This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LPT.2016.2627000, IEEE Photonics
Technology Letters

Direct Modulation of a Laser Using 112 Gb/s

16-QAM Nyquist Subcarrier Modulation
Yuliang Gao, Member, IEEE, John C. Cartledge, Fellow, IEEE, Aazar S. Kashi, Student Member, IEEE,
Scott. S.-H. Yam, Member, IEEE, and Yasuhiro Matsui
Abstract—A 112 Gb/s single-carrier, single-polarization short [14]. DMLs in high-speed short reach systems introduce
reach transmission system using a high-bandwidth 1310 nm di- nonlinear signal distortion due to the modulation dynamics
rectly modulated laser is demonstrated. Spectrally-efficient signal of the laser [25]. The pattern-dependent distortion results
generation is achieved by using (i) half-cycle Nyquist subcarrier
modulation with 16-ary quadrature-amplitude-modulation, (ii) from the resonant properties of the laser and the coupling
pre-compensation for the frequency response of the end-to- between the carrier and photon densities. The resultant penalty
end system, (iii) and a Volterra nonlinear equalizer to post- can be mitigated by reducing the modulation current or by
compensate for the nonlinear modulation dynamics of the laser. back calculation of the rate equations at the transmitter [26],
Digital signal processing techniques for signal recovery enable however, they either sacrifice the receiver sensitivity or require
transmission over 20 km of standard single mode fiber with a
bit error ratio below the 7% overhead hard-decision forward high sampling rate digital-to-analog converters. Thus, receiver
error correction coding threshold of 4:6 10 3 at a received post-compensation techniques for the DML nonlinear signal
optical power of 5:25 dBm. distortion are of interest for short reach systems.
Index Terms—Directly modulated laser, subcarrier modula- In this letter, a single-carrier, single-polarization 112 Gb/s
tion, intensity modulation, direct detection. short-reach transmission system is demonstrated using 16-
QAM half-cycle Nyquist subcarrier modulation (Nyquist-
I. I NTRODUCTION SCM) and a buried heterostructure distributed feedback laser
HE provisioning of high bandwidth internet services with a wavelength of 1310 nm [27]. This approach exhibits
T requires the operation of short-reach systems with bit
rates of 100 Gb/s and above. For an aggregate bit rate of
a good power efficiency with relatively low complexity as it
does not require clipping, Fourier transform operations in the
400 Gb/s, systems comprised of four 100 Gb/s signals with signal processing, adaptive power and bit loading of subcar-
wavelengths around 1310 nm are being investigated. Recently, riers that reflect received signal-to-noise ratios, or orthogonal
several single-carrier, single-polarization, short-reach 100 Gb/s filters. It allows the effective implementation of compensation
transmission experiments have been reported using external techniques and offers flexibility in terms of the subcarrier
modulators, electroabsorption modulated lasers or directly modulation format. A previous report of 112 Gb/s 16-QAM
modulated lasers. In these experiments, four-level pulse ampli- half-cycle Nyquist-SCM used an integrated electroabsorption
tude modulation (PAM-4) [1]-[6], discrete multi-tone (DMT) modulated laser [20]. Compared with our earlier results on
modulation [6]-[11], orthogonal PAM-4 DMT modulation half-cycle Nyquist SCM using a directly modulated laser [17],
[12], carrierless amplitude phase (CAP) modulation [6], [13]- the bit and symbol rates have been doubled as a result of
[16], Nyquist subcarrier modulation (SCM) [17]-[22], and on- including pre-compensation for the frequency response of the
off-keying have been used [23]. Intensity modulation/direct de- end-to-end system, and more importantly, post-compensation
tection (IM/DD) implementations that use directly modulated for the nonlinear signal distortion due to the laser, in a similar
lasers (DMLs) are an attractive approach for achieving a small way as has been done for PAM-4 [5]. Although nonlinear
footprint, low cost, low power dissipation, and high fiber- post-compensation increases the computational complexity, it
coupled power, which is particularly important for unamplified reduces the pre-FEC BER floor allowing the use of simple
applications [24]. Alternatively, a bit rate of 100 Gb/s can be FEC codes with low overhead (e.g., KP-4). The 112 Gb/s 16-
achieved by using polarization division multiplexing but this QAM half-cycle Nyquist-SCM signal meets the hard decision
impacts the implementation complexity. forward error correction (FEC) coding threshold of 4:6 10 3
In the context of single-carrier, single-polarization DML- [28] with a transmission distance of 20 km for standard single-
based solutions, lasers with emission in the O-band or C- mode fiber (SSMF) at a received optical power of 5:25 dBm.
band have been used, with of course, the implications of fiber
dispersion being important for the latter. A bit rate of 100 II. P RINCIPLE AND E XPERIMENTAL S ETUP
Gb/s using a DML has been demonstrated for PAM-4 [1], [4], The low-pass representation of an SCM signal s(t) can be
[5], DMT [7], orthogonal PAM-4 DMT [12], and multiCAP expressed as
X
This work was supported by the Natural Sciences and Engineering Research
Council of Canada under the CREATE NGON Project (385027-2010). Y.
s(t) = Ii p(t iTs ) cos (2 fs t) + Qi p(t iTs ) sin (2 fs t)
Gao, J. C. Cartledge, A. S. Kashi, and S. S.-H. Yam are with the Department i
of Electrical and Computer Engineering, Queen’s University, Kingston, ON, (1)
Canada (email: [email protected]). Y. Matsui is with Finisar Corp., where Ii and Qi denote the ith in-phase and quadrature
Fremont, CA, USA. symbols, p(t) is the square-root raised-cosine pulse shape,
Copyright (c) 2016 IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be Ts is the symbol period, and cos (2 fs t) and sin (2 fs t) are
obtained from the IEEE by sending a request to [email protected]. the two orthogonal carriers generated at a radio frequency

