1. INTRODUCTION

Optical fiber links are an emerging technology for high resolution time and frequency transfer and comparison. Single-mode optical fibers are a very promising medium to carry metrological signals because the guided propagation ensures excellent stability of the propagation delay, an unambiguous propagation path, and a much higher signal and lower noise than propagation in free space over long hauls.

In order to reach the lowest transfer instability, propagation noise has to be cancelled [1]. Due to the excellent reciprocity of the phase accumulated back and forth, the noise can be measured using a round-trip signal [2]. This can be done by retroreflecting part of the signal, or by using a laser which is offset phase-locked to the incoming signal [3]. This latter technique has numerous advantages, as the light injected in the backward direction is constant in power. It is a key technique for the regeneration of laser light in a cascaded link, where the link length is subdivided in smaller spans, allowing for better noise rejection [4]. In addition, it provides the opportunity to deliver a useful signal to an end user [5]. This technique was developed over 10 years in the optical and microwave domain [6–8] and was recently implemented on a long-haul link of $2\times 710\mathrm{km}$, with four cascaded spans [5]. Cascaded-link technique was used in Europe for clock comparisons between National Metrology Institutes [National Physical Laboratory (NPL), Physikalisch Technische Bundesanstalt (PTB), and Laboratoire National de Métrologie et d’Essais – Système de Références Temps-Espace (LNE-SYRTE)] [9–11].

It motivates a wide community of scientists, from those in fundamental physics to those in applied physics and industry, to consider this technique for various applications [12–19]. Projects aiming at creating metrological networks at national and continental scale over telecommunication fiber networks are pushing the technology forward, as their implementation requires a high level of maturity [20–23]. Such metrological networks require the know-how to be transferred from academia to industrial partners who can in turn provide the scientific community with a high quality of service. As a first step towards such a wide-scale metrological network, we report here on the first long-haul coherent fiber link built, set up, and managed by an industrial consortium, with supervision abilities integrated into the network operating center of a national research and education network.

The paper is organized as follows. First, we discuss the scientific objectives and subsequent requirements of a sustainable metrological network, the way the fiber is accessed, and the equipment to be built. Then we show one of the pieces of equipment we have developed for this network, an industrial-grade version of the repeater laser station (RLS) [5]. We focus our discussion on the remote abilities and the supervision that was implemented to guarantee a proper operability by a network operation center. We report on the performance obtained using a dedicated test bench, and describe the implementation and operation of one link of the network, connecting two cities in France over 680 km of coherent fiber link. We show our experimental results and compare with the state of the art.

2. REFIMEVE+ PROJECT

The frame of the work presented in this paper is the REFIMEVE+ project [23]. It aims at disseminating an ultrastable optical frequency standard emitting at a wavelength in the C-band of the International Telecommunication Union (ITU) grid over 4000 km of an active and operational telecommunication fiber network (see Fig. 1). The network is operated by RENATER, the French National Research and Education Network (NREN). The ultrastable optical signal is generated at the French National Metrology Institute for Time and Frequency, LNE-SYRTE. It consists in continuous-wave coherent light from a laser that is phase-locked to an ultrastable cavity that can be related to the absolute frequency of frequency standards using optical frequency combs. The ultrastable signal has to be disseminated to about 20 research laboratories in France, for applications ranging from comparison of optical clocks and frequency standards to tests of advanced physics theories, such as temporal variation of fundamental constants and general relativity, ultrahigh resolution atomic and molecular spectroscopy, radio astronomy, and applied physics topics such as the detection of atmospheric pollutants. At the user end, the ultrastable signal can be used directly, or with the help of an optical frequency comb.

 

Fig. 1. Map of France with the REFIMEVE+ network. Red arrows, operational links; blue arrows, planned links; orange arrows, international cross-border links towards Braunschweig (Germany), London (UK), and Torino (Italy). In this paper, we focus on the link between Paris and Lille (thick red line).

Download Full Size | PDF

The scientific requirements are a short-term relative frequency stability of the optical signal at the level of ${10}^{-15}$ at 1 s averaging time, and below ${10}^{-19}$ on one day. The accuracy of the transferred signal must be limited by the frequency standards themselves. Finally, the uptime must be as close as possible to 100%, in order to provide an efficient service to the scientific community.

