1. Introduction

The phase noise characteristics of widely tunable lasers are of great interest when considering such devices for implementation in burst switched coherent optical networks that employ higher order phase sensitive modulation formats [1]. In such systems, the linewidth of both the transmitter and receiver local oscillator lasers impact on the bit error rate performance of the system and in particular may cause cycle slips in the carrier and phase recovery. Cycle slips occur when a π/2 phase rotation (for QPSK) is missed by the carrier and phase recovery algorithm, after which the receiver will begin to decode the received symbols in the incorrect quadrants, causing catastrophic errors and subsequent network outage until the receiver is reset. The constraint on laser linewidth to achieve a tolerable cycle slip probability (10⁻¹⁸) is approximately two orders of magnitude lower than that required to obtain a low Q-factor penalty [2]. If the cycle slip probability is too high, differential decoding and data precoding will be required. Therefore it is imperative to accurately characterize the stabilization time, in terms of the instantaneous frequency and linewidth, of widely tunable lasers after a wavelength switching event has occurred, in order to avoid bit errors due to laser phase noise and catastrophic bit errors due to cycle slips.

Typically for optical burst or packet switched schemes, the switching time of the fast tuning transmitter or receiver is characterized and in order to negate the transmission performance degradation associated with a switching event, the output of the device is blanked during the stabilization period. Such systems have traditionally employed the on-off keyed (OOK) modulation format and therefore only the laser switching time and frequency stability have been extensively characterized [3]. However, as previously discussed, the stabilization time of the laser linewidth is also important when considering coherent optical networks that utilize phase sensitive modulation formats. The static linewidth of widely tunable lasers has been previously characterized using the delayed self heterodyne technique [4] and a dynamic characterization has been reported using an optical heterodyne technique with a measurement resolution of 5MHz [5]. Coherent heterodyne measurements of laser linewidth and phase noise characteristics have also been reported, however both of these schemes were performed on static low linewidth lasers [6, 7].

Conversely, in this paper we perform dynamic linewidth measurements on a fast switching digital supermode distributed Bragg reflector (DSDBR) tunable laser and characterize the linewidth of the laser in a 2 and 5-channel WDM test bed. From this characterization an accurate assessment of the laser stabilization time, in terms of lasing frequency and linewidth, can be determined. In addition to this the individual noise sources (relative intensity, phase and 1/f noise) which compose the linewidth of a digital supermode distributed Bragg reflector tunable laser is also investigated. The linewidth estimation technique is validated using a linewidth emulator and a low linewidth external cavity laser.

2. Linewidth measurement technique

The digital intradyne coherent receiver setup for verifying the laser linewidth measurement technique is illustrated in Fig. 1(a). An external cavity laser with a 10kHz linewidth, operating at a wavelength of 1544nm, was split into two paths using a 3dB fibre coupler. One arm was passed directly into the local oscillator (LO) port of a digital coherent receiver, while the second arm was passed through a linewidth emulator stage. The linewidth emulator consisted of a nested Mach-Zehnder (MZM) IQ modulator and an arbitrary waveform generator (AWG) as outlined in [8]. Theoretical phase noise was digitally generated as a Wiener process through Eq. (1):

 

Fig. 1 (a) Linewidth characterization experimental setup and (b) power spectral density of coherently received signal complex field with a Lorentzian fit.

Download Full Size | PDF

A standard technique for measuring the laser linewidth is to fit a Lorentzian curve to the power spectral density of the received complex field. This method is illustrated Fig. 1(b), which shows the spectrum of the external cavity laser when the linewidth emulator was set at 2MHz. The full width half maximum linewidth, measured at the 20dB point ( $\Delta v_{FWHM}=\Delta v_{20dB}/\sqrt{99}$), was 2.13MHz. This technique will suffice for very stable sources where long FFT lengths can be used to calculate the laser spectrum, however it is not an adequate solution when the laser source exhibits low frequency drift or when it is switching. If a laser switching event occurs the intermediate frequency (IF) of the beat note between the signal under test and the LO will change rapidly, therefore very low FFT lengths are required to track this fast frequency variation, thus diminishing the resolution of the measurement. Additionally, if the absolute frequency of the laser exhibits a very slow variation, the FFT will compute the average of the laser phase noise and this slow frequency variation over the measurement time window. This will result in a broader linewidth and therefore the Lorentzian fit will over estimate the true value of the laser phase noise. To overcome this we have employed a time domain frequency estimator which calculates the instantaneous frequency of the beat note for each sample point k, through Eq. (2), where x_in is the complex field of the coherently received signal [9].

