in Section 4 we present theresults obtained from our simulations based on which wecompare and discuss the performances of the consideredMCM systems in Section 5 we state the conclusions weobtained from our research finally in Section 6 we explainwhich will be the next steps in our investigation

2 Multicarrier Modulations

The MCM systems we analyse are based on OQAM insteadof classical IQ modulations (eg N-PSK and N-QAM) Thereason why we focus on OQAM-based schemes is that MCMsystems cannot simultaneously keep good time-frequencylocalization (TFL) Nyquist symbol rate and orthogonalitybetween transmitted symbols if conventional IQmodulationsare used as it is stated by the Balian-LowTheorem (BLT) [23]According to BLT intersymbol interference (ISI) and inter-carrier interference (ICI) introduced by the prototype filtersof these MCM systems make it impossible to achieve perfectreconstruction of complex valued symbols transmitted atNyquist rate Consequently in case matched filter receiversare used FBMC and GFDM perform worse than OFDMin terms of BER One solution to overcome the constraintsdefined by the BLT is to send alternately real and imaginaryvalued symbols at twice the Nyquist rate instead of complexsymbolsThis technique is known asOQAM[5]This strategyallows the filtered data symbols to remain orthogonal andkeep both good TFL and Nyquist symbol rate all at the sametime

Hence we first introduce the OQAM technique and nextwe explain the rest of the MCM schemes

21 OQAM OQAM consists basically in splitting onecomplex symbol into two semisymbols one real and oneimaginary After this operation the whole OQAM symbolrsquosduration remains equal to the duration of the original com-plex symbol while the duration of each semisymbol is halfthe whole symbol as shown in Figure 1Thus OQAMsymbolstructurersquos sampling rate is twice the original complex symbolstructurersquos It is worth mentioning that this oversamplingdoes not carry any change in the total bandwidth of the

Journal of Computer Networks and Communications 3

Re

Im j

zminus1 + x(k)

c(l)

c(l) 2

2

Re

Im j zminus1

+ x(k)

k even

k odd

2

2

Figure 2 Block diagram of the OQAM symbols arrangement in amulticarrier system For frequency domain alternation one sampledelay is applied to imaginary semisymbols in even subcarriers andto real semisymbols in odd subcarriers

Re

Im j

zminus1

zminus1

+2

2

Re

Im j

+

k even

k odd

2

2

x(k)

x(k)

c(l)

c(l)

Figure 3 Block diagram of the OQAM demodulation process in amulticarrier systemThis diagram performs the inverse operation tothe one shown in Figure 2

transmitted signal when OQAM is used in MCM systemsAs we explain in Section 22 when the multicarrier OQAMsignal is serialized its symbol rate is the same as it would bein a conventional IQ modulation

The OQAM scheme discussed up to now can reduce ISIcaused by pulse-shaping processes Since imaginary and realsemisymbols are transmitted alternately interference comingfrom adjacent semisymbols can be ignored in the receiver(ie interference caused by a real semisymbol to its adjacentimaginary semisymbol can be ignored and vice versa)

Similarly in frequency domain pulse-shaping makesadjacent subcarriers overlap each other thus causing severeICI If those subcarriers are organised so that real andimaginary ones are alternated one next to the other ICI fromadjacent subcarriers will be avoided because interferencecoming from them can be ignored in the receiver Figure 2shows the block diagram of an OQAM MCM systemFigure 3 on the other hand shows the demodulation processto recover the original complex data symbols

More details about OQAM can be found in [5]

22 FBMC-OQAM FBMC consists in filtering each sub-channel in order to get well-localized subcarriers Figure 4shows the basic representation of a FBMC-OQAM transmul-tiplexer The main blocks are the OQAM preprocessing ormodulator block the synthesis filter bank the analysis filterbank and OQAM postprocessing or demodulator

The key components of the FBMC-OQAM scheme arethe synthesis and analysis filter banks which are formed by

one filter per subchannel The filter of the 119898th subchannel isdefined as

119892119898 (119899) = 119901 (119899) exp(1198952120587119898119899119872 ) (1)

where 119901(119899) is the prototype filter of length 119871119901 and 119899 =0 119871119901 minus 1 The exponential factor corresponds to the119898thsubcarrier where119872 is the total number of subcarriers

Note that the FBMC-OQAM schemeworks as amultiratesystem Before filtering the 119909119898(119896)OQAM data symbols theyare upsampled by 1198722 using a zero padding process to getthe oversampled signal 119886119898(119899)

119886119898 (119899) = sum119896isinZ

119909119898 (119896) 120575 (119899 minus 1198961198722 ) 119899 isin Z (2)

The reason why the upsampling factor is1198722 and not119872 isthat OQAMpreprocessing already introduces an upsamplingfactor of 2 Thus the FBMC-OQAM signal can be expressedas

119910 (119899) =119872minus1

sum119898=0

sum119897isinZ

119886119898 (119897) 119892119898 (119899 minus 119897) 119899 isin Z (3)

At the receiver side the FBMC-OQAM signal 119910(119899) isdemodulated so that in first place each subchannel is filteredin order to get only its corresponding signal

119886119898 (119899) = sum119897isinZ

119910 (119897) 119892119898 (119899 minus 119897) 119899 isin Z (4)

where119898 = 0 119872 minus 1Then 119886119898(119899) is sampled in the time domain so that the

estimation of the original OQAM signal 119909119898(119896) is obtained119909119898 (119896) = 119886119898 (1198961198722 ) 119896 isin Z (5)

Transmission channel noise equalization and otherissues have been omitted in order to focus just on FBMC-OQAM waveform modulation and demodulation Moredetails about FBMC-OQAM can be found in [5]

23 GFDM-OQAM The structure of a GFDM-OQAMtransmultiplexer is similar to that of the FBMC-OQAMtransmultiplexer shown in Figure 4 GFDM-OQAM is alsobased on OQAM modulation and subchannel filtering Thedifference is that inGFDM-OQAMfiltering is performed by acircular convolution instead of the linear convolution used inFBMC-OQAMThis way the modulation system now adoptsa block-based signal structure so that a CP can be addedto provide orthogonality against multipath channels withoutaffecting signalrsquos TFL

We consider a GFDM-OQAM system with119872 subchan-nels and 119870 complex valued data symbols in each block (ie2119870OQAM symbols in each block) Thus if we consider 119910(119899)as one GFDM-OQAM block it can be expressed as

119910 (119899) =119872minus1

sum119898=0

119872119870minus1

sum119897=0

119886119898 (119897) 119892119898 (119899 minus 119897 + 119872119870 minus 1) (6)

4 Journal of Computer Networks and Communications

c0(l)

c1(l)

cMminus1(l)

x0(k)

x1(k)

xMminus1(k)

a0(n)

a1(n)

aMminus1(n)

g0(n)

g1(n)

gMminus1(n)

g0(n)

g1(n)

gMminus1(n)

M2

M2

M2

M2

M2

M2OQ

AM

pos

tpro

cess

ing

OQ

AM

pos

tpro

cess

ing

+

y(n)

c0(l)

c1(l)

cMminus1(l)

x0(k)

x1(k)

xMminus1(k)

a0(n)

a1(n)

aMminus1(n)

Figure 4 Basic representation of a FBMC-OQAM transmultiplexer This scheme shows the basic idea behind FBMC-OQAM where eachsymbol of each subchannel is modulated as OQAM multiplied by its corresponding subcarrier and filtered by the prototype filter In thereceiver the inverse process is performed

where 119899 = 0 119872119870 minus 1 and 119892119898 filters are obtained by theperiodic repetition of 119892119898 filters defined in (1) So 119892119898(119896) =119892119898[mod(119896119872119870)]

At the receiver side the GFDM-OQAM signal 119910(119899) isdemodulated so that in first place each subchannel is filteredin order to get only its corresponding signal Once againthe difference between FBMC-OQAM and GFDM-OQAMreceivers consists in the circular convolution

119886119898 (119899) =119872119870minus1

sum119897=0

119910 (119897) 119892119898 (119899 minus 119897 + 119872119870 minus 1) (7)

where119898 = 0 119872 minus 1Then 119886119898(119899) is sampled in the time domain so that the

estimation of the original OQAM signal 119909119898(119896) is obtainedjust as it is shown in (5)

The insertion of a CP transmission channel noiseequalization and other issues have been omitted in orderto focus just on GFDM-OQAM waveform modulation anddemodulation More details about GFDM-OQAM can befound in [19 20] In those works the idea of GFDM-OQAMis introduced although the authors refer to this scheme asFBMC-COQAM or just as COQAM

24 WCP-COQAM WCP-COQAM is introduced in [21]This technique can be considered as a GFDM-OQAM systemwith a windowing process which improves the PSD withrespect to GFDM-OQAM

In WCP-COQAM both the CP and the windowing arerelated The CP is divided into two parts the guard interval(GI) and the window interval (WI) (referred to as RI in [21])So the total length of the CP will be equal to the sum of thelengths of the GI and WI parts

119871CP = 119871GI + 119871WI (8)

where GI is the part of the CP aimed at protecting againstmultipath channels and WI is the part that will be used forwindowing

Let us consider a GFDM-OQAM signal 119904GFDM-OQAM asa queue of several GFDM-OQAM blocks each with its CPextension of length 119871CP Equally we will consider 119872 to bethe total number of subcarriers and 119870 to be the number of

symbols in each blockThen the 119897th block of aWCP-COQAMsignal is defined as

119904WCP-COQAM (119896)

=119897+1

sum119903=119897minus1

119904WCP-COQAM (119896 + 119903119871WI) 119908 (119896 minus 119903119876) (9)

where 119896 = 0 119872119870 + 119871CP minus 1 119876 = 119872119870 + 119871GI and thewindow function 119908(119896) is defined by

119908 (119896)

=

window coeff 119896 = 0 119871WI minus 11 119896 = 119871WI 119872119870 + 119871GI minus 1window coeff 119896 = 119876 119872119870 + 119871CP minus 10 otherwise

(10)

In our work we use a Hamming window for the windowcoefficients Figure 5 shows the block diagram that representsthis whole process

It must be noted from (9) and (10) that the samples ofthe CP corresponding to the WI part and also the last 119871WIsamples of the block aremultiplied by thewindow coefficientsand they are overlapped with the WI regions of the previousand next blocks respectively By overlapping the WI regionsthese extra samples are prevented from reducing SE as shownin Figure 6

The receiver structure is depicted in Figure 7 It is worthnoticing that Figure 7 only shows the recovery process ofone GFDM-OQAM block from a received WCP-COQAMsignal Once the GFDM-OQAM signal is recovered there areseveral algorithms for its demodulation In Section 23 of thispaper the basic demodulation scheme for GFDM-OQAM isexplained and in [21] the authors introduce a computationallyefficient demodulation scheme

In [21] for this scheme perfect synchronization and anideal channel are assumed So in the first place the first119872119870 + 119871GI samples of one WCP-COQAM block are takenNext the first 119871GI (which correspond to the CP extension)samples are removed while the last 119871WI samples are left toprocess them within the next block Finally in order to getall the samples back to their original positions within the

Journal of Computer Networks and Communications 5

s＄－-１－(k) SP

CPinsertion

times

times

times

w(0)

w(1)

w(MK+ L０ minus 1)

Q

Q

Q +

+

zminus1

zminus1

zminus1

s７０-C１－(k)

Figure 5 Block diagram of the cyclic prefix insertion and windowing operation for WCP-COQAM A WCP-COQAM signal is made froma GFDM-OQAM signal and a CP and a window function are applied

WI

CP MK

Block l

Block l minus 1 Block l + 1

GI M ７） ）

middot middot middot middot middot middot

Figure 6 Overlapping of adjacent WCP-COQAM blocks The regions affected by the window of one block are overlapped with the regionsaffected by the window of the previous and the next blocks These regions correspond to the first and the last 119871WI samples of each block

original GFDM-OQAM block a cyclic shift is carried outwhich can be expressed as

1199041015840GFDM-OQAM (119896) = 119904 [mod (119896 + 119871WI119872119870)] (11)

where 119896 = 0 119872119870minus 1 1199041015840GFDM-OQAM(119896) is the estimation ofthe original 119904WCP-COQAM(119896) signal and 119904(119896) is the signal afterthe removal of the windowing effect and the CP

3 Orthogonality Condition and SpectralEfficiency Analysis of WCP-COQAM

Several analyses on 5Gmodulation candidates are available inthe bibliography [24 25] some of them even includingWCP-COQAM modulation scheme [26 27] However to the bestof our knowledge none of them address the implications ofwindowing in the WCP-COQAM waveform as we do here

In this section we contribute some additional detailsregarding how windowing affects the orthogonality againstmultipath channels and the SE in WCP-COQAM Specifi-cally we state the condition that the GI region of the CPextension must fulfil in order to provide full orthogonality inthe presence of multipath channels Besides we also compareWCP-COQAM and GFDM-OQAM (due to the similaritiesbetween both MCM systems) in terms of SE and we showthat in this aspect the former is outperformed by the latter ifequal multipath protection is assumed

31 Orthogonality Condition In WCP-COQAM multipathchannel protection is directly related to the CP extensionand also to the windowing process In order to explain thisfact we show in detail the steps to form a WCP-COQAMblock starting from a basic GFDM-OQAM block (as seen inSection 23)

Once a GFDM-OQAMblock is formed the CP extensionis added to it Figure 8(a) shows the details of this first stagein the construction of WCP-COQAM signal where119872 is thetotal number of subcarriers 119870 is the number of symbols perblock WI is the region of length 119871WI devoted to windowingand overlapping GI is the region of length 119871GI that protectsagainst multipath channel effect and the CP extension hasa total length of 119871CP = 119871WI + 119871GI CP

1015840 WI1015840 and GI1015840 arethose last samples of the block which are copied in orderto form the CP extension As is represented in Figure 8(a)before the windowing is applied this signal can be consideredas a GFDM-OQAM block with CP extension

Figure 8(b) represents the WCP-COQAM block rightafter windowing is applied As shown in Figure 6 and in (9)and (10) the window function is applied on the first andthe last 119871WI samples so that these samples are somehowdistorted Because of that CPWI and GI are no longer equalto CP1015840 WI1015840 and GI1015840 respectively

Since the last 119871WI samples are affected by the window wedefine a new region for them which we call GI21015840 Thus now

6 Journal of Computer Networks and Communications

SP

CP amp lastL２）

symbolsremoval

s７０-C１－(k)

s (0)

s(L） minus 1)

s(L）)

s(MK + L） minus 1)

s(MK + L）)

s(MK + L０ minus 1)

Cyclicshift

s＄－-１－(0)

s＄－-１－(1)

s＄－-１－(MK minus 1)

Figure 7 Block diagram of one GFDM-OQAM block recovery process from a WCP-COQAM block

the original GI and GI1015840 regions are divided into GI1 and GI2and GI11015840 and GI21015840 respectively so that WI = WI1015840 GI1 =GI11015840 and GI2 = GI21015840 As for the lengths of these regions119871WI = 119871WI1015840 = 119871GI2 = 119871GI21015840 and 119871GI1 = 119871GI11015840 Note that GI1and GI11015840 are now the only really redundant parts Because ofthat only GI1 and not the whole GI (as it is explained in [21])acts as a real guard interval against multipath channels whenfrequency domain equalization (FDE) is performed Hence119871GI1 ge 119871CH minus 1 must be fulfilled in order to maintain fullorthogonality through a multipath channel where 119871CH is thetransmission channel length Thus considering that 119871GI2 =119871WI and so 119871GI1 = 119871GI minus119871GI2 = 119871GI minus119871WI we can define thefull orthogonality condition for WCP-COQAM as

119871GI ge 119871CH minus 1 + 119871WI (12)

It is worth mentioning that the FDE should be performed forsamples between GI2 and GI11015840 inclusive because that is thesection corresponding to the cyclic convolution between GI1and GI11015840 with the transmission channel

32 Spectral Efficiency Analysis Figure 9 shows how over-lapping between adjacent blocks is performed in WCP-COQAM Every windowed region is overlapped with thenext or the previous block so that the WI region of eachblock is overlapped with the GI21015840 region of the previousblock Thus the samples within WI regions do not supposean overload and they do not affect the SE Based on this factin [21] the authors claim that the use of the window function

has no effect on SE However this statement is not preciseAlthough the samples inWI do not directly affect SE becausethey overlap with the samples of GI21015840 of the previous blockwindowing implies adding theGI2 region in order to fulfil theorthogonality condition defined in (12) Note that the use ofthe window function and overlapping ofWI andGI21015840 regionsdistort those samples in GI21015840 region This and the fact thatFDE is performed between GI2 and GI11015840 inclusive implythat the GI2 region is necessary in order to recover thosesamples originally placed in the GI21015840 region At this pointwe conclude that both GI1 and GI2 are essential parts withinthe GI region of WCP-COQAM

In order to make a fair analysis of SE in WCP-COQAMwe compare it to GFDM-OQAM because both modula-tion schemes only differ in windowing and overlappingprocesses In order to maintain a homogeneous notationbetween GFDM-OQAM andWCP-COQAM we assume thefollowing considerations (for the sake of readability fromthis point on parameters corresponding to OFDM will beexpressed with the subindex 119874 parameters correspondingto GFDM-OQAM will be expressed with the subindex 119866and parameters corresponding to WCP-COQAM will beexpressed with the subindex119882)

CP119866 = GI119866 = GI1119866CP119882 = [WI119882GI119882] GI119882 = [GI1119882GI2119882]

(13)

Journal of Computer Networks and Communications 7

WI

CP

(GFDM-OQAM block with CP extension)

GI M

MK

(GFDM-OQAM block without CP extension)

Block l

CP

７） ）

(a)

WI GI2

Block l

GI1 ７） ）1 ）2

(b)

Figure 8Windowing operation of aWCP-COQAM block (a) AWCP-COQAM block before windowing is equal to a GFDM-OQAM blockwith CP extension (b) The first and the last 119871WI samples of a WCP-COQAM block are distorted by the window

WI GI2

(block l minus 1)

Block l

(block l)GI1

WI GI2(block l + 1)

GI1

(block l)

Block l minus 1 Block l + 1

７）

middot middot middot middot middot middot

）1 ）2

７） ）1 ）2

Figure 9 Overlapping between adjacent WCP-COQAM blocks and CP extensionrsquos structure Only GI1 and GI11015840 remain equal in order toemulate a circular convolution with the channel and perform frequency domain equalization so GI2 implies an overload that affects SE

The criterion we applied for this comparison is to provideequal protection against multipath channels for both modu-lations so that 119871GI1119866 = 119871GI1119882 Thus

SE119866 = 119872119870119871GI119866 +119872119870 = 119872119870

119871GI1119866 +119872119870

SE119882 = 119872119870119871GI119882 +119872119870 = 119872119870

119871GI1119882 + 119871GI2119882 +119872119870(14)

From this analysis it is clear that if full orthogonalityis to be maintained the windowing process brings an extraGI2 region to the CP which has effects on the SE of a WCP-COQAM signal In particular SE is reduced by a factor of(119871GI1 + 119871GI2 +119872119870)(119871GI1 +119872119870) compared to a windowlesssystem with equal multipath channel protection

4 Simulation Results and Discussion

In this section we show and discuss the results obtainedfrom the simulations of the systems previously describedBased on these results we assess the performances of thesimulated systems in terms of BER PSD and SE In ouranalysis we prioritize robustness against highly dispersivechannels so that we use the BER of transmissions throughRayleigh fading multipath channels in order to assess therobustness of the MCM systems In addition to BER we alsotake into consideration the PSD of the transmitted signalsfor it is an important factor when it comes to CR systemsRegarding SE we use it to represent the cost of the CPextensions windowing process and convolution tails causedby the prototype filters in FBMC-OQAM so that we canobtain a fair comparison between different MCM systems

8 Journal of Computer Networks and Communications

Table 1 Basic simulation parameters Unless opposite is stated the results shown in this work have been obtained under the conditionsdefined by these parameters

OFDM FBMC-OQAM GFDM-OQAM WCP-COQAMSignal bandwidth 20MHzIQ modulation QPSK119872 (subcarriers) 512119870 (symbolsblock overlapping factor) 1 4 4 4119871CP (samples) 32 mdash 32 224119871WI (samples) mdash mdash mdash 96119871GI (samples) 32 mdash 32 128119871GI1 (samples) mdash mdash mdash 32119871GI2 (samples) mdash mdash mdash 96Prototype filter mdash PHYDYAS [5]119871119901 (samples) mdash 2048

Since we are evaluating MCM systems we considerit appropriate to compare the analysed FMC modulationtechniques with an OFDM reference system in order toassess their performance Table 1 shows the configurationparameters we used in most of our simulations We chooseparameters suitable for low-band transmissions in ultrareli-able and low-latency indoor scenarios For that we consideraspects like subcarrier bandwidth and transmission channelrsquoscoherence bandwidth data block duration and windowlength for PSD improvement

41 Robustness against Multipath Channels In this subsec-tion we analyse the robustness of each MCM system againstmultipath channels In order to make a fair comparisonwe simulate every system with equal protection againstmultipath channels matching effective GI lengths (119871GI119874 119871GI119866 and 119871GI1119882) These parameters are selected as shown inTable 1

The equalization technique we use in our simulations isone-tap zero-forcing (ZF) FDE for all the MCM systemsWhile OFDM GFDM-OQAM and WCP-COQAM employCP extensions to prevent ISI caused by the transmissionchannel FBMC-OQAM could be considered a special casein this aspect Since it is not a blockwise modulation andno CP extension is used one-tap FDE is not a straightfor-ward equalization technique for FBMC-OQAM Howeverdue to the subcarrier bandwidth and symbol duration nearto transmission channelrsquos coherence bandwidth and timerespectively we consider the one-tap FDE as a suitableequalization method for FBMC-OQAM in this case An in-depth analysis about the doubly dispersive channelsrsquo impacton FBMC systems is given in [28]

We simulated transmissions through a Rayleigh fadingchannel model with exponential power delay profile (PDP)whose root mean square (rms) delay spread and channellength are 119905rms = 185 ns and 119871CH = 22 taps Besides wesimulate uncoded and coded communications with turbocodes soft decoding and a code rate of 13 In Figure 10

BER

100

10minus1

10minus2

10minus3

10minus4

10minus5

10minus6

0 5 10 15 20 25 30 35

OFDM (uncoded)QFDM (coded)FBMC-OQAM (coded) WCP-COQAM (coded)GFDM-OQAM (coded)

FBMC-OQAM (uncoded)WCP-COQAM (uncoded)GFDM-OQAM (uncoded)

EbN0 (dB)

Figure 10 BER curves of the analysedMCM systems in a multipathchannel We employed turbo encodingdecoding with soft decodingand a code rate of 13

we compare the BER versus 1198641198871198730 of the MCM systemsover these channelmodels assuming perfect synchronizationand full channel state information (CSI) We got these BERcurves by simulating transmissions through 104Monte Carloiterations for each MCM system and each channel modelWe transmit 12288 random data bits and we generate newrandom channels at each Monte Carlo iteration

Here we show that for uncoded transmissions in termsof error rate FBMC-OQAMperforms slightly worse than therest of MCM systems under the simulated conditions Whileat low 1198641198871198730 values every MCM system provides similarBER an error floor appears at high 1198641198871198730 values because ofthe lack of GI in FBMC-OQAMOn the other hand although

Journal of Computer Networks and Communications 9

Frequency normalised to subcarrier number

minus60

minus50

minus40

minus30

minus20

minus10

0

10

Mag

nitu

de (d

B)

minus30 minus20 minus10 0 10 20 30

OFDMFBMC-OQAM

WCP-COQAMGFDM-OQAM

Figure 11 PSD of the analysedMCM systems with some subcarriersdeactivated Only the central 32 subcarriers out of 512 are activated

OFDM GFDM-OQAM andWCP-COQAMprovide similarperformances in terms of BER providing equal multipatheffect protection OFDM slightly outperforms the other twoMCM schemes due to its perfect orthogonality

Finally it is worth noting that for coded transmissionsall the simulated MCM systems present similar BER curvesso that they provide similar robustness against multipathchannels

42 Power Spectral Density Analysis PSD is a relevant char-acteristic in order to design a CR communication systemsuitable for scenarios with high spectrum occupation In thissubsection we compare the spectrums of the signals of theanalysed MCM systems in order to assess their potentialsuitability for CR based systems

Figure 11 shows the spectrums of OFDM GFDM-OQAM WCP-COQAM and FBMC-OQAM signals one ontop of the other For this test we only activate the 32 centralsubcarriers in order to make the power leakage into adjacentsubcarriers visible

As shown in Figure 11 the FB based MCM systems carrysignificant reduction of OOB radiation compared to OFDMHowever even between these MCM schemes there is stilla remarkable difference In this sense FBMC-OQAM out-performs both GFDM-OQAM and WCP-COQAM On theother hand thanks to the windowing WCP-COQAM showsconsiderably lower OOB radiation than GFDM-OQAM Itis worth mentioning that the longer the WI region is thenarrower WCP-COQAM signalrsquos spectrum gets Since extrasamples within the GI region are needed for windowing asdescribed in Section 3 PSD improvement of WCP-COQAMover GFDM-OQAM comes at the cost of lower SE Moredetailed analysis on SE is given in Section 43

43 Spectral Efficiency Analysis In this subsection we anal-yse SE in order to evaluate the cost of obtaining higherrobustness against dispersive channels and better PSD bymeans of CP extensions and windowing For this analysis weconsider the parameters defined in Table 1 so that effectivemultipath channel protection is equal for OFDM GFDM-OQAM and WCP-COQAM Equations (15) show the SEvalues of the three blockwise MCM systems

SE119874 = 119872119870119874119871GI119874 +119872119870119874 = 09412

SE119866 = 119872119870119866119871GI119866 +119872119870119866 = 09846

SE119882 = 119872119870119882119871GI119882 +119872119870119882 = 09412

(15)

