For Peer Review Only
1.25
-
2.5 Gb/s Simpl
e Nyquist Transmitters for Coherent
UDWDM-PON with Enhanced Spectral Efficiency
Journal:
Fiber and Integrated Optics
Manuscript ID
UFIO-2018-2188.R2
Manuscript Type:
Original Article
Keywords:
beat phase modulation, DFB, Nyquist shaping, UDWDM-PON, coherent
detection
URL: http:/mc.manuscriptcentral.com/ufio Email: h2@alum.mit.edu
Fiber and Integrated Optics
For Peer Review Only
1.25 - 2.5 Gb/s Simple Nyquist Transmitters for Coherent UDWDM-
PON with Enhanced Spectral Efficiency
A coherent ultra-dense wavelength division multiplexing passive optical network
(UDWDM-PON) with high spectral efficiency is experimentally tested. The transmitters
consist of directly phase modulated distributed feedback (DFB) lasers through beat signals
whose duty-cycle and amplitude are digitally adjusted. The benefit of using Gaussian or
Nyquist pulse-shaping filters at the transmitter is evaluated through numerical simulations
and experiments. The results show that transmitting a Nyquist-shaped signal achieves a
25% spectral saving allowing to place 2.5 Gb/s/user data in 6.25 GHz channels. The
proposed transmitter tolerates differential link-losses of 15 dB with Rx sensitivity of -44
dBm at BER=10
-3
with intradyne detection.
Keywords: beat phase modulation, DFB, directly modulated laser (DML), Nyquist
shaping, UDWDM-PON, coherent detection.
1. Introduction
Motivated by the need of high spectral efficiency and aggregated capacity, coherent ultra-dense
wavelength division multiplexing (UDWDM) solutions have gained interest [1, 2]. These systems
are attractive because they can inherently filter the optical signal by selecting the wavelength, and
provide higher sensitivities compared with direct detection thus avoiding optical amplifiers [3].
However, for the competitive access market, low complexity transceivers with simplified digital
signal processing (DSP) need to be designed while keeping high the performance [4].
In UDWDM-PON, a major concern is the unwanted spectral side-lobes resulting when
modulating the optical carrier. Since the frequency grid is very narrow, these spectral components
cause interference to adjacent channels, yielding to Bit Error Rate (BER) degradation. The use of
spectral shaping filtering at the transmitter (Tx) thus can play an important role reducing the
modulated spectral width and suppressing interference from adjacent channels [5]. As a result, it
can improve the network spectral efficiency as demonstrated in [6]. However, in [6] the Tx was
highly complex because of using an external modulator.
In previous work, we proposed direct phase modulation of a distributed feedback (DFB)
laser for simplifying the Tx. The modulating data was coded and had three levels. In addition, its
amplitude and duty cycle (d-c) were digitally adjusted to achieve the phase variations [7]. In this
paper, we analyze the benefit of using Gaussian or Nyquist shaping filters in the Tx to increase
the spectral efficiency, extending the results of [8]. We first do an evaluation through numerical
simulations. Then we present the experimental validation in an UDWDM-PON. This is, to the
best of our knowledge, the first simple directly phase modulated Tx with spectral confinement for
a coherent UDWDM-PON. At the receiver (Rx), an intradyne detector recovered the data
achieving sensitivity values of -49 dBm and -44 dBm for bitrates (Rb) of 1.25 Gb/s and 2.5 Gb/s
respectively at BER of 10
-3
. Furthermore, sensitivity penalties < 1 dB are observed in channels
separated only by 6.25 GHz and tolerating a differential link-loss of 15 dB [9].
2. Numerical Simulations
The simulations were performed with VPITranmissionMaker® and MATLAB® following the
Monte-Carlo method. The simulation setup was composed of two users with Tx consisting of an
ideal phase modulator (PM) which generated a 0 - 180º phase shift keying (PSK) signal. Both
users were combined in a 3dB optical coupler as presented in Figure 1a.
The modulating data was a 2
15
-1 Pseudo Random Binary Sequence (PRBS) differentially
encoded at
of 1.25 Gb/s with rise-time
0.1
, where T
b
is the bit period.
Additional electrical filtering was applied to data before optical modulation to modify the
dynamics of the electrical pulses and shaping its spectrum.
