IEEECOMMUNICATION SURVEYS &TUTORIALS,VOL.17,NO.2

IEEE COMMUNICATION SURVEYS & TUTORIALS, VOL. 17, NO. 2, SECOND QUARTER 2015 627

Performance Improvement and Cost ReductionTechniques for Radio Over Fiber Communications

Varghese Antony Thomas, Mohammed El-Hajjar, and Lajos Hanzo

Abstract—Advanced cost reduction and performance improve-ment techniques conceived for Radio Over Fiber (ROF) commu-nications are considered. ROF techniques are expected to formthe backbone of the future fifth generation of wireless networks.The achievable link performance and the associated deploymentcost constitute the most salient metrics of an ROF architecture. Inthis paper, we commence by providing a rudimentary overview ofthe ROF architecture, and then we elaborate on ROF techniquesdesigned for improving the attainable system performance. Weconclude by describing the ROF techniques conceived for reducingthe ROF system installation costs.

Index Terms—Radio over fiber, optical modulation, wave-length interleaved wavelength division multiplexing, carrier sup-pression, low biasing, balanced photo-detection, optical injectionlocking, optical feed-forward linearization, linearized opticalmodulators, optical filtering, pre-distortion, post-compensation,chirped optical signals, optical phase conjugation, digitized radioover fiber, fiber-to-the-home networks, wavelength re-use, opticalsingle side band modulation, optical double side band modulation.

I. INTRODUCTION

DUE TO the proliferation of various high-bandwidth appli-cations, the most benign, low frequency radio spectrum

is becoming over-crowded. A natural option is to use highercarrier frequencies, such as for example millimeter-waves (mm-waves) at say 60 GHz, where these high-frequency carriers havea potentially high bandwidth, but are characterized by a lowwireless propagation range [1]. Therefore, a large number ofRadio Access Points (RAPs) are required for providing seam-less coverage. In order to cope with the increasing bandwidthdemand per user, network operators often have to split theexisting cells into smaller cells. However, increasing the num-ber of base-stations is not always a feasible option due to thehigher infrastructure costs involved [2].

One of the major access network solutions for future high-bandwidth wireless communication systems is based on opticalfibers for the transmission of radio signals between the BaseStation (BS) and RAPs, which is generally referred to as aRadio Over Fiber (ROF) solution [2]–[6]. Fig. 1 shows thesimplified diagram of a ROF link, where the wireless Radio

Manuscript received January 21, 2014; revised July 9, 2014 andNovember 3, 2014; accepted December 21, 2014. Date of publication January21, 2015; date of current version May 19, 2015. This work was supported inpart by the Engineering and Physical Sciences Research Council, U.K. underthe auspices of the India–U.K. Advanced Technology Centre and in part by theEuropean Research Council under the Advanced Fellow Grant.

The authors are with the School of Electronics and Computer Science, Uni-versity of Southampton, Southampton SO17 1BJ, U.K. (e-mail: [email protected]; [email protected]; [email protected]; http://www-mobile.ecs.soton.ac.uk).

Digital Object Identifier 10.1109/COMST.2015.2394911

Fig. 1. Simple ROF link. (E-O - Electrical-to-Optical; O-E - Optical-to-Electrical).

Fig. 2. Cellular Architecture (BS - Base Station; RAP - Radio Access Point;MS - Mobile Station). (a) Without using ROF. (b) Using ROF.

Frequency (RF) signal is converted into an optical signal in anelectrical-to-optical (E-O) converter at the BS [7]. The opticalsignal is transmitted through the fiber and detected at the RAP,where an optical-to-electrical (O-E) converter recovers the orig-inal RF signal, which is amplified and transmitted from theRAP antenna to the Mobile Station (MS), as shown in Fig. 1.

Fig. 2 shows two cellular architectures, namely one withand one without using ROF assistance. Fig. 2(a) represents theexisting cellular systems that are based on macrocells, while

1553-877X © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

628 IEEE COMMUNICATION SURVEYS & TUTORIALS, VOL. 17, NO. 2, SECOND QUARTER 2015

Fig. 2(b) is a futuristic cellular architecture that relies on a ROFbackhaul designed for supporting smaller cells. The architectureseen in Fig. 2(a) has a lower-quality wireless coverage, a lowerpower efficiency and a limited ability to use small cells, whilstthese challenges are overcome in the architecture of Fig. 2(b) atthe added cost of laying more fiber, as discussed below.

As seen from Fig. 1 and Fig. 2, ROF solutions belong to thefamily of hybrid techniques, since they rely on both optical andwireless communication. The advantage of a ROF system is itsability to amalgamate the benefits of these two diverse com-munication techniques. Explicitly, the advantages of an ROFsystem can be summarized as follows [8]–[10].

1) Low fiber attenuation: The low attenuation of glass op-tical fiber1 (0.2 dB/Km) effectively facilitates the creationof a geographically distributed large system of antennasthat are connected to the central BS, as shown in Fig. 2(b),where the BS is connected to four RAPs via opticalfiber. This architecture is referred to as a DistributedAntenna System (DAS) [5], where each BS supports alarge geographical coverage area by using optical fibersas long as 30 Km without amplification.

2) Large fiber bandwidth: The optical fiber is character-ized by large bandwidths and we will discuss the tech-niques that exploit this large bandwidth in Section IV ofthis manuscript.

3) Economically viable small cells: In a ROF system, eachBS serves a number of RAPs as shown in Fig. 2(b), wherecentralized signal processing takes place at the BS, whilethe RAPs have a low cost, hence reducing the cost of set-ting up a cellular system that employs small cells [11]. Thisis one of the major aims of the architectures and tech-niques discussed in Sections II and VI of this paper. More-over, the concentration of most of the signal processingtasks in the BS ensures that upgrades in technology andhardware may be readily carried out in conjunction withefficient dynamic resource allocation techniques [12].

4) Improved wireless coverage: As shown in Fig. 2(a),there exists regions that do not receive the wireless signalwith sufficient field strength in the existing cellular ar-chitecture. The area of these low-signal-strength regionsis minimized in a ROF assisted cellular architecture, asshown in Fig. 2(b) [2].

5) Ability to implement Multiple Input Multiple Output(MIMO) schemes [13]: The wireless communicationschemes that employ multiple transmitters and receiversfor supporting a single mobile user are referred to asMIMO schemes [14], [15]. The parallel MIMO linkseither create multiple streams and hence increase theachievable throughput, or provide a diversity gain byexploiting the fact that each RF link fades independently.The DAS architecture facilitates the design of networks inwhich each MS of Fig. 2(b) is served by multiple RAPs,as in the Co-ordinated Multipoint (CoMP) architecturediscussed in [8]. Thus, efficient MIMO schemes may beconstructed with the aid of a ROF network [3].

1The terminology of glass optical fibers refers to fibers made from silica.

6) Power Efficiency: The deployement of smaller cells, asis seen from Fig. 2(b), provides a stronger line of sightlink and ensures a lower wireless path loss [2]. Hence,a lower level of wireless transmit power is needed at theRAPs, which has the added advantage of reduced inter-cell interference [16].

Fig. 4 illustrates the various design challenges of the ROFnetwork. For example, the deployement of higher RF fre-quencies results in higher fiber dispersion, which degrades theBit Error Ratio (BER). The BER may be reduced by usingdispersion compensation mechanisms, which however increasethe implementational complexity and cost of the system. Al-ternatively, the BER may be reduced by using lower-ordermodulation schemes, which would however reduce the overallthroughput. On the other hand, the overall throughput maybe improved by ensuring better optical spectral efficiencythrough schemes like Dense Wavelength Division Multiplexing(DWDM), which however would increase the implementationalcomplexity and cost. The network outage rate or probabil-ity featuring in Fig. 4 is defined as the proportion of timefor which the network is unable to serve the MSs. A com-mon reason for outage is that the RAP is overloaded by alarge number of MSs and hence it is unable to maintain theminimum signal to interference plus noise ratio required forcommunication with each MS [17]. The outage rate is alsoincreased by shadowing imposed by large obstructing struc-tures, as in Fig. 2(a) [17]. The network outage rate can bereduced by having a larger number of RAPs, which wouldreduce both shadowing as well as the load (or MSs) per RAP.However, this would increase the system’s cost and complex-ity. Thus, most design questions usually boil down to a costversus performance tradeoff. This manuscript presents a rangeof techniques that are either aimed at cost reduction or atperformance improvement.

Comparison With Other Backhaul Techniques: Wirelessbackhauls like the microwave backhaul are competing backhaultechniques. The advantage of the wireless backhaul is that theinstallation cost does not scale significantly with the distancebetween the transmitter and receiver. Moreover, a wirelessbackhaul can be readily installed as well as disassembled andset up elsewhere [18]. However, a wireless backhaul also suffersfrom several disadvantages. Firstly, it requires wireless spec-trum, which could otherwise have been employed for wirelesscommunication with mobile devices. Moreover, in the case of amicrowave backhaul, line of sight communication is a necessity[18]. Finally, a wireless backhaul is influenced by the climaticconditions [18]. On the other hand, a fiber-based backhaul doesnot require wireless spectrum and has much higher usable band-width than the wireless backhaul. Additionally, a fiber-basedbackhaul does not require line of sight communication and itis not influenced by the climatic conditions [18]. However, theinstallation cost of a fiber-based backhaul is higher than thatfor a wireless backhaul, especially when the terrain is roughand this cost increases as the distance between the transmitterand receiver increases. Finally, a fiber-based backhaul, unlikethe wireless backhaul, cannot be deployed or moved easily[18]. At this point it is worth noting that a copper wire based

THOMAS et al.: PERFORMANCE IMPROVEMENT AND COST REDUCTION TECHNIQUES FOR ROF COMMUNICATIONS 629

Fig. 3. Block diagram of a ROF link (BS - Base Station; RAP - Radio AccessPoint; MS - Mobile Station; BPF - Band Pass Filter; DSP - Digital SignalProcessing).

backhaul has severe bandwidth restrictions and is being gradu-ally phased out.

When employing the fiber-based backhaul, the designer hastwo options, namely the standard digital fiber-based backhaul orthe ROF backhaul proposed in this paper. In case of the digitalPCM-over-fiber backhaul, each small cell of Fig. 2(b) wouldneed a full-fledged BS for detecting/transmitting the wirelesssignal as well as for transmitting/detecting the bits to/from thebackhaul, which employs protocols like the Ethernet protocol.However, as discussed earlier, a ROF backhaul will ensure thatmultiple cells are served by a single BS, as shown in Fig. 2(b),thereby reducing the network’s installation cost. Moreover, theROF technique involves the transmission of the analog RFsignal to the BS and hence it will reduce the latency, becausethe signal processing carried out in the RAP is minimized. ROFis a potential backhaul technique for the Long Term Evolution(LTE) advanced [19], [20] standard as well as for future mil-limeter wave standards [21]. The fewer BSs of a ROF assistedcellular network may be served by a digital fiber optic backhaulor a wireless backhaul. At this point, it is worth mentioningthat standards like the Open Base Station Architecture Initiative(OBSAI) ensure compatibility between the different BS mod-ules manufactured by different companies, thereby resulting incost savings.

This paper firstly discusses the components and modulationschemes of a basic ROF link in Sections II and III, respectively.A range of optical multiplexing techniques are then described inSection IV, followed by a discussion on diverse ROF link per-formance improvement techniques in Section V. Afterwards,a discussion of cost reduction techniques is presented inSection VI, followed by our design guidelines and conclusionsin Sections VIII and IX, respectively. It must be pointed out thatthroughout this paper, fi refer to frequencies, while ωi refer to thecorresponding angular frequency, which are related as follows:

ωi = 2π fi. (1)

Note that we have provided a list of the commonly usedabbreviations at the end of the paper in Table VIII.

II. ROF COMMUNICATIONS BASICS

The block diagram of a typical ROF link is shown in Fig. 3[7]. The downlink (DL) baseband (BB) signal is generated inthe BS and upconverted to the carrier frequency in order togenerate the RF signal. After the upconversion stage of Fig. 3,

Fig. 4. Stylized representation of the design challenges involved in the designof the ROF network.

Fig. 5. Design trade-offs in the design of a conventional ROF link (ODSB -Optical Double Side Band; OSSB - Optical Single Side Band; OCS - OpticalCarrier Suppressed).

E-O conversion is achieved by modulating the optical carrier bythe analog RF signal [22], [23] using the techniques discussedin Section II-A. Then, as seen in Fig. 3, the signal propagatesthrough the fiber to the RAP. The various fiber impairmentsthat affect the propagating optical signal are discussed inSection II-B. At the RAP of Fig. 3, O-E conversion is achievedusing the techniques that will be discussed in Section II-C,where the electronic signal generated by the O-E conversionis filtered using a Band Pass Filter (BPF), in order to obtain theRF signal. As shown in Fig. 3, this RF signal is then amplifiedand transmitted wirelessly by the antenna to the MS. During theuplink (UL) transmission, the RF signal received by the RAPfrom the MS is bandpass filtered and E-O converted in order togenerate the optical UL signal, which is then transmitted overthe fiber and subsequently O-E converted, followed by signaldetection. It is worth mentioning that the wireless system ofFig. 3 is also referred to as a Fi-Wi system, whose basics havebeen discussed in [9], along with a discussion of FiWi channelestimation and equalization algorithms.

Fig. 5 shows the trade-offs in the design of a conventionalROF link, including the various design options in the optical


Fig. 6. Direct modulation (LD - Laser Diode; RF Tx - RF Transmitter).(a) Direct modulation setup. (b) Power vs. drive current curve of a laser.

Fig. 7. Direct modulation bandwidth of the laser.

transmitter, optical fiber link, optical receiver and in opticalmultiplexing. These design options present the designer witha cost versus performance trade-off. The techniques shown inFig. 5 are discussed in this Section (Section II) as well as inSections III and IV.

A. Modulation Techniques

As shown in Fig. 3, E-O conversion is performed by opticalmodulation. Modulation of the optical carrier can be achievedby either directly modulating the optical source or by using aseparate optical modulator. These two techniques are typicallyreferred to as direct modulation and external modulation, re-spectively.

In direct modulation, a Laser Diode (LD) is driven by acurrent Idrive in order to generate an optical output [24]. Asshown in Fig. 6(a), modulation of the optical output power2

of a laser can be inexpensively accomplished by varying thedrive current level using an electronic RF Transmitter, referredto as RF Tx in Fig. 6(a). This is normally referred to asdirect modulation of lasers. Fig. 6(b) illustrates the Power vs.Current (P-I) characteristic of a laser, showing in a stylizedmanner the output optical power level for different values ofthe drive current. Within the linear region of the P-I charac-teristic in Fig. 6(b), the output power is related to the drivecurrent as follows: P(t) = k(Idrive− Ith) = k(Ibias+ IRF(t)− Ith),where Ith is the threshold current of the laser, Ibias is the

2Optical power can also be measured in terms of optical intensity, which isdefined as the optical power per unit area

Fig. 8. Single and Dual-drive Mach–Zehnder Modulator (MZM) externalmodulators (LD- Laser Diode). (a) Single drive MZM. (b) Dual drive MZM.

bias current, IRF(t) is the modulating RF signal, while k isthe proportionality constant that depends on the laser sourceemployed.

Fig. 7 plots the modulation response versus the modulationfrequency, where the modulation response at fm Hz is thestrength of the modulated signal when a sinusoidal modulatingsignal having a frequency of fm Hz is employed. As seen fromFig. 7, the modulation response peaks at fm = fr, beyond whichthe modulation response decreases dramatically. Here, fr iscalled the relation resonance frequency and it decides the 3-dBmodulation bandwidth f3 dB of the laser, where f3 dB =

√3 fr

[24]. The designer should ensure that the modulating currenthas frequencies which are well below fr.

Furthermore, the optical carrier is a sinusoidal signal, alsooften referred to as a tone, that has an amplitude, phase andfrequency. The Frobenius norm of the tone is referred to asthe intensity. Both the intensity [25] and the angle [26] of theoptical carrier can be modulated using external modulators.External intensity modulation can be implemented using aMach-Zehnder Modulator (MZM) or an Electro-AbsorptionModulator (EAM).

A MZM [27] is one of the popular optical external modula-tors, where the incoming signal is split into two arms, as shownin Fig. 8. The MZMs, which apply a drive voltage to either oneor to both of the arms are referred to as single-drive and dual-drive MZMs, respectively. The output optical field Esingle(t)of a single-drive MZM that is biased at a voltage of Vbias

and driven by an electronic signal V (t) can be represented asfollows:

Esingle(t) =12

[e

j( πVbias

Vπ +πV (t)

Vπ

)+1

]√2Pine jωct

= cos

(πVbias

2Vπ+

πV (t)2Vπ

)e

j( πVbias

2Vπ +πV (t)2Vπ

)√2Pine jωct ,

(2)

where fc is the optical carrier frequency, Vπ is the switchingvoltage of the MZM and Pin is the output optical power of thelaser feeding the MZM. The switching voltage of a MZM isthe voltage that changes the phase of the propagating signal byπ radians, when applied to the arm of a MZM. In the case ofthe dual-drive MZM of Fig. 8, the bias voltage Vbias is applieddifferentially via voltages Vdc1 and Vdc2, while the modulatingvoltage V (t) is applied differentially via the voltages V1(t) andV2(t). Mathematically, this is formulated as follows: Vbias =


Fig. 9. Biasing for Mach–Zehnder Modulator (MZM) (MATP-MaximumTransmission Point; QP-Quadrature Point; MITP-Minimum TransmissionPoint). (a) MZM transmittance. (b) MZM bias points.

Vdc1 −Vdc2 and V (t) = V1(t)−V2(t). The output optical fieldof the dual-drive MZM is given by the following equation:

Edual(t) =12

[e

j( πVdc1

Vπ +πV1(t)

Vπ

)+ e

j( πVdc2

Vπ +πV2(t)

Vπ

)]√2Pine jωct

= cos

(π(Vdc1 −Vdc2)

2Vπ+

π(V1(t)−V2(t))2Vπ

)

ej

[π(Vdc1+Vdc2)

2Vπ +π(V1(t)+V2(t))

2Vπ

]√2Pine jωct (3)

= cos

(πVbias

2Vπ+

πV (t)2Vπ

)

ej

[π(Vdc1+Vdc2)

2Vπ +π(V1(t)+V2(t))

2Vπ

]√2Pine jωct (4)

The output intensity Psingle(t) = |Esingle(t)|2 and Pdual(t) =|Edual(t)|2 of the single- and dual-drive MZMs, respectively,can be expressed as follows using (2) and (4):

Pdual(t) = Psingle(t) = Pin

[1+ cos

(πVbias

Vπ+

πV (t)Vπ

)]. (5)

Fig. 9(a) shows the transmittance plot of a MZM, where theoutput optical power P(t) of the MZM is plotted for variousvalues of the bias voltage Vbias. Modulation can be interpretedas a voltage variation around the bias point. It can be seen from(2) and (4) that the output optical field has exponential terms ofthe form e jφ(t). Thus, there is residual phase modulation, whichis referred to as chirp. This residual phase modulation is poten-tially converted to intensity modulation during its transmissionover the fiber, thereby corrupting the signal that was transmittedusing intensity modulation. The chirp in a dual-drive scenariois adjustable, where the dual-drive MZM modulation scheme ofFig. 8 would be chirp free if we have: V1(t) = −V2(t) =

V (t)2 .

This is referred to as the push-pull mode of operation, becausethe voltages V1(t) and V2(t) are equal in magnitude and oppositein sign.

Fig. 9(b) shows the various MZM biasing techniques that canbe employed, which include Quadrature Point (QP) biasing,Minimum Transmission Point (MITP) biasing and MaximumTransmission Point (MITP) biasing. QP biasing corresponds toVbias = Vπ/2 + m2Vπ or Vbias = −Vπ/2 + m2Vπ, while MITPbiasing corresponds to Vbias = Vπ + m2Vπ, where m is aninteger. Furthermore, MATP biasing corresponds to Vbias =0 + m2Vπ. When a sinusoidal modulating signal of V (t) =VLO cos(ωLOt) is employed, the mathematical expression for the

Fig. 10. External modulation using an Electro-Absorption Modulator (EAM)(LD- Laser diode; RF Tx- RF Transmitter). (a) Optical modulation using anEAM. (b) EAM transmittance vs. applied voltage.

Fig. 11. Optical phase modulator (LD- Laser Diode; RF Tx- RF Transmitter).

photo-detected signals in each of these biasing scenarios can beobtained from (5) as follows:

IQP(t) ∝ PQP(t)

= Pin

[1∓ sin

(πVLO cos(ωLOt)

Vπ

)]

= Pin

[12±

∞

∑n=1

(−1)nJ2n−1

(πVLO

Vπ

)cos((2n−1)ωLOt)

](6)

IMIT P(t) ∝ PMIT P(t)

= Pin

[1−cos

(πVLO cos(ωLOt)

Vπ

)](7)

= Pin

[1−J0

(πVLO

Vπ

)−2

∞

∑n=1

(−1)nJ2n

(πVLO

Vπ

)cos(2nωLOt)

]

(8)IMAT P(t) ∝ PMAT P(t)

= Pin

[1+cos

(πVLO cos(ωLOt)

Vπ

)]

= Pin

[1+J0

(πVLO

Vπ

)+2

∞

∑n=1

(−1)nJ2n

(πVLO

Vπ

)cos(2nωLOt)

]

(9)

On the other hand, as shown in Fig. 10(a), an Electro-Absorption Modulator (EAM) modulates the intensity of theoptical signal by using an electric voltage that controls theabsorption3 of the modulator, which is referred to as the Franz-Keldysh effect [28]. As the reverse bias voltage Vbias of theEAM increases, the amplitude of the output optical field isreduced by a factor of T (V ), where T (V ) is the amplitudetransmittance.

Fig. 10(b) shows the variation of T (V ) as a function of thereverse bias voltage applied[29], where T (V ) = 0 correspondsto complete extinction of the MZM output, while T (V ) = 1corresponds to the extinction-free scenario. The output opticalfield E(t) of the EAM modulator seen in Fig. 10(a), that is bi-ased at a voltage of Vbias and driven by an electronic signal V (t)is given by the following expression: E(t) ∝ T (Vbias +V (t)).

3The absorption of an optical device refers to the phenomenon of a portionof the propagating optical power being absorbed by the device material.


Fig. 12. Multi-mode and single-mode fiber.

Additionally, external optical phase modulation is imple-mented in ROF links using an optical phase modulator. Thisis implemented by using the same concept as in the MZMs ofFig. 8 with the difference that a phase modulator, unlike theMZM, has a single arm. If an electronic signal V (t) is used fordriving a phase modulator as shown in Fig. 11, then the output

optical field may be written as: E(t) = e j πV (t)2Vπ

√2Pine jωct , where

fc is the optical carrier frequency, Vπ is the switching voltage ofthe phase modulator and Pin is the operating laser power.

External modulation is more expensive than direct modula-tion due to the need for additional equipment, while the directmodulation of lasers suffers from several drawbacks, therebygiving rise to the cost versus performance trade-off alluded to inFig. 5. Firstly, the modulation bandwidth of directly modulatedlasers, shown in Fig. 7, restricts the spectrum of usable modu-lating signals, while the saturation of the P-I characteristics ofFig. 6(b) limits the output power of directly modulated lasers.Additionally, direct modulation of the laser’s output intensityis accompanied by undesired frequency modulation. Finally, ifa laser is directly modulated using signals having frequenciesclose to fr in Fig. 7, then both the non-linear products and noisein the modulated output increased.

Let us now discuss the various aspects of the fiber channelemployed in a ROF link.

B. Fiber Impairments

An optical fiber consists of an inner core through whichthe light propagates and an outer cladding. Optical fibers mayoperate either in a single-mode or in multi-mode fashion,depending on whether single or multiple propagation modesexist in the fiber. Fig. 12 shows the propagation modes of thesingle-mode and multi-mode fiber, where the cross-section ofthe core in the multi-mode fiber is larger than that of the single-mode fiber [30]. All the architectures considered in this paperemploy Single-Mode Fiber (SMF), because it provides a largeroptical bandwidth and hence it is capable of supporting longertransmission distances without a repeater [30]. However, wehasten to add that the cost of multi-mode fibers is lower thanthat of a single-mode fiber, thereby giving rise to the cost versusperformance trade-off alluded to in Fig. 5.

Fiber impairments refer to the fiber characteristics that affectthe signal transmitted in the fiber. This includes attenuation,dispersion and non-linearities, as discussed below.

Firstly, the signal power decreases as the signal propagatesthrough the optical fiber, which is due to the impurities ofthe material and owing to Rayleigh scattering [24]. This re-sults in fiber attenuation. Mathematically, the net signal powerattenuation imposed by the optical fiber can be expressed asfollows: Preceived = Ptransmitted e−αz, where Ptransmitted is theoptical power that is coupled into the optical fiber, Preceived is

the optical power after propagation through z km of the fiberand α is the fiber attenuation constant, which has a typical valueof 0.2 dB/Km [24].

Secondly, there is fiber dispersion, where the refractiveindex n(ω) of the fiber depends on the frequency f = ω

2π ofthe propagating optical wave [24]. Mathematically, we havevp(ω) = c/n(ω), where vp(ω) is the phase velocity of theoptical wave as it propagates through the fiber. Thus, thepropagation velocity depends on the optical frequency, and thisphenomenon results in dispersion. The extent of dispersion isquantified by the dispersion parameter D [24]:

D =−2πcλ2

cβ2 where β2 =

[1c

(2

dndω

+ωd2ndω2

)]ω=ωc

(10)

where fc and λc are the frequency and wavelength of theoptical carrier. This fiber dispersion is also referred to as GroupVelocity Dispersion (GVD) or chromatic dispersion [24].

Thirdly, a dielectric medium (e.g., a fiber) is an electricalinsulator that can be polarized by an appropriately appliedelectric field. Fiber non-linearity arises from the fact that thepolarization is a non-linear function of the electric field E(t)of the light, especially when the optical intensity is high [24].This results in a non-linear refractive index as a function ofthe total propagating signal power P(t). This time-varying P(t)-dependent refractive index then modulates the phase of the opti-cal signal. Time-varying phase shifts in turn result in an instan-taneous optical frequency deviation from the nominal value offc, which is referred to as frequency chirping. Phase modulationarising from fiber-nonlinearity includes Self-Phase Modula-tion and Cross-Phase Modulation [24]. Self-Phase Modulation(SPM) occurs because of the time-varying nature of the prop-agating signal’s own instantaneous power [24], while, Cross-Phase Modulation (XPM) occurs when multiple optical signalsare transmitted simultaneously. In XPM, the extent of nonlineardistortion imposed on each channel depends not only on itsown optical power, but also on the optical power of the co-propagating signals [24].

