See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/352438509

Phase noise modeling and characterization of a self-started VCSEL-based

harmonic frequency generator (SVHFG)

Article  in  Optics & Laser Technology · June 2021

DOI: 10.1016/j.optlastec.2021.107176

CITATION READS

1 54

4 authors:

Christian Daniel Muñoz Gloria Margarita Varón Durán

10 PUBLICATIONS   16 CITATIONS   
National University of Colombia
63 PUBLICATIONS   210 CITATIONS   
SEE PROFILE
SEE PROFILE

Fabien Destic Angélique Rissons

University of Toulouse Institut Supérieur de l'Aéronautique et de l'Espace (ISAE)
22 PUBLICATIONS   102 CITATIONS    52 PUBLICATIONS   278 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Hybrid optical fiber-wireless communication to support Tactile Internet View project

Internet of the future View project

All content following this page was uploaded by Christian Daniel Muñoz on 17 November 2021.

The user has requested enhancement of the downloaded file.

 Phase noise modeling and characterization of a self-started
VCSEL-based harmonic frequency generator (SVHFG)

Christian Daniel Muñoz, Margarita Varón, Fabien Destic, Angélique Rissons

Preprint Version

Abstract
This article presents the experimental characterization of the first self-started closed-loop sys-
tem for the generation of microwave harmonics using a long-wavelength VCSEL modulated in a
gain-switched regime. A 15-GHz microwave signal is generated with a phase noise of -104 dBc/Hz
at 10 kHz from the carrier when the fundamental frequency is f0 = 2.5 GHz. The side-mode
suppression ratio (SMSR) for all the microwave frequencies generated is higher than 30 dB. We
propose a predictive model to estimate the phase noise of the fundamental oscillating frequency,
considering the noise contributions of each loop component. Several simulation scenarios are in-
cluded to demonstrate the significant influence of the electrical power, the laser frequency noise,
and the optical fiber dispersion over the phase noise performance. We also present the carrier-to-
noise ratio deterioration of the microwave harmonics distributed through an optical fiber link.

Keywords— g ain-switching, laser frequency noise, microwave harmonics generation, phase noise, VCSEL

1 Introduction
The transport, generation, and processing of microwave signals using optoelectronic components be-
came popular in the 1990s under the name of microwave photonics (MWP) [1]. Its success has been
associated with the large component bandwidths covering the microwave band, but also the millimeter
waves (mm-waves) and THz frequencies [2]. In addition, the development of low relative intensity noise
(RIN) and high bandwidth lasers, external modulators with lower Vπ voltage, and fast photodetectors
with high responsivity and bandwidth has boosted the MWP improvement [3]. Consequently, MWP
is positioned as an attractive solution for analog microwave transmission to remote systems, e.g., the
radio astronomy Square Kilometre Array (SKA) [4], and a wide variety of applications such as the
measurement of physical variables [5], modern instrumentation [2], healthcare [6][7], remote sensing
[8][9], and radio-over-fiber (RoF) systems [10][11].
Currently, several mechanisms for the generation of microwave signals are available in the literature.
One of the simplest ways is the optical heterodyning of two free-running lasers in a fast photodetector.
Though extremely high frequencies are achieved, the phase noise performance is limited because the
optical phases are un-correlated and additional techniques, such as optical phase lock loop (OPLL),
must be applied [12]. Conversely, the optoelectronic oscillator (OEO) is the most popular mechanism
for generating spectrally pure microwave signals (low phase noise) [13]. Two of the main advantages
of the OEO architecture are its high configuration flexibility and the possibility of extracting the
generated signal in the electrical and optical domains. Some modifications include a laser source
modulated directly [14] or through an intensity modulator, phase modulator [15][16] or polarization
modulator [17][18]. In 2006, an OEO based on a directly modulated vertical-cavity surface-emitting
laser (VCSEL) was introduced [19]. This architecture, called VCSEL-based optoelectronic oscillator
(VBO), takes advantage of the VCSEL characteristics, simplifies the classical OEO architecture [20][21],
and has been used for generating low jitter optical pulses at 850 nm [22]. The main drawback of VBO
is the technical difficulty of producing mirrors with high reflectivity (> 99%) and high gain active
zones with materials based on ternary and quaternary alloys in O and C-band.
In order to generate microwave signals at higher frequency bands, such as the Ka-band (27 - 40
GHz) and W-band (75 - 110 GHz), some authors have designed optical frequency multipliers where a
reference frequency can be duplicated [23][24], quadrupled [25][26], sextupled [27], and even, octupled

1
[28]. A common factor among many topologies is the use of Mach-Zehnder modulators owing to
their non-linear transfer function, that allows the generation of different modes carefully selected to
produce high-frequency signals in a photodetector [29][30]. Similarly, an all-optical harmonic frequency
generator (HFG) incorporates a Faraday mirror to modulate an electrical reference microwave signal
by passing the optical wave twice through a phase modulator [31].
In this article, we propose and demonstrate the first harmonic frequency generator using a straight-
forward closed-loop architecture and a directly modulated long-wavelength VCSEL in gain-switching
(GS) condition. The repetition frequency, also called fundamental frequency, is modified employing
two band-pass filters centered at 1.25 GHz and 2.5 GHz. In both cases, nine harmonics are extracted
and characterized in terms of phase noise. Two long-wavelength VCSELs emitting in the O and C-
band are used in order to verify the impact of the optical fiber dispersion on the phase noise of the
generated microwave signals. We also provide and validate a predictive phase noise model based on
the linear feedback systems theory that includes the noise contributions of the system components.
Finally, the harmonics generated are extracted in the optical domain (optical pulses) and distributed
through an optical link. The carrier-to-noise ratio of three harmonics is measured for different optical
powers at the receiver input.
This paper is structured as follows. In Section 2, we describe the architecture of the proposed
microwave harmonics generator. The predictive phase noise model is detailed in Section 3, and some
phase noise estimations are also presented in order to verify the impact of components characteristics
over the overall noise performance. Then, the system is validated experimentally and the results are
condensed in Section 4. In the same section are included the phase noise predictions obtained after
applying the proposed model. The transmission results of the microwave harmonics are described and
summarized in Section 5. Finally, we propose some conclusions about our work.

2 SVHFG principle and architecture

The self-started VCSEL-based harmonic frequency generator (SVHFG) is a delay-line based and self-
feeding ring system for microwave signal generation. A VCSEL is directly modulated in gain-switching
regime at a fundamental frequency f0 established by a microwave pass-band filter. The gain-switched
VCSEL generates an optical pulse train with a pulse-to-pulse time (train periodicity) controlled by
the repetition rate of the drive electrical signal. In the frequency domain, the pulse train corresponds
to a frequency comb where f0 provides the space between modes. By direct photodetection, the
optical pulses are converted into an electric pulse train composed of microwave harmonics at n × f0
frequencies. The driving conditions significantly influence both the quality of the optical comb and the
noise level of the microwave harmonics. Indeed, the bias current and the amplitude and frequency of
the driving signal have more impact on the performance than the type of laser cavity or other internal
laser parameters. As in other optoelectronic oscillators, the resonant cavity consists of a low-loss
single-mode glass fiber and its length directly impacts the phase noise performance of the microwave
signals.
The architecture of the SVHFG is illustrated in Fig. 1. The optical beam coming from the VCSEL
is photodetected and immediately amplified by microwave amplifier 1. The bandwidth of both the
photodetector and the amplifier must be higher than the frequency of the highest order harmonic
desired. The microwave harmonics 2f0 , 3f0 . . . nf0 , are recovered through the electrical coupler 1
(EC1). The variable attenuator and the microwave amplifier 2 ensure the electrical power level to
modulate the VCSEL directly in gain-switching condition. The band-pass filter is responsible for
filtering the fundamental frequency f0 that can be extracted through the electrical coupler 2 (EC2).
Hence, the bandwidth of microwave amplifier 2 must cover at least the fundamental frequency f0 .
The gain Barkhausen condition is satisfied by adjusting the variable attenuator. The free spectral
range (FSR) is the reciprocal of the closed-loop time delay τ given by the optical and electrical delays
(τo and τe , respectively). The optical fiber acts as a resonant cavity and determines the optical delay
through τo = nF L/c, where L is the optical fiber delay-line (ODL) length, nF is the optical fiber
refractive index, and c is the speed of the light in vacuum. The quality factor Q of the resonant cavity
is directly related to the optical delay by Q = πf0 τo .
Concerning the phase noise of the generated microwave signal, it is strongly related to the optical
fiber length, so when the length increases, the Q factor increases, and the noise is decreased. A trade-
off between Q and FSR must be achieved to obtain an acceptable phase noise level and an FSR that

