1. Introduction

Coherent optical communication is one of the most promising ways to achieve highest receiver sensitivity, excellent spectral efficiency and longest transmission distance for the next generation of optical communication systems. Already in the late 1980s and early 1990s coherent systems attracted a lot of attention as it was a promising way to improve the receiver sensitivity. Coherent PSK systems can in principle achieve a receiver sensitivity of 9 photons/bit, which is the best receiver sensitivity of all known modulation formats. But after the development of erbium-doped fiber amplifiers (EDFA) and the introduction of wavelength division multiplexing (WDM) the interest in coherent systems faded.

In the last years, with the new possibilities offered by high-speed digital circuits, the interest in coherent optical communication systems revived. Apart from receiver sensitivity the interest lies now in the increase of the spectral efficiency as well as in the tolerance against dispersion effects and fiber nonlinearities. These are today the most limiting factors in ultralong haul communication systems. Intensity modulated systems with direct detection (IM/DD) can achieve a maximum spectral efficiency of only 1 b/s/Hz. Coherent systems allow to use multilevel modulation formats such as QPSK or QAM, which can raise the spectral efficiency up to several b/s/Hz. Additionally, in contrast to IM/DD systems or differential PSK systems, the received electrical signal in coherent receivers is proportional to the electrical field vector of the optical signal. Therefore the system becomes linear, which means that all linear distortions like chromatic dispersion as well as polarization mode dispersion can theoretically be compensated without any losses – and also nonlinear effects can be equalized very efficiently.

However, these advantages have their price. A coherent optical receiver is much more complex than a simple direct detection receiver. High speed analog-to-digital converters (ADC) are needed to convert the received signal into the digital domain. Then these ADCs must be interfaced with a digital signal processing unit, which performs polarization control, equalization and finally the carrier and data recovery. To develop these components for the next generation of optical communication systems, which will run at 40 Gbit/s or even 100 Gbit/s poses a big challenge to scientists and engineers.

This paper reviews the achievements made until today towards a realtime coherent communication system running at 40 Gbit/s. First the potential of the technology is demonstrated by showing some main results from offline coherent experiments. Then the components needed to realize these systems in realtime are discussed. In section 4 the milestones in realizing realtime coherent systems are resumed, and finally the measurement results of a realtime coherent synchronous QPSK system with a 3×3 coupler in the receiver are presented and discussed.

2. Potential of coherent optical communication systems

Coherent optical communication promises ultimate performance and dispersion tolerance in upgraded or newly built fiber links. Many offline experiments have already been conducted to evaluate the possibilities offered by this technology. Most of them concentrated on the evaluation of polarization-multiplexed quadrature phase shift keying (QPSK) systems [1–3]. But also 8-level phase shift keying (8-PSK) [4] and quadrature amplitude modulation (QAM) have already been investigated [5].

The results of these experiments clearly demonstrate the potential of coherent optical communication. For polarization multiplexed QPSK transmission data rates up to 86 Gbit/s have been reached [6]. In another experiment 42.8 Gbit/s polarization multiplexed QPSK data was transmitted over 6400 km of single mode fiber (SMF) with purely electronic dispersion compensation [7]. This represents the longest reach achieved for transmission without optical dispersion compensation for any 10 GBaud system operating over standard fiber as well as for all 40Gbit/s systems. The highest number of bit/symbol achieved in a coherent transmission experiment was 12 bit/symbol in a polarization multiplexed 64 QAM testbed [8], reaching a spectral efficiency as high as 6 bit/s/Hz.

3. Components for realtime coherent QPSK transmission

In the following the DSP components are described that are essential for the realization of a high-speed realtime coherent optical receiver.

3.1 High-speed analog-to-digital converter

The coherent receiver requires a very fast analog-to-digital converter (ADC) with a sampling rate equivalent to the symbol rate to digitize the incoming signal. If CD compensation or PMD compensation should be applied a sampling rate even twice as high as the symbol rate is recommended. A very good architecture for such high-speed ADCs is the full-flash topology where 2^N-1 parallel comparators are utilized to digitize the signal with a resolution of N bits in one step. The practical resolution of such flash ADCs is limited to 4–6 bits due to the loading of the input buffer with the 2^N-1 comparators, which restricts the achievable bandwidth and sampling speed.

Table 1 shows the specifications of a 5 bit 10 Gs/s ADC developed for the European synQPSK project. A fast circuit technology was necessary to accomplish the required sampling rate of >10 Gs/s, together with a fairly high level of integration to incorporate the encoding stages of the ADC. A 120 GHz SiGe hetero bipolar technology (IHP, Germany) turned out to be sufficient. The chip layout of the ADC is depicted in Fig. 1(a).