On the other hand, the non-linear interaction of multiple opti-cal signals that are propagating in a single optical fiber leads tothe generation of optical signals at new frequencies. This phe-nomenon is typically referred to as Four-Wave Mixing (FWM)[24]. For example, when the signals having optical frequenciesof f1, f2 and f3 Hz are propagating in the fiber, FWM results inthe generation of an optical signal at fFWM = f1 ± f2 ± f3 Hz,where the most significant contributions are those at [24]:

fFWM = fi + f j − fk (11)

where, we have i, j,k ∈ 1,2,3. The above expression includesfrequencies of the form

fFWM,1 = 2 fi − f j (12)


FWM becomes most significant at low absolute values of thedispersion parameter D [24].

Finally, there are other major non-linear effects includingStimulated Brillouin Scattering (SBS) and Stimulated RamanScattering (SRS) [24], where photons of the incident field getannihilated in both cases. In other words, optical energy istransferred from a signal propagating at a particular opticalfrequency to a signal at a lower frequency. SBS occurs in thebackwarddirection,wherethelowerfrequencyopticalsignalpro-pagates in a direction that is opposite to that of the incident higherfrequency signal, while SRS can occur in both directions [24].

Now that we have discussed both optical modulation andfiber based transmission, we will continue our discourse byconsidering optical photo-detection.

C. Photo-Detector

Photo-detection is the process of detecting an optical signalusing a photo-diode (PD), where a PD generates an electroniccurrent that is proportional to the incident optical power. Thisprocess is referred to as square-law photo-detection, where theresponsivity R of a PD is the ratio of the output photo-current tothe incident optical power. The basic types of photo-detectionare direct photo-detection [22] and coherent photo-detection[24]. In the direct photo-detection of 13(a), if Es(t) is the opticalfield of the received signal, then the generated photo-currentI(t) is I(t) = R|Es(t)|2 [22]. Thus, in direct photo-detection,the intensity |Es(t)|2 of the received optical signal should bemodulated by electronic modulating signal in order to enablethe reconstruction of the modulating signal using direct photo-detection in the receiver. On the other hand, in coherent photo-detection, the received optical signal is mixed with the outputof a laser that serves as a local oscillator, followed by their jointphoto-detection. Mathematically, this can be represented asI(t) = R|Es(t)+Elo(t)|2, where Es(t) and Elo(t) are the opticalfields of the received optical signal and the local oscillatorsignal, respectively. Expanding this expression would result inone of the terms being a product of the two fields, which isreferred to as the beat signal. In heterodyne photo-detection,the optical frequency of the flo = ωlo

2π Hz local oscillator isdifferent from the carrier frequency fs =

ωs2π Hz of the incident

optical signal. In this scenario, the photo-detected signal I(t) isgiven by:

I(t) =R|Asejωst+φs +Aloe jωlot+φlo |2

=R[A2

s +A2lo +2|As||Alo|cos(ωdi f f t +φs −φlo)

], (13)

where fdi f f = fs − flo, As and Alo are the amplitudes of thereceived optical signal and the local oscillator signal, respec-tively, while φs and φlo are the phases of the two opticalfields. Fig. 13(b) shows how heterodyne photo-detection canbe carried out for the upconversion of the BB/IF signal to anRF frequency of fr f Hz, which can be done by ensuring that wehave fdi f f = fr f and then filtering the photo-detected signal toretain the beat signal.

Direct photo-detection is simpler and hence more cost-effective than coherent photo-detection, while coherent photo-detection provides a better performance, thereby giving rise tothe cost versus performance trade-off seen in Fig. 5.

Fig. 13. Basic types of photo-detection (PD- Photo-diode). (a) Direct photo-detection. (b) Heterodyne photo-detection.

The discussions so far assumed a noise-free optical link,while the next section details the various sources of noise ina ROF link.

D. Noise in a ROF Link

There are various sources of noise in a ROF link, includingthe Relative Intensity Noise (RIN), Amplified SpontaneousEmission (ASE) noise, shot noise and the receiver’s thermalnoise, where Δ f is the bandwidth over which the noise poweris measured at the receiver [24]. The output of a laser that isdriven using a constant current exhibits certain fluctuations inthe output phase, frequency and intensity[24], which constitutesthe above-mentioned RIN. The primary source of this is thespontaneous emission of photons, where the intensity noiseresults in a SNR degradation, while the phase/frequency noiseleads to a non-zero spectral linewidth (spillage) around theoutput frequency [24]. Mathematically, the RIN noise poweris formulated as follows [24]:

σ2RIN = KRINI2

dcΔ f (14)

where KRIN is a device dependent value, which is usuallyexpressed in dB/Hz and Idc is the average photo-current value.Typically KRIN has values around −150 dB/Hz. It can be seenfrom (14) that the RIN is proportional to the square of the meanoptical power. However, when the depth of optical modulationis high, especially in a ROF link that employs the subcarriermultiplexing of several RF signals, an improved expression forRIN was derived in [31], which reflected the enhanced presenceof RIN owing to its dependence on the depth of optical modu-lation [31]. The concept of subcarrier multiplexing will be dis-cussed in Section IV-A. Additionally, the ASE noise alluded toabove is produced by spontaneous emission in an optical ampli-fier, which is then amplified by the amplifier’s gain mechanism[24]. Furthermore, Shot Noise is a quantum noise effect that ispresent in the photo-detected signal, which arises from the factthat the optical field and the electric current consists of discreteentities referred to as photons and electrons, respectively [24].Mathematically, the shot noise power is given by [24]:

σ2shot = 2eIdcΔ f (15)

where e is the charge carried by an electron. Finally, the loadresistor and the electronic amplifiers in the optical receiverimpose Thermal Noise, owing to the temperature-dependentrandom Brownian motion of the electrons [24]. Mathematically,the thermal noise power can be expressed as [24]:

σ2thermal =

4kbTRL

Δ f (16)


Fig. 14. Block diagram of losses/gains in ROF enabled DL wireless transmis-sion, as discussed in [32].

where kb is the Boltzmann constant, T is the absolute temper-ature, RL is the load resistance, while Fn is the amplifier noisefigure.

E. Wireless Channel

The wireless channel is normally modeled using Gaussian,Rayleigh or Rician distributions, where the signal y receivedover a wireless channel H can me modeled as [15]:

y = Hx+n (17)

where x is the transmitted symbol matrix, n is the Gaussiannoise matrix, while H is the channel matrix. H is unitary for aGaussian wireless channel, while it consists of Rayleigh andRician distributed channel co-efficients for a fading chan-nel obeying a Rayleigh and Rician distribution, respectively[15]. In the case of millimeter wave communications, thewireless channel has a higher free-space pathloss, which canhowever be overcome by employing beamforming using di-rectional antennas. Directional antennas consist of an arrayof antennas, whose outputs constructively interfere in somedirections, while destructively interfering in other directions.For a detailed study of the millimeter-wave wireless channeland beamforming, the reader is referred to [21] and [33].Furthermore, for a comprehensive study of the wireless chan-nel, the reader is referred to [15], since this paper focuseson the optical aspects of the ROF link. Nevertheless, in thissection we focus on radio-communication issues, which havea direct bearing on the ROF backhaul design, including powerbudgeting and the presence of accumulating noise [31], [32].The significance of these issues are illustrated by consideringthe simple case of a ROF DL that does not employ opticalamplification and has a Gaussian wireless channel. Fig. 14and Fig. 15 portray a diagrammatic representations of thegains and losses in such an ROF link. The RF signal xBS

RF(t)having a power of PBS

RF is attenuated by a factor of Lop bythe ROF link, resulting in the photo-detected signal xRAP

RF (t)having a power of PRAP

RF . The attenuation Lop includes thatarising from the E-O conversion, fiber propagation, insertionloss of optical components and the O-E conversion. Using (14)–(16), the noise nop(t) imposed by the ROF link has a power of:

Popn = σ2

RIN +σ2shot +σ2

thermal . (18)

It is worth noting here that when spread spectrum wirelesstechniques like Code Division Multiple Access (CDMA) areemployed, a larger wireless bandwidth is desired. However, thisalso increases Pop

n . The optical noise is accounted for in Fig. 14by adding it to xRAP

RF (t) for generating xRAPRF,noisy(t), as shown in

Fig. 14.

Fig. 15. power budgeting as discussed in [32].

As shown in Fig. 14 and Fig. 15, the noise contaminatedphoto-detected signal is then amplified by a factor of Gop,where Pant

RF is the power of the signal (without noise) that istransmitted from the antenna. The transmitted signal undergoesa wireless pathloss induced attenuation of Lwl , as shown inFig. 14, and is finally received at the MS, where the noise nwl(t)having a power of Pwl

n is added by the wireless link. The signalthat is finally employed in the MS is xMS

RF,noisy(t). It must benoted that in Fig. 14, Lwl and Gamp are amplitude attenuationand amplification values, while power attenuation/amplificationrefers to their squared values.

Radio cell size calculations based on cumulating noise andfiber attenuation is derived in [32] and [9]. Following thatmodel, (27) can be obtained as follows.

It can be seen from Fig. 15 that the Signal to Noise Ratio(SNR) of the ROF link (measured after photo-detection in RAP)as well as that of the total link including the optical and wirelesstransmission (measured in the MS), are as follows [32]:

SNRROF =PRAP

RF

Popn

(19)

SNRtot =PMS

RF

Ptotn

(20)

where PRAPRF depends on the operating laser power, on the

depth of optical modulation and on Lop [32]. Furthermore, thefollowing is evident from Fig. 15 [32]:

PMSRF =PRAP

RF

(Gamp

Lwl

)2

(21)

Ptotn =Pop

n

(Gamp

Lwl

)2

+Pwln . (22)

Assuming that the wireless noise power is k times that of theoptical noise power, (22) can be written as:

Ptotn = Pop

n

(Gamp

Lwl

)2

[1+ k

(Lwl

Gamp

)2

] (23)

where k ≈ 1 corresponds to both the wireless and optical noisehaving similar powers, while k � 1 or k � 1 corresponds to thewireless noise being much stronger or much weaker than theoptical noise, respectively. In the following analysis, we focusour attention on the scenario, where k ≈ 1, because the other


scenarios can be analyzed in a similar manner. Furthermore,as stated earlier, Lwl is the signal amplitude attenuation ofwireless propagation. However, the wireless path loss Ldb isusually specified in decibels and it refers to the power attenua-

tion during wireless propagation, i.e., L2wl = 10

LdB10 . Using this

relation as well as substituting (19), (21) and (23) into (20), wearrive at [32]:

SNRtot =SNRROF1

1+ k 10LdB10

G2amp

(24)

=SNRROF1

1+ 10LdB10

G2amp

for k ≈ 1. (25)

Equation (25) helps us understand the limitations imposed onthe design of the ROF link by the requirements of the wirelesslink, as discussed in terms of the minimum RAP amplifier gainand maximum cell size in the following sections.

a) Minimmum RAP Amplifier Gain: Given a ROF link hav-ing a SNR of SNRROF , (25) allows us to calculate the minimumgain Gmin

amp of the electronic amplifier in the RAP, which is nec-essary for supporting a cell size that corresponds to a maximumloss of Lmax

dB at the cell boundary, where SNRtot ≥ SNRworsttot

always has to be maintained in the MS. Using (25), Gminamp is

formulated as follows [32]:

Gminamp =

√√√√ LmaxdB10

SNRROFSNRworst

tot−1

for k ≈ 1. (26)

A fundamental design limitation that can be observed from(26) is that SNRROF should be higher than the worst-case SNRexpected at the cell boundary [32].

b) Cell Size: Assuming that the electronic amplifier in theRAP has a gain of Gamp and that a worst-case SNR of SNRworst

tothas to be maintained at the cell boundary using a ROF linkhaving a SNR of SNRROF , we can calculate the maximumaffordable wireless pathloss to be [32]:

LmaxdB = 10log

([SNRROF

SNRworsttot

−1

]G2

amp

)for k ≈ 1. (27)

It can be observed from (27) that a larger cell, i.e., a larger LmaxdB ,

requires a larger SNRROF , i.e., a lower fiber length.It can be concluded from the above discussion that SNRROF

maximization is critical for the design of a ROF link. It can beseen from Fig. 15 that SNRROF is maximized by limiting thelength of the fiber used. Alternatively, it can be improved byreducing the E-O and O-E conversion losses through the useof reactive matching circuits instead of resistive matching cir-cuits. However, this would limit the RF bandwidth of the ROFlink [32].

Our discourse on modulation techniques in Section II-Aintroduced the various optical modulators, while the discussionin Section III will cover the commonly employed modulationschemes.

III. COMMON ROF OPTICAL MODULATIONS SCHEMES

ROF systems may rely on modulating either the intensityor the angle, i.e., phase/frequency, of the optical carrier. In

Fig. 16. Optical modulation formats (ODSB - Optical Double Side Band;OSSB - Optical Single Side Band; OCS - Optical Carrier Suppression).(a) ODSB signal. (b) OSSB signal. (c) OCS signal.

the following sections we will describe the optical intensitymodulation and the optical angle modulation.

A. Optical Intensity Modulation

By definition, Intensity Modulation (IM) involves the modu-lation of the intensity4 of the optical signal by the modulatingelectronic signal, where detection can be carried out by usingboth the direct or coherent photo-detection schemes discussedin Section II-C. Intensity modulation may be of single-sidebandor double-sideband nature or may result in a carrier suppressedoutput. The corresponding Optical Double Side Band (ODSB),Optical Single Side Band (OSSB) and Optical Carrier Suppres-sion (OCS) intensity modulation schemes are described in thefollowing sections.

1) Optical Double Side Band Intensity Modulation:Fig. 16(a) shows the stylized spectrum of an ODSB signal,which consists of the optical carrier frequency fc as well as ofthe lower and upper optical sidebands. If the center frequenciesof the lower and upper optical sidebands are flower and fupper re-spectively, then in Fig. 16(a) we have fc− flower = fupper− fc =fr f [34], where fr f is the center frequency of the modulatingRF signal. Direct modulation of lasers and external modulationusing an EAM can be invoked for generating an ODSB signal[23]. Additionally, an ODSB signal may be generated by thesingle-drive MZM of Fig. 8(a), by setting Vbias =Vπ/2+m2Vπor Vbias = −Vπ/2 + m2Vπ in (5), where m is an integer [25],[35] and V (t) is the RF signal to be transmitted. The specificbias condition employed for generating ODSB signals isreferred to as quadrature biasing. The same condition is truefor the dual-drive MZM of Fig. 8(b), along with the additionalcondition that there should be a phase difference of π betweenthe RF signals V1(t) and V2(t) in (5).

While the ODSB signal of Fig. 16(a) has two modulationsidebands, the OSSB signal to be discussed in the next sectionhas a single modulation sideband.

2) Optical Single Side Band Intensity Modulation:Fig. 16(b) shows the stylized spectrum of an OSSB signal,which consists of the optical carrier frequency fc and only oneof the two possible sidebands at fsideband Hz. As seen fromFig. 16(b), we have fc − fsideband = fRF , where fRF is thefrequency of the modulating RF signal. There are severaltechniques that can be employed for generating OSSB signals,including the employment of the Fiber Bragg Gratings (FBGs)[36] and the Dual-Drive MZM of Fig. 8(b).

4The intensity of an optical signal is the square of its amplitude.


Fig. 17. OSSB generation using a Fiber Bragg Grating (FBG) (LD - LaserDiode; RF Tx - RF Transmitter).

Fig. 18. Structure and transmission spectrum of a Fiber Bragg Grating.(a) FBG filter. (b) FBG transmission.

A FBG can be used for filtering out one of the sidebands of anODSB signal, as shown in Fig. 17 [36]. A FBG filter consist of aperiodic perturbation of the effective refractive index along thelength of the fiber core, as shown in Fig. 18(a) [37]. The lengthof one period of this perturbation is referred to as the gratingperiod [37]. A FBG filter has a transmission profile as shownin Fig. 18(b), where the signal at frequency fr is reflected. Thevalue of fr depends on both the periodicity of the perturbationand on the effective refractive index of the fiber core. Thistechnique conceived for OSSB signal generation needs high-selectivity filters for retaining only the desired sideband, espe-cially when the RF carrier involved is not high. Moreover, thetemperature-dependent variations of the refractive index resultsin the reflected wavelength value being temperature-sensitive[38]. Furthermore, an OSSB signal can also be generated by theDual-Drive MZM of Fig. 8(b), when it is quadrature biased bysetting Vbias=Vπ/2+m2Vπ or Vbias =−Vπ/2+m2Vπ in Equation(5), where m is an integer, whilst simultaneously ensuring that aphase shift of 90◦ is maintained between the RF signals appliedto its two arms seen in Fig. 8(b) [27], [35], [39].

Recall that the ODSB and OSSB signals of Fig. 16(a) and (b),respectively, include the optical carrier, but this is not the casefor the OCS signal discussed in the next section.

3) Optical Carrier Suppression Modulation: As seen fromFig. 16(c), an Optical Carrier Suppression (OCS) based signalconsists of a suppressed optical carrier frequency at fc Hz andtwo sidebands at fc± fr f Hz, where the frequency of the RF sig-nal employed is fRF Hz. An OCS signal can be generated by theMZM of Fig. 8, when it is MITP biased, as illustrated in Fig. 9.This is achieved by setting Vbias = Vπ + m2Vπ in (5), wherem is an integer [40]. The photo-detection of the OCS signalgenerates a signal at 2 fr f Hz, which is a result of the above-mentioned beating between the two optical sidebands duringphoto-detection [41].

Employing the ODSB modulation of Fig. 16(a), especiallywhen using the direct modulation of Fig. 6(a), requires asimple [34] and inexpensive [23] set-up. Even when employ-

Fig. 19. Phase modulation using a Mach–Zehnder delay Interferometer (MZI)(LD - Laser Diode; RF Tx - RF Transmitter; PD - Photo-diode).

ing the external modulators of Fig. 8, the ODSB signals canbe generated using a single-drive MZM, while OSSB signalgeneration requires the more sophisticated dual-drive MZM.However, ODSB signals suffer from a higher chromaticdispersion-induced power penalty than OSSB signals. Thisis because the phase-change imposed by chromatic disper-sion is frequency-dependent, thereby resulting in a phase-difference between the upper and lower sidebands in theODSB signal of Fig. 16(a). The photo-detected signal isproportional to the optical power P(t) = |E(t)|2, with E(t)being the optical field-strength of the ODSB signal, wherethe optical power is attenuated owing to the destructive in-terference between the beating sidebands [34]. This phe-nomenon is discussed in greater detail in Section V-K.Moreover, OSSB signals have a higher optical spectral effi-ciency. The OCS signals too are tolerant to chromatic disper-sion, but they have a frequency doubled output. Thus, there is acost vs. performance trade-off, which was alluded to in Fig. 5.

Now that we have become familiar with the basic intensitymodulation techniques, we move on to discussing optical anglemodulation in the next section.

B. Optical Angle Modulation

Optical angle modulation involves the modulation of thephase or frequency of the optical signal, which are typicallyreferred to as optical Phase Modulation (PM) and FrequencyModulation (FM), respectively. Optical angle modulated sig-nals can be detected coherently, as shown in Fig. 13(b) ordirectly, as shown in Fig. 13(a), where coherent detection isstraight-forward, while direct detection requires a prior conver-sion of the angle modulated signal to an intensity modulatedsignal using a discriminator in the O-E blocks of Fig. 3. Adiscriminator is an optical component which has a frequency/phase dependent output intensity. Mach–Zehnder delay inter-ferometers of Fig. 19 [42] are the most commonly employeddiscriminators, as detailed below.

Again, the Mach–Zehnder delay Interferometer (MZI) isshown in Fig. 19, where the output of a laser operating at afrequency of fc Hz and at an optical power of Pin is phasemodulated and transmitted over the fiber. The angle modulatedsignal entering the MZI is Ein(t) =

√2Pine jωct+φ(t), where φ(t)

depends on the modulating electronic signal and the intensity|Ein(t)|2 of the angle modulated signal is constant. The incom-ing angle modulated signal is then fed into the two arms of theMZI of Fig. 19, where the signal in one of the arms experiencesa delay of τ seconds with respect to the other. The outputs of the


two arms are combined to generate an optical field Eout, whoseintensity Pout is as follows [42]:

Pout = |Eout|2 =Pin

2

∣∣∣∣e jωct+φ(t) + e jωc(t−τ)+φ(t−τ)∣∣∣∣2

=Pin

2[1+1+2cos(ωcτ+Δφ(t))] (28)

where we have Δφ(t) = φ(t)−φ(t − τ) and the MZI is quadra-ture biased with ωcτ = −π/2. Subsequently, using the small-signal approximation that sin(Δφ(t))≈Δφ(t), we get [42], [43]:

Pout = Pin [1+ sin(Δφ(t))] = Pin [1+Δφ(t)] (29)

Thus, the optical signal is now intensity modulated and canbe detected by the direct photo-detection scheme of Fig. 13(a)[42]. Additionally, the MZI of Fig. 19 generates a comple-mentary output, whose intensity is Pout,2 = Pin[1−Δφ(t)] [42].Balanced photo-detection can be achieved by subtracting thetwo photo-detected outputs, as shown in Fig. 19 [42], wherethe associated advantages of balanced photo-detection will bediscussed in Section V-D. The dynamic range of an anglemodulated link that relies on direct photo-detection, largelydepends on both the non-linearity of the descriminator and onthe residual IM present in the phase-modulated signal that istransmitted [44].

Intensity modulated signals can be detected using the simplerdirect photo-detection discussed in Section II-C, while anglemodulated signals require an additional discriminator for directphoto-detection or they employ the more complex coherentphoto-detection discussed in Section II-C. On the other hand,the cross-talk level in multi-channel systems is lower, whenemploying phase modulation than upon employing intensitymodulation [42]. Moreover, optical intensity modulation usingthe MZMs of Fig. 8 or EAM of Fig. 10(a) requires biasing,while the phase modulator of Fig. 11 is not biased. Hence,intensity modulation suffers from the drifting bias voltage andhence needs bias stabilization circuits [45]. Thus, there is acost versus performance trade-off, which was portrayed in thestylized illustration of Fig. 5.

Having covered the basics of a single ROF link, we nowdiscuss the various ROF multiplexing techniques in Section IV.

IV. MULTIPLEXING

Optical multiplexing employed in multi-user/multi-RAP sce-narios allows us to increase the throughput of an optical fiberlink. There are two basic types of optical frequency multiplex-ing schemes, which can be employed in ROF systems, namelySub-Carrier Multiplexing (SCM) [46] and Wavelength DivisionMultiplexing (WDM) [47].

A. Sub-Carrier Multiplexing

Sub-Carrier Multiplexing (SCM) refers to the frequency-domain multiplexing of multiple modulating signals, which aretermed as sub-carriers, for example signals at f1 and f2 Hz.The combined signal in then input to the direct modulationscheme [48] of 6(a) or to the external modulation [49] schemesof Fig. 8 and Fig. 10(a), as shown in Fig. 20. The spectrum of

Fig. 20. A Sub Carrier Multiplexing (SCM) transmitter model (LD- LaserDiode; RF Tx- RF Transmitter).

Fig. 21. Sub Carrier Multiplexing (SCM) optical spectrum.

the modulating electronic signal and of the modulated opticalsignal is shown in Fig. 20, where the desired sidebands, alsoreferred to as optical channels, of the ODSB modulated SCMsignal are located at fc ± f1 Hz and fc ± f2 Hz.

The optical spectrum shown in Fig. 20 is an idealized repre-sentation, while that in Fig. 21 shows the optical harmonics andintermodulation products that are generated during SCM. If thenumber of SCM channels is N =2, then the desired sidebandsare at fc± f1 Hz and fc± f2 Hz. The spectrum of Fig. 21 alsocontains other sidebands, where the sidebands at fc±m f1 Hzand fc±m f2 Hz, with m being an integer, are referred to asoptical harmonics. The sidebands located at fc±(2 f1− f2) Hz,fc±(2 f2− f1) Hz and fc ± ( f1 ± f2) Hz in Fig. 21 are referredto as optical intermodulation products. Similarly, the harmonicsin the photo-detected signal are located at m f1 Hz and m f2

Hz, while the second-order and third-order intermodulationproducts are located at fi ± f j Hz and fh ± fi − f j Hz, re-spectively, where h, i, j ∈ {1,2, . . .N}, and h, i and j are SCMchannel indices [50]. For example, the non-linearity in a ROFlink that employs the SCM of two RF frequencies, namelythe WCMDA frequency of f1 = 1.9 GHz and the WLANIEEE 802.11b frequency of f2 = 2.4 GHz, is discussed in[51]. As seen in Fig. 9 and Fig. 10(b), the intensity is mod-ulated in a non-linear fashion around the bias point in exter-nal modulation, which creates harmonics and inter-modulationproducts.

B. Wavelength Division Multiplexing

While SCM employs a single optical carrier to transmitmultiple modulating signals, Wavelength Division Multiplex-ing (WDM) involves the fiber-based transmission of multipleoptical carrier frequencies. As shown in Fig. 22(a), each carrierfrequency is modulated by an electronic signal, thereby gener-ating sidebands around that carrier. Subsequently, the signalsare multiplexed to generate the spectrum seen in Fig. 22(a),


Fig. 22. Wavelength Division Multiplexing (WDM) (LD - Laser Diode; RF Tx - RF Transmitter; AWG - Arrayed Waveguide Grating). (a) A ODSB-WDMtransmitter model. (b) A basic MUX/DEMUX. (c) An interleaver.

Fig. 23. The various cost reduction and performance improvement techniques for ROF communications (LD - Laser Diode; RF Tx - RF Transmitter).

where the spectrum consists of three optical carriers. The cor-responding sidebands appear on either side of each carrier. InDense Wavelength Division Multiplexing (DWDM) [47], thechannel spacing Δωc of Fig. 22(a) has standardized values of12.5, 25, 50, 100 or 200 GHz [52]. Various optical componentsare employed in WDM based ROF networks Fig. 23, includinga multiplexer/demultiplexer and Optical Add Drop Multiplexer(OADM), which are detailed below.

