SCIENCE CHINA

Information Sciences
. REVIEW . December 2022, Vol. 65 221401:1–221401:64
https://doi.org/10.1007/s11432-021-3524-0

Microwave photonics
Jianping YAO1* & Jose CAPMANY2,3*
1
Microwave Photonics Research Laboratory, School of Electrical Engineering and Computer Science,
University of Ottawa, Ottawa ON K1W 1K3, Canada;
2
ITEAM Research Institute, Universitat Politècnica de València, València 46022, Spain;
3
iPronics, Programmable Photonics, 7 C. Pedro Duque, València 46022, Spain

Received 19 October 2021/Revised 7 April 2022/Accepted 24 June 2022/Published online 25 August 2022

Abstract Microwave photonics, an interdisciplinary field that combines microwave engineering and pho-
tonic technology for the generation, transmission, processing, and control of microwave signals, to take advan-
tage of the broad bandwidth, high frequency, and low loss offered by modern photonics, has been intensively
researched for the last few decades, and numerous solutions have been proposed and demonstrated. In this
article, an overview about microwave photonics is provided which covers the basic concepts of microwave
photonics, photonic-assisted microwave signal generation, photonic-assisted microwave signal processing, and
true time delay beamforming. The implementation of microwave photonic systems based on photonic inte-
grated circuits is also reviewed, including the design, fabrication, and material platforms, application-specific
photonic integrated circuits for microwave photonics, and programmable integrated microwave photonics.
The challenges and future applications of microwave photonics are also discussed.
Keywords microwave photonics, photonic integrated circuits, microwave photonic link, microwave pho-
tonic signal processor, optoelectronic oscillator, radio over fiber

Citation Yao J P, Capmany J. Microwave photonics. Sci China Inf Sci, 2022, 65(12): 221401, https://doi.org/
10.1007/s11432-021-3524-0

1 Introduction
Microwave photonics is an area that deals with the use of photonic techniques to generate, process, con-
trol, and transmit microwave signals to take advantage of the high frequency, large bandwidth and low loss
offered by modern photonics for applications such as broadband wireless communications, radar, sensing,
imaging and instrumentation. Digital electronics is the most widely used approach nowadays for these
applications. However, its speed is less than a few GHz, limited mainly by the electronic sampling rate.
The unique capabilities offered by modern photonics for the generation and processing of ultra-wideband,
high-frequency microwave signals make it a promising alternative for wideband microwave applications.
For the past four decades, numerous new techniques have been proposed and demonstrated. A few
articles reviewing the microwave photonics techniques have been published [1–4], but the emphasis of
these studies was placed on the implementation of the microwave photonics techniques based on discrete
components. In this article, a comprehensive review of microwave photonics techniques with their im-
plementation based on both discrete components and photonic integrated circuits (PICs) is performed.
Specifically, we first review the basic concepts of microwave photonics including the general architecture
of a microwave photonics link and its performance measures including the link gain, noise figure, and
spurious free dynamic range. Then, techniques to generate high frequency and low phase noise microwave
signals, including optical injection locking, optical phase lock loop, external modulation, and optoelec-
tronic oscillation, are presented. Techniques to process microwave signals, especially the implementation
of microwave photonic filters, are also reviewed. Microwave arbitrary waveform generation based on direct
space-to-time mapping, spectral shaping and wavelength-to-time mapping, and temporal pulse shaping is
also reviewed. Broadband phased array beamforming is an important application of microwave photonics
and is reviewed.
* Corresponding author (email: [email protected], [email protected])

c Science China Press and Springer-Verlag GmbH Germany, part of Springer Nature 2022 info.scichina.com link.springer.com
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:2

Figure 1 (Color online) General layout of a microwave photonic link.

Thanks to the fast advancement in PIC technology, all the microwave photonic techniques previously
demonstrated based on discrete components can be implemented using PICs. In this article, integrated
microwave photonics (IMWP) is also reviewed. First, the materials systems that can be employed for
the implementation of microwave photonic subsystems and systems including indium phosphide (InP),
silicon on insulator (SOI), silicon nitride (Si3 N4 ) and lithium niobate on insulator (LOI), are reviewed.
Then, recent progress in application-specific photonic integrated circuits, where a particular circuit con-
figuration to optimally perform a particular microwave photonics functionality, is discussed. General
purpose programmable microwave photonic signal processors that are capable of performing all the main
microwave signal generation and processing functionalities by suitable software programming are also
reviewed. The challenges in implementing fully integrated microwave photonics systems based on het-
erogeneous integration are discussed.

2 Basic concepts for microwave photonics

2.1 Microwave photonics links

A microwave photonic link (MWPL) is a basic building block of any system or network in charge of the
distribution of radiofrequency (RF) signals by optical means [5–12]. A general layout of an MWPL is
shown in Figure 1 and it will serve us to introduce some basic definitions and concepts.
The starting point is an input RF signal with its spectrum having two sidebands centered at frequency
±fRF , as shown in point 1. This signal, which can have its origin either from an RF generator or from
the reception by means of a single or an arrayed antenna, modulates the light from an optical source
to up-convert its spectrum to the optical region of the spectrum (point 2), such that the sidebands are
now centered at ν ± fRF , where ν represents the central frequency of the optical source. The combined
optical signal is then processed by an optical system composed of several photonic devices (including the
possibility of an optical fiber link) and characterized by an overall lumped optical field transfer function
H (ν). The MWPL modifies the spectral characteristics of the sidebands as illustrated in point 3. Finally,
a photodetector (PD) is employed to down-convert the sidebands again to the RF part of the spectrum by
suitable beating with the optical carrier, so the recovered RF signal, processed by the MWPL (as shown
in point 4), is ready to be sent to an RF receiver or to be re-radiated. The overall electrical transfer
function H (fRF ) is also shown and must not be confused with the optical field transfer function of the
auxiliary optical system.
MWPLs can range from very simple to very complex configurations depending, fundamentally, on the
nature of the auxiliary optical system transfer function H (ν). The simplest configuration, which histor-
ically was the first to be considered, corresponds to the case where the optical system is a dispersionless
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:3

Figure 2 (Color online) Diagram of a simple MWPL showing the relevant signal nomenclature at different points of the link and
the input (lower left hand-side)/output (lower right hand-side) signal spectra for an input RF tone.

optical fiber link. Though limited in its applications this configuration is quite adequate for introducing
the concepts related to the operation and figures of merit.

2.2 Figures of merit

Figure 2 shows the diagram of a simple MWPL [5], where the optical transmitter and the optical receiver
are connected by means of a lossy and dispersionless optical fiber link of a length of L. Fiber losses are
given by an attenuation coefficient α in dB/km.
If an input RF tone at frequency ΩRF is injected into the link as shown in the lower left hand-side
inset, the output RF signal, shown in the lower right hand-side inset, is generally composed of three
contributions: the desired input tone itself with a decreased amplitude due to the system losses, noise
spread over a given spectral region, and nonlinear harmonic distortion contributions appearing at nΩRF
n = 2, 3, . . . . These three effects characterize the quality of the MWPL in terms of three figures of merit
(FOMs): the RF link gain GRF , the noise figure NF, and the spurious free dynamic range SFDR. In very
simple terms, GRF accounts for the overall end-to-end system losses, NF accounts for the overall noise
level introduced by the system components, and the SFDR qualifies the impact of nonlinearities.
RF gain. Referring to the configuration of the MWPL in Figure 2, let pi represent the RF power of the
input tone at the frequency ΩRF injected to the MWPL. In the same way, po represents the RF power
of the output tone at frequency ΩRF extracted from the MWPL. The RF link gain in natural units is
defined as [5–7, 9]
po
GRF = . (1)
pi
It is convenient to re-express (1) in terms of some intermediate auxiliary values of the optical power at
the input and the output of the optical fiber link. If we define pi,o as the root mean square (RMS) power
magnitude of the time varying optical power at the link input and po,o as the RMS power magnitude of
the time varying optical power at the link output, then
αL
po,o = 10− 10 pi,o = TM-D pi,o , (2)

where TM-D represents the optical transmittance function from the optical transmitter to the optical
detector. Using the relationship given by (2) in (1), we get [6]
!
2
p
 
po i,o
GRF = × TM2 -D × . (3)
p2o,o pi

Eq. (3) seems in principle an unnecessary complication of the simple version given in (1). However, it
is useful in practice, as it is the product of three factors that separately characterize the performance of
the optical receiver, the optical link, and the optical transmitter, respectively.
The p2i,o /pi factor represents the equivalent input-output RF power transmission at the optical trans-
mitter. For an external Mach-Zehnder modulator (MZM) employed as shown in Figure 3, then the optical
power transmission versus input modulation voltage of the MZM is given by
  
pi,o αMZM πvM
TMZM = = 1 + cos , (4)
Po 2 Vπ
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:4

(a) (b)

(c)

Figure 3 (Color online) (a) Optical transmitter employing a continuous-wave (CW) optical source and a Mach-Zehnder modulator
(MZM) external modulator; (b) optical power vs. input voltage transfer characteristic for an MZM; (c) resistive impedance matching
circuit for an external modulator.

where Po represents the output power from the continuous-wave (CW) laser (input power to the mod-
ulator), αMZM is the MZM optical loss, and the modulation voltage is expressed as VT = VM + vM (t),
where VM is the bias voltage and vM (t) is the time varying RF signal. The modulation efficiency in
terms of output optical power vs. input electrical current for an MZM biased at the quadrature point
(VM = Vπ /2) is given by
     
dpi,o dpi,o dT (vM ) παMZM Po Rin
sm , = Rin = Rin Po = . (5)
diM IM dvM VM dvM VM 2Vπ
When using a resistively matched network between the voltage source and the modulator, as shown in
Figure 3(c), then it is straightforward to show by means of circuit analysis that
2
p2i,o s2

παMZM Po
≈ m = Rin . (6)
pi Rin 2Vπ
The po /p2o,o factor represents the equivalent input-output RF power transmission at the optical receiver.
Again, it is advisable to employ electrical circuits to match the output or load resistance to that of the
optical receiver. In that case it is possible to make the following approximation [5, 6]:
po
≈ ℜ2 Rout , (7)
p2o,o
where ℜ is the detector responsivity. Introducing (2), (6), (7) into (3), one gets
 2
πIdc
GRF = Rin Rout , (8)
Vπ
where the average received photocurrent is given by
αL
αMZM ℜPo 10− 10
Idc = . (9)
2
Noise figure. Figure 2 also shows the noise power at the input Nin and output Nout of the MWPL.
The link noise figure NF is defined as the quotient between the signal to noise ratio (SNR) at the input
and the SNR at the output, given by
   
pi /Nin Nout
NF = 10 log = 10 log , (10)
po /Nout GRF Nin
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:5

where we have made use of (1).

There are two important consequences derived from (10). First, if no noise is added by the link NF
= 0 dB, then we have Nout = GRF Nin . Second, NF does not depend on the RF signal power at the link
input. In practice, however, noise is added by several electronic and photonic components in the link
(laser, photodetector, electronic components). This implies that, in general, Nout = GRF Nin + Nadd , with
Nadd representing the power of additional noise sources at the system output. Hence, we have
   
GRF Nin + Nadd Nadd
NF = 10 log = 10 log 1 + . (11)
GRF Nin GRF Nin

Note that the effect of the additional noise can be referred to as the system input as an additional
noise source with the power given by Nadd /GRF . Furthermore, Eq. (11) can be generalized to include
noise sources appearing at intermediate points in the system. In this case, if Nadd,i represents the noise
power generated at the system output at point i and GRF,i represents the RF gain between the system
input and point i, then the overall contribution to the noise figure is Nadd,i /GRF,i , given by [6, 8]
 P
i (Nadd,i /GRF,i )

NF = 10 log 1 + . (12)
Nin

Depending on the system complexity, more or less noise sources will have to be considered. However,
the most important ones, i.e., those that have to be considered in any system, are the thermal noise
and shot noise at the receiver and the laser relative intensity (RIN) noise. For a system with an RF
bandwidth given by B, the thermal noise power is given by
2
No,th = 4kT Fn B = RINo,th Idc Rout B, (13)

where k is the Boltzman constant and Fn is a noise factor that takes into account the contribution of all
electronic components at the receiver. In (13), the noise power is defined as well in terms of an equivalent
relative intensity noise given by
4kT Fn
RINo,th = 2 . (14)
Rout Idc
The shot noise power is given by
2
Nshot = 2eIdc Rout B = RINshot Idc BRout , (15)

where e represents the charge of an electron. As with the thermal noise, the shot noise power is defined
as well in terms of an equivalent relative intensity noise given by
2e
RINshot = . (16)
Idc
Finally, the laser noise is directly defined through its relative intensity noise parameter
2
Nlaser = RINlaser Idc Rout B. (17)

The overall noise contribution can be represented by

2
Nout = No,th + Nshot + Nlaser = RINtot Idc Rout B, (18)

where RINtot = RINo,th + RINshot + RINlaser . In general, more contributions to the noise will have to be
considered for a more complex system including the noise terms from filters and amplifiers. A detailed
treatment of this is presented in [12].
Dynamic range. For a given RF gain, Eq. (1) implies that a higher output RF power will be achieved
by increasing the input RF power. Unfortunately, the input RF power cannot be increased at will since
nonlinear terms start to appear in the output RF spectrum that can distort the signal. The dynamic
range gives a measure of how much RF power can be injected into an MWPL before the effects of
nonlinear terms become comparable to the link noise. Figure 4 reproduces again the configuration of a
simple MWPL but this time the input and output RF signals are outlined, and an external modulation
configuration is considered.
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:6

Figure 4 (Color online) Simple MWPL configuration showing explicitly the input and output electronic signals.

Let us assume that it is possible to establish a relationship between the output current and the input
voltage by means of a function f [6],

iT,out = f [V ] = f [VM + vM (t)]

1 d2 f 1 d3 f
     
df 2
= f (VM ) + vM (t) + v (t) + v 3 (t) + · · ·
dV VM 2! dV 2 VM M 3! dV 3 VM M
2 3
= IB,out + f1 vM (t) + f2 vM (t) + f3 vM (t) + · · · . (19)

In general, the fi coefficients can be both time and frequency dependent but for the sake of simplicity
we will, however, assume no such dependence in the following discussion. An analysis of the harmonic
distortion due to nonlinearities can be performed if we assume a sinusoidal modulation for the input RF
voltage where the amplitude vs represents the RMS value delivered by the source,
√
vM (t) = 2vs cos ΩRF t. (20)

Substitution of (20) in (19) yields the following nonlinear harmonic distortion terms:
√ 2 1 √ 2
2
vRF = 2vs cos ΩRF t = 2vs [1 + cos (2ΩRF t)] ,
2
√ 3 1 √ 3

3
vRF = 2vs cos ΩRF t = 2vs [3 cos (ΩRF t) + cos (3ΩRF t)] , (21)
4
√ 4 1 √ 4

4
vRF = 2vs cos ΩRF t = 2vs [3 + 4 cos (2ΩRF t) + cos (4ΩRF t)] ,
8
···

Note from (21) and (19) that the fk term contributes to the kth harmonic distortion term at kΩRF t,
but also contributes to terms with frequencies ranging from (k − 2)ΩRF t to (k − 2m)ΩRF t terms with an
integer m and k−2m >0. Substituting (21) into (19), we get

f2 √ 2 6f4 √ 4
iT,out = IB,out + 2vs + 2vs + · · ·
  2  16
√ 3f3 √ 10f5 √ 5

  3
+ f1 2vs + 2vs + 2vs + · · · cos (ΩRF t)
4 16
f2 √ 2 4f4 √ 4
  
+ 2vs + 2vs + · · · cos (2ΩRF t)
2 8
f3 √ 3 5f5 √ 5
  
+ 2vs + 2vs + · · · cos (3ΩRF t) + · · · . (22)
4 16

The inspection of (22) reveals several interesting features. In the first place, the dependence between
the output RF current iT,out and the input RF voltage vs is only linear if the contributions from fi , i >1
can be neglected, otherwise harmonic distortion terms must be considered. However, in practice usually,
f1 > f2 > f3 > · · · , so within a reasonable range of vs the fundamental tone is much larger than the
second harmonic which in turn is larger than the third harmonic, etc. Thus, in practice, for the majority
of systems we need only to consider f1 , f2 , and f3 in the series expansion given in (21). Note that the
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:7

largest coefficient for the n-th harmonic contains a vsn term. Hence, the RF power of the n-th harmonic
increases at least n times faster than that of the fundamental. It is also interesting to point out that the
output current DC component is shifted from the IB,out value by a term proportional to f2 .
The total output RF power can be computed from (22) [5, 6],
ΩRF
R 2π/ΩRF
i2T ,out Rout dt =
P
po = 2π 0 pk ,
k
p1 = 4f12 Rout Rin pi ,
p2 = 8f22 Rout Rin
2 2
pi , (23)
p3 = 16f32 Rout Rin
3 3
pi ,
···
pk = 2k+1 fk2 Rout Rin
k k
pi .

That is, the total RF output power is the sum of the RF powers of the individual terms in (22) since
the cross-terms yield zero average:

ΩRF 2π/ΩRF
Z
cos (mΩRF t) cos (nΩRF t)dt = δm,n . (24)
2π 0
For the following discussion, it is useful to express the relationships given in (23) in logarithmic form.
Thus, we have

P1 (dBm) = Pi (dBm) + 10log10 4f12 Rout Rin ,

2
P2 (dBm) = 2Pi (dBm) + 10log10 8f22 Rout Rin


,
2 3


P3 (dBm) = 3Pi (dBm) + 10log10 16f3 Rout Rin , (25)
···
Pk (dBm) = kPi (dBm) + 10log10 2k+1 fk2 Rout Rik .

A similar procedure can be followed for the analysis of the intermodulation distortion. In this case,
one has to assume at least a two-tone RF input to the system,
√
vM (t) = 2vs [cos ΩRF,1 t + cos ΩRF,2 t] . (26)

If substituted in (10) taking up to the third order term shows the appearance of the second order terms
in 2Ω1 , 2Ω2 , Ω1 +Ω2 , Ω1 −Ω2 , and the third order terms in 3Ω1 , 3Ω2 , 2Ω1 +Ω2 , 2Ω1 −Ω2 , 2Ω2 +Ω1 , 2Ω2 −Ω1 .
In the so-called sub-octave applications where the bandwidth is less than one octave, the 2nd order
distortion terms fell out of the band and can be filtered out. Of the remaining terms, the sum terms
can be usually neglected so the limitation in the dynamic range comes from the terms in 2Ω1 − Ω2
and 2Ω2 − Ω1 which are usually known as the third order intermodulation distortion (IMD3) terms. A
straightforward but lengthy calculation renders for the amplitude of the cos(2Ω2 − Ω1 )t term the value
√
3
3f3 2vs /4 which is three times bigger than the corresponding third order harmonic distortion (HD3)
term given by (22). Thus, the power of the IMD3 terms is 9 times higher than that of the HD3 terms.
In general, the calculation of an analytic expression of f (V ) for an end-to-end link is not possible, but
a simplified analysis can be carried out by identifying the component that causes most of the nonlinear
distortion and treating the rest of the link as linear. In many cases, this component is the external
modulator. Alternatively, a numerical calculation for the end-to-end function f (V ) can be employed.
In any case, it is possible to represent the values of the output RF power values for the signal, harmonic,
and intermodulation distortion terms as a function of the RF power of the input signal to the system.
This is usually done in logarithmic form, as given by (25). Figure 5 shows a typical representation that
we will take as a reference in order to explain several important parameters related to the dynamic range.
In particular, we have represented the straight lines corresponding to the RF signal or fundamental, the
second order harmonic distortion (HD2), IMD3 and HD3. The lines show a different slope as expected
from (24) with a value of one for the fundamental term, two for HD2, and three for IMD3 and HD3. We
2
have also plotted the noise floor η=RINout Idc Rout (W/Hz) due to the contributions of all the relevant
noise sources.
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:8

Figure 5 (Color online) Logarithmic representation of the output RF power vs. input RF power for the signal (fundamental),
second and third order distortion terms in an MWPL.

For the region of input powers within points 1 and 2, the relevant distortion source is noise and
the system performance is given in terms of the signal to noise dynamic range or SNDR. As it can be
immediately observed this value is obtained at any point in this region as the difference between the
output RF power level corresponding to the fundamental and the noise floor level. SNDR is given in
dB·Hz as the noise floor is given per Hz of bandwidth. From this value and the knowledge of the system
RF bandwidth B it is straightforward to obtain the real SNR limited by noise:

SNR (dB) = SNDR (dB) − 10log10 ([B (Hz)]) . (27)

When the input RF power reaches the value given by point 2, then the second order nonlinear distortion
becomes the dominant factor over the noise. The input power margin between points 1 and 2 generates
an output signal free from the second order nonlinearities. This input power range is known as the second
order spurious free dynamic range (SFDR2 ). In a similar way, the input power margin between points
1 and 3 generates an output signal free from the third order nonlinearities; hence this power range is
known as the third order spurious free dynamic range (SFDR3 ).
The dependence of SFDR2 on the system bandwidth can be computed as follows referring to Figure 5.
From (25) we know that

P1 (dBm) = Pi (dBm) + 10log10 4f12 Rout Rin → p1 = c1 pi ,

P2 (dBm) = 2Pi (dBm) + 10log10 8f22 Rout Rin 2

→ p2 = c2 p2i .


(28)

But at point 2, the RF power for HD2 equals that of the noise per unit bandwidth, we have
r
η
p2 = c2 p2i = η → pi = . (29)
c2

When substituted into the first expression in (28), we have

s s
c21 η p1 c21
p 1 = c1 p i = → SFDR2 = = . (30)
c2 η c2 η

Since p1 is in W, then the units of SFDR2 are Hz1/2 or dB·Hz1/2 if we express it in logarithmic form.
To distinguish the fact that this quantity is for a 1 Hz bandwidth we will name it SFDR1Hz 2 . If the
second order nonlinearities are the limiting factor, then SFDR2 for the system bandwidth of B Hz can
be computed by
SFDR2 (dB) = SFDR1Hz 2 (dB) − 5log10 ([B (Hz)]) . (31)
A similar procedure can be employed to obtain the system dependence of SFDR1Hz
3 on the system
bandwidth. Again, from (25) we know

p 1 = c1 p i ,
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:9

Figure 6 (Color online) The second and third order optical interception points for the distortion terms in an MWPL.

p3 = c3 p3i . (32)

According to (1) c1 = GRF . But at point 3 the RF power for IMD3 equals that of the noise per unit
bandwidth, we have r
η
p3 = c3 p3i = η → pi = 3 . (33)
c3
When substituted into the first expression in (32), we have
s s
3 c3
3 c1 η p 1
p 1 = c1 p i = → SFDR1Hz
3 = = 3 12 . (34)
c3 η c3 η

Since p1 is in W, then the units of SFDR1Hz

3 are Hz2/3 or dB·Hz2/3 if we express it in logarithmic form.
If the third order nonlinearities are the limiting factor (for example in a sub-octave system), then the
SFDR3 for a system bandwidth B can be computed by
20
SFDR3 (dB) = SFDR1Hz (dB) − log10 ([B (Hz)]) . (35)
3
3
In practice, the values of SFDR2 and SFDR3 are usually computed through the so-called interception
points, which are shown in Figure 6.
The interception points (IIP2 , OIP2 ) and (IIP3 , OIP3 ) are the intersection between the lines represent-
ing the fundamental and those representing the second and third order distortions, respectively. These
points can be easily obtained after experimental measurements. Referring to Figure 6 and (25),
)
OIP2 = c1 IP2
⇒ IP2 = cc12 ,
OIP2 = c2 IP22
) (36)
OIP3 = c1 IP3 q
c1
⇒ IP3 = c3 .
OIP3 = c3 IP33

When substituted in (30) and (34), we have

s s   s s
1Hz c21 c1 GRF IP2 OIP2
SFDR2 = = IP2 = = ,
ηc2 η η η
s s r
GRF IP2 OIP2 OIP2
SFDR2 = = = , (37)
ηB ηB Nout

s ss 2/3
2 2
c312

1Hz 3 3 c 1 2 3 GRF IP3 OIP3
SFDR3 = =
IP = = ,
η2 3
η 2 c3 η2 η
s 2/3  2/3
2 2 
3 GRF IP3 OIP3 OIP3
SFDR3 = = = . (38)
η2 B 2 ηB Nout
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:10

Or, in logarithmic form,

IP2 (dBm) + GRF (dB) − Nout (dBm)
SDFR2 (dB) =
2
OIP2 (dBm) − Nout (dBm)
= , (39)
2

2
SDFR3 (dB) = [IP3 (dBm) + GRF (dB) − Nout (dBm)]
3
2
= [OIP3 (dBm) − Nout (dBm)] . (40)
3
In summary, in this section, we have reviewed the basic concepts of MWPLs including the general
architecture and the performance measures including the link gain, noise figure, and the spurious free
dynamic ranges for both the second order harmonic distortions and the third order intermodulation
distortions. The same analysis can also be applied to microwave photonic filters since, ideally, a microwave
photonic filter is a linear system which is expected to have a large dynamic range. The implementation
of a microwave photonic filter, however, involves an optical modulator, which is inherently a nonlinear
device, which would make the dynamic range decrease. The techniques developed for MWPLs can be
applied to increase the dynamic range of microwave photonic filters.

3 Photonic-assisted microwave signal generation

The primary motivation of generating microwave signals in the optical domain based on photonics is to
generate high frequency, frequency tunable, and low phase noise microwave signals which cannot be easily
generated based on pure electronics. Numerous approaches have been proposed for the last few decades.
In general, all these approaches can be classified into two categories, microwave generation based on
heterodyne beating of two optical waves, and microwave generation based on optoelectronic oscillation.