2
 VCSEL

Fundamental
frequency
Optical path Electrical path

Figure 1: Self-started VCSEL-based harmonic frequency generator architecture. EC: Electrical coupler.

can be rejected by the bandpass filter. Therefore, the SVHFG operation will depend on the filter
bandwidth, which should be as narrow as possible to allow only one oscillating mode.

3 Predictive phase noise model

3.1 Model description
The SVHFG phase noise can be modeled by the linear feedback system theory proposed by Rubiola
[32] for delay-line oscillators, whereby the noise power spectral density (PSD) includes the noise con-
tributions of the different loop elements and their transfer functions. In the phase noise model, shown
in Fig. 2, all signals are Laplace transforms of the phase fluctuations. Therefore, ϕ(s) represents the
phase fluctuations of the fundamental frequency, ψ(s) are the noise sources further detailed, and Bd
and Bf are the transfer functions of the delay line and the band-pass filter, respectively.

(s)
Delay
line
+
Bd(s) + ∑

(s) Bf (s)
Filter

Figure 2: Phase noise model of the SVHFG. ϕ(s): phase fluctuations of the fundamental frequency.
ψ(s) noise sources. Bd and Bf : transfer functions of the delay line and the band-pass filter, respectively.

The phase fluctuations of the fundamental frequency are stated as:

ϕ(s) = ϕ(s) · Bd (s) · Bf (s) + ψ(s) · Bf (s) (1)

which leads to:

Bf (s)
ϕ(s) = · ψ(s) = H(s) · ψ(s) (2)
1 − Bd (s) · Bf (s)
In the frequency domain, the single-side band (SSB) power spectral density of the SVHFG phase-
noise is given by:
2
Sϕ (f ) = |H(jf )| · Sψ (f ) (3)
In this way, an expression for phase noise can be obtained by analyzing the noise sources Sψ (f )
and the transfer function H(s) separately.

3
 Z ph ZA

Figure 3: Circuit model of additive noise sources.

3.1.1 Closed-loop transfer function

For the SVHFG, the resonator is composed of an optical delay line and a band-pass filter tuned at
f0 . The phase impulse response in the time domain for the delay and filter are denoted as bd and bf
respectively and are expressed as [33]:
bd (t) = δ (t − τ ) (4)
1 − τtf
bf (t) = e (5)
τf
with τf the filter group delay given by τf = Q/πf0 . By applying a Laplace transform to bd and bf ,
the transfer function of the feedback loop corresponds to:

Bf (s) 1
H(s) = = (6)
1 − Bd (s) · Bf (s) 1 + sτf − e−sτ

Therefore, the magnitude squared of the closed-loop transfer function is obtained from Eq. 6
replacing s = jω,
2 1
|H(jω)| = (7)
2 + ω 2 τf2 − 2 cos (ωτ ) + 2ωτf sin (ωτ )

3.1.2 Noise sources

The initial conditions are guaranteed by the optical and electrical components noises, which are injected
into the loop through the summing block in Fig. 2. Some of these noises are produced by external
factors (mechanical and thermal disturbances), whereas others are caused by the intrinsic properties
of the component materials and are referred to as internal noise sources. The internal noise sources in
the SVHFG can be divided into two groups [34]: additive noise generated by random processes, e.g.,
thermal noise, and multiplicative noise coming from the microwave amplifier and VCSEL frequency
noise. Indeed, the power spectral density Sψ is:

Sψ (f ) = Sadd (f ) + Smult (f ) (8)

Each  additive noise source is represented by a current source in the electrical circuit model in Fig.
3. i2tot is the total noise contributions, ZA is the input impedance of the first amplifier, and Zph the
impedance load of the photodetector.
The total current noise is expressed as:
q q
hi2tot i = hi2RIN i + hi2sh i + hi2th i + hi2scat i (9)




with
i2RIN the VCSEL relative intensity noise (RIN), i2sh the shot noise, i2th the thermal noise,
and i2scat the intensity-to-phase noise conversion processes caused by the intensity variations inside
the optical fiber due to Raman and Rayleigh scattering. The current passing through ZA is:
Zph
q
iA = hi2tot i · (10)
ZA + Zph

4
 Hence, the power spectral density of the additive noise sources referred to the microwave amplifier
impedance is defined as:
 2
Zph
Sadd (f ) = i2tot · ZA · = i2tot · Zeq



(11)
ZA + Zph

The stimulated Raman scattering (SRS) and the stimulated Brillouin scattering (SBS) are not
considered in this noise analysis because the power emitted by the VCSEL is below the threshold powers
for all fiber lengths used in the experimental characterization. Conversely, the Rayleigh scattering,
caused by microscopic variations, is included through two phenomena: double Rayleigh backscattering
(DRBS) and single Rayleigh backscattering - single reflection (SRBS-SR) [35].
The two main multiplicative noise processes of the SVHFG are the phase noise of the microwave
amplifier and the VCSEL frequency noise to phase noise conversion due to the optical fiber disper-
sion effect. Assuming that both noise sources are uncorrelated, the power spectral density of the
multiplicative noise corresponds to:

Smult (f ) = Samp (f ) + Sf req (f ) (12)

The power spectral density of the microwave amplifier phase noise includes white and flicker noise
[32]. Hence,
1
Samp (f ) = b0 · f 0 + b−1 · (13)
f
The white noise component is estimated from the equivalent noise spectrum density N and the
carrier power P0 at the amplifier input [36],

N F kT0
b0 = = (14)
P0 P0
with F the noise figure and kT0 the thermal energy. The SVHFG architecture shown in Fig. 1
incorporates two cascaded amplifiers, so the Friss formula must be applied and Eq. 14 becomes:
 
F2 − 1 kT0
b0 = F1 + · (15)
A21 P0

In contrast, flicker noise is independent of the carrier power at the amplifier input over a wide
frequency range. Thus, b−1 is constant and its value is related to the size of the amplifier active region
and its gain. For microwave amplifiers, b−1 is in the range of 10−10 (fair quality) to 10−14 (best quality)
rad2 /Hz [32]. In the case of SVHFG, b−1 is given by the sum of the individual contributions b1−1 and
b2−1 of the two amplifiers.
The second multiplicative noise process is produced by the optical delay length fluctuations gener-
ated in turn by the laser frequency fluctuations. For the SVHFG, these delay fluctuations are converted
into phase fluctuations inside the optical fiber due to the in-phase closed-loop condition. The power
spectral density of the phase noise induced by the laser frequency noise is given by [37]:

Sf req (f ) = Cϕ2 · Sν (f ) (16)

where Sν (f ) is the power spectral density of the VCSEL frequency noise and the conversion factor Cϕ
obeys to:
L
Cϕ = 2π · f0 · λ20 · Dλ · (17)
c
where λ0 is the VCSEL wavelength and Dλ is the dispersion value of the optical fiber at λ0 . Eq.
16 reveals that the dispersion conversion of laser frequency noise can deteriorate the SVHFG phase
noise performance through the Dλ value. Furthermore, the phase noise can be improved by using
dispersion-shifted fibers (DSF) and C-band VCSELs, or O-band VCSELs with standard single-mode
fibers (such as SMF-28).