The fiber access is provided by RENATER. The metrological signal propagates in the same fiber as telecom data traffic using the dense wavelength division multiplexing (DWDM) technology [24]. REFIMEVE+ uses the channel #44 centered at 194.4 THz (1542.14 nm). This approach is essential for two reasons. First, it reduces the costs, as the fiber is shared and as many maintenance operations are handled by an operating center. Second, it enables the operation of a scientific instrument over the long term, with an uptime as close as possible to 100%.

Thermal and acoustic fluctuations in the fiber introduce phase noise, which deteriorates the frequency stability of the metrological signal. To overcome this limitation, an active noise compensation is implemented, the round-trip phase fluctuations in the fiber being measured and corrected in real time. This requires the metrological signal to be bidirectional for optimal noise rejection. Any telecom or network equipment is bypassed thanks to bidirectional optical add/drop multiplexers (OADMs). Channel #44 is added and dropped in the Internet network with 0.2 dB insertion loss for the data traffic. The OADM’s insertion loss for the metrological signal is about 0.8 dB.

Moreover, the signal attenuation is about 0.3 dB/km in average in-field, because of losses both in the fiber and on multiple optical interfaces. In every telecom shelter where a unidirectional erbium doped fiber amplifier (EDFA) has to be bypassed, a bidirectional EDFA amplifies the metrological signal. However, the gain of these amplifiers has to be kept quite low (typically 12–21 dB) to prevent self-oscillations due to parasitic Fabry–Perot cavity effects. Therefore, losses cannot be fully compensated with the bidirectional EDFAs. For long-range fiber links, better amplification techniques such as fiber Brillouin amplification or regeneration techniques, for example, using the RLSs described below, are used [5,25]. For the REFIMEVE+ project, the solution is to set up RLSs along the link every 300–500 km, depending on the link noise and the link loss net budget. This solution allows us to set up cascaded links with better noise rejection and to build a fiber network that is both robust and reliable.

Beyond the RLSs and the bidirectional EDFAs, building a nationwide network of 4000 km requires other specific modules, such as multiuser stations, central hubs, and end-user modules. For that purpose, a specific technological development was conducted by Laboratoire de Physique des Lasers and co-workers. Technology transfer has been implemented to develop industrial grade equipment, a necessity for a robust and reliable operation of the REFIMEVE+ network. Not only manufacturing, but the whole process of installing, powering up, optimizing, and operating the link in-field must be carried out according to industry standards so that REFIMEVE+ becomes a large effective and productive research infrastructure with upscale capacity.

Two key devices have been developed and used in the work presented here: the bidirectional amplifiers and the RLSs. The bidirectional amplifiers were developed by Keopsys [26]. The minimal input power is $-70\mathrm{dBm}$, and they achieve a noise figure below 5 dB with input signals as low as $-50\mathrm{dBm}$. Even with strong input signals, output power is limited by saturation to $+5\mathrm{dBm}$ to comply with the operational requirements of the network. This equipment has been conceived to incorporate two bidirectional amplifiers in one ${19}^{\prime \prime }$ (48.26 cm) rack of 1 U height and 250 mm depth. This is compliant with the specifications of the remote shelters and nodes on the network. Moreover, they are remotely controlled through IP access and included into the supervision and monitoring software discussed below (Section 3.C). The RLSs have been developed by the French company Muquans with a first step of network supervision, and a detailed presentation is shown below.

3. INDUSTRIAL-GRADE RLS

A. Concepts

The concept and principle of operation of RLSs are described in detail elsewhere [6]. Here, we introduce this optoelectronic device from an engineering system point of view. The functional diagram displayed in Fig. 2 shows the main subsystems of an RLS.

 

Fig. 2. Functional sketch of the remote controlled RLS. OC, optical coupler; FM, Faraday mirror; AOM, acousto-optic modulator; PC, polarization controller; PD, photodiode; PLL, phase lock loop; CTL, control. The tracking oscillators used after each photodetection are not shown in this picture. The pink rectangle represents the temperature-controlled box.

Download Full Size | PDF

An RLS has three main roles [5]. First, it regenerates the incoming signal before sending it backward, enabling the compensation of the noise of link $N$. Second, it sends regenerated light to the $N+1$ link and actively stabilizes it. Third, the RLS allows the dissemination of the compensated signal to a local user. This output can also be used without compensation to measure the end-to-end frequency fluctuations between two RLSs, as will be described below.

