Wavelength tunable coherent burst-mode receiver design under transient frequency offset

Bo Xu; Kun Qiu

doi:10.1364/OE.21.019498

1. Introduction

High-speed optical fiber communication systems use polarization-diverse optical coherent receiver with multilevel modulation and polarization multiplexing to achieve 100-Gb/s transmission rate and above per channel for high total capacity and spectral efficiency [1,2]. Different digital signal processing (DSP) functions are included in the coherent receiver for performance improvement plus other benefits like flexibility and low cost [3,4]. With a wavelength tunable laser at the receiver, a selected wavelength channel can be received by tuning the laser wavelength. This wavelength selectivity is desired in the realization of dynamic optical networks.

In order to make better usage of the wavelength resources and increase the network flexibility for wavelength division multiplexing (WDM) networks, sub-wavelength provisioning is a good choice with data transmission in optical packets or bursts. Optical burst-mode receivers (BMRs) are necessary for data reception under this situation [5,6]. Moreover, wavelength tunable BMRs are necessary to receive optical bursts carried by different wavelengths. A fast tuning coherent BMR was firstly demonstrated by Simsarian et al where the studied BMR allows for the reception on one of the 16 wavelength channels and the BMR could be initialized within 200 ns [7]. BMR designs with parallel DSP implementation have also been studied later in [8–10]. Non-data-aided frequency offset estimation (FOE) with Viterbi-Viterbi algorithm is used together with a constant modulus algorithm (CMA)-based MIMO equalizer for polarization demultiplexing.

However, fiber dispersion is assumed to be known and compensated before frequency offset estimation and polarization demultiplexing in the previous works. Different from the above works, the amount of accumulated fiber dispersion for each burst is assumed to be different and unknown in this work. A different optical coherent BMR design with least-mean-square (LMS)-based equalizer and FOE with training sequence is proposed as an effective way for simultaneous dispersion compensation and frequency offset estimation. Training sequence is used both in the LMS-based equalizer and FOE for faster convergence speed than BMR design with individual CMA-based equalizer and non-data-aided FOE. The interaction between dispersion equalization module and FOE module is studied for the first time in this work through simulations. The new BMR design can also work under the case when non-negligible transient frequency offset from wavelength tuning exists in the system.

Section II discusses the structure of the designed optical coherent BMRs. Simulations are used to study the performance of such BMRs in Section III. Section IV describes the effect of transient frequency offset on both the performance and the initialization speed of the receiver. Section V concludes the paper.

2. Optical coherent burst-mode receivers

Channel equalization is one of the most important DSP functions in high-speed optical coherent receivers. The channel equalizer can be used to compensate not only the accumulated chromatic dispersion of the transmission fiber, but also polarization-mode dispersion (PMD). Both frequency-domain and time-domain equalization have been widely studied [11,12]. For non-burst-mode system, a frequency domain equalizer (FDE) is often used to compensate the bulk fiber dispersion and can support much longer transmission distance [13–15]. The FDE needs to know the amount of accumulated fiber dispersion which should be estimated accordingly. Several methods to estimate fiber dispersion have been proposed in [15,16]. However, an accurate fiber dispersion estimation usually requires observations of several thousand symbols [15,16]. For burst-mode receivers, dispersion compensation with FDE is included in the optical coherent BMR in [10] where the amount of the fiber dispersion is assumed to be known or estimated in advance. However, if the amount of fiber dispersion has to be estimated on a per-burst basis, the length of the training sequence should be long enough for both dispersion estimation and following equalizer adaptation. For long transmission distance, a two-stage approach with FDE plus time-domain equalizer (TDE) for polarization demultiplexing is a better choice. However, for a fiber transmission distance up to 200 km as studied in this work, LMS-based TDE can achieve a much faster convergence speed as shown in the following.

Self-adaptive TDE can instead adapt its coefficients to different transmission parameters at minor performance loss compared to the FDE. Different coefficient adaptation methods exist and this work uses a LMS-based method with training sequence or decision feedback. The training sequence is a pseudo-random sequence known for both transmitter and receiver and it is included in the burst head. The equalizer uses the training sequence during its initialization and moves to a decision feedback mode later. The length of the training sequence is one of the key parameters for the BMR design and should be as small as possible for high transmission efficiency.