1041-1135 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LPT.2016.2627000, IEEE Photonics
Technology Letters

(RF) frequency fs [17]. To avoid inter-carrier interference,

the carrier frequency fs has to be set equal to or greater
than half the bandwidth of the baseband signal. The smallest
possible value of fs is half of the symbol rate 1=(2Ts ) for
Nyquist pulse shaping with a roll-off factor of 0, namely half-
cycle Nyquist-SCM. For 112 Gb/s single-polarization IM/DD
systems, the 16-QAM half-cycle Nyquist-SCM signal, with
four modulation levels for each of the in-phase and quadrature
components, is attractive for short-reach data transmission.
The experimental setup for 112 Gb/s half-cycle Nyquist-
SCM transmission is shown in Fig. 1. At the transmitter, a
215 de Bruijn bit sequence is mapped into the sequences of in-
phase and quadrature symbols fIi g and fQi g to produce the Fig. 1. Block diagram of the experimental setup.
four level signals. The symbols are up-sampled and applied
to square-root raised-cosine pulse shaping filters with a roll-
off factor of 0 to maximize the spectral efficiency. The on 16000 training symbols. The coefficients were then fixed, as
baseband signals were then up-converted to a carrier at 14 the nonlinear distortion introduced by the laser is essentially
GHz and re-sampled at a rate of 16/7 samples/symbol to deterministic, and training symbols were not inserted in the
produce the 112 Gb/s 16-QAM half-cycle Nyquist-SCM drive data sequences. The order three and a memory length of
signal. The waveform was pre-emphasized using the measured N = 7 represent a good compromise between performance
transfer function for the end-to-end system and loaded into and complexity for this particular application. Thus, the VNLE
the programmable memory of a digital-to-analog converter compensation requires L = 399 real multiplications for
(DAC) with a bandwidth of 25 GHz, a resolution of 8 bits, processing each input sample value. After the VNLE, the
and a sampling rate of 64 GSa/s. An electrical amplifier with post-compensated signal was down-converted to separate the
a bandwidth of 50 GHz was used to generate the required I and Q signals and band-limited to 14 GHz using a anti-
modulation current Im for the DML. A DML with a 3-dB aliasing filter. After the I and Q components of the signal
modulation bandwidth of 25 GHz and output power of 11 were orthogonalized using the Gram-Schmidt algorithm [30],
dBm at a wavelength of 1310 nm, as described in [27], was a digital square and filter clock recovery algorithm was used to
used enabling a setup without optical amplification and dis- downsample the received signal from 80/28 samples/symbol
persion compensation. Both the modulation and bias currents to 2 samples/symbol [31] and a digital equalizer was used
of the DML were optimized as they impact the modulation to compensate for residual signal distortion. The transversal
bandwidth and nonlinear signal distortion due to the laser Ts =2 fractionally-spaced linear equalizer was adapted using
modulation dynamics. After 20 km of SSMF, a variable optical the constant modulus algorithm for pre-convergence followed
attenuator (VOA) was used to adjust the received optical by the radius directed algorithm to compensate for the residual
power (ROP). The fiber attenuation was 0:35 dB/km, the zero linear distortions [32], [33]. For both algorithms the equalizer
dispersion wavelength was 1312 nm, and the dispersion slope had 21 taps. The carrier phase was recovered by using the
was 0:092 ps/(nm2 km). The signal was then detected by a p- blind phase search algorithm with 32 test phases [34]. Finally,
i-n photodiode (PIN) and integrated transimpedance amplifier symbol to bit mapping and bit error counting were performed.
(TIA) with a bandwidth of 38 GHz. The received signal was Results were obtained for a back-to-back system and for
digitized by an 80 GSa/s analog-to-digital converter (ADC) transmission over 20 km of SSMF.
using a real-time sampling oscilloscope (2:8 105 symbols)
with 32 GHz bandwidth for offline processing. III. R ESULTS
At the receiver, the captured 16-QAM Nyquist-SCM signal The small-signal S21 response of the DML obtained using
was normalized, match filtered, and equalized using a Volterra a 65 GHz vector network analyzer is shown in Fig. 2. Results
nonlinear equalizer (VNLE). The time domain discrete formu- are presented for a bias current of Ib = 60 mA with a peak-to-
lation for the Volterra series expansion up to third-order for a peak modulation current of 1.83 mA. The S21 response of the
nonlinear system with finite memory is given by bias-tee was de-embedded from the measured results. The 3dB
3 N modulation bandwidth is 25 GHz. The frequency responses
X X1 X1
N
y(n) = hp (k1 ; kp ) of the other opto-electronic and electronic components in the
p=1 k1 =0 kp =0 system adversely impact the signal quality. Thus, the end-to-
x (n k1 ) x (n kp ) (2) end magnitude and phase responses were measured for each
value of the bias current by transmitting 320 frequency tones,
where x (n) and y (n) are the input and output signals of equally distributed over 32 GHz, and determining the change
the VNLE [29]. p denotes the order of the Volterra kernels, in the amplitudes and phases of the stimulus frequencies at
N is the memory length, and hp ( ) is the pth -order Volterra the receiver. To reduce the impact of additive noise, eight
kernel.PThe total number of kernel coefficients for VNLE is independent measurements were averaged.
3 i
L = i=1 N . The kernel coefficients were obtained in a Figure 2 also illustrates the normalized magnitude and
training phase using the recursive least square algorithm based phase responses of the end-to-end system including all the

1041-1135 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LPT.2016.2627000, IEEE Photonics
Technology Letters

Fig. 2. Magnitude and phase responses of the normalized S21 parameter of Fig. 4. Dependence of the bit error ratio on the peak-to-peak modulation
the DML and the end-to-end system frequency response, Ib = 60 mA. current, Ib = 60 mA and 70 mA. Received optical power is 3 dBm.