A narrow linewidth laser diode ($<5\mathrm{kHz}$) with an output power of approximately 20 mW is at the heart of the optoelectronic device. The laser of station $N$ is combined with the incoming signal of the link to generate a beat note, optimized using a polarization controller. This beat note is used to automatically phase-lock the station’s laser onto the link signal, so that the phase fluctuations of the link are copied onto the optical phase of the laser. Typically, 1 to 2 mW of regenerated laser light travel back in the network fiber. The optical signal-to-noise ratio outside the bandwidth of the lock is the one of the laser diode, free of the accumulated spontaneous emission of the EDFAs. This light will experience on the way back the same delay fluctuations before reaching station $N-1$, where the round-trip propagation noise of the link $N$ is detected using a strongly unbalanced interferometer. The propagation noise is corrected by the station $N-1$ using an acousto-optic modulator (AOM).

The next span ($N+1$) is fed with 1 to 2 mW of this regenerated laser light. Again, the same process occurs; the noise of the next span is compensated after regeneration of the light in RLS $N+1$.

B. Design and Realization

The RLSs were developed by Muquans in strong interaction with the scientific academic partners and the RENATER network experts, so that both scientific and telecom requirements were taken into account from the design stage.

Muquans developed the physical package and all the optical subsystems, including the control of the laser source. The electronic subsystem is composed of three main modules: the laser control, the interferometer temperature control, and the link and phase lock loop (PLL) control. The latter was developed by Syrlinks, a company specializing in on-board systems for defense and space applications. The RLS includes advanced automated servo loops and an embedded operating system for local supervision and remote control.

The interferometric setup in the RLS is critical because it houses the optical phase detectors. The stability of the reference arms of the interferometer is essential for a good frequency stability at short and long integration times, because temporal phase fluctuations in these arms will be copied on the link [27]. Specific work was conducted on the interferometer in order to control the optical lengths of the arms. We implemented an integration procedure to splice the components and control the fiber lengths to a few millimeters during the assembly of the interferometers. Given a fiber thermal sensitivity of 37 fs/K/m, this ensures that the temperature dependence is greatly reduced. The components were also selected so that the interferometers have reproducible performance from one to the other. The optical fiber setup is finally housed in a temperature-controlled box, which is itself isolated from the environment, so that results are as much as possible reproducible. The internal temperature of the interferometer is controlled to better than 0.05°C for a 1°C external temperature variation. The temperature of the interferometer is remotely controlled and monitored.

With this production method, a set of 33 interferometers have already been built and characterized for the REFIMEVE+ project. Average thermal sensitivity is 0.8 fs/K with a standard deviation of 0.7 fs/K, which is comparable to the best values obtained in the prototype stations installed in the Paris-Strasbourg link [5], but with a systematic approach.

The electronic processing and control of the signal is an essential subsystem of this technology. Thorough work for rugged design was conducted, with careful selection of robust components. The RLS integrates a complex multifunction electronic board with a standalone and smart mode of operation. This monolithic board handles all the functionalities required to stabilize a coherent link, such as detecting, amplifying, and filtering the signals, tracking the incoming signals, providing the reference and correction signals, and automatically adjusting the polarization.

The industrial version of the RLS includes all the functionalities and automatic control of the laboratory prototypes, including its cycle-slip detector. In addition, several new functionalities and more control and set points were added.

The board is equipped with an field programmable gate array (FPGA) and processor, which control the amplifier gain, the servo loop parameters, and manage the input/output functions. The board also features monitoring of the temperature, input and output levels of the beat notes and RF signals. Several checkpoints of the RF processing chain can be monitored remotely and automatically. Fast Fourier transform of the error signals can be calculated in real time. The bandwidth of the phase lock loops can be measured, saved, and exported remotely, allowing for an efficient remote setting.

The laser device is also remotely operated. In particular, the RLS features a scanning mode, used to look for the beat note with the incoming signal of link $N$.

An embedded computer manages the serial communications with the subsystems, and enables a backup communication with a local operator. Events, such as detected cycle-slips, are logged and time-tagged. The computer saves the data on local memory and communicates with servers through an ethernet port and Internet protocols.