Here we show a particular case with some specific parame-ters On one hand GFDM-OQAM presents higher SE thanOFDM because its GI precedes 119870 symbols while in OFDMthe GI precedes only one symbol On the other hand becauseof the windowing operation the CP extension of WCP-COQAM specifically the GI part has 119871GI2 extra sampleswith respect to GFDM-OQAM and that is the reason whyWCP-COQAM also presents lower SE than GFDM-OQAMHowever under these conditions with a high number ofsubcarriers we can conclude that the differences in the SEbetweenOFDMGFDM-OQAM andWCP-COQAMare notreally significant

Regarding FBMC-OQAM once again we consider itdifferent from the rest of MCM systems because it is nota blockwise modulation and it uses no GI protection formultipath effect So its SE depends on the length of thetransmitted frames with respect to the convolution tailsFor long data transmissions its SE will tend to 1 while forshort data transmissions it should tend to 05 assuming aminimum amount of data samples equivalent to a GFDM-OQAM orWCP-COQAM blockThis calculation is simple ifwe introduce 4 complex data symbols (8 OQAM symbols) in(2)

119886119898 (119899) =7

sum119896=0

119909119898 (119896) 120575 (119899 minus 1198961198722 ) 119899 isin [0 81198722 minus 1] (16)

and we introduce these upsampled data in (3) The resultant119910(119899) signal will contain 4095 samples of which half willcorrespond to actual data and the rest to convolution tailsfrom subcarrier filtering

Considering the low-latency scenarios our researchfocuses on SE might suppose a significant drawback forFBMC-OQAM compared to GFDM-OQAM and WCP-COQAM during short data transmissions

5 Conclusions

In this article we make a comparison between three candi-dates for PHY layer modulation in 5G and a reference OFDM

10 Journal of Computer Networks and Communications

system in order to assess their suitability for industrial wire-less communications based on CR For that we simulate low-band communications and highly dispersive indoor channelscenarios Under these conditions we show the performancesand features of these MCM systems from different points ofview

Regarding robustness against multipath channels weconclude that all the analysed MCM systems provide similarperformances under the simulated conditions Although foruncoded transmissions FBMC-OQAM presents an errorfloor at high 1198641198871198730 values this error floor is corrected whencoding is used

In terms of PSD and OOB radiation these filtered MCMschemes provide more restrained spectrum than OFDM Inparticular FBMC-OQAM outperforms the rest of the modu-lation schemes in this aspect so it could be a suitable candi-date for the CR scenarios considered in our research On theother hand although not as good as FBMC-OQAM WCP-COQAM still has considerably better PSD than GFDM-OQAM which only provides less than 10 dB of differencein OOB radiation with respect to OFDM Therefore weconclude that circular filtering only is not sufficient andwindowing must be applied in order to ensure more efficientuse of the spectrum Moreover we consider that due to itsalso restrained spectrum WCP-COQAM might be anothersuitable modulation scheme for CR applications

As for the SE analysis as discussed in Section 43the difference between OFDM GFDM-OQAM and WCP-COQAM is not really significant for the parameters weused in our simulations On the other hand FBMC-OQAMmight suffer from severe degradation of SE during short datatransmissions in the low-latency scenarios we consider in ourresearch

6 Future Work

In this work we considered perfect synchronization andfull channel state information at the receiver but it is alsoimportant to consider more realistic simulation conditionsin the future So the next step in our investigation will beto analyse and propose time-frequency synchronization andchannel estimation methods for GFDM-OQAM and WCP-COQAM

The other key point in our research will be to simulatechannelmodels that fairly emulate industrial wireless channelconditions For that we will simulate severe multipath con-ditions time-varying channels impulsive noise and interfer-ence from other wireless systems

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

This work was partly supported by TIPOTRANS(ETORTEK) and CIIRCOS (PC2013-68) projects ofthe Basque Government (Spain) and by COWITRACC

(TEC2014-59490-C2-1-P2-P) project of the Spanish Minis-try of Economy and Competitiveness

References

[1] H Zhang D L Ruyet M Terre et al ldquoApplication of theFBMC physical layer in a Cognitive Radio scenariordquo Tech RepPHYDYAS-Physical Layer for Dynamic Access and CognitiveRadio 2009

[2] I Gaspar and GWunder ldquo5G cellular communications scenar-ios and system requirementsrdquo Tech Rep 5th Generation Non-Orthogonal Waveforms for Asynchronous Signalling 2013

[3] P Popovski V Braun H Mayer et al ldquoScenarios requirementsandKPIs for 5Gmobile andwireless systemrdquo Tech RepMobileand wireless communications Enablers for the Twenty-twentyInformation Society (METIS) 2013

[4] X Mestre D Gregoratti M Renfors et al ldquoICT-EMPhAtiCfinal project reportrdquo Tech Rep Enhanced Multicarrier Tech-niques for Professional Ad-Hoc and Cell-Based Communica-tions 2015

[5] A Viholainen M Bellanger and M Huchard ldquoPrototypefilter and structure optimizationrdquo PHYDYAS-Physical Layer forDynamic Access and Cognitive Radio 2009

[6] N Michailow M Matthe I S Gaspar et al ldquoGeneralizedfrequency division multiplexing for 5th generation cellularnetworksrdquo IEEE Transactions on Communications vol 62 no9 pp 3045ndash3061 2014

[7] V Vakilian T Wild F Schaich S Ten Brink and J-F FrigonldquoUniversal-filtered multi-carrier technique for wireless systemsbeyond LTErdquo in Proceedings of the 2013 IEEE Globecom Work-shops GC Wkshps 2013 pp 223ndash228 IEEE December 2013

[8] M Agiwal A Roy and N Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 18 no 3 pp 1617ndash1655 2016

[9] X Zhang M Jia L Chen J Ma and J Qiu ldquoFiltered-OFDM-enabler for flexible waveform in the 5th generation cellular net-worksrdquo in Proceedings of the 58th IEEE Global CommunicationsConference (GLOBECOM rsquo15) pp 1ndash6 IEEE December 2015

[10] L Zhang A Ijaz P Xiao A Quddus and R Tafazolli ldquoSubbandfiltered multi-carrier systems for multi-service wireless com-municationsrdquo IEEE Transactions on Wireless Communicationsvol 16 no 3 pp 1893ndash1907 2017

[11] A Lizeaga M Mendicute P M Rodriguez and I Val ldquoEval-uation of WCP-COQAM GFDM-OQAM and FBMC-OQAMfor industrial wireless communications with Cognitive Radiordquoin Proceedings of the 2017 IEEE International Workshop ofElectronics Control Measurement Signals and their Applicationto Mechatronics (ECMSM) pp 1ndash6 Donostia Spain May 2017

[12] P M Rodriguez I Val A Lizeaga and M MendicuteldquoEvaluation of cognitive radio for mission-critical and time-critical WSAN in industrial environments under interferencerdquoin Proceedings of the 2015 11th IEEEWorld Conference on FactoryCommunication Systems (WFCS rsquo15) pp 1ndash4 May 2015

[13] P M Rodrıguez R Torrego F Casado et al ldquoDynamicspectrumaccess integrated in awideband cognitiveRF-ethernetbridge for industrial control applicationsrdquo Journal of SignalProcessing Systems vol 83 no 1 pp 19ndash28 2016

[14] A K Somappa K Ovsthus and L Kristensen ldquoAn industrialperspective on wireless sensor networks - a survey of require-ments protocols and challengesrdquo IEEE Communications Sur-veys Tutorials vol 16 pp 1391ndash1412 2014

Journal of Computer Networks and Communications 11

[15] S Vitturi ldquoWireless networks for the factory floor require-ments available technologies research trends and practicalapplicationsrdquo in Proceedings of the IEEE International Confer-ence on Emerging Technologies and Factory Automation 2014

[16] P Popovski ldquoUltra-reliable communication in 5G wirelesssystemsrdquo in Proceedings of the 2014 1st International Conferenceon 5G for Ubiquitous Connectivity (5GU rsquo14) pp 146ndash151November 2014

[17] N Brahmi ON C Yilmaz KWHelmersson S A Ashraf andJ Torsner ldquoDeployment strategies for ultra-reliable and low-latency communication in factory automationrdquo in Proceedingsof the IEEEGlobecomWorkshops (GCWkshps rsquo15) pp 1ndash6 IEEEDecember 2015

[18] O N Yilmaz ldquoUltra-Reliable and Low-Latency 5G Communi-cationrdquo in Proceedings of the European Conference on Networksand Communications (EuCNC rsquo16) 2016

[19] H Lin andP Siohan ldquoFBMCCOQAM an enabler for cognitiveradiordquo in Proceedings of the 4th International Conference onAdvances in Cognitive Radio (COCORA rsquo14) pp 8097ndash8101May2014

[20] H Lin and P Siohan ldquoAn advanced multi-carrier modulationfor future radio systemsrdquo in Proceedings of the 2014 IEEE Inter-national Conference on Acoustics Speech and Signal Processing(ICASSP rsquo14) pp 8097ndash8101 May 2014

[21] H Lin and P Siohan ldquoMulti-carrier modulation analysis andWCP-COQAMproposalrdquoEurasip Journal onAdvances in SignalProcessing vol 2014 no 1 article 79 2014

[22] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations chapter 8 JohnWiley amp Sons Ltd ChichesterUK 2016

[23] J J Benedetto C Heil and D F Walnut Gabor Systems and theBalian-LowTheorem Birkhauser chapter 2 BostonMass USA1998

[24] A Ijaz L Zhang P Xiao and R TafazolliAnalysis of CandidateWaveforms for 5G Cellular Systems chapter 1 InTech 2016

[25] R Gerzaguet N Bartzoudis L G Baltar et al ldquoThe 5Gcandidate waveform race a comparison of complexity andperformancerdquo Eurasip Journal onWireless Communications andNetworking vol 2017 no 1 article 13 2017

[26] A B Ucuncu and A O Yilmaz ldquoPulse shaping meth-ods for OQAMOFDM and WCP-COQAMrdquo httpsarxivorgabs150900977

[27] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations JohnWiley amp Sons Ltd Chichester UK 2016

[28] L Zhang P Xiao A Zafar A ul Quddus and R TafazollildquoFBMC system an insight into doubly dispersive channelimpactrdquo IEEE Transactions on Vehicular Technology vol 66 no5 pp 3942ndash3956 2017

Journal of Computer Networks and Communications 3

Re

Im j

zminus1 + x(k)

c(l)

c(l) 2

2

Re

Im j zminus1

+ x(k)

k even

k odd

2

2

Figure 2 Block diagram of the OQAM symbols arrangement in amulticarrier system For frequency domain alternation one sampledelay is applied to imaginary semisymbols in even subcarriers andto real semisymbols in odd subcarriers

Re

Im j

zminus1

zminus1

+2

2

Re

Im j

+

k even

k odd

2

2

x(k)

x(k)

c(l)

c(l)

Figure 3 Block diagram of the OQAM demodulation process in amulticarrier systemThis diagram performs the inverse operation tothe one shown in Figure 2

transmitted signal when OQAM is used in MCM systemsAs we explain in Section 22 when the multicarrier OQAMsignal is serialized its symbol rate is the same as it would bein a conventional IQ modulation

The OQAM scheme discussed up to now can reduce ISIcaused by pulse-shaping processes Since imaginary and realsemisymbols are transmitted alternately interference comingfrom adjacent semisymbols can be ignored in the receiver(ie interference caused by a real semisymbol to its adjacentimaginary semisymbol can be ignored and vice versa)

Similarly in frequency domain pulse-shaping makesadjacent subcarriers overlap each other thus causing severeICI If those subcarriers are organised so that real andimaginary ones are alternated one next to the other ICI fromadjacent subcarriers will be avoided because interferencecoming from them can be ignored in the receiver Figure 2shows the block diagram of an OQAM MCM systemFigure 3 on the other hand shows the demodulation processto recover the original complex data symbols

More details about OQAM can be found in [5]

22 FBMC-OQAM FBMC consists in filtering each sub-channel in order to get well-localized subcarriers Figure 4shows the basic representation of a FBMC-OQAM transmul-tiplexer The main blocks are the OQAM preprocessing ormodulator block the synthesis filter bank the analysis filterbank and OQAM postprocessing or demodulator

The key components of the FBMC-OQAM scheme arethe synthesis and analysis filter banks which are formed by

one filter per subchannel The filter of the 119898th subchannel isdefined as

119892119898 (119899) = 119901 (119899) exp(1198952120587119898119899119872 ) (1)

where 119901(119899) is the prototype filter of length 119871119901 and 119899 =0 119871119901 minus 1 The exponential factor corresponds to the119898thsubcarrier where119872 is the total number of subcarriers

Note that the FBMC-OQAM schemeworks as amultiratesystem Before filtering the 119909119898(119896)OQAM data symbols theyare upsampled by 1198722 using a zero padding process to getthe oversampled signal 119886119898(119899)

119886119898 (119899) = sum119896isinZ

119909119898 (119896) 120575 (119899 minus 1198961198722 ) 119899 isin Z (2)

The reason why the upsampling factor is1198722 and not119872 isthat OQAMpreprocessing already introduces an upsamplingfactor of 2 Thus the FBMC-OQAM signal can be expressedas

119910 (119899) =119872minus1

sum119898=0

sum119897isinZ

119886119898 (119897) 119892119898 (119899 minus 119897) 119899 isin Z (3)

At the receiver side the FBMC-OQAM signal 119910(119899) isdemodulated so that in first place each subchannel is filteredin order to get only its corresponding signal

119886119898 (119899) = sum119897isinZ

119910 (119897) 119892119898 (119899 minus 119897) 119899 isin Z (4)

where119898 = 0 119872 minus 1Then 119886119898(119899) is sampled in the time domain so that the

estimation of the original OQAM signal 119909119898(119896) is obtained119909119898 (119896) = 119886119898 (1198961198722 ) 119896 isin Z (5)

Transmission channel noise equalization and otherissues have been omitted in order to focus just on FBMC-OQAM waveform modulation and demodulation Moredetails about FBMC-OQAM can be found in [5]

23 GFDM-OQAM The structure of a GFDM-OQAMtransmultiplexer is similar to that of the FBMC-OQAMtransmultiplexer shown in Figure 4 GFDM-OQAM is alsobased on OQAM modulation and subchannel filtering Thedifference is that inGFDM-OQAMfiltering is performed by acircular convolution instead of the linear convolution used inFBMC-OQAMThis way the modulation system now adoptsa block-based signal structure so that a CP can be addedto provide orthogonality against multipath channels withoutaffecting signalrsquos TFL

We consider a GFDM-OQAM system with119872 subchan-nels and 119870 complex valued data symbols in each block (ie2119870OQAM symbols in each block) Thus if we consider 119910(119899)as one GFDM-OQAM block it can be expressed as

119910 (119899) =119872minus1

sum119898=0

119872119870minus1

sum119897=0

119886119898 (119897) 119892119898 (119899 minus 119897 + 119872119870 minus 1) (6)

4 Journal of Computer Networks and Communications

c0(l)

c1(l)

cMminus1(l)

x0(k)

x1(k)

xMminus1(k)

a0(n)

a1(n)

aMminus1(n)

g0(n)

g1(n)

gMminus1(n)

g0(n)

g1(n)

gMminus1(n)

M2

M2

M2

M2

M2

M2OQ

AM

pos

tpro

cess

ing

OQ

AM

pos

tpro

cess

ing

+

y(n)

c0(l)

c1(l)

cMminus1(l)

x0(k)

x1(k)

xMminus1(k)

a0(n)

a1(n)

aMminus1(n)

Figure 4 Basic representation of a FBMC-OQAM transmultiplexer This scheme shows the basic idea behind FBMC-OQAM where eachsymbol of each subchannel is modulated as OQAM multiplied by its corresponding subcarrier and filtered by the prototype filter In thereceiver the inverse process is performed

where 119899 = 0 119872119870 minus 1 and 119892119898 filters are obtained by theperiodic repetition of 119892119898 filters defined in (1) So 119892119898(119896) =119892119898[mod(119896119872119870)]

At the receiver side the GFDM-OQAM signal 119910(119899) isdemodulated so that in first place each subchannel is filteredin order to get only its corresponding signal Once againthe difference between FBMC-OQAM and GFDM-OQAMreceivers consists in the circular convolution

119886119898 (119899) =119872119870minus1

sum119897=0

119910 (119897) 119892119898 (119899 minus 119897 + 119872119870 minus 1) (7)

where119898 = 0 119872 minus 1Then 119886119898(119899) is sampled in the time domain so that the

estimation of the original OQAM signal 119909119898(119896) is obtainedjust as it is shown in (5)

The insertion of a CP transmission channel noiseequalization and other issues have been omitted in orderto focus just on GFDM-OQAM waveform modulation anddemodulation More details about GFDM-OQAM can befound in [19 20] In those works the idea of GFDM-OQAMis introduced although the authors refer to this scheme asFBMC-COQAM or just as COQAM

24 WCP-COQAM WCP-COQAM is introduced in [21]This technique can be considered as a GFDM-OQAM systemwith a windowing process which improves the PSD withrespect to GFDM-OQAM

In WCP-COQAM both the CP and the windowing arerelated The CP is divided into two parts the guard interval(GI) and the window interval (WI) (referred to as RI in [21])So the total length of the CP will be equal to the sum of thelengths of the GI and WI parts

119871CP = 119871GI + 119871WI (8)

where GI is the part of the CP aimed at protecting againstmultipath channels and WI is the part that will be used forwindowing

Let us consider a GFDM-OQAM signal 119904GFDM-OQAM asa queue of several GFDM-OQAM blocks each with its CPextension of length 119871CP Equally we will consider 119872 to bethe total number of subcarriers and 119870 to be the number of

symbols in each blockThen the 119897th block of aWCP-COQAMsignal is defined as

119904WCP-COQAM (119896)

=119897+1

sum119903=119897minus1

119904WCP-COQAM (119896 + 119903119871WI) 119908 (119896 minus 119903119876) (9)

where 119896 = 0 119872119870 + 119871CP minus 1 119876 = 119872119870 + 119871GI and thewindow function 119908(119896) is defined by

119908 (119896)

=

window coeff 119896 = 0 119871WI minus 11 119896 = 119871WI 119872119870 + 119871GI minus 1window coeff 119896 = 119876 119872119870 + 119871CP minus 10 otherwise

(10)

In our work we use a Hamming window for the windowcoefficients Figure 5 shows the block diagram that representsthis whole process

It must be noted from (9) and (10) that the samples ofthe CP corresponding to the WI part and also the last 119871WIsamples of the block aremultiplied by thewindow coefficientsand they are overlapped with the WI regions of the previousand next blocks respectively By overlapping the WI regionsthese extra samples are prevented from reducing SE as shownin Figure 6

The receiver structure is depicted in Figure 7 It is worthnoticing that Figure 7 only shows the recovery process ofone GFDM-OQAM block from a received WCP-COQAMsignal Once the GFDM-OQAM signal is recovered there areseveral algorithms for its demodulation In Section 23 of thispaper the basic demodulation scheme for GFDM-OQAM isexplained and in [21] the authors introduce a computationallyefficient demodulation scheme

In [21] for this scheme perfect synchronization and anideal channel are assumed So in the first place the first119872119870 + 119871GI samples of one WCP-COQAM block are takenNext the first 119871GI (which correspond to the CP extension)samples are removed while the last 119871WI samples are left toprocess them within the next block Finally in order to getall the samples back to their original positions within the

Journal of Computer Networks and Communications 5

s＄－-１－(k) SP

CPinsertion

times

times

times

w(0)

w(1)

w(MK+ L０ minus 1)

Q

Q

Q +

+

zminus1

zminus1

zminus1

s７０-C１－(k)

Figure 5 Block diagram of the cyclic prefix insertion and windowing operation for WCP-COQAM A WCP-COQAM signal is made froma GFDM-OQAM signal and a CP and a window function are applied

WI

CP MK

Block l

Block l minus 1 Block l + 1

GI M ７） ）

middot middot middot middot middot middot

Figure 6 Overlapping of adjacent WCP-COQAM blocks The regions affected by the window of one block are overlapped with the regionsaffected by the window of the previous and the next blocks These regions correspond to the first and the last 119871WI samples of each block

original GFDM-OQAM block a cyclic shift is carried outwhich can be expressed as

1199041015840GFDM-OQAM (119896) = 119904 [mod (119896 + 119871WI119872119870)] (11)

where 119896 = 0 119872119870minus 1 1199041015840GFDM-OQAM(119896) is the estimation ofthe original 119904WCP-COQAM(119896) signal and 119904(119896) is the signal afterthe removal of the windowing effect and the CP

3 Orthogonality Condition and SpectralEfficiency Analysis of WCP-COQAM

Several analyses on 5Gmodulation candidates are available inthe bibliography [24 25] some of them even includingWCP-COQAM modulation scheme [26 27] However to the bestof our knowledge none of them address the implications ofwindowing in the WCP-COQAM waveform as we do here

In this section we contribute some additional detailsregarding how windowing affects the orthogonality againstmultipath channels and the SE in WCP-COQAM Specifi-cally we state the condition that the GI region of the CPextension must fulfil in order to provide full orthogonality inthe presence of multipath channels Besides we also compareWCP-COQAM and GFDM-OQAM (due to the similaritiesbetween both MCM systems) in terms of SE and we showthat in this aspect the former is outperformed by the latter ifequal multipath protection is assumed

31 Orthogonality Condition In WCP-COQAM multipathchannel protection is directly related to the CP extensionand also to the windowing process In order to explain thisfact we show in detail the steps to form a WCP-COQAMblock starting from a basic GFDM-OQAM block (as seen inSection 23)

Once a GFDM-OQAMblock is formed the CP extensionis added to it Figure 8(a) shows the details of this first stagein the construction of WCP-COQAM signal where119872 is thetotal number of subcarriers 119870 is the number of symbols perblock WI is the region of length 119871WI devoted to windowingand overlapping GI is the region of length 119871GI that protectsagainst multipath channel effect and the CP extension hasa total length of 119871CP = 119871WI + 119871GI CP

1015840 WI1015840 and GI1015840 arethose last samples of the block which are copied in orderto form the CP extension As is represented in Figure 8(a)before the windowing is applied this signal can be consideredas a GFDM-OQAM block with CP extension

Figure 8(b) represents the WCP-COQAM block rightafter windowing is applied As shown in Figure 6 and in (9)and (10) the window function is applied on the first andthe last 119871WI samples so that these samples are somehowdistorted Because of that CPWI and GI are no longer equalto CP1015840 WI1015840 and GI1015840 respectively

Since the last 119871WI samples are affected by the window wedefine a new region for them which we call GI21015840 Thus now

6 Journal of Computer Networks and Communications

SP

CP amp lastL２）

symbolsremoval

s７０-C１－(k)

s (0)

s(L） minus 1)

s(L）)

s(MK + L） minus 1)

s(MK + L）)

s(MK + L０ minus 1)

Cyclicshift

s＄－-１－(0)

s＄－-１－(1)

s＄－-１－(MK minus 1)

Figure 7 Block diagram of one GFDM-OQAM block recovery process from a WCP-COQAM block

the original GI and GI1015840 regions are divided into GI1 and GI2and GI11015840 and GI21015840 respectively so that WI = WI1015840 GI1 =GI11015840 and GI2 = GI21015840 As for the lengths of these regions119871WI = 119871WI1015840 = 119871GI2 = 119871GI21015840 and 119871GI1 = 119871GI11015840 Note that GI1and GI11015840 are now the only really redundant parts Because ofthat only GI1 and not the whole GI (as it is explained in [21])acts as a real guard interval against multipath channels whenfrequency domain equalization (FDE) is performed Hence119871GI1 ge 119871CH minus 1 must be fulfilled in order to maintain fullorthogonality through a multipath channel where 119871CH is thetransmission channel length Thus considering that 119871GI2 =119871WI and so 119871GI1 = 119871GI minus119871GI2 = 119871GI minus119871WI we can define thefull orthogonality condition for WCP-COQAM as

119871GI ge 119871CH minus 1 + 119871WI (12)

It is worth mentioning that the FDE should be performed forsamples between GI2 and GI11015840 inclusive because that is thesection corresponding to the cyclic convolution between GI1and GI11015840 with the transmission channel

32 Spectral Efficiency Analysis Figure 9 shows how over-lapping between adjacent blocks is performed in WCP-COQAM Every windowed region is overlapped with thenext or the previous block so that the WI region of eachblock is overlapped with the GI21015840 region of the previousblock Thus the samples within WI regions do not supposean overload and they do not affect the SE Based on this factin [21] the authors claim that the use of the window function

has no effect on SE However this statement is not preciseAlthough the samples inWI do not directly affect SE becausethey overlap with the samples of GI21015840 of the previous blockwindowing implies adding theGI2 region in order to fulfil theorthogonality condition defined in (12) Note that the use ofthe window function and overlapping ofWI andGI21015840 regionsdistort those samples in GI21015840 region This and the fact thatFDE is performed between GI2 and GI11015840 inclusive implythat the GI2 region is necessary in order to recover thosesamples originally placed in the GI21015840 region At this pointwe conclude that both GI1 and GI2 are essential parts withinthe GI region of WCP-COQAM

In order to make a fair analysis of SE in WCP-COQAMwe compare it to GFDM-OQAM because both modula-tion schemes only differ in windowing and overlappingprocesses In order to maintain a homogeneous notationbetween GFDM-OQAM andWCP-COQAM we assume thefollowing considerations (for the sake of readability fromthis point on parameters corresponding to OFDM will beexpressed with the subindex 119874 parameters correspondingto GFDM-OQAM will be expressed with the subindex 119866and parameters corresponding to WCP-COQAM will beexpressed with the subindex119882)

CP119866 = GI119866 = GI1119866CP119882 = [WI119882GI119882] GI119882 = [GI1119882GI2119882]

(13)

