A preview of this full-text is provided by Optica Publishing Group.
Content available from Optics Express
This content is subject to copyright. Terms and conditions apply.
Research Article Vol. 29, No. 15 / 19 July 2021 / Optics Express 23671
Analysis of the effects of jitter, relative intensity
noise, and nonlinearity on a photonic
digital-to-analog converter based on optical
Nyquist pulse synthesis
CHRISTIAN KRES S,*MEYSAM BAHMANIAN, TOBIAS SCH WA BE,
AND J. CHRI STOPH SC HEY TT
Paderborn University, Department of Circuit & System Technology, Heinz Nixdorf Institute, Warburger Str.
100, 33098 Paderborn, Germany
*christian.kress@uni-paderborn.de
Abstract:
An analysis of an optical Nyquist pulse synthesizer using Mach-Zehnder modulators
is presented. The analysis allows to predict the upper limit of the effective number of bits of this
type of photonic digital-to-analog converter. The analytical solution has been verified by means
of electro-optic simulations. With this analysis the limiting factor for certain scenarios: relative
intensity noise, distortions by driving the Mach-Zehnder modulator, or the signal generator phase
noise can quickly be identified.
© 2021 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
In the worldwide digital communication networks the demand for data transmission capacity is
continuously increasing. Quite often it is the bandwidth and the resolution of the data converters,
i.e. the digital-to-analog converters (DAC) and the analog-to-digital converters which limit the
achievable link capacity. Broadband electronic DACs can be implemented using binary and
unary weighted current switches [1], time-interleaving techniques [2] [3], with state-of-the-art
performance of 4.1 effective number of bits (ENOB) at an analog bandwidth of 58.6 GHz [4],
and digital-bandwidth-interleaving, with state-of-the-art performance of less than 3 ENOB at an
analog bandwidth of 100 GHz [5]. Photonic DACs have the potential to outperform electronic
DACs, especially in terms of bandwidth [6]. Furthermore, photonic integration using silicon
photonics technology enables the integration of photonic and electronic circuits on a single chip
which allows to combine the advantages of electronic and photonic signal processing and greatly
reduces system complexity, size, and cost.
This has motivated research into different electronic-photonic DAC concepts. Several groups
have published DACs with segmented Mach-Zehnder modulators (MZM) with binary electronic
drivers [7] [8] [9]. However these concepts are limited by the electro-optical bandwidth
of integrated MZMs and the electronic drivers and, hence, don‘t provide significant speed
improvements compared to electronic DACs. Furthermore, segmented MZMs provide quite
limited resolution of the photonic DAC. Others have presented photonic DACs using incoherent
addition of multiple, digitally modulated MZMs from either one laser diode, using suitable
splitting and attenuation in each MZM branch [10] or using multiple laser diodes with specific
powers and a wavelength division multiplexer [11]. This method has advantages in jitter and
resistance in electro-magnetic interferences, but is similarly limited by the utilized MZM’s
electro-optical bandwidths. The best reported ENOB for this method was only 3.55 at a bitrate of
12.5 Gb/s [12]. Another method is optical pulse generation with time-interleaved MZMs, being
driven by electronic DACs allowing to effectively add the individual DAC bandwidths. One
approach uses Mode-locked lasers (MLLs) to generate sharp, low-jitter amplitude-modulated
pulses for signal synthesis [13][14]. This DAC architecture increases complexity, because precise
#427424 https://doi.org/10.1364/OE.427424
Journal © 2021 Received 22 Apr 2021; revised 22 Jun 2021; accepted 24 Jun 2021; published 12 Jul 2021
Research Article Vol. 29, No. 15 / 19 July 2021 / Optics Express 23672
optical filtering of the ultra-broadband MLL pulses is required to generate the DAC output
signal. Furthermore, to this date, low-jitter MLLs have not been integrated in silicon photonics
technology, which increases system complexity and cost significantly.
Another promising approach was recently introduced as Nyquist pulse synthesizer [15][16],
using optical time-interleaving and Nyquist pulse synthesis from continuous-wave laser diodes
(CW-LD). This method allows to achieve precise Nyquist pulse properties without filtering as
the frequency comb is generated from an electrical signal generator (SG) with a fixed frequency
applied to a MZM under certain bias conditions [17]. Another advantage is that the optical
output signal bandwidth is up to three times higher than the applied electrical frequency [18].