3.1 Microwave generation based on heterodyne beating

The fundamental concept behind the generation of microwave signals based on photonics is to beat
two optical waves at a photodetector. If the two wavelengths are phase correlated, a low-phase noise
heterodyne beat signal with its frequency corresponding to the wavelength spacing is generated. The
key is to generate two wavelengths that are phase correlated, to ensure low-phase noise microwave signal
generation. In general, there are three main approaches that are employed to generate two phase-
correlated wavelengths, to beat to generate a high frequency and low phase noise microwave signal. These
techniques include optical injection locking [13–18], optical phase locked loop [19–27], optical external
modulation [28–37], and optoelectronic oscillation [38–52].
Figure 7 shows an injection-locking-based microwave signal generation system [13–18]. As can be seen,
a laser diode is directly modulated by a low-frequency microwave modulation signal. Due to frequency
modulation at the laser diode, a frequency-modulated optical signal with multiple sidebands is generated.
By injecting the frequency-modulated optical signal into two slave laser diodes, with one sideband to
injection lock one laser diode, two wavelengths corresponding to the two sidebands are selected. Since
the two wavelengths are generated due to direct modulation at the master laser diode with the same
microwave source, the two wavelengths are phase correlated. The beating of the two wavelengths at
a photodetector will generate a low phase noise microwave signal. Figure 7 shows an example of an
injection locking system for microwave signal generation. As can be seen, a low-frequency microwave
modulation signal is applied to the master laser diode to generate a frequency-modulated optical signal.
If the two slave laser diodes have free running wavelengths close to those of the ±2nd sidebands, when
the frequency-modulated optical signal is injected into the two slave laser diodes, the wavelengths from
the slave laser diodes are injection locked by the ±2nd sidebands. By beating the two wavelengths at
the photodetector, a microwave signal with its frequency being 4 times the frequency of the microwave
modulation signal is generated. Note that the system in Figure 7 can not only generate a high frequency
and low phase noise microwave signal, but also distribute the microwave signal by transmitting the two
wavelengths over a single-mode fiber (SMF). By beating the two wavelengths remotely, a microwave
signal at a remote location can be generated.
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:11

FM master laser

λ
λ0 λ0

RF

Slave LD1
4fm

Master LD PD
SMF
Isolator
Slave LD2

Figure 7 (Color online) Optical injection locking of two wavelengths for microwave signal generation and transmission. The
master laser diode is directly modulated by a microwave modulation signal with its output injected into the two slave laser diodes.
The slave laser diodes are injection-locked by the ±2nd sidebands from the output of the master laser diode. The beat at the
photodetector will generate a microwave signal with a frequency that is 4 times that of the microwave modulation signal.

Loop control

LD1
Mixer Amplifier

Lowpass
PD
filter

LD2 SMF

RF

Figure 8 (Color online) Optical phase lock loop to phase lock two optical wavelengths for microwave generation and transmission.
LD: laser diode. PD: photodetector.

A limitation of the approach is that once the salve laser diodes are selected, the free running wavelengths
are fixed; thus the frequency of the generated microwave signal is not tunable or has a very limited
frequency tunable range.
A second approach to generate two phase-correlated wavelengths and by beating the two wavelengths to
generate a low phase noise microwave signal is to use an optical phase lock loop [19–27]. Assume two free
running laser diodes have a wavelength spacing corresponding to a frequency fm . Since the wavelengths
are not phase locked, the phase noise terms from the two laser diodes after beating at a photodetector
will be translated to the generated microwave signal. To make the two wavelengths phase correlated, an
optical phase lock loop can be employed. As can be seen from Figure 8, the two wavelengths from the
two laser diodes are applied to a photodetector to generate a beat note. The beat signal is mixed with
a microwave reference signal. After low-pass filtering, a current or voltage signal that is proportional
to the phase difference between the two microwave signals is generated, which is fed back to one laser
diode to control its phase to be identical to that of another laser diode. If the loop length is sufficiently
short, the two wavelengths can be effectively phase locked. Again, the microwave signal can be generated
remotely. If the two phase-locked wavelengths are transmitted over a fiber and beat at a photodetector,
then a microwave signal at a remote location can be generated.
Again, the limitation of the approach is that the frequency has limited tunability since the wavelengths
of the laser diodes are fixed or have limited tunability. A solution is to use the external modulation
approach in which the frequency of the generated microwave signal can be widely tunable.
A third approach to generate two phase-correlated wavelengths and by beating the two wavelengths
to generate a low phase noise microwave signal is to use external modulation [28–37]. Figure 9 shows
the approach to generating a microwave signal based on optical external modulation [31, 34, 37]. The key
advantage of this approach is that the frequency of the generated microwave signal can be continuously
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:12

2fm +2 -2

λ λ λ
λ0 λ0 λ0

FBG
4fm
LD MZM PD
SMF
PC
DC
bias
RF
λ
λ0

Figure 9 (Color online) Microwave signal generation based on external modulation using an MZM and a wavelength-fixed optical
filter, such as an FBG. MZM: Mach-Zehnder modulator; FBG: fiber Bragg grating; PD: photodetector.

tunable. As shown in Figure 9, a linearly polarized light wave from a laser diode is applied to an MZM
through a polarization controller. The polarization controller is employed to control the polarization
direction of the light wave from the laser diode to be aligned with the principal axis of the MZM to
minimize the polarization-dependent loss. A low frequency microwave modulation signal is applied to
the MZM via its RF port to modulate the light wave. By biasing the MZM at the minimum transmission
point (MITP), the odd order sidebands are fully suppressed. By using an optical notch filter with its
center wavelength located at the wavelength of the optical carrier, such as a fiber Bragg grating (FBG),
the optical carrier is filtered out, and only the two ±2nd sidebands are obtained (the ±4th and higher
even order sidebands are very weak and are ignored). By beating the two sidebands at a photodetector,
a microwave signal with a frequency that is four times the frequency of the microwave modulation signal
is generated. Again, by transmitting the two ±2nd sidebands over a fiber and beating the two sidebands
remotely at a photodetector, a microwave signal at a remote location is generated.
Note that the MZM can be replaced by a phase modulator (PM). The advantage of using a PM is
that a PM is not biased; thus the bias drifting problem does not exist [32]. In addition, one may use two
cascaded MZMs, to generate a microwave signal with a higher frequency [33]. For example, a microwave
signal with a frequency multiplication factor up to 12 can be generated [34]. The use of this approach
to generate a sub-THz signal was also reported [35, 36]. Phase-noise analysis of a microwave signal with
external modulation was also studied theoretically and experimentally [37].
All the three approaches based on heterodyne beating of two wavelengths to generate a high frequency
and low-phase noise microwave signal need a microwave reference source. The phase noise of the gen-
erated microwave signal is directly related to the phase noise of the reference microwave source. If the
frequency multiplication factor is N , the phase noise performance of the generated microwave signal will
be deteriorated by 20log10 N (dB) [38]. For example, if N is 4, then the phase noise performance is
reduced by 12 dB. To generate a microwave signal with a very low phase noise, a microwave reference
source with an ultra-low phase noise is needed, which makes the implementation highly costly. On the
other hand, we may improve the phase noise performance through frequency division. For example, if the
frequency of a microwave signal is divided by 4, then the phase noise performance is increased by 12 dB.
The approach was employed to generate an ultrastable microwave signal with its stability comparable to
the best microwave oscillators, but without the need for cryogenic temperature control [39].

3.2 Microwave generation based on an optoelectronic oscillator

A solution to generate a low phase noise microwave signal without the need for a low-phase noise mi-
crowave reference source is to use an optoelectronic oscillator (OEO) [40–52]. An OEO is an optical and
electronic hybrid system with an amplified feedback loop consisting of an optical path and an electrical
path. For the optical path, the length can be made long thanks to the ultra-low loss of the state-of-the-art
optical fibers; thus the Q factor of the feedback loop can be high, making the generated microwave signal
have ultra-low phase noise. This is the key and unique feature of OEOs for low-phase noise microwave
generation. Figure 10 shows the configuration of an OEO, which consists of a laser diode, a polarization
controller, a Mach-Zehnder modulator, a length of single mode fiber, a microwave bandpass filter, an
electronic amplifier (EA) to provide the loop gain, and a photodetector. A light wave from the laser
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:13

LD MZM PD
SMF
PC

DC bias
Optical path

Electrical path
Bandpass
EA
Output filter
EC

Figure 10 (Color online) An optoelectronic oscillator. LD: laser diode; MZM: Mach-Zehnder modulator; SMF: single mode fiber;
PD: photodetector; EA: electronic amplifier.

diode is sent to the Mach-Zehnder modulator through the polarization controller. Again, the polariza-
tion controller is used to control the polarization state of the incident light wave to the Mach-Zehnder
modulator, to minimize the polarization-dependent loss. The modulated light wave is then sent to the
photodetector after passing through the single mode fiber. At the output of the photodetector, the op-
tical signal is converted to a microwave signal, which is then filtered by the microwave bandpass filter,
amplified by the electronic amplifier and sent back to the Mach-Zehnder modulator, to close the feedback
loop. If the gain provided by the electronic amplifier is sufficiently large to make the loop gain greater
than the loop loss, oscillation will start. The frequency of the generated microwave signal is equal to the
center frequency of the passband of the microwave bandpass filter.
Since the frequency tunability of an electronic microwave filter is very limited, one may use a microwave
photonic filter to perform frequency tuning and mode selection. A microwave photonic filter can have a
large frequency tunable range as compared to an electronic microwave filter. Figure 11 shows a frequency-
tunable optoelectronic oscillator in which a phase-shifted fiber Bragg grating (PS-FBG) is employed to
implement, jointly with a phase modulator and a photodetector, a frequency-tunable microwave photonic
filter [46]. As can be seen, the light wave from a tunable laser source (TLS) is directed to a phase
modulator, where a double-sideband plus carrier (DSB+C) optical signal is generated. Note the DSB+C
signal generated using a phase modulator is different from a DSB+C generated using a Mach-Zehnder
modulator. A DSB+C signal generated by a phase modulator has two first sidebands that are out of
phase. Thus, by applying a phase-modulated DSB+C signal to a photodetector, no microwave signal will
be generated except a DC, since the beating between one sideband with the optical carrier will generate
a microwave signal that will be fully canceled by another microwave signal resulting from the beating
between the other sideband with the optical carrier. To convert a phase-modulated DSB+C signal to an
intensity-modulated signal, a simple way is to remove one sideband. As can be seen, the DSB+C signal is
sent to a phase-shifted fiber Bragg grating where one sideband is removed due to the notch of the phase-
shifted fiber Bragg grating, thus the phase-modulated DSB+C signal is converted to a single-sideband
plus carrier (SSB+C) signal. The detection of the SSB+C at a photodetector will generate a microwave
signal. The central frequency of the microwave photonic filter is equal to the wavelength spacing between
the notch wavelength and the wavelength of the optical carrier. Thus, by tuning either the wavelength of
the optical carrier from the tunable laser source or the notch wavelength, the frequency of the microwave
filter can be tuned.
The key challenge in implementing an optoelectronic oscillator using a long fiber is the mode selection
difficulty. An optoelectronic oscillator with a long feedback loop has a large number of closely spaced
longitudinal modes separated by its free spectral range (FSR) within the passband of the microwave or
microwave photonic bandpass filter. To enable single mode oscillation, a microwave or microwave photonic
filter with an ultra-narrow passband must be used to select a single mode, which is very challenging
considering the small free spectral range of a long optoelectronic oscillator loop. The use of dual or
multiple feedback loops may make an optoelectronic oscillator have a large FSR due to the Vernier
effect [42], but the implementation is more complicated and the frequency tunability is limited.
Instead of using a long feedback loop to increase the Q factor, one may use a high Q factor optical
resonator to serve as a mode selection component. For example, OEwaves has demonstrated a miniatur-
ized optoelectronic oscillator in which the long fiber was replaced by a high Q factor whispering gallery
mode (WGM) resonator [47]. OEwaves has developed mm-scale whispering gallery mode resonators with
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:14

Optical path

Electrical path PS-FBG

λ
λ0
2
1
TLS PM PD
3
PC OC

EC
Output EA
λ

Figure 11 (Color online) A frequency tunable OEO. The oscillation frequency of the OEO is determined by the central frequency
of the microwave photonic filter. TLS: tunable laser source; PC: polarization controller; PM: phase modulator; PD: photodetector;
EA: electronic amplifier; PS-FBG: phase-shifted fiber Bragg grating.

Figure 12 (Color online) Miniature optoelectronic oscillator incorporating a lithium niobate whispering gallery mode (WGM)
resonator (from OEwaves [47]).

the Q factor and finesse in excess of 1011 and 107 , which are beyond what is possible with bulk or inte-
grated optical resonators. By using a whispering gallery mode resonator, a microwave signal with a phase
noise as low as −110 dBc/Hz at 10 kHz offset was generated. Figure 12 shows an OEwaves miniature
optoelectronic oscillator incorporating a lithium niobate WGM resonator in the optoelectronic oscillator
loop. In addition to the ease of mode selection, the use of a whispering gallery mode resonator can also
significantly reduce the size of the optoelectronic oscillator, making an optoelectronic oscillator to have
a small form factor, a feature that is highly needed for airborne applications.
The use of a WGM resonator can solve the mode selection problem, but the implementation is still
challenging especially the coupling between a fiber and a whispering gallery mode resonator, making the
optoelectronic oscillator highly costly. In addition, the spectral response of a whispering gallery mode
resonator is fixed or with very limited tunability, thus the frequency of the generated microwave signal
by a whispering gallery mode resonator based optoelectronic oscillator has no or very limited frequency
tunability.
Recently, a new concept, called parity time symmetry, has been introduced to the microwave photonics
field [48, 49]. A parity-time (PT) symmetric system is a special non-Hermitian system in which its
Hamiltonian possesses real eigenvalues [50–59]. The inherent mode selection feature has made the concept
extremely useful for mode selection in a laser or an optoelectronic oscillator. The key advantage of using
parity time symmetry in an optoelectronic oscillator is that high performance mode selection can be
implemented, making stable single-mode oscillation possible while maintaining a high Q factor to ensure
an ultra-low phase noise.
Figure 13 shows a parity-time symmetric optoelectronic oscillator in which two optoelectronic loops
have identical geometry but with one having a gain coefficient and the other a loss coefficient, identical in
magnitude, are implemented [48]. Once the gain/loss coefficient is greater than the coupling coefficient,
the PT symmetry condition is broken, and the mode that has the highest gain is selected.
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:15

SMF

PC1 MZM PC2 PBS TDL BPD

PD1

LD

Σ
1 −
Signal analyzer +
S21
2 EC PD2

ESA EA

Figure 13 (Color online) Diagram of a PT-symmetric optoelectronic oscillator. LD: laser diode; PC: polarization controller; MZM:
Mach-Zehnder modulator; SMF: single-mode fiber; PBS: polarization beam splitter; TDL: tunable delay line; PD: photodetector;
BPD: balanced photodetector; Σ: microwave combiner; EA: electrical amplifier; ESA: electrical spectrum analyzer [48].

Mathematically, the coupled mode equations relating to the modes in the two loops are given by

dan
= −iωn an + iκbn + γan an , (41)
dt
dbn
= −iωn bn + iκan + γbn bn , (42)
dt
where an and bn are the amplitudes of the n-th modes in the gain and loss loops, respectively, ωn is
the eigenfrequency of the longitudinal modes of the two loops without PT symmetry coupling, κ is the
coupling coefficient between the two loops, and γan and γbn are the gain coefficients of the gain and loss
loops for the n-th mode. By solving (41) and (42), we can get the eigenfrequencies of the PT symmetric
system, given by s  2
γ a + γ bn γan − γbn
ωn(1,2) = ωn + i n ± κ2n − . (43)
2 2
Assuming that the exact PT symmetry condition is satisfied for one longitudinal mode, say n = 0, i.e.,
γa0 = −γb0 = γ0 , Eq. (43) can be written as
q
(1,2)
ω0 = ω0 ± κ20 − γ02 . (44)

It can be seen from (44) that, when the gain/loss coefficient is smaller than the coupling coefficient,
the two loops exhibit broken degeneracy with eigenfrequency splitting. However, when the gain/loss
coefficient is greater than the coupling coefficient, the eigenfrequency becomes a complex number with a
non-zero imaginary part. The electrical field can then be written as
(1,2)
E0 = exp (jω0 t) exp (± |g0 | t) , (45)
p
where g0 = κ20 − γ02 . As can be seen from (45), parity-time symmetry is broken when the gain/loss
coefficient is greater than the coupling coefficient. Amplifying and decaying modes with an identical
frequency will be generated, and the amplifying mode would be the dominating mode that oscillates in
the optoelectronic oscillator.
Stable single-mode oscillation supported by broken parity-time symmetry without the need of an ultra
narrowband optical or microwave filter was evaluated experimentally based on the system shown in
Figure 13 and the experimental results are shown in Figure 14 [48]. The two loops have an identical loop
length of 9.166 km. Clearly, the loops are very long and a large number of closely spaced longitudinal
modes exist in the gain spectrum of the optoelectronic loop. Without broken parity-time symmetry, the
optoelectronic oscillator will oscillate in multimode. Figure 14(a) shows the spectrum of the microwave
signal generated by the optoelectronic oscillator. It is clearly seen multiple mode oscillation is resulted.
To achieve single-mode oscillation, enabled by the broken PT symmetry, the gain and loss in the two
loops are tuned to be balanced and to be greater than the coupling coefficient, which is done by tuning the
polarization controller before the polarization beam splitter (PBS). Figures 14(b)–(d) show the spectra
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:16

10 10
(a) (b)

−10 −10

Power (dBm)
Power (dBm)

−30 −30

−50 −50

−70 −70
−50 −30 −10 10 30 50 −50 −30 −10 10 30 50
Offset frequency (MHz) Offset frequency (MHz)
10
(c) 0 (d)

−10 −20
Power (dBm)

Power (dBm)
−30 −40

−60
−50
−80

−70 −100
−500 −300 −100 100 300 500 −500 −300 −100 100 300 500
Offset frequency (kHz) Offset frequency (Hz)

Figure 14 (Color online) The spectra of the microwave signals generated by the PT-symmetric optoelectronic oscillator. The
spectra are measured at a central frequency of 9.867 GHz. (a) Multi-mode oscillation measured with an RBW of 3 MHz;
(b) single-mode oscillation measured with an RBW of 3 MHz; (c) single-mode oscillation measured with an RBW of 9.1 kHz;
(d) single-mode oscillation measured with an RBW of 9 Hz [48].

of the generated microwave signal measured at different resolution bandwidths (RBWs). Clearly, single-
mode oscillation is achieved. Figure 14(b) shows the spectrum with a 100 MHz span and Figure 14(c)
shows the spectrum with a 1 MHz span. The oscillating mode can be identified, and the mode spacing
between the oscillating mode and the adjacent sidemode is measured to be 22.73 kHz and the oscillating
mode is 26.4 dB higher than the highest sidemode. No mode hopping is observed in the experiment for
a measurement time window of over ten minutes. Further reducing the frequency span to 1 kHz, the
spectrum of the microwave signal is shown in Figure 14(d). It can be seen the microwave signal has an
extremely narrow bandwidth thanks to the high Q factor of the optoelectronic loop due to the use of a
long fiber in the optoelectronic oscillator.
The configuration shown in Figure 13 has two physically separated loops. To achieve broken parity-time
symmetry, the lengths of the two loops must be made identical and the gain and loss must be balanced,
which makes the implementation difficult. In addition, the stability is affected due to the use of two
physically separated loops. The PT symmetry can also be implemented in a single physical resonator or
loop [54, 60–62]. By using different multiplexing techniques, two equivalent loops with balanced gain and
loss can be implemented. Figure 15 shows a configuration using a polarization-dependent Sagnac loop
to implement two equivalent loops [61]. As can be seen, the Sagnac loop, consisting of a PBS and two
polarization controllers, is employed to function equivalently to two mutually coupled feedback loops,
with the gain, loss, and coupling coefficient being independently controllable by tuning the polarization
controllers. Since only a single physical loop is used, the implementation is greatly simplified, and the
stability is highly improved. The same concept has been employed to implement a single mode fiber
ring laser, in which a single physical loop was employed to operate equivalently as two mutually coupled
loops [63].
In addition to single-frequency and low phase-noise microwave signal generation, the generation of
frequency-chirped microwave signals is also a topic of interest and can find applications in spread spec-
trum communications, radar, and microwave imaging. A review on the photonic generation of chirped
microwave waveform has recently been published [64]. In general, a chirped microwave waveform can
be generated based on spectral shaping and wavelength-to-time (SS-WTT) mapping, based on external
modulation, and Fourier-domain mode locking [65]. Here a discussion on Fourier-domain mode locking
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:17

TLS PC2
SMF
PC1 PM Cir.1
1 2 OC
PBS
P
3

ESA 1
2 PC3
Cir.2
3 PS-FBG

EDFA

Div.
EA PD

Figure 15 (Color online) A PT-symmetric optoelectronic oscillator with a single physical loop. TLS: tunable laser source; PC:
polarization controller; PM: phase modulator; SMF: single-mode fiber; Cir.: circulator; OC: optical coupler; PBS: polarization
beam splitter; PS-FBG: phase-shifted fiber Bragg grating; EDFA: erbium-doped fiber amplifier; PD: photodetector; EA: electrical
amplifier; Div.: divider; ESA: electrical spectrum analyzer [62].

Cavity time delay

f

Electronic Frequency-
amplifier scanning BPF
EPS
RF
output
PD

CW

Long fiber laser

MZM PC

Optical Electrical

Figure 16 (Color online) A Fourier domain mode locked OEO for generating a chirped microwave waveform. EPS: electronic
power splitter; MZM: Mach-Zehnder modulator; PC: polarization controller [64].

of an optoelectronic oscillator for chirped microwave signal generation is discussed.

A frequency chirped microwave waveform can also be generated by a frequency tunable optoelectronic
oscillator through fast tuning the central frequency of a microwave bandpass filter in the loop, but the
phase noise performance is poor due to the fact that the optoelectronic oscillator, when tuned from
one frequency to another, will need a buildup time to stabilize the oscillation at the new frequency. A
solution is to use Fourier domain mode locking. In a Fourier domain mode locked (FDML) OEO, all
modes co-exist and are phase locked in the OEO loop. The switching from one mode to another mode
does not need a buildup time, thus a chirped microwave waveform generated by a Fourier domain mode
locked optoelectronic oscillator will not increase the low phase noise [65].
Figure 16 shows the schematic of a Fourier domain mode locked optoelectronic oscillator. It is similar
to a regular optoelectronic oscillator except that the microwave bandpass filter is frequency tunable and
the tuning is controlled by a sawtooth waveform with its repetition rate to be identical to the free spectral
range of the optoelectronic loop or one of its multiples. The use of a Fourier domain mode locked laser
to generate a linearly chirped optical waveform, and by beating the linearly chirped optical waveform
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:18

with an optical carrier, a linearly chirped microwave waveform with a much broader bandwidth could be
generated [66].
In summary, microwave signal generation can be implemented either by heterodyne beating two phase-
correlated wavelengths or using an optoelectronic oscillator. Since the use of an optoelectronic oscillator
does not need a reference microwave source, it is an approach that is more suitable for applications where
no reference microwave source is available. Similar to the heterodyne beating approaches where the mi-
crowave signal can be transmitted over a long fiber and generated at a remote location, an optoelectronic
oscillator can also generate a microwave signal remotely. By taping part of the optical signal at the input
of the photodetector, distributing the optical signal over a fiber, and detecting the signal at a second
photodetector, a microwave signal can be generated at a remote location.

4 Photonic-assisted microwave signal processing

Microwave signals can also be processed in the optical domain to take advantage of the high frequency
and wide bandwidth offered by modern photonics. Among the numerous signal processing functions,
microwave filtering is one of the major functions and has been investigated for the last few years.
Microwave filters can be implemented in the optical domain using photonics to support operations
at high frequency with a wide frequency tunable range [1, 67–69]. The implementations of microwave
photonic filters can be, in general, classified into two categories: incoherent and coherent implementations.
In general, an incoherent microwave photonic filter has a delay-line configuration with a finite impulse
response (FIR) or infinite impulse response (IIR). In the configuration, a microwave signal is modulated
on an optical carrier or optical carriers. The time delays and tap coefficients are produced in the optical
domain and the time-delayed and weighted optical signals are detected at a photodetector. To avoid
optical interferences, which are extremely sensitive to environmental changes, an incoherent light source
that enables an incoherent combination of the time-delayed optical signals at a photodetector is needed.
The limitation of an incoherent microwave photonic filter is that the weights or tap coefficients are all
positive, making a microwave photonic filter have a low-pass response only, or special designs have to be
employed to produce negative [70–72] or complex [73,74] coefficients. The use of nonuniformly spaced taps
can also produce a delay line filter with equivalent negative or complex coefficients. A comprehensive study
about the design and implementation of nonuniformly spaced microwave photonic filters was reported
in [75]. Microwave photonic filters in the second category are implemented using an ultra-narrow band
optical filter, to translate its frequency response in the optical domain to the microwave domain. Thus, the
frequency response of the microwave filter is uniquely determined by the spectral response of the optical
filter [76–78]. Instead of using an incoherent light source, a single-frequency light source is employed.
The microwave photonic filters in the second category are also called coherent microwave photonic filters.
Since the microwave photonic filters do not have a delay line configuration, optical interference is not an
issue that would affect the stable operation of the filters.
Considering that microwave photonic filters based on a delay-line structure have been heavily studied
and a few review or tutorial articles have been published [1,67–69], here, only coherent microwave photonic
filters that translate the spectral response of an optical filter from the optical domain to the microwave
domain will be discussed.
A general structure of a coherent microwave photonic filter is shown in Figure 17. An optical carrier
from a laser source is sent to a phase modulator. Assume small signal modulation, at the output of the
phase modulator, a DSB+C signal is generated. Note that it is different from intensity modulation using
a Mach-Zehnder modulator to generate a DSB+C signal with the two sidebands being in phase, the two
first sidebands of a phase modulated signal are out of phase. Thus, the detection of a phase-modulated
DSB+C signal at a photodetector will not generate a microwave signal except a DC since the beating
between the optical carrier and the lower sideband will cancel completely the beating between the optical
carrier and the upper sideband. However, if one of the sidebands is removed by an optical notch filter
in transmission or using a dual passband filter in reflection [76], such as a fiber Bragg grating or two
cascaded fiber Bragg gratings, then the phase-modulated signal is converted to an intensity-modulated
SSB+C signal and the detection of the signal at a photodetector will generate a microwave signal.
Figure 17 shows the schematic of a coherent microwave photonic filter, in which an optical notch filter,
such as a fiber Bragg grating in transmission, is used to filter out one sideband of a phase-modulated
signal, thus achieving phase-modulation to intensity-modulation conversion. As can be seen, the entire
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:19

Input Output
PC
Optical
LD PM filter PD

Spectrum of the
optical filter

λ λ
λ0 λ λ0
λ0

Figure 17 (Color online) Schematic of a coherent microwave photonic filter, in which an optical notch filter is used to filter out
one sideband of a phase-modulated signal, thus achieving phase-modulation to intensity-modulation conversion.

operation is equivalent to a microwave filter with the bandwidth determined by the bandwidth of the notch
of the optical notch filter and the center frequency is determined by the wavelength difference between
the optical carrier and the center wavelength of the notch. Thus, by simply tuning the center frequency of
the notch filter or the wavelength of the optical carrier, the center frequency of the microwave bandpass
filter can be tuned. When tuning, the spectral shape is maintained unchanged. This is an advantage as
compared with an incoherent microwave photonic filter, where the spectral shape is changing when the
filter is tuned, or complex coefficients must be used to avoid such a spectral shape change.
Instead of using an optical filter in transmission, one may use two cascaded optical filters such as two
fiber Bragg gratings in reflection, by using one fiber Bragg grating to select the optical carrier and the
other fiber Bragg grating to select one sideband, and a phase-modulated signal is converted to a single-
sideband intensity-modulated signal [76]. A microwave photonic filter with the spectral shape identical
to the notch of the second fiber Bragg grating is generated. The major limitation of the technique
in [76] is the wide passband since two uniform fiber Bragg gratings were employed. The passband of the
microwave photonic filter was determined by the bandwidth of one of the two fiber Bragg gratings to
select the sideband. By using a high Q-factor ring resonator [77], the passband can be smaller, but still
too large for most applications.
Due to the ultra-narrow bandwidth of the stimulated Brillouin scattering (SBS) gain or loss spectrum,
a microwave photonic filter with a narrow passband can be implemented based on phase modulation and
phase-modulation to intensity-modulation conversion in an SBS gain or loss spectrum. Figure 18 shows
a narrowband microwave photonic filter based on SBS [78]. A light wave from a laser source at ω0 is split
into two paths. At the upper path, the light is modulated by a local oscillator (LO) signal at a dual-
parallel MZM (DP-MZM) to generate a single-sideband with suppressed carrier (SSB-SC) signal. Assume
that the frequency of the local oscillator signal is ωLO , the generated lower sideband is at ω0 − ωLO . The
SSB-SC signal is sent to an erbium-doped fiber amplifier to increase its power and then launched into a
long single-mode fiber to stimulate the SBS effect. The SBS gain and loss are located at ωO − ωLO − ωB
and ωO − ωLO + ωB , respectively. If one sideband of a phase-modulated signal is amplified by the SBS
gain, then the phase-modulated signal is converted to an intensity-modulated signal, and the detection of
the intensity-modulated signal at a photodetector will generate a microwave signal at ωLO − ωB , as shown
in Figure 18. The entire operation of the system is equivalent to a microwave bandpass filter with the
passband determined by the SBS gain spectrum. The advantage of this approach is that the filter can be
continuously tunable by tuning the LO frequency and the passband is ultra-narrow. However, compared
with the use of a fiber Bragg grating or a ring resonator, the system is very complicated, especially a
high power erbium-doped fiber amplifier and a long single-mode fiber are needed.
To have an ultra-narrow passband while maintaining a simple system, a solution is to use an optical
filter with an ultra-narrow passband. Figure 19 shows the implementation of a coherent microwave
photonic filter using a phase-shifted fiber Bragg grating (PS-FBG) [79]. A phase-shifted fiber Bragg
grating is a special fiber Bragg grating with a phase shift introduced to the fiber Bragg grating during
the inscription process [79]. It is known for a uniform fiber Bragg grating, if a phase shift is introduced
to the fiber Bragg grating, an ultra-narrow notch with a phase jump in the notch would be generated in
the reflection band [80]. If a phase-modulated optical signal is injected into the phase-shifted fiber Bragg
grating that is employed to modify the magnitude and the phase of one sideband, phase-modulation
to intensity-modulation conversion would be achieved, which would lead to the implementation of an
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:20

LO
v
ω
0 90° −ωLO ω0 +ωLO

DP-MZM EDFA

PC1 1
3
LD SMF PD

PC2 2
Output
PM

SSB gain spectrum

Input
ω
ω0−ωLO−ωB ω0 ωLO+ωB Ω

Figure 18 (Color online) A coherent microwave photonic filter implemented based on phase modulation and phase-modulation
to intensity-modulation using the SBS gain.