Journal of Computer Networks and Communications 7

WI

CP

(GFDM-OQAM block with CP extension)

GI M

MK

(GFDM-OQAM block without CP extension)

Block l

CP

７） ）

(a)

WI GI2

Block l

GI1 ７） ）1 ）2

(b)

Figure 8Windowing operation of aWCP-COQAM block (a) AWCP-COQAM block before windowing is equal to a GFDM-OQAM blockwith CP extension (b) The first and the last 119871WI samples of a WCP-COQAM block are distorted by the window

WI GI2

(block l minus 1)

Block l

(block l)GI1

WI GI2(block l + 1)

GI1

(block l)

Block l minus 1 Block l + 1

７）

middot middot middot middot middot middot

）1 ）2

７） ）1 ）2

Figure 9 Overlapping between adjacent WCP-COQAM blocks and CP extensionrsquos structure Only GI1 and GI11015840 remain equal in order toemulate a circular convolution with the channel and perform frequency domain equalization so GI2 implies an overload that affects SE

The criterion we applied for this comparison is to provideequal protection against multipath channels for both modu-lations so that 119871GI1119866 = 119871GI1119882 Thus

SE119866 = 119872119870119871GI119866 +119872119870 = 119872119870

119871GI1119866 +119872119870

SE119882 = 119872119870119871GI119882 +119872119870 = 119872119870

119871GI1119882 + 119871GI2119882 +119872119870(14)

From this analysis it is clear that if full orthogonalityis to be maintained the windowing process brings an extraGI2 region to the CP which has effects on the SE of a WCP-COQAM signal In particular SE is reduced by a factor of(119871GI1 + 119871GI2 +119872119870)(119871GI1 +119872119870) compared to a windowlesssystem with equal multipath channel protection

4 Simulation Results and Discussion

In this section we show and discuss the results obtainedfrom the simulations of the systems previously describedBased on these results we assess the performances of thesimulated systems in terms of BER PSD and SE In ouranalysis we prioritize robustness against highly dispersivechannels so that we use the BER of transmissions throughRayleigh fading multipath channels in order to assess therobustness of the MCM systems In addition to BER we alsotake into consideration the PSD of the transmitted signalsfor it is an important factor when it comes to CR systemsRegarding SE we use it to represent the cost of the CPextensions windowing process and convolution tails causedby the prototype filters in FBMC-OQAM so that we canobtain a fair comparison between different MCM systems

8 Journal of Computer Networks and Communications

Table 1 Basic simulation parameters Unless opposite is stated the results shown in this work have been obtained under the conditionsdefined by these parameters

OFDM FBMC-OQAM GFDM-OQAM WCP-COQAMSignal bandwidth 20MHzIQ modulation QPSK119872 (subcarriers) 512119870 (symbolsblock overlapping factor) 1 4 4 4119871CP (samples) 32 mdash 32 224119871WI (samples) mdash mdash mdash 96119871GI (samples) 32 mdash 32 128119871GI1 (samples) mdash mdash mdash 32119871GI2 (samples) mdash mdash mdash 96Prototype filter mdash PHYDYAS [5]119871119901 (samples) mdash 2048

Since we are evaluating MCM systems we considerit appropriate to compare the analysed FMC modulationtechniques with an OFDM reference system in order toassess their performance Table 1 shows the configurationparameters we used in most of our simulations We chooseparameters suitable for low-band transmissions in ultrareli-able and low-latency indoor scenarios For that we consideraspects like subcarrier bandwidth and transmission channelrsquoscoherence bandwidth data block duration and windowlength for PSD improvement

41 Robustness against Multipath Channels In this subsec-tion we analyse the robustness of each MCM system againstmultipath channels In order to make a fair comparisonwe simulate every system with equal protection againstmultipath channels matching effective GI lengths (119871GI119874 119871GI119866 and 119871GI1119882) These parameters are selected as shown inTable 1

The equalization technique we use in our simulations isone-tap zero-forcing (ZF) FDE for all the MCM systemsWhile OFDM GFDM-OQAM and WCP-COQAM employCP extensions to prevent ISI caused by the transmissionchannel FBMC-OQAM could be considered a special casein this aspect Since it is not a blockwise modulation andno CP extension is used one-tap FDE is not a straightfor-ward equalization technique for FBMC-OQAM Howeverdue to the subcarrier bandwidth and symbol duration nearto transmission channelrsquos coherence bandwidth and timerespectively we consider the one-tap FDE as a suitableequalization method for FBMC-OQAM in this case An in-depth analysis about the doubly dispersive channelsrsquo impacton FBMC systems is given in [28]

We simulated transmissions through a Rayleigh fadingchannel model with exponential power delay profile (PDP)whose root mean square (rms) delay spread and channellength are 119905rms = 185 ns and 119871CH = 22 taps Besides wesimulate uncoded and coded communications with turbocodes soft decoding and a code rate of 13 In Figure 10

BER

100

10minus1

10minus2

10minus3

10minus4

10minus5

10minus6

0 5 10 15 20 25 30 35

OFDM (uncoded)QFDM (coded)FBMC-OQAM (coded) WCP-COQAM (coded)GFDM-OQAM (coded)

FBMC-OQAM (uncoded)WCP-COQAM (uncoded)GFDM-OQAM (uncoded)

EbN0 (dB)

Figure 10 BER curves of the analysedMCM systems in a multipathchannel We employed turbo encodingdecoding with soft decodingand a code rate of 13

we compare the BER versus 1198641198871198730 of the MCM systemsover these channelmodels assuming perfect synchronizationand full channel state information (CSI) We got these BERcurves by simulating transmissions through 104Monte Carloiterations for each MCM system and each channel modelWe transmit 12288 random data bits and we generate newrandom channels at each Monte Carlo iteration

Here we show that for uncoded transmissions in termsof error rate FBMC-OQAMperforms slightly worse than therest of MCM systems under the simulated conditions Whileat low 1198641198871198730 values every MCM system provides similarBER an error floor appears at high 1198641198871198730 values because ofthe lack of GI in FBMC-OQAMOn the other hand although

Journal of Computer Networks and Communications 9

Frequency normalised to subcarrier number

minus60

minus50

minus40

minus30

minus20

minus10

0

10

Mag

nitu

de (d

B)

minus30 minus20 minus10 0 10 20 30

OFDMFBMC-OQAM

WCP-COQAMGFDM-OQAM

Figure 11 PSD of the analysedMCM systems with some subcarriersdeactivated Only the central 32 subcarriers out of 512 are activated

OFDM GFDM-OQAM andWCP-COQAMprovide similarperformances in terms of BER providing equal multipatheffect protection OFDM slightly outperforms the other twoMCM schemes due to its perfect orthogonality

Finally it is worth noting that for coded transmissionsall the simulated MCM systems present similar BER curvesso that they provide similar robustness against multipathchannels

42 Power Spectral Density Analysis PSD is a relevant char-acteristic in order to design a CR communication systemsuitable for scenarios with high spectrum occupation In thissubsection we compare the spectrums of the signals of theanalysed MCM systems in order to assess their potentialsuitability for CR based systems

Figure 11 shows the spectrums of OFDM GFDM-OQAM WCP-COQAM and FBMC-OQAM signals one ontop of the other For this test we only activate the 32 centralsubcarriers in order to make the power leakage into adjacentsubcarriers visible

As shown in Figure 11 the FB based MCM systems carrysignificant reduction of OOB radiation compared to OFDMHowever even between these MCM schemes there is stilla remarkable difference In this sense FBMC-OQAM out-performs both GFDM-OQAM and WCP-COQAM On theother hand thanks to the windowing WCP-COQAM showsconsiderably lower OOB radiation than GFDM-OQAM Itis worth mentioning that the longer the WI region is thenarrower WCP-COQAM signalrsquos spectrum gets Since extrasamples within the GI region are needed for windowing asdescribed in Section 3 PSD improvement of WCP-COQAMover GFDM-OQAM comes at the cost of lower SE Moredetailed analysis on SE is given in Section 43

43 Spectral Efficiency Analysis In this subsection we anal-yse SE in order to evaluate the cost of obtaining higherrobustness against dispersive channels and better PSD bymeans of CP extensions and windowing For this analysis weconsider the parameters defined in Table 1 so that effectivemultipath channel protection is equal for OFDM GFDM-OQAM and WCP-COQAM Equations (15) show the SEvalues of the three blockwise MCM systems

SE119874 = 119872119870119874119871GI119874 +119872119870119874 = 09412

SE119866 = 119872119870119866119871GI119866 +119872119870119866 = 09846

SE119882 = 119872119870119882119871GI119882 +119872119870119882 = 09412

(15)

Here we show a particular case with some specific parame-ters On one hand GFDM-OQAM presents higher SE thanOFDM because its GI precedes 119870 symbols while in OFDMthe GI precedes only one symbol On the other hand becauseof the windowing operation the CP extension of WCP-COQAM specifically the GI part has 119871GI2 extra sampleswith respect to GFDM-OQAM and that is the reason whyWCP-COQAM also presents lower SE than GFDM-OQAMHowever under these conditions with a high number ofsubcarriers we can conclude that the differences in the SEbetweenOFDMGFDM-OQAM andWCP-COQAMare notreally significant

Regarding FBMC-OQAM once again we consider itdifferent from the rest of MCM systems because it is nota blockwise modulation and it uses no GI protection formultipath effect So its SE depends on the length of thetransmitted frames with respect to the convolution tailsFor long data transmissions its SE will tend to 1 while forshort data transmissions it should tend to 05 assuming aminimum amount of data samples equivalent to a GFDM-OQAM orWCP-COQAM blockThis calculation is simple ifwe introduce 4 complex data symbols (8 OQAM symbols) in(2)

119886119898 (119899) =7

sum119896=0

119909119898 (119896) 120575 (119899 minus 1198961198722 ) 119899 isin [0 81198722 minus 1] (16)

and we introduce these upsampled data in (3) The resultant119910(119899) signal will contain 4095 samples of which half willcorrespond to actual data and the rest to convolution tailsfrom subcarrier filtering

Considering the low-latency scenarios our researchfocuses on SE might suppose a significant drawback forFBMC-OQAM compared to GFDM-OQAM and WCP-COQAM during short data transmissions

5 Conclusions

In this article we make a comparison between three candi-dates for PHY layer modulation in 5G and a reference OFDM

10 Journal of Computer Networks and Communications

system in order to assess their suitability for industrial wire-less communications based on CR For that we simulate low-band communications and highly dispersive indoor channelscenarios Under these conditions we show the performancesand features of these MCM systems from different points ofview

Regarding robustness against multipath channels weconclude that all the analysed MCM systems provide similarperformances under the simulated conditions Although foruncoded transmissions FBMC-OQAM presents an errorfloor at high 1198641198871198730 values this error floor is corrected whencoding is used

In terms of PSD and OOB radiation these filtered MCMschemes provide more restrained spectrum than OFDM Inparticular FBMC-OQAM outperforms the rest of the modu-lation schemes in this aspect so it could be a suitable candi-date for the CR scenarios considered in our research On theother hand although not as good as FBMC-OQAM WCP-COQAM still has considerably better PSD than GFDM-OQAM which only provides less than 10 dB of differencein OOB radiation with respect to OFDM Therefore weconclude that circular filtering only is not sufficient andwindowing must be applied in order to ensure more efficientuse of the spectrum Moreover we consider that due to itsalso restrained spectrum WCP-COQAM might be anothersuitable modulation scheme for CR applications

As for the SE analysis as discussed in Section 43the difference between OFDM GFDM-OQAM and WCP-COQAM is not really significant for the parameters weused in our simulations On the other hand FBMC-OQAMmight suffer from severe degradation of SE during short datatransmissions in the low-latency scenarios we consider in ourresearch

6 Future Work

In this work we considered perfect synchronization andfull channel state information at the receiver but it is alsoimportant to consider more realistic simulation conditionsin the future So the next step in our investigation will beto analyse and propose time-frequency synchronization andchannel estimation methods for GFDM-OQAM and WCP-COQAM

The other key point in our research will be to simulatechannelmodels that fairly emulate industrial wireless channelconditions For that we will simulate severe multipath con-ditions time-varying channels impulsive noise and interfer-ence from other wireless systems

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

This work was partly supported by TIPOTRANS(ETORTEK) and CIIRCOS (PC2013-68) projects ofthe Basque Government (Spain) and by COWITRACC

(TEC2014-59490-C2-1-P2-P) project of the Spanish Minis-try of Economy and Competitiveness

References

[1] H Zhang D L Ruyet M Terre et al ldquoApplication of theFBMC physical layer in a Cognitive Radio scenariordquo Tech RepPHYDYAS-Physical Layer for Dynamic Access and CognitiveRadio 2009

[2] I Gaspar and GWunder ldquo5G cellular communications scenar-ios and system requirementsrdquo Tech Rep 5th Generation Non-Orthogonal Waveforms for Asynchronous Signalling 2013

[3] P Popovski V Braun H Mayer et al ldquoScenarios requirementsandKPIs for 5Gmobile andwireless systemrdquo Tech RepMobileand wireless communications Enablers for the Twenty-twentyInformation Society (METIS) 2013

[4] X Mestre D Gregoratti M Renfors et al ldquoICT-EMPhAtiCfinal project reportrdquo Tech Rep Enhanced Multicarrier Tech-niques for Professional Ad-Hoc and Cell-Based Communica-tions 2015

[5] A Viholainen M Bellanger and M Huchard ldquoPrototypefilter and structure optimizationrdquo PHYDYAS-Physical Layer forDynamic Access and Cognitive Radio 2009

[6] N Michailow M Matthe I S Gaspar et al ldquoGeneralizedfrequency division multiplexing for 5th generation cellularnetworksrdquo IEEE Transactions on Communications vol 62 no9 pp 3045ndash3061 2014

[7] V Vakilian T Wild F Schaich S Ten Brink and J-F FrigonldquoUniversal-filtered multi-carrier technique for wireless systemsbeyond LTErdquo in Proceedings of the 2013 IEEE Globecom Work-shops GC Wkshps 2013 pp 223ndash228 IEEE December 2013

[8] M Agiwal A Roy and N Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 18 no 3 pp 1617ndash1655 2016

[9] X Zhang M Jia L Chen J Ma and J Qiu ldquoFiltered-OFDM-enabler for flexible waveform in the 5th generation cellular net-worksrdquo in Proceedings of the 58th IEEE Global CommunicationsConference (GLOBECOM rsquo15) pp 1ndash6 IEEE December 2015

[10] L Zhang A Ijaz P Xiao A Quddus and R Tafazolli ldquoSubbandfiltered multi-carrier systems for multi-service wireless com-municationsrdquo IEEE Transactions on Wireless Communicationsvol 16 no 3 pp 1893ndash1907 2017

[11] A Lizeaga M Mendicute P M Rodriguez and I Val ldquoEval-uation of WCP-COQAM GFDM-OQAM and FBMC-OQAMfor industrial wireless communications with Cognitive Radiordquoin Proceedings of the 2017 IEEE International Workshop ofElectronics Control Measurement Signals and their Applicationto Mechatronics (ECMSM) pp 1ndash6 Donostia Spain May 2017

[12] P M Rodriguez I Val A Lizeaga and M MendicuteldquoEvaluation of cognitive radio for mission-critical and time-critical WSAN in industrial environments under interferencerdquoin Proceedings of the 2015 11th IEEEWorld Conference on FactoryCommunication Systems (WFCS rsquo15) pp 1ndash4 May 2015

[13] P M Rodrıguez R Torrego F Casado et al ldquoDynamicspectrumaccess integrated in awideband cognitiveRF-ethernetbridge for industrial control applicationsrdquo Journal of SignalProcessing Systems vol 83 no 1 pp 19ndash28 2016

[14] A K Somappa K Ovsthus and L Kristensen ldquoAn industrialperspective on wireless sensor networks - a survey of require-ments protocols and challengesrdquo IEEE Communications Sur-veys Tutorials vol 16 pp 1391ndash1412 2014

Journal of Computer Networks and Communications 11

[15] S Vitturi ldquoWireless networks for the factory floor require-ments available technologies research trends and practicalapplicationsrdquo in Proceedings of the IEEE International Confer-ence on Emerging Technologies and Factory Automation 2014

[16] P Popovski ldquoUltra-reliable communication in 5G wirelesssystemsrdquo in Proceedings of the 2014 1st International Conferenceon 5G for Ubiquitous Connectivity (5GU rsquo14) pp 146ndash151November 2014

[17] N Brahmi ON C Yilmaz KWHelmersson S A Ashraf andJ Torsner ldquoDeployment strategies for ultra-reliable and low-latency communication in factory automationrdquo in Proceedingsof the IEEEGlobecomWorkshops (GCWkshps rsquo15) pp 1ndash6 IEEEDecember 2015

[18] O N Yilmaz ldquoUltra-Reliable and Low-Latency 5G Communi-cationrdquo in Proceedings of the European Conference on Networksand Communications (EuCNC rsquo16) 2016

[19] H Lin andP Siohan ldquoFBMCCOQAM an enabler for cognitiveradiordquo in Proceedings of the 4th International Conference onAdvances in Cognitive Radio (COCORA rsquo14) pp 8097ndash8101May2014

[20] H Lin and P Siohan ldquoAn advanced multi-carrier modulationfor future radio systemsrdquo in Proceedings of the 2014 IEEE Inter-national Conference on Acoustics Speech and Signal Processing(ICASSP rsquo14) pp 8097ndash8101 May 2014

[21] H Lin and P Siohan ldquoMulti-carrier modulation analysis andWCP-COQAMproposalrdquoEurasip Journal onAdvances in SignalProcessing vol 2014 no 1 article 79 2014

[22] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations chapter 8 JohnWiley amp Sons Ltd ChichesterUK 2016

[23] J J Benedetto C Heil and D F Walnut Gabor Systems and theBalian-LowTheorem Birkhauser chapter 2 BostonMass USA1998

[24] A Ijaz L Zhang P Xiao and R TafazolliAnalysis of CandidateWaveforms for 5G Cellular Systems chapter 1 InTech 2016

[25] R Gerzaguet N Bartzoudis L G Baltar et al ldquoThe 5Gcandidate waveform race a comparison of complexity andperformancerdquo Eurasip Journal onWireless Communications andNetworking vol 2017 no 1 article 13 2017

[26] A B Ucuncu and A O Yilmaz ldquoPulse shaping meth-ods for OQAMOFDM and WCP-COQAMrdquo httpsarxivorgabs150900977

[27] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations JohnWiley amp Sons Ltd Chichester UK 2016

[28] L Zhang P Xiao A Zafar A ul Quddus and R TafazollildquoFBMC system an insight into doubly dispersive channelimpactrdquo IEEE Transactions on Vehicular Technology vol 66 no5 pp 3942ndash3956 2017

4 Journal of Computer Networks and Communications

c0(l)

c1(l)

cMminus1(l)

x0(k)

x1(k)

xMminus1(k)

a0(n)

a1(n)

aMminus1(n)

g0(n)

g1(n)

gMminus1(n)

g0(n)

g1(n)

gMminus1(n)

M2

M2

M2

M2

M2

M2OQ

AM

pos

tpro

cess

ing

OQ

AM

pos

tpro

cess

ing

+

y(n)

c0(l)

c1(l)

cMminus1(l)

x0(k)

x1(k)

xMminus1(k)

a0(n)

a1(n)

aMminus1(n)

Figure 4 Basic representation of a FBMC-OQAM transmultiplexer This scheme shows the basic idea behind FBMC-OQAM where eachsymbol of each subchannel is modulated as OQAM multiplied by its corresponding subcarrier and filtered by the prototype filter In thereceiver the inverse process is performed

where 119899 = 0 119872119870 minus 1 and 119892119898 filters are obtained by theperiodic repetition of 119892119898 filters defined in (1) So 119892119898(119896) =119892119898[mod(119896119872119870)]

At the receiver side the GFDM-OQAM signal 119910(119899) isdemodulated so that in first place each subchannel is filteredin order to get only its corresponding signal Once againthe difference between FBMC-OQAM and GFDM-OQAMreceivers consists in the circular convolution

119886119898 (119899) =119872119870minus1

sum119897=0

119910 (119897) 119892119898 (119899 minus 119897 + 119872119870 minus 1) (7)

where119898 = 0 119872 minus 1Then 119886119898(119899) is sampled in the time domain so that the

estimation of the original OQAM signal 119909119898(119896) is obtainedjust as it is shown in (5)

The insertion of a CP transmission channel noiseequalization and other issues have been omitted in orderto focus just on GFDM-OQAM waveform modulation anddemodulation More details about GFDM-OQAM can befound in [19 20] In those works the idea of GFDM-OQAMis introduced although the authors refer to this scheme asFBMC-COQAM or just as COQAM

24 WCP-COQAM WCP-COQAM is introduced in [21]This technique can be considered as a GFDM-OQAM systemwith a windowing process which improves the PSD withrespect to GFDM-OQAM

In WCP-COQAM both the CP and the windowing arerelated The CP is divided into two parts the guard interval(GI) and the window interval (WI) (referred to as RI in [21])So the total length of the CP will be equal to the sum of thelengths of the GI and WI parts

119871CP = 119871GI + 119871WI (8)

where GI is the part of the CP aimed at protecting againstmultipath channels and WI is the part that will be used forwindowing

Let us consider a GFDM-OQAM signal 119904GFDM-OQAM asa queue of several GFDM-OQAM blocks each with its CPextension of length 119871CP Equally we will consider 119872 to bethe total number of subcarriers and 119870 to be the number of

symbols in each blockThen the 119897th block of aWCP-COQAMsignal is defined as

119904WCP-COQAM (119896)

=119897+1

sum119903=119897minus1

119904WCP-COQAM (119896 + 119903119871WI) 119908 (119896 minus 119903119876) (9)

where 119896 = 0 119872119870 + 119871CP minus 1 119876 = 119872119870 + 119871GI and thewindow function 119908(119896) is defined by

119908 (119896)

=

window coeff 119896 = 0 119871WI minus 11 119896 = 119871WI 119872119870 + 119871GI minus 1window coeff 119896 = 119876 119872119870 + 119871CP minus 10 otherwise

(10)

In our work we use a Hamming window for the windowcoefficients Figure 5 shows the block diagram that representsthis whole process

It must be noted from (9) and (10) that the samples ofthe CP corresponding to the WI part and also the last 119871WIsamples of the block aremultiplied by thewindow coefficientsand they are overlapped with the WI regions of the previousand next blocks respectively By overlapping the WI regionsthese extra samples are prevented from reducing SE as shownin Figure 6

The receiver structure is depicted in Figure 7 It is worthnoticing that Figure 7 only shows the recovery process ofone GFDM-OQAM block from a received WCP-COQAMsignal Once the GFDM-OQAM signal is recovered there areseveral algorithms for its demodulation In Section 23 of thispaper the basic demodulation scheme for GFDM-OQAM isexplained and in [21] the authors introduce a computationallyefficient demodulation scheme

In [21] for this scheme perfect synchronization and anideal channel are assumed So in the first place the first119872119870 + 119871GI samples of one WCP-COQAM block are takenNext the first 119871GI (which correspond to the CP extension)samples are removed while the last 119871WI samples are left toprocess them within the next block Finally in order to getall the samples back to their original positions within the

Journal of Computer Networks and Communications 5

s＄－-１－(k) SP

CPinsertion

times

times

times

w(0)

w(1)

w(MK+ L０ minus 1)

Q

Q

Q +

+

zminus1

zminus1

zminus1

s７０-C１－(k)

Figure 5 Block diagram of the cyclic prefix insertion and windowing operation for WCP-COQAM A WCP-COQAM signal is made froma GFDM-OQAM signal and a CP and a window function are applied

WI

CP MK

Block l

Block l minus 1 Block l + 1

GI M ７） ）

middot middot middot middot middot middot

Figure 6 Overlapping of adjacent WCP-COQAM blocks The regions affected by the window of one block are overlapped with the regionsaffected by the window of the previous and the next blocks These regions correspond to the first and the last 119871WI samples of each block

original GFDM-OQAM block a cyclic shift is carried outwhich can be expressed as

1199041015840GFDM-OQAM (119896) = 119904 [mod (119896 + 119871WI119872119870)] (11)

where 119896 = 0 119872119870minus 1 1199041015840GFDM-OQAM(119896) is the estimation ofthe original 119904WCP-COQAM(119896) signal and 119904(119896) is the signal afterthe removal of the windowing effect and the CP

3 Orthogonality Condition and SpectralEfficiency Analysis of WCP-COQAM

Several analyses on 5Gmodulation candidates are available inthe bibliography [24 25] some of them even includingWCP-COQAM modulation scheme [26 27] However to the bestof our knowledge none of them address the implications ofwindowing in the WCP-COQAM waveform as we do here

In this section we contribute some additional detailsregarding how windowing affects the orthogonality againstmultipath channels and the SE in WCP-COQAM Specifi-cally we state the condition that the GI region of the CPextension must fulfil in order to provide full orthogonality inthe presence of multipath channels Besides we also compareWCP-COQAM and GFDM-OQAM (due to the similaritiesbetween both MCM systems) in terms of SE and we showthat in this aspect the former is outperformed by the latter ifequal multipath protection is assumed

31 Orthogonality Condition In WCP-COQAM multipathchannel protection is directly related to the CP extensionand also to the windowing process In order to explain thisfact we show in detail the steps to form a WCP-COQAMblock starting from a basic GFDM-OQAM block (as seen inSection 23)

Once a GFDM-OQAMblock is formed the CP extensionis added to it Figure 8(a) shows the details of this first stagein the construction of WCP-COQAM signal where119872 is thetotal number of subcarriers 119870 is the number of symbols perblock WI is the region of length 119871WI devoted to windowingand overlapping GI is the region of length 119871GI that protectsagainst multipath channel effect and the CP extension hasa total length of 119871CP = 119871WI + 119871GI CP