Furthermore, the MZM can be operated well beyond its electro-optical 3 dB-bandwidth for the
purpose of pulse generation as long as the optical output signal exhibits sufficient power [19].
Using ultra-broadband MZMs, e.g. in LNOI technology [20], will potentially allow for sampling
rates beyond 300 GSa/s and continuous optical bandwidths above 300 GHz without frequency
interleaving.
In this contribution, we analyze the proposed Nyquist pulse synthesizer DAC with regards
to relative intensity noise (RIN) of the utilized laser diode, phase noise of the electrical signal
generator, and the influence of unwanted side bands resulting from the MZM’s non-linear transfer
function. A formula predicting the effective number of bits (ENOB) for the proposed photonic
Nyquist pulse synthesizer DAC is derived for the first time.
2. System analysis
2.1. System and ideal periodic Nyquist pulse description
Nyquist pulses are widely used in communication and signal processing applications due to
their inherent properties in time and frequency domain. In the frequency domain, a sinc-shaped
Nyquist pulse is a rectangular function. If this rectangular function is multiplied with a frequency
comb, the sinc-shaped pulses are convoluted with a Dirac Delta sequence. Since sinc-shaped
pulses possess the property of zero intersymbol-interference (ISI), the different time shifted
copies of the sinc-shaped pulses do not interfere with each other and the result is a perfect
sinc-pulse sequence [21], [22]. These sinc-pulse sequences are called periodic Nyquist pulses in
the following. The electric field amplitude of an ideal optical Nyquist comb with N numbers of
lines can be mathematically written as
ENyquist(t,N)=E0
N−1
2
n=−N−1
2
ei2π(fc+n∆f)t+iφ
N(1)
=E0
sin(πN∆ft)
Nsin(π∆ft)ei2πfct+iφ(2)
where f
c
and
∆
fare the optical carrier and the frequency offset between the lines. The peak
amplitude is normalized to E
0
[21]. The bandwidth of the centered Nyquist pulse for a fixed
frequency offset scales with N (s. Figure 1(b)).
Periodic Nyquist pulses allow for a time-interleaving system as proposed in [15]. A continuous-
wave laser diode emits light to a MZM. A signal generator modulates the laser line and, under
certain bias and drive conditions, generates periodic Nyquist pulses (s. sec. 2.2). The periodic
Nyquist pulse is then splitted into N arms. The pulses of each arm needs to be delayed by kT
/
N
with k
∈ {
0, 1,
. . .
,N
−
1
}
progressively from top to bottom regarding Fig. 1(a), where T
=
1
/∆
f
is the repetition period coming from the electrical signal from the SG. A weighting in each arm
is performed by variable attenuators or electro-optic modulators. For this purpose electronic
DACs need to be used. Subsequently, the suitably delayed signals of each branch are superposed
using combiners. In Fig. 1(a) optical delays are used while in [15] electrical phase shifters
Research Article Vol. 29, No. 15 / 19 July 2021 / Optics Express 23673
Fig. 1.
(a) Optical Nyquist pulse synthesizer system: A continuous-wave laser diode
(CW-LD) launches light into a MZM. A signal generator (SG) applies a high-frequency
sinusoidal to the MZM, which generates a precise sinc-sequence under certain bias and
RF power conditions. The sinc-sequence is splitted into N branches and modulated with
electro-optical systems with N times less bandwidth demand. Suitable delays and Nyquist
properties allow for perfect, time-aligned combination of the branches. (b) Normalized
periodic Nyquist sinc-sequence in respect of T
=1
∆f
. The bandwidth of the (isolated)
Nyquist sinc-sequence scales with the number of comb lines N. (c) Exemplary visualization
of a Nyquist pulse synthesizer system with N
=
5 arms. The different color codes represent
the respective signal in each arm. The black curve depicts the superposition of those signals.
are proposed for the Nyquist pulse generation. Although the system principles are identical,
the advantage of the system in 1(a) is that Nyquist pulse generation has to be performed only
once, thus simplifying the overall system. Furthermore, from a standpoint of chip integration
controlling the biasing of multiple MZMs and electronic phase shifters adds complexity and is
more prone to cross-talk. The proposed Nyquist pulse synthesizer allows to add the effective
bandwidths of the electronic DACs, whose individual bandwidths only need to be
∆
f. Then, the
total bandwidth of the system amounts to N∆f.