PS-FBG spectrum
Optical path
PS-FBG

Input
Electrical path

λ
PC
2
1
TLS PM PD Output
3
OC

−1 −1
+1 +1

λ λ λ
λ0 λ0 λ0

Figure 19 (Color online) A coherent microwave photonic filter implemented based on phase modulation and phase-modulation
to intensity-modulation conversion using a PS-FBG.

ultra-narrow passband.
The same concept to implement a coherent microwave photonic filter based on phase-modulation
to intensity-modulation conversion to translate the spectral response of an optical filter to that of a
microwave photonic filter can be used to implement a microwave photonic filter with a flat top [81].
Instead of using the SBS or a phase-shifted fiber Bragg grating, a specially designed superstructured
fiber Bragg grating (SFBG) was used. A superstructured fiber Bragg grating is a special fiber Bragg
grating that is spatially modulated by a periodic sampling function along the grating. Due to the
spatial sampling, the spectral response has multiple channels. If one period of the sampling function
is intentionally increased by a half period, a π phase shift is introduced equivalently to the grating,
leading to a narrow notch. The technique is called the equivalent phase shift (EPS) technique [82]. The
advantage of introducing a phase shift through the EPS technique is that the sampling period is in the
order of tens micrometers, which is 2–3 orders larger than the period of a uniform fiber Bragg grating,
making the grating inscription system have better accuracy [82]. To have a flat bottom in the notch,
the superstructured fiber Bragg grating was designed to have two closely spaced notches which were
achieved by introducing two equivalent phase shifts to the structure. Each phase shift would produce
a Lorentz-shaped notch and the combination of the two closely spaced Lorentz-shaped notches would
lead to a notch with a flat bottom. When the superstructured fiber Bragg grating with a flat-bottom
notch is incorporated to perform phase-modulation to intensity-modulation conversion, a passband with
a flat top was achieved. The reflection bandwidth and notch width of the superstructured fiber Bragg
grating, which determine the frequency tunable range and bandwidth of the passband, can be controlled
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:21

1.0 Reflection (a) 1.0 Reflection (b)

bandwidth bandwidth
0.8 0.8

Magnitude
Magnitude

0.6 0.6

0.4 0.4

0.2 0.2
Notch Notch
0.0 0.0
1551.9 1552.2 1551.9 1552.2
Wavelength (nm) Wavelength (nm)

... ... ...

EPS1 (c) EPS2

Figure 20 (Color online) (a) Simulated reflection spectrum of a phase-shifted fiber Bragg grating with a single phase shift;
(b) simulated reflection spectrum of a superstructured fiber Bragg grating with two phase shifts; (c) structure of a superstructured
fiber Bragg grating [81].

0 0 10 20
(a) (b) 3-dB: 143 MHz
Normalized magnitude (dB)

20-dB: 370 MHz 15

−5 0
−5
Magnitude (dB)

Phase (rad)

Phase (rad)
−10 10
Notch
−10 −10
−20 5

−15 −15
−30 0

−20 −20 −40 −5

1557.45 1557.55 1557.65 1557.75 5.0 5.2 5.4 5.6 5.8 6.0
Wavelength (nm) Frequency (GHz)

Figure 21 (Color online) (a) Measured reflection spectrum and phase response of a fabricated superstructured fiber Bragg grating;
(b) frequency response and phase response of the microwave photonic filter using a superstructured fiber Bragg grating [81].

by controlling the length and maximum index modulation (MIM) of the superstructured fiber Bragg
grating. Figure 20(a) shows a simulated reflection spectrum of a PS-FBG with a single-phase shift and
Figure 20(b) shows a simulated reflection spectrum of an SFBG with two phase shifts. Figure 20(c) shows
the structure of an SFBG in which two EPSs are incorporated.
Figure 21(a) shows the magnitude and phase response in reflection of a superstructured fiber Bragg
grating with a length of 32.4 mm. Due to the limited resolution of the optical vector analyzer, the notch
depth and shape cannot be precisely shown. Figure 21(b) shows the frequency response of the microwave
photonic filter. As can be seen, it has a narrow passband with a flat top. The flatness of the passband
is estimated to be within ±0.25 dB. The phase response of the microwave photonic filter is also shown in
Figure 21(b). As can be seen, the phase response is linear within the passband. The 3-dB bandwidth was
estimated to be 143 MHz. The 20-dB bandwidth was 370 MHz. The shape factor, which is defined as
the ratio between the 20-dB and 3-dB bandwidths, was calculated to be 2.6, which is much smaller than
those reported in [79]. A smaller shape factor represents a better selectivity. By tuning the wavelength
of the optical carrier, the central frequency of the passband can be tuned. The magnitude response
and bandwidth of the passband remained unchanged during the tuning, a feature that is important for
applications where the shape of the passband and bandwidth are required to be constant.
For certain applications, microwave filters with dual or multiple passbands are needed due to the in-
creasing demand for multiband/multifunctional microwave systems that support various modern services.
The same concept can also be used to implement a microwave photonic filter with two or multiple pass-
bands [83]. For a dual- or multi-band microwave photonic filter, the key component is the optical filter
which should have dual or multiple passbands. This can be achieved again using an equivalent phase-
shifted fiber Bragg grating (EPS-FBG), which could be fabricated by using the equivalent phase shift
technique [84]. A phase-shifted fiber Bragg grating has multiple channels due to the spatial sampling of
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:22

Reflection response of the EPS-FBS

TLS1 TLS2
+1st −1st
+1st 0th ω1 ω2

Refection
−1st
1.0
...
0.8 1 1
Reflection

0.6 0 0
5 pm/div 5 pm/div
Wavelength
0.4
Frequency response of the MPF

0.2 1st 2nd

Magntin
0
1553 1556 1559
Wavelength (nm)
ω1 ω2 Frequency

Figure 22 (Color online) Simulated reflection spectrum of an Figure 23 (Color online) Illustration of the generation of the
equivalent phase-shifted FBG with two phase shifts [83]. two passbands using an equivalent phase-shifted FBG based on
phase-modulation to intensity-modulation conversion [83].

PC1
DSF
EPS-FBG
2 OC2
TLS1
OC1 2 3
PM
1 3
Coupler 4 1
TLS2 EDFA1

PC2
Filter input PC3
VNA
PD
Filter output
EDFA2

Figure 24 (Color online) Schematic of a dual-passband microwave photonic filter. TLS: tunable laser source; PC: polarization
controller; PM: phase modulator; EDFA: erbium-doped fiber amplifier; EPS-FBG: equivalent phase-shifted fiber Bragg grating;
OC: optical circulator; DSF: dispersion-shifted fiber; PD: photodetector; VNA: vector network analyzer [83].

the grating structure. Equivalent phase shifts introduced to the ±1st channels are realized by changing
the sampling function. Therefore, the fabrication of a phase-shifted fiber Bragg grating is significantly
simplified, since the control of the spatial sampling is in a micrometer scale, while the control of the
phase shift in the fabrication of a true phase-shifted fiber Bragg grating is in a nanometer scale. In the
design, two π phase shifts are introduced to both of the ±1st channels to produce an ultra-narrow and
flat-bottom notch in each channel. Figure 22 shows a simulated reflection spectrum of an equivalent
phase-shifted fiber Bragg grating with two π phase shifts. As can be seen, the notches are flat due to the
use of two phase shifts. A bandpass filter is achieved due to the phase-modulation to intensity-modulation
conversion by filtering out one sideband of a phase-modulated signal. In a dual-band microwave photonic
filter, the phase-modulation to intensity-modulation conversion is performed at both the ±1st channels;
thus two independently tunable passbands are generated, as shown in Figure 23. Because two π phase
shifts are introduced to achieve flat-bottom notches, the passbands have small shape factors. Here, the
shape factor is again defined as the ratio between the 20-dB and 3-dB bandwidths.
To produce two independent passbands, two optical carriers from two tunable laser sources (TLS1 and
TLS2) are tuned to locate at the ±1st channels of the equivalent phase-shifted fiber Bragg grating, as
shown in Figure 24. When one sideband of a phase-modulated signal is suppressed by the notch in the
+1st channel, a passband with a central frequency that is the frequency difference between the optical
carrier and the notch in the +1st channel, will be produced due to the phase-modulation to intensity-
modulation conversion. Similarly, when one sideband of a phase-modulated signal is suppressed by the
notch in the −1st channel, a passband with a central frequency that is the frequency difference between
the optical carrier and the notch in the −1st channel, will also be produced. Therefore, there will be
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:23

−10 50 −10 10
−15
(a) (b) (c)
The 1st The 2nd −20 40 −20 5

Magnitude (dB)
passband

Magnitude (dB)
passband
Magnitude (dB)

Phase (rad)
Phase (rad)
−25 −30 0
−30 30

−40 20 −40 −5
−35
−50 10 −50 −10

−60 0 −60 −15

−45 0 0.5
0 1 2 3 4 5 6 7 8 −0.5 0 0.5 −0.5
Frequency (GHz) Offset frequency (GHz) Offset frequency (GHz)

Figure 25 (Color online) (a) Frequency response of the dual-passband microwave photonic filter; magnitude and phase responses
of (b) the 1st passband and (c) the 2nd passband [83].

two passbands at two different central frequencies. Again, the central frequencies of the passbands can
be tuned by shifting the wavelengths of TLS1 and TLS2 independently. Figure 25 shows the spectral
response of the dual passband filter. The bandwidth and shape factor of the 1st passband were estimated
to be 167.3 MHz and 3.8, and those of the 2nd passband were 143.4 MHz and 3.3. The 1st and 2nd
passband have frequency tunable ranges of 5.4 and 7.4 GHz, respectively. The central frequencies of both
passbands can be continuously tuned. Again, the shape of the spectral response of the two passbands
was maintained unchanged during the frequency tuning. The variations of the magnitude over the entire
frequency tunable ranges were maintained within ±0.5 dB.
The SFDR and NF of a microwave photonic filter are two important parameters that define the linearity
and noise performance of the filter and should also be evaluated. A few techniques have been proposed
to improve the spurious free dynamic range and the noise figure of a microwave photonic filter [83–87].
One simple solution to increase the spurious free dynamic range and reduce the noise figure is to partially
suppress the optical carriers while maintaining the powers at the input of the photodetector unchanged.
This operation is equivalent to increasing the filter gain without increasing the noise floor (considering the
dominant noise is the shot noise which is generated at the photodetector and is proportional to the optical
power at the input of the photodetector). This can be done, for example, using an SBS-assisted filter,
such as the one reported in [85] where the SBS-assisted filter was implemented using a dispersion-shifted
fiber (DSF). When the light waves pass through the DSF, and the powers of the optical carriers are
above the SBS threshold, Stokes waves are generated, which are traveling along the DSF in an opposite
direction of the optical carriers. Note that, under small-signal modulation conditions, the powers of the
sidebands are below the SBS threshold; thus SBS occurs only at the optical carriers and the sidebands are
not affected. As the Stokes waves circulate in the loop, the optical carriers are suppressed. If the optical
power at the input of the PD is maintained fixed, the suppression of the optical carriers will increase
the gain of the system. The spurious free dynamic range is increased. Since the powers of the optical
carriers are reduced, the short noise at the PD is reduced, which makes the noise figures reduced. In [85],
the spurious free dynamic ranges were improved by about 7 dB, and the noise figures were decreased by
10 dB.
In summary, a microwave photonic filter can be implemented in the optical domain to take advantage
of the high frequency and wide bandwidth offered by photonics. Compared with an electronic microwave
filter, a microwave photonic filter can operate at a much higher frequency and can be widely frequency
tunable. These are the two key features that make microwave filters implemented in the optical domain
attractive. Between incoherent and coherent microwave photonic filters, a coherent microwave photonic
filter has a simpler structure, since a single light source is needed, which provides a high potential for
photonic integration. The key challenge in implementing an integrated coherent microwave photonic
filter is that the optical filter, which is used to translate its spectral response to the microwave domain,
should have a high Q factor to ensure an ultra-narrow bandwidth. The low loss of Si3 N4 may make the
implementation possible, but heterogeneous integration may be needed since no active devices can be
implemented using Si3 N4 .

4.1 Microwave arbitrary waveform generation

Microwave arbitrary waveforms are widely used in radar, communications, microwave imaging, and mod-
ern instrumentation systems. Microwave arbitrary waveforms can be generated electronically based on
analog or digital electronics. Due to the limited bandwidth or sampling rate, microwave arbitrary wave-
forms generated based on analog or digital electronics have low frequency and small bandwidth. For
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:24

t
t t

Optical pulse
Photodetector
Input pulse shaper
Pulse Microwave
sequence waveform

Figure 26 (Color online) Direct space-to-time (DST) mapping for microwave arbitrary waveform generation.

many applications, however, high frequency and large bandwidth microwave waveforms are needed. One
solution is to perform microwave arbitrary waveform generation using photonics, to take advantage of the
high speed and broad bandwidth offered by modern photonics [88]. A few techniques have been proposed
to generate microwave arbitrary waveforms, including (1) direct space-to-time pulse shaping [89–96],
(2) spectral-shaping and wavelength-to-time mapping [97–102] and (3) temporal pulse shaping [103–106].
These techniques can be implemented using free space optics or fiber optics. Free-space optics-based mi-
crowave arbitrary waveform generators have relatively large sizes and higher losses. Fiber optics can make
microwave arbitrary waveform generators smaller and less lossy, and have a high potential for photonic
integration.

4.2 Direct space-to-time mapping

Microwave arbitrary waveform generation can be realized based on direct space-to-time (DST) map-
ping [89–96], in which an optical input pulse is converted to a pulse burst in the optical domain. After
photodetection to a high-speed photoreceptor, a microwave waveform is generated. The key to this
approach is to control the time delays of the pulses in the pulse burst. Figure 26 shows a direct space-
to-time mapping system in which an ultrashort optical pulse from a mode-locked laser source is sent to
an optical pulse shaper to generate a pulse burst. The pulse sequence is then applied to a PD. Due to
the bandwidth-limited nature of the photodetector, the high-frequency components are eliminated, and
a smooth microwave waveform is generated. The theory behind the generation of a microwave arbitrary
waveform using a direct space-to-time mapping system is that the microwave waveform to be generated
can be obtained by filtering the pulse burst using a band-limited filter. If the temporal spacing is in-
creasing linearly, then a chirped microwave waveform is generated. A chirped microwave waveform has a
large time bandwidth product (TBWP), which is widely used in radar systems to increase the detection
distance while maintaining a high range resolution through pulse compression (or matched filtering).
The key device in a direct space-to-time mapping system is the optical pulse shaper. It could be
implemented in free space, but the size is large. A simpler but more effective approach is to use a PIC,
such as an integrated arrayed waveguide grating. An ultra-short pulse is applied to the input of an
arrayed waveguide grating, at the output of the arrayed waveguide grating, a pulse burst with the time
spacing controlled by switching on and off some of the waveguide channels is generated. The limitation
of the approach is that an arrayed waveguide grating developed for optical communications has a small
number of channels, making the generated microwave arbitrary waveforms have a small bandwidth.

4.3 Spectral-shaping and wavelength-to-time mapping

Microwave arbitrary waveforms can also be generated based on spectral-shaping and wavelength-to-
time mapping [97–102]. Figure 27 shows a microwave arbitrary waveform generation system based on
spectral-shaping and wavelength-to-time mapping. As can be seen, the system consists of an ultra-
short pulse source, a spectral shaper, a dispersive element, and a photodetector. The spectral shaper is
used to modify the spectrum from the pulsed laser source, which can be a mode-locked laser source. The
shaped spectrum is then applied to a dispersive element where wavelength-to-time mapping is performed.
The dispersive element can be a length of dispersive fiber or a linearly chirped fiber Bragg grating. A
microwave waveform is generated in the electrical domain at the output of a high-speed photodetector.
If the dispersive element is a length of fiber with a value of the dispersion of Φ̈, the impulse response is
2
given by h(t) = exp(j 2tΦ̈ ). For an input pulse g(t) with a temporal width of ∆t0 , the signal at the output
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:25

Spectral
Optical spectrum response Shaped spectrum

Output waveform
λ λ λ

t
Optical
Pulsed laser Dispersive
spectral Photodetector
source device
shaper

Figure 27 (Color online) Microwave arbitrary waveform generation based on spectral-shaping and wavelength-to-time mapping.
By using a spectral shaper with a linearly increasing free spectral range, a linearly chirped microwave waveform can be generated
after linear wavelength-to-time mapping.

of the dispersive element is calculated by

" #
 2 Z ∞ 2
t (t − τ )
y(t) = g (t) × h (t) = g (t) × exp j = g (τ ) × exp j dτ
2Φ̈ −∞ 2Φ̈
 2 Z ∞  2    
t τ t
= exp j × g (τ ) × exp j × exp −j τ dτ
2Φ̈ −∞ 2Φ̈ Φ̈
 2 Z ∞    
t t
≈ exp j × g (τ ) × exp −j τ dτ
2Φ̈ −∞ Φ̈
 2
t
= exp j × G (ω)|ω= t , (46)
2Φ̈ Φ̈

where G (ω) is the Fourier transform of g (t). As can be seen the envelope of the output signal is
proportional to the Fourier transform of the envelope of the input signal. Note that Eq. (46) is obtained
if the temporal duration of the input ultrashort pulse ∆t0 and the dispersion Φ̈ of the dispersive element
satisfy the condition, given by
∆t0 2
≪ 1, (47)
2Φ̈
2 2 2 2
which means the phase term 2τΦ̈ in (46) satisfies 2τΦ̈ 6 ∆t 0
2Φ̈
≪ 1, thus we have exp(j 2τΦ̈ ) ≈ 1.
If the input to the dispersive device is a rectangular pulse, then the output temporal waveform should
be a sinc function. As can be seen, the key device in the microwave arbitrary waveform generator is
the spectral shaper, which should be designed to have a magnitude response that can make the shaped
spectrum have the same shape as the microwave waveform to be generated. Again, the spectral shaper
can be implemented using a spatial light modulator that can be updated in real time. A fiber-optics based
spectral shaper can also be implemented. For example, we can use a special fiber Bragg grating with
its spectral response designed to have a shape identical to that of the microwave arbitrary waveform.
The key limitation of using a fiber Bragg grating is that the spectral response is fixed or has limited
updatability. Thus, the generated waveform is fixed or can be slightly tuned with limited tunability.

4.4 Temporal pulse shaping

Microwave arbitrary waveforms can also be generated based on temporal pulse shaping (TPS) [103–106].
A typical temporal pulse shaping system consists of an ultra-short pulse source, a pair of complementary
dispersive elements, and an optical modulator. The optical modulator can be an MZM or a phase
modulator. Figure 28 shows the schematic of a temporal pulse shaping system, in which the modulator
is a Mach-Zehnder modulator.
Again, if |∆t0 2 /Φ̈| ≪ 1, where ∆t0 is the temporal width of the input ultra-short short pulse, the
electrical field of the short pulse g(t) after transmitting through the first dispersive element with a value
of the dispersion of −Φ̈ can be expressed as
" #
 2 Z ∞ 2
t (t − τ )
p(t) = g (t) × exp j = g (τ ) × exp j dτ
2Φ̈ −∞ 2Φ̈
 2 Z ∞    
t t
≈ exp j g (τ ) × exp j − τ dτ
2Φ̈ −∞ Φ̈
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:26

MZM
g(t) p(t) q(t) y(t)
MLL
ɺɺ
Φ ɺɺ
−Φ
Vbias x(t)

Sync Pattern
generator

Figure 28 (Color online) A TPS system for arbitrary waveform generation. An ultra-short pulse is generated by a mode locked
laser (MLL) [104].

 2
t
= exp j G (ω)|ω= t , (48)
2Φ̈ Φ̈

where G (ω) is the Fourier transform of g(t). The signal at the output of the Mach-Zehnder modulator
is given by  2  
t t
q (t) = p (t) × x (t) = exp j ×G × x (t) , (49)
2Φ̈ Φ̈
where x(t) is the input microwave signal to the MZM.
After passing through the second dispersive element with a value of chromatic dispersion −Φ̈, we obtain
the output signal, which is the convolution between q(t) and the impulse response of the dispersive element
h(t) = exp(−jt2 /2Φ̈ ),

t2
 
y(t) = q (t) × h (t) = q (t) × exp −j
2Φ̈
" #
Z ∞ 2
(t − τ )
= q (τ ) × exp −j dτ
−∞ 2Φ̈
t2
     
t
= exp −j ×F G × x (t)
2Φ̈ Φ̈ t
ω=− Φ̈

t2
    
t
= 2π Φ̈ exp −j × g (−t) × X − , (50)
2Φ̈ Φ̈

where F denotes the Fourier transform operation and X(ω) is the Fourier transform of x(t).
As can be seen from (50) the output waveform is a convolution between the input optical pulse and
the Fourier transform of the input modulation signal. If the input optical pulse is ultrashort, such as a
unit impulse function, then the generated waveform is the Fourier transform of the modulation signal
since the convolution of a function with a unit impulse is the function itself. One important feature of
this approach is that, to generate a fast waveform with a narrow temporal width, a slow waveform with a
wide temporal width is needed, making the implementation simplified. Note that the major difficulty in
implementing such a system is that the input optical pulse must be precisely synchronized with the input
modulation signal, to ensure the dispersed optical pulse is modulated by the input modulation signal.
In summary, microwave arbitrary waveforms can be generated in the optical domain. The key advan-
tages include wide bandwidth and high reconfigurability. Among the three different techniques, the one
based on spectral shaping and wavelength-to-time mapping can provide the best flexibility in which the
spectral shaper can be implemented using a spatial light modulator with fast reconfigurability. The use
of a photonic integrated circuit [99] as a spectral shaper can have both small size and fast reconfigura-
bility, which is considered a potential solution for future fully integrated microwave arbitrary waveform
generators.

5 True time delay beamforming

Microwave beamforming based on phased array antennas is playing an important role in modern radar
and wireless communication systems to increase transmission and receiving performance. The beam
steering in a conventional phased array antenna is realized by introducing phase shifts to a microwave
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:27

∆L
∆φ

θ θ

Wav
θ θ
Wav

ef
d
ef

ront
ront

B
∆R ∆R

(a) (b)

Figure 29 (Color online) Beam steering using (a) a phase shifter and (b) a delay line.

signal to be radiated to the free space using electronic phase shifters. The advantage of electronic phase
shifters is that they can be implemented using electronic integrated circuits, making the beamforming
network have a compact size with small power consumption and good stability. However, microwave beam
steering based on electronic phase shifters suffers from the well-known beam squint problem, which would
make the radiated microwave beam point in different directions. Thus, a phase array antenna based on
phase shifters can only support narrowband operations. For many applications, such as wideband radar
and broadband wireless communications, it is highly desirable that phase array antennas can support
broadband operations. An effective solution is to replace the phase shifters with true-time delay (TTD)
lines [1].