1015840 WI1015840 and GI1015840 arethose last samples of the block which are copied in orderto form the CP extension As is represented in Figure 8(a)before the windowing is applied this signal can be consideredas a GFDM-OQAM block with CP extension

Figure 8(b) represents the WCP-COQAM block rightafter windowing is applied As shown in Figure 6 and in (9)and (10) the window function is applied on the first andthe last 119871WI samples so that these samples are somehowdistorted Because of that CPWI and GI are no longer equalto CP1015840 WI1015840 and GI1015840 respectively

Since the last 119871WI samples are affected by the window wedefine a new region for them which we call GI21015840 Thus now

6 Journal of Computer Networks and Communications

SP

CP amp lastL２）

symbolsremoval

s７０-C１－(k)

s (0)

s(L） minus 1)

s(L）)

s(MK + L） minus 1)

s(MK + L）)

s(MK + L０ minus 1)

Cyclicshift

s＄－-１－(0)

s＄－-１－(1)

s＄－-１－(MK minus 1)

Figure 7 Block diagram of one GFDM-OQAM block recovery process from a WCP-COQAM block

the original GI and GI1015840 regions are divided into GI1 and GI2and GI11015840 and GI21015840 respectively so that WI = WI1015840 GI1 =GI11015840 and GI2 = GI21015840 As for the lengths of these regions119871WI = 119871WI1015840 = 119871GI2 = 119871GI21015840 and 119871GI1 = 119871GI11015840 Note that GI1and GI11015840 are now the only really redundant parts Because ofthat only GI1 and not the whole GI (as it is explained in [21])acts as a real guard interval against multipath channels whenfrequency domain equalization (FDE) is performed Hence119871GI1 ge 119871CH minus 1 must be fulfilled in order to maintain fullorthogonality through a multipath channel where 119871CH is thetransmission channel length Thus considering that 119871GI2 =119871WI and so 119871GI1 = 119871GI minus119871GI2 = 119871GI minus119871WI we can define thefull orthogonality condition for WCP-COQAM as

119871GI ge 119871CH minus 1 + 119871WI (12)

It is worth mentioning that the FDE should be performed forsamples between GI2 and GI11015840 inclusive because that is thesection corresponding to the cyclic convolution between GI1and GI11015840 with the transmission channel

32 Spectral Efficiency Analysis Figure 9 shows how over-lapping between adjacent blocks is performed in WCP-COQAM Every windowed region is overlapped with thenext or the previous block so that the WI region of eachblock is overlapped with the GI21015840 region of the previousblock Thus the samples within WI regions do not supposean overload and they do not affect the SE Based on this factin [21] the authors claim that the use of the window function

has no effect on SE However this statement is not preciseAlthough the samples inWI do not directly affect SE becausethey overlap with the samples of GI21015840 of the previous blockwindowing implies adding theGI2 region in order to fulfil theorthogonality condition defined in (12) Note that the use ofthe window function and overlapping ofWI andGI21015840 regionsdistort those samples in GI21015840 region This and the fact thatFDE is performed between GI2 and GI11015840 inclusive implythat the GI2 region is necessary in order to recover thosesamples originally placed in the GI21015840 region At this pointwe conclude that both GI1 and GI2 are essential parts withinthe GI region of WCP-COQAM

In order to make a fair analysis of SE in WCP-COQAMwe compare it to GFDM-OQAM because both modula-tion schemes only differ in windowing and overlappingprocesses In order to maintain a homogeneous notationbetween GFDM-OQAM andWCP-COQAM we assume thefollowing considerations (for the sake of readability fromthis point on parameters corresponding to OFDM will beexpressed with the subindex 119874 parameters correspondingto GFDM-OQAM will be expressed with the subindex 119866and parameters corresponding to WCP-COQAM will beexpressed with the subindex119882)

CP119866 = GI119866 = GI1119866CP119882 = [WI119882GI119882] GI119882 = [GI1119882GI2119882]

(13)

Journal of Computer Networks and Communications 7

WI

CP

(GFDM-OQAM block with CP extension)

GI M

MK

(GFDM-OQAM block without CP extension)

Block l

CP

７） ）

(a)

WI GI2

Block l

GI1 ７） ）1 ）2

(b)

Figure 8Windowing operation of aWCP-COQAM block (a) AWCP-COQAM block before windowing is equal to a GFDM-OQAM blockwith CP extension (b) The first and the last 119871WI samples of a WCP-COQAM block are distorted by the window

WI GI2

(block l minus 1)

Block l

(block l)GI1

WI GI2(block l + 1)

GI1

(block l)

Block l minus 1 Block l + 1

７）

middot middot middot middot middot middot

）1 ）2

７） ）1 ）2

Figure 9 Overlapping between adjacent WCP-COQAM blocks and CP extensionrsquos structure Only GI1 and GI11015840 remain equal in order toemulate a circular convolution with the channel and perform frequency domain equalization so GI2 implies an overload that affects SE

The criterion we applied for this comparison is to provideequal protection against multipath channels for both modu-lations so that 119871GI1119866 = 119871GI1119882 Thus

SE119866 = 119872119870119871GI119866 +119872119870 = 119872119870

119871GI1119866 +119872119870

SE119882 = 119872119870119871GI119882 +119872119870 = 119872119870

119871GI1119882 + 119871GI2119882 +119872119870(14)

From this analysis it is clear that if full orthogonalityis to be maintained the windowing process brings an extraGI2 region to the CP which has effects on the SE of a WCP-COQAM signal In particular SE is reduced by a factor of(119871GI1 + 119871GI2 +119872119870)(119871GI1 +119872119870) compared to a windowlesssystem with equal multipath channel protection

4 Simulation Results and Discussion

In this section we show and discuss the results obtainedfrom the simulations of the systems previously describedBased on these results we assess the performances of thesimulated systems in terms of BER PSD and SE In ouranalysis we prioritize robustness against highly dispersivechannels so that we use the BER of transmissions throughRayleigh fading multipath channels in order to assess therobustness of the MCM systems In addition to BER we alsotake into consideration the PSD of the transmitted signalsfor it is an important factor when it comes to CR systemsRegarding SE we use it to represent the cost of the CPextensions windowing process and convolution tails causedby the prototype filters in FBMC-OQAM so that we canobtain a fair comparison between different MCM systems

8 Journal of Computer Networks and Communications

Table 1 Basic simulation parameters Unless opposite is stated the results shown in this work have been obtained under the conditionsdefined by these parameters

OFDM FBMC-OQAM GFDM-OQAM WCP-COQAMSignal bandwidth 20MHzIQ modulation QPSK119872 (subcarriers) 512119870 (symbolsblock overlapping factor) 1 4 4 4119871CP (samples) 32 mdash 32 224119871WI (samples) mdash mdash mdash 96119871GI (samples) 32 mdash 32 128119871GI1 (samples) mdash mdash mdash 32119871GI2 (samples) mdash mdash mdash 96Prototype filter mdash PHYDYAS [5]119871119901 (samples) mdash 2048

Since we are evaluating MCM systems we considerit appropriate to compare the analysed FMC modulationtechniques with an OFDM reference system in order toassess their performance Table 1 shows the configurationparameters we used in most of our simulations We chooseparameters suitable for low-band transmissions in ultrareli-able and low-latency indoor scenarios For that we consideraspects like subcarrier bandwidth and transmission channelrsquoscoherence bandwidth data block duration and windowlength for PSD improvement

41 Robustness against Multipath Channels In this subsec-tion we analyse the robustness of each MCM system againstmultipath channels In order to make a fair comparisonwe simulate every system with equal protection againstmultipath channels matching effective GI lengths (119871GI119874 119871GI119866 and 119871GI1119882) These parameters are selected as shown inTable 1

The equalization technique we use in our simulations isone-tap zero-forcing (ZF) FDE for all the MCM systemsWhile OFDM GFDM-OQAM and WCP-COQAM employCP extensions to prevent ISI caused by the transmissionchannel FBMC-OQAM could be considered a special casein this aspect Since it is not a blockwise modulation andno CP extension is used one-tap FDE is not a straightfor-ward equalization technique for FBMC-OQAM Howeverdue to the subcarrier bandwidth and symbol duration nearto transmission channelrsquos coherence bandwidth and timerespectively we consider the one-tap FDE as a suitableequalization method for FBMC-OQAM in this case An in-depth analysis about the doubly dispersive channelsrsquo impacton FBMC systems is given in [28]

We simulated transmissions through a Rayleigh fadingchannel model with exponential power delay profile (PDP)whose root mean square (rms) delay spread and channellength are 119905rms = 185 ns and 119871CH = 22 taps Besides wesimulate uncoded and coded communications with turbocodes soft decoding and a code rate of 13 In Figure 10

BER

100

10minus1

10minus2

10minus3

10minus4

10minus5

10minus6

0 5 10 15 20 25 30 35

OFDM (uncoded)QFDM (coded)FBMC-OQAM (coded) WCP-COQAM (coded)GFDM-OQAM (coded)

FBMC-OQAM (uncoded)WCP-COQAM (uncoded)GFDM-OQAM (uncoded)

EbN0 (dB)

Figure 10 BER curves of the analysedMCM systems in a multipathchannel We employed turbo encodingdecoding with soft decodingand a code rate of 13

we compare the BER versus 1198641198871198730 of the MCM systemsover these channelmodels assuming perfect synchronizationand full channel state information (CSI) We got these BERcurves by simulating transmissions through 104Monte Carloiterations for each MCM system and each channel modelWe transmit 12288 random data bits and we generate newrandom channels at each Monte Carlo iteration

Here we show that for uncoded transmissions in termsof error rate FBMC-OQAMperforms slightly worse than therest of MCM systems under the simulated conditions Whileat low 1198641198871198730 values every MCM system provides similarBER an error floor appears at high 1198641198871198730 values because ofthe lack of GI in FBMC-OQAMOn the other hand although

Journal of Computer Networks and Communications 9

Frequency normalised to subcarrier number

minus60

minus50

minus40

minus30

minus20

minus10

0

10

Mag

nitu

de (d

B)

minus30 minus20 minus10 0 10 20 30

OFDMFBMC-OQAM

WCP-COQAMGFDM-OQAM

Figure 11 PSD of the analysedMCM systems with some subcarriersdeactivated Only the central 32 subcarriers out of 512 are activated

OFDM GFDM-OQAM andWCP-COQAMprovide similarperformances in terms of BER providing equal multipatheffect protection OFDM slightly outperforms the other twoMCM schemes due to its perfect orthogonality

Finally it is worth noting that for coded transmissionsall the simulated MCM systems present similar BER curvesso that they provide similar robustness against multipathchannels

42 Power Spectral Density Analysis PSD is a relevant char-acteristic in order to design a CR communication systemsuitable for scenarios with high spectrum occupation In thissubsection we compare the spectrums of the signals of theanalysed MCM systems in order to assess their potentialsuitability for CR based systems

Figure 11 shows the spectrums of OFDM GFDM-OQAM WCP-COQAM and FBMC-OQAM signals one ontop of the other For this test we only activate the 32 centralsubcarriers in order to make the power leakage into adjacentsubcarriers visible

As shown in Figure 11 the FB based MCM systems carrysignificant reduction of OOB radiation compared to OFDMHowever even between these MCM schemes there is stilla remarkable difference In this sense FBMC-OQAM out-performs both GFDM-OQAM and WCP-COQAM On theother hand thanks to the windowing WCP-COQAM showsconsiderably lower OOB radiation than GFDM-OQAM Itis worth mentioning that the longer the WI region is thenarrower WCP-COQAM signalrsquos spectrum gets Since extrasamples within the GI region are needed for windowing asdescribed in Section 3 PSD improvement of WCP-COQAMover GFDM-OQAM comes at the cost of lower SE Moredetailed analysis on SE is given in Section 43

43 Spectral Efficiency Analysis In this subsection we anal-yse SE in order to evaluate the cost of obtaining higherrobustness against dispersive channels and better PSD bymeans of CP extensions and windowing For this analysis weconsider the parameters defined in Table 1 so that effectivemultipath channel protection is equal for OFDM GFDM-OQAM and WCP-COQAM Equations (15) show the SEvalues of the three blockwise MCM systems

SE119874 = 119872119870119874119871GI119874 +119872119870119874 = 09412

SE119866 = 119872119870119866119871GI119866 +119872119870119866 = 09846

SE119882 = 119872119870119882119871GI119882 +119872119870119882 = 09412

(15)

Here we show a particular case with some specific parame-ters On one hand GFDM-OQAM presents higher SE thanOFDM because its GI precedes 119870 symbols while in OFDMthe GI precedes only one symbol On the other hand becauseof the windowing operation the CP extension of WCP-COQAM specifically the GI part has 119871GI2 extra sampleswith respect to GFDM-OQAM and that is the reason whyWCP-COQAM also presents lower SE than GFDM-OQAMHowever under these conditions with a high number ofsubcarriers we can conclude that the differences in the SEbetweenOFDMGFDM-OQAM andWCP-COQAMare notreally significant

Regarding FBMC-OQAM once again we consider itdifferent from the rest of MCM systems because it is nota blockwise modulation and it uses no GI protection formultipath effect So its SE depends on the length of thetransmitted frames with respect to the convolution tailsFor long data transmissions its SE will tend to 1 while forshort data transmissions it should tend to 05 assuming aminimum amount of data samples equivalent to a GFDM-OQAM orWCP-COQAM blockThis calculation is simple ifwe introduce 4 complex data symbols (8 OQAM symbols) in(2)

119886119898 (119899) =7

sum119896=0

119909119898 (119896) 120575 (119899 minus 1198961198722 ) 119899 isin [0 81198722 minus 1] (16)

and we introduce these upsampled data in (3) The resultant119910(119899) signal will contain 4095 samples of which half willcorrespond to actual data and the rest to convolution tailsfrom subcarrier filtering

Considering the low-latency scenarios our researchfocuses on SE might suppose a significant drawback forFBMC-OQAM compared to GFDM-OQAM and WCP-COQAM during short data transmissions

5 Conclusions

In this article we make a comparison between three candi-dates for PHY layer modulation in 5G and a reference OFDM

10 Journal of Computer Networks and Communications

system in order to assess their suitability for industrial wire-less communications based on CR For that we simulate low-band communications and highly dispersive indoor channelscenarios Under these conditions we show the performancesand features of these MCM systems from different points ofview

Regarding robustness against multipath channels weconclude that all the analysed MCM systems provide similarperformances under the simulated conditions Although foruncoded transmissions FBMC-OQAM presents an errorfloor at high 1198641198871198730 values this error floor is corrected whencoding is used

In terms of PSD and OOB radiation these filtered MCMschemes provide more restrained spectrum than OFDM Inparticular FBMC-OQAM outperforms the rest of the modu-lation schemes in this aspect so it could be a suitable candi-date for the CR scenarios considered in our research On theother hand although not as good as FBMC-OQAM WCP-COQAM still has considerably better PSD than GFDM-OQAM which only provides less than 10 dB of differencein OOB radiation with respect to OFDM Therefore weconclude that circular filtering only is not sufficient andwindowing must be applied in order to ensure more efficientuse of the spectrum Moreover we consider that due to itsalso restrained spectrum WCP-COQAM might be anothersuitable modulation scheme for CR applications

As for the SE analysis as discussed in Section 43the difference between OFDM GFDM-OQAM and WCP-COQAM is not really significant for the parameters weused in our simulations On the other hand FBMC-OQAMmight suffer from severe degradation of SE during short datatransmissions in the low-latency scenarios we consider in ourresearch

6 Future Work

In this work we considered perfect synchronization andfull channel state information at the receiver but it is alsoimportant to consider more realistic simulation conditionsin the future So the next step in our investigation will beto analyse and propose time-frequency synchronization andchannel estimation methods for GFDM-OQAM and WCP-COQAM

The other key point in our research will be to simulatechannelmodels that fairly emulate industrial wireless channelconditions For that we will simulate severe multipath con-ditions time-varying channels impulsive noise and interfer-ence from other wireless systems

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

This work was partly supported by TIPOTRANS(ETORTEK) and CIIRCOS (PC2013-68) projects ofthe Basque Government (Spain) and by COWITRACC

(TEC2014-59490-C2-1-P2-P) project of the Spanish Minis-try of Economy and Competitiveness

References

[1] H Zhang D L Ruyet M Terre et al ldquoApplication of theFBMC physical layer in a Cognitive Radio scenariordquo Tech RepPHYDYAS-Physical Layer for Dynamic Access and CognitiveRadio 2009

[2] I Gaspar and GWunder ldquo5G cellular communications scenar-ios and system requirementsrdquo Tech Rep 5th Generation Non-Orthogonal Waveforms for Asynchronous Signalling 2013

[3] P Popovski V Braun H Mayer et al ldquoScenarios requirementsandKPIs for 5Gmobile andwireless systemrdquo Tech RepMobileand wireless communications Enablers for the Twenty-twentyInformation Society (METIS) 2013

[4] X Mestre D Gregoratti M Renfors et al ldquoICT-EMPhAtiCfinal project reportrdquo Tech Rep Enhanced Multicarrier Tech-niques for Professional Ad-Hoc and Cell-Based Communica-tions 2015

[5] A Viholainen M Bellanger and M Huchard ldquoPrototypefilter and structure optimizationrdquo PHYDYAS-Physical Layer forDynamic Access and Cognitive Radio 2009

[6] N Michailow M Matthe I S Gaspar et al ldquoGeneralizedfrequency division multiplexing for 5th generation cellularnetworksrdquo IEEE Transactions on Communications vol 62 no9 pp 3045ndash3061 2014

[7] V Vakilian T Wild F Schaich S Ten Brink and J-F FrigonldquoUniversal-filtered multi-carrier technique for wireless systemsbeyond LTErdquo in Proceedings of the 2013 IEEE Globecom Work-shops GC Wkshps 2013 pp 223ndash228 IEEE December 2013

[8] M Agiwal A Roy and N Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 18 no 3 pp 1617ndash1655 2016

[9] X Zhang M Jia L Chen J Ma and J Qiu ldquoFiltered-OFDM-enabler for flexible waveform in the 5th generation cellular net-worksrdquo in Proceedings of the 58th IEEE Global CommunicationsConference (GLOBECOM rsquo15) pp 1ndash6 IEEE December 2015

[10] L Zhang A Ijaz P Xiao A Quddus and R Tafazolli ldquoSubbandfiltered multi-carrier systems for multi-service wireless com-municationsrdquo IEEE Transactions on Wireless Communicationsvol 16 no 3 pp 1893ndash1907 2017

[11] A Lizeaga M Mendicute P M Rodriguez and I Val ldquoEval-uation of WCP-COQAM GFDM-OQAM and FBMC-OQAMfor industrial wireless communications with Cognitive Radiordquoin Proceedings of the 2017 IEEE International Workshop ofElectronics Control Measurement Signals and their Applicationto Mechatronics (ECMSM) pp 1ndash6 Donostia Spain May 2017

[12] P M Rodriguez I Val A Lizeaga and M MendicuteldquoEvaluation of cognitive radio for mission-critical and time-critical WSAN in industrial environments under interferencerdquoin Proceedings of the 2015 11th IEEEWorld Conference on FactoryCommunication Systems (WFCS rsquo15) pp 1ndash4 May 2015

[13] P M Rodrıguez R Torrego F Casado et al ldquoDynamicspectrumaccess integrated in awideband cognitiveRF-ethernetbridge for industrial control applicationsrdquo Journal of SignalProcessing Systems vol 83 no 1 pp 19ndash28 2016

[14] A K Somappa K Ovsthus and L Kristensen ldquoAn industrialperspective on wireless sensor networks - a survey of require-ments protocols and challengesrdquo IEEE Communications Sur-veys Tutorials vol 16 pp 1391ndash1412 2014

Journal of Computer Networks and Communications 11

[15] S Vitturi ldquoWireless networks for the factory floor require-ments available technologies research trends and practicalapplicationsrdquo in Proceedings of the IEEE International Confer-ence on Emerging Technologies and Factory Automation 2014

[16] P Popovski ldquoUltra-reliable communication in 5G wirelesssystemsrdquo in Proceedings of the 2014 1st International Conferenceon 5G for Ubiquitous Connectivity (5GU rsquo14) pp 146ndash151November 2014

[17] N Brahmi ON C Yilmaz KWHelmersson S A Ashraf andJ Torsner ldquoDeployment strategies for ultra-reliable and low-latency communication in factory automationrdquo in Proceedingsof the IEEEGlobecomWorkshops (GCWkshps rsquo15) pp 1ndash6 IEEEDecember 2015

[18] O N Yilmaz ldquoUltra-Reliable and Low-Latency 5G Communi-cationrdquo in Proceedings of the European Conference on Networksand Communications (EuCNC rsquo16) 2016

[19] H Lin andP Siohan ldquoFBMCCOQAM an enabler for cognitiveradiordquo in Proceedings of the 4th International Conference onAdvances in Cognitive Radio (COCORA rsquo14) pp 8097ndash8101May2014

[20] H Lin and P Siohan ldquoAn advanced multi-carrier modulationfor future radio systemsrdquo in Proceedings of the 2014 IEEE Inter-national Conference on Acoustics Speech and Signal Processing(ICASSP rsquo14) pp 8097ndash8101 May 2014

[21] H Lin and P Siohan ldquoMulti-carrier modulation analysis andWCP-COQAMproposalrdquoEurasip Journal onAdvances in SignalProcessing vol 2014 no 1 article 79 2014

[22] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations chapter 8 JohnWiley amp Sons Ltd ChichesterUK 2016

[23] J J Benedetto C Heil and D F Walnut Gabor Systems and theBalian-LowTheorem Birkhauser chapter 2 BostonMass USA1998

[24] A Ijaz L Zhang P Xiao and R TafazolliAnalysis of CandidateWaveforms for 5G Cellular Systems chapter 1 InTech 2016

[25] R Gerzaguet N Bartzoudis L G Baltar et al ldquoThe 5Gcandidate waveform race a comparison of complexity andperformancerdquo Eurasip Journal onWireless Communications andNetworking vol 2017 no 1 article 13 2017

[26] A B Ucuncu and A O Yilmaz ldquoPulse shaping meth-ods for OQAMOFDM and WCP-COQAMrdquo httpsarxivorgabs150900977

[27] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations JohnWiley amp Sons Ltd Chichester UK 2016

[28] L Zhang P Xiao A Zafar A ul Quddus and R TafazollildquoFBMC system an insight into doubly dispersive channelimpactrdquo IEEE Transactions on Vehicular Technology vol 66 no5 pp 3942ndash3956 2017

Journal of Computer Networks and Communications 5

s＄－-１－(k) SP

CPinsertion

times

times

times

w(0)

w(1)

w(MK+ L０ minus 1)

Q

Q

Q +

+

zminus1

zminus1

zminus1

s７０-C１－(k)

Figure 5 Block diagram of the cyclic prefix insertion and windowing operation for WCP-COQAM A WCP-COQAM signal is made froma GFDM-OQAM signal and a CP and a window function are applied

WI

CP MK

Block l

Block l minus 1 Block l + 1

GI M ７） ）

middot middot middot middot middot middot

Figure 6 Overlapping of adjacent WCP-COQAM blocks The regions affected by the window of one block are overlapped with the regionsaffected by the window of the previous and the next blocks These regions correspond to the first and the last 119871WI samples of each block

original GFDM-OQAM block a cyclic shift is carried outwhich can be expressed as

1199041015840GFDM-OQAM (119896) = 119904 [mod (119896 + 119871WI119872119870)] (11)

where 119896 = 0 119872119870minus 1 1199041015840GFDM-OQAM(119896) is the estimation ofthe original 119904WCP-COQAM(119896) signal and 119904(119896) is the signal afterthe removal of the windowing effect and the CP

3 Orthogonality Condition and SpectralEfficiency Analysis of WCP-COQAM

Several analyses on 5Gmodulation candidates are available inthe bibliography [24 25] some of them even includingWCP-COQAM modulation scheme [26 27] However to the bestof our knowledge none of them address the implications ofwindowing in the WCP-COQAM waveform as we do here

In this section we contribute some additional detailsregarding how windowing affects the orthogonality againstmultipath channels and the SE in WCP-COQAM Specifi-cally we state the condition that the GI region of the CPextension must fulfil in order to provide full orthogonality inthe presence of multipath channels Besides we also compareWCP-COQAM and GFDM-OQAM (due to the similaritiesbetween both MCM systems) in terms of SE and we showthat in this aspect the former is outperformed by the latter ifequal multipath protection is assumed

31 Orthogonality Condition In WCP-COQAM multipathchannel protection is directly related to the CP extensionand also to the windowing process In order to explain thisfact we show in detail the steps to form a WCP-COQAMblock starting from a basic GFDM-OQAM block (as seen inSection 23)

Once a GFDM-OQAMblock is formed the CP extensionis added to it Figure 8(a) shows the details of this first stagein the construction of WCP-COQAM signal where119872 is thetotal number of subcarriers 119870 is the number of symbols perblock WI is the region of length 119871WI devoted to windowingand overlapping GI is the region of length 119871GI that protectsagainst multipath channel effect and the CP extension hasa total length of 119871CP = 119871WI + 119871GI CP

1015840 WI1015840 and GI1015840 arethose last samples of the block which are copied in orderto form the CP extension As is represented in Figure 8(a)before the windowing is applied this signal can be consideredas a GFDM-OQAM block with CP extension

Figure 8(b) represents the WCP-COQAM block rightafter windowing is applied As shown in Figure 6 and in (9)and (10) the window function is applied on the first andthe last 119871WI samples so that these samples are somehowdistorted Because of that CPWI and GI are no longer equalto CP1015840 WI1015840 and GI1015840 respectively

Since the last 119871WI samples are affected by the window wedefine a new region for them which we call GI21015840 Thus now

6 Journal of Computer Networks and Communications

SP

CP amp lastL２）

symbolsremoval

s７０-C１－(k)

s (0)

s(L） minus 1)

s(L）)

s(MK + L） minus 1)

s(MK + L）)

s(MK + L０ minus 1)