5
3.2 SVHFG phase noise estimation
The predictive phase noise model presented in the previous section is a useful computational tool to
estimate the phase noise performance of a closed-loop system where the laser source is directly modu-
lated. Likewise, our model can be used to establish the technical characteristics of a microwave signal
generator given a desired phase noise behavior. For the SVHFG system, its phase noise performance
is evaluated using technical parameters obtained from component characterizations, laboratory com-
ponent data sheets, or the literature. Four simulation scenarios (shown in Fig. 4) are established to
estimate the impact of the optical fiber length, electrical power level at the first input amplifier, VC-
SEL RIN, and VCSEL frequency noise, over the phase noise of the fundamental microwave frequency
at 1.25 GHz.
Firstly, the phase noise for three fiber lengths is estimated by assuming identical electrical and
optical conditions, i.e., the same component characteristics and identical electrical and optical power
levels within the loop. Fig. 4 a. shows the predicted phase noise curve when L = 1, 2, and 5
km. During the three lower frequency offset decades, the predicted phase noise deteriorates when the
optical fiber length is reduced from 5 km to 1 km by about 12 dB as an effect of the resonant cavity
quality factor reduction. For higher frequency offset values, the phase noise curves exhibits sharp
peaks (also called non-oscillating modes) equally spaced at ω = 2πn/τ , and minimum values when
ω = (2n − 1)2π/τ with n a positive integer. The peak magnitudes are governed by the influence of the
optical delay τ and the filter group delay τf over the closed-loop transfer function expressed in Eq. 7.
For the input electrical power of the first microwave amplifier, the phase noise is improved by
increasing the electrical power, as shown Fig. 4 b.. The predicted phase noise at 10 kHz from the
carrier drops from -119 dBc/Hz to -137 dBc for a total power increment of 20 dB (from -55 dBm to
-35 dBm). This improvement is achieved because the far-from-carrier phase noise is dominated by the
white noise of the first amplifier, which varies inversely with the input power (see Eq. 14). Concerning
the close-to-carrier phase noise, higher electrical power produces a higher corner frequency, and the
Leeson effect on the amplifier flicker noise is anticipated.
The VCSEL RIN effect on SVHFG phase noise is depicted in Fig. 4 c. For high-intensity noise
levels (-110 dB/Hz), the phase noise close-to and far-from-carrier is deteriorated. This deterioration is
caused because the RIN contribution exceeds the other noise sources, including the microwave amplifier
noise. For low RIN values (< -130 dB/Hz), the phase noise remains unchanged owing to its reduced
contribution.
Finally, Fig. 4 d. shows the effect of VCSEL frequency noise on phase noise. In all cases, the power
spectral density of the frequency noise was assumed with a slope of -10 dB/dec. The reduction of the
frequency noise from 1 × 1014 /f to 1 × 1011 /f Hz2 /Hz generates a non-representative improvement of
the predicted phase noise of about 1.6 dB at 10 kHz from the carrier. The most outstanding effect
occurs close to the carrier because the contribution of frequency noise exceeds the contributions of the
RIN and the Leeson effect on the amplifier Flicker noise.
According to simulation results, the far-from-carrier phase noise (offset frequencies higher than 1
kHz) is highly influenced by the optical fiber length, electrical power level, and VCSEL RIN. In the
first case, even if the optical fiber length can be enlarged in order to reduce the noise level, non-
oscillating modes closer to the carrier are produced. Due to the impact of these modes over the
jitter performance, they must be reduced or eliminated by including a narrower pass-band filter in
the electrical path. Concerning the electrical power level at the first microwave amplifier input, lower
white noise contributions can be obtained by including a higher optical power VCSEL and a variable
optical attenuator between the optical fiber and the photodetector shown in Fig. 1. Through this
attenuator, the optical power at the photodetector input can be adjusted in order to increase the
electrical power at the photodetector output, and ensuring at the same time, the oscillating and gain-
switching condition. Lastly, a lowest VCSEL RIN contribution can be achieved by biasing the laser
at the highest bias current at which the GS regime is reached. The phase noise enhancement of the
fundamental frequency directly impacts the phase noise of the generated microwave harmonics, which
can be extracted and amplified through the simple use of a tuned filter and a reduced bandwidth
microwave amplifier.

6
 a. -40
-50

SSB phase noise (dBc/Hz)

-60
-70
-80
-90
-100
-110
-120 1 km
2 km
-130
5 km
-140
101 102 103 104 105
-50
b.
-60
SSB phase noise (dBc/Hz)

-70
-80
-90
-100
P0 = -35 dBm
-110
P0 = -40 dBm
-120
P0 = -45 dBm
-130 P0 = -50 dBm
-140 P0 = -55 dBm
-150
101 102 103 104 105
-50
c.
-60
SSB phase noise (dBc/Hz)

-70
-80
-90
-100
-110
RIN = -150 dB/Hz
-120 RIN = -130 dB/Hz
RIN = -120 dB/Hz
-130
RIN = -110 dB/Hz
-140
101 102 103 104 105
-50
d.
-60
SSB phase noise (dBc/Hz)

-70
-80
-90
-100
-110
S
(f) = 1011 /f Hz2/Hz
-120
S (f) = 1013/f Hz2/Hz
-130 S (f) = 1014/f Hz2/Hz
-140
101 102 103 104 105
Offset frequency (Hz)

Figure 4: Predicted phase noise curves by modifying a. optical fiber length, b. input power of the
first microwave amplifier, c. VCSEL RIN, and d. VCSEL frequency noise.

7
4 SVHFG experimental characterization and model validation
The SVHFG is implemented according to the configuration shown in Fig. 1. The microwave amplifier
connected to the photodetector has a 38-dB flat gain, a notably wide band from 0.5 to 18 GHz, and
a typical noise figure of 3 dB. The gain of the second amplifier is 24 dB at 1.25 GHz and 39 dB at
2.5 GHz. Two cavity filters tuned at 1.25 GHz and 2.5 GHz are used to establish the fundamental
frequency. The optical delay line length is fixed at 5 km of standard single mode fiber (SMF) and
the VCSEL bias current at 6 mA for all implementations. In order to eliminate the power reflections
towards microwave amplifier 2, an electrical circulator is inserted before the band-pass filter. The
coupled outputs of EC1 and EC2 are attenuated by 10 and 20 dB, respectively. Finally, a 17-GHz
bandwidth fast photodetector and C-band and O-band VCSELs are used.
The experimental characterization presented in this section include the electrical microwave har-
monics spectra and measurements of system stability in the time and frequency domain.

4.1 Microwave harmonics spectra

The electrical spectra of nine microwave harmonics generated employing different VCSELs and f0 =
1.25 GHz are depicted in Fig. 5. Even though only the laser sources are modified, the noise level in
the case of the C-band VCSEL is above the floor noise. This noise increase is caused by the electrical
power difference within the loop. The emission power of the C-band VCSEL is lower than the O-
band VCSEL, so to achieve the oscillating condition, it is necessary to compensate the power through
the tunable electrical attenuator. In this way, a higher electrical power directly modulates the laser,
reducing both the carrier-to-noise ratio around the main optical mode and the phase noise of the
harmonics.
-20 f0 1st 2nd 3rd
4th
-30 5th
6th
-40 7th
-50 8th 9th
-60
Electrical power (dBm)

-70
-80
-90
-20 f0 1st 2nd 3rd 4th
-30 5th 6th 7th
8th 9th
-40
-50
-60
-70
-80
-90
1.25 2.50 3.75 5.00 6.25 7.50 8.75 10.00 11.25 12.50
Frequency (GHz)
C-band VCSEL O-band VCSEL

Figure 5: Electrical spectra of the microwave harmonics using a C-band and an O-band VCSEL for
f0 = 1.25 GHz.