Important work was conducted on the thermal management. We optimized the design and selected low power-consumption components. This slows down the aging of the electronic boards and reduces perturbations on the optical interferometer. The dimensions and weight were defined, taking into account the allocated footprint in the network nodes. The RLS is composed of two ${19}^{\prime \prime }$ (48.26 cm) rack system units, one for the general power supply and the second for the optoelectronic system. The depths are 300 mm and 600 mm, and the heights 2 U and 5 U, respectively. The power consumption is 110 W, and the total weight is 17 kg.

C. Remote Control and Supervision

Thanks to these features, the automation and remote operation of the RLS have been improved. All the lock-loops and their parameters are digitally controlled and optimized through remote access.

With this high level of automation, the operation of the link is made in only three top-level steps. First, thermal regulations and the laser are turned on. Then, the RLS checks the incoming signal and locks the laser diode on it. During this step, the polarization is automatically adjusted. The RF gains can also be automatically set if needed. The RLS optimizes all the parameters until the laser is locked. Finally, the $N+1$ and local links are locked when a return signal is detected.

The RLS is programmed to automatically lock once it is started. If the RLS detects an input signal failure, it will run the parameters optimization loop until a satisfying lock is found again. Alarm signals are automatically generated, and an operator can remotely control the station in manual mode if necessary.

Each piece of equipment can be controlled individually. The scientific operators of the network can monitor the status and check measurement data of all the instruments. They can tune parameters in order to optimize the performance and stability of the metrological network. On top of that, a global supervision system was implemented with a dedicated database and several data visualization functions. Furthermore, RENATER’s Network Operation Center (NOC) has direct access to supervise, monitor, and control all the equipment. In case of a maintenance operation happening on the network, the NOC can shut down any of the REFIMEVE+ equipment at any time for the best safety at work and service priority organization. Although we have not encountered emergency situations requiring the use of this function, we have validated it in our laboratory. Our work is still in progress, but in the near future, end users will also be able to connect to the supervision server in order to monitor the status of the metrological service and the validity of the ultrastable signal delivered to their laboratory in real time.

D. Operational Validation

The quality control of each subsystem is carried out before integration by Muquans. Here we focus on the final stage, where the performance of the overall system is assessed.

Indeed, all the RLSs constructed for the REFIMEVE+ project are thoroughly tested in the laboratory before being installed on the RENATER network. At present, 22 RLSs have been manufactured and tested.

For that purpose, we have built a test bench, sketched out in Fig. 3. We use an ultranarrow laser diode as reference laser (OEWaves Ultra-Narrow Linewidth Laser Module-Gen 3), with frequency noise of about $3{\mathrm{Hz}}^2/\mathrm{Hz}$ for a Fourier frequency above 1 kHz. The laser frequency is kept close to 194.4 THz and is monitored on a regular basis using a calibrated wavemeter.

 

Fig. 3. Scheme of the test bench with a simulated link of 200 km, including two bidirectional OADMs and one bidirectional EDFA. For the noise floor measurement, the 200 km link is replaced by a short single-mode fiber.

Download Full Size | PDF

We use two RLSs $N$ and $N+1$, $N$ being the device under test and $N+1$ being used both to send regenerated light backward in the link and to measure its stability. Station $N$ is locked on the reference laser. As a first step, optimal settings for the laser of this station $N$ are found; then its output port ${\mathrm{Out}}_{N+1}$ is connected to the input port ${\mathrm{In}}_{N+1}$ of the second RLS, either by a short link of 1 m or by a 200 km fiber link made of fiber spools, two OADMs and one bidirectional Keopsys EDFA. On this 200 km fiber link, we expect the overall noise level to be higher than on a real link of the same length because of the phase fluctuations due to the highly correlated temperature changes. Nevertheless, it permits the evaluation of the RLS functions over a long fiber link before installation and gives an upper value of the link instability.