5.1 Beam squint phenomenon

The beam squint phenomenon is characterized by the position of the mainlobe being oriented at different
angles for different microwave frequency components. In other words, the energy associated with different
frequency components is oriented in different directions. To ensure that the beam is pointing to one
direction, the microwave signal radiated to the free space via a phased array antenna must have a narrow
bandwidth, which would limit the phased array antenna to operate for narrow-band only. The squint
problem is caused due to the use of phase shifters. If true-time delay lines are employed, then the squint
problem is solved [1].
The difference between the use of a phase shifter and a true-time delay line is illustrated in Figure 29.
As can be seen from Figure 29(a), to steer the beam to a direction with angle of θ with respect to the
broadside direction, the phase shifter should provide a phase shift given by
 
d sin θ
∆φ = 2π , (51)
λ
where λ is the wavelength of the microwave signal and d is the antenna array element spacing. The beam
pointing direction is then given by  
−1 ∆φλ
θ = sin . (52)
2πd
As can be seen by changing the phase shift, the beam pointing direction is changed. However, the beam
pointing direction is also a function of the microwave wavelength. For a microwave signal with a finite
bandwidth, the frequency components will be pointing in different directions, a phenomenon known as
beam squint. Thus, to reduce the impact of beam squint, a beamforming network using electronic phase
shifters should only operate for a narrowband signal. The problem can be solved if the phase shifters are
replaced by optical true-time delay lines. As shown in Figure 29(b), the phase shifter is replaced by a
delay line. The beam pointing direction is now given by
 
−1 ∆L
θ = sin , (53)
d
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:28

90
90 1
1
120 60
120 60
0.8 0.8

0.6 0.6
150 30 150 30
0.4 0.4

0.2 0.2

180 0 180 0

210 330 210 330

d = 1 cm d = 1 cm
240 300 240 300
N=6 N=6
270
270

Figure 30 Radiation pattern of a six-element array antenna Figure 31 Beam squint effect for a phased array antenna
with d = λ0 /2 at f0 = 15 GHz. based on electronic phase shifters operating in a frequency
range from 10 to 20 GHz.

where ∆L is the physical length of the delay line. It can be seen that the beam pointing direction
is independent of the microwave wavelength or frequency. Thus, microwave beamforming with a wide
instantaneous bandwidth is achieved, and the system is squint free.
Figure 30 gives an example of an array antenna having six array elements with adjacent array elements
separated by a distance of 1 cm (d = λ0 /2) operating at a central frequency of 15 GHz. To illustrate
the beam squint effect, the bandwidth of a microwave signal to be radiated is assumed to be 10 GHz,
covering a frequency range from 10 to 20 GHz. The far-field radiation pattern for the central frequency
pointing at 120◦ is given in Figure 30.
Figure 31 illustrates the far-field radiation pattern for the microwave signal with a bandwidth of
10 GHz, from 10 to 20 GHz. As can be clearly seen, the beam pointing direction of the mainlobe varies
with the feed signal frequency. This phenomenon would decrease significantly the performance of the
beamforming system.
To eliminate the beam squint, a solution is to use true-time delays. This method is to introduce a
time delay progression to the multi-channel feed signals instead of a phase progression. For a delay line,
the time delay is constant for all frequencies and thus translates into a linearly variable phase shift with
respect to frequency. Figure 32 shows the far-field radiation pattern by using the same phased array
antenna operating at the same central frequency of 15 GHz. The behavior of the antenna is studied
again for the microwave signal with a bandwidth covering a frequency range from 10 to 20 GHz. This
time, true-time delay lines are used instead of electronic phase shifters. These elements introduce a time
progression of 16.67 ps which corresponds to a phase of π/2 at the frequency of 15 GHz. As can be seen,
the orientation of the mainlobe does not vary with the feed signal frequency. Thus, the beam squint
problem is resolved.

5.2 Implementation of true-time delay beamforming

Conventional beamforming networks are implemented using electronic phase shifters. To eliminate beam
squint, phase shifters have to be replaced by true-time delay lines. Electronic delay lines with tunable
time delays are difficult to implement and the performance in terms of bandwidth and loss is very limited.
For example, copper wires exhibit high loss at high frequencies resulting in a limited bandwidth. With
the fast advancement in photonics technology, true-time delay lines implemented in the optical domain
have been considered a strong candidate for broadband squint-free beamforming. In addition to the
inherent broad bandwidth offered by modern photonics, the implementation of true-time delay lines in
the optical domain can also benefit from the ultra-low loss of optical fibers, enabling not only effective
true-time delay generation, but also low loss and broadband microwave signal transmission over fiber.
True-time delay beamforming based on photonics has been extensively researched in the past few
years [107–125]. A beamforming network can be implemented using free space optics [107, 108], but with
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:29

90
1
120 60
0.8

0.6
150 30
0.4

0.2

180 0

210 330

d = 1 cm
240 300 N=6
270

Figure 32 Far-field radiation pattern of a phased array antenna based on true-time delay lines operating in a frequency range
from 10 to 20 GHz.

PD1
RF Input

PD2
Optical splitter

MZM

PD3

PD4
TLS

λ1 λ2 λ3 λ4 λ5

Figure 33 A photonic true-time delay beamforming system based on fiber Bragg grating based delay lines.

large size and high loss. A beamforming network can also be implemented based on fiber optics [109–125].
Since the loss of state-of-the-art fibers is very low, a fiber-optics-based beamforming network has better
performance in terms of smaller size and lower loss. Figure 33 shows a fiber Bragg grating based true-
time delay beamforming network. As can be seen, it consists of a tunable laser source, a modulator, four
channels of fiber Bragg grating delay lines, and four photodetectors. By tuning the wavelength of the
light from the tunable laser source, the four-channel signals will be reflected at different locations and
the beam pointing direction can be steered. Note that the time delays are discretely tunable due to the
use of fiber Bragg grating delay lines. If all the fiber Bragg gratings in the delay lines are replaced by
chirped gratings, continuous beam steering would be realized.
In summary, the implementation of microwave beamforming based on photonic delay lines has a key
advantage of squint free, which is highly needed to support beam steering with wide instantaneous
bandwidth. A photonic true-time delay beamforming network can be implemented based either on
free space optics, but the size is large and the weight is high, or fiber optics with smaller size and
lower weight, but the stability is poor. Advanced packaging may be needed to reduce the sensitivity to
environmental interferences, but the cost will be increased. With the advances in photonic integrated
circuits, a microwave beamforming system on an integrated chip could be implemented with a much
smaller size and lower weight, and is able to support broadband beamforming with a large number of
channels.
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:30

6 Integrated microwave photonics

The microwave photonic subsystems and systems discussed in the previous sections are implemented based
on discrete components. Despite its tremendous potential, and a considerable number of functionalities
that have been demonstrated over the last 40 years, the widespread use and application of microwave
photonics have been limited by the high cost, bulky, complex and power-consuming nature of the systems.
A major challenge in microwave photonics implies a reduced cost, size, weight and power consumption
(SWaP) of its devices, subsystems and systems. Integrated photonics has the potential to change the
scaling laws of high-bandwidth systems through proper architectural choices that combine photonics
with electronics to optimize the performance, power, footprint, and cost [126]. Drastic space and weight
reductions are immediate gains from integration.
IMWP deals with the application of integrated photonics technologies to microwave photonics systems
[127,128]. During the last few years, IMWP has become probably the most active area of current research
and development in the discipline of microwave photonics [129], capitalizing upon the outstanding progress
of integrated photonics in various material platforms such as InP, SOI, silicon nitride (Si3 N4 ), and LOI.
In this section, we will first outline the salient characteristics of available material platforms that can
be employed for the implementation of IMWP chips. We will compare the features of the more mature
material platforms and then describe the two salient approaches that are available for the implementation
of IMWP chips from a functional point of view. On one hand, we review the recent progress in application-
specific photonic integrated circuits (ASPICs), where a particular circuit configuration is designed to
optimally perform a particular microwave photonics functionality. Examples of different functionalities
recently reported will be presented, including tunable filtering, optoelectronic oscillation, instantaneous
frequency measurement, and frequency up and down conversion. On another hand, recent progress will
be reported on a radically different approach, programmable photonics, that enables the implementation
of general purpose programmable microwave photonics signal processors that can be integrated on a chip
and are capable of performing all the main functionalities by suitable software programming of its control
signals. This approach is inspired by the flexibility of digital signal processors, where a common hardware
is shared by multiple functionalities through a software-defined approach (or programmability), leading
to significant cost reduction in the hardware fabrication. Further coverage of IMWP can be found in two
existing monographic reviews [127, 129].

6.1 Design, fabrication, and material platforms

IMWP design and fabrication processes must align with the state-of-the-art generic integration (GIM)
and generic foundry (GFM) models [130] to leverage the already existing ecosystem. In the GIM, a small
set of standardized basic building blocks (BBBs) are provided to control the basic properties of light and
more complex circuits or compound building blocks (CBBs) are then built by connecting several BBBs.
In the GFM, the foundry provides a shared open access to its generic integration process through multi
project wafers, where different designs or ASPICs from different users are combined on the same wafer
providing cost sharing and additional advantages, such as on-wafer testing. Since the GIM and GFM
allow a broad range of functionalities to be synthesized from a small set of BBBs, they have been identified
as the only path leading to mass production and subsequent cost reduction for PIC manufacturing, where
figures in the order of 10 ¤/mm2 can already be reached at volumes of 1000 chips for InP technology [130].
Three material platforms: (1) InP, (2) SOI, and (3) Si3 N4 , have reached the required degree of maturity
to be considered as viable options for the implementation of complex photonic integrated circuits, either
monolithic or hybrid or heterogeneous. These operate mainly in the 1.3 and 1.55 µm wavelength regions.
Recently, a new platform based on lithium niobate on insulator (LNOI) is starting to be considered a
viable alternative for the implementation of IMWP circuits requiring high-performance modulators. We
now describe the salient features of each of them.
Indium phosphide PIC technology. Indium phosphide is a III-V compound semiconductor material and
the only technology capable of the monolithic integration of active (i.e., featuring optical amplification)
and passive photonic components [131, 132]1). A variety of techniques, including butt-joint regrowth, se-
lective area growth, offset and dual quantum well placing and quantum well intermixing, can be employed
to integrate regions with different absorption/gain properties along a single waveguide. Waveguide types

1) JePPIX Roadmap 2015. http://www.jeppix.eu.

 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:31

(a)

(b)

(c)

Figure 34 (Color online) Waveguide structures for the InP (a), SOI (b), and Si3 N4 -SiO2 (c) platforms.

include shallow (low loss and long bending radius) and deeply etched (higher losses and lower bending
radius) designs (see Figure 34(a)).
Shallow waveguides have typical widths of 2 µm and this figure is approximately 1.5 µm for deeply
etched designs. Typical losses are around 1.5 and 3 dB/cm for shallow and deeply etched waveguides,
respectively, and this is usually quoted as one of the disadvantages of this technology. These losses can
be reduced, however, to figures below 1 dB/cm by removing the p-type top layer in the non-active parts
of the chip. Minimum bending radii are between 10 and 500 µm for deeply etched and shallow etched
waveguides, respectively. Another quoted limitation of the InP platform is related to the difficulty of its
integration with electronics into the same chip. However, highly promising work is underway towards the
integration of full photonic functionality in a single InP membrane on Silicon that could be implemented
both in a CMOS or an InP fab, merging InP photonics with silicon electronics.
InP provides the most complete list of available components for integration [131] including passives,
optical filters, arrayed waveguide gratings, semiconductor optical amplifiers, saturable absorbers, optical
sources (Fabry-Perot, distributed Bragg reflector, distributed feedback, and mode-locked lasers), detectors
(pin single and balanced configurations), amplitude and phase modulators and polarization handling
devices. Moreover, it has shown record performance in integrated component count on the same chip
with figures exceeding 450 and now targeting values in the thousands range [132].
Silicon photonics PIC technology. Also known as SOI [133–136], silicon photonics is a semiconductor
technology where components are etched/patterned/fabricated in a 180–220 nm silicon layer placed on
top of a 1–3 µm insulator layer. Silicon passives are formed by initial few mask layers through par-
tial and/or full silicon etching steps after which multiple ion implantations are conducted for “active”
devices such as Ge photodetectors and silicon modulators. Coupling into and outside the chip can be
performed via edge couplers (with typical losses of 1 dB/facet) or vertically, via silicon surface gratings
(3.5–6 dB/coupler with 40–70 nm 3-dB bandwidth). The main advantage of the SOI technology resides
in its potential compatibility with the well-developed CMOS fabrication processes and the infrastructure
used in microelectronics and thus in the potential for monolithic integration of the electronic and photonic
parts of the same chip. Refractive index contrast is over 100% (3.4 for silicon and 1.45 for SiO2 ), leading
to small footprint circuits.
Two main types of waveguides are available (see Figure 34(b)). Ridge or shallow (1–8 µm width),
which exhibit relatively low losses down to 0.1–0.5 dB/cm, but are limited in bending radius to around
100 µm, and strip waveguides (500 nm width) that exhibit much higher losses (1–3 dB/cm) but support
lower values for minimum bending radius (5–20 µm). Integration density on a chip is currently below
100 components [136], but the component count integration trend is exceeding the rate given by Moore’s
law indeed. Several basic photonic components are available in monolithic SOI, including passives, such
as arrayed waveguide gratings and optical filters, Ge photodetectors, ring and traveling-wave electro-
refractive modulators (up to 50 GHz). The main disadvantage of monolithic SOI technology is that it
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:32

does not support optical sources and other active components such as optical amplifiers. To overcome this
limitation, III-V functionalities have to be integrated into the SOI platform by means of molecular [137],
adhesive [138] wafer bonding and more recently by transfer printing [139]. This approach, known as
hybrid silicon or hybrid integration technology has succeeded in incorporating amplifiers, saturable ab-
sorbers, optical sources (Fabry-Perot, distributed Bragg reflector, distributed feedback, and mode-locked
lasers), optical amplifiers, electroabsorption (>70 GHz), and Mach-Zehnder (>25 GHz) modulators, pho-
todetectors (>35 GHz) and polarization handling components into SOI chips. However, GFM and GIM
have not yet been developed for hybrid integration technology.
Silicon nitride Si3 N4 -SiO2 PIC technology. This waveguide technology is based on a combination of
stoichiometric Si3 N4 as waveguide layers, filled by and encapsulated with silica (SiO2 ) as cladding layers
grown on a silicon wafer [140, 141]. SiO2 and Si3 N4 layers are fabricated with CMOS-compatible in-
dustrial standard low-pressure chemical vapour deposition equipment that enables cost-effective volume
production. A special technology known as TriPleXTM [140], developed by a Dutch company, LioniX
International, allows the fabrication of waveguides with minimized surface roughness allowing high refrac-
tive index contrast (20%–30%) and low scattering losses. The TriPleXTM platform offers seven waveguide
cross-sectional geometries [140], some of which are shown in Figure 34(c). Their salient characteristics
at 1550 nm are low index contrast box-shaped (1 µm × 1 µm) that features reduced birefringence, 60.2
dB/cm losses and minimum bending radius (Rmin ) around 500 µm and mode field area (MFA) of 3.6
µm × 3.6 µm; high-index contrast featuring Rmin of 150 µm and MFA of 1.4 µm × 1.4 µm; single-stripe
(Rmin of 2000 µm, <0.03 dB/cm losses and MFA of 4.7 µm × 2.9 µm); double-stripe (1 µm × 1 µm,
Rmin of 100 µm and MFA of 1.6 µm × 1.7 µm) obtained by removing the sidewalls of the box configu-
ration, which leads to the significant reduction of the waveguide losses that are now in the <0.1 dB/cm
range. In/out coupling is achieved by means of adiabatically tapered spot-size converters to low index
contrast double-stripe cross-sections with <1 dB coupling loss. Several fundamental building blocks are
available including the optical waveguide, tuning elements, and directional and multimode interference
couplers. From these, more complex subsystems have been demonstrated. For the double-stripe geom-
etry, a library of standard optical components with predictable characteristics is available. The main
disadvantage of this technology is that no optical sources, detectors, amplifiers, and modulators are avail-
able. The integration of these components requires a hybrid approach with separately fabricated InP
platform chips [141]. The recent development of InP device transfer printing techniques [139, 142] and
InP membranes on Silicon [143] opens the path for the compact and versatile implementation of this
required hybrid integration approach.
Lithium niobate on silicon technology. Lithium niobate (LiNbO3 or LN) on insulator (LNOI) is a
promising material platform for integrated photonics due to single crystal LiNbO3 film’s wide transparent
window, high refractive index, and high second-order nonlinearity [144]. Compared with traditional
material platforms described above, LNOI has several advantages, including strong electro-optic effect
(largest r33 = 27 pm/V at 1500 nm), large refractive index (no = 2.21 and ne = 2.14 at 1550 nm),
wide transparency wavelength (from 400 nm to 5 µm), and stable physical and chemical characteristics,
thus making it a competitive material for linear and nonlinear integrated photonics [145]. Similar to
SOI, LNOI on insulator consists of a submicrometer LN film on a silica buried layer, which is on top
of a substrate made from silicon or LN. To fabricate an LNOI device, a single-crystal sub-micrometer
LN film is first obtained by ion slicing a bulk crystal and then is bonded to a low index substrate by
using benzocyclobutene bonding [146] or crystal bonding [147]. This fabrication process has already
matured and a typical LNOI wafer has an LN film thickness of hundreds of nanometers (typically 300–
900 nm) and a diameter of 3 or 4 inches2) . Nanostructuring techniques include wet etching, dry etching,
chemical mechanical polishing (CMP), diamond dicing, femtosecond laser direct writing, and focused ion
beam (FIB). All these techniques enable fabricating LN ridge waveguides. Although propagation loss
of LN waveguide has a theoretical limitation down to 0.001 dB/cm [148, 149]. So far, even the best LN
waveguide shows a loss higher than this limitation due to radiation loss caused by waveguide bending
and scattering loss induced by surface roughness. Nevertheless, progress in improving losses has led to
losses as low as 0.027 dB/cm while a standard figure is 0.32 dB/cm. Several devices have been reported
in this technology, including modulators (featuring bandwidths up to 70 GHz) cavity ring and micro-disk
resonators, and periodically poled waveguides and electrooptic comb sources.

2) Jinan Jingzheng Electronics Co., Ltd. https://www.nanoln.com.

 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:33

Table 1 Excerpt of recent IMWP devices after [127, 150]a)

Function Material FSR CF BW R

Filter InP 23.5 GHz or 47 GHz 28 GHz 1.9–5.4 GHz 32 dB
Filter InP – 10–40 GHz 3.9–7.1 GHz 40 dB
Filter TriPleX – 2–8 GHz 247–840 MHz 60 dB
Function Material Pulse rate Pulse width BW FCC
AWG SiN 595 GHz 300 fs – –
AWG As2S3 Chalc. – – 4–15 GHz Yes
AWG InP – – 4–15 GHz Yes
AWG TriPleX – – 4–15 GHz Yes
Function Material PS PS BW
TBPS SOI 0◦ –90◦ 16–20 GHz – –
Function Material Delay Delay BW
TTTD SOI 0–25.7 ps 8 GHz – –
TTTD TriPleX 0–396 ps 1 GHz – –
TTTD SOI 21–96 ps – – –
Function Material Range Error Power
IFM InP 5–15 GHz 200 MHz [−3, 7] dBm –
IFM TriPleX 0.5–4 GHz 93.6 MHz – –
Function Material Lock range Loop BW
OPLL InP [−9, 7.5] GHz 400 MHz – –
Function Material RF band Inst. BW # AE
BFM TriPleX 2–10 GHz 8 GHz 16 –
a) FSR: free spectral range; CF: center frequency; BW: bandwidth; R: rejection; FCC: US federal communications commission
compliance; PS: phase shift; Inst.: instantaneous; AE: antenna element.

6.2 Application-specific photonic integrated circuits for microwave photonics

Microwave photonics with a variable degree of complexity arising from the combination of building blocks
has been reported for a variety of functionalities during the last years. In particular, ASPICS has been
demonstrated. For some of these functions, an excerpt of relevant studies is provided in the following
subsections, and summarized in Table 1. For further details refer to [127–129, 150]. We now review some
salient results.
Filters. As far as integrated coherent filtering is concerned, many of the preliminary approaches
reported so far have been based mainly on single cavity ring resonators (RRs). A few, however, have
also focused on more elaborated designs involving more than one cavity and programmable features.
Representative results for single and multiple cavity filters or ring resonators can be found in [151–157].
For instance, Norberg et al. [154] reported the results for a unit cell, shown in Figure 35(b) that can be
employed as a basic building block of more complex filters as displayed in Figure 35(a).
The unit cell integrated in InP-InGaAsP is a ring loaded Mach-Zehnder interferometer. By selectively
biasing the SOAs and phase modulators placed in the arms of the unit cell, filters with a single pole, a
single zero or a combination of both can be programmed as shown in Figure 35(c). The cell concatenation
scheme shown in Figure 35(a) to form complex filters consists of the uncoupled cascade of a set of N unit
cells, where each one can be programmed to independently act as an FIR, IIR or FIR+IIR cell. Synthesis
algorithms developed in [158] can be therefore applied to implement discrete-time optical filtering with
arbitrary complexity.
A hybrid version incorporating silicon waveguides has also been reported [156, 157] that combines III-
V quantum well layers bonded with low loss passive silicon waveguides. Low loss waveguides allow for
long loop delays while III-V quantum devices provide active tuning capability. The same group involved
in [154] reported results of more complex designs involving two and up to three coupled ring cavities inside
the Mach-Zehnder structure [155], which enables second- and third-order filters as shown in Figure 36.
In particular, for this design the frequency tuning range spans around 100 GHz.
Multistage, coupled cavity filters, also known as CROWS (coupled resonant optical waveguides), have
been reported in SOI technology. In [159, 160], 1–2 GHz-bandwidth filters with very high extinction
ratios (∼50 dB) have been demonstrated. The silicon waveguides employed to construct these filters
had propagation losses of ∼0.5 dB/cm and insertion losses (excluding fiber to waveguide coupling) in the
range of 2–3.5 dB. Each ring was thermally controlled by means of metal heaters, placed on the top of the
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:34

(a)

(b)
(c)

Figure 35 (Color online) Integrated InP-InGaAsP first-order MWP coherent filter providing one pole and one zero reported
in [154]. (a) Cascade configuration of unit cells to implement high-order filters; (b) unit cell configuration; (c) examples of
measured transfer functions for one pole and one zero configurations.

(a) (b)

Figure 36 (Color online) (a) Layout of a multistage InP filter with tunable poles and zeros (after [155]); (b) detail of the resonance
shapes for different reconfiguration parameter setups with their corresponding pole positions.

(a) (b)

Figure 37 (Color online) (a) Optical images of fully fabricated 2nd-order and 5th-order CROW filters (after [159]); (b) measured
(drop and through) transfer functions for the filters.

ring. With a power dissipation of ∼72 mW, the ring resonance could be tuned by one free spectral range,
resulting in wavelength-tunable optical filters. Both the second-order and fifth-order ring resonators were
demonstrated. Figure 37 shows the relevant results for this filter.
More recently filters with a higher degree of sophistication obtained by the cascade of several CROW
stages have been reported for channel selection in dense wavelength division multiplexed systems. Fig-
ure 38 shows the results from Romero-Garcı́a and co-workers [161]. The reported devices were based
on three cascaded stages of CROWS loaded on a common bus waveguide. As shown in the layout of
Figure 38(a) and the fabricated device in Figure 38(b), the stages differed in the number of resonators but
were implemented with exactly identical unit cells, composed of a matched racetrack resonator layout and
a uniform spacing between cells. The different number of resonators in each stage allowed a high rejection
in the through port response enabled by the interleaved distribution of zeros. The exact replication of a
unique cell avoided the passband ripple and high lobes in the stopband that typically arise in apodized
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:35

(b)

(a)
(d)
(c)

(e)

Figure 38 (Color online) Multiple stage CROW filter in SOI (after [161]). (a) Filter layout; (b) fabricated device in SOI; (c) a
detail of the electrodes driving the phase shifters in each CROW stage; measurement results of the three individual stages of the
filter and combined response (d) before and (e) after tuning with thermal phase shifters.

(a) (c) (d)

(e) (f)

(b)

(g) (h)

Figure 39 (Color online) (a) and (b) Layout of a Si3 N4 reconfigurable SCISSOR (after [164]). (c)–(h) Different transfer functions
and frequency tunings obtained by reconfiguration. (c) Band-pass and (d) band-stop filter response. (e) and (f) Tuning of filter
center frequency. (g) and (h) Variations of filter response shape and bandwidth.

coupled resonator optical waveguide-based filters due to fabrication and coupling induced variations in
the effective path length of each resonator. The filters could be reconfigured by means of thermal phase
shifters incorporated in each CROW stage as shown in Figure 38(c). The fine tuning enabled the selection
of WDM channels with an increased rejection ratio as shown in Figure 38(d) and (e).
Rasras and co-workers [162, 163] demonstrated bandpass and notch filters based on Mach-Zehnder
interferometer (MZI) tunable couplers and RRs fabricated in silicon-buried channel waveguides with a
width of 2 µm and a propagation loss of 0.25 dB/cm.
High-performance filters have also been demonstrated on other platforms. For instance, in the silicon
nitride platform researchers from Monash University in Australia reported a complex side coupled inte-
grated series of optical resonators (SCISSOR) filter composed of the serial cascade of three ring resonators
of different FSRs [164–166], as illustrated in Figures 39(a) and (b), and their filter responses shown in
Figures 39(c)–(h).
Here, a pair of rings with identical FSRs were employed for filter shape synthesis and a third ring
with a smaller FSR was used to perform the function as a modulation translator [165, 166] that enables
a separate manipulation of the optical carrier phase. A reconfigurable filter was demonstrated using
such a processor where all the reconfiguration operations including filter frequency tuning, bandwidth
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:36

(a) (b)

(c)

(d) (e)

Figure 40 (Color online) (a) RAMZI filter layout based on (b) an asymmetric Mach-Zehnder interferometer. (c) Photograph of
the fabricated device in Si3 N4 (after [168]). (d) and (e) Different filter transfer functions obtained by parameter reconfiguration in
one ring and two ring RAMZIs.

variation, and switching between bandpass and bandstop filter shapes were implemented as shown in
Figures 39(c)–(h) using the tuning mechanism based on DC-voltage-driven resistor heaters.
Another example of a reconfigurable and versatile filter that combines both FIR and IIR sections is
the ring assisted MZI (RAMZI) shown in Figure 40. It comprises an asymmetric MZI with each arm
coupled with an RR. This filter can be subject to different parameter settings resulting in the possibility
of implementing either an FIR, an IIR or an FIR + IIR response as demonstrated in [159]. A team of
researchers from Lionix international, the University of Twente in the Netherlands and Monash University
in Australia reported a tunable RAMZI circuit [167, 168] employing a total of seven tuning elements,
including two tunable couplers at the input and output of the MZI, two tunable couplers of the two RRs,
and three tunable phase shifters in the ring loops and MZI delay line, all implemented using resistor-based
heaters. This RAMZI design enabled a variety of different implementations of optical filters as shown in
Figure 40. The same team has reported a wide variety of reconfigurable filters for microwave photonics
applications. A detailed description of most of these designs can be found in [169]. RAMZI filters have
also been fabricated in SOI technology [170] and a monolithic version for radiofrequency filtering has
been recently reported in InP [171].
To date, most filter implementations are incomplete in the sense that not all the required elements are
included in the chip. The exception is a recently reported monolithic MWP filter implemented in InP
technology [171]. The filter layout shown in Figure 41(a) integrates all the required elements into the same
substrate, including a tunable distributed Bragg reflector laser (DBR), a dual-drive modulator, a tunable
optical filter based on a RAMZI where poles and zeros can be selected by changing the coupling constant
of a set of 3-dB MZI couplers via the thermo optic effect and a photodetector. The chip photograph
shown in Figure 41(b) includes also other testing structures. The chip was fully packaged as shown in
Figure 41(c) and its reconfigurable operation was tested in the laboratory where the transfer function
of the RF-Photonic filter could be reconfigured by changing the bias current operating over the thermo
optic phase shifters inserted in the arms of the 3-dB MZI tunable couplers as shown in Figure 41(d).
Optoelectronic oscillators. Recently several attempts have been reported towards the full integration of
the photonic part of an OEO on a chip [172,173]. In [172], a silicon photonic integrated OEO that includes
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:37

(a) (b)

(c) (d)

Figure 41 (Color online) (a) Schematic of the InP chip, showing all the main building blocks of a monolithic tunable RF-Photonic
filter; (b) image of a fabricated die (6 mm × 4 mm); (c) packaged chip; (d) measured magnitude transfer function response for three
different current sets in the thermo-optic heaters. TL: tunable laser; MZM: dual-drive Mach-Zehnder modulator; PD: photodetector;
PM: phase modulator; MZI: Mach-Zehnder interferometer (after [171]).