As expected, both the fundamental frequency and the first three harmonics of the C-band SVHFG
are higher than the O-band SVHFG. From the fourth harmonic, the power of the C-band SVHFG
decreases drastically at a rate higher than 4.5 dB per harmonic. In contrast, the O-band SVHFG
decreases at a rate lower than 3 dB per harmonic (except for the 8th harmonic). Hence, the carrier
power at 12.5 GHz of the O-band implementation is 20 dB higher than the C-band. This performance
difference is attributed to the electrical access of the VCSELs and the deterioration of the optical
modes at higher modulation powers.
Focusing on the fundamental frequency and the microwave harmonics individually, each spectrum
comprises the main spike at the microwave frequency and several equally spaced non-oscillating modes
(Fabry-Pérot cavity effect of the optical fiber). The frequency distance or FSR is preserved in all the
harmonics generated, and the optical fiber length mainly determines its value. In our case, the FSR is
39.8 kHz when L = 5 km, as shown in Fig. 6 for the 1.25-GHz SVHFG using an O-band VCSEL.

8
 In turn, the side-mode suppression ratio (SMSR) is impacted by electrical power reduction at higher
frequencies. The highest SMSR value is 50 dB for the O-band SVHFG (51 dB in C-band SVHFG case)
and is obtained at the fundamental frequency. Then, an SMSR reduction of about 7 dB is produced
for the first harmonic of both optical bands. In any case, including the ninth harmonic (f = 10 ∗ f0 ),
the SMSR is higher than 30 dB for all implementations presented in this paper.

-20
-40 SMSR = 50 dB
-60 FSR
39.8 kHz
-80

Electrical power (dBm)

-100
-120
1.250163 1.250214 1.250265 1.250316
Frequency (GHz)
-20
-40 SMSR = 43 dB
-60 FSR
39.8 kHz
-80
-100
-120
2.500683 2.500734 2.500785 2.500836
Frequency (GHz)
Fundamental First harmonic

Figure 6: Electrical spectra of the fundamental frequency (f0 = 1.25 GHz) and first harmonic when
an O-band VCSEL is used.

For application purposes, each harmonic can be filtered and amplified with a narrow band-pass
filter and a tuned microwave amplifier. Likewise, the power level of the harmonics can be improved
by using electrical couplers with less attenuated coupled outputs and a bandwidth adjusted to the
required frequencies.

4.2 Frequency-domain stability characterization - phase noise

The phase noise measurement for the fundamental frequency and the microwave harmonics generated
by the O- and C-band SVHFG are performed using the direct measurement method. The phase noise
curves of the fundamental frequency when f0 = 1.25 GHz are shown in Fig. 7. According to the slopes
of each curve, three noise processes are observed. From right to left, the white phase noise process (f 0 )
is present in both cases, appears from 300- to about 10-kHz offset, and corresponds to the contribution
of all white noise sources inside the loop. Then, a slope of -20 dB/dec from 10-kHz to 120-Hz and
200-Hz offset (C and O-band, respectively) is the signature of a white frequency noise process (f −2 )
and is generated by the Leeson effect on the white noise of the amplifier. The third noise process, with
a -25 dB/dec slope, covers the frequencies closest to the carrier. This noise process is characterized by
up-conversion of the noise generated by Rayleigh scattering, the VCSEL frequency noise, and mainly
the higher flicker noise contribution by the two cascade microwave amplifiers.
The phase noises at 10 kHz from the carrier are -126 dBc/Hz and -125 dBc/Hz in C and O-band,
respectively. We obtained nearby values because the noise characteristics of VCSELs are similar at
1.25 GHz (RIN = -140dB/Hz), and the power conditions within the loop were adjusted to be the same
through a tunable attenuator.
As in any frequency multiplication method, the phase noise of the microwave harmonics is degraded
by a factor of 20×logN , where N is the multiplication factor [31]. Fig. 8 presents the phase noise curves
of three harmonics generated by the O-band SVHFG at 2.5, 3.75, and 5 GHz. Although the noise
level is getting worse for higher harmonics, in all cases, the noise processes are preserved in almost the
whole frequency range, except for low frequencies where the measurement method accuracy is limited.
The measured phase noises at 10 kHz offset are -114.4, -112.6, and -109.4 dBc/Hz, respectively. These
values represent a phase noise increase of 10.7, 12.5, and 15.7 dB relative to the fundamental frequency
phase noise and are higher than the expected theoretical values (6, 9.5, and 12 dB) due to the non-flat
frequency response of the optoelectronic segment composed of the photodetector and the amplifier.
Similarly to the fundamental frequency, the phase noise curves include a bump around 1 kHz offset

9
Figure 7: Phase noise of fundamental frequency at 1.25 GHz using a. C-band VCSEL b. O-band
VCSEL.

produced by the higher impact of the Rayleigh scattering at O-band wavelengths and long optical fiber
lengths.

-40 First harmonic - 2.5 GHz

Second harmonic - 3.75 GHz
-50
SSB phase noise (dBc/Hz)

Third harmonic - 5 GHz

-60

-70

-80

-90

-100

-110

-120
101 102 103 104 105
Offset frequency (Hz)

Figure 8: Phase noise curves of the first three harmonics extracted from the O-band SVHFG at 1.25
GHz.

The phase noise evolution of microwave harmonics measured at 10 kHz from the carrier using O
and C-band VCSEL are depicted in Fig. 9. The phase noise of both implementations is quite similar
up to the third harmonic. After that, the phase noise of the C-band SVHFG deteriorates fast, as
predicted by the optical carrier-to-noise ratio reduction and the accelerated decrease in the power level

10
of the harmonics. In the case of the O-band SVHFG, the phase noise degrades logarithmically with
an average offset of 4 dB compared to the expected values (red dashed line).
-85

-90

-95

Phase noise (dBc/Hz)

-100

-105

-110

-115

-120

-125

0 1 2 3 4 5 6 7 8 9
Harmonic
C-band O-band Expected value

Figure 9: Phase noise evolution at 10 kHz from the carrier of the microwave harmonics generated by
the C-band and O-band SVHFG.

Table 1 shows a performance comparison of the carriers generated at 5, 10, and 15 GHz for several
optoelectronic oscillator topologies. The results presented include an O-band SVHFG with fundamen-
tal frequencies f0 = 1.25 and f0 = 2.5 GHz. For the three selected carriers, the SVHFG performance,
in terms of phase noise, is close to other recent implementations available in the literature. At 10 GHz,
the SVHFG phase noise is lower than more complex architectures, including several loops and different
modulation schemes (intensity and phase). In some systems, the oscillation frequency can be adjusted
by controlling the laser wavelengths [38, 39] or the cascaded finite impulse response (FIR) sections of
a multiple source microwave photonic filter [40]. Although the fundamental frequency of our system is
not tunable, the straightforward SVHFG architecture enables the generation of adjustable frequency
carriers through a tunable bandpass filter and microwave amplifiers in the desired bands. It is worth
pointing out that the fundamental frequency is limited mainly by the VCSEL electrical access and its
bandwidth. Higher fundamental and harmonic frequencies can be obtained employing commercially
available on-chip VCSELs (> 10 GHz).

