Duobinary formats are today considered as being one of the most promising cost-effective solutions for the deployment of 40 Gb/s technology with direct detection on existing 10 Gb/s WDM long-haul (metropolitan and core) transmission infrastructures. Various methods for generating duobinary formats have been developed in the past few years but to our knowledge their respective performances for 40 Gb/s transmission have never been really compared experimentally. Here, we propose to evaluate at 40 Gb/s their respective robustness with respect to the most stringent transmission impairments, namely ASE noise, chromatic dispersion, polarization mode dispersion and nonlinear effects. We demonstrate that, owing to its enhanced resistance to intra-channel nonlinearities as compared to non-return-to-zero, duobinary can permit to reach transmission distances compliant with metropolitan and core applications on G.652 standard single mode fibre when quasi single-channel transmission conditions are met. We show furthermore that shifting optical duobinary filtering from the transmitter output to the receiver input can be of high interest to improve further the system maximum reach. We show also that phase-shaped binary transmission (PSBT) formats are fully compliant with 50-GHz channel spacing and that they are, in terms of transmission performance, as good as partial differential phase shift keying (Partial-DPSK), which is considered by equipment suppliers as the preferential transport solution for deployment of 40 Gb/s technology with direct detection on existing 10 Gb/s WDM metropolitan and core transmission infrastructures.
© 2012 OSA
Duobinary formats are known for their low spectral occupancy and high tolerance to residual chromatic dispersion [1–14]. These particular features make them very attractive for both high bit-rate and high spectral efficiency direct-detected optical transmissions: up to 0.8 bit/s/Hz at 40 Gb/s per channel has been demonstrated recently . Even if coherent technology is today considered as the most promising solution for 40 Gb/s long-haul transmission, duobinary techniques can still be interesting to address “niche” applications in the metropolitan and core transport networks for which equipment cost constitutes the main decision parameter of customers. Various types of duobinary transmitter, based on delay-and-add method  or low-pass filtering , applied in the electrical or optical domain, have been developed in the past few years but to our knowledge their respective performances for 40 Gb/s long-haul (metropolitan and core) transmission have never been really compared experimentally.
In this paper, after having described the various methods for generating duobinary formats, we make an extensive experimental evaluation at 40 Gb/s of their robustness to accumulation of amplified spontaneous emission (ASE) noise, chromatic dispersion (CD), polarization mode dispersion (PMD) but also to nonlinear transmission impairments when using the largely deployed G.652 standard single-mode fibre (SSMF). We show that transmission distances compliant with metropolitan and core applications can be reached when quasi single-channel transmission conditions (with respect to linear and nonlinear WDM crosstalk) are met. We also evaluate what is the best location for the optical duobinary filtering (at the transmitter output or at the receiver input) when duobinary is produced by using a DPSK transmitter as generator. In order to assess the ability of 40 Gb/s duobinary to be deployed on existing 10 Gb/s transport infrastructures, we carry out 50-GHz channel spacing WDM transmission experiments by using the two most promising duobinary formats (namely, “electrical” and “optical” PSBT). We compare their performance with that of Partial-DPSK  which is today considered by equipment suppliers as the preferential transport solution for deployment of 40 Gb/s technology with direct detection on existing 10 Gb/s long-haul transmission infrastructures. We show that long-haul transmission distances can be reached for these formats with or without the periodic insertion of 50-GHz wavelength selective switches (WSS). This last part of our study dedicated to the PSBT versus Partial-DPSK format comparison in the 50-GHz channel spacing configuration is totally new when compared to our previously published work  and conclude harmoniously the extensive numerical simulation study carried out in reference .
2. Theoretical description and experimental set-up
Duobinary modulations are known as 3-levels formats that are generated by using differential precoding and electrical or optical filtering. Compared to NRZ, duobinary formats possess a phase modulation in addition to amplitude modulation. Typically, in duobinary, a π-phase shift takes place between two groups of “1”s when the number of “0”s in-between is odd. Among various developed duobinary transmitters, we can distinguish two categories of formats, which are currently denominated as standard duobinary and phase-shaped binary transmission (PSBT) formats. Classically, duobinary and PSBT formats are generated by means of an electrical or optical filtering, based either on the delay-and-add method (duobinary)  or on the low-pass filtering technique (PSBT) . While earlier duobinary and PSBT formats were generated through electrical filtering, optical filtering was proposed recently to generate high bit-rate duobinary and PSBT (in particular at 40 Gb/s) while overcoming complexities inherent to high-speed electronics [3–9]. In this study, we have tried to implement each of these two techniques.