Frequency offset estimation and compensation must also be included in an optical coherent receiver. Similar to the TDE, both data-aided and non-data-aided FOE methods can be used [8,17]. An M-th power-based non-data-aided FOE method using the Viterbi-Viterbi algorithm can obtain an FOE range of ± R/8 where R is the symbol rate for QPSK modulation [8] while the data-aided FOE method can extend the FOE range to a much wider range of ± R/2 [17]. However, in the above studies on FOE, fiber dispersion is not considered. This work uses a data-aided method with the help of the training sequence included in the burst head for both LMS-based TDE and FOE. The frequency offset is re-estimated for each burst as each burst comes from a different transmitter.

The interaction between the fiber dispersion and frequency offset is an important issue for optical coherent receivers. In the optical coherent BMR design of [10], a FDE works independently with the FOE and a searching method for dispersion estimation is used for the FDE. In the work from [18], 240 km fiber transmission is included in the experiment. However, this work focuses a MIMO equalizer with CMA for polarization demultiplexing. The fiber dispersion is believed to be compensated using another independent TDE or FDE with known amount of accumulated fiber dispersion.

Different from previous works with independent channel equalizer and FOE, this paper focuses on simultaneous compensation of fiber dispersion and frequency offset and their interaction. The schematic of the optical coherent BMR design with LMS-based TDE and FOE is shown in Fig. 1. A TDE in the conventional butterfly structure is used for dispersion compensation and polarization demultiplexing. For LMS-based method, the tap coefficients are updated using an error signal as

ε_{x (y), k} = z_{x (y), k} - I_{x (y), k} e^{j θ_{k}} = z_{x (y), k} - I_{x (y), k} e^{j (θ_{k - 1} + Δ ω T)} .

For the k-th symbol, z_x(y)_,_k is the output signals of the channel equalizer on x- or y-polarization and I_x(y)_,_k is the transmitted symbol from either the training sequence or decision feedback.

θ_{k}

is the phase rotation of the current symbol and is the sum of the phase rotation

θ_{k - 1}

from the previous symbol and the phase rotation contribution

Δ ω T

from the frequency offset.

Δ ω

is the frequency offset between the transmitter and the receiver and is expressed in angular form with unit radian/s. T is the symbol time period. A maximum-likelihood method similar to [17] is used in the FOE where the phase rotation induced by the frequency offset is estimated as

Δ ω T = \arg \sum_{N} ({\tilde{z}}_{x, k} {\tilde{z}}_{x, k - 1}^{*} + {\tilde{z}}_{y, k} {\tilde{z}}_{y, k - 1}^{*})

where the modulated phase information is removed from z_x(y)_,_k as

{\tilde{z}}_{x (y), k} = z_{x (y), k} I_{x (y), k}^{*} .

N is the number of symbols used in the summation for noise reduction. The output signal of the equalizer is used as input to the FOE and the estimated frequency offset is feedback to the equalizer for computing the error signal. During the initialization stage, both the equalizer and the FOE might give incorrect outputs. The convergence speed of the receiver is affected by the interaction of the two modules as shown later by simulation results.

Fig. 1 Schematic of the optical coherent BMR design.

Download Full Size | PDF

If the fiber transmission is dominated by the second-order fiber dispersion, the fiber channel response c(t) is expressed in the following as the inverse Fourier transform of the dispersive channel’s frequency response

c (t) = F^{- 1} {\exp (- \frac{j}{2} β_{2} ω^{2} L)} .

β₂ is the second-order fiber dispersion coefficient and L is the accumulated fiber transmission length. F⁻¹ stands for the inverse Fourier transform with ω as the angular frequency. The dispersion compensation equalizer under zero frequency offset is simply

h (t) = F^{- 1} {\exp (\frac{j}{2} β_{2} ω^{2} L)}

if the accumulated fiber dispersion is known. The dispersion compensation can be implemented as either an FDE or a TDE for each polarization.