A Multiplexer (MUX) and a Demultiplexer (DEMUX) isused for combining and separating the optical signals appear-ing at different optical carrier frequencies, respectively. MostDEMUXs are also capable of operating as a MUX, whenoperated in the reverse direction, i.e., when light enters fromthe output ports. As seen in Fig. 22(b), the most basic MUX/

DEMUX consists of a (1 × N) Arrayed Waveguide Grating(AWG), which relies on the principle of diffracting the light inorder to separate/combine the wavelengths [53]. Each channelconsists of the optical carrier along with its sidebands, as rep-resented by (ωc,i,si) in Fig. 22(b), while the multiplexed signalis indicated as (∑ωc,i,si). The MUX of Fig. 22(b) multiplexesindividual frequency bands, while the interleaver of Fig. 22(c) isemployed for bundling sets of even and odd indexed frequencybands. As shown in Fig. 24, the interleaver can be operated inthe reverse direction to implement de-interleaving. The OADMof Fig. 25 is employed at the RAP to “drop”, i.e., demultiplex, adownlink WDM channel, while “adding”, i.e., multiplexing, anuplink signal at the same or possibly different optical frequency.Fig. 25 shows the block diagram of an OADM, where the


Fig. 24. An interleaver operated as a de-interleaver.

Fig. 25. Optical Add Drop Multiplexer (OADM).

downlink signal carried by the optical carrier at fc,2 =ωc,22π Hz

is dropped and an uplink signal at the same optical carrierfrequency is added.

The throughput that can be achieved using WDM is higherthan that, which can be achieved using SCM. However, thiscomes at the cost of having to use multiple laser sources,MUXs, DEMUXs and OADMs, thereby giving rise to the costversus performance trade-off indicated in Fig. 5. It is also pos-sible to combine and use both SCM and WDM simultaneously,as mentioned in Fig. 5.

Equipped with a basic understanding of the different com-ponents of a simple ROF link, we now move on to discussinga range of advanced performance improvement techniques inSection V.

V. ROF PERFORMANCE IMPROVEMENT TECHNIQUES

There are numerous challenges associated with efficientmillimeter-wave ROF communication, including the generationand reliable transmission of the wireless signal over the opticalchannel. In this section we describe several techniques thatmay be employed for improving the ROF link’s performancein terms of throughput, Noise Figure (NF) or dynamic range.

A. Wavelength Interleaved Multiplexing

As per definition, in a Wavelength Interleaved WavelengthDivision Multiplexing (WI-WDM) system, the frequencyspacing between the optical carrier at fc,i =

ωc,i2π Hz and its

sideband at si2π Hz is higher than the frequency spacing Δ fc =

Δωc2π Hz between the adjacent optical carriers, which results

in the interleaved spectrum shown in Fig. 26. The spec-trum of Fig. 26 is generated by multiplexing signals that aregenerated using the OSSB modulation schemes discussed inSection III-A2 [47]. In other words, as shown in Fig. 26, wehave fr f > Δ fc, where fr f is the frequency of the modulatingRF signal. This technique is usually implemented along withthe DWDM technique that was discussed in Section IV-B.

Fig. 26. WI-DWDM signal.

Fig. 27. Wavelength Interleaved Wavelength Division Multiplexing (WI-DWDM) Demultiplexer (DEMUX) using Fabry–Perot (FP) filter (AWG -Arrayed Waveguide Grating; OC - Optical Circulator).

Fig. 28. Wavelength Interleaved Wavelength Division Multiplexing (WI-DWDM) Multiplexer (MUX) using Fabry–Perot (FP) filter (AWG - ArrayedWaveguide Grating; WDM - Wavelength Division Multiplexing; OC - OpticalCirculator).

WI-WDM results in a higher spectral efficiency than the con-ventional non-interleaved WDM, but requires more sophis-ticated techniques for multiplexing and demultiplexing thesignals [47], [54]. In the following paragraphs we highlightsome of these techniques, assuming OSSB modulation.

a) (2 × N) AWG with Fabry–Perot (FP) filter [47]: A(a×b) AWG consists of a input ports and b output ports, whilea Fabry–Perot (FP) filter is an optical filter having alternatingtransmission and reflection bands. Hence, effectively it acts as acomb filter. The incident optical wave will be either transmittedthrough or reflected by the FP filter, depending on the opticalfrequency. A DEMUX and MUX pair using an AWG and aFP filter is shown in Figs. 27 and 28, respectively [47]. Thecorresponding spectra at points A, B, C, and D of the DEMUXarchitecture using the FP filter are shown in Figs. 26 and 27.In the DEMUX of Fig. 27, the WI-DWDM signal, whose spec-trum is shown in Fig. 26, enters port 1 of the Optical Circulator(OC) and leaves from its port 2, where the set of sidebands seenin the WI-WDM signal of Fig. 26 are reflected, while the set ofcarriers are transmitted by the FP filter. The reflected sidebandsre-enter port 2 of the OC and leave from port 3. The separatedcarriers-set and sidebands-set are then fed into the two inputports of the AWG of Fig. 27, whose output consists of thedemultiplexed signals. As indicated by the spectrum at pointD of Fig. 27, each separated signal contains the carrier ωi andits corresponding sideband si [47].

The MUX of Fig. 28 operates in a similar fashion, exceptthat the direction of signal-flow is reversed [47]. As shown in


Fig. 29. Multiplexer/Demultiplexer (MUX/DEMUX) employing an ArrayedWaveguide Grating (AWG) in loopback configuration (OC- Optical Circulator).

Fig. 28, the independent channels enter the MUX, while the setof carriers and the set of sidebands appear at the top and bottomoutput port of the AWG in Fig. 28, respectively. The top andbottom outputs of the AWG in Fig. 28 have stylized spectrasimilar to those seen at points B and C of Fig. 27, respectively.The set of sidebands then enter port 1 and exit from port 2 of theOC, where, as shown in Fig. 28, these sidebands are reflectedby the FP filter, re-enter port 2 and finally exit from port 3 ofthe OC. On the other hand, the set of carriers that exited fromthe top port of the AWG of Fig. 28 are transmitted through theFP filter. Hence, they too exit from port 3 of the OC. Thus, themultiplexed signal becomes available from port 3 of the OC[47] of Fig. 28.

b) (2N+2)×(2N+2) AWG with loopbacks: A loopbackarchitecture is one in which the output ports are connected tothe input ports of the AWG. The architecture shown in Fig. 29implements simultaneous demultiplexing of the N downlinkchannels and multiplexing of the N uplink channels or vice-versa in a ROF system that employs WI-WDM. While thediscussion below assumes a scenario in which the downlinkchannels are demultiplexed and uplink signals are multiplexed,the processs is similar for the case of uplink demultiplexing anddownlink multiplexing.

In Demultiplexing [55], the downlink WI-DWDM signal,which consist of the N optical carriers and their correspondingsingle sidebands, is portrayed in Fig. 29 as ∑(ωd

c,i,sdi ). The

downlink power is split by a coupler and it enters ports A1 andA4 of the AWG via the circulators OCa and OCb of Fig. 29,respectively. The optical carriers along with their respectivesidebands exit from ports B1,B3 · ·B2N−1 of Fig. 29. These areobtained from port 3 of OC1 to OCN and are indicated in Fig. 29as (ωd

c,i,sdi ) [55].

In Multiplexing [55], the N uplink optical signals enter portsB1,B3 · ·B2N−1 via the OCs and are represented as (ωu

c,i,sui )

in Fig. 29. The optical carriers exit from port A4 while thesidebands exit from port A1 in Fig. 29. The output of port A4 isindicated as ∑ωu

c,i in Fig. 29, which are then looped back intoport B2. This results in the N separated optical carriers ωi outputat A3,A5 · ·A2N+1, which are then again looped back to portsB4,B6 · ·B2N+2. This results in the combined optical carriers∑ωu

c,i leaving from port A1 of Fig. 29. Thus, both the opticalcarriers and their sidebands exit from port A1. This WI-DWDMoptical signal then exits from port 3 of OCa and it is representedas ∑(ωu

c,i,sui ) in Fig. 29.

Fig. 30. Optical Add-Drop Multiplexer (OADM) (FBG- Fiber Bragg Gratingfilter: OC- Optical Circulator).

c) Wavelength-interleaved optical add-drop multiplexers:In Fig. 30, the optical OSSB-WDM signal of Fig. 26 enters intoport 1 of OC1 and leaves from port 2 [54]. Subsequently, theoptical carrier ωc,i+1, that is to be dropped, is reflected by theFBG filter 1 (FBG1), while the corresponding sideband si+1 isreflected by FBG2 [54], as shown in Fig. 30. The rest of thesignal enters port 1 of OC2 The reflectivity profile of the FBG1and FBG2 filters is as shown in Fig. 18(b), where the centerfrequency fr =

ωr2π has values of fc,i+1 =

ωc,i+12π Hz and si+1

2π Hz,respectively. The signals that are reflected by FBG1 & FBG2back-propagate and re-enter port 2 of OC1 and they are droppedfrom port 3, where the spectrum of the signal exiting fromport 3 is shown at point B of Fig. 30. The uplink signal, whosespectrum is shown at point C of Fig. 30, enters port 3 of OC2and exits from port 1. This uplink signal having a carrier ωup

c,i+1

and sideband supi+1 gets reflected by the FBG filters, because they

are at the same frequencies as the previously dropped signal andthey re-enter port 1 of OC2. Hence, the total signal enteringport 1 of OC2 includes the channels within the signal at pointA, which were not reflected by the FBG filters, along with theadded uplink channel. The signal entering port 1 finally exitsfrom port 2 of OC2, where the spectrum exiting from port2 is shown at point D of Fig. 30. Thus, a downlink channelis dropped from the spectrum of Fig. 26 and replaced by theuplink channel to obtain the spectrum at point D of Fig. 30.

Upon concluding our discussion on WI-WDM, recall thatthe first MUX/DEMUX architecture relies on a (2×N) AWG,a Fabry–Perot (FP) filter as well as on an Optical Circulator(OC) and can either perform multiplexing or demultiplexing.On the other hand, the second MUX/DEMUX architectureemploys the more complex (2N + 2)× (2N + 2) AWG withloopbacks along with several OCs, but can achieve simultane-ous multiplexing and demultiplexing of the signals. However,none of these MUX/DEMUX architectures permits access to anindividual channel or to a set of channels within the multiplexedsignal without having to demultiplex all the channels. This canbe implemented using the OADM alluded to in the previousdiscussion. Fig. 31 shows some commonly used ROF networktopologies along with the BSs, Remote Nodes (RNs) and RAPs,where the MUX/DEMUX and OADMs are employed assumingthat duplex communication with each RAP is achieved usinga unique wavelength. Fig. 31 also shows the operations thatMUX/DEMUXs and OADMs perform on both the Downlink(DL) and Uplink (UL) ROF signals.

Let us now move on from the family of techniques conceivedfor improving optical spectral efficiency to carrier suppressiontechniques, which are capable of increasing the depth of opticalmodulation, thereby improving the receiver sensitivity.


Fig. 31. Use of Multiplexers/Demultiplexers (MUXs/DEMUXs) and Optical Add Drop Multiplexers (OADMs) in different ROF topologies (BS- Base Station;RN- Remote Node; RAP- Radio Access Point; DL Downlink; UL- Uplink.

Fig. 32. Carrier suppression (LD- Laser Diode; RF Tx- RF Transmitter; FBG-Fiber Bragg Grating; PD- Photo-diode; BPF- Band Pass Filter).

B. Carrier Suppression

Suppressing the optical carrier with respect to the modulationsidebands increases the depth of optical modulation, therebybeneficially improving the attainable receiver sensitivity [49],[56], [66], [67], where the typical value of the optimum Carrierto Sideband Ratio (CSR) is 0 dB. Fig. 32 portrays an archi-tecture that implements carrier suppression, where the outputof the laser is intensity modulated using the RF signal andtransmitted over the fiber to the RAP. In the RAP, the modulatedoptical signal enters port 1 of the OC of Fig. 32 and then exitsfrom port 2, where it enters the FBG filter that implements car-rier suppression. The FBG filter, originally discussed in Fig. 18,reflects a portion of the optical carrier power corresponding tothe desired suppression, but ideally allows 100% transmissionof the optical sidebands. The reflected optical carrier powerenters port 2 and then exits from port 3 of the OC, where it canbe re-used for uplink modulation. The FBG filter of Fig. 32 canalso be employed before fiber transmission in order to simplifythe RAP architecture [56]. However, this latter technique wouldprevent the implementation of wavelength re-use, since thereflected carrier power would no longer be available at the RAPbut only at the BS.

Following this rudimentary introduction to the pertinent is-sues of carrier suppression, we now consider the family oftechniques designed for reducing the noise imposed by the ROFlink, which in turn improves both the optical receiver sensitivity

Fig. 33. Biasing the optical modulator at an optical power below that ofQuadrature Point (QP) biasing (MITP- Minimum Transmission Point).

as well as the dynamic range. These techniques include low-biasing the optical modulator, balanced photo-detection as wellas Optical Injection Locking (OIL) and optical feed-forwardlinearization.

C. Low-Biasing the Optical Modulator

As discussed in Section III, if the ODSB/OSSB modulationof Fig. 16 was carried out at the transmitter using a RF signal atfRF Hz, then the desired photo-detected signal is also generatedat fRF Hz. However, no signal is generated at fRF Hz for theOCS modulation scheme of Fig. 16, where a frequency-doubledphoto-detected signal is generated at 2 fRF Hz instead. ODSB/OSSB modulation requires a QP biased MZM that is associatedwith Vbias = Vπ/2, while OCS modulation requires a MITP bi-ased MZM with Vbias =Vπ. If no frequency doubling is desiredthen, the QP biased MZM of Fig. 8 is employed. Fig. 33 showsthe trade-off between the optical noise and optical signal power,when the bias point indicated by an arrow moves from the QPtowards the MITP of Fig. 33 [58]. It can be seen from (6) and(7) that unlike in the QP biased scenario of Fig. 33, the MITPbiased scenario of Fig. 33 has a suppressed first harmonic,which implies that the power of the photo-detected fRF Hz RFsignal decreases, when the bias point indicated by the arrow


moves from the QP towards the MITP. However, it can be seenfrom (6) and (7), that the DC value in the photo-detected signalis higher for the case of QP biasing than for MITP biasing.Thus, observe from (14) and (15) that the shot noise, RIN andsignal-ASE beat noise level, which together constitute the op-tical noise, decrease at a faster rate, when the bias point movesfrom the QP towards MITP [58], [59], as shown in Fig. 33. Ifthe dominant source of noise in an optical link that employsa QP biased MZM is the optical noise, then the output SNRincreases as the bias is shifted from the QP towards the MITP,which can be deduced from Fig. 33. The SNR increases till thephoto-detector’s thermal noise becomes the dominant sourceof noise, beyond which the SNR starts to decrease. Therefore,the optimum bias value corresponding to the maximum SNRlies between the QP and MITP, where its actual value dependson the optical power at the MZM’s input. The expression forthe Noise Figure (NF) and its relation to the Spurious FreeDynamic Range (SFDR) of an optical link is as follows on adecibel (dB) scale [59], [68]:

NF [dB] =SNRin[dB]−SNRout[dB] (30)

SFDR[dB] =23{IP3[dB]− (−174+NF [dB])} (31)

where IP3 is the level of the third-order inter-modulation prod-ucts. Hence, optimizing the bias point not only reduces thenoise figure, but also results in an increase of the link dynamicrange. A similar performance improvement can be achievedby low-biasing the EAM of Fig. 10(a) [69], [70] to an opticalpower level below that shown in Fig. 10(b).

However, the limitation of this technique is that it assumesthat the effect of the stronger second-order distortion, includingharmonics and intermodulation distortion, generated by low-biasing can be neglected. It can be seen from (6) and (7) thatunlike in the QP biased scenario, the MITP biased scenariohas a strong second order harmonic. In a SCM system havingN channels, the second order harmonics do not lie within thetransmission bandwidth if we have 2 fi > fN ∀i, where fi isthe electronic frequency of the ith channel being modulatedonto the optical carrier. The non-linearity that generates second-order harmonics also results in the generation of second-orderintermodulation products of the form of fi ± f j, where i, j ∈{1,2, . . .N}. Second-order intermodulation products can beneglected if they do not lie within the SCM signal a bandwidthi.e., ( fi ± f j) < f1 or ( fi ± f j) > fN ∀i, j. The conditionsfor neglecting the second-order harmonic and intermodulationdistortion are met, if the SCM signals are within an octave, i.e.,fN < 2 f1 [58].

Let us now discuss a second technique that assists in reducingthe noise imposed by the optical link, namely balanced photo-detection.

D. Balanced Photo-Detection

Balanced photo-detection schemes involve the subtraction oftwo received signals, which results in the desired componentsin those signals being added, while the noise being mitigated.Fig. 34 shows the block diagram of balanced photo-detection,

Fig. 34. Balanced photo-detection (LD- laser Diode; PD- Photo-diode; MZM-Mach–Zehnder Modulator).

where the laser output is fed to the dual-output MZM thatgenerates a pair of modulated outputs [60]. The first output isthe standard output of an MZM with an intensity of P1(t) =

Pin

[1+ cos

(πVbias

Vπ+ πV (t)

Vπ

)], while the other output has an

intensity of [71]:

P2(t) = |E(t)|2 =∣∣∣∣sin

(πVbias

2Vπ+

πV (t)2Vπ

)√2Pine jωct

∣∣∣∣2

=Pin

[1− cos

(πVbias

Vπ+

πV (t)Vπ

)], (32)

Both outputs are photo-detected after transmission through thefiber and as shown in Fig. 34, balanced photo-detection involvesthe subtraction of the two photo-detected signals, where thebalanced photo-detector output is given by [60]:

Ibalanced = I1(t)− I2(t) = R ·P1(t)−R ·P2(t)

=2R ·Pin · cos

(πVbias

Vπ+

πV (t)Vπ

)(33)

where R is the responsivity of the photo-diode. Again, it canbe seen from (33) that the desired signals are added, whilethe intensity noise that is common to the two photo-detectedsignals is suppressed by the difference operation [60]. A linkthat has a sub-10 dB NF can be achieved using balancedphoto-detection [60].

The next two techniques belonging to the family of opticalnoise-reduction techniques, namely Optical Injection Locking(OIL) and optical feed-forward linearization, also belong tothe family of techniques designed for increasing the linearityof optical modulation. We now discuss OIL, which assists ingenerating high-power, low-noise laser outputs.

E. Optical Injection Locking

Optical Injection Locking (OIL) is a technique conceived forgenerating a high-power laser output that has both a low inten-sity noise and a low phase noise. Fig. 35 shows the schematic ofOIL, where the output of the low-power, low-noise master-laserof Fig. 35 is injected into a high-power slave-laser. The slave-laser is constrained to operate at the frequency of the injectedlaser, provided that this frequency is sufficiently close to itsown free-running frequency. The difference between the free-running frequency and the injection-locked frequency of theslave-laser is referred to as the frequency detuning, where usinga higher injected power results in a higher allowable frequencydetuning. As shown in Fig. 35, the output of the slave-laseris directly modulated by the electronic RF transmitter beforetransmission over the fiber, while the isolator ensures that noreflected signals enter the master-laser.


Fig. 35. Optical Injection Locking (LD- Laser Diode; PD- Photo-diode; RFTx- RF Transmitter; BPF- Band Pass Filter).

Fig. 36. Optical Sideband Injection Locking (BB- Baseband; IF - Intermedi-ate Frequency; PD- Photo-diode; BPF- Band Pass Filter).

OIL has several advantages [63]. For example, when usingOIL, the laser’s relaxation oscillation frequency fr, as shownin Fig. 7, increases. Hence, direct modulation results in weakerdynamic non-linearity induced harmonics and inter-modulation[72]. Additionally, it can be seen from Fig. 7 that the increase infr is accompanied by an increase in the modulation bandwidthof the laser [64]. Furthermore, it is possible to directly modulatethe laser using frequencies that are close to its relaxationoscillation frequency without the usual increase in the RINnoise [64]. When OIL is used, it becomes possible to generatea high-fidelity, high-power optical signal, which enables thedesigner to achieve a high link-SNR. Finally, OIL can also beemployed for generating OSSB signals [73].

Two variations of OIL that employ heterodyne photo-detection are discussed in the following sub-sections, includingOptical Sideband Injection Locking and Optical Injection PhaseLocked Loops. These techniques result in performance im-provements at the cost of an increased transmitter complexity.

a) Optical sideband injection locking: A so-called dual-mode signal consists of two optical frequencies, where hetero-dyne detection may be used for generating a RF signal, as dis-cussed in Section II-C. The performance of a link that employsthis technique can be improved by additionally invoking theconcept of OIL. Typically, it is the slave laser’s output that isRF modulated during OIL. However, as shown in Fig. 36, it isalso possible to injection-lock two -rather than just one- slavelasers to different sidebands of a master-laser [74] that has beenmodulated using a RF tone at fLO = ωLO

2π Hz. This OIL operationis then followed by the modulation of either one or of bothsidebands using the BB/IF signal at fIF = ωIF

2π Hz. As shownin Fig. 36, the subsequent heterodyne photo-detection of thesehigh-power sidebands generates the high-fidelity upconvertedRF signal appearing at ]N · fLO − fIF ] Hz. More explicitly, Op-tical Sideband Injection Locking (OSIL) facilitates the injectionlocking of the slave-lasers to higher-order optical sidebands,i.e., to harmonics that are present within the output of the RF-modulated master-laser [61], where the upconverted photode-tected signals can be generated at [N · fLO ± fIF ] Hz with Nbeing an integer.

The optical phase noise can be further reduced by amalga-mating OIL with Optical Phase Lock Loops (OPLLs), which

Fig. 37. Optical Injection Phase Locked Loop(OIPLL) (PD- Photo-diode; BBTx- Baseband Transmitter; BPF- Band Pass Filter).

Fig. 38. Optical feed-forward linearization (RF Tx - RF transmitter; LD -Laser Diode; PD - Photo-diode; BPF - Band Pass Filter).

however results in added transmitter complexity. This hybridtechnique is discussed in the next section.

b) Optical injection phase locked loop: The OPLL relieson an architecture designed for maintaining phase coherencebetween the signals that are employed in coherent photo-detection. An OPLL reduces the optical phase noise imposedon the signal transmitted over a ROF link. OIL can be com-bined with the OPLL technique in order to amalgamate thebenefits of both techniques. Such a transmitter is referred to asan Optical Injection Phase Locked Loop (OIPLL) [62], [75],which is shown in Fig. 37. The master-laser operating at anoptical frequency of fc =

ωc2π Hz, is modulated by a RF tone at

fLO = ωLO2π Hz, as shown in Fig. 37. The master-laser output is

split by the coupler C1 into two parts, one of which enters port 1and exits via port 2 of the OC, where this signal is employed ininjection locking. The slave-laser in Fig. 37 is injection lockedto one of the first-order (N = 1) or higher-order sidebands(N > 1) in the output of the master-laser. The slave-laser inFig. 37 generates the injection locked output at [ fc +N · fLO]Hz and also reflects the output of the master-laser. As shown inFig. 37, the output of the slave-laser exits from port 3 of the OC,followed by a splitting operation by the coupler C2 [75]. Theupper output of C2 in Fig. 37 is combined with the upper outputof C1 to generate the OIPLL output, while the lower output ofC2 is locally heterodyne photo-detected to generate a RF toneat N · fLO Hz, which is then input to the feedback loop. In thefeedback loop, output of the fLO Hz oscillator is frequency-multiplied by N and mixed with the photo-detected N · fLO HzRF tone. The mixer output depends on the phase differencebetween the two signals. As shown in Fig. 37, this mixer outputis fed to a loop filter which generates a feedback that modulatesthe slave-laser. Hence, the loop-filter assists in phase-matching.Meanwhile, the OIPLL output is intensity modulated by theBaseband Transmitter (BB Tx), then transmitted over the fiberand finally photo-detected to generate an upconverted signal atN · fLO Hz [75].

In addition to noise reduction, OIL also resulted in theenhanced linearity of the direct optical modulation scheme of


TABLE IRADIO OVER FIBER (ROF) PERFORMANCE IMPROVEMENT TECHNIQUES (PART 1)

Fig. 6(a) [63]. We now move on to another technique, namelyoptical feed-forward linearization, that helps enhance the lin-earity of both direct as well as external optical modulation, apartfrom suppressing the optical noise.

F. Optical Feed-Forward Linearization

Optical feed-forward linearization is a technique that maybe employed at the transmitter for pre-compensating the non-linear effects of optical modulation. This technique is imple-mented using the architecture seen in Fig. 38 [65]. As seen inFig. 38, the input RF signal power is split into two branchesby the RF coupler CRF,1, where the upper branch modulates theoptical carrier generated by laser L1, while the lower branch isused as an uncorrupted reference signal. The modulated opticaloutput of L1 seen in Fig. 38 contains both the RF signal aswell as the inter-modulation and harmonic distortion terms,which are additionally corrupted by RIN noise. This opticalsignal from L1 is then split using a 50:50 optical coupler Cop,1,as seen in Fig. 38, where the lower-branch output of Cop,1is photo-detected by the photo-diode D1, amplified and thensubtracted from the lower branch output of the RF couplerCRF,1. The output of the RF coupler CRF,1 does not suffer fromthe degradations and noise imposed by optical modulation,because it is generated by the electronic RF transmitter and notby photo-detecting the modulated optical signal [65]. Hence,the subtraction operation generates a difference signal thatideally contains only the distortion terms imposed by opticalmodulation and the RIN noise [65]. This electronically delayeddifference signal is then inverted in Fig. 38 and modulated ontothe output of laser the L2. Note that the laser L2 operates atan optical frequency fL2 =

ωL22π , which is significantly different

from the optical frequency fL1 =ωL12π of L1. As seen in Fig. 38,

the modulated output of laser L2 is then combined with the op-

tically delayed modulated output of laser L1 using the 90:10 op-tical coupler Cop,2 and transmitted over the fiber. If A1(t)e jωL1t

and A2(t)e jωL2t are the optical signals received at photo-detectorD2 from lasers L1 and L2, respectively, then their coher-ent photo-detection by photo-detector D2 results in the signal:

Plinear =R∣∣A1(t)e

jωL1t +A2(t)ejωL2t

∣∣2=R |A1(t)|2 +R |A2(t)|2+2R |A1(t)| |A2(t)|cos [ j(ωL1 −ωL2)t] (34)

where again, R is the photo-diode’s responsivity. Since the op-erating frequencies of the two lasers are significantly different,i.e., (ωL1−ωL2) is large, the third term of (34) would be locatedwithin the electronic spectrum at a frequency that is far higherthan that of the first two terms. The first two terms represent thecoherent addition of the two signals. Since R|A2(t)|2 carries aninverted replica of the distortion plus RIN noise in R|A1(t)|2,this addition results in the cancellation of the distortion plusRIN noise term [65]. Hence, as per (31), the SFDR will besignificantly improved, because the feed-forward linearizationtechnique of Fig. 38 suppresses the both inter-modulationproducts as well as the laser RIN over a large bandwidth.