Let us firstly describe the main properties of duobinary transformation from a theoritical point of view. Following Lender and Sekey [16,17], we consider a binary sequence an in which the values an = 1 or 0 are assumed with the same probabilities p(1) = p(0) = 1/2. A duobinary sequence bn is thus generated by the rules detailed hereafter:
- • if an = 0, then bn = 0
- • if an = 1, then the polarity of bn depends on the polarity of bn-k corresponding to the last an-k = 1
- • if an and an-k are separated by an even number of “0”s, bn = + bn-k
- • if an and an-k are separated by an odd number of “0”s, bn = - bn-k
- • if the binary sequence an = 000011001110110110001101010000111011
- • then the duobinary sequence bn = 0000 + + 00 + + + 0–000 + + 0-0 + 0000 + + + 0–
To better illustrate the respective specificities of binary and duobinary modulation, Fig. 2 shows intensity and phase characteristics of an identical “1110010100111” sequence implemented with the NRZ, duobinary and PSBT formats, respectively. The average power of the three sequences is identical. When compared to what has been explained before where only amplitude of binary and duobinary sequences is considered, both intensity and phase are plotted here. The correspondence is then very simple: a “+” digit (as previously introduced) corresponds to a mark in terms of intensity and a “0” in terms of phase, while a “-” digit corresponds to a mark in terms of intensity and a “π” in terms of phase. At the opposite of NRZ and duobinary formats for which energy is absent of the “0”s, PSBT formats are characterized by the presence of small bounces of energy between consecutive spaces, which are furthermore combined with π-phase shift occurring in their middle. This association limits drastically the impact of inter-symbol interference (ISI) , but the presence of energy inside the spaces limits in the same time the eye diagram opening, which itself degrades the back-to-back OSNR sensitivity of PSBT formats as compared to NRZ (as it is shown later in the manuscript). With respect to PSBT, duobinary shows useful π-phase shifts only when the number of “0”s between the groups of marks is odd. The other phase shifts, observed on Fig. 2, occurring on the duobinary sequence (i.e. when the number of “0”s is even) are inefficient to limit ISI as any energy is associated to them. While, with PSBT format, each group of marks is protected from the ISI generated by its neighbors owing to the previously described bounces of energy associated with π-phase shifts, with duobinary only the groups of “1”s separated by an odd number of “0”s are resistant to ISI.
Figure 3 shows the configuration of the three duobinary and PSBT transmitters under study as well as their eye diagrams and spectra. “Optical” duobinary is obtained by means of the delay-and-add method applied in the optical domain: optical filtering takes place just after a DPSK transmitter and is realized through the utilization of the constructive port of a 1-bit delay-line interferometer (DLI) [3–5]. “Electrical” PSBT is carried out through the insertion of an electrical 5th order Bessel low-pass filter (LPF) located just after the pulse pattern generator (PPG): its cut-off frequency is ~11.2 GHz which represents ~28% of the data rate [2,11]. “Optical” PSBT is generated by the combination of a DPSK transmitter and an optical Gaussian band-pass filter (BPF) with a 3dB-bandwidth of ~22.4 GHz which represents 56% of the data rate [6–10]. A single-drive Lithium-Niobate Mach-Zehnder modulator (MZM) fed by a laser diode (LD) is used to generate the duobinary and PSBT formats: it is biased to the null transmission point and driven by a RF amplifier whose output peak-to-peak voltage is equal to 2Vπ (Vπ is the modulator switching voltage). Differential precoding is represented over the schemes of Fig. 3, but is not used in our experiments. Indeed, differentially precoding pseudo-random bit sequences (PRBS) which are delivered by the PPG results in identical pseudo-random bit sequences at the output of the differential precoder: it is thus useless to insert such a costly precoder in a laboratory experiment which uses PRBS.
The following comments can be immediately inferred from the observation of the spectra and eye diagrams of Fig. 3: “Electrical” PSBT is the most spectrally-efficient modulation format with a 3-dB bandwidth of ~0.140 nm, followed immediately by “Optical” PSBT whose spectrum has a 3-dB bandwidth of ~0.177 nm and “Optical” duobinary whose entire spectrum occupies a bandwidth of ~1.5 nm. As expected, the eye diagram of “Optical” duobinary looks like that of NRZ. However, it can be noticed that marks are noisy when compared to those of NRZ, while crossing points are shifted towards the bottom part of the eye diagram. From their side, the two PSBT transmitters show eye diagrams characterized by energy bounces inside the spaces, which are a bit more marked in the case of “Electrical” PSBT than in the case of “Optical” PSBT. Note that NRZ is used here as a reference and that quantum-limit PSBT as described in  is not studied here (due to its difficult practical implementation).
Figure 4 below depicts the scheme of the experimental set-up which is used to evaluate the back-to-back sensitivities of NRZ, duobinary and PSBT formats. The transmitter is constituted of eight laser diodes (LD) combined together with an optical multiplexer (MUX) which are ranged on the 200-GHz ITU grid from 1545.32 to 1556.56 nm. In this first experiment which has for objective to evaluate the single-channel back-to-back sensitivities of modulation formats, only one MZM is used: indeed, the effect of crosstalk between WDM channels is not under study here, as the channel spacing is as large as 200 GHz. Each of the four transmitters described over Fig. 3 is successively implemented and evaluated. The pulse pattern generator delivers 231-1 PRBS. Note that, as already mentioned, no differential precoder is used here. A RF driver amplifies the binary sequences in order to feed the MZM with the appropriated peak-to-peak voltage. Chromatic dispersion (CD) is emulated through pieces of positive or negative dispersion fibre and varies from −200 ps/nm to + 200 ps/nm. Differential group delay (DGD) is produced by a commercial first order PMD emulator (PMDE), and adopts values between −13 ps and + 13 ps. A 200-kHz polarization scrambler is located at its input. An amplified spontaneous emission (ASE) broadband noise source is coupled with the signal before the receiver in order to make vary the received optical signal to noise ratio (OSNR). A square flat-top optical filter is located at the receiver entrance to remove the ASE noise which is outside of the signal bandwidth. A 40-GHz bandwidth PIN photodiode detects the signal, feeds a RF pre-amplifier which is itself connected to a 40 Gb/s bit error rate tester (BERT).