Suppose that s_x(y)(t) are the transmitted signals at the transmitter, $s_{x (y)} (t) \otimes c (t)$ are then the signals after fiber transmission and before mixing with the local oscillator in the receiver. $\otimes$ stands for convolution. If the frequency offset $Δ ω$ between the transmitter and the receiver is non-zero, the received signals are now

r_{x (y)} (t) = (s_{x (y)} (t) \otimes c (t)) e^{j Δ ω t} .

The Fourier transform of the received signals are then

\begin{array}{l} R_{x (y)} (j ω) = {C (j ω) S_{x (y)} (j ω) |}_{ω = ω - Δ ω} \\ = \exp (- \frac{j}{2} β_{2} {(ω - Δ ω)}^{2} L) S_{x (y)} (j (ω - Δ ω)) . \end{array}

After the channel equalizer defined in Eq. (5), the output signals in frequency-domain are

\begin{matrix} Z_{x (y)} (j ω) = H (j ω) R_{x (y)} (j ω) \\ = \exp (+ \frac{j}{2} β_{2} ω^{2} L) \exp (- \frac{j}{2} β_{2} {(ω - Δ ω)}^{2} L) S_{x (y)} (j (ω - Δ ω)) \\ = \exp (j β_{2} ω Δ ω L) \exp (- \frac{j}{2} β_{2} Δ ω^{2} L) S_{x (y)} (j (ω - Δ ω)) . \end{matrix}

The inverse Fourier transform of the above signals are found to be

z_{x (y)} (t) = e^{- j \frac{1}{2} β_{2} Δ ω^{2} L} s_{x (y)} (t + β_{2} Δ ω L) e^{j Δ ω t} .

The frequency offset introduces both a phase rotation and a time shifting t₀ = β₂△ωL to the signal. The phase rotation can be estimated together with the unknown random carrier phase. The time shifting can be compensated with a linear phase rotation in the frequency domain.

e^{j Δ ω t}

is the phase rotation term due to frequency offset and should be compensated by the FOE.

Based on the above analysis, it is confirmed that fiber dispersion compensation and frequency offset compensation can work independently if the fiber transmission effect is dominated by the second-order dispersion. Under this case, the accumulated fiber dispersion can firstly be compensated by FDE or TDE if the amount of fiber dispersion is known or estimated. However, the interaction between the unknown dispersion and unknown frequency offset has great impact on the convergence speed of the receiver. Figure 2 shows the convergence speed of the TDE with different amounts of frequency offsets. Instead of using mean square error, symbol error rate is used to show the combined effect of the TDE and the FOE. During the training period, decisions are made over the frequency offset compensated signals as usual. The number of erroneous decisions are counted and compared with the number of training sequence to compute the symbol error rate in groups of 128 symbols. A pseudo-random training sequence is used. The fiber transmission distance is fixed at 100 km with standard single-mode fiber. A step-size of 0.005 is preferred for a faster convergence speed.

Fig. 2 Convergence speed of the equalizer’s coefficient for different frequency offsets and different step-sizes.

Download Full Size | PDF

It is observed from Fig. 2 that large frequency offset of 5 GHz and 6 GHz clearly need much longer training sequence to converge than the case with 0 GHz or 2 GHz. The reason is due to the interaction of the TDE and the FOE. The larger the frequency offset, a larger phase rotation exists between the consecutive symbols. At the early stage of initialization when the FOE has incorrect outputs, the TDE coefficient adaptation needs longer time to catch up with the larger phase rotation. A longer training sequence is thus needed. It is also observed from Fig. 2 that a frequency offset of 2 GHz has a slightly faster convergence speed than the case without frequency offset. If noise is considered, a non-zero frequency offset of 2 GHz helps the output of the FOE to stabilize and in turn helps to fasten the convergence of the TDE. However, when the frequency offset is increased to 5 GHz or 6 GHz, the interaction between the dispersion compensation and frequency offset estimation with non-stabilized outputs at the initial stage slows down the convergence speed considerably.

However, if a time-varying transient frequency offset exists in the system, the output signals from the channel equalizer defined in Eq. (5) is much more complicated. Both the phase rotation and time shifting from Eq. (9) become time-varying. The equalizer tap coefficients and frequency offset estimator need to follow these time-varying effects for the correct operation of the channel equalizer and frequency offset estimator.