The linewidth is subsequently calculated by obtaining the variance of the instantaneous frequency through Eq. (3), where μ is the mean instantaneous frequency. When the laser is operating in static mode, the variance is calculated over the entire measurement time window, however for the dynamic linewidth characterization the variance is calculated over a specified bus width (N_win), which is required for a real time application that utilizes a parallelized digital signal processing implementation [1].

3. Static linewidth characterization

3.1. Low linewidth ECL and phase noise emulator

To initially verify the performance of our linewidth estimation technique, the phase modulation applied to the linewidth emulator was varied from 100kHz to 100MHz and the linewidth of the received signal was calculated using both a Lorentzian fit to the power spectral density and also from the variance of the instantaneous frequency calculated through Eq. (2) and Eq. (3), where the time window was 50μs. Figure 2(a) illustrates the measured linewidth as a function of the theoretical value. The continuous blue line indicates the digitally generated linewidth used in the emulator. For the time domain frequency estimator, the received complex field was resampled to reduce the measurement bandwidth to 4GHz to ensure that the linewidth was calculated over a bandwidth that only contained the beat note frequencies, thus limiting the impact of out of band white noise originating in the receiver. From Fig. 2(a) it is evident that our proposed linewidth measurement technique accurately estimates the theoretically specified linewidth and also agrees with the values obtained from the Lorentzian fits of each laser spectrum.

 

Fig. 2 (a) Measured linewidth as a function of the theoretical value and (b) measured linewidth as a function of the measurement bandwidth.

Download Full Size | PDF

In order to investigate the impact of the measurement bandwidth on the time domain estimator, the linewidth was calculated as a function of the resample rate. Therefore the measurement bandwidth was varied by resampling and digitally filtering (5th order Kaiser filter) the captured signal over a range of sample rates (2–50GS/s). Figure 2(b) illustrates the calculated linewidth as a function of the measurement bandwidth when the emulator was set to a linewidth of 2MHz, which is indicated by the flat line. As the linewidth emulator applied the phase noise to the ECL on a 3GHz carrier, harmonics of this carrier are also present on the received signal. Therefore if the resample rate is too high, these harmonics combined with the out of band white noise originating in the receiver contribute to the estimated phase noise. Conversely, at lower sample rates the signal is filtered too tightly, resulting in the laser phase noise being under estimated. In addition to this, it is also imperative to utilize the full range of the analog to digital (ADC) convertor in the the coherent receiver, otherwise excess quantization noise will be added to the received signal, which will also subsequently increase the estimated laser phase noise. Here we see that the optimal measurement bandwidth for a laser source with a linewidth of 2MHz is approximately 4GHz.

As previously mentioned, discrepancies between the Lorentzian fit and the linewidth measured using the time domain frequency estimator may arise if there is a slow variation in the laser’s absolute frequency. Therefore it is also important to consider the sources of noise that contribute to the phase variation. These can be identified by examining the complex field optical spectrum, the FM noise spectrum and the amplitude noise spectrum, as these provide an indication of each contribution to the overall composition of the laser linewidth [7]. Figure 3(a) illustrates the single sided power spectral densities of the received complex field and the amplitude noise for the ECL when the linewidth emulator was set at 2MHz. The slope of the complex field decreases at a rate of 20dB per decade, which is expected for a Wiener phase noise process and results in a Lorentzian spectral profile, where the noise is proportional to 1/f². The PSD of the signal amplitude is equivalent to the relative intensity noise (RIN) of the laser, where the measurement noise floor is approximately −80dB.