2.2. Nyquist Comb Generation in Mach-Zehnder Modulators
A single Mach-Zehnder modulator can be used to generate a three-line Nyquist comb by
setting a suitable input voltage amplitude and bias condition of the modulator [22]. The output
characteristic for the electrical field vector in a push-pull driven MZM can be written as
Eout =E0cos π
Vπ
vi+ΦMZM(3)
where
E0=
E
0
e
i(ωct+φc)
is the normalized carrier,
ωc
the optical carrier frequency,
ϕc
the
(random) optical phase, V
π
the required voltage to achieve a
π
shift in the MZM, and
ΦMZM
the
Research Article Vol. 29, No. 15 / 19 July 2021 / Optics Express 23674
accumulation of all influences and adjustments regarding the MZM’s phase. When applying a
sinusoidal input to the MZM v
i=
V
S·sin(ωe
t
+ϕe)
, the real part of the MZM’s output can be
written as:
Eout =E0cos(ΦMZM)J0(α)
+2
∞
k=1
cos(ΦMZM)J2k(α)cos 2k(ωet+ϕe)
−2
∞
k=1
sin(ΦMZM)J2k−1(α)sin (2k−1)(ωet+ϕe)
(4)
with the normalized amplitude
α=π
Vπ
V
S
and J
n
the Bessel function of n-th order. In order to
generate a three-line comb one has to find suitable
Φ
and
α
so that the amplitudes at the carrier
ωc
and first order harmonics
ωc±ωe
are equal while the higher order harmonics need to be
respectively small, ideally zero. In order to make the amplitudes of carrier and first harmonics
equal, the bias phase Φneeds to be set to
ΦMZM =tan−1J0(α)
J1(α). (5)
This optical phase adjustment can be achieved in a LiNbO3 MZM by means of applying a
differential voltage to the electrodes. For silicon MZMs it relies on phase shifters based on the
plasma dispersion or temperature effect. There is a trade-off in finding the a suitable amplitude V
S
:
Too small V
S
results in a too small overall power of the periodic Nyquist pulse. But the higher V
S
,
the more power is transferred to the higher order harmonics, which deteriorate the Nyquist pulse
waveform, loosing Nyquist properties, which are needed for the time-interleaved superposition
of signals in the branches of the optical Nyquist pulse synthesizer (s. Figure 1(a) and (c)). As
metric for the quality of the MZM-generated 3 line-comb the signal-to-distortion-ratio (SDR)
can be used:
SDRNyquist =(cos(Φ)J0(α))2+2(sin(Φ)J1(α))2
∞
k=−∞
k≠0,1
(sin(Φ)J2k−1(α))2+
∞
k=−∞
k≠0
(cos(Φ)J2k(α))2
(6)
where the weights of the carrier and 1st order harmonics contribute to the signal power and the
amplitudes of higher order harmonics are taken into account for the distortions. Although the
periodic Nyquist pulses itself are not the signal of interest in a DAC configuration as described in
Fig. 1(a), the harmonics deteriorate the signal to be synthesized (s
1
,
. . .
,s
N
) through convolution
in the frequency domain, and thereby will affect the achievable ENOB. The ENOB depending
on SDR can be calculated as
ENOB(SDR) =10 log(SDR)−1.77
6.02
. Figure 2(a) shows that the ENOB
limitation by harmonics is severe, even when neglecting noise and other undesirable effects.
While additional reasons will be discussed in following paragraphs, the influence of the harmonics
can be concluded as a major degradation factor.
2.3. Synthesizer signal power
The optical source for periodic Nyquist pulses is a CW-LD having a laser power of P
CW
. As
indicated in Fig. 2(b), the Nyquist pulse intensity is scaled to P
CW
as peak value. However, in the
synthesizer system, where the periodic Nyquist pulse in split into N arms, the power will be evenly
distributed. This will reduce the maximum signal power in each branch to P
CW /
N. Figure 2(b)
visualizes that argument for a system with N
=
3 branches. In a real system, additional losses
have to considered like MZM insertion loss, which typically ranges from 3-8 dB, optical coupling
Research Article Vol. 29, No. 15 / 19 July 2021 / Optics Express 23675
Fig. 2.