3. Back-to-back sensitivity measurements: OSNR, chromatic dispersion, PMD
Figure 5 below presents the back-to-back OSNR sensitivity curves of the modulation formats under study for the central channel at 1550.12 nm. For this experiment, the CD emulator, polarization scrambler and PMD emulator are removed. Only the ASE broadband noise source is coupled to the signal before entering into the optical filter which has for role to suppress the ASE noise outside of the signal bandwidth. OSNR is degraded by decreasing the losses introduced by the variable optical attenuator (VOA) inserted just after the ASE noise source. Note that OSNR are measured with an optical spectrum analyzer (OSA) that has been tuned over the 0.5 nm resolution bandwidth. Even if the OSNR sensitivity measurements are often displayed in 0.1 nm resolution bandwidth, they are in reality done in 0.5 nm at 40 Gb/s, because the whole signal spectrum has to be included into the resolution bandwidth of the OSA. As the extra-rate required by the forward error correction (FEC) code has not been considered here (due to the PSBT 5th Bessel electrical LPF which has been ordered with a bandwidth of 11.2 GHz, i.e. 28% of the 40 Gb/s data rate), we consider hereafter that the error-free criteria is reached for a BER = 10−9. The 3-dB bandwidth of the square flat-top optical filter located at the receiver entrance has been optimized to obtain the best OSNR sensitivity, and the result of this optimization gives 75 GHz for both NRZ and “Optical” duobinary, and 50 GHz for “Electrical” and “Optical” PSBT in these quasi single-channel conditions. As observed on Fig. 5, NRZ format presents as expected the best OSNR sensitivity: for a BER = 10−9, the required OSNR in 0.5 nm is 15.2 dB. “Optical” duobinary comes just after with a required OSNR in 0.5 nm of 16.8 dB for a BER = 10−9. This 1.6 dB difference between NRZ and “Optical” duobinary can be explained by the worst quality of the eye diagram of “Optical” duobinary when compared to NRZ: on Fig. 3 above, we clearly look at that marks of “Optical” duobinary are noisy when compared to those of NRZ. At the opposite, NRZ eye diagram is very clean and “1”s are as thin as “0”s. We ascribe this phenomenon to the transmitter imperfections, and more particularly to the defaults of the (RF driver + MZM) couple which generates the DPSK sequence. Results of  are fully compliant with our own results, as the measured OSNR sensitivity difference between NRZ and “Optical” duobinary is in the range [1.3; 2.3] dB depending of the bandwidth tuning of the optical filter located at the receiver input. The OSNR sensitivity of “Electrical”and “Optical” PSBT for a BER = 10−9 is respectively 4.6 dB and 5.8 dB worse than that of NRZ (OSNR = 19.8 dB for “Electrical” PSBT, OSNR = 21 dB for “Optical” PSBT against OSNR = 15.2 dB for NRZ). As previously mentioned, the worst performance of “Electrical”and “Optical” PSBT when compared to NRZ is due to the remaining energy located in the “0” bit slots, which leads to eye closure and extinction ratio degradation (as shown on Fig. 3). The slightly degraded behavior of “Optical” PSBT when compared to “Electrical” PSBT is, as previously, explained by the imperfections of the DPSK transmitter and conversion Gaussian optical filter, which are used to generate the “Optical” PSBT format. Re-opening the eye diagram owing to a CD residue ( + 150 ps/nm) permits to improve substantially the OSNR sensitivity of “Electrical” PSBT, as shown in Fig. 5. A ~2-dB OSNR sensitivity improvement at BER = 10−9 (with respect to the 0 ps/nm case) is observed.
We now evaluate the sensitivity to cumulated chromatic dispersion (CD) of the modulation formats under study. We thus introduce pieces of positive or negative dispersion fibre in order to make vary CD at 1550.12 nm from −200 ps/nm to + 200 ps/nm. The ASE broadband noise source is used to make vary the received OSNR. Figure 6 plots the OSNR (in 0.5 nm) that it is necessary to have over the receiver to reach a BER = 10−9. We can firstly note that the degraded behaviour of “Electrical” and “Optical” PSBT with respect to NRZ in terms of OSNR sensitivity is counter-balanced by a particularly good resistance to residual CD. The acceptance window for “Electrical” and “Optical” PSBT at 1-dB OSNR penalty is respectively of 380 ps/nm and 320 ps/nm, whereas it is only 70 ps/nm for both NRZ and “Optical” duobinary. “Optical” duobinary has in fact the quasi same behaviour than NRZ: its comparable eye diagram as well as the absence of energy in “π” phase shifts which occur when the number of “0”s is even (as previously explained in Fig. 2) explains that its CD resilience is quite similar to that of NRZ. In the case of PSBT formats, the combination of energy bounces inside the “0”s with π phase shifts in their middle succeeds in limiting broadening of marks on the spaces, whether the number of “0”s between the groups of “1”s is odd or even . Of course, it is not the case with NRZ where timing leakage due to accumulated chromatic dispersion results in constructive interference of marks inside the “0”s.