 

Fig. 3 (a) Power spectral density of the complex field and amplitude noise and (b) FM noise spectrum.

Download Full Size | PDF

Figure 3(b) illustrates a flat FM noise spectrum and estimated linewidth (flat red line) which is expected for a stable laser source, where the amplitude is proportional to the linewidth. The FM noise spectrum illustrates the single sided linewidth, therefore an amplitude of 1MHz is visible from Fig. 3(b). However if the laser under test exhibits slow frequency variations, the FM noise spectrum will deviate from a flat profile and the PSD of the complex field will broaden, which result in an over estimated Lorentzian fit. The influence of slow frequency drift on the linewidth measurement scheme is examined in the proceeding section using a commercially available DSDBR tunable laser.

3.2. DSDBR tunable laser

The tunable laser was a temperature controlled device and consisted of eight front mirror sections, gain, phase, rear and SOA sections, as detailed in [10]. Front mirror sections 3 and 4 were biased at 1.7mA and 3.1mA respectfully using ultra low noise laser current drivers. The gain, SOA, rear and phase sections were biased at 110mA, 80mA, 14.3mA and 0.6mA respectfully, resulting in a lasing wavelength of 1558.65nm and a side mode suppression ratio (SMSR) greater than 50dB. The experimental setup used to measure the phase noise characteristics of the tunable laser is shown in Fig. 4(a). The output of the tunable laser was passed into the signal port of the coherent receiver and a low linewidth (100kHz) tunable external cavity laser was used as the local oscillator.

 

Fig. 4 (a) Static linewidth characterization experimental setup and (b) power spectral density of the received complex field and the corresponding Lorentzian fit.

Download Full Size | PDF

The received beat note between the LO and the signal from the tunable laser was digitally down converted to baseband, down sampled and filtered to a measurement bandwidth of 4GHz. Figure 4(b) illustrates the power spectral density of the received complex field and the corresponding Lorentzian fit. The full width half maximum of the Lorentzian fit, measured at the 20dB point was 8.8MHz. The value estimated using the time domain frequency estimator was 2.1MHz, which is significantly lower than the value obtained using the conventional Lorentzian fit. The reason for this can be intuitively understood by considering the FM noise spectra of the DSDBR tunable laser, as depicted in Fig. 5. The instantaneous frequency is approximately flat from 100kHz out to the resample frequency (single sided: 2GHz), with the spikes corresponding to the multiple suppressed sub-peaks visible in the PSD of the received signal complex field (Fig. 4b). However, the DSDBR laser exhibits considerable 1/f noise power at frequencies below 100kHz, which is in contrast to the stable ECL used in the phase noise emulator. The low frequency noise arises from 1/f carrier noise [11] and injection recombination shot noise (IRSN) in the passive tuning regions of the laser [12]. The flat portion of the FM noise spectrum beyond 100kHz gives rise to a double sided linewidth of 2.2MHz, which is consistent with the 2.1MHz obtained from the time domain frequency estimation technique.

 

Fig. 5 FM noise power spectral density.

Download Full Size | PDF

In high bit-rate coherent communication systems the white FM noise is more detrimental to system performance. Therefore the measure of linewidth using the Lorentzian fit of the complex field PSD is not a good measure of phase noise in widely tunable lasers, which employ the free carrier plasma effect to tune the emission frequency, as it may be dominated by 1/f noise. Therefore the white FM noise component, which is measured using the time domain frequency estimation technique described in section 2, is important for evaluating the performance widely tuneable lasers in high speed coherent communications systems.