4.3 Time-domain stability characterization - Allan deviation

Today, applications such as global positioning system (GPS) and digital communications networks
need stable and accurate timing sources in order to avoid data errors. A source is stable when its
frequency does not change significantly over time and is accurate when the average frequency is equal
to the nominal frequency [44]. The stability verification process in the time-domain, also called long-
term stability, consists in measuring the source frequency variations in a specific time interval and
processing these values through statistical tools (due to their random nature).
The Allan variance [45], known as two-sample variance and denoted as σy2 (τ ), is a convergent
measurement of fractional-frequency fluctuations of timing sources with multiple noise processes. There
are several versions for the Allan variance, which allow identifying some non-visible noise processes
when the standard Allan variance is applied. The square root σy (τ ) is called Allan deviation (ADEV)
and is usually employed to express measurements in the time domain.
The long-term stability of the SVHFG is analyzed using the Allan deviation and the overlapping
Allan deviation in order to decrease variability and improve the statistical confidence of statistical
estimators. The procedure consists in measuring the fractional frequency every τA seconds via a
frequency counter during a specific time. Fig. 10 shows both Allan deviations of the C-band SVHFG
at 1.25 GHz. The fractional frequencies are measured every second (τA = 1 s) during 60 hours. The
confidence intervals are evaluated through a χ2 distribution and the degrees of freedom are estimated
assuming a dominant white frequency noise process (n = −2).
Based on both deviations, the same three noise processes can be identified. Even though there is
an equivalence between the phase noise curves and the Allan deviation, the these processes are not

11
 Table 1: SVHFG performance comparison with other OEO topologies.
OEO topology Oscillation Phase noise @ Modulation Tunable Harmonics Reference
frequency 10 kHz offset genera-
(GHz) (dBc/Hz) tion
Dual loop OEO 5.4 -115 External No No [41]
SVHFG - 5 -112 Direct No Yes This work
f0 = 1.25 GHz
SVHFG - 5 -117 Direct No Yes This work
f0 = 2.5 GHz
Delay line based 10 -110 Both No No [42]
OEO
Dual loop OEO 10 -100 External Yes No [38]
Dual loop coupled 10 -90 External Yes No [39]
OEO
Resonator based 10 -128 External No No [43]
OEO
SVHFG - 10 -106 Direct No Yes This work
f0 = 1.25 GHz
SVHFG - 10 -112 Direct No Yes This work
f0 = 2.5 GHz
Multi-loop OEO 15 -118 External Yes No [40]
SVHFG - 15 -104 Direct No Yes This work
f0 = 2.5 GHz

Figure 10: Allan deviation of the O-band SVHFG at 1.25 GHz.

observed in the phase noise curves. This effect is mainly due to the limitations of the phase noise
measurement method (direct method via an electrical spectrum analyzer) for frequencies below 100
Hz.
From left to right, the first process is Flicker frequency noise, which is characterized by a 0-slope
(τ 0 ). This noise process vanishes quickly due to the predominant conversion process of the flicker noise
from the amplifiers within the loop. The second process, called random walk frequency noise, has a τ 1/2
slope and is attributed to the frequency and intensity noise (at low frequencies) of the VCSEL. Finally,
frequency drifts have a slope proportional to τ 1 and are caused by system temperature variations,
mechanical vibrations, and aging of components. The gain-switched VCSEL contributes significantly
to these frequency drifts owing to the temperature fluctuations of the active zone produced by the fast
injection of carriers.

12
 The frequency stability at 1-s averaging time is 2.96 × 10−10 for both deviations. For higher
averaging times, the Allan deviations gradually increase to 6.33 × 10−8 and 8.97 × 10−7 at 1000- and
10000-s averaging time. As stated before, this stability reduction is attributed to the non-temperature
control and the mechanical vibrations faced by the SVHFG. Although our system is more stable at
1000-s averaging time than other free-running OEO architectures (1×10−6 ) [46], its long-term stability
can be improved by controlling the VCSEL and optical fiber temperature.

4.4 Predictive phase-noise model validation

The phase noise model proposed in section 3 is validated using the technical characteristics summarized
in Table 2 when C and O-band VCSELs are employed at 1.25 GHz. Some component values were
measured, and others were taken from the literature or technical component description.

Table 2: Parameter used to validate the predictive phase noise model when the fundamental frequency
is 1.25 GHz.
Parameter Value Source
τf 0.157 µs Measured
L 1.09, 2.10, and 5.11 km Measured
λ0 1528 nm and 1317 nm Measured
Dλ1550nm 17 ps/(nm·km) Manufacturer
Dλ1310nm 0 ps/(nm·km) Manufacturer
ZA 50 Ω Manufacturer
Zph 100 Ω Manufacturer
P0 -47 dBm Measured
T0 25 ◦ C Assumed
F1 , F2 3 dB Manufacturer
A1 40 dB Manufacturer
b1−1 , b2−1 -100 rad2 /Hz [32]
RIN @ 1.25 GHz -140 dB/Hz Measured
Sν (f ) 1013 /f Hz2 /Hz [47]

Firstly, we identify the open-loop noise contributions from the VCSEL and both microwave am-
plifiers, i.e., the additive and multiplicative noise, between 10 Hz and 300 kHz. The additive noise,
produced by the VCSEL RIN, Rayleigh scattering, thermal noise, and shot noise, is the lowest noise
contribution and does not dramatically impact the overall phase noise, as shown in Fig. 11. As for the
multiplicative noise, the power level at the first amplifier input and the optical fiber length affect the
phase noise level. For all the model validations presented in this section, the electrical power is assumed
to be constant, and only the optical fiber length is modified. For L = 1 and 2 km, the Flicker and
white noise from the microwave amplifiers are dominant in the whole frequency range. When L = 5
km, the phase noise is divided into two regions. From 10 Hz to 1.5 kHz, the close-to-carrier phase noise
is dominated by the VCSEL frequency noise conversion through the optical fiber dispersion. From 1.5
kHz, the noise level is determined by the white noise of amplifiers. Assuming that only one microwave
amplifier is used to achieve the GS and oscillating condition, the noise close to the carrier can be
reduced by around 5 dB (see Fig 11). It is worth mentioning that, in all cases, the far-from-carrier
noise is dominated by the white noise of the amplifiers and can be reduced by increasing the electrical
power inside the loop (see Eq. 14).
Secondly, we estimate the closed-loop phase noise for three optical fiber lengths and both optical
bands. Fig. 12 shows the predicted and measured phase noise curves when L = 1, 2, and 5 km. For all
the fiber lengths, we can identify the same three noise processes detailed in section 4.2: white phase
noise, white frequency noise, and the up-conversion Flicker noise. The predicted phase noise curves
fit the measured curves well, but they include a fourth noise process close to the carrier called Flicker
frequency noise. Its slope is -30 dB/dec, and it is caused by the Leeson effect of the amplifier’s flicker
noise at frequencies below the corner frequency fc and by the VCSEL frequency noise. This last noise
process does not appear in the measurements due to the low sensitivity of the measurement method at
frequencies below 1 kHz [48]. The non-oscillating modes at frequencies above 30 kHz from the carrier

13
 -100
Additive noise
VCSEL frequency noise - 1 km

SSB phase noise (dBc/Hz)

VCSEL frequency noise - 2 km
-110 VCSEL frequency noise - 5 km
Two amplifiers
One amplifier
-120

-130

-140
101 102 103 104 105
Offset frequency (Hz)

Figure 11: Estimated open-loop noise contributions.

