ECOC 2010, 19-23 September, 2010, Torino, Italy
Tu.3.A.4
Investigation on the Robustness of a Nyquist-WDM Terabit Superchannel to Transmitter and Receiver Non-Idealities G. Bosco(1) , , A. Carena(1) ,V. Curri(1) , P. Poggiolini(1) ,E. Torrengo(1) . F. Forghieri(2) (1) (2)
Dip. di Elettronica, Politecnico di Torino, C.so Duca Degli Abruzzi 24, Torino (Italy), B
[email protected] Cisco Photonics Italy srl, Via Philips 12, Monza (Italy), B
[email protected]
Abstract We analyze by simulation the performance of a Terabit superchannel composed of 10 PMQPSK sub-channels at 111 Gbit/s with tight channel spacing in the presence of system non-idealities both at the transmitter and at the receiver side. Introduction Increasing the spectral efficiency (SE) has become one of the main challenges in long-haul high-capacity coherent optical systems 1 . Recently, 96x112G PDM-RZ-QPSK channels with a SE of 4 bit/s/Hz have been successfully transmitted over more than 4,368 km of ultra-large area fiber 2 . This result was achieved by using an aggressive optical pre-filtering at the transmitter (Tx), which helped to reduce the crosstalk between adjacent channels. Very tight optical prefiltering was also employed in 3 . Preliminary results on the ultimate performance of this approach, called Nyquist WDM from bandlimiting the transmitted pulses while trying to satisfy the Nyquist criterion for no intersymbolinterference (ISI), indicate that, under ideal conditions, Nyquist WDM can achieve the optimum matched filter performance in ASE-noise limited systems 4 . In practice, penalties are to be expected when the ideal constraints on Nyquist WDM implementation are relaxed. In this work, we analyze both Tx and receiver (Rx) constraints, focusing in particular on: the use of a realistic Tx optical shaping filter; the bandwidth of the Rx; the finite resolution of the ADC; the limited number of samples per symbol available at the Rx. We performed the analysis for both back-to-back (btb) and long-haul transmission, including non-linear effects. Test Set-Ups Description The test system layout is shown in Fig. 1. Each 111 Gbit/s PM-QPSK Tx is based on two QPSK integrated modulators whose outputs are multiplexed through a polarization beam splitter. Modulation is non-return-to-zero (NRZ). Before wavelength multiplexing, each channel is shaped by an optical filter whose function is to limit the crosstalk between the channels. The coherent Rx front-end includes a local oscillator (LO) that is mixed to the incoming signal in two 90◦ hybrids. LO frequency matching is assumed ideal, while the linewidth of both Tx and
978-1-4244-8535-2/10/$26.00 ©2010 IEEE
Tx 1
NRZ-PM-QPSK 111 Gb/s Shaping Tx 5 filter
Tx 10
SpS, BRx NADC
Shaping filter
Shaping filter
BPD
90° hybrid
ADC
BPD
ADC
BPD
ADC
BPD
ADC
D S P
LO
Fig. 1: Layout of the analyzed system.
Rx lasers is equal to 100 kHz. Four balanced photo-detectors are used to detect the received signal components. The photo-detected signals are filtered by 5-pole Bessel filters with bandwidth BRx which are meant to account for all the bandwidth limitations of the Rx electronics. Then the signals are sampled by four ADCs, with variable numbers of resolution bits and samples per symbol (SpS). A digital signal-processing (DSP) section follows, consisting of a first stage which performs chromatic dispersion (CD) compensation. A second stage follows, with four complex 15tap finite-impulse-response (FIR) filters in ‘butterfly’ configuration 6 , which recovers the polarization frame and performs post-equalization. When the number of SpS is lower than 2, a digital interpolator is inserted before the second stage equalizer in order to resample the signal up to 2 SpS. The spacing between filter taps is then equal to T /2 for all sampling rates, where T = 1/Rs is the symbol time and Rs is the baud-rate. The second stage is driven by the constant modulus algorithm (CMA) 6 , which is active throughout the simulation. A Viterbi-Viterbi stage follows for phase recovery. All simulations were carried out transmitting 10 adjacent channels. The number of simulated symbols was 216 , i.e., 218 =262144 bits. Optical pre-filter shape In order to use a sub-channel spacing equal to the baud-rate without either crosstalk or ISI penalty, the optical filter transfer function should be such that the signal spectrum at its output had a rect-
Tu.3.A.4
Fig. 2: OSNR at BER=4 · 10−3 vs. normalized channel spacing using a 2nd order SG optical filter in btb. The Rx bandwidth is BRx = (0.4 Rs ). Inset: Nyquist filter, and 2-nd order SG filter with bandwidth 0.85 Rs .