Cyclicshift

s＄－-１－(0)

s＄－-１－(1)

s＄－-１－(MK minus 1)

Figure 7 Block diagram of one GFDM-OQAM block recovery process from a WCP-COQAM block

the original GI and GI1015840 regions are divided into GI1 and GI2and GI11015840 and GI21015840 respectively so that WI = WI1015840 GI1 =GI11015840 and GI2 = GI21015840 As for the lengths of these regions119871WI = 119871WI1015840 = 119871GI2 = 119871GI21015840 and 119871GI1 = 119871GI11015840 Note that GI1and GI11015840 are now the only really redundant parts Because ofthat only GI1 and not the whole GI (as it is explained in [21])acts as a real guard interval against multipath channels whenfrequency domain equalization (FDE) is performed Hence119871GI1 ge 119871CH minus 1 must be fulfilled in order to maintain fullorthogonality through a multipath channel where 119871CH is thetransmission channel length Thus considering that 119871GI2 =119871WI and so 119871GI1 = 119871GI minus119871GI2 = 119871GI minus119871WI we can define thefull orthogonality condition for WCP-COQAM as

119871GI ge 119871CH minus 1 + 119871WI (12)

It is worth mentioning that the FDE should be performed forsamples between GI2 and GI11015840 inclusive because that is thesection corresponding to the cyclic convolution between GI1and GI11015840 with the transmission channel

32 Spectral Efficiency Analysis Figure 9 shows how over-lapping between adjacent blocks is performed in WCP-COQAM Every windowed region is overlapped with thenext or the previous block so that the WI region of eachblock is overlapped with the GI21015840 region of the previousblock Thus the samples within WI regions do not supposean overload and they do not affect the SE Based on this factin [21] the authors claim that the use of the window function

has no effect on SE However this statement is not preciseAlthough the samples inWI do not directly affect SE becausethey overlap with the samples of GI21015840 of the previous blockwindowing implies adding theGI2 region in order to fulfil theorthogonality condition defined in (12) Note that the use ofthe window function and overlapping ofWI andGI21015840 regionsdistort those samples in GI21015840 region This and the fact thatFDE is performed between GI2 and GI11015840 inclusive implythat the GI2 region is necessary in order to recover thosesamples originally placed in the GI21015840 region At this pointwe conclude that both GI1 and GI2 are essential parts withinthe GI region of WCP-COQAM

In order to make a fair analysis of SE in WCP-COQAMwe compare it to GFDM-OQAM because both modula-tion schemes only differ in windowing and overlappingprocesses In order to maintain a homogeneous notationbetween GFDM-OQAM andWCP-COQAM we assume thefollowing considerations (for the sake of readability fromthis point on parameters corresponding to OFDM will beexpressed with the subindex 119874 parameters correspondingto GFDM-OQAM will be expressed with the subindex 119866and parameters corresponding to WCP-COQAM will beexpressed with the subindex119882)

CP119866 = GI119866 = GI1119866CP119882 = [WI119882GI119882] GI119882 = [GI1119882GI2119882]

(13)

Journal of Computer Networks and Communications 7

WI

CP

(GFDM-OQAM block with CP extension)

GI M

MK

(GFDM-OQAM block without CP extension)

Block l

CP

７） ）

(a)

WI GI2

Block l

GI1 ７） ）1 ）2

(b)

Figure 8Windowing operation of aWCP-COQAM block (a) AWCP-COQAM block before windowing is equal to a GFDM-OQAM blockwith CP extension (b) The first and the last 119871WI samples of a WCP-COQAM block are distorted by the window

WI GI2

(block l minus 1)

Block l

(block l)GI1

WI GI2(block l + 1)

GI1

(block l)

Block l minus 1 Block l + 1

７）

middot middot middot middot middot middot

）1 ）2

７） ）1 ）2

Figure 9 Overlapping between adjacent WCP-COQAM blocks and CP extensionrsquos structure Only GI1 and GI11015840 remain equal in order toemulate a circular convolution with the channel and perform frequency domain equalization so GI2 implies an overload that affects SE

The criterion we applied for this comparison is to provideequal protection against multipath channels for both modu-lations so that 119871GI1119866 = 119871GI1119882 Thus

SE119866 = 119872119870119871GI119866 +119872119870 = 119872119870

119871GI1119866 +119872119870

SE119882 = 119872119870119871GI119882 +119872119870 = 119872119870

119871GI1119882 + 119871GI2119882 +119872119870(14)

From this analysis it is clear that if full orthogonalityis to be maintained the windowing process brings an extraGI2 region to the CP which has effects on the SE of a WCP-COQAM signal In particular SE is reduced by a factor of(119871GI1 + 119871GI2 +119872119870)(119871GI1 +119872119870) compared to a windowlesssystem with equal multipath channel protection

4 Simulation Results and Discussion

In this section we show and discuss the results obtainedfrom the simulations of the systems previously describedBased on these results we assess the performances of thesimulated systems in terms of BER PSD and SE In ouranalysis we prioritize robustness against highly dispersivechannels so that we use the BER of transmissions throughRayleigh fading multipath channels in order to assess therobustness of the MCM systems In addition to BER we alsotake into consideration the PSD of the transmitted signalsfor it is an important factor when it comes to CR systemsRegarding SE we use it to represent the cost of the CPextensions windowing process and convolution tails causedby the prototype filters in FBMC-OQAM so that we canobtain a fair comparison between different MCM systems

8 Journal of Computer Networks and Communications

Table 1 Basic simulation parameters Unless opposite is stated the results shown in this work have been obtained under the conditionsdefined by these parameters

OFDM FBMC-OQAM GFDM-OQAM WCP-COQAMSignal bandwidth 20MHzIQ modulation QPSK119872 (subcarriers) 512119870 (symbolsblock overlapping factor) 1 4 4 4119871CP (samples) 32 mdash 32 224119871WI (samples) mdash mdash mdash 96119871GI (samples) 32 mdash 32 128119871GI1 (samples) mdash mdash mdash 32119871GI2 (samples) mdash mdash mdash 96Prototype filter mdash PHYDYAS [5]119871119901 (samples) mdash 2048

Since we are evaluating MCM systems we considerit appropriate to compare the analysed FMC modulationtechniques with an OFDM reference system in order toassess their performance Table 1 shows the configurationparameters we used in most of our simulations We chooseparameters suitable for low-band transmissions in ultrareli-able and low-latency indoor scenarios For that we consideraspects like subcarrier bandwidth and transmission channelrsquoscoherence bandwidth data block duration and windowlength for PSD improvement

41 Robustness against Multipath Channels In this subsec-tion we analyse the robustness of each MCM system againstmultipath channels In order to make a fair comparisonwe simulate every system with equal protection againstmultipath channels matching effective GI lengths (119871GI119874 119871GI119866 and 119871GI1119882) These parameters are selected as shown inTable 1

The equalization technique we use in our simulations isone-tap zero-forcing (ZF) FDE for all the MCM systemsWhile OFDM GFDM-OQAM and WCP-COQAM employCP extensions to prevent ISI caused by the transmissionchannel FBMC-OQAM could be considered a special casein this aspect Since it is not a blockwise modulation andno CP extension is used one-tap FDE is not a straightfor-ward equalization technique for FBMC-OQAM Howeverdue to the subcarrier bandwidth and symbol duration nearto transmission channelrsquos coherence bandwidth and timerespectively we consider the one-tap FDE as a suitableequalization method for FBMC-OQAM in this case An in-depth analysis about the doubly dispersive channelsrsquo impacton FBMC systems is given in [28]

We simulated transmissions through a Rayleigh fadingchannel model with exponential power delay profile (PDP)whose root mean square (rms) delay spread and channellength are 119905rms = 185 ns and 119871CH = 22 taps Besides wesimulate uncoded and coded communications with turbocodes soft decoding and a code rate of 13 In Figure 10

BER

100

10minus1

10minus2

10minus3

10minus4

10minus5

10minus6

0 5 10 15 20 25 30 35

OFDM (uncoded)QFDM (coded)FBMC-OQAM (coded) WCP-COQAM (coded)GFDM-OQAM (coded)

FBMC-OQAM (uncoded)WCP-COQAM (uncoded)GFDM-OQAM (uncoded)

EbN0 (dB)

Figure 10 BER curves of the analysedMCM systems in a multipathchannel We employed turbo encodingdecoding with soft decodingand a code rate of 13

we compare the BER versus 1198641198871198730 of the MCM systemsover these channelmodels assuming perfect synchronizationand full channel state information (CSI) We got these BERcurves by simulating transmissions through 104Monte Carloiterations for each MCM system and each channel modelWe transmit 12288 random data bits and we generate newrandom channels at each Monte Carlo iteration

Here we show that for uncoded transmissions in termsof error rate FBMC-OQAMperforms slightly worse than therest of MCM systems under the simulated conditions Whileat low 1198641198871198730 values every MCM system provides similarBER an error floor appears at high 1198641198871198730 values because ofthe lack of GI in FBMC-OQAMOn the other hand although

Journal of Computer Networks and Communications 9

Frequency normalised to subcarrier number

minus60

minus50

minus40

minus30

minus20

minus10

0

10

Mag

nitu

de (d

B)

minus30 minus20 minus10 0 10 20 30

OFDMFBMC-OQAM

WCP-COQAMGFDM-OQAM

Figure 11 PSD of the analysedMCM systems with some subcarriersdeactivated Only the central 32 subcarriers out of 512 are activated

OFDM GFDM-OQAM andWCP-COQAMprovide similarperformances in terms of BER providing equal multipatheffect protection OFDM slightly outperforms the other twoMCM schemes due to its perfect orthogonality

Finally it is worth noting that for coded transmissionsall the simulated MCM systems present similar BER curvesso that they provide similar robustness against multipathchannels

42 Power Spectral Density Analysis PSD is a relevant char-acteristic in order to design a CR communication systemsuitable for scenarios with high spectrum occupation In thissubsection we compare the spectrums of the signals of theanalysed MCM systems in order to assess their potentialsuitability for CR based systems

Figure 11 shows the spectrums of OFDM GFDM-OQAM WCP-COQAM and FBMC-OQAM signals one ontop of the other For this test we only activate the 32 centralsubcarriers in order to make the power leakage into adjacentsubcarriers visible

As shown in Figure 11 the FB based MCM systems carrysignificant reduction of OOB radiation compared to OFDMHowever even between these MCM schemes there is stilla remarkable difference In this sense FBMC-OQAM out-performs both GFDM-OQAM and WCP-COQAM On theother hand thanks to the windowing WCP-COQAM showsconsiderably lower OOB radiation than GFDM-OQAM Itis worth mentioning that the longer the WI region is thenarrower WCP-COQAM signalrsquos spectrum gets Since extrasamples within the GI region are needed for windowing asdescribed in Section 3 PSD improvement of WCP-COQAMover GFDM-OQAM comes at the cost of lower SE Moredetailed analysis on SE is given in Section 43

43 Spectral Efficiency Analysis In this subsection we anal-yse SE in order to evaluate the cost of obtaining higherrobustness against dispersive channels and better PSD bymeans of CP extensions and windowing For this analysis weconsider the parameters defined in Table 1 so that effectivemultipath channel protection is equal for OFDM GFDM-OQAM and WCP-COQAM Equations (15) show the SEvalues of the three blockwise MCM systems

SE119874 = 119872119870119874119871GI119874 +119872119870119874 = 09412

SE119866 = 119872119870119866119871GI119866 +119872119870119866 = 09846

SE119882 = 119872119870119882119871GI119882 +119872119870119882 = 09412

(15)

Here we show a particular case with some specific parame-ters On one hand GFDM-OQAM presents higher SE thanOFDM because its GI precedes 119870 symbols while in OFDMthe GI precedes only one symbol On the other hand becauseof the windowing operation the CP extension of WCP-COQAM specifically the GI part has 119871GI2 extra sampleswith respect to GFDM-OQAM and that is the reason whyWCP-COQAM also presents lower SE than GFDM-OQAMHowever under these conditions with a high number ofsubcarriers we can conclude that the differences in the SEbetweenOFDMGFDM-OQAM andWCP-COQAMare notreally significant

Regarding FBMC-OQAM once again we consider itdifferent from the rest of MCM systems because it is nota blockwise modulation and it uses no GI protection formultipath effect So its SE depends on the length of thetransmitted frames with respect to the convolution tailsFor long data transmissions its SE will tend to 1 while forshort data transmissions it should tend to 05 assuming aminimum amount of data samples equivalent to a GFDM-OQAM orWCP-COQAM blockThis calculation is simple ifwe introduce 4 complex data symbols (8 OQAM symbols) in(2)

119886119898 (119899) =7

sum119896=0

119909119898 (119896) 120575 (119899 minus 1198961198722 ) 119899 isin [0 81198722 minus 1] (16)

and we introduce these upsampled data in (3) The resultant119910(119899) signal will contain 4095 samples of which half willcorrespond to actual data and the rest to convolution tailsfrom subcarrier filtering

Considering the low-latency scenarios our researchfocuses on SE might suppose a significant drawback forFBMC-OQAM compared to GFDM-OQAM and WCP-COQAM during short data transmissions

5 Conclusions

In this article we make a comparison between three candi-dates for PHY layer modulation in 5G and a reference OFDM

10 Journal of Computer Networks and Communications

system in order to assess their suitability for industrial wire-less communications based on CR For that we simulate low-band communications and highly dispersive indoor channelscenarios Under these conditions we show the performancesand features of these MCM systems from different points ofview

Regarding robustness against multipath channels weconclude that all the analysed MCM systems provide similarperformances under the simulated conditions Although foruncoded transmissions FBMC-OQAM presents an errorfloor at high 1198641198871198730 values this error floor is corrected whencoding is used

In terms of PSD and OOB radiation these filtered MCMschemes provide more restrained spectrum than OFDM Inparticular FBMC-OQAM outperforms the rest of the modu-lation schemes in this aspect so it could be a suitable candi-date for the CR scenarios considered in our research On theother hand although not as good as FBMC-OQAM WCP-COQAM still has considerably better PSD than GFDM-OQAM which only provides less than 10 dB of differencein OOB radiation with respect to OFDM Therefore weconclude that circular filtering only is not sufficient andwindowing must be applied in order to ensure more efficientuse of the spectrum Moreover we consider that due to itsalso restrained spectrum WCP-COQAM might be anothersuitable modulation scheme for CR applications

As for the SE analysis as discussed in Section 43the difference between OFDM GFDM-OQAM and WCP-COQAM is not really significant for the parameters weused in our simulations On the other hand FBMC-OQAMmight suffer from severe degradation of SE during short datatransmissions in the low-latency scenarios we consider in ourresearch

6 Future Work

In this work we considered perfect synchronization andfull channel state information at the receiver but it is alsoimportant to consider more realistic simulation conditionsin the future So the next step in our investigation will beto analyse and propose time-frequency synchronization andchannel estimation methods for GFDM-OQAM and WCP-COQAM

The other key point in our research will be to simulatechannelmodels that fairly emulate industrial wireless channelconditions For that we will simulate severe multipath con-ditions time-varying channels impulsive noise and interfer-ence from other wireless systems

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

This work was partly supported by TIPOTRANS(ETORTEK) and CIIRCOS (PC2013-68) projects ofthe Basque Government (Spain) and by COWITRACC

(TEC2014-59490-C2-1-P2-P) project of the Spanish Minis-try of Economy and Competitiveness

References

[1] H Zhang D L Ruyet M Terre et al ldquoApplication of theFBMC physical layer in a Cognitive Radio scenariordquo Tech RepPHYDYAS-Physical Layer for Dynamic Access and CognitiveRadio 2009

[2] I Gaspar and GWunder ldquo5G cellular communications scenar-ios and system requirementsrdquo Tech Rep 5th Generation Non-Orthogonal Waveforms for Asynchronous Signalling 2013

[3] P Popovski V Braun H Mayer et al ldquoScenarios requirementsandKPIs for 5Gmobile andwireless systemrdquo Tech RepMobileand wireless communications Enablers for the Twenty-twentyInformation Society (METIS) 2013

[4] X Mestre D Gregoratti M Renfors et al ldquoICT-EMPhAtiCfinal project reportrdquo Tech Rep Enhanced Multicarrier Tech-niques for Professional Ad-Hoc and Cell-Based Communica-tions 2015

[5] A Viholainen M Bellanger and M Huchard ldquoPrototypefilter and structure optimizationrdquo PHYDYAS-Physical Layer forDynamic Access and Cognitive Radio 2009

[6] N Michailow M Matthe I S Gaspar et al ldquoGeneralizedfrequency division multiplexing for 5th generation cellularnetworksrdquo IEEE Transactions on Communications vol 62 no9 pp 3045ndash3061 2014

[7] V Vakilian T Wild F Schaich S Ten Brink and J-F FrigonldquoUniversal-filtered multi-carrier technique for wireless systemsbeyond LTErdquo in Proceedings of the 2013 IEEE Globecom Work-shops GC Wkshps 2013 pp 223ndash228 IEEE December 2013

[8] M Agiwal A Roy and N Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 18 no 3 pp 1617ndash1655 2016

[9] X Zhang M Jia L Chen J Ma and J Qiu ldquoFiltered-OFDM-enabler for flexible waveform in the 5th generation cellular net-worksrdquo in Proceedings of the 58th IEEE Global CommunicationsConference (GLOBECOM rsquo15) pp 1ndash6 IEEE December 2015

[10] L Zhang A Ijaz P Xiao A Quddus and R Tafazolli ldquoSubbandfiltered multi-carrier systems for multi-service wireless com-municationsrdquo IEEE Transactions on Wireless Communicationsvol 16 no 3 pp 1893ndash1907 2017

[11] A Lizeaga M Mendicute P M Rodriguez and I Val ldquoEval-uation of WCP-COQAM GFDM-OQAM and FBMC-OQAMfor industrial wireless communications with Cognitive Radiordquoin Proceedings of the 2017 IEEE International Workshop ofElectronics Control Measurement Signals and their Applicationto Mechatronics (ECMSM) pp 1ndash6 Donostia Spain May 2017

[12] P M Rodriguez I Val A Lizeaga and M MendicuteldquoEvaluation of cognitive radio for mission-critical and time-critical WSAN in industrial environments under interferencerdquoin Proceedings of the 2015 11th IEEEWorld Conference on FactoryCommunication Systems (WFCS rsquo15) pp 1ndash4 May 2015

[13] P M Rodrıguez R Torrego F Casado et al ldquoDynamicspectrumaccess integrated in awideband cognitiveRF-ethernetbridge for industrial control applicationsrdquo Journal of SignalProcessing Systems vol 83 no 1 pp 19ndash28 2016

[14] A K Somappa K Ovsthus and L Kristensen ldquoAn industrialperspective on wireless sensor networks - a survey of require-ments protocols and challengesrdquo IEEE Communications Sur-veys Tutorials vol 16 pp 1391ndash1412 2014

Journal of Computer Networks and Communications 11

[15] S Vitturi ldquoWireless networks for the factory floor require-ments available technologies research trends and practicalapplicationsrdquo in Proceedings of the IEEE International Confer-ence on Emerging Technologies and Factory Automation 2014

[16] P Popovski ldquoUltra-reliable communication in 5G wirelesssystemsrdquo in Proceedings of the 2014 1st International Conferenceon 5G for Ubiquitous Connectivity (5GU rsquo14) pp 146ndash151November 2014

[17] N Brahmi ON C Yilmaz KWHelmersson S A Ashraf andJ Torsner ldquoDeployment strategies for ultra-reliable and low-latency communication in factory automationrdquo in Proceedingsof the IEEEGlobecomWorkshops (GCWkshps rsquo15) pp 1ndash6 IEEEDecember 2015

[18] O N Yilmaz ldquoUltra-Reliable and Low-Latency 5G Communi-cationrdquo in Proceedings of the European Conference on Networksand Communications (EuCNC rsquo16) 2016

[19] H Lin andP Siohan ldquoFBMCCOQAM an enabler for cognitiveradiordquo in Proceedings of the 4th International Conference onAdvances in Cognitive Radio (COCORA rsquo14) pp 8097ndash8101May2014

[20] H Lin and P Siohan ldquoAn advanced multi-carrier modulationfor future radio systemsrdquo in Proceedings of the 2014 IEEE Inter-national Conference on Acoustics Speech and Signal Processing(ICASSP rsquo14) pp 8097ndash8101 May 2014

[21] H Lin and P Siohan ldquoMulti-carrier modulation analysis andWCP-COQAMproposalrdquoEurasip Journal onAdvances in SignalProcessing vol 2014 no 1 article 79 2014

[22] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations chapter 8 JohnWiley amp Sons Ltd ChichesterUK 2016

[23] J J Benedetto C Heil and D F Walnut Gabor Systems and theBalian-LowTheorem Birkhauser chapter 2 BostonMass USA1998

[24] A Ijaz L Zhang P Xiao and R TafazolliAnalysis of CandidateWaveforms for 5G Cellular Systems chapter 1 InTech 2016

[25] R Gerzaguet N Bartzoudis L G Baltar et al ldquoThe 5Gcandidate waveform race a comparison of complexity andperformancerdquo Eurasip Journal onWireless Communications andNetworking vol 2017 no 1 article 13 2017

[26] A B Ucuncu and A O Yilmaz ldquoPulse shaping meth-ods for OQAMOFDM and WCP-COQAMrdquo httpsarxivorgabs150900977

[27] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations JohnWiley amp Sons Ltd Chichester UK 2016

[28] L Zhang P Xiao A Zafar A ul Quddus and R TafazollildquoFBMC system an insight into doubly dispersive channelimpactrdquo IEEE Transactions on Vehicular Technology vol 66 no5 pp 3942ndash3956 2017

6 Journal of Computer Networks and Communications

SP

CP amp lastL２）

symbolsremoval

s７０-C１－(k)

s (0)

s(L） minus 1)

s(L）)

s(MK + L） minus 1)

s(MK + L）)

s(MK + L０ minus 1)

Cyclicshift

s＄－-１－(0)

s＄－-１－(1)

s＄－-１－(MK minus 1)

Figure 7 Block diagram of one GFDM-OQAM block recovery process from a WCP-COQAM block

the original GI and GI1015840 regions are divided into GI1 and GI2and GI11015840 and GI21015840 respectively so that WI = WI1015840 GI1 =GI11015840 and GI2 = GI21015840 As for the lengths of these regions119871WI = 119871WI1015840 = 119871GI2 = 119871GI21015840 and 119871GI1 = 119871GI11015840 Note that GI1and GI11015840 are now the only really redundant parts Because ofthat only GI1 and not the whole GI (as it is explained in [21])acts as a real guard interval against multipath channels whenfrequency domain equalization (FDE) is performed Hence119871GI1 ge 119871CH minus 1 must be fulfilled in order to maintain fullorthogonality through a multipath channel where 119871CH is thetransmission channel length Thus considering that 119871GI2 =119871WI and so 119871GI1 = 119871GI minus119871GI2 = 119871GI minus119871WI we can define thefull orthogonality condition for WCP-COQAM as

119871GI ge 119871CH minus 1 + 119871WI (12)

It is worth mentioning that the FDE should be performed forsamples between GI2 and GI11015840 inclusive because that is thesection corresponding to the cyclic convolution between GI1and GI11015840 with the transmission channel

32 Spectral Efficiency Analysis Figure 9 shows how over-lapping between adjacent blocks is performed in WCP-COQAM Every windowed region is overlapped with thenext or the previous block so that the WI region of eachblock is overlapped with the GI21015840 region of the previousblock Thus the samples within WI regions do not supposean overload and they do not affect the SE Based on this factin [21] the authors claim that the use of the window function

has no effect on SE However this statement is not preciseAlthough the samples inWI do not directly affect SE becausethey overlap with the samples of GI21015840 of the previous blockwindowing implies adding theGI2 region in order to fulfil theorthogonality condition defined in (12) Note that the use ofthe window function and overlapping ofWI andGI21015840 regionsdistort those samples in GI21015840 region This and the fact thatFDE is performed between GI2 and GI11015840 inclusive implythat the GI2 region is necessary in order to recover thosesamples originally placed in the GI21015840 region At this pointwe conclude that both GI1 and GI2 are essential parts withinthe GI region of WCP-COQAM

In order to make a fair analysis of SE in WCP-COQAMwe compare it to GFDM-OQAM because both modula-tion schemes only differ in windowing and overlappingprocesses In order to maintain a homogeneous notationbetween GFDM-OQAM andWCP-COQAM we assume thefollowing considerations (for the sake of readability fromthis point on parameters corresponding to OFDM will beexpressed with the subindex 119874 parameters correspondingto GFDM-OQAM will be expressed with the subindex 119866and parameters corresponding to WCP-COQAM will beexpressed with the subindex119882)

CP119866 = GI119866 = GI1119866CP119882 = [WI119882GI119882] GI119882 = [GI1119882GI2119882]

(13)

Journal of Computer Networks and Communications 7

WI

CP

(GFDM-OQAM block with CP extension)

GI M

MK

(GFDM-OQAM block without CP extension)

Block l

CP

７） ）

(a)

WI GI2

Block l

GI1 ７） ）1 ）2

(b)

Figure 8Windowing operation of aWCP-COQAM block (a) AWCP-COQAM block before windowing is equal to a GFDM-OQAM blockwith CP extension (b) The first and the last 119871WI samples of a WCP-COQAM block are distorted by the window

WI GI2

(block l minus 1)

Block l

(block l)GI1

WI GI2(block l + 1)

GI1

(block l)

Block l minus 1 Block l + 1

７）

middot middot middot middot middot middot

）1 ）2

７） ）1 ）2

Figure 9 Overlapping between adjacent WCP-COQAM blocks and CP extensionrsquos structure Only GI1 and GI11015840 remain equal in order toemulate a circular convolution with the channel and perform frequency domain equalization so GI2 implies an overload that affects SE