4. Dynamic linewidth characterization

4.1. Two-channel switching event

In order to analyze the dynamic performance of the linewidth estimation technique the fast switching DSDBR tunable laser was implemented in the experimental setup as shown Fig. 6(a). The tunable laser was switched between two wavelength channels by applying a 5kHz square wave with a 3.2ns rise time to one of the front mirror sections. The gain and SOA sections were biased at 110mA and 80mA respectfully, while the rear and phase sections remained unbiased. Switching the front mirror section resulted in a supermode wavelength switch of approximately 6.3nm, which altered between 1544.1nm (channel 1) and 1550.4nm (channel 2), with a side mode suppression ratio (SMSR) greater that 50dB for both channels. Utilizing the front mirror sections and un-biasing the phase and rear sections minimized the contribution to the phase and 1/f noise. This provided an optimum initial linewidth for the tunable laser to experimentally verify the performance of the dynamic linewidth estimation technique, however a more practical switching implementation of the DSDBR laser is investigated in section 4.2. Two low linewidth (∼100kHz) external cavity lasers (ECL) were tuned to match the switching wavelengths exhibited by the DSDBR laser and were combined using a 3dB fibre coupler to provide the LO for the coherent receiver. As both external cavity lasers were running in continuous wave mode, the instantaneous frequency variation and linewidth of the fast tuning laser could be tracked throughout the switching transition.

 

Fig. 6 (a) Dynamic linewidth characterization experimental setup and (b) percentage variation of the simulated 2MHz linewidth as a function of the bus width.

Download Full Size | PDF

For real time performance monitoring of the tunable laser stabilization time, a parallelized DSP implementation would have to be used. Therefore the linewidth and frequency estimation is calculated for a given bus width [1]. A small bus width should be used to ensure the greatest temporal resolution, however as the number of samples used to calculate the variance of the instantaneous frequency (Eq. (3)) is reduced, the error associated with the measurement increases. Figure 6(b) illustrates the estimated linewidth as a function of bus width for a 2MHz simulated phase noise, which was generated using Eq. (1). It is evident that as the bus width is reduced the variation in the estimated linewidth increases. A typical bus width of 128 samples was used in our dynamic linewidth characterization setup, which corresponded to a temporal resolution of 25.6ns (128 samples at 5GS/s). By using a bus which is 128 samples wide the variation of the linewidth estimates were approximately ±10%. The accuracy of the linewidth estimate could be improved by using a larger timing window, as indicated by Fig. 6(b), however this would come at the expense of the temporal resolution of the measurement. To overcome this limitation, a number of traces could be recorded from the real time oscilloscope sequentially. A smaller bus width could then be taken from each scope trace at the same instant in time and an average of all the samples could be computed. The bus would then move to the next number of samples of each scope trace and the process is repeated. While this will provide greater granularity, the measurement will be a time average of all the received scope traces and is therefore not suitable for a real time application.

Figure 7(a) illustrates the instantaneous frequency and time resolved linewidth of the DS-DBR tunable laser as it switches between channel 1 and channel 2. The intermediate frequency of the beat note for each channel was different (CH1:∼−1.7GHz and CH2:∼1.8GHz), therefore each burst was digitally down converted to baseband before the entire signal was down sampled and filtered. From this dynamic characterization an accurate representation of the laser switching dynamics in terms of frequency stabilization and linewidth can be determined. Each laser channel had a dwell time of 100μs which was determined by the frequency (5kHz) of the switching voltage applied to the laser mirror section. Therefore after 100μs elapsed the laser switched from channel 1 to channel 2 and a sharp change in the instantaneous frequency was experienced, with the maximum variation that can be measured limited by the resample rate (5GS/s) to ±2.5GHz. Figure 7(b) illustrates the switch from channel 1 to channel 2, displayed over a smaller time period. From this figure it is evident that the tunable laser switched from channel 1 to within 2.5GHz of the stabilized channel 2 frequency in ∼70ns. During the switching event the laser linewidth was initially dominated by the rapid change in the intermediate frequency but settled to a value of 1 to 2MHz within 70ns. A similar stabilization time of ∼70ns was experienced as the laser switched back from channel 2 to channel 1, however channel 2 exhibited a slightly lower linewidth which was consistently of the order of 1–1.5MHz which could be attributed to slightly different bias positions for each channel.