We now conclude our discussions on the family of noise-reduction techniques. Low-biasing of the optical modulatoris simpler than the subsequently discussed balanced photo-detection, but the former technique constrains the usable SCMsignals to be within an octave. While the extent of noise-reduction obtained from the various noise-reduction techniquesdiscussed in this paper depends on the specifications of thecomponents used, OIL in general provides a much higherperformance improvement than the rest. However, this comesat the cost of a higher transmitter complexity.

Before we move on to the next section, we provide, in Table I,a list of the seminal papers of the performance improvement


Fig. 39. Linearization using dual parallel MZMs (LD- Laser Diode; MZM-Mach–Zehnder Modulator; PD- Photo-diode; BPF- Band Pass Filter).

techniques discussed so far in Sections V-A, V-B, V-C, V-D,V-E and V-F.

As mentioned previously, the last two techniques in thefamily of optical noise-reduction techniques, namely OIL andoptical feed-forward linearization, also assist in increasingthe linearity of optical modulation. Let us now continue ourdiscourse by moving on to another subset of techniques thathelp increase the linearity of optical modulation, namely to thefamily of linearized optical modulators.

G. Linearized Optical Modulators

There are several architectures that linearize the transfer-characteristics of the MZM of Fig. 9(a), which result in theefficient suppression of the even-order and third-order inter-modulation products in the photo-detected signal. As a resultof linearization, the SFDR of the links employing these archi-tectures is maximized, since the SFDR of (31) now depends onthe much weaker fifth-order inter-modulation products, ratherthan on the third-order inter-modulation products. Some ofthe architectures within this subset include both the paralleland serial MZMs architectures, as well as the dual wavelengthequalization and polarization mixing to be detailed below.

a) Parallel and serial MZMs: It can be observed from (6)that the Quadrature Point (QP) biasing of a MZM results innegligible second-order distortions, but significant third-orderdistortions in the photo-detected signal, where the polarity ofthe third-order distortion depends on the quadrature bias volt-age that is applied in Fig. 8. Dual-parallel linearization drawsits motivation from the fact that the third-order distortions canbe suppressed by superimposing signals in which the polarity ofthe third-order distortion is opposite [76]. As shown in Fig. 39,the output of the LD is fed to the coupler C1 having a powersplitting ratio of η1, where the pair of outputs gleaned fromthe coupler C1 are fed to the MZMs Ma and Mb, which areQP biased at Vπ/2 and −Vπ/2, respectively [76]. As shownin Fig. 39, the pair of MZMs are driven by the same RFsignal V (t), which have a voltage ratio of γ. The outputs ofthe pair of MZMs are combined by the coupler C2 of Fig. 39having a power combining ratio of η2. Using (3), the expressiondescribing the photo-detected signal becomes:

I(t) = I0

∣∣∣∣√1−η1

√1−η2 cos

(π4+

V (t)2Vπ

)e jφa

+√

η1√

η2 cos

(−π

4+

γV (t)2Vπ

)e jφb

∣∣∣∣2

, (35)

where I0 is the photo-detected signal, if there was no dual-parallel modulation. The optical phase shifter of Fig. 39 ensures

that the phase-difference of the optical fields at the output ofthe pair of modulators is φa − φb = π/2. Then, the expressionof (35) becomes [76]:

I(t) = I0

[(1−η1)(1−η2)cos2

(π4+

V (t)2Vπ

)

+η1η2 cos2(−π

4+

γV (t)2Vπ

)]

= I0

[(1−η1)(1−η2)

(1− sin

(V (t)2Vπ

))

+η1η2

(1+ sin

(γV (t)2Vπ

))]

=∞

∑k=0

ck [V (t)]k . (36)

Using the Taylor series expansion of the sin(x) in the (36), itmay be readily shown that there are no even harmonics in thephoto-detected signal at all. Hence, naturally there would be nosecond-order or fourth-order distortions in the photo-detectedsignal either. The third-order inter-modulation product IIM3 andthe desired RF signal Isignal is proportional to the Taylor seriescoefficients c3 and c1, respectively. Hence [76], we have

IIM3 ∝c3 ∝[(1−η1)(1−η2)−η1η2γ3] (37)

Isignal ∝c1 ∝ [(1−η1)(1−η2)−η1η2γ] . (38)

It may be observed from (37) that if the condition of (1−η1)×(1−η2)=η1η2γ3 is satisfied, then this would ensure having c3 =0 and hence we suppress the third-order inter-modulation prod-ucts. However, this would also reduce the value of c1 in (38),which would attenuate the desired signal. Hence, this lineariza-tion comes at the cost of a weaker desired signal, which wouldin turn result in a reduced output SNR and an increased NF.

This concept of using those values of power-splitting ratioand drive-voltage ratio that suppress the third-order distortioncan be extended to the architecture of Fig. 40 as well, where twodual-drive MZMs, namely MZM-A and MZM-B, are connectedin series and are QP biased with Vbias =Vπ/2 [77]. The splitteras well as the coupler of MZM-B has an asymmetric powercoupling ratio of η, while the ratio of the RF drive voltagesto the MZMs is γ. Assuming the MZMs of Fig. 40 operate inthe push-pull mode of operation discussed in Section II-A, thephoto-detected signal is:

I(t) = Io

∣∣∣∣12

ej( πVbias

2Vπ +πV (t)2Vπ

)+

12

e− j

( πVbias2Vπ +

πV (t)2Vπ

)∣∣∣∣2

×∣∣∣∣√η

√ηe

j( πVbias

2Vπ +πγV (t)

2Vπ

)

+√

1−η√

1−ηe− j

( πVbias2Vπ +

πγV (t)2Vπ

)∣∣∣∣2

=0.25Io

[1+ cos

(πVbias

Vπ+

πV (t)Vπ

)]

×[

0.5η2+0.5(1−η)2+η(1−η)cos

(πVbias

Vπ+

πγV (t)Vπ

)]

=0.25Io

[1− sin

(πV (t)

Vπ

)]

×[

0.5η2 +0.5(1−η)2 −η(1−η)sin

(πγV (t)

Vπ

)](39)

=∞

∑k=0

ck (V (t))k . (40)


Fig. 40. Linearization using serial MZMs (LD - Laser Diode; MZM - Mach–Zehnder Modulator; PD - Photo-diode; BPF - Band Pass Filter).

Fig. 41. Dual-wavelength equalization (LD - Laser Diode; MZM - Mach–Zehnder Modulator; PD - Photo-diode).

Using the Taylor series expansion of (39) we get ck = 0 forall even k in (40), while those values of γ and η are chosenwhich would result in c3 = 0, thereby suppressing the third-order distortions.

It is seen from Fig. 39 and Fig. 40 that linearizing the opticalmodulation using parallel or serially connected MZMs relieson a single laser and on multiple MZMs. We now discuss analternate technique designed for linearizing optical modulation,namely dual-wavelength equalization, which relies on multiplelasers and on a single MZM.

b) Dual-wavelength equalization: The strength of boththe required signal and of the intermodulation products in thephoto-detected signal depend on the switching voltage Vπ ofthe MZM in Fig. 8. If two optical signals modulated by thesame data are photo-detected and subtracted from each other,as seen in Fig. 41, then it is possible to suppress the third-order intermodulation products without totally suppressing therequired signal, provided that the switching voltages employedin the two modulations schemes are different [78]. To expanda little further, Fig. 41 shows the block diagram of an architec-ture that implements dual-wavelength equalization, where theoutputs of two laser diodes L1 and L2 are multiplexed using theWDM scheme of Fig. 22(b) and fed into a MZM modulatorthat is driven by the RF signal V (t) and biased at Vbias Volts.Each of the two laser diodes operates at different wavelengths,where we set the ratio of the MZM switching voltages Vπ of(5) at these wavelengths to r [78]. As shown in Fig. 41, aftertransmission through the fiber, the signal is demultiplexed usingthe scheme of Fig. 22(b) and the two wavelengths are separatelyphoto-detected, amplified and then subtracted. The ratio of thephoto-detected signals after amplification is A, when there isno optical modulation. The mathematical expression of thedifference signal is as follows:

I(t) = Io

[1+ cos

(πVbias

Vπ+

πV (t)Vπ

)]

−AIo

[1+ cos

(πVbias

Vπ+

πV (t)rVπ

)]

=∞

∑k=0

ck [V (t)]k . (41)

Fig. 42. Linearization using polarization mixing (LD- Laser Diode; MZM-Mach–Zehnder Modulator; PD- Photo-diode; TE- Transverse Electric; TM-Transverse Magnetic).

Using the Taylor series expansion of the cos(x) terms and ex-ploiting the fact that the desired RF component Isignal as well asthe third-order inter-modulation product IIM3 are proportionalto the co-efficients c1 and c3, respectively, we arrive at [78]:

Isignal ∝c1 ∝[

sin

(πVbias

Vπ

)− A

rsin

(πVbias

rVπ

)](42)

IIM3 ∝c3 ∝[

sin

(πVbias

Vπ

)− A

r3 sin

(πVbias

rVπ

)]. (43)

The bias voltage Vbias is chosen so that the MZM is QP biasedor at least near-QP biased for both wavelengths, where wehave πVbias

Vπ= 2m+1

2 π and πVbiasrVπ

= 2n+12 π with m and n being

integers. When this specific biasing is used, we have c2 =[cos

(πVbias

Vπ

)− A

r2 cos(

πVbiasrVπ

)]≈ 0, i.e., the photo-detected sig-

nal would only contain highly suppressed second-order inter-modulation products. When the MZM is QP biased, IIM3 of

(43) becomes [78]: IIM3 ∝[1− A

r3

]. Hence, the third-order

inter-modulation products can be suppressed, provided that thecondition of A = r3 is maintained [78]. However, maintainingthis would reduce the strength of Isignal in (42) and hence wouldincrease the NF.

The techniques conceived for linearizing the MZM trans-fer curve of Fig. 9(a) that have been discussed so far eitheremployed multiple MZMs, as seen from Fig. 39 and Fig. 40,or employed multiple lasers, as seen from Fig. 41. We nowcontinue our discourse about linearized optical modulators bydiscussing a technique that relies on a single MZM and a singlelaser to linearize optical modulation, but requires a polarizer.

c) Polarization mixing: The polarization of light charac-terizes the direction of its electric field vector. If an opticalsignal that is propagating along the y-axis and is polarized atan angle θ with respect to the z axis, as seen from Fig. 42,enters the modulator, then it results in a super-position of apair of waves within the MZM that have orthogonal polariza-tions, which are typically referred to as the Transverse Electric(TE) and the Transverse Magnetic (TM) waves [84]. Since theMZMs are sensitive to the polarization of the optical signal, theswitching voltages V T E

π and V T Mπ are different for each of these

polarizations, as shown in the MZM transfer characteristicsof Fig. 42. The switching voltages and the bias voltage Vbias

applied satisfy the following conditions:

πVbias

V T Eπ

=π(4m+1)

2and

πVbias

V T Mπ

=π(4n+3)

2(44)


where m as well as n are integers and γ = V T Eπ

V T Mπ

may be usedfor denoting the ratio of V T E

π and V T Mπ . It can be seen from the

MZM transfer characteristics of Fig. 42 that for a bias voltageof Vbias both waves experience QP biased modulation, and thatthe transfer characteristics have opposite slopes at this biasvoltage. The pair of waves can be mathematically describedby the orthogonal components of a vector. Hence, the photo-detected signal becomes:

I(t) = I0

∣∣∣∣sin(θ)cos

(π4+

V(t)2Vπ

)x+cos(θ)cos

(−π

4+

γV(t)2Vπ

)z

∣∣∣∣2

= I0

[sin2(θ)cos2

(π4+

V(t)2Vπ

)+cos2(θ)cos2

(−π

4+

γV(t)2Vπ

)]

= I0

[sin2(θ)

(1−sin

(V(t)2Vπ

))+cos2(θ)

(1+sin

(γV(t)2Vπ

))]

=∞

∑k=0

ck (V (t))k . (45)

Here I0 is the unmodulated photo-detected signal. Using theTaylor series expansion of the trigonometric terms in (45), itmay be readily observed that there are no even harmonics (i.e.,we have ck = 0 for k being even) in the photo-detected signal.This is because the MZM is QP biased for both modes. Thethird-order inter-modulation IIM3 and the desired RF signalIsignal are proportional to the Taylor series coefficients c3 andc1, respectively. Hence from (45) we have,

IIM3 ∝c3 ∝[sin2(θ)− cos2(θ)γ3] (46)

Isignal ∝c1 ∝[sin2(θ)− cos2(θ)γ

](47)

If the condition of cos2(θ)sin2(θ)

= 1γ3 is satisfied, then this would

ensure c3 = 0 in (46) and hence we would suppress third-order inter-modulation products in the photo-detected signal

I(t) of (45) [84], where physically cos2(θ)sin2(θ)

is the ratio of the

optical power in the TE and TM modes of Fig. 42. However,according to (47), this would also reduce the value of c1, i.e.,it would reduce the desired signal. On the other hand, unlikethe architecture of Fig. 42 discussed here, the one in [79] didnot rely on the TE and TM modes having opposite slopes at theQP of the MZM transfer curves in Fig. 42. Instead, it relied ontwo linear polarizers for suppressing the non-linear sidebandswithin the optical field. Explicitely, the authors of [79] can-celled some of the optical sidebands that contribute to thethird-order distortion of the photo-detected signal, namely thesidebands at fc + 2 f1 − f2 Hz and fc + 2 f2 − f1 Hz. Here,fc Hz is the optical frequency of the laser whose output wasexternally modulated using a dual-tone RF signal at f1 Hz andf2 Hz by the linearized modulator. Finally, similar polarizationmixing techniques may also be employed for linearizing a phasemodulator [85].

Having studied the subset of linearized modulators withinthe family of techniques designed for increasing the linearityof optical modulation, we now study how optical and electricalfiltering can be invoked for increasing the linearity of opticalmodulation in the next two techniques to be discussed below.

Fig. 43. Optical spectrum of non-linear OSSB modulation.

Fig. 44. Linearization using optical filtering (LD - Laser Diode; PD - Photo-diode; BPF - Band Pass Filter).

H. Optical Filtering

Fig. 43 [80] shows the optical spectrum after an opticalcarrier at fc=

ωc2π Hz has been OSSB modulated by two RF tones

having frequencies of f1 =ω12π Hz and f2 =

ω22π Hz. Fig. 43 [80]

also shows the spectrum of the photo-detected signal. As shownin Fig. 43, the most significant inter-modulation products in thephoto-detected signal are the third-order inter-modulation prod-ucts which are located at (2ω1−ω2) and (2ω2−ω1). The majorpairs in the optical spectrum of Fig. 43 that beat together duringphoto-detection and generate the third-order inter-modulationproduct at (2ω1−ω2) are those at {ωc,ωc−2ω1+ω2} and{ωc−ω1,ωc+ω1−ω2}. Similarly, the major pairs in the opticalspectrum of Fig. 43, that beat together during photo-detectionand generate the inter-modulation product at (2ω2−ω1) arethose at {ωc,ωc−2ω2+ω1} and {ωc−ω2,ωc−ω1+ω2}. Whilethe sidebands at (ωc−ω1) and (ωc−ω2) in Fig. 43 are therequired modulation sidebands, those at (ωc−2ω1+ω2) and(ωc−2ω2+ω1) are too close to the required sidebands to beeasily filtered out using low-cost optical filters. By contrast, thesidebands at (ωc+ω1−ω2) and (ωc−ω1+ω2) may be readilyfiltered out using inexpensive optical filters. The architecturethat implements this filtering technique is shown in Fig. 44,where the unmodulated laser output power is split into twohalves using the coupler C1. One of the outputs of C1, namelythe upper branch, is OSSB modulated followed by optical filter-ing for removing the sidebands at (ωc+ω1−ω2) and (ωc−ω1+ω2) in Fig. 43. The optical filter also filters out the opticalcarrier at ωc. Then, after accurate attenuation and phase-shift,the optical carrier is re-introduced using the other output ofcoupler C1, namely the lower branch of Fig. 44. This techniquecan also be extended to three-tone modulation by filtering outthe components at (ωc±ωm∓ωn) for m,n∈{1,2,3}.

We now study the final technique designed for increasingthe linearity of optical modulation, namely pre-distortion andpost-compensation. These techniques, unlike the previous tech-niques of the family, are implemented in the electronic domain,rather than in the optical domain.


Fig. 45. Linearization using pre-distortion and post-compensation filters, as discussed in [82]}(RAP- Radio Access Point; O-E- Optical to Electronic; E-O-Electronic to Optical).

I. Pre-Distortion and Post-Compensation

The non-linearities encountered in optical modulation can becompensated for at the optical transmitter and receiver by pre-distortion and post-compensation, respectively, using electro-nic filters. [82, Fig. 45] shows the block diagram of anarchitecture that employs transmitter pre-distortion and receiverpost-compensation filters in the BS, where again, the pre-distortion filter compensates for the non-linear effects of theROF Downlink (DL) prior to the actual transmission, while thepost-compensation filter removes the higher order sidebandsgenerated by the non-linear effects after the ROF Uplink (UL)transmission [82]. Both these filters are located at the BS andhence have the advantage of centralized signal processing.This filter-based compensation technique may be used in ROFlinks employing both the direct [83] as well as the externalmodulation schemes of Fig. 6(a) and Fig. 8, respectively.

During DL transmission, the data signal is firstly pre-distorted using the non-linear characteristic of the filter, thenupconverted to RF and finally employed in modulating the DLoptical carrier, as seen in Fig. 45. This DL optical signal istransmitted to the RAP, where it is photo-detected and thenwirelessly transmitted to the MS. The UL wireless signal re-ceived at the RAP of Fig. 45 is employed for modulating the ULoptical carrier followed by transmission to the BS, where theUL ROF signal is photo-detected and then downconverted. Fi-nally, the signal is fed to the post-compensation filter of Fig. 45and then decoded. These filters can also be made adaptive, sincethe modulated response varies significantly between variouslasers [82]. The sinusoidal transfer function of a MZM shownin Fig. 9(a) can be compensated by applying a arcsine filter5 tothe modulating signal prior to optical modulation [86].

We now conclude our discussions on the family of techniquesconceived for increasing the linearity of optical modulation.While the OIL techniques of Figs. 35–37 significantly increasethe linearity of direct optical modulation, they cannot be usedfor increasing the linearity of the external optical modulationof Fig. 8. Moreover, they significantly increase the complexityof the optical transmitter. It requires polarization matchingbetween the master- and slave-laser, which are connected usingsophisticated polarization maintaining fibers. Unlike OIL, theoptical feed-forward linearization technique of Fig. 38 may bebeneficially applied for eliminating the distortions in the case of

5A filter in which the output is related to the input through a inverse sinefunction.

both direct as well as external modulation [65]. However, thistechnique requires both accurate amplitude and phase matchingfor attaining a beneficial distortion and noise reduction, whichmay be achieved by using the amplifiers and delay elementsof Fig. 38. The phase-mismatch introduced by the chromaticdispersion of the fiber may be compensated, provided that thetransmitter is designed for a fixed fiber length, which wouldhowever relatively limit the employment of such a transmitter.Additionally, it is worth noting that this linearization techniqueis also associated with a relatively high implementational cost[83]. Employing the linearized MZMs of Figs. 39–42 resultsin the suppression of the distortion, but it is accompanied bythe suppression of the desired signal as well. The subsequentlydiscussed optical filter-based linearization of optical modula-tion. as shown in Fig. 44, results in significant performanceimprovements. However, this scheme relies on the filtered side-bands found at (ωc+ω1−ω2) and (ωc−ω1+ω2) in the opticalspectrum of Fig. 43 being sufficiently far from the desiredsidebands at (ωc−ω1) and (ωc−ω2), thereby imposing restric-tions on the usable RF frequencies. We finally discussed theelectronic filter-based technique of Fig. 45, which is effectivein the case of external modulation, but cannot easily suppressboth the higher harmonics and the inter-modulation products indirect modulation [65]. Moreover, the operational bandwidth ofthese electronic techniques is limited [65]. Nevertheless, theirstrength is their relatively low implementational cost, whencompared to other optical techniques [83].

Before we move on to the next section, we provide, inTable II, a list of the seminal papers of the performance im-provement techniques discussed in Sections V-G, V-H and V-I.

Having focussed our attention on the benefits of increasingthe linearity of optical modulation, let us now concentrate ourattention on the family of techniques that help in overcomingfiber-nonlinearity.

J. Overcoming the Effects of Fiber Non-Linearity

As was discussed in Section II-B, XPM and SPM are the un-desired consequences of the fiber’s non-linearity. The detrimen-tal effects of these phenomena can be greatly reduced by usingconstant-amplitude optical modulation schemes like optical fre-quency and optical phase modulation, which however requiresadditional components for photo-detection, as was discussedin Section III-B. Additionally, two basic techniques have beenproposed in the literature for suppressing the effects of fibernon-linearity, including multiple fiber transmission [87] and


TABLE IIRADIO OVER FIBER (ROF) PERFORMANCE IMPROVEMENT TECHNIQUES (PART 2)

Fig. 46. Multiple fiber transmission for overcoming the effects of fiber non-linearity (LD-Laser Diode; MZM- Mach–Zehnder Modulator; PD- Photo-diode; WDM- Wavelength Division Multiplexing).

carrier suppression [87]. Multiple fiber transmission is similarto the balanced photo-detection technique of Fig. 34, since bothtechniques involve the removal of the undesired components byadding two replicas of the same signal. However, multiple fibertransmission requires the transmission of the signals through apair of identical fibers at different wavelengths. For example,as seen in Fig. 46, a pair two signals s1 and s2 may bemodulated onto the wavelengths λ1 generated by laser LD1 andλ2 generated by laser LD2, respectively and then transmittedthrough the fiber F1 for photo-detection. Similarly, these signalsare also modulated onto the wavelengths λ2 generated by laserLD4 and λ1 generated by laser LD3 and then transmittedthrough the fiber F2 for photo-detection, as portrayed in Fig. 46.The specific choice of these four wavelengths should meet thefollowing criteria [87]: λ1−λ2 = λ1− λ2 and D1 =−D2, whereD1 and D2 are the dispersion parameters of the fibers F1 and F2,respectively, as defined in Section II-B. In such a scenario theXPM crosstalk contamination imposed on the received signalsin the pair of optical fibers are 180◦ out of phase and hencethey would cancel out each other after their addition [87]. Thedisadvantage of this technique is the need to transmit the samesignal through multiple fibers, thereby reducing the achievablethroughput.

On the other hand, as discussed in Section V-B, carriersuppression improves the receiver sensitivity of the link, whichimplies that the transmitter can operate at a lower transmitpower, thereby reducing the effects of fiber non-linearity.

Let us now continue our discourse by considering the familyof techniques designed for overcoming chromatic dispersion.

K. Overcoming Chromatic Dispersion

Chromatic dispersion, originally discussed in Section II-B,results in symbol-broadening [24] and an associated power-reduction penalty [34], and hence increases the BER of the ROFlink. A ROF link that employs the ODSB modulation techniqueof Fig. 16(a) along with the direct photo-detection technique ofFig. 13(a) constitutes the most common ROF implementation,which is a direct consequence of its low cost as well as low com-plexity. However, the disadvantage of employing ODSB mod-ulation is the above-mentioned chromatic dispersion-inducedpower penalty, where the chromatic dispersion imposes aphase-shift of the optical signal propagating through the fiber.The frequency-dependent phase-change imposed by chromaticdispersion results in a phase-difference between the upperand lower sidebands in the ODSB signal of Fig. 16(a). Thephoto-detected signal is proportional to the optical powerP(t) = |E(t)|2, with E(t) being the optical field-strength ofthe ODSB signal, where the optical power is attenuated owingto the destructive interference between the beating side-bands. Hence, this optical power-attenuation is referred to asdispersion-induced power-penalty. The attenuation of the op-tical power results in the attenuation of the electronic powerPr f of the photo-detected signal at fr f Hz, where the extent ofattenuation is given by [34]:

Pr f ∝ cos2

[π ·L ·D ·λ2

c · f 2r f

c ·[1− 2

π · arctan(αchirp)]]

(48)

where D is the chromatic-dispersion parameter (also calledGVD parameter) of (10) discussed in Section II-B, L is thelength of the fiber, λc is the wavelength of the optical carrier, cis the speed of light in vacuum and αChirp is the chirp parameterof the signal. Chirping in an intensity-modulated optical signalrefers to a time-dependent variation of the instantaneous opticalfrequency from its ideal value of fc =

cλc

Hz. Thus, chromaticdispersion limits the both the affordable fiber lengths L and themodulating-signal frequencies fr f , when ODSB modulation is


Fig. 47. Optical Single Side Band (OSSB) signal spectrum. (a) Using Elec-tronic Double Side Band (EDSB) signal; (b) Using Electronic Single Side Band(ESSB) signal.

employed [94]. Chromatic dispersion can be compensated byusing specialized fibers called Dispersion Compensation Fibers(DCF) [24]. If Dcomp is the dispersion parameter of the DCF,then chromatic dispersion can be overcome by propagating theoptical signal through Lcomp meters of the DCF, after it haspropagated through L meters of the standard SMF, provided thatthe following condition is met [24]:

Dcomp ·Lcomp =−D ·L. (49)

The compensation of non-linearity in optical modulation us-ing electronic filters was discussed in Section V-I. Similarly,even chromatic dispersion can be compensated using electronicequalizers [10].

Additionally, other techniques that have been proposed in theliterature for overcoming chromatic dispersion include employ-ing dispersion-tolerant schemes like the OSSB and OCS mod-ulation schemes of Fig. 16(b) and (c), apart from employingthe fiber-gratings of Fig. 49, chirped optical signals, the opticalphase conjugation technique of Fig. 53 and the exploitation ofSPM [89], which are detailed below.

a) Dispersion-tolerant modulation schemes: Employingthe OSSB and OCS optical modulation formats of Fig. 16(b)and (c), respectively, results in a high degree of immunity tothe dispersion-induced power penalty [39], [91]. Additionally,the BS transmitter may modulate a BB signal onto the opti-cal carrier, while, at the RAP, the heterodyne photo-detectiontechnique of Fig. 13(b) would ensure the generation of an up-converted RF signal that does not suffer from the dispersion-induced power penalty.

The detrimental effect of chromatic dispersion increasesas the optical bandwidth of the signal increases. Hence, theresilience of OSSB modulation to chromatic dispersion canbe further improved by employing electronic single-sidebandmodulating signals during optical modulation, instead of theusual electronic double-sideband signal [34]. Fig. 47(a) and(b) illustrate the optical spectra of OSSB signals that havebeen generated using electronic double-sideband and single-sideband signals, respectively [34]. Even when OSSB modula-tion is employed, some residual chromatic dispersion persistsbecause the modulating signal has a finite bandwidth of 2Baround the RF carrier fRF , as seen in Fig. 47(a). As seenfrom Fig. 47(b), the excess-bandwidth can be eliminated byemploying an electronic single-sideband signal during opticalmodulation, instead of the usual electronic DSB signal [34],which leads to a further reduction of the detrimental effects ofchromatic dispersion.