The criterion we applied for this comparison is to provideequal protection against multipath channels for both modu-lations so that 119871GI1119866 = 119871GI1119882 Thus

SE119866 = 119872119870119871GI119866 +119872119870 = 119872119870

119871GI1119866 +119872119870

SE119882 = 119872119870119871GI119882 +119872119870 = 119872119870

119871GI1119882 + 119871GI2119882 +119872119870(14)

From this analysis it is clear that if full orthogonalityis to be maintained the windowing process brings an extraGI2 region to the CP which has effects on the SE of a WCP-COQAM signal In particular SE is reduced by a factor of(119871GI1 + 119871GI2 +119872119870)(119871GI1 +119872119870) compared to a windowlesssystem with equal multipath channel protection

4 Simulation Results and Discussion

In this section we show and discuss the results obtainedfrom the simulations of the systems previously describedBased on these results we assess the performances of thesimulated systems in terms of BER PSD and SE In ouranalysis we prioritize robustness against highly dispersivechannels so that we use the BER of transmissions throughRayleigh fading multipath channels in order to assess therobustness of the MCM systems In addition to BER we alsotake into consideration the PSD of the transmitted signalsfor it is an important factor when it comes to CR systemsRegarding SE we use it to represent the cost of the CPextensions windowing process and convolution tails causedby the prototype filters in FBMC-OQAM so that we canobtain a fair comparison between different MCM systems

8 Journal of Computer Networks and Communications

Table 1 Basic simulation parameters Unless opposite is stated the results shown in this work have been obtained under the conditionsdefined by these parameters

OFDM FBMC-OQAM GFDM-OQAM WCP-COQAMSignal bandwidth 20MHzIQ modulation QPSK119872 (subcarriers) 512119870 (symbolsblock overlapping factor) 1 4 4 4119871CP (samples) 32 mdash 32 224119871WI (samples) mdash mdash mdash 96119871GI (samples) 32 mdash 32 128119871GI1 (samples) mdash mdash mdash 32119871GI2 (samples) mdash mdash mdash 96Prototype filter mdash PHYDYAS [5]119871119901 (samples) mdash 2048

Since we are evaluating MCM systems we considerit appropriate to compare the analysed FMC modulationtechniques with an OFDM reference system in order toassess their performance Table 1 shows the configurationparameters we used in most of our simulations We chooseparameters suitable for low-band transmissions in ultrareli-able and low-latency indoor scenarios For that we consideraspects like subcarrier bandwidth and transmission channelrsquoscoherence bandwidth data block duration and windowlength for PSD improvement

41 Robustness against Multipath Channels In this subsec-tion we analyse the robustness of each MCM system againstmultipath channels In order to make a fair comparisonwe simulate every system with equal protection againstmultipath channels matching effective GI lengths (119871GI119874 119871GI119866 and 119871GI1119882) These parameters are selected as shown inTable 1

The equalization technique we use in our simulations isone-tap zero-forcing (ZF) FDE for all the MCM systemsWhile OFDM GFDM-OQAM and WCP-COQAM employCP extensions to prevent ISI caused by the transmissionchannel FBMC-OQAM could be considered a special casein this aspect Since it is not a blockwise modulation andno CP extension is used one-tap FDE is not a straightfor-ward equalization technique for FBMC-OQAM Howeverdue to the subcarrier bandwidth and symbol duration nearto transmission channelrsquos coherence bandwidth and timerespectively we consider the one-tap FDE as a suitableequalization method for FBMC-OQAM in this case An in-depth analysis about the doubly dispersive channelsrsquo impacton FBMC systems is given in [28]

We simulated transmissions through a Rayleigh fadingchannel model with exponential power delay profile (PDP)whose root mean square (rms) delay spread and channellength are 119905rms = 185 ns and 119871CH = 22 taps Besides wesimulate uncoded and coded communications with turbocodes soft decoding and a code rate of 13 In Figure 10

BER

100

10minus1

10minus2

10minus3

10minus4

10minus5

10minus6

0 5 10 15 20 25 30 35

OFDM (uncoded)QFDM (coded)FBMC-OQAM (coded) WCP-COQAM (coded)GFDM-OQAM (coded)

FBMC-OQAM (uncoded)WCP-COQAM (uncoded)GFDM-OQAM (uncoded)

EbN0 (dB)

Figure 10 BER curves of the analysedMCM systems in a multipathchannel We employed turbo encodingdecoding with soft decodingand a code rate of 13

we compare the BER versus 1198641198871198730 of the MCM systemsover these channelmodels assuming perfect synchronizationand full channel state information (CSI) We got these BERcurves by simulating transmissions through 104Monte Carloiterations for each MCM system and each channel modelWe transmit 12288 random data bits and we generate newrandom channels at each Monte Carlo iteration

Here we show that for uncoded transmissions in termsof error rate FBMC-OQAMperforms slightly worse than therest of MCM systems under the simulated conditions Whileat low 1198641198871198730 values every MCM system provides similarBER an error floor appears at high 1198641198871198730 values because ofthe lack of GI in FBMC-OQAMOn the other hand although

Journal of Computer Networks and Communications 9

Frequency normalised to subcarrier number

minus60

minus50

minus40

minus30

minus20

minus10

0

10

Mag

nitu

de (d

B)

minus30 minus20 minus10 0 10 20 30

OFDMFBMC-OQAM

WCP-COQAMGFDM-OQAM

Figure 11 PSD of the analysedMCM systems with some subcarriersdeactivated Only the central 32 subcarriers out of 512 are activated

OFDM GFDM-OQAM andWCP-COQAMprovide similarperformances in terms of BER providing equal multipatheffect protection OFDM slightly outperforms the other twoMCM schemes due to its perfect orthogonality

Finally it is worth noting that for coded transmissionsall the simulated MCM systems present similar BER curvesso that they provide similar robustness against multipathchannels

42 Power Spectral Density Analysis PSD is a relevant char-acteristic in order to design a CR communication systemsuitable for scenarios with high spectrum occupation In thissubsection we compare the spectrums of the signals of theanalysed MCM systems in order to assess their potentialsuitability for CR based systems

Figure 11 shows the spectrums of OFDM GFDM-OQAM WCP-COQAM and FBMC-OQAM signals one ontop of the other For this test we only activate the 32 centralsubcarriers in order to make the power leakage into adjacentsubcarriers visible

As shown in Figure 11 the FB based MCM systems carrysignificant reduction of OOB radiation compared to OFDMHowever even between these MCM schemes there is stilla remarkable difference In this sense FBMC-OQAM out-performs both GFDM-OQAM and WCP-COQAM On theother hand thanks to the windowing WCP-COQAM showsconsiderably lower OOB radiation than GFDM-OQAM Itis worth mentioning that the longer the WI region is thenarrower WCP-COQAM signalrsquos spectrum gets Since extrasamples within the GI region are needed for windowing asdescribed in Section 3 PSD improvement of WCP-COQAMover GFDM-OQAM comes at the cost of lower SE Moredetailed analysis on SE is given in Section 43

43 Spectral Efficiency Analysis In this subsection we anal-yse SE in order to evaluate the cost of obtaining higherrobustness against dispersive channels and better PSD bymeans of CP extensions and windowing For this analysis weconsider the parameters defined in Table 1 so that effectivemultipath channel protection is equal for OFDM GFDM-OQAM and WCP-COQAM Equations (15) show the SEvalues of the three blockwise MCM systems

SE119874 = 119872119870119874119871GI119874 +119872119870119874 = 09412

SE119866 = 119872119870119866119871GI119866 +119872119870119866 = 09846

SE119882 = 119872119870119882119871GI119882 +119872119870119882 = 09412

(15)

Here we show a particular case with some specific parame-ters On one hand GFDM-OQAM presents higher SE thanOFDM because its GI precedes 119870 symbols while in OFDMthe GI precedes only one symbol On the other hand becauseof the windowing operation the CP extension of WCP-COQAM specifically the GI part has 119871GI2 extra sampleswith respect to GFDM-OQAM and that is the reason whyWCP-COQAM also presents lower SE than GFDM-OQAMHowever under these conditions with a high number ofsubcarriers we can conclude that the differences in the SEbetweenOFDMGFDM-OQAM andWCP-COQAMare notreally significant

Regarding FBMC-OQAM once again we consider itdifferent from the rest of MCM systems because it is nota blockwise modulation and it uses no GI protection formultipath effect So its SE depends on the length of thetransmitted frames with respect to the convolution tailsFor long data transmissions its SE will tend to 1 while forshort data transmissions it should tend to 05 assuming aminimum amount of data samples equivalent to a GFDM-OQAM orWCP-COQAM blockThis calculation is simple ifwe introduce 4 complex data symbols (8 OQAM symbols) in(2)

119886119898 (119899) =7

sum119896=0

119909119898 (119896) 120575 (119899 minus 1198961198722 ) 119899 isin [0 81198722 minus 1] (16)

and we introduce these upsampled data in (3) The resultant119910(119899) signal will contain 4095 samples of which half willcorrespond to actual data and the rest to convolution tailsfrom subcarrier filtering

Considering the low-latency scenarios our researchfocuses on SE might suppose a significant drawback forFBMC-OQAM compared to GFDM-OQAM and WCP-COQAM during short data transmissions

5 Conclusions

In this article we make a comparison between three candi-dates for PHY layer modulation in 5G and a reference OFDM

10 Journal of Computer Networks and Communications

system in order to assess their suitability for industrial wire-less communications based on CR For that we simulate low-band communications and highly dispersive indoor channelscenarios Under these conditions we show the performancesand features of these MCM systems from different points ofview

Regarding robustness against multipath channels weconclude that all the analysed MCM systems provide similarperformances under the simulated conditions Although foruncoded transmissions FBMC-OQAM presents an errorfloor at high 1198641198871198730 values this error floor is corrected whencoding is used

In terms of PSD and OOB radiation these filtered MCMschemes provide more restrained spectrum than OFDM Inparticular FBMC-OQAM outperforms the rest of the modu-lation schemes in this aspect so it could be a suitable candi-date for the CR scenarios considered in our research On theother hand although not as good as FBMC-OQAM WCP-COQAM still has considerably better PSD than GFDM-OQAM which only provides less than 10 dB of differencein OOB radiation with respect to OFDM Therefore weconclude that circular filtering only is not sufficient andwindowing must be applied in order to ensure more efficientuse of the spectrum Moreover we consider that due to itsalso restrained spectrum WCP-COQAM might be anothersuitable modulation scheme for CR applications

As for the SE analysis as discussed in Section 43the difference between OFDM GFDM-OQAM and WCP-COQAM is not really significant for the parameters weused in our simulations On the other hand FBMC-OQAMmight suffer from severe degradation of SE during short datatransmissions in the low-latency scenarios we consider in ourresearch

6 Future Work

In this work we considered perfect synchronization andfull channel state information at the receiver but it is alsoimportant to consider more realistic simulation conditionsin the future So the next step in our investigation will beto analyse and propose time-frequency synchronization andchannel estimation methods for GFDM-OQAM and WCP-COQAM

The other key point in our research will be to simulatechannelmodels that fairly emulate industrial wireless channelconditions For that we will simulate severe multipath con-ditions time-varying channels impulsive noise and interfer-ence from other wireless systems

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

This work was partly supported by TIPOTRANS(ETORTEK) and CIIRCOS (PC2013-68) projects ofthe Basque Government (Spain) and by COWITRACC

(TEC2014-59490-C2-1-P2-P) project of the Spanish Minis-try of Economy and Competitiveness

References

[1] H Zhang D L Ruyet M Terre et al ldquoApplication of theFBMC physical layer in a Cognitive Radio scenariordquo Tech RepPHYDYAS-Physical Layer for Dynamic Access and CognitiveRadio 2009

[2] I Gaspar and GWunder ldquo5G cellular communications scenar-ios and system requirementsrdquo Tech Rep 5th Generation Non-Orthogonal Waveforms for Asynchronous Signalling 2013

[3] P Popovski V Braun H Mayer et al ldquoScenarios requirementsandKPIs for 5Gmobile andwireless systemrdquo Tech RepMobileand wireless communications Enablers for the Twenty-twentyInformation Society (METIS) 2013

[4] X Mestre D Gregoratti M Renfors et al ldquoICT-EMPhAtiCfinal project reportrdquo Tech Rep Enhanced Multicarrier Tech-niques for Professional Ad-Hoc and Cell-Based Communica-tions 2015

[5] A Viholainen M Bellanger and M Huchard ldquoPrototypefilter and structure optimizationrdquo PHYDYAS-Physical Layer forDynamic Access and Cognitive Radio 2009

[6] N Michailow M Matthe I S Gaspar et al ldquoGeneralizedfrequency division multiplexing for 5th generation cellularnetworksrdquo IEEE Transactions on Communications vol 62 no9 pp 3045ndash3061 2014

[7] V Vakilian T Wild F Schaich S Ten Brink and J-F FrigonldquoUniversal-filtered multi-carrier technique for wireless systemsbeyond LTErdquo in Proceedings of the 2013 IEEE Globecom Work-shops GC Wkshps 2013 pp 223ndash228 IEEE December 2013

[8] M Agiwal A Roy and N Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 18 no 3 pp 1617ndash1655 2016

[9] X Zhang M Jia L Chen J Ma and J Qiu ldquoFiltered-OFDM-enabler for flexible waveform in the 5th generation cellular net-worksrdquo in Proceedings of the 58th IEEE Global CommunicationsConference (GLOBECOM rsquo15) pp 1ndash6 IEEE December 2015

[10] L Zhang A Ijaz P Xiao A Quddus and R Tafazolli ldquoSubbandfiltered multi-carrier systems for multi-service wireless com-municationsrdquo IEEE Transactions on Wireless Communicationsvol 16 no 3 pp 1893ndash1907 2017

[11] A Lizeaga M Mendicute P M Rodriguez and I Val ldquoEval-uation of WCP-COQAM GFDM-OQAM and FBMC-OQAMfor industrial wireless communications with Cognitive Radiordquoin Proceedings of the 2017 IEEE International Workshop ofElectronics Control Measurement Signals and their Applicationto Mechatronics (ECMSM) pp 1ndash6 Donostia Spain May 2017

[12] P M Rodriguez I Val A Lizeaga and M MendicuteldquoEvaluation of cognitive radio for mission-critical and time-critical WSAN in industrial environments under interferencerdquoin Proceedings of the 2015 11th IEEEWorld Conference on FactoryCommunication Systems (WFCS rsquo15) pp 1ndash4 May 2015

[13] P M Rodrıguez R Torrego F Casado et al ldquoDynamicspectrumaccess integrated in awideband cognitiveRF-ethernetbridge for industrial control applicationsrdquo Journal of SignalProcessing Systems vol 83 no 1 pp 19ndash28 2016

[14] A K Somappa K Ovsthus and L Kristensen ldquoAn industrialperspective on wireless sensor networks - a survey of require-ments protocols and challengesrdquo IEEE Communications Sur-veys Tutorials vol 16 pp 1391ndash1412 2014

Journal of Computer Networks and Communications 11

[15] S Vitturi ldquoWireless networks for the factory floor require-ments available technologies research trends and practicalapplicationsrdquo in Proceedings of the IEEE International Confer-ence on Emerging Technologies and Factory Automation 2014

[16] P Popovski ldquoUltra-reliable communication in 5G wirelesssystemsrdquo in Proceedings of the 2014 1st International Conferenceon 5G for Ubiquitous Connectivity (5GU rsquo14) pp 146ndash151November 2014

[17] N Brahmi ON C Yilmaz KWHelmersson S A Ashraf andJ Torsner ldquoDeployment strategies for ultra-reliable and low-latency communication in factory automationrdquo in Proceedingsof the IEEEGlobecomWorkshops (GCWkshps rsquo15) pp 1ndash6 IEEEDecember 2015

[18] O N Yilmaz ldquoUltra-Reliable and Low-Latency 5G Communi-cationrdquo in Proceedings of the European Conference on Networksand Communications (EuCNC rsquo16) 2016

[19] H Lin andP Siohan ldquoFBMCCOQAM an enabler for cognitiveradiordquo in Proceedings of the 4th International Conference onAdvances in Cognitive Radio (COCORA rsquo14) pp 8097ndash8101May2014

[20] H Lin and P Siohan ldquoAn advanced multi-carrier modulationfor future radio systemsrdquo in Proceedings of the 2014 IEEE Inter-national Conference on Acoustics Speech and Signal Processing(ICASSP rsquo14) pp 8097ndash8101 May 2014

[21] H Lin and P Siohan ldquoMulti-carrier modulation analysis andWCP-COQAMproposalrdquoEurasip Journal onAdvances in SignalProcessing vol 2014 no 1 article 79 2014

[22] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations chapter 8 JohnWiley amp Sons Ltd ChichesterUK 2016

[23] J J Benedetto C Heil and D F Walnut Gabor Systems and theBalian-LowTheorem Birkhauser chapter 2 BostonMass USA1998

[24] A Ijaz L Zhang P Xiao and R TafazolliAnalysis of CandidateWaveforms for 5G Cellular Systems chapter 1 InTech 2016

[25] R Gerzaguet N Bartzoudis L G Baltar et al ldquoThe 5Gcandidate waveform race a comparison of complexity andperformancerdquo Eurasip Journal onWireless Communications andNetworking vol 2017 no 1 article 13 2017

[26] A B Ucuncu and A O Yilmaz ldquoPulse shaping meth-ods for OQAMOFDM and WCP-COQAMrdquo httpsarxivorgabs150900977

[27] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations JohnWiley amp Sons Ltd Chichester UK 2016

[28] L Zhang P Xiao A Zafar A ul Quddus and R TafazollildquoFBMC system an insight into doubly dispersive channelimpactrdquo IEEE Transactions on Vehicular Technology vol 66 no5 pp 3942ndash3956 2017

Journal of Computer Networks and Communications 7

WI

CP

(GFDM-OQAM block with CP extension)

GI M

MK

(GFDM-OQAM block without CP extension)

Block l

CP

７） ）

(a)

WI GI2

Block l

GI1 ７） ）1 ）2

(b)

Figure 8Windowing operation of aWCP-COQAM block (a) AWCP-COQAM block before windowing is equal to a GFDM-OQAM blockwith CP extension (b) The first and the last 119871WI samples of a WCP-COQAM block are distorted by the window

WI GI2

(block l minus 1)

Block l

(block l)GI1

WI GI2(block l + 1)

GI1

(block l)

Block l minus 1 Block l + 1

７）

middot middot middot middot middot middot

）1 ）2

７） ）1 ）2

Figure 9 Overlapping between adjacent WCP-COQAM blocks and CP extensionrsquos structure Only GI1 and GI11015840 remain equal in order toemulate a circular convolution with the channel and perform frequency domain equalization so GI2 implies an overload that affects SE

The criterion we applied for this comparison is to provideequal protection against multipath channels for both modu-lations so that 119871GI1119866 = 119871GI1119882 Thus

SE119866 = 119872119870119871GI119866 +119872119870 = 119872119870

119871GI1119866 +119872119870

SE119882 = 119872119870119871GI119882 +119872119870 = 119872119870

119871GI1119882 + 119871GI2119882 +119872119870(14)

From this analysis it is clear that if full orthogonalityis to be maintained the windowing process brings an extraGI2 region to the CP which has effects on the SE of a WCP-COQAM signal In particular SE is reduced by a factor of(119871GI1 + 119871GI2 +119872119870)(119871GI1 +119872119870) compared to a windowlesssystem with equal multipath channel protection

4 Simulation Results and Discussion

In this section we show and discuss the results obtainedfrom the simulations of the systems previously describedBased on these results we assess the performances of thesimulated systems in terms of BER PSD and SE In ouranalysis we prioritize robustness against highly dispersivechannels so that we use the BER of transmissions throughRayleigh fading multipath channels in order to assess therobustness of the MCM systems In addition to BER we alsotake into consideration the PSD of the transmitted signalsfor it is an important factor when it comes to CR systemsRegarding SE we use it to represent the cost of the CPextensions windowing process and convolution tails causedby the prototype filters in FBMC-OQAM so that we canobtain a fair comparison between different MCM systems

8 Journal of Computer Networks and Communications

Table 1 Basic simulation parameters Unless opposite is stated the results shown in this work have been obtained under the conditionsdefined by these parameters

OFDM FBMC-OQAM GFDM-OQAM WCP-COQAMSignal bandwidth 20MHzIQ modulation QPSK119872 (subcarriers) 512119870 (symbolsblock overlapping factor) 1 4 4 4119871CP (samples) 32 mdash 32 224119871WI (samples) mdash mdash mdash 96119871GI (samples) 32 mdash 32 128119871GI1 (samples) mdash mdash mdash 32119871GI2 (samples) mdash mdash mdash 96Prototype filter mdash PHYDYAS [5]119871119901 (samples) mdash 2048

Since we are evaluating MCM systems we considerit appropriate to compare the analysed FMC modulationtechniques with an OFDM reference system in order toassess their performance Table 1 shows the configurationparameters we used in most of our simulations We chooseparameters suitable for low-band transmissions in ultrareli-able and low-latency indoor scenarios For that we consideraspects like subcarrier bandwidth and transmission channelrsquoscoherence bandwidth data block duration and windowlength for PSD improvement

41 Robustness against Multipath Channels In this subsec-tion we analyse the robustness of each MCM system againstmultipath channels In order to make a fair comparisonwe simulate every system with equal protection againstmultipath channels matching effective GI lengths (119871GI119874 119871GI119866 and 119871GI1119882) These parameters are selected as shown inTable 1

The equalization technique we use in our simulations isone-tap zero-forcing (ZF) FDE for all the MCM systemsWhile OFDM GFDM-OQAM and WCP-COQAM employCP extensions to prevent ISI caused by the transmissionchannel FBMC-OQAM could be considered a special casein this aspect Since it is not a blockwise modulation andno CP extension is used one-tap FDE is not a straightfor-ward equalization technique for FBMC-OQAM Howeverdue to the subcarrier bandwidth and symbol duration nearto transmission channelrsquos coherence bandwidth and timerespectively we consider the one-tap FDE as a suitableequalization method for FBMC-OQAM in this case An in-depth analysis about the doubly dispersive channelsrsquo impacton FBMC systems is given in [28]

We simulated transmissions through a Rayleigh fadingchannel model with exponential power delay profile (PDP)whose root mean square (rms) delay spread and channellength are 119905rms = 185 ns and 119871CH = 22 taps Besides wesimulate uncoded and coded communications with turbocodes soft decoding and a code rate of 13 In Figure 10

BER

100

10minus1

10minus2

10minus3

10minus4

10minus5

10minus6

0 5 10 15 20 25 30 35

OFDM (uncoded)QFDM (coded)FBMC-OQAM (coded) WCP-COQAM (coded)GFDM-OQAM (coded)

FBMC-OQAM (uncoded)WCP-COQAM (uncoded)GFDM-OQAM (uncoded)

EbN0 (dB)

Figure 10 BER curves of the analysedMCM systems in a multipathchannel We employed turbo encodingdecoding with soft decodingand a code rate of 13

we compare the BER versus 1198641198871198730 of the MCM systemsover these channelmodels assuming perfect synchronizationand full channel state information (CSI) We got these BERcurves by simulating transmissions through 104Monte Carloiterations for each MCM system and each channel modelWe transmit 12288 random data bits and we generate newrandom channels at each Monte Carlo iteration

Here we show that for uncoded transmissions in termsof error rate FBMC-OQAMperforms slightly worse than therest of MCM systems under the simulated conditions Whileat low 1198641198871198730 values every MCM system provides similarBER an error floor appears at high 1198641198871198730 values because ofthe lack of GI in FBMC-OQAMOn the other hand although

Journal of Computer Networks and Communications 9

Frequency normalised to subcarrier number

minus60

minus50

minus40

minus30

minus20

minus10

0

10

Mag

nitu

de (d

B)

minus30 minus20 minus10 0 10 20 30

OFDMFBMC-OQAM

WCP-COQAMGFDM-OQAM

Figure 11 PSD of the analysedMCM systems with some subcarriersdeactivated Only the central 32 subcarriers out of 512 are activated

OFDM GFDM-OQAM andWCP-COQAMprovide similarperformances in terms of BER providing equal multipatheffect protection OFDM slightly outperforms the other twoMCM schemes due to its perfect orthogonality

Finally it is worth noting that for coded transmissionsall the simulated MCM systems present similar BER curvesso that they provide similar robustness against multipathchannels

42 Power Spectral Density Analysis PSD is a relevant char-acteristic in order to design a CR communication systemsuitable for scenarios with high spectrum occupation In thissubsection we compare the spectrums of the signals of theanalysed MCM systems in order to assess their potentialsuitability for CR based systems

Figure 11 shows the spectrums of OFDM GFDM-OQAM WCP-COQAM and FBMC-OQAM signals one ontop of the other For this test we only activate the 32 centralsubcarriers in order to make the power leakage into adjacentsubcarriers visible

As shown in Figure 11 the FB based MCM systems carrysignificant reduction of OOB radiation compared to OFDMHowever even between these MCM schemes there is stilla remarkable difference In this sense FBMC-OQAM out-performs both GFDM-OQAM and WCP-COQAM On theother hand thanks to the windowing WCP-COQAM showsconsiderably lower OOB radiation than GFDM-OQAM Itis worth mentioning that the longer the WI region is thenarrower WCP-COQAM signalrsquos spectrum gets Since extrasamples within the GI region are needed for windowing asdescribed in Section 3 PSD improvement of WCP-COQAMover GFDM-OQAM comes at the cost of lower SE Moredetailed analysis on SE is given in Section 43

43 Spectral Efficiency Analysis In this subsection we anal-yse SE in order to evaluate the cost of obtaining higherrobustness against dispersive channels and better PSD bymeans of CP extensions and windowing For this analysis weconsider the parameters defined in Table 1 so that effectivemultipath channel protection is equal for OFDM GFDM-OQAM and WCP-COQAM Equations (15) show the SEvalues of the three blockwise MCM systems