These CD behaviors of NRZ and PSBT formats are very well exhibited on Fig. 7 below, where two characteristic binary sequences “...0010100...” and “00100100...” are represented when they are stressed by 0 ps/nm and + 170 ps/nm of residual CD for respectively NRZ and “electrical” PSBT formats. All the details on the numerical simulations that have permitted to obtain these results are given in reference . We clearly see that a CD residue of + 170 ps/nm distorts significantly the two “...0010100...” and “00100100...” binary sequences in the case of the NRZ format, but not in the case of “electrical” PSBT. For NRZ, a significant part of energy is transferred from the two marks towards the central “0” bit slots. It is absolutely not the case with “electrical” PSBT, for which energy of “1”s remains confined in marks whatever the type of binary sequences. We can also observe in the case of “electrical” PSBT that a CD residue of + 170 ps/nm increases the pulse peak power and thus re-opens the eye diagram. This particular feature explains why the required OSNR is lower (or equivalently, the performance is better) for PSBT formats when a certain amount of positive or negative CD has been accumulated (cf. the purple and red curves of Fig. 6) when compared to the 0 ps/nm area. This particular feature explains also very well why the OSNR sensitivity curve of “electrical” PSBT is better for + 150 ps/nm of cumulated CD than for 0 ps/nm, as shown on Fig. 5 above.
Finally, we investigate the tolerance of modulation formats to differential group delay (DGD). DGD is emulated thanks to a first-order PMD emulator described in Fig. 4. A 200-kHz fast polarization scrambler is inserted at the PMDE entrance, in order to average the PMD impact over all the states of polarization. This type of measurement is very reliable and avoids stabilizing the input state of polarization into the PMDE at 45° of its principal polarization axis. Furthermore, it gives nearly the same results than those obtained with the worst state of polarization method. CD is fixed to 0 ps/nm. Figure 8 below plots the OSNR penalty for a BER = 10−9 as a function of DGD. All formats offer equivalent performance with a typical acceptance window for a 1-dB OSNR penalty of ~12 ps. The various size of the spectrum bandwidth between the modulation formats under study does not seem to have a major impact over the DGD robustness.
4. Robustness to intra-channel transmission impairments
We investigate now the tolerance to intra-channel nonlinearities of the various modulation formats under study. Figure 9 presents the experimental set-up. The transmitter is still constituted of eight LD spaced by 200 GHz and ranged from 1545.32 to 1556.56 nm. They are combined together by a multiplexer before attacking a MZM. The MZM is driven by 231-1 PRBS. The architecture of each of the four transmitters under study is detailed over the Fig. 3 above. A pre-compensation stage of −700 ps/nm (at 1550.12 nm) is inserted between the transmitter and the recirculating loop, which is constituted of four 100-km G.652 SSMF spans compensated at 97% by slope-matched dispersion compensation fiber modules (DCF). In order to minimize loop polarization effects, a polarization scrambler synchronously modulated with the loop circulation period is included. The SSMF/DCF map losses are compensated by means of hybrid distributed Raman / Erbium-doped fiber amplifier (EDFA). The backward Raman gain was fixed to ~15 dB in the SSMF, while an EDFA was added to each span (after the DCF module) to provide the additional ~15 dB gain necessary to compensate for SSMF/DCF pair losses. The noise figure of EDFA is ~5.5 dB. The 1450-nm Raman pump is a fibre laser which provides ~500 mW of power to each of the four 100-km SSMF spans constituting the recirculating loop. A dynamic gain equalizer (DGE) flattens the loop gain and suppresses the ASE noise outside of the WDM multiplex bandwidth. After a pre-determined number of loop round trips, the measured channel at 1550.12 nm is selected with the square flat-top optical filter previously defined. The post-compensation is optimized at last by means of a tuneable dispersion compensation module (TDCM).
Figure 10 shows the BER as a function of the span input power per channel for the wavelength at 1550.12 nm and the various formats at 1200 km. Optimal span input powers are 1 dB higher for “Electrical” and “Optical” PSBT (~-1 dBm per channel) than for NRZ and “Optical” duobinary (~-2 dBm per channel), showing that PSBT formats are more resistant to intra-channel nonlinear effects than NRZ and “Optical” duobinary . Nonetheless, due to their significantly worse back-to-back OSNR sensitivity and in spite of a slightly better OSNR after 1200 km (1 dB more), “Electrical” and “Optical” PSBT have a worse BER than NRZ and “Optical” duobinary at their optimum span input power per channel. “Optical” PSBT presents the worst BER because it has, in back-to-back, the worst OSNR sensitivity and is probably a bit more sensitive to transmission impairments than “Electrical” PSBT.