Page 1 of 7
URL: http:/mc.manuscriptcentral.com/ufio Email: h2@alum.mit.edu
Fiber and Integrated Optics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
For Peer Review Only
Figure 1. (a) Simulation Setup. (b) Sensitivity penalty @BER=10
-3
vs normalized filter BW for the three
pulse-shaping schemes.
We considered two shaping filter types: Gaussian and Raised-cosine. In the first one, the
sharp symbol transitions were smoothed to Gaussian dynamics with a rise-time
depending on
the 3-dB filter bandwidth (BW) calculated as 0.35
⁄
. On the second hand, the Raised–
cosine adjusted the excess filter BW beyond the Nyquist BW: 12
⁄
. The parameter that defined
the filter was the roll-off (α) which ranged between 0 and 1. For this shaping technique, the lower
α, the lower the total filter BW and the higher spectral compression. The filter BW was expressed
as: 1 α
2
⁄
.
The optical power of each generated differential PSK (DPSK) signal was set to 0 dBm
and launched through 25 km of single-mode fiber (SMF). Then a variable optical attenuator
(VOA) emulated further splitting and adjusted the power at the Rx. The state of polarization
(SOP) of the transmitted signal was matched with that of the local oscillator (LO) by means of a
polarization controller (PC). The signal was detected with a coherent intradyne Rx based on a 3x3
optical coupler [10].
In order to find the optimal parameters (
andαfor the shaping filters, we set the input
power at R
X
at the value that produced BER = 10
-3
for Non-return-to-zero (NRZ) format. Then,
we computed the power penalty due to the Gaussian filter BW. The results are shown in Figure
1b. A power penalty larger than 1 dB was observed for 1.2
⁄
. Therefore, the Gaussian
filter was set at 1.2
⁄
(corresponding to
0.29
. For the Raised-cosine filter, the
power penalty at maximum bandwidth 1
⁄
(corresponding to α = 1) was 3.5 dB.
Figure. 2 depicts the original NRZ electrical spectrum before and after the filters. In the NRZ
signal, we observe that the sharp transitions between symbols result in high-frequency harmonic
components. There is significant power beyond 1.25 GHz and the main to secondary spectral
lobes power suppression (MSPS) is 14dB. With electrical Gaussian filtering, the symbol
transitions are smoothed and the MSPS is 18 dB. When using the Raised-cosine filter, the
electrical pulses are sinc-shaped and ideally exhibit a square spectrum. As seen in Figure 2, the
MSPS is >60 dB.
Figure 2. Electrical spectra of 1.25 Gb/s PRBS data and eye diagrams for each case: (a) NRZ, (b) Gaussian,
(c) Raised-cosine pulse shaping at the Tx.
Intradyne
Rx
1544.9nm
LO
25kmSMF
VOA
PC
DPSKTx
2
7.5GHz
DPSKTx
1
PM
CW
Pulse‐
shapingfilter
PRBS
Differential
encoder
(a) (b)
Page 2 of 7
URL: http:/mc.manuscriptcentral.com/ufio Email: h2@alum.mit.edu
Fiber and Integrated Optics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
For Peer Review Only
A second user was added into the network with identical pulse shaping. Figure 3 plots
the optical spectra of both channels separated 6.25 GHz. The NRZ, Gaussian shaped and Raised-
cosine filtered signals show an MSPS of 12 dB, 13 dB and 19 dB respectively. The latter thus is
expected to reduce the channel spectral separation. It is interesting to note that the optical MSPS
values are lower than the electrical because of the non-linear characteristic of the phase
modulation. This produced strong Bessel harmonic components in the optical domain.
Next, the minimum channel spacing was evaluated. For this, the second user optical
frequency was shifted from -7.5 GHz to 7.5 GHz with respect to the other user emission
frequency. The BER degradation was measured in the fixed user (Figure 4a). These values were
translated into power penalty at a sensitivity of BER=10
-3
considering the single channel as
reference (Figure 4b). The values at 1 dB penalty were 2.8 GHz, 2.2 GHz and 1.8 GHz for NRZ,
Gaussian and Raised-cosine pulse-shaping respectively. Notably, there was no penalty at a 6.25
GHz spacing and even less than 2.5 GHz could be achieved when shaping the pulse.