Fig. 48. Chromatic dispersion compensation using a Tapered LinearlyChirped Grating (TLCG) (LD - Laser Diode; RF Tx - RF Transmitter; PD -Photo-diode; BPF - Band Pass Filter).

Fig. 49. Linearly chirped Fiber Bragg Grating (FBG).

Fig. 50. Optical delay of Tapered Linearly Chirped Grating (TLCG).

However, unlike the ODSB modulation, the OSSB and OCSmodulations, primarily rely on the external modulation tech-nique of Fig. 8(b) that employs a dual-drive MZM [34]. Thus,if we are constrained to employ ODSB modulation, then wemay overcome the limitations imposed by chromatic dispersionby relying on the techniques discussed below, where employingfiber-gratings is discussed first.

b) Fiber gratings: As discussed in Section II-B, chro-matic dispersion occurs because the propagation velocity ofthe different frequencies contained in the optical signal aredifferent. Thus, chromatic dispersion manifests itself in theform of the various frequency components having differentdelays. This delay difference can be equalized by using thetechnique of Fig. 48 that relies on the Tapered Linearly ChirpedGrating (TLCG)6 optical filter [88], where the output of a laseroperating at an optical frequency of fc =

ωc2π Hz is modulated

using a RF signal and then transmitted over the fiber. At thereceiver of Fig. 48, the optical signal enters port 1 and exitsfrom port 2 of the optical circulator (OC). The signal then entersthe TLCG and it is reflected back into port 2 of the OC. Finally,the signal exits from port 3 and then it is photo-detected. Asseen from Fig. 50, within its reflectivity bandwidth7 that iscentered around the optical carrier of ωc, the TLCG introducesan optical delay that is a near-linear function of the opticalfrequency. Thus, as seen from Fig. 50, the frequency-dependent

6The linearly chirped FBG of Fig. 49, unlike the conventional FBG ofFig. 18(a), has a linear increase/decrease of the otherwise constant grating pe-riod. The TLCG consists of a linearly chirped fiber grating that is implementedin a tapered fiber.

7Band of frequencies pre-dominantly reflected by the optical filter.


Fig. 51. Power attenuation induced by chromatic dispersion [27].

dispersion-induced delay introduced by the optical fiber can becompensated by using a TLCG [88].

While this technique is potentially capable of compensatingthe effects of chromatic dispersion, it also requires an additionaloptical filter with a carefully designed optical delay profile. Inthe next section, we discuss a technique, namely the employ-ment of chirped optical signals, that is easier to implement.

c) Chirped optical signals: The dispersion-inducedpower-reduction of ODSB signals was formulated in (48),where it can be observed that the extent of power-reductiondepends on both the magnitude and polarity of the chirpparameter αchirp of the transmitted signals. Fortunately, thepower-reduction could be lessened by introducing chirp in thetransmitted signal, as shown in Fig. 51, while ensuring that thischirp has the reverse polarity with respect to the chirp inducedby the chromatic dispersion [27]. It can be seen from (3)that the dual-drive MZM of Fig. 8(b) introduces an adjustablechirp, where the value of αchirp depends on the amplitudesof both the drive voltages that are applied to the arms of thedual-drive MZM, i.e., on V1(t) and V2(t). As discussed inSection II-A, zero-chirp operation is possible, if the dual-driveMZM of Fig. 8(b) operates in the push-pull mode. If V1(t) andV2(t) are sinusoidal signals associated with amplitudes of V1

and V2, respectively, then we have [27]:

αchirp =V1 +V2

V1 −V2tan

(π(Vbias −Vπ)

2Vπ

)

= ± V1 +V2

V1 −V2(ODSB signal) (50)

where Vbias =(2n+1)Vπ

2 for ODSB modulation for n = 0,1, . . ..The dual-drive MZM typically has a non-ideal extinction ratioof γ �= 1, where for the case of ODSB signal transmission

we have αchirp = V1+γ2V2γ(V1−V2)

[95]. As discussed in Section II-A,the ideal value of γ is 1. Fig. 51 compares the dispersion-induced power attenuation when using unchirped and chirpedoptical signals, where the dispersion-induced power attenuationis shown for the case of ODSB transmission, that employs anoptical wavelength of 1550 nm, a sinusoidal RF modulatingsignal of 12 GHz and a fiber having a dispersion parameter ofD = 17 ps/nm/km [27]. It can be seen from Fig. 51, that using

Fig. 52. Chromatic dispersion compensation using Self-Phase Modulation(SPM) (LS- laser Diode; RF Tx- RF Transmitter; PD- Photo-diode; BPF- BandPass Filter; RF Rx- RF Receiver).

Fig. 53. Chromatic dispersion compensation using Optical Phase Conjugation(LS - laser Diode; RF Tx - RF Transmitter; PD - Photo-diode; BPF - Band PassFilter).

negative values of αchirp reduces the dispersion-induced powerattenuation [27].

While the above techniques rely on a MZM for generatinga chirp that compensates the dispersion-induced chirp, thetechnique discussed in the next section relies on the otherwiseundesired effect of SPM for compensating the dispersion-induced chirp.

d) Beneficial exploitation of SPM: As discussed inSection II-B, SPM is imposed by the fiber’s non-linearity andit becomes significant, when a high optical transmit-poweris employed. This is the case in the architecture of Fig. 52owing to inserting an optical amplifier in the transmitter. SPMintroduces a chirp that has the opposite polarity to the chirpthat is introduced by chromatic dispersion [89]. Thus, SPM canbe employed for reducing the effects of chromatic dispersion[89], where the extent of SPM-induced chirp depends both onthe optical power levels and on the fiber length. However, thetechnique of Fig. 52 may need the use of an additional opticalequipment, namely an optical amplifier, which also adds ASEnoise. The main limitation of this technique is the inevitableaccompanying non-linear phenomenon of Stimulated BrillouinScattering (SBS), which was discussed in Section II-B. Hence,the resultant SBS must be suppressed using other techniques.Additionally, the high optical power levels used in this tech-nique may result in fiber’s non-linearity inducing significantinter-modulation in SCM systems and XPM in WDM systems.

We now continue our discourse on the family of techniquesdesigned for overcoming chromatic dispersion by consider-ing the optical phase conjugation technique that aims foreliminating the dispersion-induced phase-difference betweenthe upper and lower sidebands in the ODSB signal of Fig. 16(a).

e) Optical phase conjugation: An ODSB signal suffersfrom the dispersion-induced power-reduction imposed by thephase-difference between the upper and lower sidebands. Thispower-penalty can be avoided by swapping the two sidebandsmidway through the fiber-length [90]. This would ensure asimilar dispersion-induced phase-shift in both the upper andthe lower sidebands, which eliminates the detrimental phase-difference. This technique is referred to as Optical Phase Conju-gation (OPC). Fig. 53 shows the block diagram of optical phaseconjugation, where an ODSB signal is generated by modulating


Fig. 54. Optical phase conjugation using Four Wave Mixing (FWM) (LSB -Lower Side Band; USB - Upper Side Band).

an optical carrier at ωc using a RF signal at ωRF . The ODSBsignal consists of the optical carrier along with lower andupper sidebands at ωc −ωRF and ωc +ωRF , respectively. Thesignal enters the phase conjugator of Fig. 53 after propagatinghalf-way through the fiber length. As shown within the phaseconjugator of Fig. 53, a high-power optical carrier at ωpump,that is referred to as the pump signal, is coupled with the opticalsignal being transmitted. This combined signal is opticallyamplified, followed by optical filtering to remove the opticalamplifier’s out-of-band ASE noise discussed in Section II-Dand it then enters a Semiconductor Optical Amplifier (SOA).The FWM process originally introduced in Section II-B, thatoccurs within the SOA between the pump signal ωpump andthe modulated signal, generates an additional signal as per (11)and (12) [90], where the signal-spectrum at the output of theconjugator is shown in Fig. 54. It should be noted that thewavelength-conversion efficiency is polarization sensitive andhence the modulated signal and pump signal must be polar-ization matched using a polarization controller, prior to theircoupling. It can be seen from Fig. 54 that the lower sidebandof the original signal corresponds to the upper sideband in thenew signal and vice-versa. The resultant optical signal is thenfiltered, amplified and transmitted over the remaining half ofthe fiber link. Additionally, the effect of SPM, originally intro-duced in Section II-B, is partially compensated by OPC [90].Naturally, perfect compensation of the power-dependent SPMwould have been possible only if the power variation along thefiber length was perfectly symmetric with respect to the positionof the phase conjugator [90]. However in reality, the opticalpower gradually reduces as the signal propagates through boththe fiber segments in Fig. 53.

We now conclude our discussions on the family of techniquesconceived for overcoming chromatic dispersion and move onto the next performance improvement technique, namely Dig-itized Radio Over Fiber (DROF). The digital optical link usedin long-distance wired optical communication provides a betterperformance than the ROF link of Fig. 3. Hence, DROF aimsfor incorporating the advantages of a digital optical link in ROFcommunication.

L. DROF Architecture

All the ROF architectures discussed so far modulate theanalog RF signal onto the optical carrier, as seen from Fig. 3,which, for the sake of contrast, we refer to as Analog RadioOver Fiber (AROF). By contrast, in Digitized ROF (DROF), the

Fig. 55. Bandpass sampling.

RF signal is sampled, digitized and then is used for modulatingthe optical carrier. The main motivation behind the design of theDROF architecture is the desire to have a digital optical link andat the same time rely on centralized signal processing at the BS[92]. In the DROF architecture of Fig. 56, the signal processingrequired for generating the DL RF signal is carried out at theBS, where the RF signal is digitized before modulating theoptical carrier [92]. Conventional sampling of the RF signal ata rate higher than twice the highest frequency present in thesignal using the Nyquist Theorem is not practically feasible,because of the high values of the wireless carrier frequency.Hence, typically critical bandpass sampling is used [92], whichis illustrated in Fig. 55, where the sampling frequency is chosento satisfy the aliasing-free condition, so that the original analogRF signal can be reconstructed by using a bandpass filtercentered at fRF = ωRF

2π Hz and having a bandwidth of B Hz. Thevalid bandpass sampling rates for a bandwidth of B are [92]:

2(

fRF + B2

)k

≤ fs ≤2(

fRF − B2

)k−1

∀1≤k ≤⌊

fRF + B2

B

⌋(51)

where w� represents the integer part of w and k is an integer.Thus, when employing bandpass sampling, the sampling ratedepends on the bandwidth of the signal, rather than, on thehighest frequency present in the signal [92].

Fig. 56 illustrates the block diagram of a DROF architecture[92], where the DL BB signal is generated, upconverted anddigitized using an Analogue to Digital Converter (ADC) at theBS. The digitized signal is then employed in modulating theoptical carrier. At the RAP, the photo-detected BB signal is fedto a Digital to Analogue Convertor (DAC) and the RF signalis reconstructed. The reconstructed RF signal is then amplifiedusing a power amplifier and transmitted over the wirelesschannel. As shown in Fig. 56, the UL signal received at the RAPis amplified, bandpass filtered, digitized and employed for mod-ulating the optical carrier. At the BS, the photo-detected signalis fed to a DAC, followed by down-conversion to a BB signal.The transmitted UL bits are then detected from the BB signal,by employing Digital Signal Processing (DSP) techniques [92].

The advantage of employing optical BB modulation is thatone can use inexpensive direct modulation [92] or an externalmodulator having a low-modulation bandwidth [96]. Further-more, typically a performance improvement is achieved by em-ploying a digital optical link. Employing a DROF architecture,that involves the transmission of optical signals carrying BB di-gital data, has a high degree of tolerance to the previously men-tioned dispersion-induced power-penalty [3] and also ensures


Fig. 56. DROF link block diagram.

TABLE IIIRADIO OVER FIBER (ROF) PERFORMANCE IMPROVEMENT TECHNIQUES (PART 3)

a higher degree of linearity of the optical-link. Hence, DROFprovides a better performance than the conventional ROF archi-tecture of Fig. 3 [92], [93]. However, the trade-off in this caseis the need for using ADCs and DACs and hence DROF has amore complex and costly RAP than the conventional ROF linkof Fig. 3. Furthermore, BB optical modulation is also employedin fiber-based wired optical communication and has been exten-sively dealt with in various tutorials [97] and books [24].

At this point it is worth mentioning that despite the lowsampling rate achieved using bandpass sampling, the bit rateof the DROF link may still be high, because it is the productof the sampling rate and the ADC’s bit-resolution. Recently, avariant of the DROF technique, namely sampled RF-over-fiber,was proposed to overcome this challenge [93]. The solution in[93] had a similar architecture as Fig. 56, with the differencethat the ADC only samples the RF signal using a sample-and-hold circuit without quantizing or encoding the sampled values.The sampled baseband signal is then transmitted over the ROFlink. In the UL, the BS’s receiver filters out the down-convertedversion of the RF signal from the spectrum of the basebandsampled signal, while in the DL, the RAP’s receiver filters theimage at the RF frequency. However, in the DL, employing ahigher carrier frequency results in a weaker filtered image.

Before we conclude our discussions on the performanceimprovement techniques, we provide, in Table III, a list ofthe seminal papers of the performance improvement techniquesdiscussed in Sections V-J, V-K and V-L.

We now conclude our discussions on the performance im-provement techniques designed for ROF communications byreferring back to Fig. 4 and Fig. 23. It becomes plausible from

our discussions on the performance improvement techniquesthat while the designer has access to the plethora of techniqueslisted in Fig. 23 to choose from, there is also a cost to bepaid, as seen from Fig. 4. The higher the implementational costand complexity of the technique, the better the performanceimprovement is expected to be. Hence, we complete our dis-course with a discussion of the various cost-reduction tech-niques designed for ROF communications, as listed in Fig. 23.

VI. COST REDUCTION TECHNIQUES

The various cost reduction techniques proposed for ROFcommunications, as seen from Fig. 23, include the family oftechniques designed for integrating ROF communication withthe existing Fiber To The Home (FTTH) optical networks, thefamily of techniques conceived for implementing wavelengthreuse and the set of techniques proposed for reducing the num-ber of lasers, which will be discussed in Section VI-A–VI-C,respectively.

A. Integration With Existing FTTH Optical Networks

The penetration of the optical fiber to the end user is in-creasing at a rapid rate, because the telecommunication serviceproviders are attempting to support increased wired data ratesfor their customers. These networks are referred to as Fiber ToThe Home (FTTH) networks. It should be noted here that wiredsignals in this manuscript refer to the fiber signals for stationarydevices like personal-computers, while wireless signals refersto the signals for mobile devices. The wired signal that is carried


Fig. 57. Using separate optical carriers for optical baseband and optical RFmodulation (LD - laser diode; RF Tx - RF transmitter; BB Tx - BasebandTransmitter; FTTH - Fiber To The Home)

by the FTTH networks is of baseband (BB) nature, while theROF signal is of RF nature. Hence, transmitting the ROF sig-nals over both the existing FTTH networks [103] and indoor op-tical networks [104] would result in significant cost advantagesand in efficiently exploiting the bandwidth of the existing fibers.This kind of integration can be achieved in the following ways:

1) Employing separate optical carriers for optical BB andoptical RF modulation [105];

2) Employing a single optical carrier for two-step optical BBand optical RF modulation;

3) Employing a single optical carrier for simultaneous opti-cal BB and optical RF modulation.

Note that the terms optical BB modulation and optical RFmodulation in this manuscript refer to the modulation of elec-tronic BB and electronic RF signals onto the optical carrier,respectively.

1) Employing Separate Optical Carriers for Optical Base-band and Optical RF Modulation: Separate optical carriers canbe employed at the BS for transmitting the BB wired signal andthe ROF wireless signal. As shown in Fig. 57, lasers operatingat optical frequencies of fbaseband = ωbaseband

2π Hz and fROF =ωROF

2π Hz are modulated by the BB wired signal and the RF wire-less signal, respectively. These two signals are then multiplexedfor generating a signal, whose spectrum is shown in Fig. 57.As shown in Fig. 57, the multiplexed signal is transported overthe FTTH network to the home of the end user, where it isdemultiplexed. The optical signal carrying the BB data directlysupports the wired devices, while the ROF signal is fed to aRAP that supports indoor wireless communication [103]. Inthis case the BS would be several kilometers away from thecustomer premises. In the case of larger houses/buildings thedemultiplexed ROF signal would enter an in-building opticalnetwork that supports several RAPs [106].

2) Employing a Single Optical Carrier for Two-Step Opti-cal Baseband and RF Modulation: Instead of using separateoptical carriers, the wired BB data and the wireless RF signalmay be modulated onto the same optical carrier at the BS intwo modulation steps and then transmitted to the RAP. Fig. 58shows the block diagram of a 2-step modulation for integratingROF and FTTH transmission [101]. Observe in Fig. 58 thatthe output of the laser operating at an optical power of Pin isBB modulated by the bit stream b(t) using an MZM having a

switching voltage of Vπ. Then, the BB modulated optical signalpower PBB is:

PBB = Pin

[1− cos

(πb(t)

Vπ

)](52)

where we have b(t) = 0 and b(t) = Vπ for bits ‘0’ and ‘1’,respectively. As shown in Fig. 58, the output of the BB modula-tion step is then OCS modulated using a RF tone VLO cos(ωLOt)that drives a MITP biased MZM to generate a signal, whoseoptical power Ptotal is:

Ptotal =PBB

[1− cos

(πVLO cos(ωLOt)

Vπ

)]

=PBB

[1− J0

(πVLO

Vπ

)

−2∞

∑n=1

(−1)nJ2n

(πVLO

Vπ

)cos(2nωLOt)

]. (53)

Then, as shown in Fig. 58, the signal is transmitted through theFTTH network and then fed to an optical coupler that splitsthe optical signal into two parts. The lower-branch is photo-detected using a low-bandwidth photo-detector for recovering

the BB wired signal represented by PBB

[1− J0

(πVLO

Vπ

)]in (53),

while upper-branch of the coupler is photo-detected using ahigh-bandwidth photo-detector to recover a RF signal repre-

sented by 2PBBJ2

(πVLO

Vπ

)cos(2ωLOt) in (53) [101].

We now continue our discourse on the family of techniquesdesigned for integrating ROF communication with the existingFTTH networks by considering simultaneous optical basebandand optical RF modulation.

3) Simultaneous Optical Baseband and Optical RF Modu-lation: Instead of using the two-step modulation of Fig. 58,the wired BB data and the wireless RF signal could be si-multaneously modulated onto the same optical carrier at theBS and then transmitted to the customer premises. This canbe achieved using several techniques, including the use of theconventional external modulator (i.e., MZM and EAM) and ofthe sophisticated nested-external modulators.

a) Simultanous optical baseband and optical RF modu-lation relying on an external modulator: Simultaneous modu-lation can be achieved using the MZM of Fig. 8 or the EAMof Fig. 10(a), where the architecture that achieves simultaneousoptical baseband and RF modulation using an external modula-tor is shown in Fig. 59. Observe in Fig. 59 that the output of thelaser is modulated by a MZM that is biased at Vbias = −Vπ/2and driven by a combination of the BB wired data b(t) and thewireless RF signal xr f (t). Subsequently, the modulated signalis transmitted from the BS to the end user over the FTTHnetwork, where it is photo-detected and filtered to separate theBB wired and RF wireless data. Using (5), the photo-detected

signal I(t) ∝ Pout(t) = Pin

[1+ sin

(π(b(t)+xr f (t))

Vπ

)], which can

be expanded using Taylor series [98] to:

Pout(t)Pin

= 1+ sin

(πb(t)

Vπ

)+ cos

(πb(t)

Vπ

)πxr f (t)

Vπ

−12

sin

(πb(t)

Vπ

)π2x2r f (t)

V 2π

− 16

cos

(πb(t)

Vπ

)π3x3r f (t)

V 3π

. . . , (54)


Fig. 58. Employing a single optical carrier for two-step optical baseband and RF modulation (LD- laser diode; MZM- Mach–Zehnder Modulator; BB Tx-Baseband Transmitter; PD- Photo-diode; BPF- Band Pass Filter; FTTH- Fiber To The Home; LPF- Low Pass Filter).

Fig. 59. Simultaneous optical baseband and optical RF modulation using an external modulator (LD - laser diode; RF Tx - RF transmitter;BB Tx - BasebandTransmitter; PD - Photo-diode; BPF - Band Pass Filter; FTTH - Fiber To The Home; LPF - Low Pass Filter).

where,

b(t) ={

V for bit ‘1’−V for bit ‘0’

In (54), the first two terms contain the desired BB signal, i.e.,

PBB(t) = 1 + sin(

πb(t)Vπ

)and the third term has the desired

RF signal, Pr f (t) = cos(

πb(t)Vπ

)πxr f (t)

Vπ, while the higher order

terms produce intermodulation distortion both at the BB and

RF frequencies. cos(

πb(t)Vπ

)is the same for both values of b(t),

which implies that the desired RF signal in the third term of(54) is not corrupted by the BB signal [98]. However, the BBtransmission affects the power of the photo-detected RF signal.For the case of pure optical RF modulation, i.e., when b(t) = 0,the power in the desired third term of (54) would be maximized,

because we have∣∣∣cos

(πb(t)

Vπ

)∣∣∣ = 1, while for simultaneous

optical modulation we have∣∣∣cos

(πb(t)

Vπ

)∣∣∣< 1, because b(t) �= 0.

The performance of BB modulation is improved by using alarger difference between the optical power PBB(t) transmittedfor bit ‘0’ and bit ‘1’. This can be achieved by increasingthe amplitude V of b(t), which, however, reduces the value of∣∣∣cos

(πb(t)

Vπ

)∣∣∣ and hence reduces Pr f (t). Hence, the power in the

detected RF signal is reduced when simultaneous optical BBand RF modulation is employed. For the case of pure optical BBmodulation, i.e., for xRF(t) = 0 in (54), we would have chosenV = Vπ/2 to ensure that the MZM is operating at its minimumand maximum PBB(t) for the bits ‘0’ and ‘1’, respectively.However, for the scenario that involves simultaneous BB andRF modulation, it is evident from (54) that choosing V =Vπ/2would result in the desired third term Pr f (t) going to zero,i.e., no RF transmission would occur. Thus, the BB opticalmodulation has to be carried out using non-optimum values ofPBB(t) [98].

Additionally, the architecture of Fig. 59 can be implementedusing the EAM of Fig. 10(a), where the condition of avoid-

Fig. 60. Optical modulation using the Electro-Absorption Modulator (EAM).

ing overmodulation imposes an upper limit on the total drivevoltage that can be applied to the EAM [99]. The BB signalb(t) is combined with the RF signal xr f (t) and the combinedsignal is then added to the bias voltage Vbias to generate thedrive voltage, which is applied to the EAM. In this scenario,the output intensity that would be photo-detected is [99]:

I(t) ∝ P(t) ∝ T 2 (Vbias +b(t)+ xr f (t)) (55)

where T (v) is the transmittance of the EAM described inSection II-A. Expanding (55) using Taylor series we get:

I(t) ∝ T 2 (Vbias +b(t))

+2T (Vbias +b(t))

[d (T (v))

dv

]v=Vbias+b(t)

xr f (t)+ . . . (56)

In (56), the first term contains the desired BB signal, whilethe second term has the desired RF signal [99]. The higherorder terms would produce intermodulation distortion at boththe BB and RF frequencies. The BB signal is represented byb(t) = V for bit ‘1’ and b(t) = −V for bit ‘0’. For the caseof pure optical RF modulation, i.e., for b(t) = 0, the power inthe desired second term would be maximized because the slope

of the transmission curve, i.e.,[

d(T (v))dv

], is maximized at V =

Vbias, which is evident in Fig. 60. However, for simultaneous


Fig. 61. Simultaneous optical baseband and optical frequency-doubled RF modulation (LD - laser diode; RF Tx - RF transmitter; BB Tx - Baseband Transmitter;PD - Photo-diode; BPF - Band Pass Filter; FTTH - Fiber To The Home; LPF - Low Pass Filter; MZM - Mach–Zehnder Modulator).

Fig. 62. Simultaneous optical baseband and optical frequency-doubled RF modulation using nested MZM (LD - laser diode; RF Tx - RF transmitter; BB Tx -Baseband Transmitter; PD - Photo-diode; BPF - Band Pass Filter; FTTH - Fiber To The Home; LPF - Low Pass Filter; MZM - Mach–Zehnder Modulator).

optical modulation, the slopes at V = Vbias ±V become lowerthan that at V = Vbias, as seen from Fig. 60 and hence thephoto-detected RF signal power drops [99]. The performanceof BB modulation is improved by using a larger difference inthe optical power transmitted for bits ‘0’ and ‘1’, which can bearranged by increasing the amplitude V of b(t). However, thisreduces the slope values at V =Vbias±V . Thus, BB modulationhas to be carried out using non-optimum values in the EAMtransmission curve of Fig. 60.

In addition to the architecture of Fig. 59, the system shownin Fig. 61 also employs simultaneous optical BB and RF modu-lation with the additional advantage of being able to achievefrequency-doubling. Frequency-doubling helps generate highRF signals using lower electronic oscillator frequencies andoptical modulators of lower electronic bandwidth. ROF trans-mission that implements frequency-doubling can be carriedout along with simultaneous BB transmission, by using aninterleaver [100]. In the architecture in Fig. 61, the output ofthe laser, operating at an optical frequency of fc =

ωc2π Hz, is

modulated using a MZM that is driven by a combination of thewireless RF signal at ωRF/2 and the wired BB signal. As shownin Fig. 61 [100], the modulated BB data is present at the opticalcarrier ωc, while the RF data is carried by the two first-orderoptical sidebands at ωc ±ωRF/2. After transmission from theBS through the FTTH network, these optical sidebands are sep-arated from the signal at the optical carrier frequency, by usingan optical interleaver in the end user’s home. Subsequently, asshown in Fig. 61, the sidebands are photo-detected to generate afrequency-doubled wireless RF signal at ωRF , while the signalat the optical carrier frequency is photo-detected to retrieve theBB wired signal [100], [107].

It may be concluded that the cost of carrying out simultane-ous optical BB and optical RF modulation, as opposed to sep-arate modulations, is that of striking a performance trade-off.