(a) (b)

Figure 42 (Color online) Schematic of the proposed integrated photonics OEO (after [172, 173]). (a) Perspective view of the
proposed OEO; the fabricated chip prototype captured by a microscope camera; perspective view of the MDR with a top-placed
micro-heater; perspective view of the MDR with a p-type doped micro-heater; image of the MDR with a top-placed micro-heater;
and an image of the MDR with a p-type doped micro-heater (after [172]). (b) Schematic and photographs of the InP OEO, a
photograph of the fabricated device, the photonic part of the OEO. DML: directly modulated laser; ODL: optical delay line; PD:
photo detector; EA: electrical amplifier; ATT: attenuator (after [173]).

a high-speed PM, a thermally tunable micro-disk resonator (MDR), and a high-speed PD has been
proposed and experimentally demonstrated. The basic features of this circuit are shown in Figure 42(a).
The OEO was able to generate a microwave signal with a frequency tuning range from 3 to 8 GHz. The
phase noise of the generated microwave signal was measured to be around −80 dBc/Hz at a 10-kHz offset
frequency, which is still far from the values achieved with fiber-based OEOs.
In [173], an InP based integrated reconfigurable optoelectronic oscillator has been reported and ex-
perimentally demonstrated. In this case, the chip contains the active elements and also all the main
RF components are included. The storage cavity here was implemented by means of a spiral waveguide
rather than by an integrated resonator. All of the devices needed in the optoelectronic oscillation loop
circuit were monolithically integrated on a chip within a size of 5 cm ×6 cm. By tuning the injection
current, the output frequency of the proposed OEO was tuned from 7.30 to 8.87 GHz featuring the phase
noise value of around −91 dBc/Hz@1 MHz.
Arbitrary radiofrequency waveform generators. Microwave arbitrary waveform generation (AWG) can
find very useful applications in future 5G communication systems, pulsed radar, and modern instru-
mentation. The currently available electronic arbitrary waveform generation is limited in frequency and
bandwidth to around 22 and 6 GHz, respectively. The use of photonic techniques provides a promising
solution to overcoming the main limitations encountered in electronics systems and a variety of tech-
niques have been proposed [88, 89, 174] including direct space-to-time mapping, spectral shaping and
wavelength-to-time mapping and temporal pulse shaping to generate microwave waveforms in the giga-
hertz and multiple gigahertz regions. Many proofs-of-concept and first demonstrations of these techniques
have been implemented through bulky fiber-opticbased systems and clearly an on-chip integrated pulse
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:38

(a)
(b)

(c)

Figure 43 (Color online) On-chip arbitrary waveform generation using cascaded of 8 microring resonators in SOI platform.
(a) Waveform generator system; (b) optical image of the pulse shaper; (c) example of waveform generation at 60 GHz (after [175]).

shaper is thus a desirable solution to overcome the limitations commonly associated with these conven-
tional bulk optics pulse shapers.
Khan and co-workers [175] at Purdue University demonstrated an integrated ultrabroadband microwave
arbitrary waveform generator that incorporates a fully programmable spectral shaper fabricated on a
silicon photonic chip. The spectral shaper is a reconfigurable filter consisting of eight add/drop micror-
ing resonators on a silicon photonics platform. Ultra-compact cross-section (500 nm×250 nm) silicon
nanowires were used to fabricate the rings. The typical bending radius used was 5 µm and the waveguide
propagation loss was around 3.5 dB/cm. This spectral shaper programmability is achieved by thermally
tuning both the resonant frequencies and the coupling strengths of the microring resonators. A cartoon
of the spectral shaper and an optical image showing two channels of ring resonators with micro-heaters
are shown in Figures 43(a) and (b), respectively.
The principle of the photonic microwave arbitrary waveform generation system implemented here is
shown in Figure 43(a). The spectral shaper is used to modify the spectrum emitted from a mode-locked
laser. The shaped spectrum then undergoes wavelength-to-time mapping in a dispersive device, which
in this case is a length (5.5 km) of optical fiber, before being converted to the electrical domain of
a microwave waveform using a high-speed photodetector. By incorporating the spectral shaper into a
arbitrary waveform generation system, a variety of different waveforms were generated, including those
with an apodized amplitude profile, multiple π phase shifts (Figure 43(c)), two-tone waveforms and
frequency-chirped waveforms at the central frequency of 60 GHz.
The main drawback of this design stems from the fact that it requires a dispersive element for the
frequency to time mapping operation. This of course can be provided on the flight by an optical fiber
link if the signal is to be generated at a centralized site and then delivered to a remote location for
radiation. If, however, the full signal generation is to be achieved on-chip, then the use of the dispersive
delay line must be avoided and then resorts to the alternative techniques. Figure 44 shows the results of
a silicon chip based on spectral shaping as well, but where the time delays for each spectral component
are provided by separate tunable delay lines, implemented with SCISSORS [176].
The AWG layout is shown in Figure 44(a). An input ultra-fast pulse is sampled by eight microring
add-drop filters of different radii Ri (λi denotes the associated resonance wavelength, i = 1, 2, . . . , 8) that
download and produce eight replica pulses. Since the wide bandwidth of the input pulse covers several
free spectral ranges of the silicon microrings, each replica generation microring downloads up to six small
slices of the input spectrum. Although in the time domain, each pulse replica actually consists of a series
of densely packed exponentially decaying amplitude pulses. Only the envelopes of the pulse replicas can
be detected during optical/electrical conversion, the width of each pulse replica’s envelope being inversely
proportional to the optical round-trip time in the ring and the ring’s power coupling coefficients to bus
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:39

(a)

(b)

(c) (d) (e) (f)

Figure 44 (Color online) Chip design (a) and fabricated layout (b) for a reconfigurable arbitrary waveform generator, using
wavelength to time mapping and selective tunable delay lines based on SCISSORS. (c)–(f) Examples of generated waveforms.
(c) 40-GHz uniform RF burst; (d) 40-GHz apodized RF burst; (e) 40-GHz burst with an intermediate p phase shift; (f) frequency
up-chirped burst from 30 to 50 GHz (after [176]).

waveguides. The frequency spacing between the selected channels in the filter bank can be arbitrary and
non-uniform as long as they are not too small to cause interference between channels.
After the pulse replication is performed in each stage for a given wavelength region of the pulse, a
selective delay is impressed by means of an on-chip tuneable delay element consisting of a fixed part and
a tuneable part shown in Figure 44(b). The fixed part is a long waveguide with a delay value at multiples
of a given set value τ ps, and each channel has τ ps more delay than the previous one. The tuneable
part consists of cascaded microring SCISSOR all-pass filters, which are activated to provide more or less
additional delay. A final stage composed of another eight microring add-drop filters of different radii Ri
provides the final amplitude for the wavelength sample and combines all the wavelength samples into a
common output waveguide bus. Figures 44(c)–(f) provide examples of several AWG waveforms generated
by reconfiguring the chip parameters.
A particular waveform of interest for radar and communication applications is the linearly chirped
microwave waveform (LCMW). For LCMW generation, the optical spectral shaper should have a spectral
response with an increasing or decreasing FSR. The approach based on spectral shaping and wavelength
to time (SS-WTT) mapping provides large flexibility, since the spectral response of the spectral shaper
can be easily reconfigured.
The key component in an SS-WTT mapping system is the spectral shaper. In fact, the configuration
proposed in [176] can provide an increasing or decreasing FSR by thermal tuning of the MRR. However,
the chirp rate that can be obtained is quite limited. To generate an LCMW with a large chirp rate, other
approaches have been recently proposed and demonstrated in silicon photonics and fabricated using a
CMOS-compatible technology with 193-nm deep ultraviolet lithography. Figure 45(a) shows the layout
of a spectral shaper [177] that consists of multiple cascaded MRRs with largely different radii. It has
an MZI structure incorporating multiple cascaded MRR. The incident light beam is split by the first
adiabatic 3-dB coupler into two beams that travel through the upper and lower arms of the MZI. Each
ring selectively transfers the optical power at its resonance wavelength from the through-port waveguide
to its drop-port waveguide (note that different colors of the rings shown in the figure indicate that each
ring is designed to have a different radius and therefore a different resonant wavelength). The spacing
between two neighboring rings is large enough to avoid mutual interferences. In the upper and lower
waveguides, S-shape waveguide bends, which are designed using Bezier curves to minimize the mode
mismatch and thus reduce the waveguide bending loss, are added to accommodate the change of the
rings. The resonant wavelengths from both waveguides are then recombined at the first adiabatic 3-dB
coupler and constructively interfere as reflected signals while non-resonant wavelengths are recombined
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:40

(a) (b) (c) (d)

(e) (f) (g)

Figure 45 (Color online) Silicon Photonics chirped microwave signal generators reported in [178]. (a) Schematic view of a
silicon-based on-chip spectral shaper consisting of five MRR; (b) measured spectral response of the spectral shaper consisting of
five cascaded MRRs; (c) and (d) generated LCMW and its spectrogram when for the spectral shaper consisting of five cascaded
MRRs; (e) perspective view of the silicon-based on-chip optical spectral shaper incorporating LC-WBGs (inset: wire waveguide
(left) and rib waveguide (right)); (f) schematic layout of the designed on-chip spectral shaper, and image of the fabricated spectral
shaper with the length of the offset waveguide equal to the length of the LC-WBG captured by a microscope camera; (g) view of
an LC-WBG (inset: simulated fundamental TE mode profile of the rib waveguide with the rib width of 500 nm (left) and 650 nm
(right)).

at the second adiabatic 3-dB coupler at the transmission port.

In order to achieve maximum reflection power, each ring is designed to work in the critical coupling
condition. By carefully designing the ring radii, at the reflection port, the reflection peaks with a linearly
increasing spacing can be achieved. Figures 45(b)–(d) show, respectively, the spectral response, the
generated waveform, and its spectrogram at the reflection port of an on-chip spectral shaper consisting
of five MRRs.
While compared with the work in [175], the distinct feature of this optical spectral shaper is that the
MRRs are designed to have largely different radii for the generation of an LCMW with a much higher
chirp rate (17.2 GHz/ns compared to 8 GHz/ns), the TBWP of this configuration is quite small (18.7);
hence, for practical applications where large pulse compression is required, an LCMW with a much greater
TBWP is needed.
A spectral shaper designed to provide a much larger TBWP [178] is shown in Figure 45(e). The shaper
layout and photograph are shown in Figure 45(f). It has an MZI structure that incorporates two identical
linearly chirped waveguide Bragg gratings (LC-WBGs) with opposite chirp rates in its two arms. An
input light wave is split by a compact Y-branch into two beams to travel through the LC-WBGs in the
upper and lower arms. Spectral components of the light waves of different wavelengths are reflected from
different positions in the LC-WBGs. By the second and the third Y-branches, the reflected light waves
are collected and recombined at the fourth Y-branch. The combined light wave is guided to an output
grating coupler to couple the light out of the chip. When recombined, due to the optical interference,
the optical spectral shaper has a spectral response with a wavelength-dependent FSR, which is required
to generate an LCMW. The central frequency of the generated LCMW can be controlled by using an
offset waveguide added in the lower arm of the spectral shaper to control the length difference between
the two arms. By properly designing the LC-WBGs and choosing the length of the offset waveguide, the
spectral response of the spectral shaper can be controlled to have a symmetrical, uniformly increasing or
decreasing FSR. The two LC-WBGs are the key components in the spectral shaper. These gratings are
realized by introducing periodic sidewall corrugations on the slab as shown in Figure 45(g). Keeping the
grating period uniform and linearly increasing the width of the rib along the grating, a linear chirp is
produced since the effective refractive index is linearly increasing as the rib width increases in a definite
range. Using this design, LCMW with TBWP values over 200 have been reported.
Frequency combs. The generation of optical combs on a chip [179] enables several applications such as
the precise measurement of optical frequencies through direct referencing to microwave atomic clocks and
the production of multiple taps for high sidelobe rejection signal processors. Recently the generation of
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:41

(a) (b)

(c)

(d)

Figure 46 (Color online) (a) Top and (b) side views of a III-V on Si anti-colliding MLL comb source design (after [180]).
(c) Microscope image of the III-V-on-Si MLL. Insets: scanning electron microscope images of various constituent parts of the laser.
(d) Optical comb generated by the passively locked 1 GHz MLL. Inset: detail of evenly spaced optical modes in the comb.

optical frequency combs through the nonlinear process of continuous-wave optical-parametric oscillation
using micro-scale resonators or (also known as optical micro-combs) has attracted significant interest
since these devices have the potential to yield highly compact and frequency agile comb sources. Several
authors have reported the generation of optical frequency combs from highly-robust CMOS-compatible
integrated microresonator optical parametric oscillators (see a complete and updated review in [179]),
however we will not delve deeper into this approach as it is based on nonlinear effects, which are outside
the scope of this article.
Combs have been traditionally based on bulky and high-cost femtosecond lasers with line spacing in
the range of 100 MHz to 1 GHz. However, their integrated and low-cost counterparts, i.e., integrated
semiconductor mode-locked lasers, are limited by their large comb spacing, a small number of lines,
and broad optical linewidth. However, a recent study [180], has reported a demonstration of a III-V-
on-Si comb laser that can function as a compact, low-cost frequency comb generator after frequency
stabilization. Figure 46 shows the layout and the emitted spectrum of the integrated comb source under
optimum operating conditions.
The use of low-loss passive silicon waveguides enables the integration of a long laser cavity, which
enables the laser to be locked in the passive mode at a repetition rate of 1 GHz. The optical spectrum
features a 12 nm bandwidth comb structure with 10-dB output optical spectrum. The very small optical
mode spacing results in a dense optical comb of over 1400 equally spaced optical lines with sub-kHz 10-dB
radio frequency linewidth. Furthermore, the narrow longitudinal mode linewidth (<400 kHz) indicates
notably stable mode-locking.
The main problem with the former approach is the reconfigurability, as it is set by the cavity length
and might require the use of external phase filters to tune the repetition frequency. One solution, which
can also be integrated on a chip is to generate the comb spectrum by cascading amplitude and phase
modulators at the output of a continuous-wave laser source [181], driven by specially tailored RF wave-
forms to generate an optical frequency comb with very high spectral flatness. In [181] this technique has
been experimentally demonstrated and shown to lead to extremely flat comb spectrum with a spectral
power variation below 1 dB. The number of comb lines that can be generated is limited by the power
handling capability of the phase modulator, and this can be scaled without compromising the spectral
flatness.
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:42

(a) (b)

(c)

Figure 47 (Color online) Integrated 4-bit switched optical delay line (after [194]). (a) Circuit layout; (b) waveguide cross section;
(c) optical output for all delay settings.

Optical delay lines. Reconfigurable optical delay lines are classified into two main categories: variable-
length delay lines (VLDLs) and variable-propagation-velocity lines (VLDLs) based on optical filters.
VLDLs can be implemented by means of the so-called switched delay lines (SDLs), where the amount of
delay is reconfigured in discrete steps by selectively using integrated optical switches in a cascade of optical
waveguide paths of different lengths. SDLs have been proposed in a completely integrated photonics
form allowing increased compactness and improved stability versus thermal and mechanical stress, with
reduced delay switching times reaching a few tens of nanoseconds [109]. Higher delay resolution (required
for antennas operating at high frequency) can also be improved with an integrated photonics approach.
A number of photonic integration platforms have been used including polymer technology [182–184],
silica [185–188], LiNbO3 [189], GaAs [190, 191], and InP [192, 193]. Figure 47 illustrates the case of a
recent example [194] using an ultralow-loss Si3 N4 substrate featuring a fully integrated 4-bit TTD line
capable of delays above 12 ns, corresponding to about 2.4 m of propagation length, on a chip area of 4.5
cm×8.5 cm, with waveguide losses as low as 1 dB/m.
VPVL lines can be implemented using optical filters or dispersive optical transmission lines. Their
main advantage as compared to VLDLs resides in their capability for continuous delay tuning that
translates into the possibility of continuous steering of the antenna beam direction. Recent efforts have
been directed towards their on-chip implementations using photonic crystals, integrated Bragg gratings
and integrated filters. Photonic crystals have been employed as a single fully integrated dispersive delay
element. Sancho et al. [195] demonstrated the use of a single, very compact, low loss photonic crystal
to implement multiple variable TTDs. The operation is based on the slow-light properties of photonic
crystals, in which the group delay can be increased to very high values depending on the wavelength
demonstrating a 1.5-mm-long device capable of generating delays up to 70 ps with losses below 10 dB
over the complete 0–50-GHz band. More recently Zhang and Yao [196] have proposed an electronically
reconfigurable Bragg grating device that can operate as an electrically tunable linearly chirped grating
providing a dispersive delay line. The device layout is shown in Figures 48(a)–(d). It consists of multiple
series-connected uniform Bragg sub-grating sections and a Fabry-Perot (FP) cavity section in the center
of the grating. Each uniform Bragg sub-grating section incorporates an independent lateral PN junction
as shown in Figure 48(b), and between two neighbouring sections there is an undoped grating to function
as an insulator. Distributed electrodes are connected to the independent PN junctions.
By applying a bias voltage to a PN junction, the refractive index of the grating in that particular section
could be tuned locally based on the free-carrier plasma dispersion effect as seen in Figure 48(c). Thus, the
entire index modulation profile of the grating can be electrically reconfigured by field programming all
the bias voltages enabling the grating to have diverse spectral characteristics for diverse functionalities.
Since the uniform sub-gratings in the left and right sections can be independently tuned, by shifting the
spectral response of one of the two uniform sub-gratings with different bias voltages, the device can be
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:43

(b)
(a)

(c)

(d)

(e) (f) (g)

(h) (i)

Figure 48 (Color online) Electrically tunable silicon Bragg grating (after [195]). (a) Perspective view of the grating on a silicon
chip. (b) Cross-sectional view of the grating rib waveguide. (c) Top-view of the grating. (d)–(g) Microscope camera images of the
fabricated grating, the input grating coupler and compact Y-branch, the FP cavity section, and the transmission and reflection
grating couplers. (h) Reflection and transmission spectra when the device is reconfigured to be a chirped grating. (i) Ime-domain
measurement when the fabricated grating is reconfigured to be a chirped grating.

reconfigured to be a chirped grating.

Figure 48(h) presents the measured reflection and transmission spectra of the chirped grating when a
maximum reverse bias voltage of +20 V is applied to the left PN junction and a forward bias voltage
of −1.04 V is applied to the right PN junction. The 3-dB bandwidth of the spectra is increased to be
1.29 nm, which is larger than the 0.79 nm corresponding to the uniform grating setting. Figure 48(i)
demonstrates the time delay function, a microwave rectangular pulse with a temporal width of 290 ps is
modulated on two optical carriers with different wavelengths and sent to the grating. The two reflected
pulses experience two different time delays since they are carried by two different wavelengths at 1537.74
and 1538.54 nm. Specifically, the pulse carried by the optical wavelength at 1537.74 nm is reflected
by the right sub-grating from its center. The pulse carried by the optical wavelength at 1538.54 nm is
reflected by the left sub-grating from its center. The time delay difference is 15 ps. Larger time delays are
possible if the length of the grating is increased, and more independent grating sections are incorporated
for high-level tuning in terms of group delay and chirp rate.
Integrated optical filters have also been used to implement broadband tunable optical delays. For
example, cascaded optical ring resonators (ORRs) have been demonstrated in silica [185, 188], TriPleX
[196–198] and SOI [199, 200]. This solution combines optical ring resonators comprising tunable phase
shifts in the ring loops and tunable couplers. The basic principle is that the optical ring resonators
can be configured to be an optical all-pass filter [159, 198], as shown in Figure 49(a), featuring a group
delay at the resonance frequency, which can be set to be much larger than the round-trip delay of the
ring loop and can be controlled via the coupler. This enables the realization of long delays using a
small chip area, which can be further increased by cascading multiple RRs in series to form a SCISSOR
structure, as illustrated in Figure 49(b). A limitation of this scheme resides in the fact that the delay ×
bandwidth product is constant and thus, increasing the available delay at resonance implies a reduction
in the available bandwidth. This can be overcome by slightly detuning the resonances of the optical
ring resonators in the cascade as reported and experimentally demonstrated in [201] for an 8 ORR
SCISSOR structure fabricated in Si3 N4 . Figure 49(c) provides the basic waveguide structure, individual
and SCISSOR layouts and a photograph of one of the fabricated optical ring resonators, including the
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:44

(a) (b)

(c) (d)

Figure 49 (Color online) Tunable resonant delay line based on optical ring resonators acting as an all-pass filter (after [196]).
(a) Single ORR structure; (b) cascade ORR structure; (c) details from an 8 ORR SCISSOR fabricated in Si3 N4 (after [201]);
(d) bandwidth increase by slightly detuning the resonances of the rings I (c) (after [201]).

electrodes placed on top of the ring waveguide to tune its resonance through the thermo optic effect.
Finally, a novel concept to implement OTTDLs is based on the principle of index-variable delay lines.
Here true-time delay control can be realized through varying the group index/ propagation velocity in the
waveguides. One efficient way to achieve this is by the inscription of subwavelength gratings (SWGs) [202]
in SOI waveguides by alternating sections of silicon and silica. These waveguides can provide a variable
group delay by changing the duty cycle relationship between the silicon section within the grating period.
In [202], a 4-tap OTTDL is reported where each tap is based on one SWG waveguide, and all waveguides
have the same length of L = 8. The SWG waveguides are realized by alternating periodically segments
of silicon and silica, with a period of Λ = 250 nm along the propagation direction. Figure 50 shows the
schematic and performance operation for an input 1 ps Gaussian pulse featuring a 9 ps incremental delay
between waveguides by choosing four different values of the SWG duty cycles (60%, 50%, 40% and 30%).
Phase shifters. In addition to delay, phase shifting is also attractive for a number of signal processing
applications. Tunable RF phase shifters have been demonstrated using SOI [203, 204] and hybrid SOI
and III-V semiconductor ring resonators [205]. Here the phase shift is obtained after beating the optical
carrier and the single sideband RF modulating signal [206]. The principle is shown in Figure 51 for a
structure composed of two optical ring resonators in cascade (to achieve a full 2π phase excursion [203]).
One of the two carriers is placed in coincidence with the resonance of the all-pass transfer function of
the resonator where it experiences a different phase-shift compared to the component that is outside the
resonance. Tuning is achieved either by moving the resonance of the ring cavity through an internal
phase shifter or by tuning the optical carrier.
Both tunable time delay and phase shift operations can be incorporated into a single photonic chip
using the separate carrier tuning (SCT) scheme, proposed by Morton and Khurgin [207]. Burla and co-
workers [208] have reported and demonstrated the SCT scheme altogether with optical single sideband
filtering monolithically integrated in a single chip. The circuit, shown in Figure 52(a) consists of a
reconfigurable optical delay line, a separate carrier tuning unit, and an optical sideband filter. The optical
sideband filter, a RAMZI, removes one of the radio frequency sidebands of a double-sideband intensity-
modulated optical carrier. The ODL and separate carrier tuning units are individually implemented
using a pair of cascaded optical ring resonators. The group delay of the signal sideband is changed by
tuning the resonance frequencies and the coupling factor of the optical ring resonators in ODL. This is
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:45

(a)

(b)
(c) (d)

(e) (f) (g)

Figure 50 (Color online) (a) Schematic and (b) an SEM image of our fabricated 4-tap OTTDL structure based on SWG waveguides
in SOI reported in [202]; (c) waveguide cross-section and (d) parameters of the SWG waveguides; (e) measured power spectral
response of the fabricated OTTDL; (f) generated time-domain pulse train at the output of the fabricated OTTDL device in
response to a single input optical pulse; (g) input and amplified output signal spectra.