 

Fig. 7 (a) Instantaneous frequency variation and dynamic linewidth as the tunable laser switched from channel 1 to channel 2 and (b) single switch from channel 1 to channel 2.

Download Full Size | PDF

4.2. Wavelength switching on a five-channel 50GHz grid

To further assess the practicalities of employing a DSDBR laser as a local oscillator in a burst switched coherent receiver, it was implemented in a 5-channel 50GHz spaced WDM system, illustrated in Fig. 8(a). Five commercially available distributed feedback lasers (DFB) with linewidths of 500kHz at a bias current of 110mA were temperature tuned to reside on a 50GHz grid from 1551.86nm to 1553.46nm. The DSDBR tunable laser was dynamically switched between the 5-channels by applying a stepped voltage signal to the rear section of the device using an arbitrary waveform generator. Each switching voltage level had a duration of 8μs. In addition to this, constant currents were applied to front mirror section 3 (5mA), mirror section 4 (0.7mA), the gain (150mA) and SOA (100mA) sections, using ultra low noise current sources. The lasers phase section remained unbiased to limit the contribution to the low frequency 1/f noise.

 

Fig. 8 (a) Dynamic linewidth characterization experimental setup and (b) instantaneous frequency variation and dynamic linewidth as the tunable laser switched sequentially from channel 1 to channel 5.

Download Full Size | PDF

The received signal contained the beat note of all five DFB channels, each at a slightly different intermediate frequency. Therefore each 8μs burst was digitally down converted to baseband before the entire signal was down sampled and filtered. Figure 8(b) illustrates the linewidth and instantaneous frequency as a function of time for the DSDBR laser for all five bursts. The first burst corresponds to the switch from channel 5 to channel 1, after which the laser switched sequentially from one channel to the next. The measured linewidth also contains a contribution from the DFB lasers as both complex fields are convolved in the receiver. Therefore this measurement provides the overall convolved linewidth between the transmitter and receiver lasers, which is an important parameter to consider for coherent optical networks that utilize higher order modulation formats and digital coherent reception. The modulation format order and baud rate, to be employed in the coherent optical network, may be dictated by the overall convolved linewidth between the transmitter and receiver laser sources.

From Fig. 8(b), it is evident that the linewidth of the DSDBR laser settles down within approximately 50ns for each of the 5 channels. The instantaneous frequency settles to within 0.5GHz of the channel 1 frequency in 50ns, which represents the largest wavelength switch in the 5-channel system. The instantaneous frequency for the remaining 4-channels returns to below 100MHz of their respective channel frequencies within the same time period. The steady-state convolved linewidth always remains below ∼3MHz after a switching event, with the best performing channel exhibiting a linewidth of approximately 1MHz. The variation in linewidth between each of the bursts stems from the different operating points for the DSDBR laser, which were optimized to match the comb wavelengths and not for best linewidth performance, which would be commensurate with a practical implementation of a burst mode transceiver. This linewidth variation is an inherent property of tunable DBR laser structures, where the lowest linewidth performance depends on the fine optimization of the control currents applied to each laser tuning section.

5. Conclusion

We have demonstrated a dynamic linewidth measurement technique using a digital intradyne coherent receiver and shown that the temporally resolved linewidth estimation gives results that are comparable to those obtained in the static case. The switching dynamics, in terms of switching time, frequency stability and linewidth of a DSDBR tunable laser were characterized. In addition to this the separate noise sources which contribute to the overall composition of laser linewidth were also investigated and it was shown that the time domain linewidth estimator detailed in this work is an optimum technique to measure the white noise component of the laser phase noise. The stabilization time of a commercially available DSDBR laser was investigated by implementing it in a coherent burst mode receiver in a 5-channel WDM test-bed. It was shown that the DSDBR tunable laser could switch to within 500MHz of the desired channel frequency with a minimum linewidth in 50ns.