Fig. 3. Comparison of the RF spectra without and with pre-compensation Fig. 5. Dependence of the bit error ratio on the received optical power for
for the end-to-end frequency response, Ib = 60 mA and Im = 58 mApp . back-to-back and 20 km systems, Ib = 60 mA and Im = 58 mApp .

components between the input of the DAC and the output of 58.0 GHz for Ib = 60 mA and Im = 58:5 mApp .
the ADC. At an RF frequency of 28 GHz, a 13.5 dB magnitude The modulation current needs to be optimized in order
loss and 250 degree phase change are observed. Denoting the to balance the trade-off between the received signal power
continuous time magnitude and phase frequency responses by and the nonlinear distortion due to the modulation dynamics
jH(f )j and '(f ), respectively, the generated signal is digitally of the DML. The dependence of the BER on the peak-to-
pre-compensated as follows peak modulation current Im without and with the VNLE are
S(k) shown in Fig. 4 for a back-to-back system configuration and
s0 (n) = IF F T (3) a received optical power of -3 dBm. A small modulation
jH(k)j ej'(k)
current yields a low signal-to-carrier ratio (and low extinction
where S(k) is the Fourier transform of the discrete time ratio (ER) after down-conversion) and thus the signal is more
112 Gb/s 16-QAM half-cycle Nyquist-SCM modulating signal susceptible to additive receiver noise. Alternatively, a large
s(n), s0 (n) is the pre-compensated modulating signal, and modulation current offers an increase in the signal-to-noise
IFFT denotes the inverse Fourier transform operation. The ratio and ER but leads to high nonlinear distortion. In the
spectrum of the received signal obtained with a microwave region of near-optimum bias and modulation currents, the
spectrum analyzer is shown in Fig. 3 for Ib = 60 mA and a variation in the BER is relatively small. For Ib = 60 mA
modulation current of Im = 58:5 mApp without and with pre- the VNLE reduces the minimum BER by a factor of 3.7
compensation. The 12.5 dB power loss at a frequency of 28 from 1:68 10 3 to 4:49 10 4 . For comparison, results are
GHz for the uncompensated signal was equalized resulting in also shown in Fig. 4 for Ib = 70 mA; the VNLE reduces
a flat RF power spectrum expected for a Nyquist-SCM signal. the minimum BER by a factor of 3.9 from 1:77 10 3
4
The optical spectrum exhibited a 25 dB spectral width of to 4:58 10 . Without and with the VNLE the optimum

1041-1135 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LPT.2016.2627000, IEEE Photonics
Technology Letters