Compared with the LMS-based TDE, CMA is a widely-used adaptation algorithm for the TDE in a non-burst-mode receiver with FDE [13,14]. There, the TDE is used for polarization demultiplexing, residual dispersion compensation and PMD compensation. For this case, a short TDE with 9-16 taps can be used since most of the fiber dispersion has been compensated by the FDE. However, if there is no FDE in the system, CMA has a slower convergence speed than the LMS algorithm. A comparison between the CMA and LMS algorithms without FDE will be included in the following section.

3. Performance of the burst-mode receivers

The performance of the proposed optical coherent BMR is studied with burst transmissions based on simulations. Polarization-division multiplexed QPSK (PDM-QPSK) with a rate of 112 Gb/s is assumed in the simulation. The sampling rate at the receiver is twice the symbol rate. The fiber channel consists of standard single-mode fiber and optical amplifiers. No optical dispersion compensation is used and the accumulated fiber chromatic dispersion is purely compensated with the TDE. PMD is also included in the simulation with an all-order PMD model from [16] as $U (j ω) = \prod_{i = 1}^{N = 20} U_{i} (j ω)$ . 20 differential group delay (DGD) segments with randomly rotating Jones matrices are used with

U_{i} (j ω) = (\begin{matrix} e^{j (ϕ_{i} + ω τ_{i}) / 2} & 0 \\ 0 & e^{- j (ϕ_{i} + ω τ_{i}) / 2} \end{matrix}) \times (\begin{matrix} \cos α_{i} & \sin α_{i} \\ - \sin α_{i} & \cos α_{i} \end{matrix}) .

ϕ_{i}

is the phase shift between the two orthogonal fast and slow axes of each DGD segment,

τ_{i}

is the differential group delay and

α_{i}

is the random polarization rotation from each DGD segment. The differential group delay is of Maxwellian distribution. The mean value of the differential group delay is set to be 38 ps in the simulations, but is found to have negligible effects on the system. For the transmission distances studied in this work, the optical pulse broadening is dominated by the fiber chromatic dispersion.

The training sequence included in the burst head is used to help the initialization of the BMR. The length of the burst head is a key parameter for the receiver design. A long training sequence can be used with a small step-size to guarantee the convergence of the equalizer, but at a cost of reduced channel efficiency. For short training sequences, a relatively large step-size is required which might induce unstable working of the TDE. Other than transmission distance, frequency offset also has great impact on the convergence speed of the BMR as shown in Fig. 3 with 100 km and 200 km transmission distances. The OSNR used in the simulation is kept at 13 dB for bit error rate (BER) to be below 10⁻³. The number of taps for each equalizer is fixed at 40. Figure 3(a) shows the convergence of the BMR and Fig. 3(b) shows the estimated frequency offset from the FOE. It is clear that during the early stage of the initialization, both the TDE and FOE have incorrect outputs. The longer the transmission distance, the larger the frequency offset, a longer training sequence is needed. At a transmission distance of 200 km, a frequency offset range of ± 5 GHz can be allowed if a training sequence less than 3K symbols is used. At a transmission rate of 112 Gb/s with QPSK, 3K symbols corresponds to an initialization time of about 120 ns.

Fig. 3 Convergence speed and the error performance of the BMR with different frequency offsets, different step-sizes and different transmission distances.

Download Full Size | PDF

It is noticed that the longer transmission distance of 200 km clearly has a slower convergence speed than the case of 100 km. The longer the transmission distance is, the optical pulses are broadened further and a slower convergence speed is expected due to more interferences caused by the signal broadening. Moreover, different signals samples have a different phase rotation due to the frequency offset. Suppose that a single pulse is send at time t = kT with phase modulation I₀, the samples of the received signal of (6) is then

r_{k} = (I_{0} δ (k) \otimes c_{k}) e^{j k Δ ω T} = I_{0} c_{k} e^{j k Δ ω T}

δ (k)

is the discrete-time delta function and c_k is the discrete-time fiber dispersive channel response. A single polarization is assumed here for simplicity. The output signal of the TDE is then

z_{k} = \sum_{m} h_{m} r_{k - m} = \sum_{m} h_{m} I_{0} c_{k - m} e^{j (k - m) Δ ω T} = I_{0} e^{j k Δ ω T} \sum_{m} h_{m} c_{k - m} e^{- j m Δ ω T} .

