All-optical 160 Gbit/s RZ data retiming system incorporating a pulse shaping fibre Bragg grating

We characterize a 160Gbit/s retimer based on flat-topped pulses shaped using a superstructured fibre Bragg grating. The benefits of using shaped rather than conventional pulse forms in terms of timing jitter reduction are confirmed by bit-error-rate measurements. Introduction As the data rates increase, and the associated pulse durations get ever shorter, managing noise-related transmission impairments (e.g. induced amplitude and timing jitter) becomes ever more critical, and optical processing of the signal increasingly desirable, if not essential. The retiming function in a regenerator, and in more general terms the synchronization between a locally generated signal used to control an optical switch, and the data signal are key requirements for high speed systems, since timing jitter is often converted into amplitude noise at the output of analogue switching systems. The key to reducing timing jitter is to establish a rectangular temporal switching window [1, 2]. This provides optimal resilience to timing jitter-induced errors, and also reduces the absolute accuracy for temporal bit alignment. In the technique that we present in this paper, we achieve timing jitter reduction in transmission systems operating at data rates up to 160 Gbit/s by linearly reshaping RZ data pulses into longer rectangular (flat-top) pulses and then switching these pulses in a nonlinear Kerr switch with a synchronous optical clock signal [3, 4]. The required pulse shaping is performed using Superstructured Fibre Bragg Grating (SSFBG) technology, which allows the implementation of optical filters with accurately controlled frequency and phase responses of almost arbitrary complexity in a single continuous grating structure [5]. SSFBGs have proven to be an extremely powerful tool for advanced photonic applications that require accurate manipulation of the exact shape of short optical pulses. Shaping of a pulse with a pre-known spectrum can be achieved by feeding it into a SSFBG which has been designed so that the reflected spectrum has the amplitude and phase characteristics of the desired waveform. Using this linear technique, we have previously demonstrated the reshaping of short pulses of a few picoseconds duration into rectangular pulses of different pulse widths (10-20ps) [4-6]. However, this extremely versatile technique is suitable for other shaping applications as well. For example, we have demonstrated the generation of transform-limited parabolic pulses [7], and by using the same principles we have achieved sophisticated phase and/or amplitude time-domain encoding of short pulses, suitable for applications in optical codedivision multiple access systems [8]. The experiment that we present in this paper is the first where rectangular pulses as short as ~5ps are generated using this technology. It should be noted that even shorter rectangular switching windows have been achieved using for example, propagation in a section of polarisation maintaining fibre [3], or a long period grating [9]. However, unlike these techniques, SSFBG pulse shaping is polarisation insensitive, making it a very robust and practical scheme. We show detailed characterisations of the SSFBGbased retimer in terms of timing jitter measurements and full bit error rate (BER) results. The retimer is successfully applied to both 40 and 160 Gbit/s data signals, obtaining clear improvement and error free performance at both bit rates. We also present comparative data using Gaussian pulse shapes to control the switch, thereby highlighting the improvement obtained with the pulse shaping grating. Principle of operation and experimental set-up Figure 1: Set-up of the retiming system and its operation principle. The experimental set-up and the operating principle of the retiming technique are shown in Fig.1. A wavelength tuneable 10 GHz semiconductor modelocked laser (TMLL) is used as the data source to generate ~1.8 ps Gaussian pulses. The operating wavelength of the laser is ~1557 nm and the laser Figure 2: a) Signal spectra both before and after the SSFBG. b) Temporal profile of flat top (solid line) and Gaussian shapes (dashed line) with same FWHM (~5ps). source has an inherent root mean square (RMS) timing jitter of τrms ~420 fs. These pulses are modulated to provide a 2-1 pseudorandom bit sequence (PRBS) and subsequently multiplexed up to either 40 or 160 Gbit/s. The data is then fed to the pulse shaping SSFBG, which converts the 1.8 ps Gaussian pulses into ~5 ps rectangular pulses. We note that the SSFBG has been designed to work optimally with ~1 ps pulses at its input. Fig.2 a) compares the spectra both before and after the filtering where the sinc-function features of the rectangular pulses can be clearly observed, and shows that the SSFBG response is broader than the input spectrum. (This data is taken at a pulse repetition rate of 10 GHz). Note that the spectral asymmetry observed results from asymmetry in the SSFBG’s spectral response. Fig.2 b) shows the corresponding temporal profile of the pulse shaping grating, measured with a cross-correlator with a ~600 fs sampling pulse. As a consequence of the relatively broad input pulses, the shaped pulses have smoother edges and a narrower flat top than optimal. Nevertheless, the flat top extends over ~2 ps which is sufficient to eliminate most of the 420 fs RMS timing jitter of the incident data pulses. We include a plot of a Gaussian pulse with similar full width at half maximum (FWHM) in Fig.2 b) as a reference. The shaped data pulses are then used as input to a Kerr switch based on 200 m of highly nonlinear fibre (HNLF) (dispersion slope 0.018 ps/nmkm, λ0=1554 nm, γ ~ 10.5 Wkm). The clock source is an Erbium-glass mode-locked laser (ERGO) with low inherent timing jitter (~210 fs RMS), operating at 1544 nm and providing pulses with a FWHM of 1.3 ps when optimally compressed in a short length of external fibre. These pulses are synchronised to the data stream and are multiplexed in an additional fibre-based multiplexer before being amplified and filtered for use as the control signal to the Kerr switch. The data signal is aligned at 90° to the Kerr switch polariser so as to be attenuated in the Kerr switch. The clock pulses co-propagating in the Kerr switch are aligned at 45° relative to the shaped data pulses. The polarisation of that part of the square pulse that overlaps with the clock pulse has its polarisation rotated within the HNLF due to cross-phase modulation and can thus be transmitted through the polariser. Since the pulse width of the control pulses is comparable to that of the data pulses, and the walk-off between the two signals inside the switch is negligible, the data pulses retain the control pulse width after retiming (FWHM~1.2ps), whilst maintaining their original wavelength. Indeed Fig.3 a) and Fig.3 b) show the spectra at the output of the retiming scheme for operation at 40 and 160Gbit/s respectively, highlighting that the spectra get broader with respect to the initial pulse spectrum (see Fig.2a) but are still centred on the original wavelength. The signal-to-noise ratios are more than 30dB in both cases. Figure 3: Spectra of the retimed data (after the Kerr switch) at 40Gbit/s (a) and 160Gbit/s. Pulse Retiming: experimental results First we study the retiming properties of the system at 40 Gbit/s. At this data rate we are able to measure the RMS timing jitter on the trailing edge of the pulse using an Agilent Precision Timebase Module. The results in Tab.1 clearly show that the switched pulses have their jitter drastically reduced from ~420 fs to ~250 fs (close to the clock pulse jitter). A slight increase in the amplitude noise can be noted for the pulses at the output of the switch. Indeed in Tab.1 the amplitude noise standard deviation values for the marks normalized to their corresponding mean values are plotted at different points of the system.