3.2 Digital signal processing unit

To facilitate the signal processing functions for carrier recovery, polarization control and CD or PMD compensation, standard-cell based silicon CMOS technology is preferable due to its higher integration capability and lower power requirements compared to a bipolar technology. A drawback of CMOS standard-cell design is the limitation of the attainable data rates. Therefore, a fast demultiplexer must be built into the DSP unit to allow a high speed interface with the input signals from the ADCs. It must be realized with a full-custom demultiplexer circuit, integrated together with the standard-cell design into a single chip. The full-custom demultiplexer can use a differential circuit topology (source coupled FET logic) to achieve the necessary data rates for the high speed interface.

Fig. 1(b) shows the layout of two full-custom 1:8 demultiplexers combined with a standard-cell DSP unit for carrier & data recovery of a synchronous QPSK transmission system at 10 Gbaud. The standard-cell based DSP unit is clocked at 625 MHz to enable the complex processing in 16 parallel units in a PLL-free approach [9]. At the output of the DSP unit 32 recovered data bits (16 channels, I&Q) are available at a data rate of 625 Mbit/s. Table 1 lists the specifications of the combined full-custom demultiplexer and the standard-cell DSP unit.

 

Fig. 1. Layout of 5 bit SiGe ADC chip (a) and CMOS chip with combined full-custom design for demultiplexing and standard-cell design for carrier & data recovery (b).

Download Full Size | PDF

Table 1. Specifications of SiGe ADC and CMOS chip

View Table

Another possible receiver design is to combine the high-speed ADC and the signal processing unit in a single chip in a BiCMOS or CMOS technology [10]. This has the advantage that the system has lower power consumption and the high-speed interface between the ADCs and the CMOS chip can be omitted. However, highest bipolar transit frequency f_T and smallest CMOS feature size can so far not be combined in the same BiCMOS technology. And in CMOS it is very difficult to achieve the needed ADC input bandwidth.

4. Recent realtime coherent transmission experiments

The first realtime coherent synchronous QPSK transmission was published in 1991 [11]. 100 Mbit/s QPSK data was received in a digital receiver using a phase-locked-loop (PLL) to compensate for the intermediate frequency and phase offset between the signal laser and the local oscillator laser. The drawback of the PLL approach is that external cavity lasers (ECL) are required. The PLL can only lock, if the laser linewidth times symbol rate ratio is below 0.0001. Another approach which is more tolerant against laser phase noise and allows to use standard DFB lasers is the feed-forward carrier recovery [12]. The first realtime coherent transmission system using this kind of receiver was demonstrated in 1992 [13], with a throughput of 565 Mbit/s of PSK modulated data. The data was recovered synchronously in a coherent receiver with analog carrier recovery and standard DFB lasers.

More than a decade later, in June 2006, the first realtime QPSK transmission with standard DFB lasers was published [14]. This time a digital synchronous coherent receiver with feed-forward carrier recovery was applied for a data rate of 800 Mbit/s. The system was realized using commercially available ADCs interfacing with a field programmable gate array (FPGA) for signal processing. In November 2006 the maximum data rate for coherent synchronous QPSK transmission was already pushed to 4.4 Gbit/s [15]. Another half year later the spectral efficiency for realtime systems was doubled by a polarization-multiplexed QPSK transmission system with fast electronic polarization tracking at a data rate of 2.8 Gbit/s [16], still realized with commercially available electronic components. But in July 2007 Nortel announced the first realtime implementation of a polarization multiplexed coherent QPSK system with a data rate of 40 Gbit/s, implemented this time in an application specific integrated circuit (ASIC) [17].

5. Realtime coherent QPSK transmission applying a 3×3 coupler in the receiver

Novel transmission formats are not only assessed by their performance, but also by their costs and complexity. Therefore the realization of coherent QPSK transmission with standard components, which can be produced at low cost, is essential. One milestone to reach this goal was the development of a phase noise tolerant carrier recovery scheme, which allowed the use of standard DFB lasers [12]. Another possibility to reduce the component costs is the replacement of the optical 90° hybrid, which is in general realized with integrated optics [18] or free-space micro-optics, by a fused symmetric 3×3 coupler. This is advantageous compared to the use of asymmetric 3×3 couplers, which can be operated as 90° hybrids as well, but without suppressing direct detection terms. This section describes how the I&Q components can be extracted from the 3×3 coupler output signals and presents the first measurement results for a realtime QPSK transmission with a 3×3 coupler applied in the receiver.