To obtain I₀ as the result, the phase rotation

e^{j k Δ ω T}

is supposed to be compensated by the FOE. Moreover, to make

\sum_{m} h_{m} c_{k - m} e^{- j m Δ ω T} = δ (k)

, the TDE tap coefficients h_m should compensate the varying phase rotation term

e^{- j m Δ ω T}

from the frequency offset. The larger the frequency offset is, the larger phase rotation should be incorporated into the tap coefficients and a longer adaptation time is needed. Consequently a slower convergence speed is experienced in general for large frequency offset. For the case of 200 km transmission distance and 5 GHz frequency offset, the estimated frequency offset oscillates considerably during the early stage of the initialization. The interaction of the unstable output signals from the TDE and the FOE slows down the convergence speed considerably.

Figure 3(c) shows the error performance of the optical coherent BMR with different values of frequency offset. Two fiber transmission distances of 100 km and 200 km are included in the simulation. The length of the training sequence is 4K symbols and the step-size is 0.005 for all cases. As long as the BMR finishes its initialization with correctly estimated frequency offset, the BMRs have almost identical error performance for different frequency offsets.

Longer fiber transmission distance can be supported, but at a cost of reduced frequency offset. Table 1 shows the allowed frequency offset range obtained from simulations as the fiber transmission distance increases. The length of the training sequence is fixed at 4K symbols and the OSNR is fixed at 13 dB for the simulations. From Table 1, it is seen that ± 5 GHz frequency offset can be tolerated if the fiber transmission distance is limited to 200 km. However, the working range of the frequency offset reduces to ± 2 GHz at a transmission distance of 800 km. Note that for each 100 km longer transmission, 20 more taps should be added to the equalizer. To support 800 km fiber transmission distance, a TDE should have about 160 taps and this large number of taps might not be realistic for implementation at the required transmission rate. Moreover, the step-size should be reduced for stable operation as the number of taps increases.

Table 1. Working range of frequency offset at different transmission distances

View Table

It is evident from the previous study that the interaction between the LMS-based TDE and the FOE has a great impact on the system performance. If CMA is used to replace the LMS algorithm, the TDE can work independently with the FOE. For CMA, the error signal for the tap coefficient updating with QPSK modulation is [19]

ε_{x (y), k} = z_{x (y), k} (1 - {| z_{x (y), k} |}^{2}) .

Thus, no phase rotation information from the FOE is required for CMA. In [19], a similar performance is found between the CMA and LMS algorithms for dispersion up to 1000 ps/nm with 14 Gbaud 16-QAM optical coherent system. However, a small step size of 0.001 is used in that study which requires more than 20K iterations for the TDE to converge. Such a small step size guarantees that both the CMA and LMS algorithms have good performances, but at a cost of very slow convergence speed.

To achieve fast initialization time for BMR, a much larger step size is required. Under this case, the LMS algorithm shows much faster convergence speed than CMA. Figure 4 compares the convergence speed of the two algorithms for both 100 km and 200 km transmission distances at a frequency offset of 2 GHz. A step size of 0.005 is used for both cases. Instead of using an OSNR of 13 dB as in Figs 3(a) and 3(b), an OSNR of 15 dB is used in Fig. 4 since CMA fails to converge at 13 dB for 200 km transmission. At a transmission distance of 100 km, the LMS algorithm reaches convergence with only 1K iterations while CMA requires 3K iterations. If the transmission distance is increased to 200 km, more than 10K iterations are necessary for CMA to converge which is much slower than the LMS algorithm. For this reason, LMS is selected for the studied optical coherent BMR without FDE.

Fig. 4 Comparison on the convergence speed of the CMA and LMS algorithms.

Download Full Size | PDF

It is also known that the recursive least-squares (RLS) algorithm can achieve a faster convergence speed than the LMS algorithm [20]. However, the RLS algorithm has a much higher computation complexity due to the inverse matrix computation in the adaptation [20] and is thus not considered in this work.