performance is very similar for the two values of the bias [10] S. Amiralizadeh, A. Yekani, and L. A. Rusch, “Discrete multi-tone
current. The optimum value of the peak-to-peak modulation transmission with optimized QAM constellations for short-reach optical
communications,” J. Lightw. Technol., vol. 34, no. 15, pp. 3515-3522,
current increases with the bias current; for Ib = 60 mA and 2016.
70 mA the optimum modulation currents are Im = 58:5 mApp [11] D. Pilori, C. Fludger, and R. Gaudino, “Comparing DMT variants in
and 63:8 mApp , respectively. medium-reach 100G optically amplified systems,” to appear J. Lightw.
Technol.
Figure 5 shows the dependence of the BER on the ROP [12] W. A. Ling, Y. Matsui, H. M. Daghighian, and I. Lyubomirsky, “112
for a back-to-back system and for transmission over 20 km of Gb/s transmission with a directly-modulated laser using FFT-based
SSMF. The results are for optimized values of Ib = 60 mA synthesis of orthogonal PAM and DMT signals,” Opt. Exp., vol. 23,
no. 15, pp. 19202-19212, 2015.
and Im = 58 mApp . For both the back-to-back and 20 km [13] M. I. Olmedo, T. Zuo, J. B. Jensen, Z. Qiwen, X. Xiaogeng, and I.
transmission systems, the results without the VNLE exhibit T. Monroy, “Multiband carrierless amplitude phase modulation for high
a BER floor of about 5 10 4 at a ROP of 0 dBm, which capacity optical data links,” J Lightw. Technol., vol. 32, no. 4, pp. 798-
804, 2014.
is mostly due to the nonlinear modulation dynamics of the [14] R. Puerta et al., “107.5 Gb/s 850 nm multi- and single-mode VCSEL
DML. By implementing the VNLE at the receiver to post- transmission over 10 and 100 m of multi-mode fiber,” Conf. Opt. Fiber
compensate for the nonlinear signal distortion, at a received Commun., paper Th5B.5, 2016.
[15] J. He, H. Dong, R. Deng, T. Li, and L. Chen, “Enhanced performance of
optical power of 0 dBm the BER is reduced by a factor of CAP system using an overlap frequency-domain equalization scheme,”
46.7 from 6:07 10 4 to 1:30 10 5 for the back-to-back IEEE Photon. J., vol. 8, no. 2, 6600308, 2016.
case. For 20 km transmission, the BER is 1:45 10 5 at a [16] L. Sun, J. Du, and Z. He, “Multiband three-dimensional carrierless
amplitude phase modulation for short reach optical communications,”
received optical power of 0 dBm. By using the VNLE, a 0.85 J Lightw. Technol., vol. 34, no. 13, pp. 3103-3109, 2016.
dB reduction in the required ROP at the 7% overhead hard [17] A. S. Karar and J. C. Cartledge, “Generation and detection of a 56-
decision FEC BER threshold of 4:6 10 3 [28] is achieved. Gb/s signal using a DML and half-cycle 16-QAM Nyquist-SCM,” IEEE
Photon. Technol. Lett., vol. 25, no. 8, pp. 757-760, 2013.