In Fig. 2 the Rx bandwidth BRx was fixed at 0.4 · Rs . In Fig. 3 we plot the system btb sensitivity as a function of BRx for different optical Tx filter shapes and subchannel spacing. One striking peculiarity of Nyquist-WDM is its good performance even at very low values of BRx . This is allowed by the DSP second-stage which effectively post-equalizes the signal by enhancing
the high frequencies attenuated by the Rx. Below 0.2 · Rs , however, the CMA algorithm fails to converge, presumably because the second-stage FIR transfer function needed to restore the signal becomes too critical. Even using the realistic SG filter, performance is flat or little impaired at least down to BRx = 0.3 · Rs . This aspect is likely to be of importance for practical application. Note that the penalty increase for large Rx bandwidth is due to aliasing, that is 2 SpS are not enough to avoid spectral folding. 18 SG filter (∆ f/Rs=1.1) 17
OSNR [dB]
angular shape with bandwidth Rs 5 . This shape is not physically realizable and even simulating it is problematic. To establish a benchmark, we used instead the Tx filter shown in the inset of Fig. 2 (dashed line). With it, the transmitted pulses have a raised-cosine spectrum with a steep rolloff=0.1 . This filter is still ISI-free 5 (i.e., it is ‘Nyquist’) but it causes some inter-subchannel crosstalk: at a subchannel spacing equal to the baud-rate, the system back-to-back (btb) sensitivity for a pre-FEC BER of 4·10−3 is OSNR=12.8 dB (0.1 nm), about 0.8 dB penalty from ideal (which is 12 dB). We then used a Tx optical filter with a 2nd order Supergaussian (SG) profile, typical of many AWGs and interleavers. Fig. 2 shows a plot of sensitivity as a function of the normalized subchannel spacing ∆f /Rs when using the SG filter with bandwidth 0.85 ∆f (optimum value found through simulations). Since the sides of this realistic filter are less steep than the benchmark Nyquist filter (see inset in Fig. 2), substantial crosstalk occurs at baud-rate spacing. ISI is present as well. As a result, SG at baud-rate spacing has a penalty of about 2 dB with respect to the Nyquist filter. Penalties can be decreased by increasing the subchannel spacing and in fact the same btb sensitivity as with the benchmark Nyquist filter is obtained at about ∆f = 1.1 · Rs .
16
SG filter (∆ f/Rs=1) Nyquist filter (∆ f/Rs=1)
15 14 13 12 0.2
0.4
0.6 BRx/Rs
0.8
1
Fig. 3: OSNR at BER=4 · 10−3 vs. normalized Rx electrical bandwidth. Comparison between ideal and realistic optical filter in btb. Two samples per symbol (SpS).
ADC resolution and speed The results shown in this section were obtained using a 2nd order SG filter with bandwidth 0.85·∆f and subchannel spacing 1.1·Rs . Fig. 4 shows the robustness of such Terabit superchannel to the finite resolution of the ADC. The quantizer is assumed to be uniform with 1% of overload. The penalty in using a finite number of resolution bits is almost negligible down to 4 bits, but even at 3 bits the penalty is limited (lower than 1 dB). However, sampling penalties are noticeably exacerbated at low Rx bandwidths (< 0.3 Rs ) because sampling noise is amplified by the DSP equalizer trying to enhance the attenuated highfrequency signal components. Fig. 5 shows the analysis of the robustness of the system to the limited speed of the ADC, which translates in a limited number of samples per symbol available in the DSP module. In all cases the resolution of ADC is assumed to be 5 bits. The OSNR penalty with respect to the 2 SpS case is equal to 0.4 dB and 1.3 dB when using 1.5 and 1.2 SpS, respectively. Note that the minima observed in Fig. 5 are due to aliasing at the ADCs: when the analog Rx bandwidth grows, a larger number of SpS is needed to avoid spectral folding due to sampling.