4. Effect of transient frequency offset

Wavelength tunability is an important property for flexible WDM networks. Fast tunable lasers (FTL) can be included in both the transmitters and the BMRs for this purpose. To achieve fast wavelength tuning [21], proposes to use SOA blanking to avoid interference to other existing channels during the mode competition. However, the instantaneous wavelength variation cannot be avoided during wavelength locking. The wavelength variation can be described as [21]

Δ λ_{OT} (t) = Δ λ_{init} e^{- ω_{E} t} \cdot \cos (ω_{O} t),

where △λ_init is the initial wavelength deviation, ω_E is the damping coefficient and ω_O controls the period of the sinusoidal oscillation. Figure 5 shows an example of the transient frequency offset using (14). The parameters are obtained from the results of [10] where the initial frequency offset from the initial wavelength deviation is 2 GHz. The oscillation period is about 250 ns and significant transient frequency offset close to 1 GHz still exists after 250 ns. The damping coefficient is set to be 0.0041 ns⁻¹.

Fig. 5 Transient frequency offset with wavelength tunable BMR.

Download Full Size | PDF

Figures 6(a) and 6(b) compare the convergence speed of the BMR with and without transient frequency offset for both 100 km and 200 km fiber transmission distances. A static frequency offset of 1 GHz is added in the simulation together with the transient frequency offset from Fig. 5. A larger step-size of 0.01 is used on the TDE for a faster convergence speed at minor OSNR penalty due to more noise included in the equalizer’s output signals. The longer the transmission distance, a slower convergence speed is observed. Moreover, for the case of 200 km transmission distance, Fig. 6(a) shows that the equalizer has reached its convergence state after 4K training symbols. However, the time-varying transient frequency offset during the period of 5K to 6K training symbols could bring the equalizer into an unstable state with non-zero error rate and deviated frequency offset estimation.

Fig. 6 Convergence speed and error performance of the BMR with and without transient frequency offset.

Download Full Size | PDF

From previous discussion with Eq. (12), the TDE tap coefficients should be able to compensate a varying phase rotation term $e^{- j m Δ ω T}$ from the frequency offset. For static frequency offset, the tap coefficients have converged to a stable state after the initialization and a stable operation is obtained. However, if transient frequency offset exists in the system, the tap coefficients have to be updated following the variation of the frequency offset $Δ ω$ . An unstable operation of the TDE is possible if the tap coefficients are not updated fast enough. From Fig. 6, it is noticed that the TDE is unstable around 2K and 5.5K symbols. The corresponding frequency offset curve is shown in Fig. 5 with red color. By taking time derivative of (14), it is found that the TDE becomes unstable around the first and second local maxima of the time derivative of the transient frequency offset, which correspond to the fastest time changing of the frequency offset. For the same reason, a larger step-size at 0.01 is preferred over 0.005 for a faster tap coefficient updating to follow the time variation of the transient frequency offset. A step-size of 0.02 shows no further improvement on the convergence speed, but has a higher OSNR penalty. If the step-size is increased over 0.02, an unstable operation of the TDE is observed. Consequently, the step-size of 0.01 is chosen for the case with transient frequency offset.

Figure 6(c) compares the error performance of the BMR for 100 km and 200 km fiber transmission distances. A training sequence of 4K symbols is assumed which corresponds to an initialization time of 160 ns. For the case of 200 km transmission distance, an OSNR penalty of 1dB is observed at the BER of 10⁻³. The BMR also shows an error floor at about 7 × 10⁻⁴ for this case. Since the training sequence has only 4K symbols, the first 2K symbols of the decision feedback mode has relatively high error rate as the transient frequency offset brings the equalizer into an unstable state.

If CMA is used to replace the LMS algorithm for the TDE with transient frequency offset, simulation results of Fig. 7 show that CMA still has a slower convergence speed than the LMS algorithm. For 200 km transmission distance, CMA requires more than 14K symbols to converge. After 14K symbols, only small time variation is left for the transient frequency offset and no unstable operation of TDE is observed. Compared with the slow convergence speed of CMA, the LMS algorithm is a better choice for fast initialization of the BMR on a per-burst basis with unknown fiber dispersion.