Figure 51 (Color online) Illustration of the basic principle behind the implementation of a tunable RF phase shifter using a
resonant integrated resonator and a single sideband modulated signal (after [206]). The tuning of the chip resonance is here
performed using a voltage activated phase-shifter inside the cavities.

supplemented by applying an additional phase shift in the [0, 2π] range to the carrier. In [208], this
module was employed to demonstrate of a two-tap microwave photonic filter whose notch position could
be shifted by 360◦ over a bandwidth of 1 GHz. The operation of the different modules in the chip is
depicted in Figures 52(b)–(d).
Waveguide Bragg gratings realized on SOI can also implement tunable ultrabroadband phase shifts in
conjunction with optical single-sideband modulation. In [207], a compact (130-nm long) device has been
employed to create broadband phase shifts for RF frequencies beyond 900 GHz.
Beamformers. In phased array antennas a beam is formed and steered to a given angle by adjusting
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:46

(b)

(a)

(c)

(d)

Figure 52 (Color online) Reconfigurable circuit simultaneously implementing tunable phase shifting and delays using the separate
carrier tuning technique (after [208]). (a) Circuit layout; (b) operation of the tunable SSB filter; (c) operation of the ODL for the
radiofrequency sideband; (d) operation of the phase shifter for the optical carrier.

the phase relationship between a given number of radiating elements [1–3]. This principle of operation
works correctly for narrowband signals but, in the case of wideband signals, a limitation arises since if
a constant phase shift is produced from element to element, the beam pointing is different for different
frequency components. This deleterious effect, known as beam squint, can be compensated by using
tunable delay lines, rather than by employing phase shifters. Thus, it turns out that both delay lines
and phase shifters can be advantageously employed to implement beam steering circuits, each one in a
particular context. Attending to the different types of delay lines and phase shifters considered in the
prior section there are three categories of integrated photonic beam formers that can be implemented:
wideband beam formers based on discretely tunable delay lines TTD [209], wideband beam formers based
on continuously tunable delay lines TTD [210–217] and narrowband beam formers based on optical phase
shifters [189, 218, 219].
Figure 53 shows the configuration and some measured results of a complete tunable broadband beam
former based on optical ring resonators arranged in a binary-tree network architecture reported in [216]
and fabricated in TriPleX waveguide technology. The beamformer was composed of a 4-element array
shown in Figure 53(a), subsequently completed by 1 × 4 RF phase shifting stages to feed a final 1 ×
16 linear antenna array as shown in Figure 53(b). Basic true-time delays are obtained using the ODL
configuration shown in Figure 53(c). Delays up to 236 ps were demonstrated over a bandwidth of 2
GHz for Ku-Band satellite communication applications [216] as shown in Figure 53(c). The beam former
elements could be configured as true delay lines or as pure phase shifters. In the first case it provided
squint-free operation in the (10.7–12.75) GHz, while in the second the quint effect was clearly appreciable,
as shown in Figure 53(d).

6.3 Programmable integrated microwave photonics

The fact that several ASPIC designs have proven to operate successfully as multifunctional circuits [220]
raise the question of whether it is possible to design a simple and replicable photonic hardware architecture
that under suitable programming could implement the different functionalities required in MWP. This
so-called programmable integrated microwave photonics would be inspired on similar grounds as the field
programmable gate array (FPGA) device in electronics or the digital signal processor (DSP) in advanced
communications. In this section we provide some insight into the recent and currently undergoing efforts
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:47

(a) (b)

(c)

(d)

Figure 53 (Color online) Reconfigurable delay and phase 1 × 16 beamformer (after [216]). (a) Reconfigurable optical 1 × 4
circuit; (b) complete beam forming network; (c) phase shift vs. frequency in the four outputs of the optical beam forming stage;
(d) squint and squint free operation in the Ku band for a targeted pointing angle of 30◦ .

RF output RF input

Control electronics

Optical RF switch RF
RF filter amplifier
source
RF core

Reconfigurable
Reconfigurable
Coupler optical Coupler
optical source External Detector
filtering
spectral shaper modulator system
system

Optical source Optical

subsystem core

Control electronics

Optical auxiliary input Opticall auxilia

auxiliary output

Figure 54 (Color online) Layout of an MWP transistor (after [221]).

in this field.
Early efforts. The starting point for the MWP programmable processor can be traced back to the
proposal of the MWP transistor [221] shown in Figure 54, which is composed of subsystems, each of
which is a collection of connected fixed-and-variable components.
The transistor configuration is obtained by opening the feedback loop of a generalized optoelectronic
oscillator. As it was shown in [221], this architecture is in principle capable of performing few, but
important, required functionalities in MWP provided that its optical subsystems can be reconfigured by
means of suitable control signals. The main limitation of the MWP transistor is however that it does not
support the reconfiguration of interconnections between its internal optical subsystems. This results in
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:48

(a)
RF input RF output
AUX
optical
input RF Interface & core

Reconfigurable
AUX
Optical optical core
E/O ORSE ORSE O/E optical
source
ORSE output

Software control C=[c1, c2, c3, ..., cn]

Optical ports
(b)

Rings Rings
switching elements (ORSE)
switching elements (ORSE)

switching elements (ORSE)

Optical routing and
Optical routing and

Optical routing and

MZIs MZIs
Optical ports

Optical ports
Gratings Gratings

Optical ports

Figure 55 (Color online) (a) Layout of the universal processor (after [222]). (b) Particular implementation of the optical core
using ad-hoc photonic components such as Mach-Zehnder interferometers (MZIs), ORRs and gratings connected by several optical
routing and switching elements (ORSEs).

an intermediate performance between ASPIC and a universal MWP processor.

As a first alternative to going one step further in terms of flexibility, the former approach can be
extended to implement a programmable MWP processor architecture [222], which incorporates optical
routing and switching elements (ORSEs) into the MWP transistor layout and replaces the reconfigurable
optical filtering system by a more powerful and versatile optical core. Figure 55(a) schematizes the
concept.
The central element of the programmable optical processor is the optical core, where the main signal
processing tasks are carried out in the photonic domain. Ideally, the optical core should be built upon a
versatile architecture capable of implementing different functionalities in response to different electronic
control signals. In practice, among the reconfigurable photonic circuits proposed in the literature, the
ones that could especially emerge as candidates for this core are either based on the cascade of finite
(i.e., Mach-Zehnder interferometers) or infinite (i.e., ring cavity) impulse response cells or a combination
of both. These structures feature periodic filters and allow for bandwidth reconfiguration and notch
tunability by moving the zeros and poles along the z-plane. Thus, to further increase the degree of
flexibility of the processor, the former configuration can be extended with a subsystem of switched optical
signal processing elements including different types of fixed and reconfigurable filtering structures and
interconnections, as illustrated in Figure 55(b). This example includes FIR filters, IIR and dispersive
delay lines (DDL) that can be accessible by the correct programming of the ORSEs that interconnect
them. The optical routing and switching elements would need to have medium-count ports depending on
the number of component banks. This architecture, as discussed in [223], is still limited in (a) flexibility,
as ad-hoc designed photonic subsystems must be included and interconnected within the optical core,
(b) scalability, as fixed size photonic components are the building blocks for more complex circuits and
(c) loss, as a considerable number of ORSEs need to be traversed to implement complex configurations.
Waveguide mesh core processors. A more versatile optical core architecture can be obtained by following
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:49

(a) (b) (c)

TBU

Mesh
interconnection- 1 2 1 2
1 3
node
2
6 3 4
3 4
5

Figure 56 (Color online) Reconfigurable mesh designs [226] (upper): (a) Hexagonal type, (b) triangular type, (c) square type,
and their associated interconnections points (bottom).

similar principles as those of the FPGAs in electronics [224]. The central concept is based on breaking
down complex circuits into a large network of identical two-dimensional (2D) unit cells implemented by
means of an MZI waveguide mesh or lattice. Zhuang and co-workers [225] have been the first to propose
a programmable optical chip architecture connecting MZI devices in a square-shaped mesh network
grid. The distinctive feature of this approach is that it enables both feedforward and feedbackward
filtering configurations, selecting the adequate path through the mesh and providing independent tuning
of circuit parameters to complex valued coefficients by introducing phase tuning elements in both arms
of the MZIs to enable independent control of amplitude and phase of light at coupler outputs. Further
to the square shaped mesh, other configurations, such as the triangular and hexagonal geometries have
been reported [226].
Figure 56 shows three different types of waveguide mesh designs. Figures 56(a)–(c) illustrate a 4-cell
arrangement for the hexagonal, triangular, and square-type meshes, respectively.
The basic building block of these meshes is a tunable coupler that must provide, independently, a
complete splitting ratio tuning and phase response. This switching/tapping/de-phasing mechanism can
be obtained either by exploiting the electro-refraction and/or the electro-absorption effect, or by means
of the thermo-optic effect in a broadband 3-dB balanced MZI. By configuring each tunable coupler placed
at each side of the cell perimeter as a switch (in either its cross or bar state) or as a tunable coupler, one
can synthesize a given photonic integrated circuit topology, as shown in [227].
The upper part of Figure 57(a) shows the tunable basic unit (TBU) composed of the tunable coupler
and its access (input/output) waveguides. The geometry of the latter is a function of the bending radius
and varies for each mesh topology due to different angles between elements. The basic unit length (BUL)
is
BUL = Laccess + LTunable-Coupler, (54)
where Laccess is the overall length of the access waveguide segment and LTunable-Coupler is the length of
the tunable coupler.
Referring to Figure 57(b), the tunable basic unit can implement 3 different states: cross state switch
(light path connects in1 to out2 and in2 to out1 ), bar state switch (light path connects in1 to out1 and in2
to out2 ) and tunable splitter. For a balanced MZI loaded with heaters on both arms, the splitting ratio
is obtained by increasing the effective index due to the Joule effect in the upper or lower arm, producing
a ϕupper and ϕlower phase shift, respectively. Once set, a common drive in both heaters will provide a
common phase shift, leading to independent control of the amplitude ratio and the phase. The device
matrix is defined by !
j∆ sin θ cos θ
hTBU = je γ, (55)
cos θ − sin θ
where θ is (ϕupper − ϕlower )/2 and ∆ is (ϕupper + ϕlower )/2. The coupling factor K is then defined as the
cos2 (θ) and the general loss term γ includes the effect of propagation losses in the access waveguides, the
tunable coupler waveguide and the insertion losses for both 3-dB couplers.
The different waveguide mesh geometries have been analyzed and compared against a set of different
figures of merit in [226], concluding that the hexagonal lattice yields the best performance for almost all
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:50

Laccess Laccess

2 LTunable-Coupler 2
in1 out1
Tunable coupler
in2 out2
BULLL=Laccess+LTunable-Coupler

BUL

φupper
3-dB 3-dB
in1 Coupler out1
Coupler
in2 Heaters
Access out2
waveguide φlower

Control electronics
(a)

in1 out1 in1 out1 in1 out1

in2 out2 in2 out2 in2 out2
θ=0 θ = [0, π/2] θ = π/2
Cross-state (CS) Tunable coupler (TC) Bar-state (BS)
(b)

Figure 57 (Color online) (a) (Upper) Labeled schematic of a general tunable coupler acting as the basic building block of the
mesh. The basic unit length (BUL) is illustrated as the sum of the tunable coupler length and the arc length of the access
waveguides. (Lower) Particular case of an integrated balanced MZI-based tunable coupler. (b) Signal flow for the different tunable
basic unit (TBU) configuration states (after [227]).

the performance metrics. Table 2 summarizes the results of that analysis.

For reference, the spatial tuning resolution step quantifies which is the minimum step in BUL units
by which the arm length mismatch or the cavity length can be increased/decreased, the reconfiguration
performance of the mesh is given by the number of filters with different spectral period values that
can be implemented given a maximum value X (in BUL units) for the path imbalance in unbalanced
Mach-Zehnder interferometers (UMZIs) or the cavity length in optical ring resonators. The replication
flexibility gives the number of possible alternative geometries for a specific filter implementation, while
the switching elements per unit area denote the amount of required BULs per surface unit to achieve
equivalent functionality. Finally, the two last rows in Table 2 quantify the overall losses in the TBUs of
triangular and hexagonal waveguide meshes as compared to those in the square mesh configuration. A
complete definition of the figures of merit can be found in [226].
With suitable programming of a selected set of TBUs in the 2D waveguide mesh, different circuit
layouts of variable complexity can be programmed over the same hardware. For example, Figures 58(a)
and (b) show two implementation examples of simple UMZI FIR filters and optical ring resonator IIR
filters with different path unbalance and cavity lengths, respectively. Filters with the increased degree of
complexity can be built upon cascading of these simple units.
Since waveguide mesh processors are a quite recent area of research, there are only a few experimental
results reported based on silicon nitride [225] and SOI [227] technologies.
Figure 59(a) shows the basic layout and photograph of the programmable optical chip architecture
connecting MZI devices in a square-shaped mesh network grid proposed by Zhuang and co-workers
[225]. The structure, fabricated in Si3 N4 , comprised two square cells and was employed to demonstrate
simple FIR and IIR impulse response filters with single and/or double input/output ports of synthetized
optical ring resonators. The processor featured an FSR of 14 GHz and is fully programmable. By
appropriate programming of this processor, Zhuang et al. have demonstrated bandpass filters with a
tunable center frequency that spans two octaves (1.6–6 GHz) and a reconfigurable band shape (including
flat-top resonance with up to passband-stopband 25 dB extinction). They also demonstrated notch filters
with up to 55 dB rejection ratio, Hilbert transformers and tunable delay lines as shown in Figure 59(b).
A recent contribution has reported the results of a waveguide mesh composed of 7 hexagonal cells
fabricated in silicon on insulator [227]. The chip photograph is shown in Figure 60. The device was
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:51

Table 2 Summary of values for the figures of merit of the different mesh designs (see the definitions in [226])

Figure of merit Triangular Square Hexagonal

ORR cavity spatial tuning 3 4 2*
resolution step in BUL The first and second steps
units (the lower the better) have a resolution of 6 and 4
MZI arm imbalance spatial tuning resolution 3 4 2
step in BUL units (the lower the better)
ORR reconfiguration performance (the higher 8 6 9
the better) (for X = 25 BUL)
MZI reconfiguration performance 8 6 12
(for X = 25 BUL)
Switching elements per unit area compared to +65.00% 0.00% −36.66%
square % (the lower the better for a fixed
value of reconfiguration performance)
Replication ratio for ORR structures up to 16 BUL 1 2.68 1.31
cavity length (the higher the better)
Replication ratio for MZI structures up to 12 BUL 1 3 3.36
interferometric length (the higher the better)
Laccess /Laccesssquare % for a fixed radius of curvature Ra +33.33% +0.00% −33.33%
(the lower the better)
Ra/Rasquare % for a fixed BUL value −25.00% +0.00% +50.00%
(the higher the better)

(a) Reconfigurable optical core (c) Targeted PIC: UMZI filters

5 8 8
5

1 TC 1 2
TC TC 2 ΔL = 4 BUL ΔL = 6 BUL
TC 2
TC
2
TC
4 TC Phase shifter
4
ΔL = 2 BUL

Not used Tunable

Cross State Bar State
(available) Coupler (TC)

(b) Reconfigurable optical core (d) Targeted PIC: ORRs filters

12 10 12
10

ΔLORR = 12 BUL ΔLORR = 12 BUL

TC TC
6 6
TC
ΔLORR = 6 BUL Phase shifter
TC

Not used Tunable

Cross State Bar State
(available) coupler (TC)

Figure 58 (Color online) (a) and (b) FIR filter implementations using a hexagonal waveguide mesh. Mesh setting for (c) three
different targeted UMZI filters and (d) three different targeted ORRs [226].

fabricated at the Southampton Nanofabrication Centre at the University of Southampton. SOI wafers
with a 220-nm thick silicon overlayer and a 3-µm thick buried oxide layer were used (for more details on
fabrication and testing, see [227]).
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:52

(a) (b)

Figure 59 (Color online) (a) Schematic and a photo of the Si3 N4 waveguide technology (TriPleX) chip implementing a 2 square
cell waveguide mesh reported in [225]; (b) different programmed circuit configurations obtained by varying phase-tuning elements
in the chip and the measurements of their corresponding frequency responses.

Figure 60 (Color online) Photograph of the 7-hexagonal cell waveguide mesh core fabricated in SOI and reported in [227].

The overall structure comprised 30 independent MZI devices and 60 thermo optic heaters. The waveg-
uide mesh MZI devices were independently tuned in power splitting ratio and overall phase shift by means
of current injection to the heaters deposited on top of the waveguides implementing the interferometer
arms. Despite the simplicity of the layout depicted in Figure 61, the 7-cell structure can implement over
100 different circuits. For MWP filtering applications basic MZI, FIR transversal filters, basic tunable
ring cavities and IIR filters, as well as compound structures such as CROWS and side-coupled integrated
spaced sequence of resonators (SCISSORS) are of particular interest. In [227], a different configuration
featuring a wide variety of FSR values was reported. Figure 61 shows, for instance, 5 examples of mi-
crowave photonic FIR and IIR filters synthesized with the same hardware just by changing the operation
state point of the different MZI elements in the mesh.
Another application of interest in microwave photonics is true-time delay lines as these are the basic
building blocks of key functionalities, including arbitrary waveform generation and optical beamforming.
The hexagonal waveguide mesh core can implement both dispersive as well as discrete time programmable
true-time delay lines [228].
Waveguide mesh MWP universal processors. The incorporation of the waveguide mesh structure in
the MWP processor core enables a more versatile processor architecture, which is the one we propose
for future development. In this architecture, which we show in Figure 62, all the subsystems are directly
connected to the reconfigurable optical core. This architecture is the first proposal for a generic-purpose
software-defined MWP processor that incorporates the possibility of modifying the control signal flow to
reconfigure the processing and interconnecting subsystems as required. It supports four different modes
of operation as far as the input/output signals are concerned: electrical/electrical, electrical/optical,
optical/electrical, and optical/optical operations.
Electrical/electrical operations are typically employed in MWP functionalities such as RF-filtering,
instantaneous frequency measurement, frequency mixing, RF and millimeter-wave arbitrary signal gen-
eration to cite a few. It requires the processor to enable an optical source, electro-optic (EO) and
optoelectronic (OE) converters as well as the reconfigurable optical core. Figures 62(a)–(d) illustrate the
signal flows for these operations. Note that if a second modulator is integrated, it can be enabled to
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:53

Lattice-filter: C = 1570.13 nm
(a) CS BS 0 MZI2: C = 1570.13 nm
AV TC 2-cascaded unbalanced mach- 0

Opt. Power (dB)

transmission (dB)
zehnder interferometer

Norm. optical
OUT1 −10
OUT2 (ΔL=2 BUL) -20
K1 K2 K3 −20
IN 2 OUT1
2 -40
4 4 OUT2 −30 -20 0 20
IN
Tunable coupler FSR=36.8 GHz Relative optical
−40 -30 -20 -10 0 10 20 30 frequency (GHz)
Relative optical frequency (GHz)
(b) CS BS Transversal filter
Trans-Filter: C = 1570.07 nm
AV TC (ΔL=4 BUL) −10

transmission (dB)
Norm. optical
11

7 −20
K2

K3

K1 K4
OUT1 −30
OUT2
IN IN 3 OUT2
OUT1 -20 -10 0 10 20
Tunable coupler
Relative optical frequency (GHz)
CS BS SCISSOR (L =6 BUL, N=2) SCISSOR6 N2: C = 1570.13 nm
(c) AV TC 0
OUT2 BW3-dB= 4.65 GHz
transmission (dB)
Norm. optical

K1u K2u −10 BW3-dB= 5.30 GHz

BW3-dB= 5.90 GHz
OUT2 −20
K1d K2d
OUT1 SLL = 7.5 dB
* PS IN −30
SLL = 16.8 dB
Tunable coupler
IN −40 SLL = 36.0 dB
OUT1 Phase shifter 0 10 20 30 40
Relative optical frequency (GHz)
CROW6 N2. C = 1570.13 nm
CROW (L=6 BUL, N=2)
(d) OUT2 −18
transmission (dB)

OUT2 K3 −20
Norm. optical

BW3-dB= 5.05 GHz

−22 BW3-dB= 3.60 GHz
PS
−24 BW3-dB= 4.40 GHz
K2
−26
K1
OUT1 IN −28
Tunable coupler −30
OUT1 IN
Phase shifter 0 10 20 30 40
Relative optical frequency (GHz)
CROW (6 BUL)
CS BS CROW6 N3. C = 1570.13 nm
(e) AV TC OUT2
K4 −20
OUT2 BW3-dB= 6.05 GHz
K3
transmission (dB)
Norm. optical

PS −25
K2

K1 −30
OUT1 IN 0 10 20 30 40
OUT1 IN Relative optical frequency (GHz)
Tunable coupler

Figure 61 (Color online) Five examples of synthesized RF-photonic filters using the same 7-cell hexagonal mesh. In each case,
the left part shows the TBU settings in the mesh. The intermediate part shows the circuit layout and the right part shows the
measured results for the RF transfer function modules. (a) Two-cascaded unbalanced 2-MZI FIR filter; (b) three-tap transversal
FIR filter; (c) two-cavity SCISSOR FIR filter; (d) two-cavity CROW filter; (e) three-cavity CROW Filter. MZI programming code:
CS: cross state switch; BS: bar state switch; TC: tunable coupler; and AV: available (unused) [227].

perform frequency mixing operations based on the cascade of two EO modulators.

Sometimes, the processed signal has to be distributed over a particular fiber link length after generation
and/or processing. The processor can leverage the inherent properties of optical fibers for distribution
purposes. The electrical/optical mode is widely employed in radio-over-fiber MWPLs. At the receiver
point of the link, another multipurpose MWP processor can be employed. In this case, the receiver would
be working in optical/electrical mode, processing the signal before the photo-detection. Optionally, the
receiver can enable its own optical source to act as a local oscillator as well. Both modes of operation
are displayed in Figures 62(b) and (c), respectively.
The last mode of operation, illustrated in Figure 62(d), is the optical input/optical output. In this
case, the input signal can be processed directly in the optical domain. Optical channel management can
then perform common optical processing operations such as add/drop, switching, and broadcasting. Note
that all the previous modes of operation may coexist for certain multi-task functionality. For example, a
modulated signal could be divided after being processed and both distributed through the optical ports
and down-converted by the photodetectors.
We have shown that the hexagonal waveguide core is the best candidate to perform the reconfigurable
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:54

Electrical Electrical Electrical Electrical

input/s output/s input/s output/s

(a) (b)
RF & high-speed electronics interface RF & high-speed electronics interface

E/O E/O E/O E/O

Optical O/E Optical O/E

source Optical source
Optical
Optical filter Optical filter
input/s Optical input/s Optical
output/s output/s
Control electronics & software Control electronics & software

Electrical Electrical Electrical Electrical

input/s output/s input/s output/s
(d)
(c)
RF & high-speed electronics interface RF & high-speed electronics Interface

E/O E/O E/O E/O

Optical O/E Optical O/E

source Switches Dispersion
source Optical
Coupler compensators
Optical filter Optical O. Filters
input/s Optical input/s Optical
output/s output/s
Control electronics & software Control electronics & software

Figure 62 (Color online) Generic purpose MWP processor architecture programmed different modes of operation. (a) Electri-
cal/electrical, (b) electrical/optical, (c) optical/electrical and (d) optical/optical [228].

optical core task: featuring subsystems interconnection as well as enabling versatile optical processing
operations. By suitable programming each element on the processor, including the reconfigurable optical
core, the same hardware can be configured to perform all the main functionalities required in MWP, as
we illustrate in Figures 63 and 64.
True-time delay lines TTDs. Discretely and continuously tunable DLs [229, 230] can be supported by
the MWP processor. Figure 63(a) illustrates, for example, the processor settings for the implementation
of a discrete true-time delay based on the SCT approach [231]. Note that four optical ring resonators
need to be enabled for the carrier tuning and optical delay line sections, as well as a double-loaded MZI
for the implementation of the optical single side-band filter.
Filtering applications. The same hardware can be programmed as well to perform versatile RF-
photonic tunable filtering. Here, the spectral characteristics of the optical filter implemented by the
optical waveguide mesh core are directly translated into the RF spectrum by means of a self-beating
homodyne technique, where part of the un-modulated optical carrier is sent separately and combined
with the RF-modulated and optically filtered prior to photodetection, as illustrated in Figure 63(b) [231].
In this example, the synthesized RF filter is implemented by means of 6 cascaded optical ring resonators.
Microwave and mm-wave tone generation. This functionality enables the photonic generation and
distribution of high-frequency RF signals, [232]. Among the different approaches reported, we illustrate
here the implementation of two. Figure 63(c) depicts the programmable processor employing the external
modulator with different modulation biasing points together with an optical filter to suppress the optical
carrier and the undesired (odd or even) modes. The filter is implemented here with a fourth-order lattice
of MZIs. The second approach consists of programming an optoelectronic oscillator [46] by closing the
feedback loop of the RF section, as shown in Figure 63(d). In this example, the reconfigurable optical
core implements a High-Q optical storage section with seven optical ring resonators.
Arbitrary waveform generation. Wavelength-to-time mapping is a promising solution and can be
integrated on a chip. A broadband optical source is filtered by a spectral shaper. The spectral shape can
be translated to the time domain by a dispersive element. Figure 64(a) illustrates the processor settings
to configure the spectral shaper by means of five cascaded optical ring resonators [175], and a dispersive
element implemented by a fourth-order lattice filter [232]. Note that versatility is the most valuable
property of the spectral shaper, and the hexagonal waveguide processor core suits perfectly for this task,
as demonstrated. Higher dispersive delays might be required and could be obtained by employing the
optical ports for the connection of dispersive fiber reels.
Beamforming networks. The use of tunable delay lines allows a stable pointing angle independent of
the transmitted frequency, enabling broadband operation. Figure 64(b) illustrates the implementation of
a beamformer based on the natural discrete delay lines of the hexagonal core. By modifying the length of
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:55

(a)
RF & High-speed electronics

fRF1 fLO CT: carrier tuning

fRF? ODL: optical delay line
E/O E/O
O/E
Optical
source O/E CT ODL O/E
Optical C
E/O
source CT ODL O/E
O/E
Bias OSB
O/E Tunable coupler filter
Phase shifter
Optical ports
(b)
RF & High-speed electronics H( fRF)
Combiner H(λ)
fRF1 fLO fRF
Splitter
λ fRF
E/O E/O 6 6
C
O/E E/O
Optical Optical 6 6 6 6
source O/E O/E
source C
O/E
O/E Tunable coupler
Phase shifter

Optical ports

(c) 2fRF
RF & High-speed electronics
C H(λ) f
fRF1 fLO fRF C+2fRF
fRF
λ
C+fRF
E/O E/O 2 2 2 2
Optical C
O/E E/O O/E
Optical source
source O/E 4 4 4 4
Tunable coupler
O/E
Phase shifter
O/E

Optical ports

(d) fRF,out
RF
RF & High-speed electronics RF Amp. Filter RF Amp.
coupler

fRF1 fLO

E/O E/O fRF

O/E 6 6 6 6
Optical Optical
source O/E E/O 6 6 O/E
source 6 6
O/E
Tunable coupler High-Q optical storage
O/E Phase shifter

Optical ports

Figure 63 (Color online) (Left) Examples of MWP processor settings for the implementation of different MWP functionalities.
For each functionality the circuit schemes are shown on the right hand side: (a) true time delay line employing the single carrier
tuning (SCT) technique, (b) self-homodyne filter FIR, microwave and mm-wave tones generation based on (c) external modulation,
and (d) optoelectronic oscillatio [228].

the paths, the differential delay can be changed. The tilt angle tuning will be limited to discrete values
in the range of [−π, π].
Instantaneous frequency measurement. These systems are employed to measure the frequency of
microwave signals in real-time applications. Figure 64(c) depicts one of the approaches that consist in
mapping the unknown microwave frequency to an optical power ratio by establishing a unique relationship,
commonly referred to as the amplitude comparison function (ACF), that is independent of both the laser
and input RF powers. The ACF can be performed by a 2×2 complementary filter [233]. In this case,
the filter is implemented by means of 2 cascaded balanced MZIs loaded with a total of 4 optical ring
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:56

(a) H(λ) H(λ) H(t)

RF & High-speed electronics

fRF1 fLO λ λ Time

E/O E/O 1 5 1 5
O/E 6 6
Optical Optical O/E
source O/E source 6 6 6
5 1 5 1
O/E
Tunable coupler Spectral Dispersive
O/E Phase shifter shaper subsystem

Optical ports
(b)
RF & High-speed electronics

fRF1 fLO O/E

ΔL
fRF O/E
E/O E/O ΔL
O/E O/E
Optical ΔL
Optical O/E
source O/E E/O
source
O/E ΔL = 2 BULs

O/E Tunable coupler

Optical ports
(c)
RF & High-speed electronics

fRF1 fLO HACF( fRF)

H(λ)
fRF?
E/O E/O
C−fRF C+fRF λ fRF
O/E
Optical
source O/E C O/E
Optical C
E/O
source O/E
O/E
O/E Tunable coupler
Phase shifter

Optical ports

RF & High-speed electronics (d)

fRF fLO C
fRF1 fLO
fRF
C+fRF+fLO
E/O E/O 2 2 2 2
Optical C
O/E E/O E/O O/E
Optical source
source O/E 4 4 4
4
O/E Tunable coupler C C+fRF

Phase shifter
O/E

Optical ports

Figure 64 (Color online) (Left) General-purpose photonic processor settings for the implementation of a particular MWP scheme
(right): (a) arbitrary waveform generation based on wavelength-to-time mapping, (b) beamforming networks based on true time
delay, (c) instantaneous frequency measurement, and (d) RF-mixing based on 2 cascaded modulators [228].

resonators. Sharper filter functions lead to an increase in the resolution at the expense of reducing
the frequency range. A key advantage of the reconfigurable processor is that the ACF can be made
reconfigurable, so multiple measurements can be done while reconfiguring an ACF for each frequency
range of measurement.
Frequency mixing. Frequency up-and/or down-conversion of RF signals is typically required in many
radio-over-fiber and intermediate-frequency-over-fiber systems [234]. Figure 64(d) illustrates the processor
configuration to achieve this functionality, where an optional optical filter, this time implemented by a
fourth-order lattice filter, is employed to suppress the possible intermodulation contributions prior to the
second modulator input.
For some of the previous functionalities, the integration of an optical amplifier subsystem or an optical
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:57

Figure 65 (Color online) Analog signal processing engine in the context of ICT systems. This concept is transversal and enables
different applications for microwave photonics systems.

amplifier array prior to photodetection would be required to overcome the electrooptic and optoelectronic
conversion losses.
The key challenge in implementing a microwave photonic signal processor is the relatively high loss,
which would limit the scale of the photonic signal processor and would make it not ready to replace a
state-of-the-art electronic signal processor. However, with the fast advances in PIC technology, especially
the maturity of the heterogeneous integration techniques, a PIC that incorporates both photonic and
electronic (including both analog and digital) devices would outperform the current electronic-only signal
processors and would find practical applications in communications and radar systems.