Figure 3. Optical spectra for two users separated by 6.25 GHz for three different pulse-shaping schemes:
(a) NRZ, (b) Gaussian, (c) Raised-cosine.
Figure 4. (a) BER against Channel Spacing for Tx
1
using three different pulse-shaping filters. (b) Sensitivity
penalty @BER=10
-3
for the three pulse-shaping schemes.
3. Experimental setup
Once the technique was evaluated through simulations, we implemented the UDWDM-PON
scenario depicted in Figure 5. The setup was composed by two identical transmitters (Tx
1
and
Tx
2
) based on direct modulated DFBs with linewidths of 4 MHz and 3 MHz and modulation BW
of 10 GHz and 2.5 GHz respectively. The DFBs were biased at 75 mA, value at which we
observed that the adiabatic chirp dominated.
Page 3 of 7
URL: http:/mc.manuscriptcentral.com/ufio Email: h2@alum.mit.edu
Fiber and Integrated Optics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
For Peer Review Only
The data consisted of two uncorrelated differentially encoded 2
15
-1 Pseudo Random
Binary Sequences (PRBS) at Rb of 1.25 Gb/s and 2.5 Gb/s. Each sequence modulated a separate
DFB. The data sequences were digitally equalized by means of a 1-tap finite impulse response
(FIR) filter with half-bit period delay. The result was a Dicode Return-to-zero (Dicode RZ) signal
whose duty-cycle was adjusted to 50%. Then, its spectrum was shaped with a Raised-cosine FIR
filter of 64-taps. The resulting samples were uploaded to a 20 GSa/s arbitrary waveform generator
(AWG). The outputs were amplified with an 8 GHz (SHF98P) and 12 GHz (TBI-781) BW
electrical amplifiers for Tx
1
and Tx
2
respectively. The waveform amplitude was optimized to
ensure 0-π phase changes when directly modulating the DFB and producing, as a result, optical
DPSK signals [7]. The optical signals were combined with a 3-dB optical coupler and sent through
25 km of single mode fibre (SMF) with a launched power of 0 dBm. A variable optical attenuator
(VOA) reproduced splitting losses and limited the power into the Rx.
The Rx consisted of an intradyne detector based on a 3x3 optical coupler which mixed
the incoming optical signal with a LO. The latter was a 100 kHz linewidth external cavity laser
(ECL) with 0 dBm optical power emitting at λ
LO
= λ
1
= 1544.9 nm. The three outputs of the optical
coupler were detected with 10 GHz p-i-n- photodiodes (PDs) followed by low-noise electrical
amplifiers. The electrical signals were low-pass filtered, sampled and processed with a 50 GSa/s
real-time oscilloscope (RTO). The three currents were combined and processed to obtain the in-
phase (I) and quadrature (Q) parts as in [10]. Each component was then differentially demodulated
with a bit-delay and multiply operation, and then both were added. Afterwards, the samples
passed through a 4th order low-pass filter with
cut-off frequency and the BER was computed.
Figure 5. Experimental setup; the inset shows the data stream with Nyquist shaping and the RF spectra of
both Txs spaced 6.25 GHz.
4. Results
Two electrical signals, one with and another without Nyquist-shaping, were digitally
generated. Figure 6a shows the electrical data at Rb = 1.25 Gb/s along with the Nyquist shaped
waveform for a roll-off factor (α) of 0.25 (upper) and 1 (lower). Figures 6b, 6c, 6d and 6e present
the photo-detected spectra at a relative intermediate frequency (IF) of 10 GHz for signals at
bitrates of 1.25 Gb/s and 2.5 Gb/s. We observe that the lateral lobes are eliminated, but the main
lobe is not compacted as expected with externally modulated Txs [6]. This is because we are
directly modulating the phase of the laser, which has nonlinear dynamics, and leads to harmonics
due to the Bessel function solution. However, the BW of the residual intensity modulation (IM)
is narrowed. This is particularly noticed for lower α value. In contrast with PM, the IM residual
term is generated with a linear modulation. Finally, we decided to use α = 1 since the effect of
lower α is barely perceived.
Page 4 of 7
URL: http:/mc.manuscriptcentral.com/ufio Email: h2@alum.mit.edu
Fiber and Integrated Optics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60