In order to take advantage of the robustness of the DPSK format with respect to fibre nonlinearities, we decided to transfer the delay line interferometer (DLI) and the 22.4-GHz Gaussian optical band-pass filter located at the output of the DPSK transmitter (before the recirculating loop) to the input of the receiver (after the recirculating loop) . The bandwidth of the square flat-top optical filter located at the receiver entrance (cf. Figure 4) is not modified when compared to the initial tunings realized previously in back-to-back. Figure 11 presents the BER versus span input power per channel curves for the “Optical” duobinary and “Optical” PSBT formats when the DLI or 22.4-GHz Gaussian optical band-pass filter is located at the transmitter output or receiver input, respectively. With “Optical” duobinary, the optimal span input power per channel is increased of 1 dB when the DLI is shifted at the loop output, while the BER is improved of ½ decade at the optimum span input power. When considering “Optical” PSBT, the span input power per channel is identical but the BER is improved of ~1 decade when the 22.4-GHz Gaussian band-pass filter is transferred after the recirculating loop. Transferring the optical filter dedicated to duobinary / PSBT conversion after the transmission line permits to benefit from the enhanced robustness of DPSK format to fibre nonlinearities, and from the additional filtering of ASE noise (accumulated in the line) provided by this optical filter. These measurements fully validate the approach consisting to convert the DPSK signal into a duobinary or PSBT one at the receiver entrance rather than at the transmitter output.
Finally, we plotted in Fig. 12 the BER versus the transmission distance of the channel at 1550.12 nm for the various modulation formats under study. For each distance and each format, both span input power per channel and post-compensation are optimized. The pre-compensation was kept at −700 ps/nm. As expected, the significant BER advantage of both NRZ and “Optical” duobinary on PSBT formats at low transmission distances is progressively reduced with the accumulation of intra-channel nonlinearities, still proving the higher resilience of PSBT formats to this degradation. At 2400 km, there is nearly no difference between NRZ and the PSBT formats, while at 800 km the difference was of ~2 decades. As expected, Fig. 12 also shows the superior transmission performance of the “Optical” duobinary and “Optical” PSBT formats in the configuration where the DLI and Gaussian optical band-pass filter, respectively, are shifted from the transmitter output to the receiver input: ½ BER decade is gained at 2400 km for each of the two modulation formats when compared to the configuration where the filtering is carried out at the transmitter output.
5. WDM transmission with 50 GHz channel spacing and performance comparison between PSBT and Partial-DPSK modulation formats
A 16x40 Gb/s WDM transmission is now implemented to evaluate the performance of the “Electrical” and “Optical” PSBT formats to operate on a 50-GHz ITU grid. Indeed, “Optical” duobinary is definitely not compliant with 50-GHz channel spacing, as its spectrum and eye diagram as well as its general behaviour (OSNR, CD, PMD sensitivity) looks like that of NRZ format. As reference for the comparison that we are going to carry out, we use Partial-DPSK, which is today considered as the preferential low-cost transport solution for the deployment of 40 Gb/s technology with direct detection on legacy transport infrastructures.
Partial-DPSK is generated thanks to the combination of a standard DPSK transmitter and a delay line interferometer (DLI) in which the delay has been modified from one bit time (Tbit) to 0.65 bit period (0.65xTbit) . This 0.65xTbit DLI is then associated with a 40-GHz balanced photo-receiver. We firstly measure the back-to-back OSNR sensitivity curves of the “Electrical” PSBT, “Optical” PSBT and Partial-DPSK modulation formats when the channel spacing is successively fixed at 100 GHz and 50 GHz. For this experiment, we use two 100-GHz interleaved combs of eight channels, separately modulated by two MZMs, which are driven by 231-1 decorrelated PRBS (i.e. by the data and complementary data). The immediate neighbours of the channel under measurement do not carry the same information than this last one, and the effect of WDM crosstalk is now well-emulated. The 3-dB bandwidth of the square flat-top optical filter located at the receiver entrance is tuned to 50 GHz for both PSBT and Partial-DPSK modulation formats when the channel spacing is 50 GHz. When the channel spacing is increased up to 100 GHz, the 3-dB bandwidth of the optical filter does not change for the PSBT formats (50 GHz) and is enlarged to 65 GHz for Partial-DPSK. Note that the Gaussian optical filter of 22.4-GHz bandwidth which converts DPSK into “Optical” PSBT is located here before the square flat-top optical filter located at the receiver entrance. At the opposite, the 0.65xTbit DLI used with Partial-DPSK is necessarily located after this square flat-top optical filter, as Partial-DPSK requires a differential detector whose two photodiodes are directly connected to the DLI outputs. Figure 13 below gives the OSNR sensitivity curves (i.e. the BER versus OSNR measured in 0.5 nm) of the “Electrical” PSBT, “Optical” PSBT and Partial-DPSK formats. For a channel spacing of 100 GHz, we can notice first that Partial-DPSK has a better OSNR sensitivity than PSBT formats: for a BER = 10−9, its OSNR sensitivity is improved of ~2.3 dB with respect to “Electrical” PSBT and of ~3.4 dB with respect to “Optical” PSBT. For a channel spacing of 50 GHz, the OSNR sensitivity of Partial-DPSK is better than those