7 Applications
Microwave photonics technology found initial niche applications mainly in security and defense systems [4].
However, with information and communication technology (ICT) systems expanding at an awesome pace
in terms of capacity demand, the number of connected end-users, and required infrastructure, it has
become evident that there is and will be in the forthcoming years a need for a flexible, scalable, and
future-proof solution for seamlessly interfacing the wireless and photonic segments of communication
networks. Microwave photonics with its inherent capability of interfacing these highly dissimilar media
is the best positioned technology to provide a flexible, adaptive, and future-proof physical layer with
unmatched characteristics by enabling the realization of key functionalities in microwave systems, which
are either complex or even not directly possible within the RF domain.
As shown in Figure 65, microwave photonics systems are usually envisioned at the heart of analog
signal processing engines, placed in between the signal acquisition devices and front-end digital signal
processors (DSPs) to accommodate the signal formats with the constraints imposed by the DSP-limited
sampling rates.
By realizing the wide variety of functionalities that we have previously described, MWP systems are
expected to span dramatically their areas of application embracing civil applications with the expected
massive takeover. Some examples shown in Figure 66 include the following.
(i) 5/6G communications where MWP filtering, Radio over fiber communications and beamforming
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:58

Figure 66 (Color online) Different application scenarios for microwave photonic systems in emerging fields (courtesy of iPronics).

techniques will be extensively exploited.

(ii) Internet of Things (IoT) and Industry 4.0 where MWP signal generation, transmission and inter-
facing capabilities will enable the integration of man-to-man, man-to-machine and machine-to-machine
communications and sensor data collection in mixed fiber-wireless scenarios.
(iii) Car-to-car communications and autonomous driving which will require advanced and flexible radar
and lidar techniques based on advanced high frequency radio signal generation and beamforming,
(iv) Avionics and drone engineering which will require the integration of advanced, compact, and
flexible RF signal generators, antenna beamformers, and broadband analog communications.
(v) High resolution sensing, molecular and bio spectroscopy, and imaging which will require the in-
tegration of advanced tunable optoelectronic oscillators, optical generation of THz waves with optical
spectrum analyzers.
These emerging paradigms are currently on the increasing slope of their S development curve and are
expected to push this pressure even further. The requirements demanded by most of these scenarios call
for novel technology developments in both the physical layer and the network architectures. In particular,
an evolution to higher frequencies is expected as 5/6G communications will target K (18–26.5 GHz), Ka
(26.5–40 GHz), Q (30–50 GHz), and E (60–90 GHz) bands, radar at the W band (75–110 GHz) and new
standards for WLAN at the V band (50–75 GHz). This will require a smaller footprint in the RF frontends
and increased processing bandwidths possibly carried by application-specific microwave photonics chips
in the first instance and programmable microwave photonics chips in a later stage as the multifunction
operation will be required.
Another important application of microwave photonics is wideband and high-resolution radars. Thanks
to the wide bandwidth offered by photonics, wideband microwave waveforms can be generated based on
photonics, which would help increase the resolution of a radar system. The first demonstration of a
microwave photonic radar was in 2014 [235] in which a mode locked laser was employed to generate a
high frequency microwave signal for transmission, and at the same time, mode locked laser was used as a
high repetition rate pulse train for high-speed sampling for receiver signal processing. Since then, a great
number of studies have been published to report different architectures of radar systems implemented
based on microwave photonics [236].

8 Discussion and conclusion

A comprehensive review of microwave photonics techniques with their implementation based on discrete
components and photonic integrated circuits has been performed. Despite the great potential offered
by microwave photonic subsystems and systems based on discrete components with unique features not
offered by state-of-the-art electronics, their applications are still limited. Recent activities in high-speed
modulators [237,238] and photodetectors [239,240] have made discrete component-based microwave pho-
tonic systems operate at much higher speed and wider bandwidth. However, the high manufacturing
costs have made them mainly for some specific applications where the costs are not a major issue. The
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:59

activities in integrated photonics to integrate both active and passive photonic devices in a single photonic
integrated chip have helped to realize integrated microwave photonics systems at significantly reduced
cost with greatly improved performance. In this article, a detailed review of the recent progress in inte-
grated microwave photonics was provided including the materials systems that are suitable for integrated
microwave photonics, and two types of integrated microwave photonics systems, application-specific pho-
tonic integrated circuits, and general-purpose programmable microwave photonic signal processors, were
provided.
We would like to emphasize that among the different materials systems, SOI has been considered a
material system with the highest potential for integrated microwave photonics thanks to the key ad-
vantages including small footprint, low loss, and low fabrication cost by using the well-developed and
mature CMOS fabrication technology. Both active and passive photonic components can be realized in
Si-based chips except for light sources and optical amplifiers. For microwave photonics applications, light
sources are always needed. In most of the demonstrations reported in the literature, an external light
source was employed, which makes the systems complicated and costly. In addition, due to fiber-to-chip
and chip-to-fiber coupling being needed, the system loss is high. To solve these problems, heterogeneous
integration should be implemented by jointly using Si and InP, with light sources and optical amplifiers
implemented based on InP/Si and all other photonic components based on Si, to make monolithically
integrated microwave photonics subsystems and systems to have a compact size, low loss, and better
stability.
References
1 Yao J. Microwave photonics. J Lightw Technol, 2009, 27: 314–335
2 Capmany J, Novak D. Microwave photonics combines two worlds. Nat Photon, 2007, 1: 319–330
3 Seeds A J, Williams K J. Microwave photonics. J Lightwave Technol, 2006, 24: 4628–4641
4 Berceli T, Herczfeld P R. Microwave photonics—a historical perspective. IEEE Trans Microwave Theor Techn, 2010, 58:
2992–3000
5 Cox III C H. Analog Photonic Links: Theory and Practice. Cambridge: Cambridge University Press, 2004
6 Chang W S C. RF Photonic Technology in Optical Fiber Links. Cambridge: Cambridge University Press, 2007
7 Iezekiel S. Microwave Photonics. Hoboken: John Wiley & Sons, 2009
8 Vilcot A, Cabon B, Chazelas J. Microwave-Photonics: From Components to Applications and Systems. Berlin: Springer,
2003
9 Rumerhald C, Algani C, Billabert A L. Microwaves Photonic Links: Components and Circuits. Hoboken: John Wiley &
Sons, 2013
10 Cox C H, Ackerman E I, Betts G E, et al. Limits on the performance of RF-over-fiber links and their impact on device
design. IEEE Trans Microwave Theor Techn, 2006, 54: 906–920
11 Roussell H V, Regan M D, Prince J L, et al. Gain, noise figure and bandwidth-limited dynamic range of a low-biased external
modulation link. In: Proceedings of Interntional Topical Meeting on Microwave Photonics, Victoria, 2007. 84–87
12 Gasulla I, Capmany J. Analytical model and figures of merit for filtered microwave photonic links. Opt Express, 2011, 19:
19758–19774
13 Goldberg L, Taylor H F, Weller J F, et al. Microwave signal generation with injection-locked laser diodes. Electron Lett,
1983, 19: 491–493
14 Goldberg L, Yurek A, Taylor H F, et al. 35 GHz microwave signal generation with injection locked laser diode. Electron
Lett, 1985, 21: 714–715
15 Bouyer P, Gustavson T L, Haritos K G, et al. Microwave signal generation with optical injection locking. Opt Lett, 1996,
21: 1502–1504
16 Cui W, Shao T, Yao J. Wavelength reuse in a UWB over WDM-PON based on injection locking of a Fabry-Pérot laser diode
and polarization multiplexing. J Lightwave Technol, 2014, 32: 220–227
17 Pan S, Tang Z, Zhu D, et al. Injection-locked fiber laser for tunable millimeter-wave generation. Opt Lett, 2011, 36:
4722–4724
18 Pan S, Yao J. Wideband and frequency-tunable microwave generation using an optoelectronic oscillator incorporating a
Fabry-Perot laser diode with external optical injection. Opt Lett, 2010, 35: 1911–1913
19 Harrison J, Mooradian A. Linewidth and offset frequency locking of external cavity GaAlAs lasers. IEEE J Quantum
Electron, 1989, 25: 1252–1255
20 Ramos R T, Seeds A J. Fast heterodyne optical phase-lock loop using double quantum well laser diodes. Electron Lett, 1992,
28: 82–83
21 Gliese U, Nielsen T N, Bruun M, et al. A wideband heterodyne optical phase-locked loop for generation of 3–18 GHz
microwave carriers. IEEE Photon Technol Lett, 1992, 4: 936–938
22 Bordonalli A C, Walton C, Seeds A J. High-performance phase locking of wide linewidth semiconductor lasers by combined
use of optical injection locking and optical phase-lock loop. J Lightwave Technol, 1999, 17: 328–342
23 Fan F Z, Dagenais M. Optical generation of a megahertz-linewidth microwave signal using semiconductor lasers and a
discriminator-aided phase-locked loop. IEEE Trans Microwave Theor Techn, 1997, 45: 1296–1300
24 Rideout H R, Seregelyi J S, Paquet S, et al. Discriminator-aided optical phase-lock loop incorporating a frequency down-
conversion module. IEEE Photon Technol Lett, 2006, 18: 2344–2346
25 Rideout H R, Seregelyi J S, Yao J. A true time delay beamforming system incorporating a wavelength tunable optical
phase-lock loop. J Lightwave Technol, 2007, 25: 1761–1770
26 Balakier K, Ponnampalam L, Fice M J, et al. Integrated semiconductor laser optical phase lock loops. IEEE J Sel Top
Quantum Electron, 2018, 24: 1500112
27 Balakier K, Shams H, Fice M J, et al. Optical phase lock loop as high-quality tuneable filter for optical frequency comb line
selection. J Lightwave Technol, 2018, 36: 4646–4654
28 O’Reilly J J, Lane P M, Heidemann R, et al. Optical generation of very narrow linewidth millimetre wave signals. Electron
Lett, 1992, 28: 2309–2311
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:60

29 O’Reilly J J, Lane P M. Remote delivery of video services using mm-waves and optics. J Lightwave Technol, 1994, 12:
369–375
30 O’Reilly J J, Lane P M. Fibre-supported optical generation and delivery of 60 GHz signals. Electron Lett, 1994, 30: 1329–
1330
31 Qi G H, Yao J P, Seregelyi J, et al. Generation and distribution of a wide-band continuously tunable millimeter-wave signal
with an optical external modulation technique. IEEE Trans Microwave Theor Techn, 2005, 53: 3090–3097
32 Qi G H, Yao J P, Seregelyi J, et al. Optical generation and distribution of continuously tunable millimeter-wave signals
using an optical phase modulator. J Lightwave Technol, 2005, 23: 2687–2695
33 Qasymeh M, Li W, Yao J. Frequency-tunable microwave generation based on time-delayed optical combs. IEEE Trans
Microwave Theor Techn, 2011, 59: 2987–2993
34 Li W, Yao J. Investigation of photonically assisted microwave frequency multiplication based on external modulation. IEEE
Trans Microwave Theor Techn, 2010, 58: 3259–3268
35 Pan S L, Yao J P. Tunable subterahertz wave generation based on photonic frequency sextupling using a polarization
modulator and a wavelength-fixed notch filter. IEEE Trans Microwave Theor Techn, 2010, 58: 1967–1975
36 Liu W L, Wang M G, Yao J P. Tunable microwave and sub-terahertz generation based on frequency quadrupling using a
single polarization modulator. J Lightwave Technol, 2013, 31: 1636–1644
37 Qi G, Yao J, Seregelyi J, et al. Phase-noise analysis of optically generated millimeter-wave signals with external optical
modulation techniques. J Lightwave Technol, 2006, 24: 4861–4875
38 Robins W P. Phase Noise in Signal Sources: Theory and Applications. London: The Institution of Engineering and Tech-
nology, 1984. 77
39 Fortier T M, Kirchner M S, Quinlan F, et al. Generation of ultrastable microwaves via optical frequency division. Nat
Photon, 2011, 5: 425–429
40 Yao X S, Maleki L. Optoelectronic microwave oscillator. J Opt Soc Am B, 1996, 13: 1725–1735
41 Yao X S, Maleki L. Optoelectronic oscillator for photonic systems. IEEE J Quantum Electron, 1996, 32: 1141–1149
42 Yao X S, Maleki L. Multiloop optoelectronic oscillator. IEEE J Quantum Electron, 2000, 36: 79–84
43 Ozdur I, Mandridis D, Hoghooghi N, et al. Low noise optically tunable optoelectronic oscillator with Fabry-Perot etalon. J
Lightw Technol, 2010, 28: 3100–3106
44 Pan S, Yao J. Wideband and frequency-tunable microwave generation using an optoelectronic oscillator incorporating a
Fabry-Perot laser diode with external optical injection. Opt Lett, 2010, 35: 1911–1913
45 Li W, Yao J. An optically tunable optoelectronic oscillator. J Lightwave Technol, 2010, 28: 2640–2645
46 Li W, Yao J. A wideband frequency tunable optoelectronic oscillator incorporating a tunable microwave photonic filter based
on phase-modulation to intensity-modulation conversion using a phase-shifted fiber Bragg grating. IEEE Trans Microwave
Theor Techn, 2012, 60: 1735–1742
47 Maleki L. The optoelectronic oscillator. Nature Photon, 2011, 5: 728–730
48 Zhang J, Yao J. Parity-time-symmetric optoelectronic oscillator. Sci Adv, 2018, 4: eaar6782
49 Liu Y, Hao T, Li W, et al. Observation of parity-time symmetry in microwave photonics. Light Sci Appl, 2018, 7: 38
50 Bender C M, Boettcher S. Real spectra in non-Hermitian Hamiltonians having PT symmetry. Phys Rev Lett, 1998, 80:
5243–5246
51 Bender C M. Making sense of non-Hermitian Hamiltonians. Rep Prog Phys, 2007, 70: 947–1018
52 Feng L, El-Ganainy R, Ge L. Non-Hermitian photonics based on parity-time symmetry. Nat Photon, 2017, 11: 752–762
53 El-Ganainy R, Makris K G, Khajavikhan M, et al. Non-Hermitian physics and PT symmetry. Nat Phys, 2018, 14: 11–19
54 Qi B, Chen H Z, Ge L, et al. Parity-time symmetry synthetic lasers: physics and devices. Adv Opt Mater, 2019, 7: 1900694
55 Hodaei H, Miri M A, Heinrich M, et al. Parity-time-symmetric microring lasers. Science, 2014, 346: 975–978
56 Feng L, Wong Z J, Ma R M, et al. Single-mode laser by parity-time symmetry breaking. Science, 2014, 346: 972–975
57 Hodaei H, Miri M A, Hassan A U, et al. Parity-time-symmetric coupled microring lasers operating around an exceptional
point. Opt Lett, 2015, 40: 4955–4958
58 Hodaei H, Miri M A, Hassan A U, et al. Single mode lasing in transversely multi-moded PT-symmetric microring resonators.
Laser Photon Rev, 2016, 10: 494–499
59 Liu W, Li M, Guzzon R S, et al. A photonic integrated parity-time symmetry wavelength-tunable single-mode microring
laser. Nat Commun, 2017, 10: 15389
60 Fan Z, Zhang W, Qiu Q, et al. Hybrid frequency-tunable parity-time symmetric optoelectronic oscillator. J Lightwave
Technol, 2020, 38: 2127–2133
61 Li P, Dai Z, Fan Z, et al. Parity-time-symmetric frequency-tunable optoelectronic oscillator with a single dual-polarization
optical loop. Opt Lett, 2020, 45: 3139–3142
62 Dai Z, Fan Z Q, Li P, et al. Frequency-tunable parity-time-symmetric optoelectronic oscillator using a polarization-dependent
Sagnac loop. J Lightw Technol, 2020, 38: 5327–5332
63 Fan Z Q, Dai Z, Qiu Q, et al. Parity-time symmetry in a single-loop photonic system. J Lightwave Technol, 2020, 38:
3866–3873
64 Chi H, Wang C, Yao J. Photonic generation of wideband chirped microwave waveforms. IEEE J Microw, 2021, 1: 787–803
65 Hao T, Cen Q, Dai Y, et al. Breaking the limitation of mode building time in an optoelectronic oscillator. Nat Commun,
2018, 9: 1839
66 Tang J, Zhu B, Zhang W, et al. Hybrid Fourier-domain mode-locked laser for ultra-wideband linearly chirped microwave
waveform generation. Nat Commun, 2020, 11: 3814
67 Capmany J, Ortega B, Pastor D. A tutorial on microwave photonic filters. J Lightwave Technol, 2006, 24: 201–229
68 Minasian R A. Photonic signal processing of microwave signals. IEEE Trans Microwave Theor Techn, 2006, 54: 832–846
69 Yao J. Photonics to the rescue: a fresh look at microwave photonic filters. IEEE Microwave, 2015, 16: 46–60
70 Capmany J, Pastor D, Martinez A, et al. Microwave photonic filters with negative coefficients based on phase inversion in
an electro-optic modulator. Opt Lett, 2003, 28: 1415–1417
71 Wang Q, Yao J. Multitap photonic microwave filters with arbitrary positive and negative coefficients using a polarization
modulator and an optical polarizer. IEEE Photon Technol Lett, 2008, 20: 78–80
72 Yan Y, Blais S R, Yao J P. Tunable photonic microwave bandpass filter with negative coefficients implemented using an
optical phase modulator and chirped fiber Bragg gratings. J Lightwave Technol, 2007, 25: 3283–3288
73 Yan Y, Yao J. A tunable photonic microwave filter with a complex coefficient using an optical RF phase shifter. IEEE
Photon Technol Lett, 2007, 19: 1472–1474
74 Shahoei H, Yao J. A continuously tunable multi-tap complex-coefficient microwave photonic filter based on a tilted fiber
Bragg grating. Opt Express, 2013, 21: 7521–7527
75 Dai Y, Yao J. Nonuniformly spaced photonic microwave delay-line filters and applications. IEEE Trans Microwave Theor
Techn, 2010, 58: 3279–3289
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:61

76 Yi X, Minasian R A. Microwave photonic filter with single bandpass response. Electron Lett, 2009, 45: 362–363
77 Palaci J, Villanueva G E, Galan J V, et al. Single bandpass photonic microwave filter based on a notch ring resonator. IEEE
Photon Technol Lett, 2010, 22: 1276–1278
78 Marpaung D, Morrison B, Pant R, et al. Frequency agile microwave photonic notch filter with anomalously high stopband
rejection. Opt Lett, 2013, 38: 4300–4303
79 Li W, Li M, Yao J. A narrow-passband and frequency-tunable microwave photonic filter based on phase-modulation to
intensity-modulation conversion using a phase-shifted fiber Bragg grating. IEEE Trans Microwave Theor Techn, 2012, 60:
1287–1296
80 Erdogan T. Fiber grating spectra. J Lightwave Technol, 1997, 15: 1277–1294
81 Gao L, Chen X, Yao J. Tunable microwave photonic filter with a narrow and flat-top passband. IEEE Microw Wireless
Compon Lett, 2013, 23: 362–364
82 Dai Y, Chen X, Jiang D, et al. Equivalent phase shift in a fiber Bragg grating achieved by changing the sampling period.
IEEE Photon Technol Lett, 2004, 16: 2284–2286
83 Gao L, Zhang J, Chen X, et al. Microwave photonic filter with two independently tunable passbands using a phase modulator
and an equivalent phase-shifted fiber Bragg grating. IEEE Trans Microwave Theor Techn, 2014, 62: 380–387
84 Yan Y, Yao J. Photonic microwave bandpass filter with improved dynamic range. Opt Lett, 2008, 33: 1756–1758
85 Li W, Yao J. A narrow-passband frequency-tunable microwave photonic filter with an improved dynamic range. In: Pro-
ceedings of Optical Fiber Communication Conference and Exposition and the National Fiber Optic Engineers Conference,
Anaheim, 2013. 1–3
86 Chen X, Li W, Yao J. Microwave photonic link with improved dynamic range using a polarization modulator. IEEE Photon
Technol Lett, 2013, 25: 1373–1376
87 Han X, Xu E, Yao J. Tunable single bandpass microwave photonic filter with an improved dynamic range. IEEE Photon
Technol Lett, 2016, 28: 11–14
88 Yao J. Photonic generation of microwave arbitrary waveforms. Optics Commun, 2011, 284: 3723–3736
89 Leaird D E, Weiner A M. Femtosecond opticalpacket generation by a direct space-to-time pulse shaper. Opt Lett, 1999, 24:
853–855
90 Leaird D E, Weiner A M. Femtosecond direct space-to-time pulse shaping. IEEE J Quantum Electron, 2001, 37: 494–504
91 Leaird D E, Weiner A M. Chirp control in the direct space-to-time pulse shaper. Opt Lett, 2000, 25: 850–852
92 McKinney J D, Leaird D E, Weiner A M. Millimeter-wave arbitrary waveform generation with a direct space-to-time pulse
shaper. Opt Lett, 2002, 27: 1345–1347
93 McKinney J D, Seo D S, Weiner A M. Photonically assisted generation of continuous arbitrary millimetre electromagnetic
waveforms. Electron Lett, 2003, 39: 309–310
94 McKinney J D, Seo J D, Leaird D E, et al. Photonically assisted generation of arbitrary millimeter-wave and microwave
electromagnetic waveforms via direct space-to-time optical pulse shaping. J Lightwave Technol, 2003, 21: 3020–3028
95 Xiao S, McKinney J D, Weiner A M. Photonic microwave arbitrary waveform generation using a virtually imaged phased-
array (VIPA) direct space-to-time pulse shaper. IEEE Photon Technol Lett, 2004, 16: 1936–1938
96 Leaird D E, Weiner A M. Femtosecond direct space-to-time pulse shaping in an integrated-optic configuration. Opt Lett,
2004, 29: 1551–1553
97 Wang C, Yao J P. Chirped microwave pulse generation based on optical spectral shaping and wavelength-to-time mapping
using a Sagnac loop mirror incorporating a chirped fiber Bragg grating. J Lightwave Technol, 2009, 27: 3336–3341
98 Wang C, Yao J P. Simultaneous optical spectral shaping and wavelength-to-time mapping for photonic microwave arbitrary
waveform generation. IEEE Photon Technol Lett, 2009, 21: 793–795
99 Yao J. Microwave photonics: arbitrary waveform generation. Nat Photon, 2010, 4: 79–80
100 Wang C, Yao J P. Large time-bandwidth product microwave arbitrary waveform generation using a spatially discrete chirped
fiber Bragg grating. J Lightwave Technol, 2010, 28: 1652–1660
101 Chi H, Yao J P. Chirped RF pulse generation based on optical spectral shaping and wavelength-to-time mapping using a
nonlinearly chirped fiber Bragg grating. J Lightwave Technol, 2008, 26: 1282–1287
102 Wang C, Yao J. Photonic generation of chirped millimeter-wave pulses based on nonlinear frequency-to-time mapping in a
nonlinearly chirped fiber Bragg grating. IEEE Trans Microwave Theor Techn, 2008, 56: 542–553
103 Azaña J, Berger N K, Levit B, et al. Reconfigurable generation of high-repetition-rate optical pulse sequences based on
time-domain phase-only filtering. Opt Lett, 2005, 30: 3228–3230
104 Chi H, Yao J. Symmetrical waveform generation based on temporal pulse shaping using amplitude-only modulator. Electron
Lett, 2007, 43: 415–417
105 Wang C, Li M, Yao J. Continuously tunable photonic microwave frequency multiplication by use of an unbalanced temporal
pulse shaping system. IEEE Photon Technol Lett, 2010, 22: 1285–1287
106 Li M, Wang C, Li W, et al. An unbalanced temporal pulse-shaping system for chirped microwave waveform generation.
IEEE Trans Microwave Theor Techn, 2010, 58: 2968–2975
107 Dolfi D, Michel-Gabriel F, Bann S, et al. Two-dimensional optical architecture for time-delay beam forming in a phased-array
antenna. Opt Lett, 1991, 16: 255–257
108 Dolfi D, Joffre P, Antoine J, et al. Experimental demonstration of a phased-array antenna optically controlled with phase
and time delays. Appl Opt, 1996, 35: 5293–5300
109 Ng W, Walston A A, Tangonan G L, et al. The first demonstration of an optically steered microwave phased array antenna
using true-time-delay. J Lightwave Technol, 1991, 9: 1124–1131
110 Esman R D, Monsma M J, Dexter J L, et al. Microwave true time-delay modulator using fibre-optic dispersion. Electron
Lett, 1992, 28: 1905
111 Esman R D, Frankel M Y, Dexter J L, et al. Fiber-optic prism true time-delay antenna feed. IEEE Photon Technol Lett,
1993, 5: 1347–1349
112 Molony A, Lin Z, Williams J A R, et al. Fiber Bragg-grating true time-delay systems: discrete-grating array 3-b delay lines
and chirped-grating 6-b delay lines. IEEE Trans Microwave Theor Techn, 1997, 45: 1527–1530
113 Corral J L, Marti J, Fuster J M, et al. True time-delay scheme for feeding optically controlled phased-array antennas using
chirped-fiber gratings. IEEE Photon Technol Lett, 1997, 9: 1529–1531
114 Zmuda H, Soref R A, Payson P, et al. Photonic beamformer for phased array antennas using a fiber grating prism. IEEE
Photon Technol Lett, 1997, 9: 241–243
115 Minasian R A, Alameh K E. Optical-fiber grating-based beamforming network for microwave phased arrays. IEEE Trans
Microwave Theor Techn, 1997, 45: 1513–1518
116 Matthews P J, Frankel M Y, Esman R D. A wide-band fiber-optic true-time-steered array receiver capable of multiple
independent simultaneous beams. IEEE Photon Technol Lett, 1998, 10: 722–724
117 Tong D T K, Wu M C. Multiwavelength optically controlled phased-array antennas. IEEE Trans Microwave Theor Techn,
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:62