match well with the measured peaks. The shape and distribution of the non-oscillating modes are
governed by the resonant cavity delay τ , and the filter group delay τf , which relates the bandwidth
and quality factor of the filter (see Eq. 6). The difference between the measured and simulated mode
level is caused by the measurement resolution used.
All frequency ranges of noise processes observed in Fig. 12 are modified when the optical fiber is
lengthened. The non-oscillating modes dominate the white phase noise processes, and their frequency
noise ranges increase as an effect of FSR reduction. Consequently, the white frequency noise process,
which represents the Leeson effect on the loop components’ white noise, is reduced. As expected for
fiber lengths lower than 2 km, the simulated phase noises close-to-carrier are very close for both optical
bands (Fig. 12 a. and b.). This effect, also observed in the measured curves, is produced because
the amplifier Flicker noise makes a stronger contribution than other noise sources. Conversely, when
L = 5 km (Fig. 12 c.), the phase noise close-to-carrier increases because there is a higher contribution
from the noise conversion process of the VCSEL frequency noise through the optical fiber dispersion
(see Eq. 16). Due to the zero-dispersion of the optical fiber in the O-band, a lower phase noise level
is measured and simulated from 10 Hz to 3 kHz in this optical band. Thus, the phase noise of the
SVHFG loop can be reduced by using a single microwave amplifier, VCSEL with reduced frequency
noise, and optical fiber with zero dispersion in the laser emission wavelength.
As intended, the phase noise at 10 kHz from the carrier is improved by lengthening the optical fiber.
Table 3 summarizes the estimated and measured phase noise values at 10 kHz from the carrier. From
L = 1 km to 5 km, the phase noise in all cases is enhanced by 10 to 13 dB. The difference between the
measured and simulated phase noise levels, which can also be seen in Fig. 12, is produced because some
crucial parameters were not measured, such as the amplifier Flicker noise and the VCSEL frequency
noise, and because for all the phase noise estimations the power input at the first amplifier (P0 ) was
assumed constant.
According to the proposed model, the phase noise value at 10 kHz in both optical bands should
be equal for each fiber length. However, the measured values show a difference of about 1.6 dB for
each length. This difference is attributed to the small power differences of two VCSELs and the direct
measurement uncertainty.

Table 3: Phase noise values measured at 10 kHz from the carrier at 1.25 GHz.
L C-band O-band C-band O-band
mea- mea- pre- pre-
sured sured dicted dicted
(dBc/Hz) (dBc/Hz) (dBc/Hz) (dBc/Hz)
1 km -113.7 -115.1 -116.3 -116.3
2 km -119.1 -121.0 -121.6 -121.6
5 km -126.7 -125.1 -129.8 -130.1

14
Figure 12: Predicted and measured closed-loop phase noise of SVHFG at 1.25 GHz for a. L = 1 km,
b. L = 2 km, and c. L = 5 km.

5 Microwave harmonics transmission

As in any other optoelectronic oscillator, the microwave harmonics generated with the SVHFG can be
extracted in both the electrical and optical domains. In the optical case, an optical coupler must be
placed between the optical isolator and the optical delay line in the setup shown in Fig. 1. The optical
pulses generated by the gain switching technique can be distributed through an optical fiber, and then
the microwave harmonics can be recovered by a fast photodetector. The quality of these harmonics is
affected by the noise contribution of the optoelectronic components included in the optical link. Fig.
13 shows the setup used to verify the carrier-to-noise ratio (CNR or C/N) of microwave harmonics
when C-band optical pulses are transmitted. Due to the reduced power of the VCSEL, the optical
pulses are amplified by an erbium-doped fiber amplifier (EDFA) with a gain fixed at 20 dB. Then, a
variable optical attenuator (VOA) makes it possible to modify the input power of the photodetector
PD2 to evaluate the impact of the input optical power on the electrical noise level.

15
 Optical path Electrical path

Figure 13: Optical link for the microwave harmonics distribution.

The CNR of the optical link can be expressed as [49]:


IP2 D2 · (m2 2)
CN R = 2 (18)
4kT 2hν·IP D2 ·F
Rph2 + 2q · IP D2 + RIN · IP2 D2 + Pin−EDF A

where IP D2 is the photodetected current, RIN is the VCSEL RIN at the frequency f0 , m the modulation
index (m > 0.1 for large modulation), 4kT /Rph2 and 2qIP D2 are the thermal and shot noise on the
photodetector PD2, respectively. The last term of the denominator represents the amplifier-generated
excess intensity noise density with F the EDFA noise factor, Pin−EDF A the optical power at the
amplifier input, and Rph2 the load resistance. The noise factor contribution by the variable optical
attenuator is assumed to be zero.
For small IP D2 values, the floor noise is governed by the thermal noise, whereas for high currents,
the main contribution is made by the intensity noise of the optical amplifier. Thus, the variable
attenuator controls both the noise floor level and the power of the fundamental frequency, i.e., the link
CNR. Fig. 14 shows the CNR measurement of the transmitted fundamental frequency and the first
three harmonics for five different attenuations.

55 Fundamental
First harmonic
50
Second harmonic
45 Third harmonic

40
C/N (dB)

35
30
25
20
15
-10 -8 -6 -4 -2 0 2 4 6
Photodetected optical power (dBm)

Figure 14: CNR evolution at different photodetected powers.

As expected, the CNR of the fundamental frequency and harmonics is reduced when the pho-
todetected optical power decreases. For all optical powers, the CNR of the fundamental frequency is
higher due to the higher power at this frequency (see Fig. 5). For both fundamental and harmonic
frequencies, a total optical attenuation of 17 dB produces an average CNR deterioration of 30 dB. For
harmonics at frequencies higher than 5 GHz, the carrier-to-noise ratio can deteriorate significantly due
to the frequency response of the photodetector.
The phase noise floor of the harmonics can be determined from Eq. 18. As for the CNR, the floor
noise is dominated by the thermal noise for low received power and by the RIN laser for high power.
In this sense, phase noise can be reduced by increasing the optical input power of the photodetector.
Additionally, the performance of the phase noise is affected by the process of amplitude-to-phase
conversion in the photodetector. To minimize this non-linear effect, the optical input power must be
adjusted in order to reduce the electrical phase fluctuations caused by the optical intensity fluctuations.

16
6 Conclusion
This paper presents the first self-started microwave harmonics generator using a straightforward single
closed-loop system based on a gain-switched VCSEL and an optical delay line. The GS technique is
applied for increasing the fundamental system frequency limited by the VCSEL bandwidth (including
the electrical access response). In the electrical domain, the generated signal comprises a fundamental
microwave frequency determined by a band-pass filter and several microwave harmonics. The funda-
mental frequency stability was characterized in time- and frequency-domain through Allan deviation
and phase noise measurements. Unlike other harmonic generation systems, our system requires nei-
ther an external modulation scheme (intensity or phase modulation) nor a microwave reference signal.
Furthermore, both the fundamental frequency and the microwave harmonics generated can be contin-
uously adjusted by replacing the band-pass filter with a tunable microwave band-pass filter according
to the electrical characteristics of the loop components.
The SVHFG predictive phase noise model is based on the linear feedback system theory and
includes the main additive and multiplicative noise sources. First model estimations demonstrated
that noise sources impact both close-to and far-from-carrier phase noise. Likewise, the model shows
that the phase noise level can be reduced by increasing the electrical power at the first amplifier input,
lengthening the optical fiber, and using VCSELs with reduced RIN noise levels. Meanwhile, the phase
variations produced by the VCSEL frequency noise and fiber dispersion impact mainly on the noise
level close to the carrier. Through our model, we have also demonstrated that the floor phase noise
level is governed by the white noise of two microwave amplifiers and that the phase noise close-to-
carrier can be reduced by using only one microwave amplifier with sufficient gain and low white noise
contribution, short fibers, and fibers with zero dispersion in the VCSEL emission wavelength.
After several experimental validations, the predicted phase noise curves agree well with the curves
obtained experimentally. However, the model reveals other noise processes close to the carrier that
are not visible in the measurements due to the low sensitivity of the measurement method. A sup-
plementary noise level offset is observed for the predicted curves over the whole analyzed frequency
range. The offset value remains constant and is produced because some model parameters were as-
sumed from the literature. Therefore, the proposed model allows us to determine the components’
noise characteristics according to the required phase noise. It is worth mentioning that even though
the model allows the design of SVHFG systems, some parameters such as the electrical power within
the loop cannot be arbitrarily established because this power is responsible for achieving the oscillating
and gain-switching conditions. We consider that the model developed can be improved by including
the VCSEL RIN at low frequencies and the noise generated from the non-linear distortions caused by
large-signal modulation of the VCSEL, e.g., clipping.
For all the microwave harmonics generated, the non-oscillating modes are conserved at a distance
given by the optical fiber length, and their SMSR values are higher than 30 dB. The phase noise
degradation of the harmonics follows the same principle as any other frequency multiplication system.
Frequencies up to 15 GHz are generated with a phase noise at 10 kHz offset below -100 dBc/Hz. As
demonstrated with the predictive model, the optical fiber dispersion and the VCSEL frequency noise
do not impact the phase noise at 10 kHz from the carrier of the fundamental frequency when different
optical fiber lengths and emission wavelengths are used.
The optical output of the SVHFG is extracted from the system and transmitted through a simple
optical transmission system. The CNR decreases with the optical power received at the photodetector
due to the reduced power of each harmonic and the higher floor noise related to the photodetected
current. In terms of phase noise, the transmission system was not validated, but the harmonics are
expected to deteriorate quickly due to additional noise sources, such as the amplified spontaneous
emission (ASE) of the EDFA amplifier.
Higher fundamental frequencies and microwave harmonics can be generated using VCSELs with
higher modulation bandwidths such as the on-chip VCSELs or by applying the gain-switching and
optical injection locking (OIL) techniques simultaneously. Through the OIL technique, we can increase
the VCSEL bandwidth and reduce the RIN level contribution to the overall phase noise performance
[50]. The SVHFG applicability can be extended by characterizing the generated optical pulses in terms
of jitter, pulse width, and amplitude.