of “Electrical” and “Optical” PSBT but only up to the BER = 10−6 and the OSNR = 15dB. Beyond this limit, the OSNR sensitivity curve of Partial-DPSK is worse than those of PSBT formats, and reaches an error floor around the BER = 10−9. At the opposite, the BER versus OSNR curves of “Electrical” and “Optical” PSBT do not exhibit an error floor for 50-GHz channel spacing and are nearly super-imposed with the sensitivity curves measured for 100-GHz channel spacing, showing the good immunity of PSBT formats to WDM crosstalk caused by 50-GHz channel spacing. The presence of an error-floor around the BER = 10−9 for Partial-DPSK, and its absence for the PSBT formats can be explained by the configuration of our experimental set-up. Indeed, the combination of odd and even channels with a 3-dB coupler before the receiver is detrimental for Partial-DPSK as the format conversion (from Classical-DPSK to Partial-DPSK) through the 0.65xTbit DLI takes place behind the square flat-top optical filter located at the receiver entrance. Linear WDM crosstalk between the 50-GHz spaced Partial-DPSK channels is thus exacerbated. This is not the case with the PSBT formats, for which the format conversion takes place before the square flat-top optical filter located at the receiver entrance. This position of the conversion filter limits drastically the linear WDM crosstalk in the 50-GHz channel spacing configuration for the PSBT formats. Moving the multiplexer located into the transmitter from the MZM input to the MZM output, as it is the case into industrial WDM systems, would have removed this error-floor in the case of Partial-DPSK and resulted in nearly similar performance for both Partial-DPSK and PSBT formats in the 50-GHz channel spacing configuration.
Furthermore, as explained in , the OSNR sensitivity of Partial-DPSK in the 50-GHz channel spacing configuration can be improved if the DLI is tuned differently of 0.65xTbit: introducing for instance a delay of 0.86xTbit improves the required OSNR of nearly 2 dB for the BER = 1x10−5. At the opposite, when in-line add-drop multiplexers are inserted, 0.65xTbit Partial-DPSK becomes better than 0.86xTbit Partial-DPSK. After only four add-drop multiplexers, reference  measures more than 2 dB of gain for the 0.65xTbit Partial-DPSK over the 0.86xTbit Partial-DPSK in terms of OSNR sensitivity (for the BER = 1x10−5). The robustness to add-drop multiplexer cascade being essential in modern meshed optical transport networks, equipment suppliers have chosen to rather implement 0.65xTbit Partial-DPSK for their industrial 40 Gb/s transponders. Consequently, in this work, in order to be as close as possible of the industrial implementation of Partial-DPSK, we have also chosen to study the 0.65xTbit Partial-DPSK format, which provides furthermore better resistance to chromatic dispersion than 0.86xTbit Partial-DPSK. Tolerance of 0.65xTbit Partial-DPSK to accumulated CD and DGD is thus reported in Fig. 14 and Fig. 15 . The acceptance window of Partial-DPSK at 1-dB OSNR penalty is nearly 120 ps/nm, which is almost two fold higher than that of NRZ (~70 ps/nm) but three fold worse than that of “Electrical” and “Optical” PSBT (~380 ps/nm and 320 ps/nm, respectively). In terms of DGD, the performance of Partial-DPSK, “Electrical” and “Optical” PSBT is quite similar, as shown in Fig. 15.
The transmission experiment is depicted in Fig. 16 below. As compared to the previous set-up detailed in Fig. 9, sixteen channels are operated and ranged on a 50-GHz ITU grid. The eight odd and even channels are separately modulated in a MZM by 231-1 decorrelated PRBS (i.e. by the data and complementary data), and combined through a polarization-maintaining 3-dB coupler. The recirculating loop is similar to the previously described one, except that a 50-GHz wavelength selective switch (WSS) is introduced into the loop. This WSS separates the incoming 50-GHz spaced WDM channels in two groups of 100-GHz spaced wavelengths, which are filtered by the WSS transfer function and sent in two different output fibres. A decorrelating fibre (i.e. a spool of fifty meters of SSMF) is inserted over one of the two channel groups at the WSS output. When Partial-DPSK is implemented, the 40 Gb/s receiver represented on Fig. 16 (adapted in the drawing below to the PSBT formats) is replaced by a module combining a 0.65xTbit DLI, a 40 GHz balanced photo-receiver and a 40 Gb/s BERT.
Figure 17 plots the BER as a function of the span input power per channel for the central wavelength at 1550.12 nm after 1200 km of propagation into the recirculating loop in diverse configurations. The curves corresponding to the quasi single-channel transmission (i.e. 200 GHz channel spacing) are plotted here for the Partial-DPSK, “Electrical” and “Optical” PSBT formats. Note that, as previously, the 22.4-GHz Gaussian optical band-pass filter is located at the receiver entrance (i.e. after the recirculating loop) in the case of “Optical” PSBT. The six other configurations correspond to the 50-GHz channel spacing propagation of our three modulation formats in the two following cases: the 50-GHz WSS is successively inserted and removed of the transmission line. We firstly observe that the optimum span input power per channel is nearly the same (i.e. −1 dBm per channel) for the various configurations under study. For this span input power, the BER is equal to ~1x10−6 for the PSBT formats and ~1x10−7 for Partial-DPSK when the channel spacing is fixed to 200 GHz. These results show that Partial-DPSK is superior to PSBT in these quasi single-channel propagation conditions. It results mainly from the better OSNR sensitivity in back-to-back of Partial-DPSK (shown in Fig. 13) when the channel spacing is large enough.