(a) Signal-to-distortion ratio of 3-line Nyquist pulse generated in a MZM by a
sinusoidal input without filtering. (b) Visualization of the achievable synthesis signal power:
The power of a 3-line periodic Nyquist pulse (black, dotted) is evenly splitted into 3 arms,
whereas the pulses are suitably delayed to increase the sampling rate (red,blue). The signal
power for the synthesis is therby reduced by factor 3.
losses, if integrated MZMs are used, and excess losses from splitters and combiners. Although
the optical signals can be amplified, the signal-to-noise-and-distortion (SINAD) ratio will not
improve. It was shown in Fig. 2(a) that SDR increases for small
α
, but it is clear that maximum
achievable signal amplitude scales with the nominator of Eq. (6). Hence, when other signal
degradation effects are taken into account, which are discussed in the following, decreasing
α
doesn‘t necessarily improve the ENOB. Although the signal power of the synthesizer system is
distributed into N arms, the SINAD does not degrade because the powers of all deterioration
effects will be as well evenly split. Therefore, the SINAD is the same before splitting, in each
arm, and as well after combining as the power is contained.
2.4. Phase noise transfer in MZMs
Equation (4) indicates that the electrical phase
ϕe
of the input signal is simply passed on to the
Bessel-weighted output of the optical signal. Hence, the phase noise of the signal generator is
likewise passed on to the optical signal E
out
. This is true for the optical intensity and therefore
for the photo detector current as well:
I=RPout =R∥Eout ∥2
=
E2
0
21+cos(2ΦMZM)J0(2α))
+2
∞
k=1
cos(2ΦMZM)J2k(2α)cos 2k(ωet+ϕe)
−2
∞
k=1
sin(2ΦMZM)J2k−1(2α)sin (2k−1)(ωet+ϕe)
(7)
We have also proven this experimentally by comparing the phase noise of our signal generator
(Anritsu MG3694B) and the phase noise when applying the signal generator to the MZM,
followed by a detection process by a 70 GHz photodiode and electrical amplification (SHF S807
C), which was necessary for the phase noise analyzer‘s sensitivity (Anapico APPH20G) at 10
GHz input frequency (see Fig. 3). We have used a LiNbO3 and an integrated silicon MZM with
linear, segmented drivers on-chip to evaluate the additional deterioration effects of the silicon
Research Article Vol. 29, No. 15 / 19 July 2021 / Optics Express 23676
modulator. For comparison, we have performed this experiment with a second signal source
(Keysight AWG M8194A).
Fig. 3.
Phase noise measurement of a signal generator (SG) / arbitrary signal generator
(AWG): simple electrical (blue) and with transition to optical using continuous-wave laser,
Mach-Zehnder modulator (MZM) and photodiode (PD). The electrical signal was amplified
(AMP) and it’s phase noise measured via phase noise analyzer (PNA).
Figure 4(a) indicates that the phase noise through the optical system ideally follows the signal
generator‘s phase noise up to an offset frequency of 1 MHz, which is the dominant part when
calculating the jitter in the time domain. Above 1 MHz the optical systems (LiNbO3 and silicon
MZMs) reach a plateau which was identified as shot noise limit of the photodiode. We see a
little bit worse performance for the silicon MZM, as we had higher optical coupling losses in the
integrated chip. There is no plateau for the AWG measurement, as we have boosted the AWG
signal power by an electrical amplifier. It is worth noting that the laser linewidth, although it
contributes to the laser phase noise, can be neglected as it vanishes in a detection process. In
contrast to laser phase noise, the phase noise of the signal generator transforms into amplitude
noise by means of the MZM.
2.5. Sampling error due to jitter
The power of a random jitter process can be modeled as a mean-free Gaussian-distribution
with variance
σtj
. In a sampling process the error caused by jitter of the clock signal can be
estimated using a first-order Taylor series approximation, leading to a analytical expression of the
signal-to-noise ratio (SNR) of
SNRX,Jit =10 log ρXX (0)
−ρ′′
XX (0) · ρtjtj(0)(8)
where
ρ
denotes the auto-correlation function, Xand t
j
referring to the data signal and jitter
distribution function respectively [23]. Although this expression was derived for analog-digital
converters, the linear approximation holds true for digital-analog conversion as well. When we
assume Xto be a sinusoidal signal with frequency fX, Eq. (8) can be derived as:
SNRsin,Jit =10 log 1
(2πfX)2·σ2
tj
=20 log 1
2πfX·σtj.