SE119874 = 119872119870119874119871GI119874 +119872119870119874 = 09412

SE119866 = 119872119870119866119871GI119866 +119872119870119866 = 09846

SE119882 = 119872119870119882119871GI119882 +119872119870119882 = 09412

(15)

Here we show a particular case with some specific parame-ters On one hand GFDM-OQAM presents higher SE thanOFDM because its GI precedes 119870 symbols while in OFDMthe GI precedes only one symbol On the other hand becauseof the windowing operation the CP extension of WCP-COQAM specifically the GI part has 119871GI2 extra sampleswith respect to GFDM-OQAM and that is the reason whyWCP-COQAM also presents lower SE than GFDM-OQAMHowever under these conditions with a high number ofsubcarriers we can conclude that the differences in the SEbetweenOFDMGFDM-OQAM andWCP-COQAMare notreally significant

Regarding FBMC-OQAM once again we consider itdifferent from the rest of MCM systems because it is nota blockwise modulation and it uses no GI protection formultipath effect So its SE depends on the length of thetransmitted frames with respect to the convolution tailsFor long data transmissions its SE will tend to 1 while forshort data transmissions it should tend to 05 assuming aminimum amount of data samples equivalent to a GFDM-OQAM orWCP-COQAM blockThis calculation is simple ifwe introduce 4 complex data symbols (8 OQAM symbols) in(2)

119886119898 (119899) =7

sum119896=0

119909119898 (119896) 120575 (119899 minus 1198961198722 ) 119899 isin [0 81198722 minus 1] (16)

and we introduce these upsampled data in (3) The resultant119910(119899) signal will contain 4095 samples of which half willcorrespond to actual data and the rest to convolution tailsfrom subcarrier filtering

Considering the low-latency scenarios our researchfocuses on SE might suppose a significant drawback forFBMC-OQAM compared to GFDM-OQAM and WCP-COQAM during short data transmissions

5 Conclusions

In this article we make a comparison between three candi-dates for PHY layer modulation in 5G and a reference OFDM

10 Journal of Computer Networks and Communications

system in order to assess their suitability for industrial wire-less communications based on CR For that we simulate low-band communications and highly dispersive indoor channelscenarios Under these conditions we show the performancesand features of these MCM systems from different points ofview

Regarding robustness against multipath channels weconclude that all the analysed MCM systems provide similarperformances under the simulated conditions Although foruncoded transmissions FBMC-OQAM presents an errorfloor at high 1198641198871198730 values this error floor is corrected whencoding is used

In terms of PSD and OOB radiation these filtered MCMschemes provide more restrained spectrum than OFDM Inparticular FBMC-OQAM outperforms the rest of the modu-lation schemes in this aspect so it could be a suitable candi-date for the CR scenarios considered in our research On theother hand although not as good as FBMC-OQAM WCP-COQAM still has considerably better PSD than GFDM-OQAM which only provides less than 10 dB of differencein OOB radiation with respect to OFDM Therefore weconclude that circular filtering only is not sufficient andwindowing must be applied in order to ensure more efficientuse of the spectrum Moreover we consider that due to itsalso restrained spectrum WCP-COQAM might be anothersuitable modulation scheme for CR applications

As for the SE analysis as discussed in Section 43the difference between OFDM GFDM-OQAM and WCP-COQAM is not really significant for the parameters weused in our simulations On the other hand FBMC-OQAMmight suffer from severe degradation of SE during short datatransmissions in the low-latency scenarios we consider in ourresearch

6 Future Work

In this work we considered perfect synchronization andfull channel state information at the receiver but it is alsoimportant to consider more realistic simulation conditionsin the future So the next step in our investigation will beto analyse and propose time-frequency synchronization andchannel estimation methods for GFDM-OQAM and WCP-COQAM

The other key point in our research will be to simulatechannelmodels that fairly emulate industrial wireless channelconditions For that we will simulate severe multipath con-ditions time-varying channels impulsive noise and interfer-ence from other wireless systems

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

This work was partly supported by TIPOTRANS(ETORTEK) and CIIRCOS (PC2013-68) projects ofthe Basque Government (Spain) and by COWITRACC

(TEC2014-59490-C2-1-P2-P) project of the Spanish Minis-try of Economy and Competitiveness

References

[1] H Zhang D L Ruyet M Terre et al ldquoApplication of theFBMC physical layer in a Cognitive Radio scenariordquo Tech RepPHYDYAS-Physical Layer for Dynamic Access and CognitiveRadio 2009

[2] I Gaspar and GWunder ldquo5G cellular communications scenar-ios and system requirementsrdquo Tech Rep 5th Generation Non-Orthogonal Waveforms for Asynchronous Signalling 2013

[3] P Popovski V Braun H Mayer et al ldquoScenarios requirementsandKPIs for 5Gmobile andwireless systemrdquo Tech RepMobileand wireless communications Enablers for the Twenty-twentyInformation Society (METIS) 2013

[4] X Mestre D Gregoratti M Renfors et al ldquoICT-EMPhAtiCfinal project reportrdquo Tech Rep Enhanced Multicarrier Tech-niques for Professional Ad-Hoc and Cell-Based Communica-tions 2015

[5] A Viholainen M Bellanger and M Huchard ldquoPrototypefilter and structure optimizationrdquo PHYDYAS-Physical Layer forDynamic Access and Cognitive Radio 2009

[6] N Michailow M Matthe I S Gaspar et al ldquoGeneralizedfrequency division multiplexing for 5th generation cellularnetworksrdquo IEEE Transactions on Communications vol 62 no9 pp 3045ndash3061 2014

[7] V Vakilian T Wild F Schaich S Ten Brink and J-F FrigonldquoUniversal-filtered multi-carrier technique for wireless systemsbeyond LTErdquo in Proceedings of the 2013 IEEE Globecom Work-shops GC Wkshps 2013 pp 223ndash228 IEEE December 2013

[8] M Agiwal A Roy and N Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 18 no 3 pp 1617ndash1655 2016

[9] X Zhang M Jia L Chen J Ma and J Qiu ldquoFiltered-OFDM-enabler for flexible waveform in the 5th generation cellular net-worksrdquo in Proceedings of the 58th IEEE Global CommunicationsConference (GLOBECOM rsquo15) pp 1ndash6 IEEE December 2015

[10] L Zhang A Ijaz P Xiao A Quddus and R Tafazolli ldquoSubbandfiltered multi-carrier systems for multi-service wireless com-municationsrdquo IEEE Transactions on Wireless Communicationsvol 16 no 3 pp 1893ndash1907 2017

[11] A Lizeaga M Mendicute P M Rodriguez and I Val ldquoEval-uation of WCP-COQAM GFDM-OQAM and FBMC-OQAMfor industrial wireless communications with Cognitive Radiordquoin Proceedings of the 2017 IEEE International Workshop ofElectronics Control Measurement Signals and their Applicationto Mechatronics (ECMSM) pp 1ndash6 Donostia Spain May 2017

[12] P M Rodriguez I Val A Lizeaga and M MendicuteldquoEvaluation of cognitive radio for mission-critical and time-critical WSAN in industrial environments under interferencerdquoin Proceedings of the 2015 11th IEEEWorld Conference on FactoryCommunication Systems (WFCS rsquo15) pp 1ndash4 May 2015

[13] P M Rodrıguez R Torrego F Casado et al ldquoDynamicspectrumaccess integrated in awideband cognitiveRF-ethernetbridge for industrial control applicationsrdquo Journal of SignalProcessing Systems vol 83 no 1 pp 19ndash28 2016

[14] A K Somappa K Ovsthus and L Kristensen ldquoAn industrialperspective on wireless sensor networks - a survey of require-ments protocols and challengesrdquo IEEE Communications Sur-veys Tutorials vol 16 pp 1391ndash1412 2014

Journal of Computer Networks and Communications 11

[15] S Vitturi ldquoWireless networks for the factory floor require-ments available technologies research trends and practicalapplicationsrdquo in Proceedings of the IEEE International Confer-ence on Emerging Technologies and Factory Automation 2014

[16] P Popovski ldquoUltra-reliable communication in 5G wirelesssystemsrdquo in Proceedings of the 2014 1st International Conferenceon 5G for Ubiquitous Connectivity (5GU rsquo14) pp 146ndash151November 2014

[17] N Brahmi ON C Yilmaz KWHelmersson S A Ashraf andJ Torsner ldquoDeployment strategies for ultra-reliable and low-latency communication in factory automationrdquo in Proceedingsof the IEEEGlobecomWorkshops (GCWkshps rsquo15) pp 1ndash6 IEEEDecember 2015

[18] O N Yilmaz ldquoUltra-Reliable and Low-Latency 5G Communi-cationrdquo in Proceedings of the European Conference on Networksand Communications (EuCNC rsquo16) 2016

[19] H Lin andP Siohan ldquoFBMCCOQAM an enabler for cognitiveradiordquo in Proceedings of the 4th International Conference onAdvances in Cognitive Radio (COCORA rsquo14) pp 8097ndash8101May2014

[20] H Lin and P Siohan ldquoAn advanced multi-carrier modulationfor future radio systemsrdquo in Proceedings of the 2014 IEEE Inter-national Conference on Acoustics Speech and Signal Processing(ICASSP rsquo14) pp 8097ndash8101 May 2014

[21] H Lin and P Siohan ldquoMulti-carrier modulation analysis andWCP-COQAMproposalrdquoEurasip Journal onAdvances in SignalProcessing vol 2014 no 1 article 79 2014

[22] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations chapter 8 JohnWiley amp Sons Ltd ChichesterUK 2016

[23] J J Benedetto C Heil and D F Walnut Gabor Systems and theBalian-LowTheorem Birkhauser chapter 2 BostonMass USA1998

[24] A Ijaz L Zhang P Xiao and R TafazolliAnalysis of CandidateWaveforms for 5G Cellular Systems chapter 1 InTech 2016

[25] R Gerzaguet N Bartzoudis L G Baltar et al ldquoThe 5Gcandidate waveform race a comparison of complexity andperformancerdquo Eurasip Journal onWireless Communications andNetworking vol 2017 no 1 article 13 2017

[26] A B Ucuncu and A O Yilmaz ldquoPulse shaping meth-ods for OQAMOFDM and WCP-COQAMrdquo httpsarxivorgabs150900977

[27] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations JohnWiley amp Sons Ltd Chichester UK 2016

[28] L Zhang P Xiao A Zafar A ul Quddus and R TafazollildquoFBMC system an insight into doubly dispersive channelimpactrdquo IEEE Transactions on Vehicular Technology vol 66 no5 pp 3942ndash3956 2017

8 Journal of Computer Networks and Communications

Table 1 Basic simulation parameters Unless opposite is stated the results shown in this work have been obtained under the conditionsdefined by these parameters

OFDM FBMC-OQAM GFDM-OQAM WCP-COQAMSignal bandwidth 20MHzIQ modulation QPSK119872 (subcarriers) 512119870 (symbolsblock overlapping factor) 1 4 4 4119871CP (samples) 32 mdash 32 224119871WI (samples) mdash mdash mdash 96119871GI (samples) 32 mdash 32 128119871GI1 (samples) mdash mdash mdash 32119871GI2 (samples) mdash mdash mdash 96Prototype filter mdash PHYDYAS [5]119871119901 (samples) mdash 2048

Since we are evaluating MCM systems we considerit appropriate to compare the analysed FMC modulationtechniques with an OFDM reference system in order toassess their performance Table 1 shows the configurationparameters we used in most of our simulations We chooseparameters suitable for low-band transmissions in ultrareli-able and low-latency indoor scenarios For that we consideraspects like subcarrier bandwidth and transmission channelrsquoscoherence bandwidth data block duration and windowlength for PSD improvement

41 Robustness against Multipath Channels In this subsec-tion we analyse the robustness of each MCM system againstmultipath channels In order to make a fair comparisonwe simulate every system with equal protection againstmultipath channels matching effective GI lengths (119871GI119874 119871GI119866 and 119871GI1119882) These parameters are selected as shown inTable 1

The equalization technique we use in our simulations isone-tap zero-forcing (ZF) FDE for all the MCM systemsWhile OFDM GFDM-OQAM and WCP-COQAM employCP extensions to prevent ISI caused by the transmissionchannel FBMC-OQAM could be considered a special casein this aspect Since it is not a blockwise modulation andno CP extension is used one-tap FDE is not a straightfor-ward equalization technique for FBMC-OQAM Howeverdue to the subcarrier bandwidth and symbol duration nearto transmission channelrsquos coherence bandwidth and timerespectively we consider the one-tap FDE as a suitableequalization method for FBMC-OQAM in this case An in-depth analysis about the doubly dispersive channelsrsquo impacton FBMC systems is given in [28]

We simulated transmissions through a Rayleigh fadingchannel model with exponential power delay profile (PDP)whose root mean square (rms) delay spread and channellength are 119905rms = 185 ns and 119871CH = 22 taps Besides wesimulate uncoded and coded communications with turbocodes soft decoding and a code rate of 13 In Figure 10

BER

100

10minus1

10minus2

10minus3

10minus4

10minus5

10minus6

0 5 10 15 20 25 30 35

OFDM (uncoded)QFDM (coded)FBMC-OQAM (coded) WCP-COQAM (coded)GFDM-OQAM (coded)

FBMC-OQAM (uncoded)WCP-COQAM (uncoded)GFDM-OQAM (uncoded)

EbN0 (dB)

Figure 10 BER curves of the analysedMCM systems in a multipathchannel We employed turbo encodingdecoding with soft decodingand a code rate of 13

we compare the BER versus 1198641198871198730 of the MCM systemsover these channelmodels assuming perfect synchronizationand full channel state information (CSI) We got these BERcurves by simulating transmissions through 104Monte Carloiterations for each MCM system and each channel modelWe transmit 12288 random data bits and we generate newrandom channels at each Monte Carlo iteration

Here we show that for uncoded transmissions in termsof error rate FBMC-OQAMperforms slightly worse than therest of MCM systems under the simulated conditions Whileat low 1198641198871198730 values every MCM system provides similarBER an error floor appears at high 1198641198871198730 values because ofthe lack of GI in FBMC-OQAMOn the other hand although

Journal of Computer Networks and Communications 9

Frequency normalised to subcarrier number

minus60

minus50

minus40

minus30

minus20

minus10

0

10

Mag

nitu

de (d

B)

minus30 minus20 minus10 0 10 20 30

OFDMFBMC-OQAM

WCP-COQAMGFDM-OQAM

Figure 11 PSD of the analysedMCM systems with some subcarriersdeactivated Only the central 32 subcarriers out of 512 are activated

OFDM GFDM-OQAM andWCP-COQAMprovide similarperformances in terms of BER providing equal multipatheffect protection OFDM slightly outperforms the other twoMCM schemes due to its perfect orthogonality

Finally it is worth noting that for coded transmissionsall the simulated MCM systems present similar BER curvesso that they provide similar robustness against multipathchannels

42 Power Spectral Density Analysis PSD is a relevant char-acteristic in order to design a CR communication systemsuitable for scenarios with high spectrum occupation In thissubsection we compare the spectrums of the signals of theanalysed MCM systems in order to assess their potentialsuitability for CR based systems

Figure 11 shows the spectrums of OFDM GFDM-OQAM WCP-COQAM and FBMC-OQAM signals one ontop of the other For this test we only activate the 32 centralsubcarriers in order to make the power leakage into adjacentsubcarriers visible

As shown in Figure 11 the FB based MCM systems carrysignificant reduction of OOB radiation compared to OFDMHowever even between these MCM schemes there is stilla remarkable difference In this sense FBMC-OQAM out-performs both GFDM-OQAM and WCP-COQAM On theother hand thanks to the windowing WCP-COQAM showsconsiderably lower OOB radiation than GFDM-OQAM Itis worth mentioning that the longer the WI region is thenarrower WCP-COQAM signalrsquos spectrum gets Since extrasamples within the GI region are needed for windowing asdescribed in Section 3 PSD improvement of WCP-COQAMover GFDM-OQAM comes at the cost of lower SE Moredetailed analysis on SE is given in Section 43

43 Spectral Efficiency Analysis In this subsection we anal-yse SE in order to evaluate the cost of obtaining higherrobustness against dispersive channels and better PSD bymeans of CP extensions and windowing For this analysis weconsider the parameters defined in Table 1 so that effectivemultipath channel protection is equal for OFDM GFDM-OQAM and WCP-COQAM Equations (15) show the SEvalues of the three blockwise MCM systems

SE119874 = 119872119870119874119871GI119874 +119872119870119874 = 09412

SE119866 = 119872119870119866119871GI119866 +119872119870119866 = 09846

SE119882 = 119872119870119882119871GI119882 +119872119870119882 = 09412

(15)

Here we show a particular case with some specific parame-ters On one hand GFDM-OQAM presents higher SE thanOFDM because its GI precedes 119870 symbols while in OFDMthe GI precedes only one symbol On the other hand becauseof the windowing operation the CP extension of WCP-COQAM specifically the GI part has 119871GI2 extra sampleswith respect to GFDM-OQAM and that is the reason whyWCP-COQAM also presents lower SE than GFDM-OQAMHowever under these conditions with a high number ofsubcarriers we can conclude that the differences in the SEbetweenOFDMGFDM-OQAM andWCP-COQAMare notreally significant

Regarding FBMC-OQAM once again we consider itdifferent from the rest of MCM systems because it is nota blockwise modulation and it uses no GI protection formultipath effect So its SE depends on the length of thetransmitted frames with respect to the convolution tailsFor long data transmissions its SE will tend to 1 while forshort data transmissions it should tend to 05 assuming aminimum amount of data samples equivalent to a GFDM-OQAM orWCP-COQAM blockThis calculation is simple ifwe introduce 4 complex data symbols (8 OQAM symbols) in(2)

119886119898 (119899) =7

sum119896=0

119909119898 (119896) 120575 (119899 minus 1198961198722 ) 119899 isin [0 81198722 minus 1] (16)

and we introduce these upsampled data in (3) The resultant119910(119899) signal will contain 4095 samples of which half willcorrespond to actual data and the rest to convolution tailsfrom subcarrier filtering

Considering the low-latency scenarios our researchfocuses on SE might suppose a significant drawback forFBMC-OQAM compared to GFDM-OQAM and WCP-COQAM during short data transmissions

5 Conclusions

In this article we make a comparison between three candi-dates for PHY layer modulation in 5G and a reference OFDM

10 Journal of Computer Networks and Communications

system in order to assess their suitability for industrial wire-less communications based on CR For that we simulate low-band communications and highly dispersive indoor channelscenarios Under these conditions we show the performancesand features of these MCM systems from different points ofview

Regarding robustness against multipath channels weconclude that all the analysed MCM systems provide similarperformances under the simulated conditions Although foruncoded transmissions FBMC-OQAM presents an errorfloor at high 1198641198871198730 values this error floor is corrected whencoding is used

In terms of PSD and OOB radiation these filtered MCMschemes provide more restrained spectrum than OFDM Inparticular FBMC-OQAM outperforms the rest of the modu-lation schemes in this aspect so it could be a suitable candi-date for the CR scenarios considered in our research On theother hand although not as good as FBMC-OQAM WCP-COQAM still has considerably better PSD than GFDM-OQAM which only provides less than 10 dB of differencein OOB radiation with respect to OFDM Therefore weconclude that circular filtering only is not sufficient andwindowing must be applied in order to ensure more efficientuse of the spectrum Moreover we consider that due to itsalso restrained spectrum WCP-COQAM might be anothersuitable modulation scheme for CR applications

As for the SE analysis as discussed in Section 43the difference between OFDM GFDM-OQAM and WCP-COQAM is not really significant for the parameters weused in our simulations On the other hand FBMC-OQAMmight suffer from severe degradation of SE during short datatransmissions in the low-latency scenarios we consider in ourresearch

6 Future Work

In this work we considered perfect synchronization andfull channel state information at the receiver but it is alsoimportant to consider more realistic simulation conditionsin the future So the next step in our investigation will beto analyse and propose time-frequency synchronization andchannel estimation methods for GFDM-OQAM and WCP-COQAM

The other key point in our research will be to simulatechannelmodels that fairly emulate industrial wireless channelconditions For that we will simulate severe multipath con-ditions time-varying channels impulsive noise and interfer-ence from other wireless systems

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

This work was partly supported by TIPOTRANS(ETORTEK) and CIIRCOS (PC2013-68) projects ofthe Basque Government (Spain) and by COWITRACC

(TEC2014-59490-C2-1-P2-P) project of the Spanish Minis-try of Economy and Competitiveness

References

[1] H Zhang D L Ruyet M Terre et al ldquoApplication of theFBMC physical layer in a Cognitive Radio scenariordquo Tech RepPHYDYAS-Physical Layer for Dynamic Access and CognitiveRadio 2009

[2] I Gaspar and GWunder ldquo5G cellular communications scenar-ios and system requirementsrdquo Tech Rep 5th Generation Non-Orthogonal Waveforms for Asynchronous Signalling 2013

[3] P Popovski V Braun H Mayer et al ldquoScenarios requirementsandKPIs for 5Gmobile andwireless systemrdquo Tech RepMobileand wireless communications Enablers for the Twenty-twentyInformation Society (METIS) 2013

[4] X Mestre D Gregoratti M Renfors et al ldquoICT-EMPhAtiCfinal project reportrdquo Tech Rep Enhanced Multicarrier Tech-niques for Professional Ad-Hoc and Cell-Based Communica-tions 2015

[5] A Viholainen M Bellanger and M Huchard ldquoPrototypefilter and structure optimizationrdquo PHYDYAS-Physical Layer forDynamic Access and Cognitive Radio 2009

[6] N Michailow M Matthe I S Gaspar et al ldquoGeneralizedfrequency division multiplexing for 5th generation cellularnetworksrdquo IEEE Transactions on Communications vol 62 no9 pp 3045ndash3061 2014

[7] V Vakilian T Wild F Schaich S Ten Brink and J-F FrigonldquoUniversal-filtered multi-carrier technique for wireless systemsbeyond LTErdquo in Proceedings of the 2013 IEEE Globecom Work-shops GC Wkshps 2013 pp 223ndash228 IEEE December 2013

[8] M Agiwal A Roy and N Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 18 no 3 pp 1617ndash1655 2016

[9] X Zhang M Jia L Chen J Ma and J Qiu ldquoFiltered-OFDM-enabler for flexible waveform in the 5th generation cellular net-worksrdquo in Proceedings of the 58th IEEE Global CommunicationsConference (GLOBECOM rsquo15) pp 1ndash6 IEEE December 2015

[10] L Zhang A Ijaz P Xiao A Quddus and R Tafazolli ldquoSubbandfiltered multi-carrier systems for multi-service wireless com-municationsrdquo IEEE Transactions on Wireless Communicationsvol 16 no 3 pp 1893ndash1907 2017

[11] A Lizeaga M Mendicute P M Rodriguez and I Val ldquoEval-uation of WCP-COQAM GFDM-OQAM and FBMC-OQAMfor industrial wireless communications with Cognitive Radiordquoin Proceedings of the 2017 IEEE International Workshop ofElectronics Control Measurement Signals and their Applicationto Mechatronics (ECMSM) pp 1ndash6 Donostia Spain May 2017

[12] P M Rodriguez I Val A Lizeaga and M MendicuteldquoEvaluation of cognitive radio for mission-critical and time-critical WSAN in industrial environments under interferencerdquoin Proceedings of the 2015 11th IEEEWorld Conference on FactoryCommunication Systems (WFCS rsquo15) pp 1ndash4 May 2015

[13] P M Rodrıguez R Torrego F Casado et al ldquoDynamicspectrumaccess integrated in awideband cognitiveRF-ethernetbridge for industrial control applicationsrdquo Journal of SignalProcessing Systems vol 83 no 1 pp 19ndash28 2016

[14] A K Somappa K Ovsthus and L Kristensen ldquoAn industrialperspective on wireless sensor networks - a survey of require-ments protocols and challengesrdquo IEEE Communications Sur-veys Tutorials vol 16 pp 1391ndash1412 2014

Journal of Computer Networks and Communications 11

[15] S Vitturi ldquoWireless networks for the factory floor require-ments available technologies research trends and practicalapplicationsrdquo in Proceedings of the IEEE International Confer-ence on Emerging Technologies and Factory Automation 2014

[16] P Popovski ldquoUltra-reliable communication in 5G wirelesssystemsrdquo in Proceedings of the 2014 1st International Conferenceon 5G for Ubiquitous Connectivity (5GU rsquo14) pp 146ndash151November 2014

[17] N Brahmi ON C Yilmaz KWHelmersson S A Ashraf andJ Torsner ldquoDeployment strategies for ultra-reliable and low-latency communication in factory automationrdquo in Proceedingsof the IEEEGlobecomWorkshops (GCWkshps rsquo15) pp 1ndash6 IEEEDecember 2015

[18] O N Yilmaz ldquoUltra-Reliable and Low-Latency 5G Communi-cationrdquo in Proceedings of the European Conference on Networksand Communications (EuCNC rsquo16) 2016

[19] H Lin andP Siohan ldquoFBMCCOQAM an enabler for cognitiveradiordquo in Proceedings of the 4th International Conference onAdvances in Cognitive Radio (COCORA rsquo14) pp 8097ndash8101May2014

[20] H Lin and P Siohan ldquoAn advanced multi-carrier modulationfor future radio systemsrdquo in Proceedings of the 2014 IEEE Inter-national Conference on Acoustics Speech and Signal Processing(ICASSP rsquo14) pp 8097ndash8101 May 2014

[21] H Lin and P Siohan ldquoMulti-carrier modulation analysis andWCP-COQAMproposalrdquoEurasip Journal onAdvances in SignalProcessing vol 2014 no 1 article 79 2014