Explicitely, the amplitudes of b(t) and xRF(t) in Fig. 59 haveto be chosen in such a way that the required Quality of Service(QoS) can be achieved for both the BB and RF transmission[99]. A similar trade-off exists for the case of simultaneousmodulation using a tri-band signal consisting of two RF signalsat microwave and millimeter-wave frequencies along with a BBsignal [29].

The stringency of this trade-off can be relaxed to some extentby using the more sophisticated nested-external modulator, asdetailed in the next section.

b) Simultanous modulation relying on a nested-externalmodulator: Simultanous optical BB and optical RF modulationmay also be implemented using the architecture of Fig. 62. thatuses a nested-MZM instead of the conventional MZM, wherethe output of the laser diode operating at fc =

ωc2π Hz is split into

two branches, which then modulate a nested MZM consistingof three MZMs referred to as MZM-a, MZM-b and MZM-c inFig. 62. Specifically, MZM-a of Fig. 62 is biased at MITP andit is driven by the RF signal at a frequency of fRF/2 = ωRF/2

2π Hzto achieve OCS modulation. On the other hand, On OFF Keying(OOK) [97] optical BB modulation may be implemented usingMZM-b, while MZM-c is biased at the MATP [102]. As seenin Fig. 62, the spectrum of the optical signal at the output ofthe nested MZM consists of the modulated BB data presentat the optical carrier ωc, while the RF data is carried by thetwo first-order optical sidebands at ωc ± ωRF/2 [102]. Aftertransmission from the BS to the customer premises through theFTTH network, the signal enters port 1 of the OC shown inFig. 62 and exits from port 2, where the FBG filter of Fig. 18(a)reflects the signal at ωc but lets the sidebands pass through.Subsequently, as shown in Fig. 62, the sidebands are photo-detected to generate a frequency-doubled wireless RF signal atfRF = ωRF

2π Hz, while the reflected optical signal enters port 2and exits from port 3 of the OC in Fig. 62. This optical signal is


TABLE IVRADIO OVER FIBER (ROF) COST REDUCTION TECHNIQUES (PART 1)

then photodetected to retrieve the BB wired signal. The majoradvantage of this technique is that the two modulation schemescan be independently optimized [102]. However, for a fixedtotal optical power received at the photo-diode, increasing themodulation depth of one of the schemes increases its modu-lated optical power and hence improves this scheme’s perfor-mance, while degrading the other scheme’s performance [102].

Thus, the family of techniques proposed for integrating ROFcommunication with existing FTTH networks involves a trade-off between the performance of the ROF link and the basebandtransmission. When designing the wireless networks, the effectof such a trade-off becomes important in the light of the limitedpower budget. It can be seen from the discussions in this sectionthat the value of PRAP

RF considered in Section II-E reduces, whenthe ROF link is integrated with the FTTH network. Thus, it canbe concluded from (19) that this reduction would degrade theSNRROF of the ROF link and hence would affect the wirelessnetwork parameters, like the cell-size and RAP amplifier gain,as discussed in Section II-E. We conclude our discussions onthe family of techniques proposed for integrating ROF com-munication with existing FTTH networks by stating that theyrely on infrastructure sharing to implement multiplexed opticalcommunication. On the other hand, the techniques discussed inthe next section, namely the family of techniques designed forwavelength reuse, relies on infrastructure sharing to implementduplexed optical communication.

Before we move on to the next section, we provide, inTable IV, a list of the seminal papers of the family of techniquesproposed for integrating ROF communication with existingFTTH networks, as discussed in Section VI-A.

B. Wavelength Re-Use

Wavelength re-use is an optical duplexing technique, whereUplink (UL) communication is achieved by re-using the Down-link (DL) optical signal at the RAP. This leads to significant costadvantages, because the RAPs do not need to have their ownoptical sources, when wavelength re-use is employed. However,allocating a portion of the DL optical power to UL communi-cation results in the degradation of the DL performance. Thevarious ways of achieving wavelength re-use include OpticalCarrier Recovery and Re-use (OCRR), Optical Re-Modulation

Fig. 63. Optical Carrier Recovery and Re-use (OCRR) in the RAP employingan optical coupler and FBG (PD - Photo-diode; BPF - Band Pass Filter; FBG -Fiber Bragg Grating).

(ORM) and employing a modulator-cum-photo-detector, whichwill be discussed in Section VI-B1–VI-B3, respectively.

1) Optical Carrier Recovery and Re-Use (OCRR): Thetechniques that involve the separation of the total or a fractionof the DL optical carrier power from the DL sidebands, forits reuse in UL communication, are referred to as OCRRtechniques.

In the RAP of Fig. 63 [115], the DL OSSB signal poweris split using an optical coupler, where the signal of the upperbranch is photo-detected, while the lower branch-signal is fedto a FBG filter, whose transmission spectrum is also shown inFig. 63. The FBG passes the optical carrier but reflects 100% ofthe sideband power. An isolator is incorporated at the input ofthe FBG filter to prevent the back propagation of the reflectedsideband. The separated optical carrier that becomes availableat the output of the FBG filter is then modulated by the UL RFsignal and transmitted from the RAP to the BS over the fiber[115], hence achieving wavelength re-use.

In the RAP of Fig. 64 [115], [118], the DL OSSB signalenters port 1 of the OC and exits from port 2, where a FBGfilter, whose spectrum is also shown in Fig. 64, reflects 50% ofthe optical carrier’s power. The remaining 50% of the opticalcarrier’s power along with 100% of the modulation sidebandpower propagates through the FBG filter and then it is photo-detected for generating the DL wireless signal. The reflected50% of the optical carrier power enters the OC from port 2, exitsfrom port 3 and then it is employed for optical UL modulation.For the case of an ODSB DL signal, the FBG filter of Fig. 64can be designed for reflecting 100% of the optical carrier power,while the pair of sidebands propagate through the filter and


Fig. 64. Optical Carrier Recovery and Re-use (OCRR) in the RAP using anOC and FBG (FBG - Fiber Bragg Grating; OC - Optical Circulator; PD - Photo-diode; BPF - Band Pass Filter).

Fig. 65. Optical Carrier Recovery and Re-use (OCRR) in the RAP using aninterleaver (PD - Photo-diode; BPF - Band Pass Filter).

are photo-detected for generating a frequency-doubled DL RFsignal [40].

OCRR can also be achieved by employing an optical inter-leaver at the RAP [119]. As shown in the RAP of Fig. 65, theDL ODSB signal is fed to an interleaver that separates the op-tical carrier at ωc from the pair of first-order sidebands at ωc ±ωRF/2, where fRF/2 = ωRF/2

2π is the frequency of the RF signalgenerated in the BS. The optical sidebands are then photo-detected to generate a frequency-doubled signal at fRF Hz[119], [119], [120]. As shown in Fig. 65, the separated opticalcarrier of the lower branch can then be employed for wave-length re-use. If second-order optical sidebands at ωc ± ωRF

are generated instead of the first-order ones, for example byusing a MATP biased MZM at the BS, then, in the RAP, theirseparation from the optical carrier followed by the subsequentphoto-detection yields a frequency-quadrupled signal [122].

As mentioned previously in Section V-A, a commonly em-ployed WDM ROF signal is the WI-DWDM signal of Fig. 26.The architecture of Fig. 66 [117] may also be employed toimplement wavelength re-use in the RAP of a ROF systememploying WI-WDM signals, where again, the input DL WI-DWDM spectrum shown in Fig. 66 was discussed in Section V-A. The WI-DWDM signal enters port 1 of the OC and exitsfrom port 2. The channel that is to be dropped at the RAP is{ωc,i+1,si+1}. Hence, as shown in Fig. 66, the signal exitingport 2 is passed through the filter FBG A, that reflects boththe optical carrier ωc,i+1 and the corresponding sideband si+1,which therefore may be referred to as a double-notch filter. Therest of the WDM signal is fed into port 6 and it finally leavesfrom port 7. Both the optical carrier ωc,i+1 and the sidebandsi+1, that were reflected by FBG A, enters port 2 and exits fromport 3, where it encounters the second filter FBG B, as shown inFig. 66. The FBG B filter is a notch-filter that reflects 50% of the

optical carrier ωc,i+1. The remaining 50% of the optical carrierpower and 100% of the sideband power is passed through andemployed in DL photo-detection [117]. The reflected 50% ofthe optical carrier power re-enters port 3 and leaves from port4, where is it employed in UL modulation. As shown in Fig. 66,the UL signal enters port 5 and exits via port 6, where itencounters FBG A, that has been designed to reflect both theoptical carrier at ωc,i+1 and the corresponding sideband. Hence,the UL signal is reflected by FBG A, enters port 6 of the OC andexits via port 7. Thus, the optical signal exiting port 7 consistsof the UL signal at ωc,i+1 and si+1, instead of the DL signalat that frequency [117]. It also contains the other carriers andsidebands of the input WI-DWDM signal.

The advantage of the design of Fig. 63 is its simplicity [115].However, it wastes 50% of the modulated sideband power. Thisdisadvantage is circumvented by the design of Fig. 64 [115],which can also be extended to the SCM transmission regime ofFig. 20, regardless of the number of RF signals involved, sincethe FBG filter of Fig. 64 only has to reflect the optical carrierfrom the spectrum of Fig. 20. However, this design requires anOC, which is more complex and hence more expensive than anoptical coupler. The design of Fig. 66 can be simultaneouslyused as an OADM and a wavelength re-use module in a bus- orring-based ROF network. Again, the disadvantage of this designis its complexity. A common design challenge in the architec-tures of Figs. 64 and 66 is the fact that the imperfect isolationof the OC ports and variations in the transmittance-reflectancecharacteristics of the temperature-sensitive FBG filters result incross-talk between the DL and UL signals [117]. The designof Fig. 65 employs an optical interleaver, which is not assensitive to temperature variations as a FBG filter [38]. How-ever, in the design of Fig. 65, the weaker second-order opticalsidebands at ωc ±ωRF have to be filtered out at the BS, other-wise they will be passed along with the separated optical carrier.This occurs because the transfer function of the interleaverconsists of alternating passbands and stopbands.

Let us now continue our discussions on the family oftechniques invented for wavelength-reuse by considering thearrangements that implement Optical Re-Modulation (ORM),which, unlike the above OCRR techniques, do not entail theseparation of the optical carrier from its sidebands using high-selectivity optical filtering, but involve other constraints.

2) Optical Re-Modulation (ORM): In ORM, the DL opticalsignal is remodulated by the UL data at the RAP, withoutrelying on filtering for recovering the optical carrier. This canbe achieved in a number of ways, as described below.

Fig. 67 [116] illustrates an architecture that employs theEAM of 10(a) to achieve ORM. In Fig. 67, the downlinksignal at fd = ωd

2π Hz and a local oscillator tone at fl =ωl2π Hz

are combined with the aid of the SCM technique of Fig. 20and the resultant signal is then used for modulating the DLoptical carrier generated by the LD at fc =

ωc2π Hz. The ODSB

modulated signal is then optically filtered for generating anOSSB signal, whose spectrum is also shown in Fig. 67 [116].After transmission of this signal from the BS to the RAP overthe fiber, as shown in Fig. 67, half of the signal power is photo-detected and bandpass filtered in the upper branch of the Fig. 67for removing the local oscillator tone and then transmitted to


Fig. 66. Optical Carrier Recovery and Re-use (OCRR) in the RAP of a ROF system employing WI-WDM signals (FBG - Fiber Bragg Grating; OC - OpticalCirculator; PD - Photo-diode; BPF - Band Pass Filter).

Fig. 67. Optical Re-Modulation (ORM) using an Electro-Absorption Modulator (RF Tx - RF Transmitter; LD - Laser Diode; PD - Photo-diode; BPF - BandPass Filter).

Fig. 68. Optical Re-Modulation (ORM) with constraints on uplink/downlinkfrequency allocation (RF Tx - RF Transmitter; LD - Laser Diode; PD -Photo-diode; BPF - Band Pass Filter; MZM - Mach–Zehnder Modulator; Rx- Receiver).

the MS, On the other hand, the other half of the signal power inthe lower branch is additionally remodulated using an EAM bythe UL signal at fu =

ωu2π Hz. It is then amplified and optically

filtered for ensuring that the UL optical signal consist of a toneat (ωc −ωl), an UL sideband at (ωc −ωu) and a residual DLsideband at (ωc−ωd) [116]. The UL optical signal is then trans-mitted from the RAP to the BS over the fiber, where it is photo-detected and bandpass filtered for removing the DL downlinksignal. Finally, the downconverted uplink signal at ωu −ωl isdetected [116].

Another ORM architecture is shown in Fig. 68, where theoutput of a LD is modulated by the DL RF signal and it isthen transmitted from the BS to the RAP. Then, a part of theDL signal power is separated using an optical coupler andadditionally remodulated by the UL signal, using the MZMof the lower-branch in Fig. 68. The remodulated UL signal istransmitted from the RAP to the BS. Meanwhile, the signal inthe upper-branch of Fig. 68 is photo-detected to generate the DLwireless signal. If the width of the UL and DL RF bands is less

Fig. 69. Optical Re-Modulation (ORM) using phase modulation and intensitymodulation for downlink and uplink optical communication, respectively (RFTx - RF Transmitter; LD - Laser Diode; PD - Photo-diode; BPF - Band PassFilter; Rx - Receiver).

than an octave8 and the frequency separation between the bandsis more than the width of the bands, then the further modulatedUL signal remains unaffected by the inter-modulation productsbetween the DL and UL signals [109]. Thus, the UL signal canbe filtered out from the signal that is photo-detected at the BS.

Furthermore, it is possible to phase-modulate the opticalcarrier during its DL transmission and then intensity modulatethe same optical carrier for its UL transmission as detailedbelow [6], [123], [124]. As shown in Fig. 69 [124], the outputof the LD is phase-modulated by the RF signal and transmittedfrom the BS to the RAP over the fiber, where it was split intotwo branches by a coupler. The DL signal in the upper-branchof Fig. 69 is balanced photo-detected after performing a phaseto intensity conversion using a delay interferometer (DI) such asthe MZI of Fig. 19, as discussed in Section III-B [124]. An MZIassists in employing balanced photo-detection, as discussedin Section V. It is worth mentioning that instead of perform-ing phase-to-intensity conversion followed by direct balanced

8A frequency band from f1 Hz to f2 Hz is equal to an octave if f2 = 2 f1


Fig. 70. Optical Re-Modulation (ORM) using a Reflective SemiconductorOptical Amplifier (RSOA) (RF Tx - RF Transmitter; LD - Laser Diode; PD -Photo-diode; BPF - Band Pass Filter; LPF - Low Pass Filter).

photo-detection, the coherent photo-detection of the phase-modulated signal may also be employed, as discussed Sec-tion II-C. The lower branch of the coupler of Fig. 69 is intensity-modulated by the UL RF signal using an external modulator.Afterwards, the UL optical signal is transmitted from the RAPto the BS over the fiber, where it is photo-detected to generatethe UL RF signal, which is subsequently bandpass filteredand fed to the RF receiver.

Additionally, the architecture of Fig. 70 can also be used,where the output of a laser source is modulated by the DLRF signal using a MZM. Subsequently, the DL optical signalenters port 1 and exits from port 2 of the OC. This signalis then transmitted from the BS to the RAP over the fiber,where the signal is split into two branches. The upper branchis photo-detected to generate the DL wireless signal [121],while the lower branch is fed to a gain-saturated ReflectiveSemiconductor Optical Amplifier (RSOA) in order to suppressthe DL optical sidebands. As shown in Fig. 70, the opticalcarrier is simultaneously modulated by the UL BB data that isgenerated by down-converting the fuplink =

ωuplink2π Hz wireless

signal from the MS, which is received at the RAP [121]. Hence,the UL transmission requires the oscillator operating at ωuplink

in the RAP. The UL optical signal exits the RSOA and then itis transmitted from the RAP to the BS through the optical fiber,where it enters port 2 and exits from port 3 of the OC. As shownin Fig. 70, the UL signal is then photo-detected and the resultantBB UL signal is filtered and detected [121].

Let us now briefly contrast the solutions in Figs. 67–70. Thedesign of Fig. 67 requires the UL signal to be amplified usingErbium Doped Fiber Amplifiers (EDFAs) which add ASE op-tical noise, as discussed in Section II-D. On the other hand, thedesign of Fig. 68 is simple, but it imposes certain conditions onthe wireless frequency assignment. The design of Fig. 69 doesnot impose such restrictions, but requires either the phase-to-intensity conversion discussed in prior to the DL direct photo-detection or it has to employ the DL coherent photo-detection.Also, fiber dispersion converts the DL phase-modulation tointensity modulation, thereby corrupting the intensity modula-tion based UL transmission of Fig. 69. Moreover, any residualintensity-modulation of the DL transmission would degrade theUL transmission of the system in portrayed Fig. 69. Finally, thedesign of Fig. 70, unlike the designs of Figs. 67–69, employs anoscillator for the down-conversion of the uplink wireless signalreceived from the MS, which may become a disadvantage inthe case of high-RF frequencies, like millimeter-wave signals.However, the design of Fig. 70 has the advantage that the uplinkoptical modulator of the RAP performs the additional func-

Fig. 71. EAM absorption characteristics. (a) EAM absorption at different op-tical wavelengths; (b) EAM absorption or transmission vs. bias voltage at λ1.

tion of an amplifier as well, thereby improving the attainableBER [121].

All the wavelength re-use techniques described so far,namely the OCRR and ORM techniques, employ a separateoptical UL modulator and DL photo-detector in the RAP.However, in the next section we discuss wavelength re-usetechniques that employ a single modulator-cum-photo-detector,thereby reducing the RAP complexity.

3) Employing a Single Device Modulator-Cum-Photo-Detector: The wavelength re-use techniques of Section VI-B1and VI-B2 need a photo-detector for detecting the DL opti-cal signal and a separate modulator for transmitting the ULoptical signal. However, it is possible to employ an EAM asa modulator-cum-photo-detector.9 Both the operating opticalwavelength and the bias-voltage has an influence on the trans-mission characteristics of the EAM. Fig. 71(a) illustrates thedependence of the EAM absorption on the operating wave-length [11]. As shown in Fig. 71(a), an optical input signal ata wavelength of λ2 would be largely absorbed, i.e., it is largelyphoto-detected, regardless of the bias voltage, while an opticalinput signal at a wavelength of λ1 would experience a bias-voltage-dependent absorption, i.e., a bias-voltage-dependentphoto-detection, or transmittance level, as shown in Fig. 71(b)and discussed in Section II-A. As shown in Fig. 71(b), a lowlevel of transmittance corresponds to a high level of absorption,i.e., a high level of photo-detection of the optical input signal bythe EAM [112]. On the other hand, a high-level of transmittancewould enable the modulation of the input optical signal bythe electronic signal applied to the EAM. This property hasbeen exploited for designing the following EAM-based opticaltransceivers.

Single-wavelength half-duplex operation can be achievedusing the EAM. When light at a wavelength λ1 of Fig. 71(a)enters an EAM, achieving a high RF modulation efficiencyrequires the EAM to be biased at a point, where the slope ofthe transmission characteristics of Fig. 71(b) is maximum. Thisis true for moderate values of the reverse bias voltage, as seenfrom Fig. 71(b) [110]. On the other hand, it can also be seenfrom Fig. 71(b) that the optical input signal is largely absorbed,i.e., it is photo-detected, when a high reverse bias voltage isemployed [112]. Thus, the EAM can be operated both as amodulator and a photo-detector by appropriately switching thebias voltage applied to it [110], as shown in the architecture ofFig. 72, where the bias voltage applied to the EAM in the RAPis switched based on whether photo-detection of the DL optical

9A single device that is capable of photo-detecting the DL optical signal aswell as modulating it with the electronic UL signal.


Fig. 72. Half-duplex operation of Electro-Absorption Modulator (EAM) asmodulator-cum-photo-detector (LD - Laser Diode; RF Tx - RF Transmitter;PD - Photo-diode; BPF - Band Pass Filter; RF Rx - RF Receiver).

Fig. 73. Full-duplex operation of Electro-Absorption Modulator (EAM) asmodulator-cum-photo-detector (LD - Laser Diode; RF Tx - RF Transmitter;PD - Photo-diode; BPF - Band Pass Filter; RF Rx - RF Receiver).

signal from the BS or transmission of the UL optical signal tothe BS is taking place. The major drawback of this technique ishowever that the bi-directional communication is half-duplex,because simultaneous photo-detection and modulation is notpossible. Additionally, the need for employing bias controlcircuits is another complexity-related drawback.

In contrast to Fig. 72 relying on a half-duplex operation,single-wavelength full-duplex operation can be used, as in thearchitecture of Fig. 73, where the operating wavelength is cho-sen to be λ1 of Fig. 71(a) and the bias voltage is not switched. Inthe architecture of Fig. 73, the EAM bias voltage of Fig. 71(b)is adjusted to a specific value that would strike the desiredcompromise between the UL modulation efficiency and the DLphoto-detection efficiency, along with the UL modulation non-linearity being at an acceptable level [111]. Hence, the EAMin the RAP is not operating in the ideal photo-detection regionof Fig. 71(b) and it photo-detects the DL optical signal, whilealso modulating the UL wireless signal onto the optical carrier.It is possible to operate the EAM in a bias-free scenario, i.e.,when a bias of zero volts is applied, but it can be seen fromFig. 71(b) that choosing a bias voltage of zero volts would resultin the EAM becoming less efficient as a photo-detector than asa modulator. Hence different antenna gain values would haveto be applied to the UL and DL signal at the mobile device toensure a similar performance [113].

Finally in contrast to Fig. 72 and Fig. 73, dual-wavelengthfull-duplex operation may also be employed. The need to strikea compromise between the modulation efficiency and the photo-detection efficiency can be circumvented by adopting a dual-wavelength operation. In the architecture of Fig. 74 [114], theoutput of the laser LD2, operating at λ2 of Fig. 71(a), is inten-

Fig. 74. Full-duplex operation of Electro-Absorption Modulator (EAM) asmodulator-cum-photo-detector (LD - Laser Diode; RF Tx - RF Transmitter;PD - Photo-diode; BPF - Band Pass Filter; RF Rx - RF Receiver).

sity modulated by a RF signal at fd = ωd2π Hz. The modulated

optical signal is then combined with the unmodulated output ofthe laser LD1 operating at λ1 of Fig. 71(a). As shown in Fig. 74,this combined signal is transmitted from the BS to the RAP overthe fiber, where the optical signal at λd is photo-detected by theEAM, while the optical signal at λ1 is modulated by the EAMusing the UL RF signal at fu = ωu

2π Hz [114]. The UL opticalsignal is transmitted from the RAP to the BS over the fiber andthen it is photo-detected, filtered and detected.

Before we move on to the next section, we provide, inTable V, a list of the seminal papers of the family of wavelengthreuse techniques discussed in Section VI-B.

Let us now consider the family of techniques proposed forreducing the number of lasers that are employed in the BS,unlike the above wavelength re-use techniques that aim to makethe RAP laser-free.

C. Reducing the Number of Lasers

While wavelength-reuse eliminates the need to have a lasersource in the RAP, the BS still has to have several laser sourcesfor supporting multiple RAPs by using the WDM signal ofFig. 22(a). The need to have several laser sources in a networkemploying WDM can be totally or partially relaxed in thefollowing ways:

1) Using a Multi-Mode Optical Source: A single-modelaser produces a tone at a single frequency, while a multi-modelaser produces tones at multiple optical frequencies [127]. Asshown in Fig. 75(a) [125], the various Δω spaced modes of amulti-mode laser are separated using a DEMUX. Then, pairsof modes having a separation of NΔω are employed in opticalmodulation using a BB/IF signal at fIF = ωIF

2π Hz, where eachpair of modes constitutes a channel and N is a positive integer.Subsequently, the modulated channels are multiplexed to gener-ate the WDM signal of Fig. 75(a), which is transmitted througha ROF optical network to the RAP, where the heterodynedetection of the modes in each channel generates the DL RFsignal at ωd =(NΔω+ωIF). It has been shown in [125] that thistechnique can be used for generating up to 360 such channels,which can be used for supporting duplex communication with180 RAPs. By additionally employing the SCM of Fig. 20 it ispossible to support up to 10080 RF channels associated with abandwidth of 250 MHz belonging to the unlicensed 60 GHzband [126]. The availability of a large number of channels


TABLE VRADIO OVER FIBER (ROF) COST REDUCTION TECHNIQUES (PART 2)

Fig. 75. Techniques to reduce the number of lasers (LD - Laser Diode; WDM- Wavelength Division Multiplexing; PD - Photo-diode; BPF - Band Pass Filter;IL - Interleaver; MMW - Millimeter Wave; BB Tx - Baseband Transmitter). (a) WDM using a multimode optical source; (b) Employing phase modulationharmonics to generate WDM signal.

can be exploited in the design of a dynamically reconfigurableWDM network carrying 60 GHz ROF signals [12].

2) Employing Modulation Harmonics: The harmonics thatare imposed by the modulation of an RF tone on the laser’soutput can be beneficially harnessed as independent opticalcarriers [3], [106]. These modulation harmonics can supportmulti-band transmission to a single RAP. Fig. 75(b) illustratesan architecture in which the upper harmonics that are gener-ated during phase modulation can be used for supporting thetransmission of WiFi, WiMAX and 60 GHz signals to the RAP[106]. At the BS, the optical carrier is phase-modulated usinga fRF = 30 GHz = ωRF

2π electronic tone at a phase modulationdepth, which would result in first-order harmonic at (ωc±ωRF)and in significant second-order sidebands at (ωc ± 2ωRF), inaddition to the presence of the optical carrier at ωc, as shownin Fig. 75(b) [106]. The optical carrier is removed by the firstinterleaver (IL1), while the second interleaver (IL2) separatesthe four sidebands, where there are two sidebands of the first-order and two of the second-order. The pair of second-ordersidebands at (ωc + 2ωRF) and (ωc − 2ωRF) are considered as

Fig. 76. Exploiting Mach–Zehnder Modulator non-linearity to generate aWavelength Division Multiplexing (WDM) signal (LD - Laser Diode; PD -Photo-diode; BPF - Band Pass Filter; IL - Interleaver; MMW - MillimeterWave; BB Tx - Baseband Transmitter).

independent optical carriers and are modulated with the BBWiFi and BB WiMAX signals, respectively. The pair of 2 fRF =60 GHz spaced first-order sidebands at (ωc +ωRF) and (ωc −ωRF) are then combined and modulated with the BB data of the


TABLE VIRADIO OVER FIBER (ROF) COST REDUCTION TECHNIQUES (PART 3)

Fig. 77. System architecture and performance results. (a) Hybrid Analog/Digitized ROF; (b) BER performance of the AROF and DROF links.

desired 60 GHz signal. The four sidebands are then combinedto generate the spectrum seen in Fig. 75(b) and this multiplexedsignal is transmitted from the BS to the RAP over the fiber. Inthe RAP, the first interleaver IL3 of Fig. 75(b) separates thefirst-order sidebands from the second-order sidebands [106].The two first-order sidebands are heterodyne photo-detected togenerate a 60 GHz millimeter-wave signal, while the secondinterleaver IL4 of Fig. 75(b) is used for separating the twosecond-order sidebands, followed by their individual photo-detection and upconversion to generate the 2.4 GHz WiFi andthe 5.8 GHz WiMAX signals [106].