The FEC BER threshold is achieved for a ROP of 5:25 dBm [18] J. C. Cartledge and A. S. Karar, “100 Gb/s intensity modulation and
for both the back-to-back system and for 20 km transmission direct detection,” J. Lightw. Technol., vol. 32 , no. 16, pp. 2809-2814,
resulting in loss margins of 16.25 dB and 8.25 dB, respectively. 2014.
[19] M. S. Erkılınç et al., “Nyquist-shaped dispersion-precompensated sub-
carrier modulation with direct detection for spectrally-efficient WDM
IV. C ONCLUSION transmission,” Opt. Exp., vol. 22, no. 8, pp. 9420-9431, 2014.
A single-polarization, single-wavelength 112 Gb/s 16-QAM [20] K. Zhong et al., “Transmission of 112 Gbit/s single polarization half-
cycle 16QAM Nyquist-SCM with 25 Gbps EML and direct detection,”
Nyquist-SCM signal has been generated using a high speed Eur. Conf. Opt. Commun., paper P.4.9, 2015.
1310 nm DML and DSP-based pre-compensation for the [21] J. Tang, J. He, D. Li, M. Chen, and L. Chen, “64/128-QAM half-cycle
end-to-end system response and post-compensation for the subcarrier modulation for short-reach optical communications,” IEEE
Photon. Technol. Lett., vol. 27, no. 3, pp. 284-287, 2015.
nonlinear modulation dynamics of the laser using a VNLE. [22] Z. Li et al., “Signal-signal beat interference cancellation in spectrally-
The simple system configuration uses neither external modu- efficient WDM direct-detection Nyquist-pulse-shaped 16-QAM subcar-
lation nor wavelength/polarization multiplexing technologies. rier modulation,” Opt. Exp., vol. 23, no. 18, pp. 23694-23709, 2015.
[23] J. Lee et al., “Serial 103.125-Gb/s transmission over 1 km SSMF for
By using a VNLE, a pre-FEC BER of 4:6 10 3 can be met low-cost, short-reach optical interconnects,” Conf. Opt. Fiber Commun.,
for a back-to-back system and for transmission over 20 km of paper Th5A.5, 2014.
SSMF at a received optical power of 5:25 dBm. [24] M. Sharif, J. K. Perin, and J. M. Kahn, “Modulation schemes for single-
laser 100 Gb/s links: single-carrier,” J. Lightw. Technol., vol. 33, no. 20,
R EFERENCES pp. 4268-4277, 2015.
[25] L. A. Coldren, S. W. Corzine, and M. L. Mashanovitch, Diode Lasers
[1] Y. Matsui et al., “55-GHz bandwidth short-cavity distributed reflector and Photonic Integrated Circuits, Wiley, 2012, Chp. 5.
laser and its application to 112-Gb/s PAM-4,” Conf. Opt. Fiber Com- [26] A. S. Karar, J. C. Cartledge, J. Harley and K. Roberts, “Electronic pre-
mun., paper Th5B.4, 2016. compensation for a 10.7-Gb/s system employing a directly modulated
[2] S. Kanazawa et al., “Transmission of 214-Gbit/s 4-PAM signal using laser,” J. Lightw. Technol., vol. 29, no. 13, pp. 2069-2076, 2011.
an ultrabroadband lumped-electrode EADFB laser module,” Conf. Opt. [27] Y. Matsui, T. Pham, T. Sudo, G. Carey, and B. Young, “112 Gb/s WDM