When the channel spacing is decreased to 50 GHz, more than one and half BER decade is lost at the optimum span input power per channel (i.e. −1 dBm per channel). In these conditions, “Electrical” PSBT and Partial-DPSK have nearly the same performances with or without the periodic insertion of the 50-GHz WSS. The “Optical” PSBT format is outperformed by “Electrical” PSBT and Partial-DPSK in all the configurations. The better performance of “Electrical” PSBT over “Optical” PSBT (when the WSS cascade is not inserted) seems proving the lower immunity of “Optical” PSBT against transmission impairments when the channel spacing is fixed at 50 GHz. However, the “Optical” PSBT format has an interesting behaviour, as it seems that the periodic insertion of the 50-GHz WSS improves the transmission performance as compared to the case where the WSS is completely removed from the recirculating loop. More than ½ BER decade difference is gained when the WSS cascade is present into the line, showing that the filtering transfer function resulting from the concatenation of the 50-GHz WSS cascade (which limits the WDM crosstalk in transmission) and the 22.4-GHz Gaussian band-pass filter at the receiver side is more adapted to this 50-GHz WDM transmission configuration than the 22.4-GHz Gaussian band-pass filter used alone at the receiver (which does not limit the WDM crosstalk in transmission).
In order to confirm these trends, the BER is plotted on Fig. 18 as a function of the transmission distance for the central channel at 1550.12 nm and for the “Electrical” and “Optical” PSBT as well as for the Partial-DPSK format. For each distance and each format, both span input power and post compensation are optimized. As expected, the quasi single-channel transmission configuration outperforms the other ones, and Partial-DPSK is the best format. When the channel spacing is fixed to 50 GHz, in particular at transmission distances superior or equal to 1200 km, “Electrical” PSBT and Partial-DPSK modulation formats have nearly the same transmission performance with or without the periodic insertion of the 50-GHz WSS cascade. At lower distances, “Electrical” PSBT works better than Partial-DPSK: it is due to the back-to-back OSNR sensitivity which is better in the case of “Electrical” PSBT than in the case of Partial-DPSK when the channel spacing is fixed to 50 GHz (cf. Figure 13 above). “Optical” PSBT is the poorest modulation format in terms of transmission performance: while “Electrical” PSBT and Partial-DPSK reach a BER = 10−3 after 2400 km of propagation (with or without the WSS), “Optical” PSBT gets to this limit after only 1600 km of transmission into the recirculating loop. It confirms what we have explained above, namely that “Optical” PSBT is less immune than the “Electrical” PSBT and Partial-DPSK formats against the crosstalk induced by the transmission impairments when the channel spacing is equal to 50 GHz. As previously observed, the periodic insertion of the 50-GHz WSS improves the performance of “Optical” PSBT (~½ BER decade can be gained) when compared to the case where no WSS is introduced into the line. It confirms that the 50-GHz WSS cascade is efficient to limit the WDM transmission crosstalk of the “Optical” PSBT format in the 50-GHz channel spacing configuration, as previously explained. This effect is also observed over the Partial-DPSK format at low transmission distances (i.e. at 400 km and 800 km): ~½ BER decade is gained. It is blurred at higher transmission reaches when the nonlinearities accumulation masks the other effects.
Our studies of the relative robustness of duobinary, PSBT, and NRZ formats have shown that “Electrical” and “Optical” PSBT are the most resistant to residual chromatic dispersion. However this improved chromatic dispersion resilience is obtained at the expense of a reduced back-to-back OSNR sensitivity. From its side, “Optical” duobinary behaves largely as NRZ. Duobinary and PSBT formats are as robust as NRZ when facing polarization mode dispersion. When considering quasi single-channel transmission conditions (i.e. with 200 GHz channel spacing), NRZ and “Optical” duobinary are the best modulation formats, followed by “Electrical” PSBT and “Optical” PSBT, which is the poorest format in terms of transmission performance. Shifting the optical filtering (i.e. the delay line interferometer in the case of “Optical” duobinary or the 22.4-GHz Gaussian band-pass optical filter in the case of “Optical” PSBT) permits to improve the performance in these quasi single-channel transmission conditions, since fibre propagation is carried out with the DPSK format, which is known for its nonlinearities robustness. NRZ and “Optical” duobinary are definitely not able to support 50-GHz channel spacing transmission, at the opposite of PSBT formats. The performance of “Electrical” and “Optical” PSBT is thus compared to that of Partial-DPSK, which is considered as the modulation format reference for 50-GHz spaced 40 Gb/s WDM transmission with direct detection. Two 50-GHz configurations have been then evaluated for each modulation format: a 50-GHz wavelength selective switch (WSS) cascade has been successively inserted and removed from the transmission line. The obtained results indicate that “Electrical” PSBT has the same performance than Partial-DPSK, with or without the insertion of the 50-GHz WSS cascade. “Electrical” PSBT and Partial-DPSK outperforms largely “Optical” PSBT, which is significantly impacted by the WDM transmission impairments. Inserting periodically a 50-GHz WSS contributes to improve the transmission performance of “Optical” PSBT in this 50-GHz channel spacing configuration. Furthermore, Partial-DPSK is more impaired by chromatic dispersion than the “Electrical” PSBT format. So, the general conclusion is that “Electrical” PSBT is a credible solution, in addition to Partial-DPSK, for cost-effectively upgrading the legacy 10 Gb/s WDM transmission systems to carry 40 Gb/s WDM channels.