(9)
Research Article Vol. 29, No. 15 / 19 July 2021 / Optics Express 23677
Fig. 4.
(a) Phase noise measurement comparing a signal generator (SG) / arbitrary signal
generator (AWG) and transfer in the optical system at 10 GHz input frequency. (b) ENOB
limitation depending on Nyquist pulse signal generator jitter.
The absolute jitter of a signal is related to the phase noise by integrating it’s spectrum between
certain frequency limits:
σtj =2f1f2SPS(f)df
2πf(10)
Since many systems can reconstruct a carrier up to kHz-range, we have integrated the measured
phase noise (s. Figure 4(a)) from 1 kHz to 100 MHz. The calculated jitter rms values for the
signal generator have been 32 fs, 41 fs and 60 fs for just the SG, the SG and a LiNbO3 MZM, and
the SG and a silicon MZM respectively. For the AWG we have measured 156 fs, 160 fs, and 162
fs for the respective cases. We can conclude that there is a slight increase of the jitter in respect
to the optical systems. When there is no detection limit (as for the measurement with the AWG),
the main contribution of the jitter is the signal source, which is the main conclusion here. Using
Eq. (8) and feeding it with the calculated jitter values leads to an estimation of the achievable
ENOB using periodic Nyquist pulses with N
=
3 lines, which means 3 times the bandwidth of
the regular signal generator output sinusoid (s. Figure 4(b)).
2.6. Relative intensity noise
Another limiting factor in the optical signal quality is relative-intensity-noise (RIN) of the optical
source (s. Figure 1(a)). RIN comprises random fluctuations in the intensity of a laser. RIN is
usually defined as a mean square power fluctuation:
δP2
o(t)=∞
0
SRIN(f)df (11)
with S
RIN
being it’s one-sided spectral density [24]. When applying an optical bandpass
filter centered at the carrier f
c
, we get approximate integration limits B
=
f
max −
f
min
, with
f
max,min =
f
c±
B
/
2, and calculate the RIN as
δP2
o(
t
)=
BS
RIN
. The bandwidth of the filter should
Research Article Vol. 29, No. 15 / 19 July 2021 / Optics Express 23678
obviously not be smaller that the bandwidth of the generated comb for the Nyquist pulses: N
∆
f
SG
,
when the filter is placed at the output of the optical DAC.
3. Results and discussion
The above discussed effects of MZM distortion, signal generator jitter and RIN on the DAC
precision can be merged into a formula to calculate the signal-to-noise-and-distortion (SINAD)
which is extended compared to the regular case by the jitter error:
SINAD(f)=Signal(f)
Jitter(f)+RIN +Distortion(f)(12)
For the signal power we have calculated
Signal(
f
)=
P
CW
H
(
f
)
G
(
V
S)
, where the maximum signal
power is normalized to P
CW
with as described in section 2.3,H
(
f
)
the bandpass filter power
transfer function of order 20 and G
(
V
S)
a correction function for the optical peak power depending
on the modulation amplitude V
S
in accordance to the nominator of Eq. (6). The jitter error
can be derived in accordance to the denominator of Eq. (8) for the periodic Nyquist pulse as
Jitter(
f
)=2
3
P
CW (
2
π
f
)2σ2
tj
for a three-line spectral comb. The distortions stemming from the
MZM non-linearity can be defined as the denominator in Eq. (6) besides that the weights of the
harmonics are additionally multiplied with the suitable values from the bandpass filter function
H
(
f
)
for the respective frequencies. The RIN is defined as
RIN =∞
0
S
RIN (
f
)
H
(
f
)
df , also shaped
by the optical bandpass filter. Lastly, we use the SINAD to calculate the ENOB for sinusoidal
signals:
ENOB(f)=10 log (SINAD(f))−1.77
6.02 . (13)
We have compared the outcomes from the analytical formula with system simulations using
Lumerical Interconnect. The analytical and simulation results are in good agreement (s. Figure 5).