Fiber Commun., paper Th5B.3, 2016. link using two directly modulated Al-MQW BH DFB lasers at 56 Gb/s,”
[3] H. Mardoyan et al., “Single carrier 168-Gb/s line-rate PAM direct Conf. Opt. Fiber Commun., paper Th5B.6, 2015.
detection transmission using high-speed selector power DAC for optical [28] F. Chang, K. Onohara, and T. Mizuochi, “Forward error correction for
interconnects,” J Lightw. Technol., vol. 34, no. 7, pp. 1593-1598, 2016. 100 G transport networks,” IEEE Commun. Mag., vol. 48, no. 3, pp.
[4] C. Yang et al., “IM/DD-based 112-Gb/s/lambda PAM-4 transmission S48-S55, 2010.
using 18-Gbps DML,” IEEE Photon. J., vol. 8, no. 3, 7903907, 2016. [29] J. Pan and C.-H. Cheng, “Nonlinear electrical compensation for the
[5] Y. Gao, J. C. Cartledge, S. S.-H. Yam, A. Rezania, and Y. Matsui, coherent optical OFDM system,” J. Lightw. Technol., vol. 29, no. 2,
“112 Gb/s PAM-4 using a directly modulated laser with linear pre- pp. 215-221, 2011.
compensation and nonlinear post-compensation,” Eur. Conf. Opt. Com- [30] I. Fatadin, S. J. Savory, and D. Ives, “Compensation of quadrature
mun., paper M.2.C.2, 2016. imbalance in an optical QPSK coherent receiver,” IEEE Photon. Technol.
[6] K. Zhong et al., “Experimental study PAM-4, CAP-16, and DMT for Lett., vol. 20, no. 20, pp. 1733-1735, 2008.
100 Gb/s short reach optical transmission systems,” Opt. Exp., vol. 23, [31] M. Oerder and H. Heyr, “Digital filter and square timing recovery,” IEEE
no. 2, pp. 1176-1189, 2015. Trans. on Commun., vol. 36, no. 5, pp. 605-612, 1988.
[7] C. Xie et al., “Single-VCSEL 100-Gb/s short-reach system using discrete [32] S. J. Savory, “Digital filters for coherent optical receivers,” Opt. Exp.,
multi-tone modulation and direct detection,” Conf. Opt. Fiber Commun., vol. 16, no. 2, pp. 804-817, 2008.
paper Tu2H.2, 2015. [33] I. Fatadin, D. Ives, and S. J. Savory, “Blind equalization and carrier phase
[8] L. Zhang et al., “Beyond 100-Gb/s transmission for 80-km SMF using recovery in a 16-QAM optical coherent system,” J. Lightw. Technol., vol.
direct-detection SSB-DMT at C-band,” J Lightw. Technol., vol. 34, no. 27, no. 15, pp. 3042-3049, 2009.
2, pp. 723-729, 2016. [34] T. Pfau, S. Hoffmann, and R. Noé, “Hardware-efficient coherent digital
[9] Z. Liu, B. Kelly, J. O’Carroll, R. Phelan, D. J. Richardson, and R. receiver concept with feedforward carrier recovery for M-QAM constel-
Slavík, “Discrete multitone format for repeater-less direct-modulation lations,” J. Lightw. Technol., vol. 27, no. 8, pp. 989-999, 2009.
direct-detection over 150 km,” J. Lightw. Technol., vol. 34, no. 13, pp.
3223-3229, 2016.

1041-1135 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.