17
Acknowledgement
This work was supported by Departmental Government of Nariño (Grant No. BPIN 2013000100092),
ECOS-Nord project through Colciencias - Colombia (Grant No. M301PR03F18) and French Ministry
of Higher Education (Grant No. C17P01), Fondation ISAE-SUPAERO.

References
[1] D. Jäger, “Traveling-wave optoelectronic devices for microwave and optical applications,” in Pro-
ceedings of the Progress in Electromagnetics Research Symposium (PIERS), p. 327, 1991.
[2] J. Yao, “A Tutorial on Microwave Photonics,” Photonics Society Newsletter, vol. 26, no. 3, pp. 4–
12, 2012.
[3] S. Iezekiel, Microwave Photonics: Devices and Applications. Wiley - IEEE, John Wiley & Sons,
Mar. 2009.
[4] P. E. Dewdney, P. J. Hall, R. T. Schilizzi, and T. J. L. W. Lazio, “The Square Kilometre Array,”
Proceedings of the IEEE, vol. 97, pp. 1482–1496, Aug. 2009.
[5] X. Zou, X. Liu, W. Li, P. Li, W. Pan, L. Yan, and L. Shao, “Optoelectronic oscillators (OEOs) to
sensing, measurement, and detection,” IEEE Journal of Quantum Electronics, vol. 52, pp. 1–16,
Jan. 2016.
[6] X. Li, Body Matched Antennas for Microwave Medical Applications. PhD thesis, Karlsruher
Institut für Technologie KIT, 2013.
[7] C. Li, M. R. Tofighi, D. Schreurs, and T.-S. J. Horng, Principles and Applications of
RF/Microwave in Healthcare and Biosensing. Elsevier Science, 2016.
[8] Y. Liu, C. Hu, Y. Dong, B. Xu, W. Zhan, and C. Sun, “Geometric accuracy of remote sensing
images over oceans: The use of global offshore platforms,” Remote Sensing of Environment,
vol. 222, pp. 244–266, Mar. 2019.
[9] J. A. Nanzer, Microwave and Millimeter-wave Remote Sensing for Security Applications. Artech
House, 2012.
[10] Y. Tong, C. Chow, G. Chen, C. Peng, C. Yeh, and H. K. Tsang, “Integrated Silicon Photonics
Remote Radio Frontend (RRF) for Single-Sideband (SSB) Millimeter-Wave Radio-Over-Fiber
(ROF) Systems,” IEEE Photonics Journal, vol. 11, pp. 1–8, Apr. 2019.
[11] Y. Tian, K. Lee, C. Lim, and A. Nirmalathas, “60 GHz Analog Radio-Over-Fiber Fronthaul
Investigations,” Journal of Lightwave Technology, vol. 35, pp. 4304–4310, Oct. 2017.
[12] U. Gliese, T. N. Nielsen, S. Nørskov, and K. E. Stubkjær, “Multifunctional Fiber-Optic Microwave
Links Based on Remote Heterodyne Detection,” IEEE Trans. Microw. Theory Tech., vol. 46,
pp. 458–468, May 1998.
[13] X. S. Yao and L. Maleki, “Optoelectronic oscillator for photonic systems,” IEEE Journal of
Quantum Electronics, vol. 32, pp. 1141–1149, Jul. 1996.
[14] H. Sung, X. Zhao, E. K. Lau, D. Parekh, C. J. Chang-Hasnain, and M. C. Wu, “Optoelectronic
Oscillators Using Direct-Modulated Semiconductor Lasers Under Strong Optical Injection,” IEEE
Journal of Selected Topics in Quantum Electronics, vol. 15, no. 3, pp. 572–577, 2009.
[15] Y. Shao, X. Han, Q. Ye, B. Zhu, Y. Dai, C. Wang, and M. Zhao, “Low Power RF Signal Detection
Using a High Gain Tunable OEO based on Equivalent Phase Modulation,” Journal of Lightwave
Technology, p. 1, 2019.
[16] T. Sakamoto, T. Kawanishi, and M. Izutsu, “Optoelectronic oscillator using a LiNbO3 phase
modulator for self-oscillating frequency comb generation,” Optics letters, vol. 31, pp. 811–813,
Mar. 2006.

18
[17] S. Pan and J. Yao, “Optical Clock Recovery Using a Polarization-Modulator-Based Frequency-
Doubling Optoelectronic Oscillator,” Journal of Lightwave Technology, vol. 27, pp. 3531–3539,
Aug. 2009.
[18] Z. Tang, S. Pan, D. Zhu, R. Guo, Y. Zhao, M. Pan, D. Ben, and J. Yao, “Tunable Optoelectronic
Oscillator Based on a Polarization Modulator and a Chirped FBG,” IEEE Photonics Technology
Letters, vol. 24, pp. 1487–1489, Sep. 2012.
[19] M. Varón-Durán, J.-M. Martin, A. Le Kernec, and C. Mollier, “VCSEL Based Oscillator for
Harmonic Frequency Generation,” in Proceedings of the 5th Topical Meeting on Optoelectronic
Distance/Displacement Measurements and Applcations, (Madrid), pp. 96–101, 2006.
[20] M. Varón Durán, A. Le Kernec, and J.-C. Mollier, “Opto-microwave source using a harmonic
frequency generator driven by a VCSEL-based ring oscillator,” in Proceedings of the European
Microwave Association, vol. 3, pp. 248–253, EUMA European Microwave Association, 2007.
[21] A. Hayat, M. Varón, A. Bacou, A. Rissons, and J. C. Mollier, “2.49 ghz low phase-noise opto-
electronic oscillator using 1.55µm vcsel for avionics and aerospace applications,” in 2008 IEEE
International Meeting on Microwave Photonics jointly held with the 2008 Asia-Pacific Microwave
Photonics Conference, (Gold Coast, Qld, Australia), pp. 98–101, Sep. 2008.
[22] P. Devgan, D. Serkland, G. Keeler, K. Geib, and P. Kumar, “An optoelectronic oscillator using
an 850-nm VCSEL for generating low jitter optical pulses,” IEEE Photonics Technology Letters,
vol. 18, no. 5, pp. 685–687, 2006.
[23] L. Wang, N. Zhu, W. Li, and J. Liu, “A Frequency-Doubling Optoelectronic Oscillator Based on
a Dual-Parallel Mach-Zehnder Modulator and a Chirped Fiber Bragg Grating,” IEEE Photonics
Technology Letters, vol. 23, pp. 1688–1690, Nov. 2011.
[24] X. Liu, W. Pan, X. Zou, D. Zheng, L. Yan, and B. Luo, “Frequency-Doubling Optoelectronic
Oscillator Using DSB-SC Modulation and Carrier Recovery Based on Stimulated Brillouin Scat-
tering,” IEEE Photonics Journal, vol. 5, no. 2, 2013.