To differentiate the effects, the graphs contain ENOB calculations of each single deterioration
effect, reducing denominator Eq. (12) to a respective deterioration. In the example graphs, we
have assumed filter bandwidths that should be easily available as standard lab equipment (0.3 &
1 nm). The jitter values of 53 fs and 160 fs were taken from the phase noise measurement of
the system with a signal generator and an arbitrary signal generator respectively (s. section 2.5).
Several statements can be concluded from the analysis and several cases: For poor lasers, the RIN
is a dominant factor limiting the ENOB. Commercial lasers, having a RIN of -145 dBc, without
any narrow filters are limiting the ENOB to around 3.5 bits (s. Figure 5(b)). In order to achieve
useful ENOB, the laser RIN should be as small as possible by either using very advanced lasers
or inserting a very narrow optical filter after the laser (s. Figure 5(a)). For comparison, RIN of
-165 dBc [25], even using a broadband 1 nm band pass filter allows for an ENOB of 7. In all
cases the modulation amplitude V
s
was maximized so that the deterioration of the distortion was
slightly less than for the RIN and the total ENOB was maximized. In order to keep the distortion
by the harmonics the least contributor, the signal amplitude should be scaled to less than 0.15 V
π
of the modulator. When V
s>
0.36 V
π
, without any additional counter measures, the distortion will
limit the ENOB to 4 for any frequency. The rising of the SDR caused by distortions is due to the
fact that the harmonics move outside the optical band pass cut-off frequency when increasing the
fundamental frequency. Hence, the power in the higher order harmonics decreases and the SDR
increases. The jitter error becomes dominant around 10 GHz for the AWG and around 20 GHz for
the SG for this setup. Hence, when focusing on high bandwidth signal synthesizers, improving
the phase noise of the electrical source is paramount. In Fig. 5(e) a very low phase noise signal
generator (LPN-SG) with 4 fs rms jitter has been used [26]. In that scenario, the electrical signal
generator is no longer the limiting factor, allowing for
≥
6 ENOB for the displayed 200 GHz.
Figure 5(e) also indicates that the ENOB could go up to 9.6 over a 50 GHz span or up to 8.5 over
Research Article Vol. 29, No. 15 / 19 July 2021 / Optics Express 23679
100 GHz span when filtering the laser with an additional narrow filter between laser and MZM.
The analysis shows that a 1 GHz filter would suffice for that purpose.
Fig. 5.
ENOB is displayed for single effects as well as the total ENOB, which is compared
to simulated value. For comparison, three cases are shown: (a) Lumerical simulation setup:
RIN is generated in CW-laser. Freqeuncy sweep is done with the signal generator (SG).
Distortions are scaled with the SG output amplitude. A jitter source is inserted, fed by
measured values. A band pass filter (BPF) filters RIN and Distortions. SINAD is measured
by a optical spectrum analyzer (OSA). An optional BPF (dashed) could decrease the RIN
significantly. (b)
RIN =−
145 dBc,
BWoptical =
1 nm,
α=
1.13,
σrms =
53 fs (SG). (c)
RIN =−
165 dBc,
BWoptical =
0.3 nm,
α=
0.44,
σrms =
160 fs (SG) . (d)
RIN =−
165 dBc,
BWoptical =0.3 nm, α=0.44, σrms =53 fs (AWG). (e) RIN =−165 dBc, BWoptical =0.3
nm, α=0.53, σrms =4 fs (LPN-SG).
As future work, the ENOB prediction can be more refined by considering non-idealities
of the utilized electro-optic modulators or variable attenuators, which are used in the parallel
branches of the Nyquist pulse synthesizer DAC system (s. Figure 1(a)). In addition, since perfect
reconstruction, as property of the Nyquist pulses, is only possible at the zero-crossings, errors of
Research Article Vol. 29, No. 15 / 19 July 2021 / Optics Express 23680
the delays in the parallel branches will be transformed to amplitude errors. Hence, circuits or
systems that fine-tune the delay and suitable calibration techniques should be taken into account
as well. If the transition from the proposed photonic DAC into electrical domain is desired,
additional effects such as photo current shot-noise, absolute optical power, linearity and thermal
noise of the photo detector and an optional, subsequent transimpedance amplifier need to be
considered.