On the other hand, instead of supporting multiband transmis-sion to a single RAP, as in Fig. 75(b), it is possible to employthe modulation harmonics as independent optical carriers thatsupport multiple RAPs [3], [5], which could support DAS-based wireless MIMO communication [5]. As shown in Fig. 76,it is possible to generate multiple optical carriers by modulatingthe output of a LD, operating at an optical output of fc =

ωc2π Hz,

using a sinusoidal modulating signal at fLO = ωLO2π Hz and a

high modulation depth [3], [5]. This is achieved using the OCSarchitecture in Section III-A3, which employs the dual-driveMZM of Fig. 8(b) that is biased at Vbias = Vπ and operatingin the push-pull mode. As observed from the spectrum seenin Fig. 76, using a high modulation index results in the firstharmonic at (ωc ± ωLO), third harmonic at (ωc ± 3ωLO) andfifth harmonic at (ωc ± 5ωLO) having a significant magnitude.These can then be used as six independent optical carriershaving an inter-carrier spacing of 2ωLO [3], [5], for supportingmultiple RAPs, one of which is shown in Fig. 76.

Before concluding our discussions on the cost reductiontechniques, we provide, in Table VI, a list of the seminal papers

of the family of ROF techniques conceived for reducing thenumber of lasers, as discussed in Section VI-A.

VII. DESIGN EXAMPLE

A hybrid architecture that uses the state-of-the-art DROFarchitecture of Fig. 56 with the conventional ROF of Fig. 3 canbe designed, where the conventional ROF is referred to as Ana-logue ROF (AROF) in this section to distinguish it from DROF.In addition to the advantages of the DROF solution discussedin Section V-L, there is also the further advantage that multipleusers may be optically Code Division Multiplexed (CDM) [3].The MSs have different clock and carrier frequencies as wellas different distances from the RAP. Hence, their UL signalsarrive asynchronously at the RAPs. Additionally, imperfectpower control as well as the inevitable multi-user interferenceresult in the asynchronous wireless UL having a lower SNRthan the synchronous DL [3]. Hence the more robust, but morecostly DROF link of Fig. 56 can be employed for UL opticaltransmission, while the low-cost conventional AROF link ofFig. 3 can be employed for the DL, as shown in the architectureof Fig. 77(a) [3]. As a further benefit, wavelength re-use relyingon the ORM technique of Fig. 68 may be invoked, becausethe UL Baseband (BB) signal of DROF may be transmitted bythe same optical carrier as that used for transmitting the DLRF signal of AROF. This also increases the optical spectralefficiency. The lower optical transmit power requirement ofthe DROF mode also supports the re-use of the DL opticalcarrier power. Moreover, the ring topology of the architecture inFig. 77(a) minimizes the cost of fiber laying [3]. The technique


of exploiting the modulation harmonics in an OCS signal, asdiscussed in Section VI-C2, is also employed in the architectureof Fig. 77(a) [3] for supporting multiple RAPs. As discussedin Section I, wireless MIMO transmission, that relies on theuse of DL transmit-antenna diversity and UL receive-antennadiversity, improves the wireless link performance of the ROFsystem in Fig. 77(a). The discussion of the architecture seen inFig. 77(a) is as follows:

a) Optical carrier generator (OCG): The spectrum of thenon-linearly modulated OCS signal is as shown in Fig. 76.This technique is used for generating the optical carriers in thesystem of Fig. 77(a), where optical sidebands B to E of thespectrum in Fig. 76 support four RAPs.

b) BS transmitter: The DL RF signals transmitted to eachRAP at frequencies fdwn1 = 2.47 GHz and fdwn2 = 5.8 GHz areSub-Carrier Multiplexed (SCM) and are then used for intensitymodulating the optical sidebands B to E in the output of OCG.Each of these sidebands now transmits a ODSB DL opticalsignal to the 4 RAPs [3]. Transmit diversity is achieved in thegeneration of the RF signals with the aid of Space Time BlockCoding (STBC) [15]. A single user per RF carrier is assumed.

c) ROF ring: The DL ROF signals are transmittedthrough the fiber, where the 4 major sidebands in the OCGoutput are used for supporting 4 RAPs in the ring. Duplex op-tical transmission is used by each RAP at a single OCG outputsideband via an OADM and ORM [3]. The block diagram of aRAP is shown in Fig. 77(a). The optical signal dropped at RAP-i is fed to a 3 dB splitter. One of the splitter outputs is photo-detected and amplified for recovering the DL RF signal, whilethe other output is modulated by the UL BB DROF signal. TheBB data modulates the reused DL optical carrier and corruptsthe two DL sidebands, which is not a problem since the DLsignal has already been detected [3]. The users’ wireless ULsignals at fup1 Hz and fup2 Hz are received at all RAPs and thencombined by a Maximal Ratio Combining (MRC) [128] at theBS [3]. In the UL optical transmitter of the RAP, the digitizationof the received RF signals is performed using the concept ofbandpass sampling. The pair of digitized RF signals are thenCode Division Multiplexed [128] prior to optical modulationand transmission over the optical UL [3].

d) BS receiver: The UL optical signal that is receivedat the RAP is fed into a × 4 AWG, each of whose outputsare photo-detected and amplified using an Automatic GainControl (AGC) amplifier. The amplified signals are Low PassFiltered (LPF) to remove the corrupted DL RF signals and theirintermodulation products. The filtered BB CDM DROF signalsare code division de-multiplexed (CDD) to separate the twodigitized UL RF signals received from each RAP, which arethen fed into DACs for recovering the RF signals [3]. Sincethe UL wireless signals arriving from the two mobile usersare received at all the RAPs, four copies of the two UL RFsignals are obtained [3]. These are then down-converted andfed to MRC detection blocks [128]. The system of Fig. 77(a)was simulated using the parameter values of Table VII. Theperformance results shown in Fig. 77(b) indicate a better per-formance for the DROF UL than for the AROF DL. It can alsobe seen that the DROF UL performance, unlike the AROF DLperformance, of all the four RAPs is similar and hence DROF is

TABLE VIIPARAMETER VALUES USED IN THE ROF ARCHITECTURE

Fig. 78. Uplink performance using MRC & BPSK modulation and DLperformance using STBC & BPSK modulation.

capable of supporting longer fiber lengths than AROF. Fig. 78shows the performance of the overall DL and UL, including thewireless transmission, where an optical power of 15 dBm wasemployed. It can be seen that the overall system performance isdominated by the wireless performance, because near error-freetransmission was achieved in the fiber link.

VIII. DESIGN GUIDELINES

The sensitivity of a ROF link is defined as the receivedoptical power at which a BER of 10−9 is achieved [24]. On theother hand, the linearity of the ROF link, among others, maybe expressed in terms of the suppression of the harmonics andintermodulation products in the photo-detected signal and alsoin terms of the BER degradation imposed by the fiber’s non-linearity. Also, the user-support capacity of the ROF systemmay be defined as the number of RAPs or as the number ofMSs that are supported by it. The first step in designing a ROFsystem is to set target values for the above performance metrics,based on the requirements of the wireless network, for which


TABLE VIIIFREQUENTLY USED ABBREVIATIONS

the ROF system forms the backhaul. Next, we go about design-ing a basic ROF link, based on the techniques listed in Fig. 5.

The simplest ROF link is one that employs direct modulationof the optical carrier’s intensity discussed in Section II-Afor generating the ODSB signal discussed in Section III-A1,which is then detected by direct photo-detection, as highlighted

in Section II-C, with multiplexing being implemented usingthe SCM technique discussed in Section IV-A. However, thedesigner would be constrained to use external modulation, ifthe RF carrier is beyond the modulation bandwidth of thelaser. For a high RF carrier, like a millimeter-wave carrier, thechromatic dispersion-induced power penalty would require thedesigner to employ the OSSB or OCS modulations discussedin Section III-A2 and III-A3, respectively. In such a scenario,the designer may use an optical filter after the external mod-ulator or alternatively employ a dual-drive MZM. However,if the sensitivity, i.e., BER requirements are too stringent foran intensity-modulated direct-detected link, then the designermay employ the optical angle modulation technique discussedin Section III-B along with the coherent photo-detection ofSection II-C. Coherent photo-detection requires an additionallaser in the receiver, which may be avoided by using the opticaldiscriminator of Section III-B. This, however, would be atthe expense of a performance degradation. Additionally, thedesigner may opt for the more complex WDM, instead of theSCM, if a large number of RAPs (or mobile users) have to besupported.

Having designed a basic ROF link, the next priority ofthe designer should be to incorporate maximum possible costsaving techniques listed in Fig. 23, while ensuring that theperformance targets are still met. In urban areas, where a signif-icant extent of fiber has already been laid for digital fiber-basedservices, the ROF network designer may employ the family ofROF techniques for integrating ROF communication with theexisting FTTH networks, as discussed in Section VI-A. Thismust, however, ensure that the quality of service of the fiber-based digital services is not compromised. The designer mayalso employ the family of techniques designed for wavelengthre-use, as discussed in Section VI-B. However, the allocation ofa portion of the DL optical power for UL communication woulddegrade the DL performance and hence it must be ensured thatthe sensitivity targets are met. The wavelength re-use techniqueof OCRR may require high-selectivity optical filtering. Thisrequirement is relaxed in the wavelength re-use technique ofORM, which however imposes the restrictive conditions dis-cussed in Section VI-B2. When employing a modulator-cum-photo-detector, conceiving the half-duplex regime of Fig. 72is straightforward. By contrast, the full-duplex communicationscenario of Fig. 73 or Fig. 74 either involves a DL versus ULperformance trade-off or employs a pair of optical wavelengths,respectively. Finally, cost saving may be achieved by reducingthe number of lasers, as discussed in Section VI-C.

When the basic ROF link does not meet one or more of theperformance targets set by the wireless network, this may bealleviated by employing the performance improvement tech-niques listed in Fig. 23. The employment of a cost-savingtechnique might provide a significant economic advantage tothe designer, but also result in the performance targets not beingmet. In this scenario too, the designer may employ the cost-saving technique in conjunction with a performance improve-ment technique, provided the total system’s cost remains low.If the target sensitivity of the ROF link is not met, then thedesigner may employ a technique from the family of tech-niques for increasing depth of optical modulation, for noise-reduction or for overcoming chromatic dispersion, which were


discussed in Section V-B–V-F and in Section V-K. Among thesetechniques, the ones employing carrier suppression, balancedphoto-detection and dispersion-tolerant modulation schemesare easier to implement. On the other hand, if the target linearityof the ROF link is not met, then the designer may increase thelinearity of optical modulation, as discussed in Section V-E–V-I or aim for overcoming the fiber non-linearity, as discussed inSection V-J. Employing linearized optical modulators reducesthe strength of the required RF signal along with the suppres-sion of the non-linear products, thereby degrading the link’ssensitivity. Both OIL and optical feed-forward linearizationprovides noise-reduction in addition to improving the linearityof optical modulation, while the technique of electronic pre-and post-compensation is the simplest to implement. Finally, ifthe target number of users cannot be supported by the ROF link,then the designer may opt for improving the optical spectralefficiency, as discussed in Section V-A.

Moreover, when it is not possible to meet the performancetargets, the analog optical link can be replaced by the DigitizedRadio Over Fiber architecture listed in Fig. 23. Alternatively,the sampled RF over fiber technique discussed in Section V-Lmay be employed. As discussed in Section V-L, this architec-ture employs the more robust digital optical link, but has a morecomplex RAP.

IX. CONCLUSION

In this paper, we introduced the concept of Radio Over Fiber(ROF) transmissions in Section I and then proceeded to discussthe various techniques and design trade-offs of the ROF linkof Fig. 3 in Sections II–IV, including the optical transmitter,the optical fiber, the optical receiver and the optical network.These trade-offs were summarized in the stylized illustrationof Fig. 5. We then continued our discourse in Section V bycharacterizing the main architectures that may be used forimproving the ROF communications system’s performance,which were categorized in Fig. 23. The wavelength-interleavedmultiplexing technique discussed in Section V-A increases theoptical spectral efficiency and hence the throughput of theROF link, while the carrier suppression technique discussed inSection V-B improves the receiver sensitivity by increasing thedepth of the optical modulation. On the other hand, the noiseimposed by the optical link can be reduced by the techniquesdiscussed in Section V-C and V-D, namely by low-biasing theoptical modulator and by employing balanced photo-detection.These techniques have been illustrated in Figs. 33 and 34,respectively. The techniques of Optical Injection Locking andoptical feed-forward linearization discussed in Section V-E andV-F, respectively, not only help in noise-reduction but alsoin increasing the linearity of the optical modulation. We thenwent on to discuss other techniques conceived for increasingthe linearity of optical modulation, including the employmentof linearized modulators, of optical filters as well as of elec-tronic pre-distortion or post-compensation filters, which wereportrayed in Figs. 39–45. We then discussed the multiple fibertransmission scheme of Fig. 46 designed for overcoming thenon-linear effects of the optical fiber, followed by a discussionof the various schemes of Figs. 47–54 aimed at overcomingchromatic dispersion. These arrangements included the use of

dispersion tolerant modulation schemes, fiber gratings, chirpedoptical signals, self-phase modulation and finally the use ofoptical phase conjugation. We then concluded our discussionsby presenting the Digitized ROF architecture of Fig. 56, whichaims for incorporating the benefits of digital optical communi-cation in the conventional ROF link of Fig. 3, at the cost of amore complex Radio Access Point (RAP) design.

Naturally, we have to strike a cost versus performance trade-off in the various ROF techniques. Since the performanceenhancement techniques of Section V increase the cost of theROF link, in Section VI we additionally detail the main costreduction techniques that may be used in the ROF link ofFig. 3. The integration of the ROF network with the Fiber ToThe Home (FTTH) network, as seen in Figs. 57–62, wouldreduce the installation costs. However, in such a scenario, thedesigner has to strike a trade-off between the performance ofthe baseband optical transmission of the FTTH network and theattainable performance of the ROF transmission. Cost reduc-tions can also be achieved by making the RAP laser-free by re-using the DL optical signal for UL communication, as seen inthe architectures of Figs. 63–74. These wavelength re-use tech-niques rely on Optical Carrier Recovery and Reuse (OCRR),Optical Re-modulation (ORM) or on the use of a modulator-cum-photo-detector, which were discussed in Section VI-B1–VI-B3, respectively. The OCRR techniques of Figs. 63–66 relyon the separation of the DL optical carrier from the DL mod-ulation sidebands, while the ORM techniques of Figs. 67–70re-modulate the optical signal carrying the DL wireless signalwith the uplink wireless signal as well. On the other hand, amodulator-cum-photo-detector detects the DL optical signal be-sides simultaneously performing UL optical modulation in theRAP. Finally, another technique of reducing the ROF system’scost in a multi-RAP and/or multi-user scenario is to decreasethe number of lasers employed in the BS, which can be achievedby using multi-mode laser sources, as shown in Fig. 75(a)or by employing the modulation harmonics as shown inFig. 75(b) and Fig. 76. We then present the design guidelines inSection VIII.

We conclude by stating that there are numerous open areasfor further research. Some of these areas include, although arenot limited to, integrating various millimeter-wave techniqueswith a ROF backhaul [129], [130], the use of mode-divisionmultiplexing in multimode optical fibers for ROF [131], em-ploying phased array antennas in a ROF link [132] and the useof plastic optical fiber for indoor personal area networks [133].Multimode optical fibers, especially plastic optical fibers, areless expensive than single mode fibers. Having a large coresize simplifies their connections and they can be used with low-cost Light Emitting Diodes (LEDs) and Vertical-Cavity SurfaceEmitting Lasers (VCSELs). However, the multimode nature oftheir propagation inevitably limits their employable bandwidthand lengths. Significant research efforts are being invested inovercoming this drawback through efficient signal processing.

REFERENCES

[1] D. Wake, A. Nkansah, and N. Gomes, “Radio over fiber link designfor next generation wireless systems,” IEEE/OSA J. Lightw. Technol.,vol. 28, no. 16, pp. 2456–2464, Aug. 2010.


[2] X. Fernando, “Radio over fiber-an optical technique for wireless access,”presented at the IEEE Communications Society Presentation, Toronto,ON, Canada, Oct., 2009.

[3] V. Thomas, S. Ghafoor, M. El-Hajjar, and L. Hanzo, “A full-duplexdiversity-assisted hybrid analogue/digitized radio over fibre for opti-cal/wireless integration,” IEEE Commun. Lett., vol. 17, no. 2, pp. 1–4,Feb. 2013.

[4] V. A. Thomas, S. Ghafoor, M. El-Hajjar, and L. Hanzo, “Baseband radioover fiber aided millimeter-wave distributed antenna for optical/wirelessintegration,” IEEE Commun. Lett., vol. 17, no. 5, pp. 1012–1015,May 2013.

[5] S. Ghafoor and L. Hanzo, “Sub-carrier-multiplexed duplex 64-qamradio-over-fiber transmission for distributed antennas,” IEEE Commun.Lett., vol. 15, no. 12, pp. 1368–1371, Dec. 2011.

[6] S. Ghafoor, V. A. Thomas, and L. Hanzo, “Duplex digitized transmissionof 64-QAM signals over a single fiber using a single pulsed laser source,”IEEE Commun. Lett., vol. 16, no. 8, pp. 1312–1315, Aug. 2012.

[7] W. Stephens, T. Joseph, and B. Chen, “Analog microwave fiber opticcommunications links,” in Proc. IEEE MTT-S Int. Microw. Symp. Dig.,Jun. 1984, pp. 533–534.

[8] N. J. Gomes, P. P. Monteiro, and A. Gameiro, Next Generation WirelessCommunications Using Radio Over Fiber. Hoboken, NJ, USA: Wiley,2012.

[9] X. N. Fernando, Radio Over Fiber for Wireless Communications: FromFundamentals to Advanced Topics. Hoboken, NJ, USA: Wiley, 2014.

[10] J. Johny, Radio Over Fiber—A New Communication Era: Future Broad-band Wireless Access. New York, NY, USA: Academic, 2012.

[11] K. Kitayama et al., “An approach to single optical component an-tenna base stations for broad-band millimeter-wave fiber-radio accesssystems,” IEEE Trans. Microw. Theory Tech., vol. 48, no. 12, pp. 2588–2595, Dec. 2000.

[12] J. J. Vegas Olmos, T. Kuri, and K. Kitayama, “Dynamic reconfigurablewdm 60-ghz millimeter-waveband radio-over-fiber access network:Architectural considerations and experiment,” IEEE/OSA J. Lightw.Technol., vol. 25, pp. 3374–3380, Nov. 2007.

[13] L. Hanzo, M. El-Hajjar, and O. Alamri, “Near-capacity wirelesstransceivers and cooperative communications in the mimo era: Evolutionof standards, waveform design, future perspectives,” Proc. IEEE, vol. 99,no. 8, pp. 1343–1385, Aug. 2011.

[14] L. Hanzo, O. Alamri, M. El-Hajjar, and N. Wu, Near-Capacity Multi-Functional MIMO Systems: Sphere-Packing, Iterative Detection andCooperation, 1st ed. Hoboken, NJ, USA: Wiley/IEEE Press, 2009.

[15] L. Hanzo, Y. Akhtman, L. Wang, and M. Jiang, MIMO-OFDM for LTE,WiFi and WiMax: Coherent Versus Non-Coherent and Co-OperativeTurbo Tranceivers, 1st ed. Hoboken, NJ, USA: Wiley, 2011.

[16] H. Al-Raweshidy and S. Komaki, Radio Over Fiber Technologies forMobile Communication Networks, 1st ed. Norwood, MA, USA: ArtechHouse, 2002.

[17] J.-M. Kelif and M. Coupechoux, “Impact of topology and shadowingon the outage probability of cellular networks,” in Proc. IEEE ICC,Jun. 2009, pp. 1–6.

[18] T. Naveh, “Mobile backhaul: Fiber vs. microwave,” Ceragon, Tel Aviv,Israel, White Paper, pp. 1–11, Oct. 2009, vol. 1.

[19] E. Dahlman, S. Parkvall, and J. Sköld, 4g: Lte/lte-Advanced for MobileBroadband. New York, NY, USA: Academic, 2011.

[20] T. Kanesan, W. P. Ng, Z. Ghassemlooy, and C. Lu, “Experimental fullduplex simultaneous transmission of lte over a dwdm directly modulatedrof system,” IEEE/OSA Journal of Optical Communications and Net-working, vol. 6, no. 1, pp. 8–17, Jan. 2014.

[21] T. Rappaport et al., “Millimeter wave mobile communications for 5gcellular: It will work!” IEEE Access, vol. 1, pp. 335–349, May 2013.

[22] C. Cox, III, E. Ackerman, R. Helkey, and G. Betts, “Direct-detectionanalog optical links,” IEEE Trans. Microw. Theory Tech., vol. 45, no. 8,pp. 1375–1383, Aug. 1997.

[23] C. Carlsson, A. Larsson, and A. Alping, “RF transmission over mul-timode fibers using VCSELs-comparing standard and high-bandwidthmultimode fibers,” IEEE/OSA J. Lightw. Technol., vol. 22, no. 7,pp. 1694–1700, Jul. 2004.

[24] G. P. Agrawal, Fiber-Optic Communication Systems, 3rd ed. Hoboken,NJ, USA: Wiley, 2002.

[25] J. Ma et al., “Fiber dispersion influence on transmission of the opti-cal millimeter-waves generated using LN-MZM intensity modulation,”IEEE/OSA J. Lightw. Technol., vol. 25, no. 11, pp. 3244–3256, Nov. 2007.

[26] R. Kalman, J. Fan, and L. Kazovsky, “Dynamic range of coherent analogfiber-optic links,” IEEE J. Lightw. Technol., vol. 12, no. 7, pp. 1263–1277, Jul. 1994.

[27] G. Smith, D. Novak, and Z. Ahmed, “Overcoming chromatic-dispersioneffects in fiber-wireless systems incorporating external modulators,”IEEE Trans. Microw. Theory Tech., vol. 45, no. 8, pp. 1410–1415,Aug. 1997.

[28] G. L. Li and P. K. Yu, “Optical intensity modulators for digital and analogapplications,” IEEE/OSA J. Lightw. Technol., vol. 21, no. 9, pp. 2010–2030, Sep. 2003.

[29] K. Ikeda, T. Kuri, and K. Kitayama, “Simultaneous three-band mod-ulation and fiber-optic transmission of 2.5-Gb/s baseband, microwave,60-GHz-band signals on a single wavelength,” IEEE/OSA J. Lightw.Technol., vol. 21, no. 12, pp. 3194–3202, Dec. 2003.

[30] H. J. R. Dutton, Understanding Optical Communications, 1st ed.Triangle Park, NC, USA: IBM, 1998.

[31] X. Fernando, “Improved expression for intensity noise in multimediaover fiber networks,” in Proc. 1st Int. Conf. Ind. Inf. Syst., Aug. 2006,pp. 425–429.

[32] X. Fernando and A. Anpalagan, “On the design of optical fiber basedwireless access systems,” in Proc. IEEE ICC, Jun. 2004, vol. 6,pp. 3550–3555.

[33] W. Roh et al., “Millimeter-wave beamforming as an enabling technol-ogy for 5g cellular communications: Theoretical feasibility and pro-totype results,” IEEE Commun. Mag., vol. 52, no. 2, pp. 106–113,Feb. 2014.

[34] G. Smith and D. Novak, “Broad-band millimeter-wave (38 GHz)fiber-wireless transmission system using electrical and optical SSBmodulation to overcome dispersion effects,” IEEE Photon. Technol.Lett., vol. 10, no. 1, pp. 141–143, Jan. 1998.

[35] J. Corral, J. Marti, and J. Fuster, “General expressions for IM/DDdispersive analog optical links with external modulation or opticalup-conversion in a Mach–Zehnder electrooptical modulator,” IEEETrans. Microw. Theory Tech., vol. 49, no. 10, pp. 1968–1976, Oct. 2001.

[36] J. Park, W. Sorin, and K. Lau, “Elimination of the fibre chromatic dis-persion penalty on 1550 nm millimetre-wave optical transmission,” IETElectron. Lett., vol. 33, no. 6, pp. 512–513, Mar. 1997.

[37] R. Kashyap, Fiber Bragg Gratings, 1st ed. New York, NY, USA: Aca-demic, 1999.

[38] J. Yu et al., “DWDM optical millimeter-wave generation for radio-over-fiber using an optical phase modulator and an optical interleaver,” IEEEPhoton. Technol. Lett., vol. 18, no. 13, pp. 1418–1420, Jul. 2006.

[39] G. Smith, D. Novak, and Z. Ahmed, “Technique for optical SSB gen-eration to overcome dispersion penalties in fibre-radio systems,” IETElectron. Lett., vol. 33, no. 1, pp. 74–75, Jan. 1997.

[40] Z. Jia, J. Yu, and G. K. Chang, “A full-duplex radio-over-fiber systembased on optical carrier suppression and reuse,” IEEE Photon. Technol.Lett., vol. 18, no. 16, pp. 1726–1728, Aug. 2006.

[41] M. Weiss et al., “60-GHz photonic millimeter-wave link for short tomedium range wireless transmission up to 12.5 Gb/s,” IEEE/OSA J.Lightw. Technol., vol. 26, no. 15, pp. 2424–2429, Aug. 2008.

[42] V. Urick, F. Bucholtz, P. Devgan, J. McKinney, and K. Williams, “Phasemodulation with interferometric detection as an alternative to intensitymodulation with direct detection for analog-photonic links,” IEEE Trans.Microw. Theory Tech., vol. 55, no. 9, pp. 1978–1985, Sep. 2007.

[43] M. LaGasse and S. Thaniyavaru, “Bias-free high-dynamic-range phase-modulated fiber-optic link,” IEEE Photon. Technol. Lett., vol. 9, no. 5,pp. 681–683, May 1997.

[44] J. Wyrwas and M. Wu, “Dynamic range of frequency modulated direct-detection analog fiber optic links,” IEEE/OSA J. Lightw. Technol.,vol. 27, no. 24, pp. 5552–5562, Dec. 2009.