The ability of the RLS $N$ and $N+1$ to actively compensate the noise of the link is checked by measuring its end-to-end stability. To achieve this, the user outputs of the two RLSs are connected with a short fiber link that includes an AOM. This setup allows for the measurement of the end-to-end frequency fluctuations of the link, using a two-way frequency comparison method as first introduced in [28]. The beat note signal between the laser of the station $N$ and the one transmitted on a short fiber from station $N+1$ and frequency-shifted with the AOM is detected on the $\mathrm{PD}{3}_N$ of station $N$. In the same way, the beat note between the laser signal of station $N+1$ and the one transmitted on a short fiber from station $N$ and frequency-shifted with the AOM is detected on the $\mathrm{PD}{3}_{N+1}$ of station $N+1$. By postprocessing these beat notes, one obtains the phase difference between the user outputs signal of the two stations, which gives the end-to-end phase fluctuations of the link. The beat note signals between the two laser stations are counted with a dead-time-free frequency counter ($K+K$ Messtechnik FXE80) operated in lambda mode with a 1 s gate time.

We first measure the noise floor of the RLS under test using a very short link with a few meters of fiber. The resulting Allan deviation is shown in Fig. 4. Frequency stabilities below $2\times {10}^{-17}$ at 1 s and a few ${10}^{-20}$ for $\tau >{10}^3\mathrm{s}$ are systematically achieved. The small excess of noise above $\tau \simeq 2000\mathrm{s}$, due to the thermal periodic perturbation of the air conditioning system acting on the station interferometer, is below $2\times {10}^{-20}$. The long-term instability is as low as $2\times {10}^{-21}$ at 1-day averaging time. This instability is well below that of the best optical clocks and shows that the RLS will not limit the link instability at the ${10}^{-20}$ level. Moreover, continuous operation of the stations over several weeks has been demonstrated.

 

Fig. 4. Fractional frequency instability calculated from data recorded with $\mathrm{\Lambda }$-counter and 1 s gate time versus averaging time. Dark red squares, modified Allan deviation for a compensated 200 km laboratory link; light red circles, modified Allan deviation for the free-running 200 km laboratory link; dark green stars, modified Allan deviation for a short link to estimate the noise floor of the RLS; light green diamonds, modified Allan deviation for the free-running short link.

Download Full Size | PDF

We then replace the short fiber link by the 200 km fiber link on spools. After optimization of the laser settings, we obtained the results that are shown in Fig. 4. The stability at 1 s is $1\times {10}^{-15}$ and decreases to below ${10}^{-19}$ for $\tau >{10}^4\mathrm{s}$, limited by the stability of the reference laser [29]. One can again observe a small excess of noise at $\tau \simeq 2000\mathrm{s}$, due to the thermal periodic perturbation of the air conditioning system acting on the fiber spools’ bank. Note that this second test stage was only carried out with the first few industrial-grade stations in order to validate all the RLS functions before in-field setup.

4. DEPLOYMENT OF A 680 km OPTICAL LINK: PARIS-LILLE

The first industrial-grade link of the REFIMEVE+ network was deployed between two cities in France, Paris and Lille, with a geographical distance of about 250 km (Fig. 1). The sketch of the experiment is depicted in Fig. 5. The link is made of two parallel 340 km long fiber spans from the RENATER network, one being for the uplink from Paris to Lille and one for the downlink back to Paris. In total, the metrological signal travels approximately 680 km. Two RLSs were installed in a RENATER node in Paris at the input and output ends of the link and one RLS in the node at Lille. Four EDFAs were installed in the telecom shelters with 10 OADMs along the link on each span. The optical attenuation on each span is approximately 80 dB. The total gain of the EDFAs is limited to 50 dB on each span of this link in order to prevent self-oscillations in the fiber.

 

Fig. 5. Sketch of the Paris-Lille-Paris link. USL, ultrastable laser. Black triangles are the unidirectional EDFA. Double red triangles are the bidirectional EDFA. Blue trapezes are the OADMs. For each span, we provide the span length and the span losses. We remotely supervise and optimize four double bidirectional EDFAs and three RLSs. Red lines refer to the ultrastable signal.

Download Full Size | PDF

The 680 km link between Paris and Lille was deployed during the summer of 2017. Installing all the RLSs and EDFAs in the RENATER nodes and shelters took approximately one week. Once all the equipment was set up, remote communication was checked, and the link was stabilized without any malfunction. At that stage, the lasers were locked, but many cycle slips degraded the stability and the accuracy of the frequency transfer. The best stability level, as shown in Fig. 6, was reached after further optimizations that lasted approximately three weeks. Because the RLSs and EDFAs were fully tested before installation, this period of time was dedicated to adjusting the amplifier and lock-loop gains, focusing on the dynamics of the link itself rather than on the equipment. Overall, the installation and adjustment process took under a month. Once the phase lock-loop parameters and the gain of the optical amplifiers were optimized, the cascaded link was operated continuously for several weeks. Even when perturbations occur because of thermal fluctuations, acoustic noise, or activity on the network, events are time-tagged and the RLSs automatically relock in less than 1 min. The interferometer temperature experienced fluctuations with a standard deviation of .065 K during the acquisition of the data.