4. Conclusion
We have derived an analysis of the impairments of a Nyquist pulse synthesizer system by laser
RIN, distortion by the MZM non-linearity and electrical signal source jitter. The analysis
allows to predict an upper limit on ENOB, which is an important metric for high-speed arbitrary
waveform signal generators and DACs. For high ENOB the CW laser signal should exhibit a
low RIN and the laser signal should be filtered with a high-Q optical filter. In addition, the
frequency synthesizer phase noise should be as low as possible as it is the main limiting factor
in high-frequency performance. The best electronic DACs achieve 4.1 ENOB over a frequency
range of up to 58.6 GHz without frequency interleaving [4]. In conclusion, we have shown that a
photonic DAC based on Nyquist pulse synthesis outperforms state-of-the-art electronic DACs
both in bandwidth and ENOB. Our analysis predicts that using an ultra-low phase noise SG [26],
a laser source with -165 dBc RIN [25] and a 1 GHz pre-filter enables more than 8 ENOB over
more than 100 GHz signal bandwidth.
Funding. Deutsche Forschungsgemeinschaft (403154102).
Disclosures. The authors declare no conflicts of interest.
Data availability.
Simulation and measurement data underlying the results presented in this paper may be obtained
from the authors upon reasonable request.
References
1.
T. Ellermeyer, R. Schmid, A. Bielik, J. Rupeter, and M. Möller, “Da and ad converters in sige technology: Speed
and resolution for ultra high data rate applications,” in 36th European Conference and Exhibition on Optical
Communication, (2010), pp. 1–6.
2.
Y. M. Greshishchev, D. Pollex, S. C. Wang, M. Besson, P. Flemeke, S. Szilagyi, J. Aguirre, C. Falt, N. Ben-Hamida,
R. Gibbins, and P. Schvan, “A 56gs/s 6b dac in 65nm cmos with 256
×
6b memory,” in 2011 IEEE International
Solid-State Circuits Conference, (2011), pp. 194–196.
3.
K. Schuh, F. Buchali, W. Idler, Q. Hu, W. Templ, A. Bielik, L. Altenhain, H. Langenhagen, J. Rupeter, U. Duemler, T.
Ellermeyer, R. Schmid, and M. Moeller, “100 gsa/s bicmos dac supporting 400 gb/s dual channel transmission,” in
ECOC 2016; 42nd European Conference on Optical Communication, (2016), pp. 1–3.
4.
M. Collisi and M. Möller, “A 120 gs/s 2:1 analog multiplexer with high linearity in sige-bicmos technology,” in 2020
IEEE BiCMOS and Compound Semiconductor Integrated Circuits and Technology Symposium (BCICTS), (2020), pp.
1–4.
5.
X. Chen, S. Chandrasekhar, S. Randel, G. Raybon, A. Adamiecki, P. Pupalaikis, and P. J. Winzer, “All-electronic
100-ghz bandwidth digital-to-analog converter generating pam signals up to 190 gbaud,” J. Lightwave Technol.
35
(3),
411–417 (2017).
6. S. Cundiff and A. Weiner, “Optical arbitrary waveform generation,” Nat. Photonics 4(11), 760–766 (2010).
7.
D. Patel, A. Samani, V. Veerasubramanian, S. Ghosh, and D. V. Plant, “Silicon photonic segmented modulator-based
electro-optic dac for 100 gb/s pam-4 generation,” IEEE Photonics Technol. Lett. 27(23), 2433–2436 (2015).
8.
I. G. López, P. Rito, D. Petousi, S. Lischke, D. Knoll, M. Kroh, L. Zimmermann, M. Ko, A. C. Ulusoy, and D.
Kissinger, “Monolithically integrated si photonics transmitters in 0.25 um bicmos platform for high-speed optical
communications,” in 2018 IEEE/MTT-S International Microwave Symposium - IMS, (2018), pp. 1312–1315.
9.
Y. Sobu, S. Tanaka, Y. Tanaka, Y. Akiyama, and T. Hoshida, “High-speed, multi-level operation of all-silicon
segmented modulator for optical dac transmitter,” in 2020 IEEE Photonics Conference (IPC), (2020), pp. 1–2.
10.
A. Yacoubian and P. Das, “Digital-to-analog conversion using electrooptic modulators,” IEEE Photonics Technol.
Lett. 15(1), 117–119 (2003).
11.
F. Zhang, B. Gao, and S. Pan, “Serial photonic digital-to-analog converter based on time and wavelength interleaving
processing,” in 2016 25th Wireless and Optical Communication Conference (WOCC), (2016), pp. 1–4.