1998, 46: 108–115

118 Tong D T K, Wu M C. Common transmit/receive module for multiwavelength optically controlled phased array antennas.
In: Proceedings of Optical Fiber Communication Conference and Exhibit, 1998. 354–355
119 Yao J P, Yang J L, Liu Y Q. Continuous true-time-delay beamforming employing a multiwavelength tunable fiber laser
source. IEEE Photon Technol Lett, 2002, 14: 687–689
120 Liu Y, Yao J, Yang J. Wideband true-time-delay unit for phased array beamforming using discrete-chirped fiber grating
prism. Optics Commun, 2002, 207: 177–187
121 Vidal B, Madrid D, Corral J L, et al. Photonic true-time delay beamformer for broadband wireless access network at 40 GHz
band. In: Proceedings of IEEE MTT-S International Microwave Symposium Digest, 2002. 3: 1949–1952
122 Liu Y Q, Yang J L, Yao J P. Continuous true-time-delay beamforming for phased array antenna using a tunable chirped
fiber grating delay line. IEEE Photon Technol Lett, 2002, 14: 1172–1174
123 Chen Y H, Chen R T. A fully packaged true time delay module for a K-band phased array antenna system demonstration.
IEEE Photon Technol Lett, 2002, 14: 1175–1177
124 Liu Y, Yao J, Yang J. Wideband true-time-delay beam former that employs a tunable chirped fiber grating prism. Appl
Opt, 2003, 42: 2273–2277
125 Jung B M, Shin J D, Kim B G. Optical true time-delay for two-dimensional X-band phased array antennas. IEEE Photon
Technol Lett, 2007, 19: 877–879
126 Blumenthal D J, Barton J, Beheshti N, et al. Integrated photonics for low-power packet networking. IEEE J Sel Top
Quantum Electron, 2011, 17: 458–471
127 Marpaung D, Roeloffzen C, Heideman R, et al. Integrated microwave photonics. Laser Photon Rev, 2013, 7: 506–538
128 Iezekiel S, Burla M, Klamkin J, et al. RF engineering meets optoelectronics: progress in integrated microwave photonics.
IEEE Microwave, 2015, 16: 28–45
129 Marpaung D, Yao J, Capmany J. Integrated microwave photonics. Nat Photon, 2019, 13: 80–90
130 Smit M, Leijtens X, Ambrosius H, et al. An introduction to InP-based generic integration technology. Semicond Sci Technol,
2014, 29: 083001
131 Coldren L A, Nicholes S C, Johansson L, et al. High performance InP-based photonic ICs—a tutorial. J Lightwave Technol,
2011, 29: 554–570
132 Kish F, Nagarajan R, Welch D, et al. From visible light-emitting diodes to large-scale III-V photonic integrated circuits.
Proc IEEE, 2013, 101: 2255–2270
133 Hochberg M, Baehr-Jones T. Towards fabless silicon photonics. Nat Photon, 2010, 4: 492–494
134 Bogaerts W, Fiers M, Dumon P. Design challenges in silicon photonics. IEEE J Sel Top Quantum Electron, 2014, 20: 1–8
135 Soref R. The past, present, and future of silicon photonics. IEEE J Sel Top Quantum Electron, 2006, 12: 1678–1687
136 Chrostowski L, Hochberg M. Silicon Photonics Design: From Devices to Systems. Cambridge: Cambridge University Press,
2015
137 Heck M J R, Bauters J F, Davenport M L, et al. Hybrid silicon photonic integrated circuit technology. IEEE J Sel Top
Quantum Electron, 2013, 19: 6100117
138 Keyvaninia S, Muneeb M, Stanković S, et al. Ultra-thin DVS-BCB adhesive bonding of III-V wafers, dies and multiple dies
to a patterned silicon-on-insulator substrate. Opt Mater Express, 2013, 3: 35–46
139 Haq B, Kumari S, van Gasse K, et al. Micro-transfer-printed III-V-on-silicon C-band semiconductor optical amplifiers. Laser
Photon Rev, 2020, 14: 1900364
140 Heideman R G, Hoekman M, Schreuder E. TriPleXTM -based integrated optical ring resonators for lab-on-a-chip and envi-
ronmental detection. IEEE J Sel Top Quantum Electron, 2012, 18: 1583–1596
141 Roeloffzen C G H, Zhuang L, Taddei C, et al. Silicon nitride microwave photonic circuits. Opt Express, 2013, 21: 22937–
22961
142 Corbett B, Loi R, Zhou W, et al. Transfer print techniques for heterogeneous integration of photonic components. Prog
Quantum Electron, 2017, 52: 1–17
143 van der Tol J J G M, Jiao Y, Shen L, et al. Indium phosphide integrated photonics in membranes. IEEE J Sel Top Quantum
Electron, 2018, 24: 1–9
144 Qi Y, Li Y. Integrated lithium niobate photonics. Nanophotonics, 2020, 9: 1287–1320
145 Nikogosyan D N. Nonlinear Optical Crystals: A Complete Survey. 1st ed. New York: Springer, 2005
146 Poberaj G, Koechlin M, Sulser F, et al. Ion-sliced lithium niobate thin films for active photonic devices. Optical Mater,
2009, 31: 1054–1058
147 Rabiei P, Gunter P. Optical and electro-optical properties of submicrometer lithium niobate slab waveguides prepared by
crystal ion slicing and wafer bonding. Appl Phys Lett, 2004, 85: 4603–4605
148 Wu R, Wang M, Xu J, et al. Long low-loss-litium niobate on insulator waveguides with sub-nanometer surface roughness.
Nanomaterials, 2018, 8: 910
149 Zhang M, Wang C, Cheng R, et al. Monolithic ultra-high-Q lithium niobate microring resonator. Optica, 2017, 4: 1536–1537
150 Muñoz P, Capmany J, Pérez D, et al. Integrated microwave photonics: state of the art and future trends. In: Proceedings
of International Conference on Transparent Optical Networks (ICTON), Graz, 2014
151 Guzzon R S, Norberg E J, Parker J S, et al. Monolithically integrated programmable photonic microwave filter with tunable
inter-ring coupling. In: Proceedings of IEEE International Topical Meeting on Microwave Photonics, Montreal, 2010. 23–26
152 Norberg E J, Guzzon R S, Parker J S, et al. A monolithic programmable optical filter for RF-signal processing.
In: Proceedings of IEEE International Topical Meeting on Microwave Photonics, Montreal, 2010. 365–368
153 Norberg E J, Guzzon R S, Parker J S, et al. Programmable photonic filters from monolithically cascaded filter stages.
In: Proceedings of Integrated Photonics Research, Silicon and Nanophotonics, Monterey, 2010
154 Norberg E J, Guzzon R S, Parker J S, et al. Programmable photonic microwave filters monolithically integrated in InP-
InGaAsP. J Lightwave Technol, 2011, 29: 1611–1619
155 Guzzon R S, Norberg E J, Parker J S, et al. Integrated InP-InGaAsP tunable coupled ring optical bandpass filters with zero
insertion loss. Opt Express, 2011, 19: 7816–7826
156 Chen H-W, Fang A W, Bovington J, et al. Hybrid silicon tunable filter based on a Mach–Zehnder interferometer and ring
resonator. In: Proceedings of IEEE International Topical Meeting on Microwave Photonics, Valencia, 2009. 1–4
157 Chen H W, Fang A W, Peters J D, et al. Integrated microwave photonic filter on a hybrid silicon platform. IEEE Trans
Microwave Theor Techn, 2010, 58: 3213–3219
158 Madsen C K, Zhao J H. Optical Filter Design and Analysis: A Signal Processing Approach. Hoboken: John Wiley & Sons,
Inc. 1999
159 Dong P, Feng N N, Feng D, et al. GHz-bandwidth optical filters based on high-order silicon ring resonators. Opt Express,
2010, 18: 23784–23789
160 Feng N N, Dong P, Feng D, et al. Thermally-efficient reconfigurable narrowband RF-photonic filter. Opt Express, 2010, 18:
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:63

24648–24653
161 Romero-Garcı́a S, Moscoso-Mártir A, Müller J, et al. Wideband multi-stage CROW filters with relaxed fabrication tolerances.
Opt Express, 2018, 26: 4723–4737
162 Rasras M S, Tu K, Gill D M, et al. Demonstration of a tunable microwave-photonic notch filter using low-loss silicon ring
resonators. J Lightwave Technol, 2009, 27: 2105–2110
163 Rasras M S, Gill D M, Patel S S, et al. Demonstration of a fourth-order pole-zero optical filter integrated using CMOS
processes. J Lightwave Technol, 2007, 25: 87–92
164 Zhuang L. Flexible RF filter using a nonuniform SCISSOR. Opt Lett, 2016, 41: 1118
165 Zhuang L, Taddei C, Hoekman M, et al. Ring resonator-based on-chip modulation transformer for high-performance phase-
modulated microwave photonic links. Opt Express, 2013, 21: 25999–26013
166 Zhuang L, Khan M R, Beeker W, et al. Novel microwave photonic fractional Hilbert transformer using a ring resonator-based
optical all-pass filter. Opt Express, 2012, 20: 26499–26510
167 Taddei C, Yen N T H, Zhuang L M, et al. Waveguide filter-based on-chip differentiator for microwave photonic signal
processing. In: Proceedings of IEEE International Topical Meeting on Microwave Photonics (MWP), Alexandria, 2013.
28–31
168 Zhuang L M, Hoekman M, Oldenbeuving R M, et al. CRIT-alternative narrow-passband waveguide filter for microwave
photonic signal processors. IEEE Photon Technol Lett, 2014, 26: 1034–1037
169 Xie Y, Geng Z, Zhuang L, et al. Programmable optical processor chips: toward photonic RF filters with DSP-level flexibility
and MHz-band selectivity. Nanophotonics, 2017, 7: 421–454
170 Orlandi P, Morichetti F, Strain M J, et al. Photonic integrated filter with widely tunable bandwidth. J Lightwave Technol,
2014, 32: 897–907
171 Fandiño J S, Muñoz P, Doménech D, et al. A monolithic integrated photonic microwave filter. Nat Photon, 2017, 11:
124–129
172 Zhang W, Yao J. Silicon photonic integrated optoelectronic oscillator for frequency-tunable microwave generation. J Light-
wave Technol, 2018, 36: 4655–4663
173 Tang J, Hao T, Li W, et al. Integrated optoelectronic oscillator. Opt Express, 2018, 26: 12257–12265
174 Lin I S, McKinney J D, Weiner A M. Photonic synthesis of broadband microwave arbitrary waveforms applicable to ultra-
wideband communication. IEEE Microw Wireless Compon Lett, 2005, 15: 226–228
175 Khan M H, Shen H, Xuan Y, et al. Ultrabroad-bandwidth arbitrary radiofrequency waveform generation with a silicon
photonic chip-based spectral shaper. Nat Photon, 2010, 4: 117–122
176 Wang J, Shen H, Fan L, et al. Reconfigurable radio-frequency arbitrary waveforms synthesized in a silicon photonic chip.
Nat Commun, 2015, 6: 5957
177 Zhang W, Zhang J, Yao J. Largely chirped microwave waveform generation using a silicon-based on-chip optical spectral
shaper. In: Proceedings of IEEE Topical Meeting Microwave Photonics, Sapporo, 2014. 20–23
178 Zhang W, Yao J. Photonic generation of linearly chirped microwave waveform with a large time-bandwidth product using
a silicon-based on-chip spectral shaper. In: Proceedings of International Topical Meeting on Microwave Photonics (MWP),
Paphos, 2015. 26–29
179 Wu J, Xu X, Nguyen T G, et al. RF photonics: an optical microcombs’ perspective. IEEE J Sel Top Quantum Electron,
2018, 24: 1–20
180 Wang Z, van Gasse K, Moskalenko V, et al. A III-V-on-Si ultra-dense comb laser. Light Sci Appl, 2017, 6: e16260
181 Wu R, Supradeepa V R, Long C M, et al. Generation of very flat optical frequency combs from continuous-wave lasers using
cascaded intensity and phase modulators driven by tailored radio frequency waveforms. Opt Lett, 2010, 35: 3234
182 Chen Y, Wu K, Zhao F, et al. Loss compensated photonic true-time delay for phased-array antenna. IEEE Antenn Propag
Soc Int Symp, 2004, 4: 4324–4327
183 Wang X, Howley B, Chen M Y, et al. Fully- integrated 4-bit true time delay module using polymer optical switches and
waveguide delay lines. In: Proceedings of the Integrated Photonics Research Applications, 2006
184 Chen Y, Wu K, Zhao F, et al. Reconfigurable true-time delay for wideband phased-array antennas. In: Proceedings of SPIE,
2004. 5363: 125–130
185 Horikawa K, Ogawa I, Ogawa H, et al. Photonic switched true time delay beam forming network integrated on silica
waveguide circuits. In: Proceedings of 1995 IEEE MTT-S International Microwave Symposium, Orlando, 1995. 65–68
186 Horikawa K, Ogawa I, Kitoh T, et al. Silica-based integrated planar lightwave true-time-delay network for microwave antenna
applications. In: Proceedings of Optical Fiber Communications, 1996. 100–101
187 LeGrange J, Kasper A, Madsen C, et al. Demonstration of an integrated, tunable high resolution true time delay line.
In: Proceedings of the 17th Annual Meeting IEEE Lasers Electro-Optics Society, 2004. 2: 790–791
188 Rasras M S, Madsen C K, Cappuzzo M A, et al. Integrated resonance-enhanced variable optical delay lines. IEEE Photon
Technol Lett, 2005, 17: 834–836
189 Horikawa K, Nakasuga Y, Ogawa H. Self-heterodyning optical waveguide beam forming and steering network integrated on
Lithium Niobate substrate. IEEE Trans Microwave Theor Techn, 1995, 43: 2395–2401
190 Ng W W, Yap D, Narayanan A A, et al. GaAs and silica-based integrated optical time-shift network for phased arrays.
In: Proceedings of SPIE, 1994. 2155: 114–123
191 Ng W, Yap D, Narayanan A, et al. High-precision detector-switched monolithic GaAs time-delay network for the optical
control of phased arrays. IEEE Photon Technol Lett, 1994, 6: 231–234
192 Stulemeijer J, van Vliet F E, Benoist K W, et al. Compact photonic integrated phase and amplitude controller for phased-
array antennas. IEEE Photon Technol Lett, 1999, 11: 122–124
193 Flamand G, de Mesel K, Moerman I, et al. InP-based PIC for an optical phased-array antenna at 1.06 µm. IEEE Photon
Technol Lett, 2000, 12: 876–878
194 Moreira R L, Garcia J, Li W, et al. Integrated ultra-low-loss 4-bit tunable delay for broadband phased array antenna
applications. IEEE Photon Technol Lett, 2013, 25: 1165–1168
195 Sancho J, Bourderionnet J, Lloret J, et al. Integrable microwave filter based on a photonic crystal delay line. Nat Commun,
2012, 3: 1075
196 Zhang W, Yao J. A fully reconfigurable waveguide Bragg grating for programmable photonic signal processing. Nat Commun,
2018, 9: 1396
197 Zhuang L, Roeloffzen C G H, Heideman R G, et al. Single-chip ring resonator-based 1 × 8 optical beam forming network in
CMOS-compatible waveguide technology. IEEE Photon Technol Lett, 2007, 19: 1130–1132
198 Zhuang L, Hoekman M, Beeker W, et al. Novel low-loss waveguide delay lines using Vernier ring resonators for on-chip
multi-λ microwave photonic signal processors. Laser Photon Rev, 2013, 7: 994–1002
199 Cardenas J, Foster M A, Sherwood-Droz N, et al. Wide-bandwidth continuously tunable optical delay line using silicon
microring resonators. Opt Express, 2010, 18: 26525–26534
 Yao J P, et al. Sci China Inf Sci December 2022 Vol. 65 221401:64

200 Morton P A, Cardenas J, Khurgin J B, et al. Fast thermal switching of wideband optical delay line with no long-term
transient. IEEE Photon Technol Lett, 2012, 24: 512–514
201 Xiang C, Davenport M L, Khurgin J B, et al. Low-loss continuously tunable optical true time delay based on Si3 N4 ring
resonators. IEEE J Sel Top Quantum Electron, 2018, 24: 1–9
202 Wang J, Ashrafi R, Adams R, et al. Subwavelength grating enabled on-chip ultra-compact optical true time delay line. Sci
Rep, 2016, 6: 30235
203 Pu M, Liu L, Xue W, et al. Widely tunable microwave phase shifter based on silicon-on-insulator dual-microring resonator.
Opt Express, 2010, 18: 6172–6182
204 Lloret J, Sancho J, Pu M, et al. Tunable complex-valued multi-tap microwave photonic filter based on single silicon-on-
insulator microring resonator. Opt Express, 2011, 19: 12402–12407
205 Lloret J, Morthier G, Ramos F, et al. Broadband microwave photonic fully tunable filter using a single heterogeneously
integrated III-V/SOI-microdisk-based phase shifter. Opt Express, 2012, 20: 10796–10806
206 Adams D B, Madsen C K. A novel broadband photonic RF phase shifter. J Lightwave Technol, 2008, 26: 2712–2717
207 Morton P A, Khurgin J B. Microwave photonic delay line with separate tuning of the optical carrier. IEEE Photon Technol
Lett, 2009, 21: 1686–1688
208 Burla M, Marpaung D, Zhuang L, et al. On-chip CMOS compatible reconfigurable optical delay line with separate carrier
tuning for microwave photonic signal processing. Opt Express, 2011, 19: 21475–21484
209 Howley B, Wang X, Chen M, et al. Reconfigurable delay time polymer planar lightwave circuit for an X-band phased-array
antenna demonstration. J Lightwave Technol, 2007, 25: 883–890
210 Meijerink A, Roeloffzen C G H, Meijerink R, et al. Novel ring resonator-based integrated photonic beamformer for broadband
phased array receive antennas-part I: design and performance analysis. J Lightwave Technol, 2010, 28: 3–18
211 Zhuang L M, Roeloffzen C G H, Meijerink A, et al. Novel ring resonator-based integrated photonic beamformer for broadband
phased array receive antennas-part II: experimental prototype. J Lightwave Technol, 2010, 28: 19–31
212 Marpaung D, Zhuang L, Burla M, et al. Towards a broadband and squint-free Ku-band phased array antenna system for
airborne satellite communications. In: Proceedings of the 5th European Conference Antennas Propagation, Rome, 2011.
2623–2627
213 Marpaung D, Zhuang L, Burla M, et al. Photonic integration and components development for a Ku-band phased array
antenna system. In: Proceedings of International Topical Meeting on Microwave Photonics jointly held with the 2011
Asia-Pacific Microwave Photonics Conference, 2011. 458–461
214 Burla M, Roeloffzen C G H, Zhuang L, et al. System integration and radiation pattern measurements of a phased array
antenna employing an integrated photonic beamformer for radio astronomy applications. Appl Opt, 2012, 51: 789–802
215 Zhuang L, Marpaung D, Burla M, et al. Low-loss, high-index-contrast Si3 N4 /SiO2 optical waveguides for optical delay lines
in microwave photonics signal processing. Opt Express, 2011, 19: 23162–23170
216 Burla M, Marpaung D, Zhuang L, et al. Integrated photonic Ku -band beamformer chip with continuous amplitude and delay
control. IEEE Photon Technol Lett, 2013, 25: 1145–1148
217 Burla M, Marpaung D A I, Zhuang L, et al. Multiwavelength-integrated optical beamformer based on wavelength division
multiplexing for 2-D phased array antennas. J Lightwave Technol, 2014, 32: 3509–3520
218 Horikawa K, Ogawa I, Kitoh T, et al. Silica-based integrated planar lightwave true-time-delay network for microwave antenna
applications. In: Proceedings of Optical Fiber Communications, 1996. 100–101
219 Grosskopf G, Eggemann R, Ehlers H, et al. Maximum directivity beam-former at 60 GHz with optical feeder. IEEE Trans
Antennas Propagat, 2003, 51: 3040–3046
220 Liu W, Li M, Guzzon R S, et al. A fully reconfigurable photonic integrated signal processor. Nat Photon, 2016, 10: 190–195
221 Capmany J, Mora J, Muñoz P, et al. A microwave photonics transistor. In: Proceedings of IEEE Topical Meeting on
Microwave Photonics 2013, Alexandria, 2013
222 Pérez D, Gasulla I, Capmany J. Software-defined reconfigurable microwave photonics processor. Opt Express, 2015, 23:
14640–14654
223 Perez D, Gasulla I, Capmany J. Honeycomb lattice meshes for reconfigurable universal microwave photonics processors.
In: Proceedings of SPIE/COS Photonics Asia, 2016
224 Capmany J, Gasulla I, Pérez D. Microwave photonics: the programmable processor. Nat Photon, 2016, 10: 6–8
225 Zhuang L, Roeloffzen C G H, Hoekman M, et al. Programmable photonic signal processor chip for radiofrequency applica-
tions. Optica, 2015, 2: 854–859
226 Pérez D, Gasulla I, Capmany J, et al. Reconfigurable lattice mesh designs for programmable photonic processors. Opt
Express, 2016, 24: 12093
227 Pérez D, Gasulla I, Crudgington L, et al. Multipurpose silicon photonics signal processor core. Nat Commun, 2017, 8: 636
228 Perez D, Gasulla I, Capmany J. Toward programmable microwave photonics processors. J Lightwave Technol, 2018, 36:
519–532
229 Wang X, Zhou L, Li R, et al. Continuously tunable ultra-thin silicon waveguide optical delay line. Optica, 2017, 4: 507–515
230 Burla M, Khan M R H, Marpaung D A I, et al. Separate carrier tuning scheme for integrated optical delay lines in photonic
beamformers. In: Proceedings of International Topical Meeting on Microwave Photonics, Singapore, 2011
231 Pérez D, Gasulla I, Capmany J, et al. Figures of merit for self-beating filtered microwave photonic systems. Opt Express,
2016, 24: 10087–10102
232 Jinguji K, Kawachi M. Synthesis of coherent two-port lattice-form optical delay-line circuit. J Lightwave Technol, 1995, 13:
73–82
233 Shao H, Yu H, Jiang X, et al. Large bandwidth and high accuracy photonic-assisted instantaneous microwave frequency esti-
mation system based on an integrated silicon micro-resonator. In: Proceedings of 2014 IEEE 11th International Conference
on de Group IV Photonics (GFP), Paris, 2014
234 Ho K, Liaw S, Lin C. Efficient photonic mixer with frequency doubling. IEEE Photon Technol Lett, 1997, 9: 511–513
235 Ghelfi P, Laghezza F, Scotti F, et al. A fully photonics-based coherent radar system. Nature, 2014, 507: 341–345
236 Pan S, Zhang Y. Microwave photonic radars. J Lightwave Technol, 2020, 38: 5450–5484
237 Trajkovic M, Blache F, Debregeas H, et al. 55 GHz EAM bandwidth and beyond in InP active passive photonic integration
platform. In: Proceedings of Conference on Lasers and Electro-Optics, 2018
238 Burla M, Hoessbacher C, Heni W, et al. 500 GHz plasmonic Mach-Zehnder modulator enabling sub-THz microwave photonics.
APL Photonics, 2019, 4: 056106
239 Tzu T C, Sun K, Costanzo R, et al. Foundry-enabled high-power photodetectors for microwave photonics. IEEE J Sel Top
Quantum Electron, 2019, 25: 3800111
240 Lischke S, Peczek A, Morgan J S, et al. Ultra-fast germanium photodiode with 3-dB bandwidth of 265 GHz. Nat Photon,
2021, 15: 925–931