[25] D. Zhu, S. Liu, D. Ben, and S. Pan, “Frequency-Quadrupling Optoelectronic Oscillator for Mul-
tichannel Upconversion,” IEEE Photonics Technology Letters, vol. 25, pp. 426–429, Mar. 2013.
[26] D. Zhu, S. Pan, and D. Ben, “Tunable Frequency-Quadrupling Dual-Loop Optoelectronic Oscil-
lator,” IEEE Photonics Technology Letters, vol. 24, pp. 194–196, Feb. 2012.

[27] S. Pan and J. Yao, “Tunable Subterahertz Wave Generation Based on Photonic Frequency Sextu-
pling Using a Polarization Modulator and a Wavelength-Fixed Notch Filter,” IEEE Transactions
on Microwave Theory and Techniques, vol. 58, pp. 1967–1975, Jul. 2010.
[28] M. Hasan and T. J. Hall, “A photonic frequency octo-tupler with reduced RF drive power and
extended spurious sideband suppression,” Optics & Laser Technology, vol. 81, pp. 115–121, Jul.
2016.
[29] Y. Fu, X. Zhang, B. Hraimel, T. Liu, and D. Shen, “Mach-Zehnder: A Review of Bias Control
Techniques for Mach-Zehnder Modulators in Photonic Analog Links,” IEEE Microwave Magazine,
vol. 14, pp. 102–107, Nov. 2013.

[30] Y. Gao, A. Wen, Q. Yu, N. Li, G. Lin, S. Xiang, and L. Shang, “Microwave Generation With
Photonic Frequency Sextupling Based on Cascaded Modulators,” IEEE Photonics Technology
Letters, vol. 26, pp. 1199–1202, Jun. 2014.
[31] P. O. Hedekvist, B. . Olsson, and A. Wiberg, “Microwave harmonic frequency generation utilizing
the properties of an optical phase Modulator,” Journal of Lightwave Technology, vol. 22, pp. 882–
886, Mar. 2004.
[32] E. Rubiola, Phase Noise and Frequency Stability in Oscillators. The Cambridge RF and Microwave
Engineering Series, Cambridge: Cambridge University Press, 2008.

19
[33] E. Rubiola and R. Brendel, “A generalization of the Leeson effect,” FEMTO-ST Institute, 2010.
[34] O. Lelièvre, V. Crozatier, P. Berger, G. Baili, O. Llopis, D. Dolfi, P. Nouchi, F. Goldfarb, F. Brete-
naker, L. Morvan, and G. Pillet, “A Model for Designing Ultralow Noise Single- and Dual-Loop
10-GHz Optoelectronic Oscillators,” Journal of Lightwave Technology, vol. 35, pp. 4366–4374,
Oct. 2017.

[35] V. Kalavally, I. D. Rukhlenko, M. Premaratne, and T. Win, “Multipath Interference in Pulse-

Pumped Fiber Raman Amplifiers: Analytical Approach,” Journal of Lightwave Technology,
vol. 28, pp. 2701–2707, Sep. 2010.
[36] E. Rubiola and V. Giordano, “On the 1/f frequency noise in ultra-stable quartz oscillators,” IEEE
Trans Ultrason Ferroelect Freq Contr, vol. 54, pp. 15–22, Jan. 2007.

[37] K. Volyanskiy, Y. K. Chembo, L. Larger, and E. Rubiola, “Contribution of Laser Frequency

and Power Fluctuations to the Microwave Phase Noise of Optoelectronic Oscillators,” Journal of
Lightwave Technology, vol. 28, pp. 2730–2735, Sep. 2010.
[38] H. Du, R. Guo, H. Peng, Y. Xu, and Z. Chen, “Tunable Optical Comb Generation Utilizing an Op-
toelectronic Oscillator Based on Stimulated Brillouin Scattering,” in 2018 Asia Communications
and Photonics Conference (ACP), pp. 1–3, oct 2018.
[39] F. Fan, J. Hu, W. Zhu, J. Wang, Y. Gu, X. Han, Z. Wu, and M. Zhao, “Tunable optoelectronic
oscillator based on coupled double loops and stimulated brillouin scattering,” in 2017 International
Topical Meeting on Microwave Photonics (MWP), pp. 1–3, oct 2017.

[40] G. Charalambous, G. K. M. Hasanuzzaman, A. Perentos, and S. Iezekiel, “High-Q wavelength

division multiplexed optoelectronic oscillator based on a cascaded multi-loop topology,” Optics
Communications, vol. 387, pp. 361–365, Mar. 2017.
[41] G. Charalambous, G. K. M. Hasanuzzaman, A. Perentos, and S. Iezekiel, “Optoelectronic recir-
culating delay line implementation of a high-q optoelectronic oscillator,” in 2017 International
Topical Meeting on Microwave Photonics (MWP), pp. 1–4, oct 2017.
[42] Y. Cheng, J. Yan, and S. Zhao, “Self-Oscillating Parametric Optical Frequency Comb Generation
Using an Electroabsorption Modulated Laser-Based Optoelectronic Oscillator,” IEEE Photonics
Journal, vol. 11, no. 4, pp. 1–9, 2019.

[43] K. Saleh, P. H. Merrer, O. Llopis, and G. Cibiel, “Optoelectronic oscillator based on fiber ring
resonator: Overall system optimization and phase noise reduction,” in 2012 IEEE International
Frequency Control Symposium Proceedings, pp. 1–6, 2012.
[44] N. Da Dalt and A. Sheikholeslami, Understanding Jitter and Phase Noise: A Circuits and Systems
Perspective. Cambridge: Cambridge University Press, 2018.

[45] D. W. Allan, “Statistics of atomic frequency standards,” Proceedings of the IEEE, vol. 54, no. 2,
pp. 221–230, 1966.
[46] K. Xu, Z. Wu, J. Zheng, J. Dai, Y. Dai, F. Yin, J. Li, Y. Zhou, and J. Lin, “Long-term stability
improvement of tunable optoelectronic oscillator using dynamic feedback compensation,” Optics
Express, vol. 23, no. 10, pp. 12935–12941, 2015.

[47] F. Gruet, A. Al-Samaneh, E. Kroemer, L. Bimboes, D. Miletic, C. Affolderbach, D. Wahl,

R. Boudot, G. Mileti, and R. Michalzik, “Metrological characterization of custom-designed 894.6
nm VCSELs for miniature atomic clocks,” Optics Express, vol. 21, pp. 5781–5792, mar 2013.
[48] D. Owen, “Good Practice Guide to Phase Noise Measureme,” tech. rep., National Physical Lab-
oratory, 2004.

[49] V. Auroux, A. Fernandez, O. Llopis, P. H. Merrer, and A. Vouzellaud, “Microwave phase noise
properties of optical links involving small signal and gain saturated optical amplifiers,” in 2014
European Frequency and Time Forum (EFTF), pp. 381–383, 2014.

20
 [50] J. Coronel, M. Varón, and A. Rissons, “Phase noise analysis of a 10-GHz optical injection-locked
vertical-cavity surface-emitting laser-based optoelectronic oscillator,” Optical Engineering, vol. 55,
no. 9, 2016.

21

View publication stats