12.
T. Zhang, Q. Qiu, Z. Fan, J. Su, and M. Xu, “Experimental study on a 4-b serial optical digital to analog convertor,”
IEEE Photonics J. 10(2), 1–9 (2018).
Research Article Vol. 29, No. 15 / 19 July 2021 / Optics Express 23681
13.
P. Kondratko, A. Leven, Y.-K. Chen, J. Lin, U.-V. Koc, K.-Y. Tu, and J. Lee, “12.5-ghz optically sampled
interference-based photonic arbitrary waveform generator,” IEEE Photonics Technol. Lett.
17
(12), 2727–2729 (2005).
14.
M. Khafaji, M. Jazayerifar, K. Jamshidi, and F. Ellinger, “Optically-assisted time-interleaving digital-to-analogue
converters,” in 2016 IEEE Photonics Society Summer Topical Meeting Series (SUM), (2016), pp. 216–217.
15.
J. Meier and T. Schneider, “Precise, high-bandwidth digital-to-analog conversion by optical sinc-pulse sequences,” in
2019 12th German Microwave Conference (GeMiC), (2019), pp. 166–169.
16.
K. Singh, J. Meier, A. Misra, S. PreuBler, J. C. Scheytt, and T. Schneider, “Photonic arbitrary waveform generation
with three times the sampling rate of the modulator bandwidth,” IEEE Photonics Technol. Lett.
32
(24), 1544–1547
(2020).
17.
A. Misra, C. Kress, K. Singh, S. PreuBler, J. C. Scheytt, and T. Schneider, “Integrated all optical sampling of
microwave signals in silicon photonics,” in 2019 International Topical Meeting on Microwave Photonics (MWP),
(2019), pp. 1–4.
18.
K. Singh, J. Meier, A. Misra, S. Preussler, and T. Schneider, “Nyquist data transmission with threefold bandwidth of
the utilized modulator,” in OSA Advanced Photonics Congress (AP) 2020 (IPR, NP, NOMA, Networks, PVLED, PSC,
SPPCom, SOF), (Optical Society of America, 2020), p. NeTu2B.6.
19.
A. Misra, C. Kress, K. Singh, S. Preußler, J. C. Scheytt, and T. Schneider, “Integrated source-free all optical
sampling with a sampling rate of up to three times the rf bandwidth of silicon photonic mzm,” Opt. Express
27
(21),
29972–29984 (2019).
20.
C. Wang, M. Zhang, X. Chen, M. Bertrand, A. Shams-Ansari, S. Chandrasekhar, P. Winzer, and M. Loncar,
“Integrated lithium niobate electro-optic modulators operating at cmos-compatible voltages,” Nature
562
(7725),
101–104 (2018).
21.
M. Soto, M. Alem, M. A. Shoaie, A. Vedadi, C.-S. Brés, L. Thévenaz, and T. Schneider, “Optical sinc-shaped nyquist
pulses of exceptional quality,” Nat. Commun. 4(1), 2898 (2013).
22.
M. A. Soto, M. Alem, M. A. Shoaie, A. Vedadi, C. Brés, L. Thévenaz, and T. Schneider, “Generation of nyquist sinc
pulses using intensity modulators,” in CLEO: 2013, (2013), pp. 1–2.
23.
N. Da Dalt, M. Harteneck, C. Sandner, and A. Wiesbauer, “On the jitter requirements of the sampling clock for
analog-to-digital converters,” IEEE Trans. Circuits Syst. I: Fundamental Theory Appl. 49(9), 1354–1360 (2002).
24.
J. M. Senior, Optical Fiber Communications (2nd Ed.): Principles and Practice (Prentice Hall International (UK)
Ltd., GBR, 1993).
25.
More Photonics, “Apic cwl-100-1550-165 ultra-low rin dfb laser,” https://morephotonics.com/products/laser-
modules/low-rin-dfb-lasers/cwl-100-1550-168/ (2021).
26.
M. Bahmanian, S. Fard, B. Koppelmann, and J. C. Scheytt, “Wide-band frequency synthesizer with ultra-low phase
noise using an optical clock source,” in 2020 IEEE/MTT-S International Microwave Symposium (IMS), (2020), pp.
1283–1286.