[45] H. Chi, X. Zou, and J. Yao, “Analytical models for phase-modulation-based microwave photonic systems with phase modulation to intensitymodulation conversion using a dispersive device,” IEEE/OSA J. Lightw.Technol., vol. 27, no. 5, pp. 511–521, Mar. 2009.

[46] R. Olshansky, “Microwave subcarrier multiplexing: New approach towideband lightwave systems,” IEEE Circuits Devices Mag., vol. 4, no. 6,pp. 8–14, Nov. 1988.

[47] H. Toda, T. Yamashita, T. Kuri, and K. Kitayama, “Demultiplexingusing an arrayed-waveguide grating for frequency-interleaved DWDMmillimeter-wave radio-on-fiber systems,” IEEE/OSA J. Lightw. Technol.,vol. 21, no. 8, pp. 1735–1741, Aug. 2003.

[48] C. Oh and W. Gu, “Fiber induced distortion in a subcarrier multiplexedlightwave system,” IEEE J. Sel. Areas Commun., vol. 8, no. 7, pp. 1296–1303, Sep. 1990.

[49] R. Hui et al., “Subcarrier multiplexing for high-speed optical transmis-sion,” J. Lightw. Technol., vol. 20, no. 3, pp. 417–427, Mar. 2002.

[50] R. Gross and R. Olshansky, “Multichannel coherent FSK experimentsusing subcarrier multiplexing techniques,” IEEE/OSA J. Lightw. Tech-nol., vol. 8, no. 3, pp. 406–415, Mar. 1990.

[51] R. Yuen and X. Fernando, “Analysis of sub-carrier multiplexed ra-dio over fiber link for the simultaneous support of WLAN andWCDMA systems,” Wireless Pers. Commun. J., vol. 33, no. 1, pp. 1–20,Apr. 2005.

[52] K. Iniewski, C. McCrosky, and D. Minoli, Network Infrastructure andArchitecture: Designing High-Availability Networks, 1st ed. Hoboken,NJ, USA: Wiley, 2008.


[53] M. Smit and C. Van Dam, “PHASAR-based WDM-devices: Principles,design and applications,” IEEE J. Sel. Topics Quantum Electron., vol. 2,no. 2, pp. 236–250, Jun. 1996.

[54] C. Lim, A. Nirmalathas, D. Novak, R. Tucker, and R. Waterhouse,“Technique for increasing optical spectral efficiency in millimetre-wavewdm fibre-radio,” IET Electron. Lett., vol. 37, no. 16, pp. 1043–1045,Aug. 2001.

[55] M. Bakaul, A. Nirmalathas, C. Lim, D. Novak, and R. Water-house, “Simultaneous multiplexing and demultiplexing of wavelength-interleaved channels in DWDM millimeter-wave fiber-radio net-works,” IEE/OSA J. Lightw. Technol., vol. 24, no. 9, pp. 3341–3352,Sep. 2006.

[56] M. Attygalle, C. Lim, G. Pendock, A. Nirmalathas, and G. Edvell,“Transmission improvement in fiber wireless links using fiber Bragggratings,” IEEE Photon. Technol. Lett., vol. 17, no. 1, pp. 190–192,Jan. 2005.

[57] H. K. Xijia Gu, Y. He, and X. Fernando, “Transmission efficiency im-provement in microwave fiber-optic link using sub-picometer optic band-pass filter,” in Proc. SPIE Conf. Photon. North, Sep. 2005, vol. 5971,Art. ID. 597123.

[58] M. Farwell, W. Chang, and D. Huber, “Increased linear dynamic rangeby low biasing the mach-zehnder modulator,” IEEE Photon. Technol.Lett., vol. 5, no. 7, pp. 779–782, Jul. 1993.

[59] E. Ackerman et al., “Signal-to-noise performance of two analog photoniclinks using different noise reduction techniques,” in Proc. IEEE/MTT-SInt. Microw. Symp., Jun. 2007, pp. 51–54.

[60] J. McKinney et al., “Sub-10-dB noise figure in a multiple-GHz analogoptical link,” IEEE Photon. Technol. Lett., vol. 19, no. 7, pp. 465–467,Apr. 2007.

[61] R.-P. Braun, G. Grosskopf, D. Rohde, and F. Schmidt, “Low-phase-noise millimeter-wave generation at 64 GHz and data transmission usingoptical sideband injection locking,” IEEE Photon. Technol. Lett., vol. 10,no. 5, pp. 728–730, May 1998.

[62] L. Johansson and A. Seeds, “Generation and transmission of millimeter-wave data-modulated optical signals using an optical injection phase-lock loop,” IEEE/OSA J. Lightw. Technol., vol. 21, no. 2, pp. 511–520,Feb. 2003.

[63] L. Chrostowski, X. Zhao, and C. Chang-Hasnain, “Microwave per-formance of optically injection-locked vcsels,” IEEE Trans. Microw.Theory Tech., vol. 54, no. 2, pp. 788–796, Feb. 2006.

[64] A. Ng’oma et al., “Performance of a multi-Gb/s 60 GHz radio overfiber system employing a directly modulated optically injection-lockedVCSEL,” IEEE/OSA J. Lightw. Technol., vol. 28, no. 16, pp. 2436–2444,Aug. 2010.

[65] T. Ismail, C.-P. Liu, J. Mitchell, and A. Seeds, “High-dynamic-rangewireless-over-fiber link using feedforward linearization,” IEEE/OSA J.Lightw. Technol., vol. 25, no. 11, pp. 3274–3282, Nov. 2007.

[66] C. Lim, M. Attygalle, A. Nirmalathas, D. Novak, and R. Waterhouse,“Analysis of optical carrier-to-sideband ratio for improving transmissionperformance in fiber-radio links,” IEEE Trans. Microw. Theory Tech.,vol. 54, no. 5, pp. 2181–2187, May 2006.

[67] M. Attygalle, C. Lim, and A. Nirmalathas, “Extending optical transmis-sion distance in fiber wireless links using passive filtering in conjunctionwith optimized modulation,” IEEE/OSA J. Lightw. Technol., vol. 24,no. 4, pp. 1703–1709, Apr. 2006.

[68] A. Karim and J. Devenport, “Noise figure reduction in externally mod-ulated analog fiber-optic links,” IEEE Photon. Technol. Lett., vol. 19,no. 5, pp. 312–314, Mar. 2007.

[69] R. Welstand et al., “Enhanced linear dynamic range property of franz-keldysh effect waveguide modulator,” IEEE Photon. Technol. Lett.,vol. 7, no. 7, pp. 751–753, Jul. 1995.

[70] C. Sun et al., “High spurious free dynamic range fibre link using a semi-conductor electroabsorption modulator,” IET Electron. Lett., vol. 31,no. 11, pp. 902–903, May 1995.

[71] J. Li et al., “Photonic-assisted microwave frequency measurement withhigher resolution and tunable range,” Opt. Lett., vol. 34, no. 6, pp. 743–745, Mar. 2009.

[72] X. J. Meng, T. Chau, and M. Wu, “Improved intrinsic dynamic distor-tions in directly modulated semiconductor lasers by optical injectionlocking,” IEEE Trans. Microw. Theory Tech., vol. 47, no. 7, pp. 1172–1176, Jul. 1999.

[73] H. K. Sung, E. Lau, and M. Wu, “Optical single sideband modulation us-ing strong optical injection-locked semiconductor lasers,” IEEE Photon.Technol. Lett., vol. 19, no. 13, pp. 1005–1007, Jul. 2007.

[74] L. Goldberg, H. Taylor, J. Weller, and D. Bloom, “Microwave signal gen-eration with injection-locked laser diodes,” IET Electron. Lett., vol. 19,no. 13, pp. 491–493, Jun. 1983.

[75] L. Johansson and A. Seeds, “36-GHz 140-Mb/s radio-over-fiber trans-mission using an optical injection phase-lock loop source,” IEEE Pho-ton. Technol. Lett., vol. 13, no. 8, pp. 893–895, Aug. 2001.

[76] S. Korotky and R. De Ridder, “Dual parallel modulation schemes forlow-distortion analog optical transmission,” IEEE J. Sel. Areas Com-mun., vol. 8, no. 7, pp. 1377–1381, Sep. 1990.

[77] H. Skeie and R. V. Johnson, “Linearization of electro-optic modulatorsby a cascade coupling of phase-modulating electrodes,” Proc. SPIE,Integr. Opt. Devices I, vol. 1583, pp. 153–164, Dec. 1991.

[78] E. Ackerman, “Broad-band linearization of a mach-zehnder electroop-tic modulator,” IEEE Trans. Microw. Theory Tech., vol. 47, no. 12,pp. 2271–2279, Dec. 1999.

[79] B. Masella, B. Hraimel, and X. Zhang, “Enhanced spurious-free dynamicrange using mixed polarization in optical single sideband mach-zehndermodulator,” IEEE/OSA J. Lightw. Technol., vol. 27, no. 15, pp. 3034–3041, Aug. 2009.

[80] C. Lim, A. Nirmalathas, K.-L. Lee, D. Novak, and R. Waterhouse, “In-termodulation distortion improvement for fiber ndash;radio applicationsincorporating ossb+c modulation in an optical integrated-access envi-ronment,” IEEE/OSA J. Lightw. Technol., vol. 25, no. 6, pp. 1602–1612,Jun. 2007.

[81] X. G. H. Kosek, Y. He, and X. Fernando, “All-optical demultiplexingof WLAN and Cellular CDMA radio signals,” IEEE/OSA J. Lightw.Technol., vol. 25, no. 6, pp. 1401–1409, Jun. 2007.

[82] X. Fernando and A. Sesay, “Adaptive asymmetric linearization of radioover fiber links for wireless access,” IEEE Trans. Veh. Technol., vol. 51,no. 6, pp. 1576–1586, Nov. 2002.

[83] L. Roselli et al., “Analog laser predistortion for multiservice radio-over-fiber systems,” IEEE/OSA Journal of Lightwave Technology, vol. 21,no. 5, pp. 1211–1223, May 2003.

[84] L. Johnson and H. Roussell, “Linearization of an interferometric mod-ulator at microwave frequencies by polarization mixing,” IEEE Photon.Technol. Lett., vol. 2, no. 11, pp. 810–811, Nov. 1990.

[85] B. Haas and T. Murphy, “A simple, linearized, phase-modulated ana-log optical transmission system,” IEEE Photon. Technol. Lett., vol. 19,no. 10, pp. 729–731, May 2007.

[86] Y. Chiu, B. Jalali, S. Garner, and W. Steier, “Broad-band electronic lin-earizer for externally modulated analog fiber-optic links,” IEEE Photon.Technol. Lett., vol. 11, no. 1, pp. 48–50, Jan. 1999.

[87] F. Yang, M. Marhic, and L. Kazovsky, “Nonlinear crosstalk andtwo countermeasures in SCM-WDM optical communication systems,”IEEE/OSA J. Lightw. Technol., vol. 18, no. 4, pp. 512–520, Apr. 2000.

[88] J. Marti, J. Fuster, and R. Laming, “Experimental reduction ofchromatic dispersion effects in lightwave microwave/millimetre-wavetransmissions using tapered linearly chirped fibre gratings,” IETElectron. Lett., vol. 33, no. 13, pp. 1170–1171, Jun. 1997.

[89] F. Ramos, J. Marti, V. Polo, and J. Fuster, “On the use of fiber-induced self-phase modulation to reduce chromatic dispersion effectsinmicrowave/millimeter-wave optical systems,” IEEE Photon. Technol.Lett., vol. 10, no. 10, pp. 1473–1475, Oct. 1998.

[90] H. Sotobayashi and K. Kitayama, “Cancellation of the signal fadingfor 60 GHz subcarrier multiplexed optical DSB signal transmissionin nondispersion shifted fiber using midway optical phase conjuga-tion,” IEEE/OSA J. Lightw. Technol., vol. 17, no. 12, pp. 2488–2497,Dec. 1999.

[91] J. Yu et al., “Optical millimeter-wave generation or up-conversion us-ing external modulators,” IEEE Photon. Technol. Lett., vol. 18, no. 1,pp. 265–267, Jan. 2006.

[92] A. Nirmalathas et al., “Digitized rf transmission over fiber,” IEEE Mi-crow. Mag., vol. 10, no. 4, pp. 75–81, Jun. 2009.

[93] Y. Yang, C. Lim, and A. Nirmalathas, “Investigation on transportschemes for efficient high-frequency broadband ofdm transmission infibre-wireless links,” J. Lightw. Technol., vol. 32, no. 2, pp. 267–274,Jan. 2014.

[94] U. Gliese, S. Norskov, and T. N. Nielsen, “Chromatic dispersion infiber-optic microwave and millimeter-wave links,” IEEE Trans. Microw.Theory Tech., vol. 44, no. 10, pp. 1716–1724, Oct. 1996.

[95] H. Kim and A. Gnauck, “Chirp characteristics of dual-drive. mach-zehnder modulator with a finite dc extinction ratio,” IEEE Photon. Tech-nol. Lett., vol. 14, no. 3, pp. 298–300, Mar. 2002.

[96] L. Chen, Y. Shao, X. Lei, H. Wen, and S. Wen, “A novel radio-over-fiber system with wavelength reuse for upstream data connection,” IEEEPhoton. Technol. Lett., vol. 19, no. 6, pp. 387–389, Mar. 2007.

[97] P. Winzer and R. Essiambre, “Advanced optical modulation formats,”Proc. IEEE, vol. 94, no. 5, pp. 952–985, May 2006.

[98] D. Blumenthal et al., “Fiber-optic links supporting baseband data andsubcarrier-multiplexed control channels and the impact of MMIC pho-tonic/microwave interfaces,” IEEE Trans. Microw. Theory Tech., vol. 45,no. 8, pp. 1443–1452, Aug. 1997.

[99] T. Kamisaka, T. Kuri, and K. Kitayama, “Simultaneous modulation andfiber-optic transmission of 10-Gb/s baseband and 60-GHz-band radiosignals on a single wavelength,” IEEE Trans. Microw. Theory Tech.,vol. 49, no. 10, pp. 2013–2017, Oct. 2001.


[100] Z. Jia, J. Yu, A. Chowdhury, G. Ellinas, and G. K. Chang, “Si-multaneous generation of independent wired and wireless servicesusing a single modulator in millimeter-wave-band radio-over-fiber sys-tems,” IEEE Photon. Technol. Lett., vol. 19, no. 20, pp. 1691–1693,Oct. 2007.

[101] Z. Jia, J. Yu, G. Ellinas, and G. K. Chang, “Key enabling technologies foroptical x2013;wireless networks: Optical millimeter-wave generation,wavelength reuse, architecture,” IEEE/OSA J. Lightw. Technol., vol. 25,no. 11, pp. 3452–3471, Nov. 2007.

[102] C.-T. Lin et al., “Hybrid optical access network integrating fiber-to-the-home and radio-over-fiber systems,” IEEE Photon. Technol. Lett.,vol. 19, no. 8, pp. 610–612, Apr. 2007.

[103] R. Llorente et al., “Ultra-wideband radio signals distribution in FTTHnetworks,” IEEE Photon. Technol. Lett., vol. 20, no. 11, pp. 945–947,Jun. 2008.

[104] M. Huchard et al., “Ultra-broadband wireless home network based on60-GHz WPAN cells interconnected via RoF,” IEEE/OSA J. Lightw.Technol., vol. 26, no. 15, pp. 2364–2372, Aug. 2008.

[105] M. Bakaul, A. Nirmalathas, C. Lim, D. Novak, and R. Waterhouse,“Hybrid multiplexing of multiband optical access technologies towardsan integrated DWDM network,” IEEE Photon. Technol. Lett., vol. 18,no. 21, pp. 2311–2313, Nov. 2006.

[106] A. Chowdhury, H.-C. Chien, Y.-T. Hsueh, and G. K. Chang, “Advancedsystem technologies and field demonstration for in-building optical-wireless network with integrated broadband services,” IEEE/OSA J.Lightw. Technol., vol. 27, no. 12, pp. 1920–1927, Jun. 2009.

[107] C. Lim, A. Nirmalathas, D. Novak, R. Waterhouse, and G. Yoffe,“Millimeter-wave broad-band fiber-wireless system incorporating base-band data transmission over fiber and remote lo delivery,” IEEE/OSA J.Lightw. Technol., vol. 18, no. 10, pp. 1355–1363, Oct. 2000.

[108] J. Wu, J. S. Wu, and H. W. Tsao, “A fiber distribution system for micro-cellular radio,” IEEE Photon. Technol. Lett., vol. 6, no. 9, pp. 1150–1152,Sep. 1994.

[109] J. S. Wu, J. Wu, and H. W. Tsao, “A radio-over-fiber network for micro-cellular system application,” IEEE Trans. Veh. Technol., vol. 47, no. 1,pp. 84–94, Feb. 1998.

[110] R. Welstand et al., “Dual-function electroabsorption waveguidemodulator/detector for optoelectronic transceiver applications,” IEEEPhoton. Technol. Lett., vol. 8, no. 11, pp. 1540–1542, Nov. 1996.

[111] L. Westbrook and D. Moodie, “Simultaneous bi-directional analoguefibre-optic transmission using an electroabsorption modulator,” IETElectron. Lett., vol. 32, no. 19, pp. 1806–1807, Sep. 1996.

[112] L. Noel et al., “Novel techniques for high-capacity 60-GHz fiber-radiotransmission systems,” IEEE Trans. Microw. Theory Tech., vol. 45, no. 8,pp. 1416–1423, Aug. 1997.

[113] D. Wake, D. Johansson, and D. Moodie, “Passive picocell: A new con-cept in wireless network infrastructure,” IET Electron. Lett., vol. 33,no. 5, pp. 404–406, Feb. 1997.

[114] A. Stohr, K. Kitayama, and D. Jager, “Full-duplex fiber-optic RF sub-carrier transmission using a dual-function modulator/photodetector,”IEEE Trans. Microw. Theory Tech., vol. 47, no. 7, pp. 1338–1341,Jul. 1999.

[115] A. Nirmalathas, D. Novak, C. Lim, and R. Waterhouse, “Wavelengthreuse in the WDM optical interface of a millimeter-wave fiber-wirelessantenna base station,” IEEE Trans. Microw. Theory Tech., vol. 49, no. 10,pp. 2006–2012, Oct. 2001.

[116] T. Kuri, K. Kitayama, and Y. Takahashi, “A single light-source configu-ration for full-duplex 60-GHz-band radio-on-fiber system,” IEEE Trans.Microw. Theory Tech., vol. 51, no. 2, pp. 431–439, Feb. 2003.

[117] M. Bakaul, A. Nirmalathas, and C. Lim, “Multifunctional WDM op-tical interface for Millimeter-wave fiber-radio antenna base station,”IEEE/OSA J. Lightw. Technol., vol. 23, no. 3, pp. 1210–1218, Mar. 2005.

[118] A. Kaszubowska, L. Hu, and L. Barry, “Remote downconversion withwavelength reuse for the radio/fiber uplink connection,” IEEE Photon.Technol. Lett., vol. 18, no. 4, pp. 562–564, Feb. 2006.

[119] L. Chen, H. Wen, and S. Wen, “A radio-over-fiber system with a novelscheme for millimeter-wave generation and wavelength reuse for up-linkconnection,” IEEE Photon. Technol. Lett., vol. 18, no. 19, pp. 2056–2058, Oct. 2006.

[120] J. Yu, Z. Jia, T. Wang, and G. Chang, “A novel radio-over-fiber config-uration using optical phase modulator to generate an optical mm-waveand centralized lightwave for uplink connection,” IEEE Photon. Technol.Lett., vol. 19, no. 3, pp. 140–142, Feb. 2007.

[121] Y. Y. Won, H. C. Kwon, and S. K. Han, “1.25-Gb/s wavelength-divisionmultiplexed single-wavelength colorless radio-on-fiber systems using re-flective semiconductor optical amplifier,” IEEE/OSA J. Lightw. Technol.,vol. 25, no. 11, pp. 3472–3478, Nov. 2007.

[122] M. F. Huang, J. Yu, Z. Jia, and G. K. Chang, “Simultaneous generationof centralized lightwaves and double/single sideband optical millimeter-wave requiring only low-frequency local oscillator signals for radio-

over-fiber systems,” IEEE/OSA J. Lightw. Technol., vol. 26, no. 15,pp. 2653–2662, Aug. 2008.

[123] X. Yu, T. Gibbon, and I. Monroy, “Bidirectional radio-over-fiber systemwith phase-modulation downlink and RF oscillator-free uplink using areflective SOA,” IEEE Photon. Technol. Lett., vol. 20, no. 24, pp. 2180–2182, Dec. 2008.

[124] H. C. Ji, H. Kim, and Y. C. Chung, “Full-duplex radio-over-fiber systemusing phase-modulated downlink and intensity-modulated uplink,” IEEEPhoton. Technol. Lett., vol. 21, no. 1, pp. 9–11, Jan. 2009.

[125] T. Kuri, T. Nakasyotani, H. Toda, and K.-I. Kitayama, “Characterizationsof supercontinuum light source for WDM millimeter-wave-band radio-on-fiber systems,” IEEE Photon. Technol. Lett., vol. 17, no. 6, pp. 1274–1276, Jun. 2005.

[126] T. Nakasyotani, H. Toda, T. Kuri, and K. Kitayama, “Wavelength-division-multiplexed millimeter-waveband radio-on-fiber system usinga supercontinuum light source,” J. Lightw. Technol., vol. 24, no. 1,pp. 404–410, Jan. 2006.

[127] P. Delfyett et al., “Optical frequency combs from semiconductor lasersand applications in ultrawideband signal processing and communica-tions,” IEEE/OSA J. Lightw. Technol., vol. 24, no. 7, pp. 2701–2719,Jul. 2006.

[128] L. Hanzo, L. L. Yang, E. L. Kuan, and K. Yen, Single and MulticarrierDS-CDMA: Multi User Detection, Space-Time Spreading, Synchroniza-tion, Standards and Networking, 1st ed. Hoboken, NJ, USA: Wiley,2003.

[129] H.-T. Huang, C.-T. Lin, Y.-T. Chiang, C.-C. Wei, and C.-H. Ho, “Simple2 × 2 mimo 60-ghz optical/wireless system with extending fiber trans-mission distance,” J. Lightw. Technol., vol. 32, no. 20, pp. 3660–3667,Oct. 2014.

[130] S. Mikroulis, O. Omomukuyo, M. Thakur, and J. Mitchell, “Investigationof a smf-mmf link for a remote heterodyne 60-ghz ofdm rof based gigabitwireless access topology,” J. Lightw. Technol., vol. 32, no. 20, pp. 3645–3653, Oct. 2014.

[131] G. Gordon, M. Crisp, R. Penty, T. Wilkinson, and I. White, “Feasibil-ity demonstration of a mode-division multiplexed mimo-enabled radio-over-fiber distributed antenna system,” J. Lightw. Technol., vol. 32,no. 20, pp. 3521–3528, Oct. 2014.

[132] K. Lau, G. Lutes, and R. Tjoelker, “Ultra-stable rf-over-fiber transport innasa antennas, phased arrays and radars (invited),” J. Lightw. Technol.,vol. 32, no. 20, pp. 3440–3451, Oct. 2014.

[133] A. Koonen and E. Tangdiongga, “Photonic home area networks,” J.Lightw. Technol., vol. 32, no. 4, pp. 591–604, Feb. 2014.

Varghese Antony Thomas received the B.E. (Hons)degree in electrical and electronics engineering fromBirla Institute of Technology and Science (Goa Cam-pus), Pilani, India, in 2010 and the M.Sc. degreein wireless communications from the University ofSouthampton, Southampton, U.K. in 2011. He iscurrently working toward the Ph.D. degree with theWireless Communications Research Group, Univer-sity of Southampton. He is the recipient of theseveral academic awards including the Common-wealth Scholarship of the Government of UK and

Mayflower Scholarship of University of Southampton. His research interestsinclude optical communications, optical wireless integration and backhaul formultiple-input–multiple-output and radio-over-fiber systems.


Mohammed El-Hajjar received the B.Eng. de-gree in electrical engineering from the AmericanUniversity of Beirut, Beirut, Lebanon, in 2004 andthe M.Sc. degree in radio-frequency communica-tion systems and the Ph.D. degree in wireless com-munications from the University of Southampton,Southampton, U.K., in 2005 and 2008, respectively.Following his doctoral studies, he joined ImaginationTechnologies as a Design Engineer, where he workedon designing and developing Imagination’s multi-standard communications platform, which resulted

in three patents. In January 2012, he joined the Southampton Wireless researchGroup, School of Electronics and Computer Science, University of Southamp-ton, Southampton, U.K., as a Lecturer. His research interests include thedevelopment of intelligent communications systems, including energy-efficienttransceiver design, cross-layer optimization for large-scale networks, multiple-input–multiple-output systems, millimeter-wave communications, and radio-over-fiber systems. He is currently a Lecturer with the School of Electronicsand Computer Science, University of Southampton, Southampton, U.K. He isthe author of a Wiley–IEEE book and in excess of 50 journal and internationalconference papers. He received several academic awards, including the DorothyHodgkin Postgraduate Award and the IEEE ICC2010 Best Paper Award.

Lajos Hanzo received the Master’s degree in elec-tronics, the D.Sc. degree, and the “Doctor HonarisCausa” degree from the Technical University of Bu-dapest, Budapest, Hungary, in 1976, 1983, and 2009,respectively. During his 38-year career in the fieldof telecommunications, he has held various researchand academic posts in Hungary, Germany, and theU.K. Since 1986, he has been with the School ofElectronics and Computer Science, University ofSouthampton, Southampton, U.K., where he holdsthe Chair in telecommunications. From 2009 to

2012, he was the Editor-in-Chief of the IEEE Press and a Chaired Professorwith Tsinghua University, Beijing, China. He has successfully supervisedabout 100 Ph.D. students, coauthored 20 John Wiley/IEEE Press books onmobile radio communications, totalling in excess of 10 000 pages, published1460 research entries at IEEE Xplore, acted both as TPC and General Chairof IEEE conferences, presented keynote lectures, and has been awarded anumber of distinctions. Currently, he is directing an academic research team,working on a range of research projects in the field of wireless multimediacommunications sponsored by industry, the Engineering and Physical SciencesResearch Council U.K., the European IST Programme, and the Mobile VirtualCentre of Excellence UK. He is an enthusiastic supporter of industrial andacademic liaison, and he offers a range of industrial courses. He is also aGovernor of the IEEE Vehicular Technology Society. He is a Fellow of theRoyal Academy of Engineering, the Institution of Engineering and Technology,and the European Association for Signal Processing.

Documents

IEEECOMMUNICATION SURVEYS &TUTORIALS,VOL.17,NO.2