Fig. 7 Comparison on the convergence speed of the CMA and LMS algorithms with transient frequency offset.

Download Full Size | PDF

5. Conclusion

Optical coherent BMRs with tunable wavelength are good choices to meet the flexibility and sub-wavelength granularity requirement of dynamic optical WDM networks. LMS-based TDE and FOE are proposed as a new BMR design for simultaneous fiber dispersion compensation and frequency offset estimation. The interaction of the two modules is studied through simulation. The range of frequency offset under which the BMR works is found for different fiber transmission distances. With a training sequence as short as 4K symbols, 200 km non-dispersion-compensated fiber transmission distance with over ± 5 GHz frequency offset can be supported at negligible OSNR penalty.

Transient frequency offset is observed during the wavelength tuning period for the wavelength tunable BMRs and its effect on the BMR is studied for the first time in this paper. If time-varying frequency offset is included in the simulation, the BMR shows worse error performance with a reduced convergence speed. For a fiber transmission distance of 200 km, the OSNR penalty increases to about 1dB with an error floor at about 7 × 10⁻⁴. A training sequence as short as 4K symbols is still enough if super-FEC is included in the system for error control.

This paper focuses on theoretic and simulation studies on the interaction of fiber dispersion and frequency offset and its effect on the BMR design. With the performance of designed BMR verified by simulation results, the BMR implementation with parallel DSP functions will be considered for the future work.

Acknowledgment

This work is supported by the 863 high technique program of China under contract number #2012AA011304.

References and links

1. J. D. Downie, J. Hurley, D. Pikula, and X. Zhu, “Ultra-long-haul 112 Gb/s PM-QPSK transmission systems using longer spans and Raman amplification,” Opt. Express 20(9), 10353–10358 (2012). [CrossRef] [PubMed]

2. P. J. Winzer, “High-spectral-efficiency optical modulation formats,” J. Lightwave Technol. 30(24), 3824–3835 (2012). [CrossRef]

3. M. Tomizawa, “DSP aspects for deployment of 100G-DWDM systems in carrier networks,” in Optical Fiber Communication Conference, OSA Technical Digest (Optical Society of America, 2012), paper CTuT2. [CrossRef]

4. A. Meiyappan, P.-Y. Kam, and H. Kim, “A complex-weighted, decision-aided, maximum-likelihood carrier phase and frequency-offset estimation algorithm for coherent optical detection,” Opt. Express 20(18), 20102–20114 (2012). [CrossRef] [PubMed]

5. D. Lavery, M. Ionescu, S. Makovejs, E. Torrengo, and S. J. Savory, “A long-reach ultra-dense 10 Gbit/s WDM-PON using a digital coherent receiver,” Opt. Express 18(25), 25855–25860 (2010). [CrossRef] [PubMed]

6. M. Tomizawa, “Real-time implementation of packet-by-packet polarization demultiplexing in a 28 Gbs burst mode coherent receiver,” in Optical Fiber Communication Conference, OSA Technical Digest (Optical Society of America, 2012), paper OM3H.6.

7. J. E. Simsarian, J. Gripp, A. H. Gnauck, G. Raybon, and P. J. Winzer, “Fast-tuning 224-Gb/s intradyne receiver for optical packet networks,” in Optical Fiber Communication Conference, OSA Technical Digest (Optical Society of America, 2010), paper PDPB5.K. [CrossRef]

8. B. C. Thomsen, R. Maher, D. S. Millar, and S. J. Savory, “Burst mode receiver for 112 Gb/s DP-QPSK with parallel DSP,” Opt. Express 19(26), B770–B776 (2011). [CrossRef] [PubMed]

9. R. Maher, D. S. Millar, S. J. Savory, and B. C. Thomsen, “Widely tunable burst mode digital coherent receiver with fast reconfiguration time for 112 Gb/s DP-QPSK WDM networks,” J. Lightwave Technol. 30(24), 3924–3930 (2012). [CrossRef]