5.1 Mathematical description

From a complex I&Q signal X obtained in a coherent receiver with 90° hybrid, a 3-phase signal (x ₁, x ₂, x ₃) similar to the output of a 3×3 coupler can be obtained as

It contains the same information as X and fulfills the additional condition x ₁+x ₂+x ₃=0.

But in receivers with 3×3 coupler the latter is not the case because of thermal noise. Nevertheless Re{X} and Im{X} can be obtained by linear combinations of the output signals x ₁, x ₂ and x ₃ of a 3×3 coupler. In order to minimize the influence of thermal, quantization, direct detection, ASE and intensity noise, each of x _1,2,3 must be weighted proportional to the contained amplitude of the respective component. A suitable transformation is

This is superior to the simple assignment Re{X}~x ₁, Im{X}~x ₂-x ₃ since subsequent application of first (2), then (1) enforces x ₁+x ₂+x ₃=0, as recommended in [19]. The approximation √3≈7/4=1.75 can be used for a hardware-efficient implementation of (2).

The weighting proportional to the desired amplitude component can also be automated, which is especially advantageous if coupler properties are not known exactly, as is the case in practice. The input transformation matrix can be adaptively optimized by correlating the data before and behind the decision circuit, similar to the polarization control algorithm described in [12]. By setting the complex input vector to

where X ₁ and X ₂ are originally the received signals in two polarizations, the algorithm [12] can also be used to automatically adapt the transformation matrix to the 3-phase input signal. In doing so the algorithm even compensates for non-ideal 3×3 coupler characteristics.

5.2 Experimental setup

The transmitter uses a DFB signal laser with a specified linewidth of 1 MHz and a QPSK modulator driven with 2×700 Mbit/s precoded PRBS data with a sequence length of 2²³-1 (Fig. 2). After transmission through 80 km of standard single mode fiber, the signal is optically preamplified and filtered by a ~20 GHz wide bandpass. The coherent receiver features a second DFB laser as its local oscillator, and manual polarization control. The two optical signals are superimposed in a fused 3×3 coupler and detected with three photodiodes. The resulting electrical I&Q signals are amplified before being sampled with 6-bit analog-to-digital converters (ADCs). The ADCs interface with a Xilinx Virtex 4 FPGA where the three phase signal is converted to an I&Q signal according to (2) or (3) and the carrier and data are recovered according to [9]. The FPGA also generates a clock phase error signal and a LO frequency error signal for NRZ clock recovery and LO frequency control.

 

Fig. 2. Setup for 1.4 Gbit/s coherent QPSK transmission with 3×3 coupler used in the receiver.

Download Full Size | PDF

5.3 Measurement results and discussion

Fig. 3 shows the averaged BER of the received I&Q channels against the preamplifier input power at the receiver. The minimum BER was 2.8·10^-5 for the receiver configuration with fixed transformation (2). With adaptive transformation ((3) combined with [12]) the best measured BER was 7.3·10^-5. For both receivers the PRBS could be detected until the preamplifier input power was set below -52 dBm.

For comparison the BER performance of a receiver with a 90° hybrid is also shown in Fig. 3. It can be seen that the receiver sensitivity is about 1 dB higher than for the 3×3 coupler and fixed transformation function in the receiver. This results from non-ideal coupling ratios in the 3×3 coupler. In contrast, the 90° hybrid allows control of coupling ratios as well as phase shift. For the 3×3 coupler receiver with adaptive transformation the receiver sensitivity is the same as with 90° hybrid because the algorithm compensates for the non-ideal 3×3 coupler – but only for low preamplifier input power levels. The BER floor is higher than for the receiver with hard-wired transformation function. Very recently, the same effect showed up after the system was re-engineered to track fast polarization fluctuations in a polarization-multiplexed QPSK transmission. A hardware error in the ADC-FPGA interface was then found, and eliminated. This has brought down the BER floors. We are confident that the same would hold for this 3×3 coupler experiment but can not test it because the system is needed for ongoing other experiments.

 

Fig. 3. Measured BER vs. preamplifier input power for 1.4 Gbit/s synchronous QPSK transmission.

Download Full Size | PDF

6. Summary

We have summarized the recent achievements in coherent optical communication, considering offline and realtime experiments. Electronic components were presented which are necessary to realize a realtime coherent QPSK transmission system at 10 Gbaud. Additionally the measurement results for the realtime implementation of a synchronous QPSK transmission system with a data rate of 1.4 Gbit/s and a 3×3 coupler in the receiver were shown. The minimum achieved bit error rate of 2.8·10^-5 is close to the one achieved with a 90° hybrid and is well below the FEC threshold.