Tu.3.A.4
17
1.2 SpS, respectively. All in all, non-linear effects do not appear to alter the btb result hierarchy.
OSNR [dB]
16 15
NADC=3,4,5,∞
2 SpS
14 1.5 SpS
13
12 0.2
0.4
0.6 BRx/Rs
0.8
1
Fig. 4: OSNR at BER=4 · 10−3 vs. normalized Rx electrical bandwidth in btb for different numbers of the ADC resolution bits and using 2 SpS. 17
1.2 SpS
16
Nyquist Filter (2 SpS, ∆f=Rs)
1.2 SpS
BRx/Rs
Fig. 6: Maximum reach vs. Rx bandwidth limitation: comparison between Nyquist filter at Rs spacing (dashed line) and SG filter at 1.1 Rs spacing (solid lines) for different values of samples per symbol (SpS).
OSNR [dB]
1.5 SpS 15
2 SpS 14 13
12 0.2
0.4
0.6 BRx/Rs
0.8
1
Fig. 5: OSNR at BER=4 · 10−3 vs. normalized Rx electrical bandwidth in btb for different values of samples per symbol (SpS) and 5 bits ADC resolution.
The curves with higher SpS extend further to the right because they can avoid aliasing for larger values of BRx . Non-linear propagation We simulated a link composed of N spans, each including a fiber stretch with Lspan =90 km and an EDFA with a noise-figure F =5 dB whose gain completely recovers the span loss. The fiber was SMF with loss α = 0.22 dB/km, dispersion D = 16.7 ps/nm/km and non-linearity coefficient γ = 1.3 (W · km)−1 . The maximum reach results at BER=4 · 10−3 are shown in Fig. 6, vs. the normalized Rx bandwidth BRx /Rs . The launch power was optimized for each value of BRx . The plot shows that the same maximum distance of 2300 km can be achieved with both the benchmark Nyquist filter at ∆f = Rs and SG filter at ∆f = 1.1 · Rs when using 2 SpS and 5 bits of ADC resolution. The optimum BRx is about 0.40.5·Rs in both cases. The maximum reach decreases to 2100 and 1800 km when using 1.5 and
Comments and Conclusion We have analyzed the impact of some key Tx and Rx non idealities on the performance of a Terabit superchannel system, composed of 10x111 Gbit/s PM-QPSK subchannels, based on (or approaching) the ‘Nyquist’ WDM condition. Even though optical ‘Nyquist’ filters may be difficult to make, we showed that using conventional AWG or interleaver profiles already allows to reach near-Baud-rate (1.1 x Baud-rate) subchannel spacing with small penalties. As far as the impact of the ADC finite resolution and sampling speed is concerned, the Nyquist WDM Terabit superchannel turned out to be quite robust. Finally, we found that such systems tolerate electrical Rx -3dB bandwidths as low as 0.3 times the Baud-rate, a circumstance that may have significant practical impact. These results suggests that ‘Nyquist’ WDM Terabit superchannels could be a promising technology for future high-spectral efficiency Tb/s per channel systems. This work was supported by CISCO Systems within a SRA contract. The simulator OptSim was supplied by RSoft Inc.
References 1 A.H.Gnauck et al., Proc. OFC 2009, PDPB8. 2 J.-X. Cai, et al., Proc. OFC’10, PDPB10 (2010) 3 G. Gavioli, et al., Proc. OFC’10, OthD3 (2010) 4 G. Bosco et al., “Performance Limits of Nyquist-WDM and CO-OFDM in High-Speed PM-QPSK Systems, submitted for publication in Phot. Technol. Lett. 5 S. Benedetto and E. Biglieri Principles of digital transmission: with wireless applications, Kluwer Academic Publ. (1999) 6 S. Savory, Opt. Express, 16, 804 (2008)