10. M. Li, N. Deng, F. N. Hauske, Q. Xue, X. Shi, Z. Feng, S. Cao, and Q. Xiong, “Optical burst-mode coherent receiver with a fast tunable LO for receiving multi-wavelength burst signals,” in Optical Fiber Communication Conference, OSA Technical Digest (Optical Society of America, 2012), paper OTu1G.3. [CrossRef]

11. M. S. Faruk and K. Kikuchi, “Adaptive frequency-domain equalization in digital coherent optical receivers,” Opt. Express 19(13), 12789–12798 (2011). [CrossRef] [PubMed]

12. A. Vgenis, C. S. Petrou, C. B. Papadias, I. Roudas, and L. Raptis, “Nonsingular constant modulus equalizer for PDM-QPSK coherent optical receivers,” IEEE Photon. Technol. Lett. 22(1), 45–47 (2010). [CrossRef]

13. J.-X. Cai, C. R. Davidson, A. Lucero, H. Zhang, D. G. Foursa, O. V. Sinkin, W. W. Patterson, A. N. Pilipetskii, G. Mohs, and N. S. Bergano, “20 Tbit/s transmission over 6860 km with sub-nyquist channel spacing,” J. Lightwave Technol. 30(4), 651–657 (2012). [CrossRef]

14. J. C. Cartledge, J. D. Downie, J. E. Hurley, X. Zhu, and I. Roudas, “Bit error ratio performance of 112 Gb/s PM-QPSK transmission systems,” J. Lightwave Technol. 30(10), 1475–1479 (2012). [CrossRef]

15. M. Kuschnerov, F. N. Hauske, K. Piyawanno, B. Spinnler, M. S. Alfiad, A. Napoli, and B. Lankl, “DSP for coherent single-carrier receivers,” J. Lightwave Technol. 27(16), 3614–3622 (2009). [CrossRef]

16. R. A. Soriano, F. N. Hauske, N. G. Gonzalez, Z. Zhang, Y. Ye, and I. T. Monroy, “Chromatic dispersion estimation in digital coherent receivers,” J. Lightwave Technol. 29(11), 1627–1637 (2011). [CrossRef]

17. X. Zhou, X. Chen, and K. Long, “Wide-range frequency offset estimation algorithm for optical coherent systems using training sequence,” IEEE Photon. Technol. Lett. 24(1), 82–84 (2012). [CrossRef]

18. R. Maher, D. S. Millar, S. J. Savory, and B. C. Thomsen, “Fast switching burst mode receiver in a 24-channel 112Gb/s DP-QPSK WDM system with 240km transmission,” in Optical Fiber Communication Conference, OSA Technical Digest (Optical Society of America, 2012), paper JW2A.57. [CrossRef]

19. I. Fatadin and S. J. Savory, “Blind equalization and carrier phase recovery in a 16-QAM optical coherent system,” J. Lightwave Technol. 27(15), 3042–3049 (2009). [CrossRef]

20. K. Roberts, M. O'Sullivan, K.-T. Wu, H. Sun, A. Awadalla, D. J. Krause, and C. Laperle, “Performance of dual-polarization QPSK for optical transport systems,” J. Lightwave Technol. 27(16), 3546–3559 (2009). [CrossRef]

21. A. Bianciotto, B. J. Puttnam, B. Thomsen, and P. Bayvel, “Optimization of wavelength-locking loops for fast tunable laser stabilization in dynamic optical networks,” J. Lightwave Technol. 27(12), 2117–2124 (2009). [CrossRef]

Transmission distance (km)	Tap number	Frequency offset (GHz)	Step size
100	20	± 7	0.01
200	40	± 6	0.005
300	60	± 5.5	0.005
400	80	± 5	0.003
500	100	± 4	0.003
600	120	± 3.5	0.003
700	140	± 3	0.003
800	160	± 2.5	0.003

Wavelength tunable coherent burst-mode receiver design under transient frequency offset

Abstract

1. Introduction

2. Optical coherent burst-mode receivers

3. Performance of the burst-mode receivers

4. Effect of transient frequency offset

5. Conclusion

Acknowledgment

References and links

Cited By

Figures (7)

Tables (1)

Equations (14)

Optics Express