References and links
1. K. Yonenaga, S. Kuwano, S. Norimatsu, and S. Shibata, “Optical duobinary transmission system with no receiver sensitivity degradation,” Electron. Lett. 31(4), 302–304 (1995). [CrossRef]
2. D. Penninckx, M. Chbat, L. Pierre, and J. P. Thiery, “The phase-shaped binary transmission (PSBT): a new technique to transmit far beyond the chromatic dispersion limit,” IEEE Photon. Technol. Lett. 9(2), 259–261 (1997). [CrossRef]
3. X. Wei, X. Liu, S. Chandrasekhar, A. H. Gnauck, G. Raybon, J. Leuthold, and P. J. Winzer, “40 Gb/s duobinary and modified duobinary transmitter based on an optical delay interferometer,” in Proceedings ECOC 2002, Copenhaguen, 4, 1–2 (2002).
4. A. H. Gnauck, X. Liu, S. Chandrasekhar, and X. Wei, “Optical duobinary format from demodulation of DPSK using athermal delay interferometer,” IEEE Photon. Technol. Lett. 18(4), 637–639 (2006). [CrossRef]
5. D. Penninckx, H. Bissessur, P. Brindel, E. Gohin, and F. Bakhti, “Optical differential phase shift keying direct detection considered as a duobinary signal,” in Proceedings ECOC 2001, 3, 456–457 (2001).
6. P. Brindel, L. Pierre, G. Ducournau, O. Latry, M. Kétata, and O. Leclerc, “Optical generation of 43 Gbit/s phase-shaped binary transmission format from DPSK signal using 50 GHz periodic optical filter,” in Proceedings ECOC 2005, 4, 847–848 (2005).
7. H. Kim and C. X. Yu, “Optical duobinary transmission system featuring improved receiver sensitivity and reduced optical bandwidth,” IEEE Photon. Technol. Lett. 14(8), 1205–1207 (2002). [CrossRef]
8. I. Lyubomirsky and B. Pitchumani, “Impact of optical filtering on duobinary transmission,” IEEE Photon. Technol. Lett. 16(8), 1969–1971 (2004). [CrossRef]
9. A. D’Errico, R. Proietti, L. Giorgi, G. Contestabile, and E. Ciaramella, “WDM-DPSK detection by means of frequency-periodic Gaussian filtering,” Electron. Lett. 42(2), 112–113 (2006). [CrossRef]
10. A. Royset and D. R. Hjelme, “Novel dispersion tolerant optical duobinary transmitter using phase modulator and Bragg grating filter,” in Proceedings ECOC 1998, Amsterdam, 225–226 (1998).
11. G. Charlet, S. Lanne, L. Pierre, C. Simonneau, P. Tran, H. Mardoyan, P. Brindel, M. Gorlier, J. C. Antona, M. Molina, P. Sillard, J. Godin, W. Idler, and S. Bigo, “Cost-optimized 6.3Tbit/s-capacity terrestrial link over 17x100kmusing Phase-Shaped Binary Transmission in a conventional all-EDFA SMF-based system,” OFC 2003, PD25 (2003).
12. C. Gosset, L. Dupont, A. Tan, A. Bezard, and E. Pincemin, “Experimental performance comparison of duobinary formats for 40 Gb/s Long-Haul Transmission,” OFC' 2008, Paper JThA55 (2008).
13. A. Tan and E. Pincemin, “Performance comparison of duobinary formats for 40 Gb/s and mixed 10/40 Gb/s long-haul WDM transmission on SSMF and LEAF fibers,” J. Lightwave Technol. 27(4), 396–408 (2009). [CrossRef]
14. G. Bosco, A. Carena, V. Curri, R. Gaudino, and P. Poggiolini, “Quantum limit of direct receivers using duobinary transmission,” IEEE Photon. Technol. Lett. 15(1), 102–104 (2003). [CrossRef]
15. B. Mikkelsen, C. Rasmussen, P. Mamyshev, and F. Liu, “Partial DPSK with excellent filter tolerance and OSNR sensitivity,” IEEE Photon. Technol. Lett. 42, 1363–1364 (2006).
16. A. Lender, “The duobinary technique for high-speed data transmission,” IEEE Trans.Commun. Electron. 82, 214–218 (1963).
17. A. Sekey, “An analysis of the duobinary technique,” IEEE Trans. Commun. Technol. 14(2), 126–130 (1966). [CrossRef]