 

Fig. 6. Fractional frequency instability calculated from data recorded with $\mathrm{\Lambda }$-counter and 1 s gate time versus averaging time. Black solid circles, modified Allan deviation for the Paris-Lille free-running link; blue solid and hollow squares, overlapping and modified Allan deviation, respectively, for the Paris-Lille-Paris compensated link.

Download Full Size | PDF

To estimate the stability of the link, we measure the end-to-end signal between the two RLSs in Paris, using again a dead-time-free frequency counter operated in $\mathrm{\Lambda }$-mode with a 1 s gate time. By measuring the correction signal applied to the AOM of the first station synchronously (station 1 in Fig. 5), we can also estimate the free-running noise on the first span of the link.

Figure 6 shows the free-running stability and the overlapping and modified Allan deviation of the compensated link during a continuous five-day measurement. When the link is compensated, we achieve a frequency stability of $5.4\times {10}^{-16}$ at 1 s integration time, which averages down to $1.7\times {10}^{-20}$ at 65,000 s. At short term, it is limited by the propagation delay and at long term by laser and interferometer noise [27,29–31]. This stability is comparable to state-of-the-art ultrastable optical frequency transfer obtained by research laboratories.

The accuracy of the frequency transfer is assessed by calculating the mean of the frequency data set, and the statistical uncertainty given by the overlapping Allan deviation at long integration time [32]. We took the upper error bar at 65,000 s for a conservative estimate. One obtains $-3\times {10}^{-21}\pm (3\times {10}^{-20})$. Note that the uncertainty is not limited by the inaccuracy of the local RF oscillator [5].

In order to better appreciate the performance of the link, Fig. 7 shows the end-to-end phase fluctuations during this measurement. Over this time period, the RLS tagged 23 cycle slips. Most of them correspond to polarization adjustments. In total, 285 data points were removed from the 432,000 s long data. The integrated phase fluctuations of the end-to end signal are in a range of 10 rad over five. The corresponding time error, expressed in femtoseconds, is less than 8 fs. The uptime over this period is as high as 99.93%.

 

Fig. 7. End-to-end phase fluctuations after propagation over a 680 km fiber link.

Download Full Size | PDF

This performance, both in terms of stability and accuracy, meets the requirements of the REFIMEVE+ project.

5. CONCLUSIONS AND PERSPECTIVES

We demonstrated the setting up and exploitation of the first industrial-grade coherent optical fiber link as part of the REFIMEVE+ network. We show a state-of-the-art relative frequency stability and accuracy at 65,000 s integration time below the ${10}^{-19}$ level over a cascaded link of 680 km. Moreover, the industrial-grade RLS contribution to the instability is below $2\times {10}^{-20}$ for integration times higher than 1000 s. We also report an uptime of 99.93% over five continuous days, which is very promising regarding the effectiveness of REFIMEVE+ to the end user.

With the work presented here, we totally fulfill the requirements of the REFIMEVE+ project. We have demonstrated the basic tools for the upcoming extension of this network. We have also built a test bench to validate the serial production of RLSs.

Furthermore, we have set up our production capacity for a full deployment of the network within the next two years. The deployment time we can achieve is now comparable to that of a telecommunication link for data traffic, which shows our high level of maturity and alleviates concerns about the control of schedule and costs. With this work, the technological readiness level (TRL) of the RLSs was increased from 5–6 to 8. The TRL of the cascaded links with EDFAs and RLSs is now set as high as 8 [33].

Regarding the hardware developments, the next steps are to reach the same level of maturity for the end user and eavesdropping setups. Another major concern is the central hub of the network that is being developed with the concept of superstation [34]. Finally, one crucial challenge relates to network supervision. Supervision software tools need to be further developed and implemented. We demonstrate here our ability to gather much information into a complex database. The next challenge is to optimally utilize this database for the most effective network and the best scientific use. Our objective of disseminating frequency standard with high availability is now close at hand.