[22] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations chapter 8 JohnWiley amp Sons Ltd ChichesterUK 2016

[23] J J Benedetto C Heil and D F Walnut Gabor Systems and theBalian-LowTheorem Birkhauser chapter 2 BostonMass USA1998

[24] A Ijaz L Zhang P Xiao and R TafazolliAnalysis of CandidateWaveforms for 5G Cellular Systems chapter 1 InTech 2016

[25] R Gerzaguet N Bartzoudis L G Baltar et al ldquoThe 5Gcandidate waveform race a comparison of complexity andperformancerdquo Eurasip Journal onWireless Communications andNetworking vol 2017 no 1 article 13 2017

[26] A B Ucuncu and A O Yilmaz ldquoPulse shaping meth-ods for OQAMOFDM and WCP-COQAMrdquo httpsarxivorgabs150900977

[27] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations JohnWiley amp Sons Ltd Chichester UK 2016

[28] L Zhang P Xiao A Zafar A ul Quddus and R TafazollildquoFBMC system an insight into doubly dispersive channelimpactrdquo IEEE Transactions on Vehicular Technology vol 66 no5 pp 3942ndash3956 2017

Journal of Computer Networks and Communications 9

Frequency normalised to subcarrier number

minus60

minus50

minus40

minus30

minus20

minus10

0

10

Mag

nitu

de (d

B)

minus30 minus20 minus10 0 10 20 30

OFDMFBMC-OQAM

WCP-COQAMGFDM-OQAM

Figure 11 PSD of the analysedMCM systems with some subcarriersdeactivated Only the central 32 subcarriers out of 512 are activated

OFDM GFDM-OQAM andWCP-COQAMprovide similarperformances in terms of BER providing equal multipatheffect protection OFDM slightly outperforms the other twoMCM schemes due to its perfect orthogonality

Finally it is worth noting that for coded transmissionsall the simulated MCM systems present similar BER curvesso that they provide similar robustness against multipathchannels

42 Power Spectral Density Analysis PSD is a relevant char-acteristic in order to design a CR communication systemsuitable for scenarios with high spectrum occupation In thissubsection we compare the spectrums of the signals of theanalysed MCM systems in order to assess their potentialsuitability for CR based systems

Figure 11 shows the spectrums of OFDM GFDM-OQAM WCP-COQAM and FBMC-OQAM signals one ontop of the other For this test we only activate the 32 centralsubcarriers in order to make the power leakage into adjacentsubcarriers visible

As shown in Figure 11 the FB based MCM systems carrysignificant reduction of OOB radiation compared to OFDMHowever even between these MCM schemes there is stilla remarkable difference In this sense FBMC-OQAM out-performs both GFDM-OQAM and WCP-COQAM On theother hand thanks to the windowing WCP-COQAM showsconsiderably lower OOB radiation than GFDM-OQAM Itis worth mentioning that the longer the WI region is thenarrower WCP-COQAM signalrsquos spectrum gets Since extrasamples within the GI region are needed for windowing asdescribed in Section 3 PSD improvement of WCP-COQAMover GFDM-OQAM comes at the cost of lower SE Moredetailed analysis on SE is given in Section 43

43 Spectral Efficiency Analysis In this subsection we anal-yse SE in order to evaluate the cost of obtaining higherrobustness against dispersive channels and better PSD bymeans of CP extensions and windowing For this analysis weconsider the parameters defined in Table 1 so that effectivemultipath channel protection is equal for OFDM GFDM-OQAM and WCP-COQAM Equations (15) show the SEvalues of the three blockwise MCM systems

SE119874 = 119872119870119874119871GI119874 +119872119870119874 = 09412

SE119866 = 119872119870119866119871GI119866 +119872119870119866 = 09846

SE119882 = 119872119870119882119871GI119882 +119872119870119882 = 09412

(15)

Here we show a particular case with some specific parame-ters On one hand GFDM-OQAM presents higher SE thanOFDM because its GI precedes 119870 symbols while in OFDMthe GI precedes only one symbol On the other hand becauseof the windowing operation the CP extension of WCP-COQAM specifically the GI part has 119871GI2 extra sampleswith respect to GFDM-OQAM and that is the reason whyWCP-COQAM also presents lower SE than GFDM-OQAMHowever under these conditions with a high number ofsubcarriers we can conclude that the differences in the SEbetweenOFDMGFDM-OQAM andWCP-COQAMare notreally significant

Regarding FBMC-OQAM once again we consider itdifferent from the rest of MCM systems because it is nota blockwise modulation and it uses no GI protection formultipath effect So its SE depends on the length of thetransmitted frames with respect to the convolution tailsFor long data transmissions its SE will tend to 1 while forshort data transmissions it should tend to 05 assuming aminimum amount of data samples equivalent to a GFDM-OQAM orWCP-COQAM blockThis calculation is simple ifwe introduce 4 complex data symbols (8 OQAM symbols) in(2)

119886119898 (119899) =7

sum119896=0

119909119898 (119896) 120575 (119899 minus 1198961198722 ) 119899 isin [0 81198722 minus 1] (16)

and we introduce these upsampled data in (3) The resultant119910(119899) signal will contain 4095 samples of which half willcorrespond to actual data and the rest to convolution tailsfrom subcarrier filtering

Considering the low-latency scenarios our researchfocuses on SE might suppose a significant drawback forFBMC-OQAM compared to GFDM-OQAM and WCP-COQAM during short data transmissions

5 Conclusions

In this article we make a comparison between three candi-dates for PHY layer modulation in 5G and a reference OFDM

10 Journal of Computer Networks and Communications

system in order to assess their suitability for industrial wire-less communications based on CR For that we simulate low-band communications and highly dispersive indoor channelscenarios Under these conditions we show the performancesand features of these MCM systems from different points ofview

Regarding robustness against multipath channels weconclude that all the analysed MCM systems provide similarperformances under the simulated conditions Although foruncoded transmissions FBMC-OQAM presents an errorfloor at high 1198641198871198730 values this error floor is corrected whencoding is used

In terms of PSD and OOB radiation these filtered MCMschemes provide more restrained spectrum than OFDM Inparticular FBMC-OQAM outperforms the rest of the modu-lation schemes in this aspect so it could be a suitable candi-date for the CR scenarios considered in our research On theother hand although not as good as FBMC-OQAM WCP-COQAM still has considerably better PSD than GFDM-OQAM which only provides less than 10 dB of differencein OOB radiation with respect to OFDM Therefore weconclude that circular filtering only is not sufficient andwindowing must be applied in order to ensure more efficientuse of the spectrum Moreover we consider that due to itsalso restrained spectrum WCP-COQAM might be anothersuitable modulation scheme for CR applications

As for the SE analysis as discussed in Section 43the difference between OFDM GFDM-OQAM and WCP-COQAM is not really significant for the parameters weused in our simulations On the other hand FBMC-OQAMmight suffer from severe degradation of SE during short datatransmissions in the low-latency scenarios we consider in ourresearch

6 Future Work

In this work we considered perfect synchronization andfull channel state information at the receiver but it is alsoimportant to consider more realistic simulation conditionsin the future So the next step in our investigation will beto analyse and propose time-frequency synchronization andchannel estimation methods for GFDM-OQAM and WCP-COQAM

The other key point in our research will be to simulatechannelmodels that fairly emulate industrial wireless channelconditions For that we will simulate severe multipath con-ditions time-varying channels impulsive noise and interfer-ence from other wireless systems

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

This work was partly supported by TIPOTRANS(ETORTEK) and CIIRCOS (PC2013-68) projects ofthe Basque Government (Spain) and by COWITRACC

(TEC2014-59490-C2-1-P2-P) project of the Spanish Minis-try of Economy and Competitiveness

References

[1] H Zhang D L Ruyet M Terre et al ldquoApplication of theFBMC physical layer in a Cognitive Radio scenariordquo Tech RepPHYDYAS-Physical Layer for Dynamic Access and CognitiveRadio 2009

[2] I Gaspar and GWunder ldquo5G cellular communications scenar-ios and system requirementsrdquo Tech Rep 5th Generation Non-Orthogonal Waveforms for Asynchronous Signalling 2013

[3] P Popovski V Braun H Mayer et al ldquoScenarios requirementsandKPIs for 5Gmobile andwireless systemrdquo Tech RepMobileand wireless communications Enablers for the Twenty-twentyInformation Society (METIS) 2013

[4] X Mestre D Gregoratti M Renfors et al ldquoICT-EMPhAtiCfinal project reportrdquo Tech Rep Enhanced Multicarrier Tech-niques for Professional Ad-Hoc and Cell-Based Communica-tions 2015

[5] A Viholainen M Bellanger and M Huchard ldquoPrototypefilter and structure optimizationrdquo PHYDYAS-Physical Layer forDynamic Access and Cognitive Radio 2009

[6] N Michailow M Matthe I S Gaspar et al ldquoGeneralizedfrequency division multiplexing for 5th generation cellularnetworksrdquo IEEE Transactions on Communications vol 62 no9 pp 3045ndash3061 2014

[7] V Vakilian T Wild F Schaich S Ten Brink and J-F FrigonldquoUniversal-filtered multi-carrier technique for wireless systemsbeyond LTErdquo in Proceedings of the 2013 IEEE Globecom Work-shops GC Wkshps 2013 pp 223ndash228 IEEE December 2013

[8] M Agiwal A Roy and N Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 18 no 3 pp 1617ndash1655 2016

[9] X Zhang M Jia L Chen J Ma and J Qiu ldquoFiltered-OFDM-enabler for flexible waveform in the 5th generation cellular net-worksrdquo in Proceedings of the 58th IEEE Global CommunicationsConference (GLOBECOM rsquo15) pp 1ndash6 IEEE December 2015

[10] L Zhang A Ijaz P Xiao A Quddus and R Tafazolli ldquoSubbandfiltered multi-carrier systems for multi-service wireless com-municationsrdquo IEEE Transactions on Wireless Communicationsvol 16 no 3 pp 1893ndash1907 2017

[11] A Lizeaga M Mendicute P M Rodriguez and I Val ldquoEval-uation of WCP-COQAM GFDM-OQAM and FBMC-OQAMfor industrial wireless communications with Cognitive Radiordquoin Proceedings of the 2017 IEEE International Workshop ofElectronics Control Measurement Signals and their Applicationto Mechatronics (ECMSM) pp 1ndash6 Donostia Spain May 2017

[12] P M Rodriguez I Val A Lizeaga and M MendicuteldquoEvaluation of cognitive radio for mission-critical and time-critical WSAN in industrial environments under interferencerdquoin Proceedings of the 2015 11th IEEEWorld Conference on FactoryCommunication Systems (WFCS rsquo15) pp 1ndash4 May 2015

[13] P M Rodrıguez R Torrego F Casado et al ldquoDynamicspectrumaccess integrated in awideband cognitiveRF-ethernetbridge for industrial control applicationsrdquo Journal of SignalProcessing Systems vol 83 no 1 pp 19ndash28 2016

[14] A K Somappa K Ovsthus and L Kristensen ldquoAn industrialperspective on wireless sensor networks - a survey of require-ments protocols and challengesrdquo IEEE Communications Sur-veys Tutorials vol 16 pp 1391ndash1412 2014

Journal of Computer Networks and Communications 11

[15] S Vitturi ldquoWireless networks for the factory floor require-ments available technologies research trends and practicalapplicationsrdquo in Proceedings of the IEEE International Confer-ence on Emerging Technologies and Factory Automation 2014

[16] P Popovski ldquoUltra-reliable communication in 5G wirelesssystemsrdquo in Proceedings of the 2014 1st International Conferenceon 5G for Ubiquitous Connectivity (5GU rsquo14) pp 146ndash151November 2014

[17] N Brahmi ON C Yilmaz KWHelmersson S A Ashraf andJ Torsner ldquoDeployment strategies for ultra-reliable and low-latency communication in factory automationrdquo in Proceedingsof the IEEEGlobecomWorkshops (GCWkshps rsquo15) pp 1ndash6 IEEEDecember 2015

[18] O N Yilmaz ldquoUltra-Reliable and Low-Latency 5G Communi-cationrdquo in Proceedings of the European Conference on Networksand Communications (EuCNC rsquo16) 2016

[19] H Lin andP Siohan ldquoFBMCCOQAM an enabler for cognitiveradiordquo in Proceedings of the 4th International Conference onAdvances in Cognitive Radio (COCORA rsquo14) pp 8097ndash8101May2014

[20] H Lin and P Siohan ldquoAn advanced multi-carrier modulationfor future radio systemsrdquo in Proceedings of the 2014 IEEE Inter-national Conference on Acoustics Speech and Signal Processing(ICASSP rsquo14) pp 8097ndash8101 May 2014

[21] H Lin and P Siohan ldquoMulti-carrier modulation analysis andWCP-COQAMproposalrdquoEurasip Journal onAdvances in SignalProcessing vol 2014 no 1 article 79 2014

[22] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations chapter 8 JohnWiley amp Sons Ltd ChichesterUK 2016

[23] J J Benedetto C Heil and D F Walnut Gabor Systems and theBalian-LowTheorem Birkhauser chapter 2 BostonMass USA1998

[24] A Ijaz L Zhang P Xiao and R TafazolliAnalysis of CandidateWaveforms for 5G Cellular Systems chapter 1 InTech 2016

[25] R Gerzaguet N Bartzoudis L G Baltar et al ldquoThe 5Gcandidate waveform race a comparison of complexity andperformancerdquo Eurasip Journal onWireless Communications andNetworking vol 2017 no 1 article 13 2017

[26] A B Ucuncu and A O Yilmaz ldquoPulse shaping meth-ods for OQAMOFDM and WCP-COQAMrdquo httpsarxivorgabs150900977

[27] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations JohnWiley amp Sons Ltd Chichester UK 2016

[28] L Zhang P Xiao A Zafar A ul Quddus and R TafazollildquoFBMC system an insight into doubly dispersive channelimpactrdquo IEEE Transactions on Vehicular Technology vol 66 no5 pp 3942ndash3956 2017

10 Journal of Computer Networks and Communications

system in order to assess their suitability for industrial wire-less communications based on CR For that we simulate low-band communications and highly dispersive indoor channelscenarios Under these conditions we show the performancesand features of these MCM systems from different points ofview

Regarding robustness against multipath channels weconclude that all the analysed MCM systems provide similarperformances under the simulated conditions Although foruncoded transmissions FBMC-OQAM presents an errorfloor at high 1198641198871198730 values this error floor is corrected whencoding is used

In terms of PSD and OOB radiation these filtered MCMschemes provide more restrained spectrum than OFDM Inparticular FBMC-OQAM outperforms the rest of the modu-lation schemes in this aspect so it could be a suitable candi-date for the CR scenarios considered in our research On theother hand although not as good as FBMC-OQAM WCP-COQAM still has considerably better PSD than GFDM-OQAM which only provides less than 10 dB of differencein OOB radiation with respect to OFDM Therefore weconclude that circular filtering only is not sufficient andwindowing must be applied in order to ensure more efficientuse of the spectrum Moreover we consider that due to itsalso restrained spectrum WCP-COQAM might be anothersuitable modulation scheme for CR applications

As for the SE analysis as discussed in Section 43the difference between OFDM GFDM-OQAM and WCP-COQAM is not really significant for the parameters weused in our simulations On the other hand FBMC-OQAMmight suffer from severe degradation of SE during short datatransmissions in the low-latency scenarios we consider in ourresearch

6 Future Work

In this work we considered perfect synchronization andfull channel state information at the receiver but it is alsoimportant to consider more realistic simulation conditionsin the future So the next step in our investigation will beto analyse and propose time-frequency synchronization andchannel estimation methods for GFDM-OQAM and WCP-COQAM

The other key point in our research will be to simulatechannelmodels that fairly emulate industrial wireless channelconditions For that we will simulate severe multipath con-ditions time-varying channels impulsive noise and interfer-ence from other wireless systems

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

This work was partly supported by TIPOTRANS(ETORTEK) and CIIRCOS (PC2013-68) projects ofthe Basque Government (Spain) and by COWITRACC

(TEC2014-59490-C2-1-P2-P) project of the Spanish Minis-try of Economy and Competitiveness

References

[1] H Zhang D L Ruyet M Terre et al ldquoApplication of theFBMC physical layer in a Cognitive Radio scenariordquo Tech RepPHYDYAS-Physical Layer for Dynamic Access and CognitiveRadio 2009

[2] I Gaspar and GWunder ldquo5G cellular communications scenar-ios and system requirementsrdquo Tech Rep 5th Generation Non-Orthogonal Waveforms for Asynchronous Signalling 2013

[3] P Popovski V Braun H Mayer et al ldquoScenarios requirementsandKPIs for 5Gmobile andwireless systemrdquo Tech RepMobileand wireless communications Enablers for the Twenty-twentyInformation Society (METIS) 2013

[4] X Mestre D Gregoratti M Renfors et al ldquoICT-EMPhAtiCfinal project reportrdquo Tech Rep Enhanced Multicarrier Tech-niques for Professional Ad-Hoc and Cell-Based Communica-tions 2015

[5] A Viholainen M Bellanger and M Huchard ldquoPrototypefilter and structure optimizationrdquo PHYDYAS-Physical Layer forDynamic Access and Cognitive Radio 2009

[6] N Michailow M Matthe I S Gaspar et al ldquoGeneralizedfrequency division multiplexing for 5th generation cellularnetworksrdquo IEEE Transactions on Communications vol 62 no9 pp 3045ndash3061 2014

[7] V Vakilian T Wild F Schaich S Ten Brink and J-F FrigonldquoUniversal-filtered multi-carrier technique for wireless systemsbeyond LTErdquo in Proceedings of the 2013 IEEE Globecom Work-shops GC Wkshps 2013 pp 223ndash228 IEEE December 2013

[8] M Agiwal A Roy and N Saxena ldquoNext generation 5Gwirelessnetworks A comprehensive surveyrdquo IEEE CommunicationsSurveys and Tutorials vol 18 no 3 pp 1617ndash1655 2016

[9] X Zhang M Jia L Chen J Ma and J Qiu ldquoFiltered-OFDM-enabler for flexible waveform in the 5th generation cellular net-worksrdquo in Proceedings of the 58th IEEE Global CommunicationsConference (GLOBECOM rsquo15) pp 1ndash6 IEEE December 2015

[10] L Zhang A Ijaz P Xiao A Quddus and R Tafazolli ldquoSubbandfiltered multi-carrier systems for multi-service wireless com-municationsrdquo IEEE Transactions on Wireless Communicationsvol 16 no 3 pp 1893ndash1907 2017

[11] A Lizeaga M Mendicute P M Rodriguez and I Val ldquoEval-uation of WCP-COQAM GFDM-OQAM and FBMC-OQAMfor industrial wireless communications with Cognitive Radiordquoin Proceedings of the 2017 IEEE International Workshop ofElectronics Control Measurement Signals and their Applicationto Mechatronics (ECMSM) pp 1ndash6 Donostia Spain May 2017

[12] P M Rodriguez I Val A Lizeaga and M MendicuteldquoEvaluation of cognitive radio for mission-critical and time-critical WSAN in industrial environments under interferencerdquoin Proceedings of the 2015 11th IEEEWorld Conference on FactoryCommunication Systems (WFCS rsquo15) pp 1ndash4 May 2015

[13] P M Rodrıguez R Torrego F Casado et al ldquoDynamicspectrumaccess integrated in awideband cognitiveRF-ethernetbridge for industrial control applicationsrdquo Journal of SignalProcessing Systems vol 83 no 1 pp 19ndash28 2016

[14] A K Somappa K Ovsthus and L Kristensen ldquoAn industrialperspective on wireless sensor networks - a survey of require-ments protocols and challengesrdquo IEEE Communications Sur-veys Tutorials vol 16 pp 1391ndash1412 2014

Journal of Computer Networks and Communications 11

[15] S Vitturi ldquoWireless networks for the factory floor require-ments available technologies research trends and practicalapplicationsrdquo in Proceedings of the IEEE International Confer-ence on Emerging Technologies and Factory Automation 2014

[16] P Popovski ldquoUltra-reliable communication in 5G wirelesssystemsrdquo in Proceedings of the 2014 1st International Conferenceon 5G for Ubiquitous Connectivity (5GU rsquo14) pp 146ndash151November 2014

[17] N Brahmi ON C Yilmaz KWHelmersson S A Ashraf andJ Torsner ldquoDeployment strategies for ultra-reliable and low-latency communication in factory automationrdquo in Proceedingsof the IEEEGlobecomWorkshops (GCWkshps rsquo15) pp 1ndash6 IEEEDecember 2015

[18] O N Yilmaz ldquoUltra-Reliable and Low-Latency 5G Communi-cationrdquo in Proceedings of the European Conference on Networksand Communications (EuCNC rsquo16) 2016

[19] H Lin andP Siohan ldquoFBMCCOQAM an enabler for cognitiveradiordquo in Proceedings of the 4th International Conference onAdvances in Cognitive Radio (COCORA rsquo14) pp 8097ndash8101May2014

[20] H Lin and P Siohan ldquoAn advanced multi-carrier modulationfor future radio systemsrdquo in Proceedings of the 2014 IEEE Inter-national Conference on Acoustics Speech and Signal Processing(ICASSP rsquo14) pp 8097ndash8101 May 2014

[21] H Lin and P Siohan ldquoMulti-carrier modulation analysis andWCP-COQAMproposalrdquoEurasip Journal onAdvances in SignalProcessing vol 2014 no 1 article 79 2014

[22] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations chapter 8 JohnWiley amp Sons Ltd ChichesterUK 2016

[23] J J Benedetto C Heil and D F Walnut Gabor Systems and theBalian-LowTheorem Birkhauser chapter 2 BostonMass USA1998

[24] A Ijaz L Zhang P Xiao and R TafazolliAnalysis of CandidateWaveforms for 5G Cellular Systems chapter 1 InTech 2016

[25] R Gerzaguet N Bartzoudis L G Baltar et al ldquoThe 5Gcandidate waveform race a comparison of complexity andperformancerdquo Eurasip Journal onWireless Communications andNetworking vol 2017 no 1 article 13 2017

[26] A B Ucuncu and A O Yilmaz ldquoPulse shaping meth-ods for OQAMOFDM and WCP-COQAMrdquo httpsarxivorgabs150900977

[27] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations JohnWiley amp Sons Ltd Chichester UK 2016

[28] L Zhang P Xiao A Zafar A ul Quddus and R TafazollildquoFBMC system an insight into doubly dispersive channelimpactrdquo IEEE Transactions on Vehicular Technology vol 66 no5 pp 3942ndash3956 2017

Journal of Computer Networks and Communications 11

[15] S Vitturi ldquoWireless networks for the factory floor require-ments available technologies research trends and practicalapplicationsrdquo in Proceedings of the IEEE International Confer-ence on Emerging Technologies and Factory Automation 2014

[16] P Popovski ldquoUltra-reliable communication in 5G wirelesssystemsrdquo in Proceedings of the 2014 1st International Conferenceon 5G for Ubiquitous Connectivity (5GU rsquo14) pp 146ndash151November 2014

[17] N Brahmi ON C Yilmaz KWHelmersson S A Ashraf andJ Torsner ldquoDeployment strategies for ultra-reliable and low-latency communication in factory automationrdquo in Proceedingsof the IEEEGlobecomWorkshops (GCWkshps rsquo15) pp 1ndash6 IEEEDecember 2015

[18] O N Yilmaz ldquoUltra-Reliable and Low-Latency 5G Communi-cationrdquo in Proceedings of the European Conference on Networksand Communications (EuCNC rsquo16) 2016

[19] H Lin andP Siohan ldquoFBMCCOQAM an enabler for cognitiveradiordquo in Proceedings of the 4th International Conference onAdvances in Cognitive Radio (COCORA rsquo14) pp 8097ndash8101May2014

[20] H Lin and P Siohan ldquoAn advanced multi-carrier modulationfor future radio systemsrdquo in Proceedings of the 2014 IEEE Inter-national Conference on Acoustics Speech and Signal Processing(ICASSP rsquo14) pp 8097ndash8101 May 2014

[21] H Lin and P Siohan ldquoMulti-carrier modulation analysis andWCP-COQAMproposalrdquoEurasip Journal onAdvances in SignalProcessing vol 2014 no 1 article 79 2014

[22] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations chapter 8 JohnWiley amp Sons Ltd ChichesterUK 2016

[23] J J Benedetto C Heil and D F Walnut Gabor Systems and theBalian-LowTheorem Birkhauser chapter 2 BostonMass USA1998

[24] A Ijaz L Zhang P Xiao and R TafazolliAnalysis of CandidateWaveforms for 5G Cellular Systems chapter 1 InTech 2016

[25] R Gerzaguet N Bartzoudis L G Baltar et al ldquoThe 5Gcandidate waveform race a comparison of complexity andperformancerdquo Eurasip Journal onWireless Communications andNetworking vol 2017 no 1 article 13 2017

[26] A B Ucuncu and A O Yilmaz ldquoPulse shaping meth-ods for OQAMOFDM and WCP-COQAMrdquo httpsarxivorgabs150900977

[27] F Luo and C J Zhang Signal Processing for 5G Algorithms andImplementations JohnWiley amp Sons Ltd Chichester UK 2016

[28] L Zhang P Xiao A Zafar A ul Quddus and R TafazollildquoFBMC system an insight into doubly dispersive channelimpactrdquo IEEE Transactions on Vehicular Technology vol 66 no5 pp 3942ndash3956 2017

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Active and Passive Electronic Components

Control Scienceand Engineering

Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

RotatingMachinery

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

Journal of

Volume 201

Submit your manuscripts athttpswwwhindawicom

VLSI Design

Hindawi Publishing Corporationhttpwwwhindawicom Volume 201

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Shock and Vibration

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation httpwwwhindawicom

Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Navigation and Observation

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

DistributedSensor Networks

International Journal of